My first job out of residency was at a rural critical access hospital where I practiced full-spectrum family medicine. I mainly saw patients in a primary care clinic, but I also worked as the hospitalist, ER doctor, and OB provider. The closest big hospital was a 3-hour drive or 45-minute flight away, so when a septic patient rolled into the ER, my first question was whether this was something we could manage or if we had to get this patient on a plane. While the medical complexity was part of the calculus, sometimes we had to ship patients because we didn’t have enough nurses.

My current job is as a family medicine residency faculty member at a large academic medical center. I mostly see patients and precept residents in the primary care setting and occasionally supervise residents on an inpatient hospitalist service. When a patient brings up a complaint like mood, my first question is whether I should manage it or refer to my psychiatry colleagues.

These two jobs are very different, but in both situations I saw myself as a primary care doctor. By that, I meant that I am the primary person responsible for coordinating your medical care. Kind of like a medical co-pilot who identifies a patient’s medical needs and then triages them to the best setting to receive that care. Sometimes that’s with me in the clinic, other times it’s with me in the hospital, yet other times it’s with someone else entirely.

There is an oft-quoted statistic that primary care doctors need 26 hours per day to adequately care for an average patient panel of 2,500 patients (Porter et al., 2022). These 26 hours are made up of 14 hours of preventive care, 7 hours of chronic disease management, 2 hours of acute concerns, and 3 hours of documentation. If an average primary care doctor works 32 patient-facing hours per week, has 20-minute appointment slots, and works 48 weeks per year, that is 4,608 appointments per year, or 1.8 visits per patient. This barely feels like enough time to go over the preventive care options, let alone any problems the patients actually bring. Primary care is actively evolving to address this 26-hour problem.

This problem is made up of three variables:

1. Hours per day.

2. Number of patients.

3. Things to do per patient.

I see a few ideas to address these variables.

One idea is to work more. Primary care doctors are already quitting in droves, and nobody in society is particularly interested in working more.

A second idea is to see fewer patients. If we brought panel sizes down to the 500-1,000 range, we could spend adequate time with patients. However, even if we have more time, our patients still have a fixed amount of time, money, and energy. Not all patients want to spend hours with their doctor.

A third idea is to delegate. For instance, rather than spending 10 minutes on an in-depth discussion of the risks and benefits of colon cancer screening, we could have RNs call patients to tell them they’re due for a Cologuard and automatically refer anyone with a positive result to the endoscopy center. This takes the problem off our hands, and it’s happening as we speak. Some recommendations can be appropriately delegated to support staff like RNs or pharmacists, but I think we need to be intentional about which ones we delegate.

The first three ideas are about increasing capacity. More clinician time, fewer patients, or more people sharing the work. My final idea is to prioritize the recommendations that make up the 26 hours.

On this Substack, I spend a lot of time addressing differences between guidelines on the same topic by different societies. For instance, the ACP recommends colon cancer screening at age 50, whereas the USPSTF recommends starting at age 45 (Qaseem et al., 2023; USPSTF et al., 2021). I think this question is important enough that I wrote an entire post about starting colon cancer at age 45 vs 50. But at the end of the day, the two guidelines are pretty similar. They both endorse colon cancer screening, just at slightly different ages.

Another example is blood pressure management. The AAFP recommends a goal <140/90, whereas the ACC/AHA recommends 130/80 (Coles et al., 2022; Jones et al., 2025). I wrote about this question also, but again, at the end of the day they both endorse treating blood pressure, just to slightly different targets.

In the primary care clinic, we have competing interests. We have the 47-year-old with a blood pressure of 135/85 who is flagging for colon cancer screening. Ideally, we discuss both things, but appointment times are always shrinking and the inbox is always growing. The more pertinent question is, will colon cancer screening or hypertension management add more value to the patient’s life. And these are not the only two competing issues.

The USPSTF is trying to address this problem, but here is the list of every USPSTF recommendation that received an A or B grade (moderate to high net benefit):

This list is illegibly tiny because it has over 50 recommendations, and this is just one list. The ACIP has immunization recommendations. The ACC/AHA has lipid management guidelines.

I think being a primary care doctor will increasingly be a balance of what the patient wants to address with what the medical system recommends for us to address. To strike this balance we need to be equipped to think beyond whether colon cancer screening is good. It is. We need to think about how good colon cancer screening is, and whether discussing that topic today is the best use of our, and our patient’s, time.

To this end, I am working on an experiment called the Primary Care Power Rankings. I envision a regularly updated tab on this Substack with a subjective ranking of objective data about common primary care interventions. The goal is not to create a definitive ranking to replace the USPSTF or to tell primary care providers what to do. Rather, I want to offer an accessible framework to compare the value of different primary care recommendations.

Admittedly, this is a lofty project. I would love to collaborate with anyone interested in writing evidence reviews and thinking through how to compare the things we do every day. If anyone is interested but feels unequipped, I am happy to provide training. Just send me a chat. Otherwise, stay tuned.

References

1. Coles S, Fisher L, Lin KW, Lyon C, Vosooney AA, Bird MD. Blood Pressure Targets in Adults With Hypertension: A Clinical Practice Guideline From the AAFP. Am Fam Physician. Dec 2022;106(6):Online.

2. Jones DW, Ferdinand KC, Taler SJ, et al. 2025 AHA/ACC/AANP/AAPA/ABC/ACCP/ACPM/AGS/AMA/ASPC/NMA/PCNA/SGIM Guideline for the Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol. Nov 4 2025;86(18):1567–1678. doi:10.1016/j.jacc.2025.05.007

3. Porter J, Boyd C, Skandari MR, Laiteerapong N. Revisiting the Time Needed to Provide Adult Primary Care. Journal of General Internal Medicine. 2022 Jul 1;38(1)doi:10.1007/s11606-022-07707-x

4. Qaseem A, Harrod CS, Crandall CJ, et al. Screening for Colorectal Cancer in Asymptomatic Average-Risk Adults: A Guidance Statement From the American College of Physicians (Version 2). Ann Intern Med. Aug 2023;176(8):1092–1100. doi:10.7326/M23-0779

5. USPSTF, Davidson KW, Barry MJ, et al. Screening for Colorectal Cancer: US Preventive Services Task Force Recommendation Statement. JAMA. May 18 2021;325(19):1965–1977. doi:10.1001/jama.2021.6238

I started writing this as my third and final colon cancer screening post. My first post broadly reviewed the RCTs, my second post was a more focused discussion of the age to start colon cancer screening, and here I started by thinking about which modality is the best.

As I was working through the evidence comparing the diagnostic tests, what I found was a few RCTs that compared different colon cancer screening modalities and measured patient-oriented outcomes like colon cancer mortality, but mostly I came across studies that evaluated various test characteristics like sensitivity and specificity. In order to figure out if these papers mattered, I went down a rabbit hole of sensitivity and specificity. Here, I will attempt to summarize what I learned by using FIT as an example diagnostic test and hopefully set the stage to compare the various modalities for colon cancer screening.

FIT

First, what is FIT? FIT is one of several stool-based colon cancer screening tests. Specifically, FIT is an antibody-based assay that detects the globin protein, which makes up hemoglobin. As globin goes through the digestive tract, it gets degraded and becomes undetectable by FIT, so it is supposed to be more specific for lower GI bleeds as opposed to upper. It also only detects human globin, so it does not cross-react with other species.

Other stool tests include fecal occult blood tests (FOBT) and stool DNA tests.

FOBT uses guaiac, which is a residue from a guaiacum tree that changes color when exposed to peroxidase activity. Heme, a component of hemoglobin, catalyzes this peroxidase reaction. Unfortunately, this peroxidase activity is very commonly found in many foods including vegetable and animal meats and patients have to restrict their diet for 3 days, so no red meat, raw vegetables or fruits, NSAIDs, iron supplements, or heartburn medications.

The stool DNA test detects DNA related to colon cancer along with the presence of hemoglobin.

The FOBT gives you a yes or no answer, and the stool DNA tests use proprietary algorithms to give a yes or no. The FIT is ideal for our discussion of sensitivity and specificity because it measures a concentration of globin that is then converted into a positive or negative answer. How we determine the cutoff level of globin to determine positive or negative can have drastic effects on its diagnostic performance.

Sensitivity and specificity

Now let’s transition into everyone’s favorite Step 1 topic: sensitivity and specificity. Here is the classic 2x2 table (True positive = TP, true negative = TN, false positive = FP, and false negative = FN):

Given a population, a certain number will be sick, and a certain number will be healthy. If you test everyone in that population, you will get some positives and some negatives. The definitions for sensitivity and specificity:

Sensitivity represents the left column.

Sensitivity is focused on the sick people. It tells you the positive test rate among sick people. A 100% sensitive test should be positive for all sick people. Another way of saying this is that a 100% sensitive test should not be negative for any sick person. This is where SNOUT (SeNsitivity rules OUT) comes from. If you have a highly sensitive test and your patient tests negative, you should be able to confidently rule out sickness in this patient.

Specificity represents the right column:

Specificity is focused on the healthy people. It tells you the negative test rate among healthy people. A 100% specific test should be negative for all healthy people. Another way of saying this is that a 100% specific test should not be positive for any healthy person. This is where SPIN (SPecificity rules IN) comes from. If you have a highly specific test and your patient tests positive, you can confidently rule in sickness for this patient.

If we update this example to colon cancer and FIT, we get this:

And now let’s get back to FIT.

What is a positive FIT? When you run a diagnostic test like a FIT, it tells you a certain concentration of globin protein in the stool which is reported as mcg of hemoglobin per g of stool. Say you run a bunch of FITs, represented below as poorly drawn dots, and each FIT gives you a certain concentration of Hgb/stool:

Now consider that we are all-knowing gods that knows the true colon cancer status of each poorly drawn dot. The green dots are healthy and the red dots have colon cancer:

If the results of FIT were reported as a concentration of hemoglobin per stool, then every time a FIT result pops into my inbox I would have to go to UpToDate to find the concentration at which to recommend colonoscopy. Thoughtfully considering the risks and benefits of a FIT cutoff is great for Substack, but in the middle of a busy clinic day I just need a quick answer so I can send the patient for colonoscopy or provide reassurance and move on. So, to make it easier on me there are certain cutoff concentration at which to consider a FIT positive or negative. Everything above the cutoff concentration is positive and everything below is negative.

If we have a low cutoff, then all the red dots would test positive, and we would have a highly sensitive test that captures everyone with colon cancer.

Say we decided as a society that everyone with colon cancer had to be sacrificed to the all-knowing gods. However, the gods got mad at us silly humans because we sacrificed too many healthy people. In response, we increase the cutoff.

Now, we have a more specific test, and no healthy people would be sacrificed. However, now the gods are mad because we aren’t sacrificing enough people. At this point you are probably wondering if there is some complicated math thing that can identify the optimal cutoff point. And that is kind of what the receiver operating characteristic (ROC) analysis does. ROC curves plot the sensitivity against the 1-specificity. The 1-specificity focuses on the same healthy population as the specificity:

1-specificity tells you the positive test rate among healthy people aka the false positive rate. Here is an example ROC curve:

I replaced 1-specificity with false positive rate to make it more intuitive. To bring this back to FIT, decreasing the cutoff moves the line to the right of the ROC curve. This allows you to detect more people with cancer, so your sensitivity goes up, but you are also capturing more healthy people.

The goal of the ROC curve is to decide the cutoff point that optimizes the sensitivity and specificity. I found many complicated statistical methods to identify the optimal point on the ROC curve, but there is also a clinical component to this optimization. Depending on the clinical situation you may value sensitivity and specificity differently. In our scenario where we sacrifice everyone with colon cancer, it seems bad to say someone has colon cancer when they really don’t, because that person would be needlessly sacrificed. In this case, a highly specific test seems better. In other cases, we may want to prioritize sensitivity.

Going back to FIT, the usual cutoff is 20 mcg of hemoglobin per g of stool. This cutoff was determined based on many studies. Here is one of them:

Brenner and Werner, 2017

• Population: Patients undergoing screening colonoscopy in Germany (n = 1,822). Patients were excluded from the analysis if they were younger than 50 or older than 80 years old, if they had a history of colon cancer or inflammatory bowel disease (IBD), if they had a recent colonoscopy, or if they had an inconclusive colonoscopy.

• Methods: FOB gold was evaluated for sensitivity, specificity, and PPV for advanced adenoma across cutoffs from 1-50 mcg hemoglobin per g stool. Advanced adenoma was defined as: >1 cm, tubulovillous or villous histology, or high-grade dysplasia.

• Results: Across the study 0.8% of patients had colon cancer, 10.6% had an advanced adenoma, and 19% had a non-advanced neoplasm.

  • Using 20 mcg (the standard) as the cutoff to detect colon cancer or advanced adenoma:

    • Sensitivity 34%

    • Specificity 95%

    • PPV 51%

  • If we increased the cutoff to 50 mcg, we get this:

    • Sensitivity 20%

    • Specificity 97%

    • PPV 54%

  • If we decrease to 10 mcg, we get this:

    • Sensitivity 44%

    • Specificity 90%

    • PPV 36%

  • If we keep decreasing to 5 mcg, we get this:

    • Sensitivity 66%

    • Specificity 60%

    • PPV 17%

  • Due to small sample size, the authors only reported selected results for colon cancer only (not including advanced adenomas) but did note that when using a cutoff of 20 mcg they detected all 14 cases of colon cancer and that the specificity was similar.

A Cochrane meta-analysis that summarized many studies like this one reported similar numbers (Grobbee et al., 2022):

• At 10 mcg, to detect colon cancer or advanced adenoma:

  • Sensitivity 33%

  • Specificity 93%

• At 20 mcg, we get this:

  • Sensitivity 26%

  • Specificity 97%

• The power of the meta-analysis is in its statistical power. Because they are pooling so many studies, they have many more colon cancer cases and can more confidently report results specifically for detecting colon cancer and not an advanced adenoma. Here are the results for 10 mcg:

  • Sensitivity 76%

  • Specificity 94%

• And 20 mcg:

  • Sensitivity 65%

  • Specificity 96%

A ROC curve analysis found the optimal cutoff point at 19.8 mcg (Vanaclocha-Espi et al., 2021).

Conclusions

Okay, so where does that leave us.

1. There is not a correct cutoff for FIT. Moving the cutoff is a balancing act between sensitivity and specificity. Looking at the data, 20 mcg looks to be a reasonable cutoff. I would have been interested to know the test characteristics at 50 mcg for colon cancer.

2. The question I’m left thinking about is: sensitivity and specificity for what? Is our goal to detect colon cancer only? Should we be trying to detect advanced adenomas so these patients can get a colonoscopy to remove the adenoma? Should we be looking for any colon abnormality that could increase risk for colon cancer? This question is a balance of the benefits of early detection with the harms of procedures like colonoscopy, and I hope to address this question in my next post where I compare different modalities.

References

1. Brenner H, Werner S. Selecting a Cut-off for Colorectal Cancer Screening With a Fecal Immunochemical Test. Clin Transl Gastroenterol. Aug 3 2017;8(8):e111. doi:10.1038/ctg.2017.37

2. Grobbee EJ, Wisse PHA, Schreuders EH, et al. Guaiac-based faecal occult blood tests versus faecal immunochemical tests for colorectal cancer screening in average-risk individuals. Cochrane Database Syst Rev. Jun 6 2022;6(6):CD009276. doi:10.1002/14651858.CD009276.pub2

3. Vanaclocha-Espi M, Ibanez J, Molina-Barcelo A, et al. Optimal cut-off value for detecting colorectal cancer with fecal immunochemical tests according to age and sex. PLoS One. 2021;16(7):e0254021. doi:10.1371/journal.pone.0254021

I have a quick read method for going through RCTs that I have been teaching to family medicine residents, and I thought I’d share it here. I call it the What, Who, Result method.

The goal of this method is to work through a paper in 5-10 minutes, and it’s targeted towards the time-strapped resident or practicing primary care doctor. I primarily use it for papers that I’m already committed to reading because someone brought it to my attention. I’ll use this paper about Paxlovid for COVID as an example (PDF here):

I heard about this paper on a podcast episode of Sensible Medicine, which is a Substack I really like about evidence-based medicine.

I don’t use this method if I’m browsing through papers trying to decide what to read. In that case, I skim the title or abstract. I also don’t use it if for an in-depth critical appraisal when I really want to understand the methodology of a trial.

The method consists of three steps: What, Who, Result.

1. WHAT was the intervention?

I start by looking at the title to make sure this is something I want to read. For outpatient COVID, the only medicine to even consider is Paxlovid. I rarely test patients for COVID, but patients frequently get positive home COVID tests and message me about whether they should take Paxlovid.

Next, I skip the abstract because I’m already interested in and committed to reading the article. I’ll get all the details in the abstract from my read-through, and the abstract usually omits pertinent details I need to decide if the results are relevant to my patients.

I also skip the introduction because it’s either clinical background that’s too basic for a practicing clinician or research intricacies that are too esoteric to be useful.

I then move on to the methods. The “what” details are usually in the middle of this section, typically under a header like Interventions, Trial Protocols, etc. In our case, on page 5 we have a section called Procedures:

The treatment is pretty straightforward, the standard dose for Paxlovid. However, the details can be much more important for other trials. For instance, if the paper is about statins, you might want details like high- vs moderate-intensity treatment.

An important missing detail here is the control group. This Procedures section doesn’t mention a placebo, so I had to look a little further to figure out how the control group was handled. I found it in a section on page 4 called Randomization and Masking:

The control group received “usual care” and did not receive a placebo pill. This is potentially important because the Paxlovid group might get a placebo boost from simply knowing that they took a pill for COVID.

2. WHO was in the study?

The first place I look for the “who” details is near the beginning of the methods section under a header like Study Population, Inclusion/Exclusion Criteria, etc. In our case, on page 4 we have a section called Objectives, Patients and Oversight:

Unfortunately, the information here is quite limited and the critical details are buried in the supplement:

Having to go back and find the supplement really kills the quick-read aspect of this method.

Finally, I move on to the results section, but I don’t read any of the text or look at any of the figures. Instead, I focus entirely on a couple tables. Typically, there is a Table 1 that shows baseline characteristics:

I like getting additional details about the kinds of patients who were actually studied, and I usually focus on things like age, sex, and comorbidities. In this paper, vaccination status is also relevant because a previous paper showed that Paxlovid decreased progression to severe COVID in unvaccinated patients.

3. And what was the RESULT?

I recommend reading the result last for the same reason I teach medical students to look at the ST segment of an EKG last. They should first go through the rate, rhythm, T waves, QRS interval, etc. Yes, ST elevation is important, but if you graduated from medical school, I promise you will not miss ST elevation on an EKG. New RBBB or T-wave inversion in V3 that wasn’t present on a previous EKG, however, might be easier to miss. So you work through the annoying details of a trial before you get to the punchline.

For the “result” details I go back to the methods and scan for a header like End Points, Outcomes, etc. Here we have Outcome Measures:

The important thing to think about here is whether this is an outcome you and your patient actually care about. In this case, if I were thinking about taking Paxlovid, I would consider avoiding a hospitalization to be a meaningful benefit. This is also important because hospitalization is a fairly objective outcome. A little placebo boost from taking a pill might make you feel less fatigued, but it is much less likely to affect whether you ultimately need to be hospitalized.

Then I move on to Table 2, which usually contains the actual outcome you care about:

The final result isn’t really important for this activity, but this study did not demonstrate a decrease in hospitalization or death with Paxlovid.

Sometimes there is also a Table 3 of adverse events. In this case, those are combined with the primary outcome results in Table 2.

Finally, I don’t read the discussion for the same reason I don’t read the introduction. It’s either clinical background that’s too basic for a practicing clinician or research intricacies that are too esoteric to be useful

Conclusions

And that’s it. It’s not perfect, and if you really want to get into the nitty-gritty of critical appraisal this isn’t the approach for you. However, from the perspective of a busy practicing doctor who heard about a new study on Paxlovid and wants to know whether it matters for their patients, I think this method is good enough.

References:

Butler CC, Pinto AD, Harris V, et al. Oral Nirmatrelvir–Ritonavir for Covid-19 in Higher-Risk Outpatients. New England Journal of Medicine. 2026–04–23;394(16)doi:10.1056/NEJMoa2502457

This is my second post on colon cancer screening. My first post reviewed the RCTs to determine the efficacy of colon cancer screening. In this post I am continuing to think about the benefit of starting colon cancer screening at age 45 instead of age 50. I would recommend reading the first post before this one because I will refer to those RCTs.

Various societies offer different starting ages for colon cancer screening. The USPSTF recommends 45, the American College of Gastroenterology (ACG) recommends 50, and the American College of Physicians (ACP) recommends 50 (Qaseem et al., 2023; Shaukat et al, 2021; USPSTF et al., 2021). So, what makes the USPSTF different?

The two pieces of evidence that USPSTF cites for this recommendation are epidemiological studies showing an increasing incidence of colon cancer in young people and modeling studies showing a benefit to starting colon cancer screening at age 45.

Epidemiology

Let’s start with the epidemiology. Below is a figure of colon cancer incidence from an NEJM review (Sinicrope et al., 2022). The x-axis is the year, and the y-axis is the number of cases per 100,000. Panel A shows the number of new colon cancers and panel B shows deaths from colon cancer.

The author describes this figure, “An alarming increase in cases of early-onset colorectal cancer, defined as diagnosis in patients younger than 50 years of age, has occurred in the United States.” Allow me a brief aside into adjectives. Adjectives are describing words: big, small, boring, alarming. Adjectives are relative. What is boring to one person may not be boring to another. But I would argue that adjectives are relatively relative. Take for instance roller coasters. I don’t like roller coasters, but it would be hard to describe them as boring. In fact, that’s part of the reason I don’t like them. At this point you are likely wondering why I didn’t edit out this paragraph, so let me get back to the point. The author describes this figure as, “An alarming increase in cases of early-onset colon cancer [emphasis mine].” Alarming is relative. It’s possible that going from 5 to 7 new diagnoses of colon cancer per 100,000 people is alarming to some people. For me, I think of alarming as an alarm going off in my head. I would consider my kitchen being on fire to be alarming. In a medical context, I consider it alarming if a patient walks into my clinic with a blood pressure of 78/52. In the context of cancer epidemiology, I’m not exactly sure what I would consider alarming, but when I saw that colon cancer incidence increased from 5 to 7 per 100,000 people, I did not feel my adrenaline ramping up because I was terrified about what would happen next.

An important thing to think about when you interpret data is context. Okay, yes, colon cancer incidence is increasing in people ages 0-49, but what is the big picture? Below is colon cancer incidence in people ages 0-49 and also people 50-64 and 65+.

The first thing you can clearly see is that even with a drastically decreased incidence of colon cancer in ages 65+, this age group still has the highest incidence of colon cancer.

Second, if you look closely, you can see that the age <50 blue line is trending upwards, but you have to squint. You could make the argument that this small increase is clinically and societally important. It is gut-wrenching to take care of a patient that dies of cancer in their 40s, but maybe we could be a little more intentional about our use of adjectives and describe this data as concerning or unnerving.

Modeling studies

The USPSTF commissioned a modeling study to evaluate the benefits of colon cancer screening by different modalities to start at different ages (Knudsen et al., 2021). I am not an expert of modeling studies, so I won’t provide a critique of the methodology, and let’s work under the assumption that this modeling study is correct.

The authors predict that colon cancer screening adds 0.1-0.3 years of life per patient. This would be 36-108 days. Various colon cancer screening strategies were tried, and the more aggressive screening regimens were found to be the most life-prolonging. The most aggressive regimen was q5 year colonoscopies from age 45-85. That would be 9 total colonoscopies if everything was completely normal.

They specifically predicted that starting colon cancer screening at age 45 instead of 50 extended lives by 0.02 years, or 7 days.

This is alarming. This modeling study is interesting, and I like the idea of modeling cancer screening methods so we can pick out the optimal strategy to test by RCT. It should not, by itself, lead to conclusive recommendations. My pre-conceived bias about modeling studies is that I do not trust them. Every time I watch an NBA game the talking heads are telling me about the Draft Kings over/under’s. It’s interesting to talk about these predictions, but they are frequently wrong. This is why basketball games are actually played instead of simulated. There is lots of money in predicting these lines, and if Draft Kings can be wrong, then scientists trying to model colon cancer mortality can be wrong.

I find it concerning that a modeling study would be used to rubber stamp the recommendation to start colon cancer screening at age 45, but I find it alarming that this was based on a benefit of 7 days of added life.

RCTs

And what does the RCT-level evidence actually say about screening in the age 45-50 population? There are no RCTs comparing screening starting at age 45 vs 50 and only two FOBT RCTs included patients age 45-50 (Faivre et al., 2004; Kronborg et al., 2004). These studies did not report subgroup analyses for this 45–50-year-old age group. The Minnesota FOBT study did do subgroup analyses by age, and they did not find a statistical effect of age on colon cancer mortality. Looking at the data below, it does look like there is a greater effect of screening on colon cancer mortality for older patients (Shaukat et al., 2013).

Conclusions

Let’s take what we know.

1. Incidence of colon cancer is increasing for young people, although both the relative increase and the absolute number of people 45-50 years old getting diagnosed with colon cancer are low.

2. A modeling study found a benefit of 7 days to colon cancer screening when started at age 45.

Now let’s take what we don’t know:

1. By starting colon cancer screening earlier, are we meaningfully detecting colon cancer?

   1. Gil Welch uses a barnyard analogy to describe three types of cancers that we can find by screening: turtles, birds, and rabbits. In this analogy the animals escaping the barnyard is akin to dying of cancer, and cancer screening is the methods we use to prevent the animals from escaping. Turtles are slow-growing cancers that were never going to harm the patient. Identifying these cancers is meaningless and can only lead to harm. Birds are aggressive cancers that have already metastasized at the time of screening. Identifying these cancers is meaningless because there is nothing we can do to help this patient. The rabbits are the one’s we want to catch.

   2. It’s possible that the colon cancers we are detecting by starting colon cancer screening at age 45 are rabbits and this is truly life-changing, but we cannot know for sure without the RCTs.

2. Whether colon cancer screening is useful at age 50-55.

   1. While I was writing this post I started to think about whether 50 is even the right age to start colon cancer screening. The most convincing RCTs for colon cancer screening are the flex sig studies, which consistently show an improvement in colon cancer incidence and colon cancer-specific mortality, and when pooled show an improvement in all-cause mortality. 3 of the 4 flex sig RCTs enroll patients starting at age 55. The one study that started enrollment at age 50 did do a subgroup analysis comparing patients age 50-55 vs 55-64 and they did not find obvious differences (Botteri et al., 2026).

I personally do not consider 7 added days of life meaningful, and I question whether an RCT would replicate this result. I would recommend starting colon cancer screening at age 50 and I am even starting to question if 55 is the more correct age.

References

1. Botteri E, Holme O, Loberg M, et al. Twenty-Three-Year Benefits of Sigmoidoscopy Screening for Colorectal Cancer : A Randomized Trial. Ann Intern Med. Jun 2026;179(6):804–811. doi:10.7326/ANNALS-25-05456

2. Faivre J, Dancourt V, Lejeune C, et al. Reduction in colorectal cancer mortality by fecal occult blood screening in a French controlled study. Gastroenterology. 2004/06/01;126(7)doi:10.1053/j.gastro.2004.02.018

3. Knudsen AB, Rutter CM, Peterse EFP, et al. USPSTF Modeling Study: Colorectal Cancer Screening. JAMA. 2021/05/18;325(19)doi:10.1001/jama.2021.5746

4. Kronborg O, Jørgensen OD, Fenger C, Rasmussen M. Randomized study of biennial screening with a faecal occult blood test: results after nine screening rounds. Scandinavian Journal of Gastroenterology. 2004–1–1;39(9)doi:10.1080/00365520410003182

5. Qaseem A, Harrod CS, Crandall CJ, et al. Screening for Colorectal Cancer in Asymptomatic Average-Risk Adults: A Guidance Statement From the American College of Physicians (Version 2). Annals of Internal Medicine. 2023–08–01;176(8)doi:10.7326/M23-0779

6. SEER*Explorer: An interactive website for SEER cancer statistics [Internet]. Surveillance Research Program, National Cancer Institute; 2026 Apr 22. [updated: 2026 Jun 22; cited 2026 Aug 19]. Available from: https://seer.cancer.gov/statistics-network/explorer/. Data source(s): SEER Incidence Data, November 2025 Submission (1975-2023), SEER 21 registries.

7. Shaukat A, Kahi CJ, Burke CA, Rabeneck L, Sauer BG, Rex DK. ACG Clinical Guidelines: Colorectal Cancer Screening 2021. Am J Gastroenterol. Mar 1 2021;116(3):458–479. doi:10.14309/ajg.0000000000001122

8. Shaukat A, Mongin SJ, Geisser MS, et al. Long-term mortality after screening for colorectal cancer. N Engl J Med. Sep 19 2013;369(12):1106–14. doi:10.1056/NEJMoa1300720

9. Sinicrope FA. Increasing Incidence of Early-Onset Colorectal Cancer. New England Journal of Medicine. 2022–04–21;386(16)doi:10.1056/NEJMra2200869

10. USPSTF, Davidson KW, Barry MJ, et al. USPSTF Recommendation: Screening for Colorectal Cancer. JAMA. 2021/05/18;325(19)doi:10.1001/jama.2021.6238

I will kick off my evidence deep dives for cancer screening with colon cancer. Before these evidence deep dives, my preconceived biases on cancer screening are that it is generally not high-yield and among them colon cancer is the highest yield. The recommendations that we generally follow are the USPSTF recommendations (USPSTF et al., 2021):

• Screening for everyone ages 45-75, and selective screening for ages 76-85.

• Recommended modalities are:

  • Fecal occult blood (FOBT; every year)

  • Fecal immunochemical test (FIT; every year)

  • Stool DNA (every 1-3 years)

  • Colonoscopy (every 10 years)

  • CT colonography (every 5 years)

  • Flexible sigmoidoscopy (flex sig; every 5 years).

The American College of Gastroenterology (ACG) recommends screening for ages 50-75 years old with moderate evidence and 45-49 with very low evidence. They recommend colonoscopy and FIT as the first-line screening tests followed by flex sig, stool DNA, CT and colon capsule (Shaukat et al., 2021).

The American College of Physicians (ACP) recommends screening for ages 50-75 years old and recommend not using stool DNA tests (cologuard) or CT colonography (Qaseem et al., 2023).

As I started reading the evidence for these guidelines, my question evolved from this vague question of whether I would recommend colon cancer screening to a more specific sequence of questions:

1. Is any colon cancer screening good?

2. How good?

3. At what age should we discuss colon cancer screening?

4. Are there specific modalities we should or should not recommend?

I decided to answer the first two questions in this post because there is RCT-level evidence to address these questions, and answer the last two questions in later posts with more observational evidence.

I will try to navigate you through a tour of the RCTs, followed by my conclusions. If you are in a super rush, I have a tldr section. If you are in a kind of rush, you could skip my “Evidence from RCTs of Screening” section. This section is mostly so I (and you!) can check my work when I come back to this question in 6 months and I have forgotten everything I read. If you have the time, I encourage you read my summaries of the RCTs, sit with them, and see if you agree with my conclusions.

tldr

1. Colon cancer screening is good.

2. At best, colon cancer screening has an NNT roughly in the 200s for all-cause mortality and 400s to prevent colon cancer.

Evidence from RCTs of Screening

It was hard for me to decide the best patient-oriented outcome. There are three relevant outcomes to consider: colon cancer incidence, colon cancer-specific mortality, and all-cause mortality. Each of the three outcomes has issues.

1. Colon cancer incidence: On one hand, it doesn’t matter if you get colon cancer, what matters is the consequences of the colon cancer. ie - whether it makes you sick, kills you, etc. On the other hand, as a definitely- not-oncologist, I do not remember seeing cases of colon cancer that we just watch and wait.

2. Colon cancer mortality: It seems good to avoid dying of colon cancer, but the issue is that there are competing causes of death, which is to say there’s a bunch of things that can kill you. The only cause-specific mortality that matters to me is suicide and homicide. Otherwise, a death is a death. From a population health/screening perspective, I think this is one of the most important pieces of data:

   

   This data is what people die of. Cancer is the number 2 cause, and accounts for about 20% of deaths. Colon cancer accounts for <10% of cancer deaths, so even if colon cancer screening somehow prevented half of colon cancer deaths, it would still prevent <1% of deaths.

   And another question about these causes of death, what do they even mean? What is a death due to cancer? If a patient with lung cancer dies of pneumonia, that seems related. But what if a patient with breast cancer in remission dies of a sudden pulmonary embolism? Or what if a patient with breast cancer successfully treated with radiation dies from lymphoma? I’ve filled out these cause of death forms, and in many situations it’s hard to identify a precise mechanism of death. This is another reason to favor all-cause mortality.

3. All-cause mortality: If there are adverse effects to cancer screening that are unexpected, they will be accounted for here. The downside is that the contribution of any one disease to all-cause mortality is small, and even with meta-analyses that pool multiple RCTs it is hard to find anything but the smallest effect.

All that said, I chose to focus on all-cause mortality and colon cancer incidence because if I can avoid dying that is good, and if I can avoid colon cancer treatments by clipping a pre-malignant polyp that also seems good. I don’t feel that the cause-specific mortality is as significant because if I can avoid dying of colon cancer, but I am just as likely to die from something else, then that seems meaningless. There were RCTs with these outcomes for colonoscopy, flex sig, and FOBT, but not for stool DNA tests or CT colonography. I separated them by modality:

Colonoscopy

COLONPREV: Castells et al., 2025

• Population: Patients from Spain ages 50-69 with no FIT testing for two years or colonoscopy or sigmoidoscopy for five years.

• Intervention: Invitation for colon cancer screening by colonoscopy (n = 26,332) vs FIT (n = 26,719). 31% of the colonoscopy group underwent screening and 39% for the FIT group.

• Follow-up: 13 years.

• Result: By ITT analysis there was no difference between FIT vs colonoscopy in colon cancer incidence or all-cause mortality. By PP analysis colonoscopy showed a decrease in colon cancer incidence (RR 0.67; NNT 238) and all-cause mortality (RR 0.83; NNT 111).

  • The intervention of these RCTs is technically the intervention to screen, so we are technically analyzing the number-needed-to-invite but for simplicity I will just use NNT.

• Adverse events: 0.3% of patients in each group had serious complications including 4 perforations in the colonoscopy group and 5 in the FIT group. The perforations in the FIT group were all from colonoscopies performed for positive FIT.

NordICC: Bretthauer et al., 2022; Kaminski et al., 2026

• Population: Patients from northern Europe ages 55-64 with no prior colon cancer screening.

• Intervention: Invitation for colon cancer screening by colonoscopy (n = 28,220) vs no invitation (n = 56,365). 42% of patients that were invited underwent colonoscopy.

• Follow-up: 13 years.

• Result: By ITT analysis there was a decrease in colon cancer incidence (HR 0.81; NNT 294) and no change in all-cause mortality. By PP analysis there was a greater decrease in colon cancer incidence (RR 0.55; NNT 125) and still no change in all-cause mortality.

• Adverse events: 0.13% of patients had clinically significant bleeding that required repeat endoscopy, and 0 patients had a perforation.

Flexible sigmoidoscopy (flex sig)

NORCCAP trial: Botteri et al., 2026

• Population: Patients from Norway ages 50-64 with no history of colon cancer.

• Intervention: Invitation for colon cancer screening by one-time flex sig + FIT (n = 10,278) vs flex sig and no FIT (n = 10,274) vs no invitation (n = 78,102). 63% of the flex sig group underwent screening. There was no effect of FIT so the flex sig with and without FIT were grouped and considered as the flex sig group.

• Follow-up: 22.5 years.

• Result: By ITT analysis there was a decrease in colon cancer incidence (HR 0.81; NNT 100) and all-cause mortality (HR 0.99; NNT 500).

• Adverse events: Not reported.

PLCO screening trial: Weissfeld et al., 2012; Miller et al., 2019

• Population: Patients from the US ages 55-74 with no history of prostate, lung, colorectal, or ovarian cancer.

• Intervention: Invitation for colon cancer screening by two-time flex sig at time 0 and 3-5 years (n = 77,443) vs no invitation (n = 77,444). 86% of the flex sig group underwent screening at least once and 50% had two flex sig’s.

• Follow-up: 16 years.

• Result: By ITT analysis there was a decrease in colon cancer incidence (RR 0.82; NNT 258).

• Adverse events: Not reported.

SCORE trial: Senore et al., 2022

• Population: Patients from Italy ages 55-64 with no history of colonoscopy or flex sig for three years.

• Intervention: Invitation for colon cancer screening by one-time flex sig (n = 17,148) vs no invitation (n = 17,144). 57% of the flex sig group underwent screening.

• Follow-up: 15 years for incidence and 18 years for mortality.

• Result: By ITT analysis there was a decrease in colon cancer incidence (RR 0.81; NNT 200) and no change in all-cause mortality. By PP analysis there was a decrease in colon cancer incidence (RR 0.67; NNT 114) and all-cause mortality (RR 0.93; 95% CI 0.87 - 1.00; NNT 32).

• Adverse events: Not reported.

UKFSST: Wooldrage et al., 2024

• Population: Patients from UK ages 55-64 with no history of colonoscopy or flex sig for three years.

• Intervention: Invitation for colon cancer screening by one-time flex sig (n = 57,099) vs no invitation (n = 112,927). 71% of the flex sig group underwent screening.

• Follow-up: 21 years.

• Result: By ITT analysis there was decrease in colon cancer incidence (HR 0.76; NNT 82) and no change in all-cause mortality. Results by PP analysis were similar.

• Adverse events: Not reported.

Fecal occult blood test (FOBT)

Burgundy Trial: Faivre et al., 2004

• Population: Patients from Burgundy, France ages 45-74.

• Intervention: Invitation for colon cancer screening by biennial FOBT (n = 45,642) vs control (n = 45,557). 69% of patients in the intervention arm completed at least one FOBT and 38% completed at least five.

• Follow-up: 11 years.

• Result: By ITT analysis there was no change in colon cancer incidence or all-cause mortality.

• Adverse events: There were no perforations or episodes of severe bleeding

Funen Trial: Kronborg et al., 2004

• Population: Patients from Funen, Denmark ages 45-74.

• Intervention: Invitation for colon cancer screening by biennial FOBT (n = 30,967) vs control (n = 30,966). 67% of patients in the intervention arm completed at least one FOBT.

• Follow-up: 17 years.

• Result: By ITT analysis there was no change in colon cancer incidence or all-cause mortality.

• Adverse events: When complications of colon cancer treatment were included with colon cancer mortality, the RR jumped to 0.89 and lost statistical significance.

Goteborg Trial: Lindholm et al., 2008

• Population: Patients from Goteborg, Sweden ages 60-64.

• Intervention: Invitation for colon cancer screening by FOBT (n = 34,144) vs control (n = 34,164). Patients were offered one or two rescreenings depending on the cohort. 70% of patients in the intervention arm completed at least one FOBT.

• Follow-up: 15.5 years.

• Result: By ITT analysis there was no change in colon cancer incidence or all-cause mortality.

• Adverse events: Adverse events were not reported.

Nottingham Trial: Hardcastle et al., 1996; Scholefield et al., 2012

• Population: Patients from the UK ages 45-74.

• Intervention: Invitation for colon cancer screening by biennial FOBT (n = 76,466) vs control (n = 76,384). 60% of patients in the intervention arm completed at least one FOBT and 38% completed all of the offered screening tests.

• Follow-up: 19.5 years.

• Result: By ITT analysis there was no change in colon cancer incidence or all-cause mortality.

• Adverse events: Adverse events were not reported.

Minnesota Colon Cancer Control Study: Mandel et al., 2000; Shaukat et al., 2013

• Population: Patients from America ages 50-80 with no prior colon cancer screening.

• Intervention: Invitation for colon cancer screening by annual FOBT (n = 15,570) vs biennial FOBT (n = 15,587) vs no invitation (n = 15,394). 90% of patients in the screening group completed at least 1 FOBT.

• Follow-up: 18 years for incidence and 30 years mortality.

• Result: By ITT analysis there was a decrease in colon cancer incidence for both annual FOBT (RR 0.80; NNT 162) and biennial (RR 0.83; NNT 198). There was no change in all-cause mortality. Results by PP analysis were similar.

• Adverse events: Of 12,246 colonoscopies, 4 patients had a perforation and 3 required surgery for severe bleeding.

Meta-analysis confirmed these findings (Kaneko et al., 2026). By ITT analysis there was a decrease in colon cancer incidence for both flex sig (RR 0.78; NNT 297) and colonoscopy (0.83; NNT 382), but not FOBT. Flex sig was found to have a borderline improvement for all-cause mortality (RR 0.97; 95% CI 0.95-1.00; NNT 240) but not colonoscopy or FOBT. Another meta-analysis found that flex sig increased lifespan by 110 days (Bretthauer et al., 2023).

Conclusions

Okay, that was a lot of evidence. Let’s start with the guidelines.

Comparing the three guidelines, the biggest difference is that the ACG and USPSTF guidelines draw a lot of inferences from modeling and observational studies, whereas the ACP guidelines are more RCT-literalist. I tend to agree with the ACP’s approach and let me address why.

There is not a doubt in my mind that people who get colon cancer screening are healthier than patients that do not. The question is whether colon cancer screening is what makes them healthier. When I think of my patient panel and I think about my patients that regularly undergo colon cancer screening, and I think about my patients that have never done colon cancer screening, I imagine two kinds of patients. One patient is not on Medicaid, white, and male. The other patient is on Medicaid and not white. This is why screening studies have to be RCT. There is too much that differentiates the patient that gets screening from the patient that does not get screening.

That said, I do not necessarily agree with the strict RCT-literalist approach either. Ultimately my goal for the patient in front of me is to predict their true benefit from colon cancer screening. The RCTs do a great job of answering the question, “Does the invitation to x screening test improve outcomes,” but this is a population level question that policy makers should be asking. This is slightly different than the scenario I am navigating with patients because the counseling I want to provide is how much they will benefit from actually doing the test, which is really one step past the question asked in the RCTs.

The most literal interpretation of the RCT data by ITT analysis would say that only flex sig has shown an improvement for all-cause mortality with an NNT around 200. By broadening our outcome of interest to colon cancer incidence there is a benefit with either flex sig or colonoscopy with an NNT around 300. The reason I don’t exactly agree with this strict interpretation is the screening compliance rates. Both the FOBT and flex sig trials were around 60% compliance, and the colonoscopy trials were in the 30-40% range. The lower the compliance, the more any true effect from colon cancer screening would be watered down.

The PP analysis addresses this concern by comparing patients who completed the screening with the control group. The downside here is the healthy-screenee bias that RCTs are meant to eliminate.

This is a very un-evidence-based-medicine thing to say, but I think the true biological benefit is probably somewhere inbetween the ITT and PP analyses, and I think it is reasonable to think of the ITT analysis as a floor of the benefit for the patient in front of you. There are additional time points of NordICC and other pending RCTs of colon cancer screening that may provide additional insight into this question but in the meantime my general impression is that colon cancer screening both prevents colon cancer and also death from all-causes. I recommend this, but with the large NNTs I do not recommend them strongly.

References

1. Botteri E, Holme O, Loberg M, et al. Twenty-Three-Year Benefits of Sigmoidoscopy Screening for Colorectal Cancer: A Randomized Trial. Ann Intern Med. Jun 2026;179(6):804–811. doi:10.7326/ANNALS-25-05456

2. Bretthauer M, Løberg M, Wieszczy P, et al. Effect of Colonoscopy Screening on Risks of Colorectal Cancer and Related Death. New England Journal of Medicine. 2022–10–27;387(17)doi:10.1056/NEJMoa2208375

3. Bretthauer M, Wieszczy P, Løberg M, et al. Estimated Lifetime Gained With Cancer Screening Tests. JAMA Internal Medicine. 2023/11/01;183(11)doi:10.1001/jamainternmed.2023.3798

4. Castells A, Quintero E, Bujanda L, et al. Effect of invitation to colonoscopy versus faecal immunochemical test screening on colorectal cancer mortality (COLONPREV): a pragmatic, randomised, controlled, non-inferiority trial. Lancet. Apr 12 2025;405(10486):1231–1239. doi:10.1016/S0140-6736(25)00145-X

5. Faivre J, Dancourt V, Lejeune C, et al. Reduction in colorectal cancer mortality by fecal occult blood screening in a French controlled study. Gastroenterology. 2004/06/01;126(7)doi:10.1053/j.gastro.2004.02.018

6. Hardcastle JD, Chamberlain JO, Robinson MH, et al. Randomised controlled trial of faecal-occult-blood screening for colorectal cancer. The Lancet. 1996/11/30;348(9040)doi:10.1016/S0140-6736(96)03386-7

7. Kaminski MF, Kalager M, Løberg M, et al. Long-term effects of colonoscopy screening on colorectal cancer incidence and mortality: a multicountry, population-based randomised controlled trial. The Lancet. 2026/05/09;407(10541)doi:10.1016/S0140-6736(26)00508-8

8. Kaneko M, Sakuraba A. Long-term effectiveness of endoscopic and stool-based colorectal cancer screening strategies: a systematic review and network meta-analysis. Ther Adv Gastroenterol. 2026;19:17562848261454186. doi:10.1177/17562848261454186

9. Knudsen AB, Rutter CM, Peterse EFP, et al. USPSTF Modeling Study: Colorectal Cancer Screening. JAMA. 2021/05/18;325(19)doi:10.1001/jama.2021.5746

10. Kronborg O, Jørgensen OD, Fenger C, Rasmussen M. Randomized study of biennial screening with a faecal occult blood test: results after nine screening rounds. Scandinavian Journal of Gastroenterology. 2004–1–1;39(9)doi:10.1080/00365520410003182

11. Lindholm E, Brevinge H, Haglind E. Survival benefit in a randomized clinical trial of faecal occult blood screening for colorectal cancer. British Journal of Surgery. 2008/07/11;95(8)doi:10.1002/bjs.6136

12. Mandel JS, Church TR, Bond JH, et al. The Effect of Fecal Occult-Blood Screening on the Incidence of Colorectal Cancer. New England Journal of Medicine. 2000–11–30;343(22)doi:10.1056/NEJM200011303432203

13. Miller EA, Pinsky PF, Schoen RE, Prorok PC, Church TR. Effect of flexible sigmoidoscopy screening on colorectal cancer incidence and mortality: long-term follow-up of the randomised US PLCO cancer screening trial. Lancet Gastroenterol Hepatol. Feb 2019;4(2):101–110. doi:10.1016/S2468-1253(18)30358-3

14. Qaseem A, Harrod CS, Crandall CJ, et al. Screening for Colorectal Cancer in Asymptomatic Average-Risk Adults: A Guidance Statement From the American College of Physicians (Version 2). Annals of Internal Medicine. 2023–08–01;176(8)doi:10.7326/M23-0779

15. Scholefield JH, Moss SM, Mangham CM, Whynes DK, Hardcastle JD. Nottingham trial of faecal occult blood testing for colorectal cancer: a 20-year follow-up. Gut. 2012–07–01;61(7)doi:10.1136/gutjnl-2011-300774

16. Senore C, Riggi E, Armaroli P, et al. Long-Term Follow-up of the Italian Flexible Sigmoidoscopy Screening Trial. Ann Intern Med. Jan 2022;175(1):36–45. doi:10.7326/M21-0977

17. Shaukat A, Kahi CJ, Burke CA, Rabeneck L, Sauer BG, Rex DK. ACG Clinical Guidelines: Colorectal Cancer Screening 2021. Am J Gastroenterol. Mar 1 2021;116(3):458–479. doi:10.14309/ajg.0000000000001122

18. Shaukat A, Mongin SJ, Geisser MS, et al. Long-Term Mortality after Screening for Colorectal Cancer. New England Journal of Medicine. 2013–09–19;369(12)doi:10.1056/NEJMoa1300720

19. USPSTF, Davidson KW, Barry MJ, et al. USPSTF Recommendation: Screening for Colorectal Cancer. JAMA. 2021/05/18;325(19)doi:10.1001/jama.2021.6238

20. Weissfeld JL, Schoen RE, Pinsky PF, et al. Flexible Sigmoidoscopy in the Randomized Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial: Added Yield from a Second Screening Examination. JNCI Journal of the National Cancer Institute. 2012 Jan 31;104(4)doi:10.1093/jnci/djr549

21. Wooldrage K, Robbins EC, Duffy SW, Cross AJ. Long-term effects of once-only flexible sigmoidoscopy screening on colorectal cancer incidence and mortality: 21-year follow-up of the UK Flexible Sigmoidoscopy Screening randomised controlled trial. Lancet Gastroenterol Hepatol. Sep 2024;9(9):811–824. doi:10.1016/S2468-1253(24)00190-0

Okay so the first thing to know about the meningococcal vaccine is that there are three types recommended by the CDC. One includes polysaccharide antigens for four serogroups of Neisseria meningitidis (serogroups A, C, W, and Y), another for one serogroup (B), and then a 5-in-1 (ABCWY). MenACWY comes in two different forms, either the CRM detoxified diphtheria toxin or TT tetanus toxoid. MenB also comes as two different forms, either the 4 component version (4CMenB) and a factor H binding protein version (MenB-FHbp).

These vaccines are to protect against invasive meningococcal disease which primarily presents as meningitis. Here are the most recent US meningococcal disease rates by serogroup:

The current ACIP recommendations are for MenACWY at 11 and 16 years, and MenB by shared decision-making at 16 (Mbaeyi et al., 2022). I will specifically evaluate studies in the routine risk population, and not high-risk patients that are recommended for early vaccination including asplenia, HIV, complement deficiency, sickle cell.

tldr

I recommend both meningococcal vaccines because the disease they are preventing is very serious and the adverse effects are not nearly as serious, but I don’t think either has strong enough evidence to reach the everyone-should-do-this level.

Efficacy from RCTs

I reviewed RCTs for any meningococcal vaccine in participants that were 11+ years old and used clinical outcomes like meningococcal disease. The majority of the studies were old and from the 70s when the disease incidence was about 1000-fold higher than today. The vaccines today are different from the one’s tested in the 70s, but they still use the same polysaccharide antigen. The newer studies were mostly immunosurrogacy studies and I did not include these.

Artenstein et al., 1970

• Population: Young healthy men in the US army.

• Treatment: Polysaccharide vaccine to serogroup C meningococci (n = 13,763) vs no vaccine (n = 54,309).

• Outcome: 38 unvaccinated participants got meningococcal disease from serogroup C compared to 1 vaccinated participant. This was a reduction of 87% and NNT 161 over 8 weeks (p <0.01).

• Control rate of disease: 0.7 cases/1000 over 8 weeks.

• Safety: No systemic adverse effects were reported.

Erwa et al., 1973

• Population: Participants from Sudan. 27% were in the 10-20 years old range.

• Treatment: Polysaccharide vaccine to serogroup A meningococci (n = 10,891) vs tetanus toxoid (n = 10,749).

• Outcome: 0 cases of serogroup A meningococcal meningitis in the meningococcal vaccine group and 7 cases in the tetanus group (p <0.05).

• Control rate of disease: 0.6 cases/1000 over 16 weeks.

• Safety: No systemic adverse effects were reported.

Wahdan et al., 1973

• Population: Schoolchildren 6-15 years old in Egypt.

• Treatment: Polysaccharide vaccine to serogroup A meningococci (n = 62,295) vs tetanus toxoid (n = 62,054).

• Outcome: 0 cases of serogroup A meningococcal meningitis in the meningococcal vaccine group and 8 cases in the tetanus group (p = 0.004).

• Control rate of disease: 0.1 cases/1000 over 8 months.

• Safety: No systemic adverse effects were reported.

Makela et al., 1975

• Population: Healthy army recruits in Finland.

• Treatment: Polysaccharide vaccine to serogroup A meningococci (n = 16,458) vs tetanus toxoid (n = 20,748).

• Outcome: 1 case of serogroup A meningococcal meningitis in the meningococcal vaccine group and 11 cases in the unvaccinated group (p = 0.004).

• Control rate of disease: 0.7 cases/1000 over 9 months.

• Safety: No systemic adverse effects were reported.

Wahdan et al., 1977

• Population: Schoolchildren 6-15 years old in Egypt.

• Treatment: Polysaccharide vaccine to serogroup A meningococci (n = 88,263) vs tetanus toxoid (n = 88,383).

• Outcome: 4 cases of serogroup A meningococcal meningitis in the meningococcal vaccine group and 12 cases in the tetanus group (p <0.01).

• Control rate of disease: 0.1 cases/1000 over 2 years.

• Safety: No systemic adverse effects were reported.

Greenwood et al., 1980

• Population: Nigeria.

• Treatment: Polysaccharide vaccine to serogroup A and C meningococci was given to 8,600 participants in 9 villages with >2 reported cases of meningococcal disease (n = 10,600) vs no vaccine in 7 control villages (n = 11,000).

• Outcome: 10 cases of meningococcal disease in the villages with meningococcal vaccine and 38 cases in the control group (p <0.001).

• Control rate of disease: 3 cases/1000 over 3 months.

• Safety: No systemic adverse effects were reported.

Blakebrough et al., 1983

• Population: Participants in Nigeria age 11-20 years old.

• Treatment: Polysaccharide vaccine to group C meningococci (n = 438) vs no vaccine (n = 874).

• Outcome: No difference in meningococcal carriage (11% in vaccine vs 12% in controls) or disease rates (0.5% vs 0.2%; p >0.1).

• Control rate of disease: 5 cases/1000 over 4 months.

• Safety: No systemic adverse effects were reported.

Bjune et al., 1991

• Population: Participants in Norway age 13-21 years old.

• Treatment: Conjugated polysaccharide vaccine to serogroup B meningococci conjugated with OM x2 over 6 weeks (n = 88,000) vs aluminum hydroxide control (n = 83,000).

• Outcome: 12 cases of meningococcal disease in the vaccine group and 24 cases in the control group (p = 0.012).

• Control rate of disease: 0.2 cases/1000 over 1 year.

• Safety: No systemic adverse effects were reported.

Read et al., 2014

• Population: Healthy 18-24 year old children in England.

• Treatment: 4CMenB x2, MenACWY-CRM, or control (Japanese encephalitis virus vaccine).

• Follow-up: 1 year

• Outcome: No difference in the primary outcome of pharyngeal carriage of disease-causing N meningitidis at one month post-vaccine. They did find differences in some of the secondary outcomes including MenACWY-CRM decreasing carriage of disease-causing N meningitidis at 2 months or later (OR 0.7; 9% in the vaccine group vs 11% for the control), 4CMenB decreasing carriage rates of some but not all serogroups of N meningitidis.

• Control rate of disease: There were no cases of meningococcal disease.

• Safety: Myalgias were increased in 4CMenB (75%) but not MenACWY-CRM (42%), as compared to the Japanese encephalitis vaccine (50%).

Marshall et al., 2020

• Population: Healthy 15-18 year old children in Australia.

• Treatment: 4CMenB vaccine x2 doses 2 months apart (n = 12,746) vs the same vaccine one year later (n = 11,523).

• Outcome: There was no statistically significant difference in PCR detection of disease-causing N meningitidis from oropharyngeal swabs.

• Control rate of disease: There were no cases of meningococcal B disease.

• Safety: Adverse events were not reported.

There was a Cochrane Review for the serogroup A polysaccharide vaccine which found a vaccine efficacy of 95% (Patel and Lee, 2005).

Efficacy from observational studies

Part of the problem with meningococcal disease is that it’s become so rare that it’s difficult to run an adequately powered RCT with clinical outcomes. Because of this, we resort to observational studies to identify a clinical effect.

Cohn et al., 2017

• Population: Meningococcal disease cases that were reported from 2006-2013 in 11-19 year old participants in the United States identified through one of two surveillance mechanisms (Active Bacterial Core surveillance and MeningNet).

• Analytic method: Case-control comparing cases with meningococcal disease (n = 181) with controls which were either friends that were referred by the case participant or children at the same school (n = 199).

• Outcome: This study found 20% of cases were vaccinated with MenACWY-D and 44% of controls for a vaccine efficacy of 69%. The efficacy dropped from 79 to 69 to 61% for participants vaccinated within 1 year, 1-3 years, and 3-8 years, respectively.

Mbaeyi et al., 2020

• Population: All reported meningococcal disease cases in the United States from 2000-2017 (n = 19,207).

• Intervention: ACIP recommended that all 11-12 year olds get MenACWY in 2005 and they recommended a booster dose at 16 years in 2010.

• Analytic method: 3 different time periods were analyzed including pre-vaccine (2000-2005), primary dose recommendation (2006-2010), booster dose recommendation (2010-2017).

• Outcome: This study found that meningococcal disease was decreasing before the vaccine from 2000-2005 (0.61 cases/100,000), continued to decrease in the primary dose years from 2006-2010 (0.35 cases/100,000), and again continued to decrease thereafter (0.15 cases/100,000). Rates for serogroup B, which should not be affected by MenACWY, also decreased throughout this time period. It’s hard for me to make sense of the statistics because there are so many outcomes and no clear patterns.

McMillan et al., 2021

• Population: 16-19 year olds in South Australia.

• Intervention: This was an observational follow-up of the Marshall 2020 RCT described above, which started in 2017, and used 4CMenB.

• Analytic method: Observational.

• Outcome: This study found that serogroup B meningococcal disease was increasing before the vaccine from 1 case in 2004 to 10 in 2015. This then decreased to 7, 5, and 1 case in 2016, 2017, and 2018 respectively. They found a 71% reduction in 2017 and 2018 compared to the predicted rates. They did not find any effect on cases in 0-4 year olds.

Wang et al., 2022

• Population: 15-21 year olds in Australia.

• Intervention: In 2019, Australia recommended 4CMenB to everyone 15-21 years old (n = 53,356 received at least one dose).

• Analytic method: Comparing rates of serogroup B meningococcal diseaase in the 8 years pre-vaccine to post-vaccine.

• Outcome: This study found that meningococcal disease decreased from 4 per year pre-vaccine to 2 and then 0 in participants 15-18 years old (p = 0.078) and found a similar pattern from 4 per year pre-vaccine to 4 and then 0 in participants 19-21 years old (p = 0.51).

Safety from observational studies

With the RCT-level evidence for efficacy being old, and the observational-level evidence being borderline, the question really comes down to safety. For a vaccine with unclear benefits, it has to be robustly safe to recommend. There were initial VAERS reports for GBS, which were addressed in the following big safety studies.

Velentgas et al., 2012

• Population: 11-18 year olds from 5 insurance databases in the US (n = 9,578,688).

• Intervention: MenACWY-D (conjugated to diphtheria toxoid).

• Analytic method: Retrospective cohort looking for GBS in the 6 weeks after vaccination.

• Outcome: There were 0 cases of GBS. The upper bound of the 95% confidence interval was 2.1 cases per million vaccines.

Tseng et al., 2017

• Population: 11-21 year olds in the US (n = 48,899).

• Intervention: MenACWY-CRM.

• Analytic method: Self-controlled case series comparing each participant to themselves during at-risk and control time periods.

• Outcome: Of 26 pre-specified outcomes only 1 showed a statistically significant increase, which was Bell’s palsy (2.9-fold; n = 8). Of the 8 cases, 6 occurred when participants received a concomitant vaccine. All 8 cases resolved spontaneously. There were 0 cases of GBS.

Conclusions

In order to get a better risk/benefit picture, let’s start with safety. The most evidence-based adverse effect is Bell’s palsy that occurs in about 1 per 6,000 vaccines. This is a self-resolving adverse effect. There are not robust safety studies for either MenB or MenABCWY and it would be great if we could learn more about the safety of these vaccines, especially the pentavalent one which would be convenient for administration.

On the other hand, the RCTs show dramatic improvements in rates of meningococcal disease. These trials used an older version of the vaccine, but they contained the same antigen as modern vaccines. The observational studies are supportive that the modern vaccine continue to prevent meningococcal disease. Current disease incidence for meningococcal disease is around 1 in a million, but it has a case-fatality ratio of 8%, so 8% of patients that get meningococcal disease will die (Wang et al., 2019).

It’s not a slam dunk, but I’m sold on the meningococcal vaccine. For my personal risk calculus, the death of a young person is an incredibly dramatic event that I would do everything possible to prevent. Even the possibility of preventing the death of a young person by the meningococcal vaccine has a much greater value to me than a more certain possibility of preventing the death of an older person by use of statins. With no serious adverse effects, and a disease this severe, I think this is worthwhile.

References

1. Artenstein MS, Gold R, Zimmerly JG, Wyle FA, Schneider H, Harkins C. Prevention of meningococcal disease by group C polysaccharide vaccine. N Engl J Med. Feb 19 1970;282(8):417–20. doi:10.1056/NEJM197002192820803

2. Bjune G, Hoiby EA, Gronnesby JK, et al. Effect of outer membrane vesicle vaccine against group B meningococcal disease in Norway. Lancet. Nov 2 1991;338(8775):1093–6. doi:10.1016/0140-6736(91)91961-s

3. Blakebrough IS, Greenwood BM, Whittle HC, Bradley AK, Gilles HM. Failure of meningococcal vaccination to stop the transmission of meningococci in Nigerian schoolboys. Ann Trop Med Parasitol. Apr 1983;77(2):175–8. doi:10.1080/00034983.1983.11811694

4. Cohn AC, MacNeil JR, Harrison LH, et al. Effectiveness and Duration of Protection of One Dose of a Meningococcal Conjugate Vaccine. Pediatrics. Feb 2017;139(2)doi:10.1542/peds.2016-2193

5. Erwa HH, Haseeb MA, Idris AA, Lapeyssonnie L, Sanborn WR, Sippel JE. A serogroup A meningococcal polysaccharide vaccine: studies in the Sudan to combat cerebrospinal meningitis caused by Neisseria meningitidis group A. Bull World Health Organ. 1973;49(3):301–5.

6. Greenwood BM, Wali SS. Control of meningococcal infection in the African meningitis belt by selective vaccination. Lancet. Apr 5 1980;1(8171):729–32. doi:10.1016/s0140-6736(80)91230-1

7. Makela PH, Kayhty H, Weckstrom P, Sivonen A, Renkonen OV. Effect of group-A meningococcal vaccine in army recruits in Finland. Lancet. Nov 8 1975;2(7941):883–6. doi:10.1016/s0140-6736(75)92125-x

8. Marshall HS, McMillan M, Koehler AP, et al. Meningococcal B Vaccine and Meningococcal Carriage in Adolescents in Australia. N Engl J Med. Jan 23 2020;382(4):318–327. doi:10.1056/NEJMoa1900236

9. Mbaeyi SA, Bozio CH, Duffy J, et al. Meningococcal Vaccination: Recommendations of the Advisory Committee on Immunization Practices, United States, 2020. MMWR Recomm Rep. Sep 25 2020;69(9):1–41. doi:10.15585/mmwr.rr6909a1

10. Mbaeyi S, Pondo T, Blain A, et al. Incidence of Meningococcal Disease Before and After Implementation of Quadrivalent Meningococcal Conjugate Vaccine in the United States. JAMA Pediatr. Sep 1 2020;174(9):843–851. doi:10.1001/jamapediatrics.2020.1990

11. McMillan M, Chandrakumar A, Wang HLR, et al. Effectiveness of Meningococcal Vaccines at Reducing Invasive Meningococcal Disease and Pharyngeal Neisseria meningitidis Carriage: A Systematic Review and Meta-analysis. Clin Infect Dis. Aug 2 2021;73(3):e609–e619. doi:10.1093/cid/ciaa1733

12. McMillan M, Wang B, Koehler AP, Sullivan TR, Marshall HS. Impact of Meningococcal B Vaccine on Invasive Meningococcal Disease in Adolescents. Clin Infect Dis. Jul 1 2021;73(1):e233–e237. doi:10.1093/cid/ciaa1636

13. Patel M, Lee CK. Polysaccharide vaccines for preventing serogroup A meningococcal meningitis. Cochrane Database Syst Rev. Jan 25 2005;2005(1):CD001093. doi:10.1002/14651858.CD001093.pub2

14. Read RC, Baxter D, Chadwick DR, et al. Effect of a quadrivalent meningococcal ACWY glycoconjugate or a serogroup B meningococcal vaccine on meningococcal carriage: an observer-blind, phase 3 randomised clinical trial. Lancet. Dec 13 2014;384(9960):2123–31. doi:10.1016/S0140-6736(14)60842-4

15. Tseng HF, Sy LS, Ackerson BK, et al. Safety of Quadrivalent Meningococcal Conjugate Vaccine in 11- to 21-Year-Olds. Pediatrics. Jan 2017;139(1)doi:10.1542/peds.2016-2084

16. Velentgas P, Amato AA, Bohn RL, et al. Risk of Guillain–Barré syndrome after meningococcal conjugate vaccination. Pharmacoepidemiology and Drug Safety. 2012/12/01;21(12)doi:10.1002/pds.3321

17. Wahdan MH, Rizk F, el-Akkad AM, et al. A controlled field trial of a serogroup A meningococcal polysaccharide vaccine. Bull World Health Organ. Jun 1973;48(6):667–73.

18. Wahdan MH, Sallam SA, Hassan MN, et al. A second controlled field trial of a serogroup A meningococcal polysaccharide vaccine in Alexandria. Bull World Health Organ. 1977;55(6):645–51.

19. Wang B, Santoreneos R, Giles L, Haji Ali Afzali H, Marshall H. Case fatality rates of invasive meningococcal disease by serogroup and age: A systematic review and meta-analysis. Vaccine. May 9 2019;37(21):2768–2782. doi:10.1016/j.vaccine.2019.04.020

20. Wang B, Giles L, Andraweera P, et al. Effectiveness and impact of the 4CMenB vaccine against invasive serogroup B meningococcal disease and gonorrhoea in an infant, child, and adolescent programme: an observational cohort and case-control study. Lancet Infect Dis. Jul 2022;22(7):1011–1020. doi:10.1016/S1473-3099(21)00754-4

My short answer was that there are lots of people talking about coronary artery calcium and new ways to evaluate cardiovascular risk. Different doctors are saying different things about these tests because we don’t have strong evidence to recommend one thing or another and we are mostly speculating. I don’t feel it adds much to the blood tests we have already done.

Here is my long answer.

The coronary artery calcium (CAC) score is a measure of calcified plaque in the coronary arteries, and it is obtained from a specific type of chest CT. It can be used as a measure of cardiovascular risk.

The ACC AHA 2026 Dyslipidemia guideline recommends using CAC scoring for the patient with a borderline 3-5% PREVENT score as a way to defer statin treatment (Blumenthal et al., 2026). If CAC >0 they recommend treatment and if CAC = 0 they recommend repeat CAC scoring in a few years:

After reading the guideline, I started my CAC deep dive by reading the UpToDate page, and next moved on to several insightful commentaries including The Skeptical Cardiologist's, Dr. Anish Koka’s and Dr. John Mandrola’s. These are three thoughtful cardiologists that give a balanced view on CAC, and I would highly recommend their commentaries.

I next moved on to the primary literature, and what I mostly found was a sea of observational studies with very limited RCT-level evidence. This is where I had to pause and think about my usual critical appraisal approach. My instinct is to search for RCTs because we consider this the best way to evaluate treatments, but CAC is a diagnostic test and these require a different approach. It’s a tool in the toolbox of evaluating cardiovascular risk, and we need to know how well it predicts cardiovascular risk, but ultimately it’s only useful if it changes our treatment decisions.

Let’s go through what I learned.

Observational Studies

There are essentially unlimited observational studies about CAC, so I want to keep my focus on the primary care wheelhouse of primary prevention. This is patients with no history of cardiovascular disease, and no symptoms. Healthy people.

South Bay Heart Watch: Greenland et al., 2004

• Population: Patients >45 years old that were asymptomatic, had no history of MI or angina, and had an 8-year Framingham risk score >10%. This analysis focused on patients without diabetes (n = 1,029). The average age was 65 and 90% were male. I would consider this a medium risk population.

• Follow-up: 6.3 years.

• Outcome: Compared to patients with CAC 0, patients with CAC 1-100 did not have a difference in cardiovascular events (MI or death from coronary heart disease; p = 0.26), CAC 101-300 was a 2-fold increase (p = 0.05), and CAC >300 was 4-fold (p < 0.001). They used this information to develop models for risk prediction with and without CAC and found that inclusion of CAC improved the AUC of the risk prediction model from 0.63 to 0.68 (p < 0.001). The AUC is a measure of discrimination, or the ability of a risk score to correctly identify a patient who will go on to have a event as high risk. A score of 1 means the score perfectly predicts every outcome, and a score of 0.5 means the score is guessing randomly. So the CAC improves the risk prediction model, but is there clinical meaning to an improved AUC of 0.05?

• They had this figure that provides an interesting visual:

• Starting from the cluster of four on the left, we have the patients with the lowest Framingham Risk Score group (0-9%). Within this group the darker bars represent higher CAC, and it looks like CAC 0, 1-100, and 101-300 are quite similar in their coronary outcomes. It looks like there is a trend towards increased coronary events for CAC >300, but this is not statistically significant. Looking at the rest of the clusters of four, it doesn’t look like there is a ton of ability for the CAC 1-300 to distinguish risk from CAC 0.

• Sources of bias: There were very few patients in the 0-9% low-risk group (n = 98) because the study only recruited medium- or higher-risk patients and the patients in that low-risk group were initially medium-risk but then found to be low-risk based on the baseline bloodwork that was obtained at 30 months. It is possible that patients who were low-risk to start with are biologically different than patients who were at one point medium- or higher-risk that then became low-risk. Also, the study population was mostly white (84%) and male (90%).

Budoff et al., 2007

• Study type: Prospective cohort study evaluating CAC as a predictor of all-cause mortality.

• Population: All patients that were referred for CAC (n = 25,253). The average age was 56, 54% were male, just 15% had hypertension and 4% had diabetes. I would consider this a low-risk population.

• Follow-up: 5.8 years.

• Outcome: 0.6% of patients with CAC 0 died. Compared with CAC 0, CAC 1-10 did not affect mortality (p = 0.29). CAC 11-100 had a 3.6-fold increase, CAC 100-399 was 3.8-fold, 400-699 was 5.7, 700-999 was 3.3, and >1000 was 9.3 (all p < 0.0001).

• Sources of bias: It’s hard to know what sort of patients were being referred for a CAC 20 years ago but it was probably a unique population as there were no guidelines recommending this.

MESA: Detrano et al., 2008; Polonsky et al., 2010

• Study type: Prospective cohort study evaluating CAC as a predictor of major coronary events (MI or death from coronary heart disease).

• Population: 45-84 years old patients across America that had no history of cardiovascular disease. The average age was 62, 47% were male, and included white (38%), black (27%), Hispanic (21%), and Chinese patients (11%). 11% had diabetes, 44% had hypertension, and 42% have a family history of cardiovascular disease. My estimated PREVENT score for the cohort was 7%. N = 6,722. I would consider this a medium risk population.

• Follow-up: At various points from 3-11 years.

• Outcome: At 3.8 years, patients with a CAC 1-100 had a 3-fold increase in major coronary events over CAC 0, and CAC >100 was 7-fold (both p < 0.001). They developed models for risk prediction with and without CAC and found that inclusion of CAC improved the AUC of the risk prediction model from 0.79 to 0.83 (p < 0.001). Similar to the South Bay Heart Watch, there was an improvement, but does it matter?

• Sources of bias: This is about as good as it gets for an observational study. The population was diverse, the design was prospective to limit recall bias, and it had impressive rates of follow-up.

Blaha et al., 2009

• Study type: Prospective cohort study evaluating the risk of CAC 0 as compared to 1-10.

• Population: Patients in America with no history of cardiovascular disease or symptoms of cardiovascular disease that were referred for CAC testing (n = 44,052). Average age 54 years old, 54% were male, and 34% had hypertension. I would consider this a low-risk population.

• Follow-up: 5.6 years.

• Outcome: 0.5% of patients with CAC 0 died. When compared to CAC 0, CAC 1-10 had a 1.9-fold increase in all-cause mortality. CAC >10 had a 5-fold increase, CAC 101-400 was 5.5-fold, and CAC >400 was 9.6.

• Sources of bias: Similar to Budoff et al 2007, this is a unique population of patients that were recommended for CAC testing despite any guidelines recommending this.

Heinz Nixdorf Recall Study; Erbel et al., 2010

• Study type: Prospective cohort study evaluating the ability of CAC to change a patient’s risk score.

• Population: 45-75 years old patients from Germany that had no history of cardiovascular disease (n = 4,129).

• Follow-up: 5 years.

• Outcome: When compared to CAC 0, CAC 1-99 had similar hard events (nonfatal MI and coronary deaths), 100-399 had an RR of 2.8, and >400 of 6.4. The AUC for the ROC curves improved from 0.71 to 0.76 with CAC (p = 0.004). Again similar to South Bay Heart Watch and MESA, there was a minimal improvement, but does it matter.

MESA; Nasir et al., 2015

• Study type: Prospective cohort study evaluating the effect of CAC on ASCVD outcomes in populations stratified by predicted ASCVD risk score.

• Population: Same as MESA population above.

• Follow-up: 10.3 years.

• Outcome: The MESA population was divided into three subgroups PCE <5%, 5-7.5%, and >7.5%. The results were in table form, so I put them into a graph:

  

• In broad strokes, this data differs from Greenland et al. 2004 in that the CAC 1-100 differentiates intermediate risk patients from CAC 0. Otherwise we have the same general trends, high CAC increases risk in low-risk patients and low CAC decreases risk in high-risk patients, but these patients are still at risk.

CARDIA; Carr et al., 2017

• Study type: Prospective cohort study evaluating the ability of CAC at mid-life (~40 years old) to predict early (<60 years old) cardiovascular events.

• Population: Patients 18-30 years old that underwent CAC scoring at 15, 20, and 25 years (n = 3,980). Patients at the start of study (year 15) were on average 40 years old, BP 113/74, and 45% male. These were young healthy patients.

• Follow-up: 12.5 years starting from the first CAC score, which was at year 15.

• Outcome: 0.8 per 100 patients with CAC 0 had a cardiovascular event (the outcomes were confusing, but I interpreted them as unstable angina, fatal and nonfatal MI). There was no difference for CAC 1-19 (p = 0.07), CAC 20-99 had a 5-fold increase, and CAC >100 a 6-fold increase (both p < 0.001).

Based on these studies I have a couple thoughts:

1. The heterogeneity in the studies makes it hard to draw conclusions. They used various outcomes including ASCVD (nonfatal and fatal MI, nonfatal and fatal stroke), CHD (nonfatal and fatal MI), and all-cause mortality. They also used a range of CAC cutoffs from CAC 1-10, CAC 1-100 together, and even all CAC >0 together.

2. There is no doubt that more CAC is associated with more all-cause mortality and more coronary heart disease. However, it’s not exactly clear where to draw the “more CAC” line. CAC >100 is pretty consistently associated with an increased risk. CAC 1-100 is muddy. This makes it especially hard to identify if CAC = 0 is the only safe CAC, or if we could be safe up to say CAC 30.

3. Looking specifically at low-risk populations, it is similarly clear that patients with high CAC are at higher risk of cardiovascular events (Greenland et al., 2004; Nasir et al., 2015). Again, it’s not exactly clear where to draw the “high CAC” line. Greenland et al. found the effect wasn’t noticeable until CAC >300, whereas Nasir et al. found an effect even by 1-100. This makes it hard to use CAC decide which low-risk patients should still be treated with statins.

4. Looking specifically at high-risk populations, even the CAC = 0 patients are at substantial cardiovascular risk (Greenland et al., 2004; Nasir et al., 2015). This makes it hard to use CAC to say high-risk patients do not need to be on statins.

RCTs

There were no RCTs purely testing +/- CAC, but I will review two RCTs of cardiovascular screening that included CAC. I read a third trial (CAUGHT-CAD) that was peripherally related to CAC but I chose not to review it because it seemed more like a trial about statins (Nerlekar et al., 2025).

DANCAVAS; Lindholt et al., 2022

• Population: Men 65-74 years old in Denmark. About half the population had hypertension.

• Intervention: Invitation to screening for cardiovascular disease by CAC score, afib, aortic and iliac aneurysms, A1c, and lipid panel (n = 16,736; 62% completed the screening). The control patients were not invited (n = 29,790).

• Follow-up: 5.6 years.

• Outcome: No difference in all-cause mortality by intention-to-treat analysis (ITT; p = 0.06). ITT pools all patients that were intended to be treated by the intervention, so in this case all 16,736 patients were analyzed even though only 62% completed the screening. This is in contrast to per-protocol analysis, which only looks at the 62% that completed the screening. I was curious to know the per-protocol results, but these were not reported. I would expect the per-protocol population that agreed to screening to be a biased healthier subpopulation, and if there was not a notable difference in all-cause mortality with the per-protocol analysis I would be even more skeptical of this intervention. They did find an 11% reduction for all-cause mortality in patients 65-69 years old (6.0 vs 6.7% in control).

• Safety: There was a trend towards increased severe bleeding with the invitation to screen (6.8 vs 6.3%, p = 0.06).

DANCAVAS 2; Lindholt et al., 2025

• Population: Men 60-64 years old in Denmark. 40% of patients had hypertension.

• Intervention: Invitation to screening for cardiovascular disease by CAC score, afib, aortic and iliac aneurysms, A1c, and lipid panel (n = 5,946; 62% completed the screening). The control patients were not invited (n = 25,322).

• Follow-up: 7.0 years.

• Outcome: No difference in all-cause mortality by ITT analysis (p = 0.16). In this case they did do the per-protocol analysis and found a 51% reduction in all-cause mortality.

• Safety: There was more hospitalization for bleeding with the invitation to screen (5.9 vs 5.0%, p = 0.007).

These trials need to be interpreted with caution. The conservative interpretation of these trials is that cardiovascular screening does not benefit all-cause mortality, and therefore should not be offered. However, there is a subtle but important difference between screening and being invited for screening.

Being invited for screening is a policy level question. Do we want doctors to spend precious primary care time discussing cardiovascular screening. If lots of patients will say no, like the 38% of patients in DANCAVAS, it may be a waste of time to have doctors talk about it. Even if you have the best intervention in the world, if no patients will do it, there is no point to talking about it.

At the individual clinician-patient level the pertinent question is whether the patient will benefit from the screening. At this point we are past the decision to invite, and we are in a slightly different position from DANCAVAS. The per-protocol population that underwent screening is a biased healthier subpopulation that is less exposed to the harms of screening. DANCAVAS 2 found that screening causes increased hospitalizations for bleeding. It’s possible that if more patients had done the screening, there would be more harm that would balance out any gains from the screening. Ultimately, we need an RCT with higher compliance to truly answer this question.

Conclusions

This is a complicated space. In medicine, complicated spaces are usually cluttered with opinions and inconclusive science.

For CAC, the observational studies are convincing that it’s associated with cardiovascular risk. It’s not clear that the improvement to risk prediction by adding CAC is clinically significant. If the effect size was giant, it would be easier to justify accepting CAC as an improved risk predictor, but with this effect size we need the RCTs.

The RCTs we do have, the DANCAVAS’s, found that CAC has no effect on all-cause mortality. These are imperfect studies. Only two-thirds of patients actually did the screening, and the intervention was a cardiovascular screening package that included CAC.

The burden of proof lies with the person proposing the change, and CAC needs more proof.

References

1. Blaha M, Budoff MJ, Shaw LJ, et al. Absence of coronary artery calcification and all-cause mortality. JACC Cardiovasc Imaging. Jun 2009;2(6):692–700. doi:10.1016/j.jcmg.2009.03.009

2. Blumenthal RS, Morris PB, Gaudino M, et al. 2026 ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the Management of Dyslipidemia: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol. Mar 13 2026;doi:10.1016/j.jacc.2025.11.016

3. Budoff MJ, Shaw LJ, Liu ST, et al. Long-term prognosis associated with coronary calcification: observations from a registry of 25,253 patients. J Am Coll Cardiol. May 8 2007;49(18):1860–70. doi:10.1016/j.jacc.2006.10.079

4. Carr JJ, Jacobs DR, Jr., Terry JG, et al. Association of Coronary Artery Calcium in Adults Aged 32 to 46 Years With Incident Coronary Heart Disease and Death. JAMA Cardiol. Apr 1 2017;2(4):391–399. doi:10.1001/jamacardio.2016.5493

5. Detrano R, Guerci AD, Carr JJ, et al. Coronary calcium as a predictor of coronary events in four racial or ethnic groups. N Engl J Med. Mar 27 2008;358(13):1336–45. doi:10.1056/NEJMoa072100

6. Erbel R, Mohlenkamp S, Moebus S, et al. Coronary risk stratification, discrimination, and reclassification improvement based on quantification of subclinical coronary atherosclerosis: the Heinz Nixdorf Recall study. J Am Coll Cardiol. Oct 19 2010;56(17):1397–406. doi:10.1016/j.jacc.2010.06.030

7. Greenland P, LaBree L, Azen SP, Doherty TM, Detrano RC. Coronary artery calcium score combined with Framingham score for risk prediction in asymptomatic individuals. JAMA. Jan 14 2004;291(2):210–5. doi:10.1001/jama.291.2.210

8. Lindholt JS, Mejldal A, Rasmussen LM, et al. Outcomes of cardiovascular screening in men aged 60-64 years: the DANCAVAS II trial. Eur Heart J. Aug 30 2025;doi:10.1093/eurheartj/ehaf704

9. Lindholt JS, Sogaard R, Rasmussen LM, et al. Five-Year Outcomes of the Danish Cardiovascular Screening (DANCAVAS) Trial. N Engl J Med. Oct 13 2022;387(15):1385–1394. doi:10.1056/NEJMoa2208681

10. Nasir K, Bittencourt MS, Blaha MJ, et al. Implications of Coronary Artery Calcium Testing Among Statin Candidates According to American College of Cardiology/American Heart Association Cholesterol Management Guidelines: MESA (Multi-Ethnic Study of Atherosclerosis). J Am Coll Cardiol. Oct 13 2015;66(15):1657–68. doi:10.1016/j.jacc.2015.07.066

11. Nerlekar N, Vasanthakumar SA, Whitmore K, et al. Effects of Combining Coronary Calcium Score With Treatment on Plaque Progression in Familial Coronary Artery Disease: A Randomized Clinical Trial. JAMA. Apr 22 2025;333(16):1403–1412. doi:10.1001/jama.2025.0584

12. Polonsky TS, McClelland RL, Jorgensen NW, et al. Coronary artery calcium score and risk classification for coronary heart disease prediction. JAMA. Apr 28 2010;303(16):1610–6. doi:10.1001/jama.2010.461

I previously wrote about statins for primary prevention in low-risk patients without diabetes. Here I wanted to look at the evidence for patients with diabetes.

The ACC AHA 2026 Dyslipidemia guideline provides the following flowchart to decide which patients with diabetes should be on a statin (Blumenthal et al., 2026):

These recommendations align with the ADA guideline so there is good consensus that everyone >40 years of age and diabetes should be on a statin (ADA, 2026).

tldr

For a patient with diabetes over 40 years of age, I would strongly recommend a statin with guidance of NNT 20-50 for major cardiovascular event and 100 for death.

I would probably tell a patient there is about a 1 in 100 chance this will save your life, and 1 in 20 chance that this will prevent a heart attack. These numbers aren’t a perfect representation of the truth and there is the ever-looming question of how to extrapolate from these 1-4 year studies to a lifetime of taking a statin, but more important than describing these numbers I pitch statins as a crucial part of diabetes care. In my counseling I emphasize the connection between diabetes and heart disease. I check annual LDL levels. If they don’t want a statin, I routinely bring up the topic at follow-up visits.

For patients under 40 years of age, I think it’s reasonable to use risk stratification like the guidelines recommend, but the evidence is weak and I wouldn’t push it.

Efficacy from RCTs

For this deep dive my initial thought was to focus on RCTs that specifically enrolled patients with diabetes and without cardiovascular disease. However, I found that there were only three trials like this, and two of these trials were quite small, so I also looked at studies that had a large population of patients with diabetes. I used
all-cause mortality as the patient-oriented outcome. I don’t have any new thoughts and see my previous post on statins in patients without diabetes for more details.

Similar to the RCTs for patients without diabetes, none of the RCTs were powered for all-cause mortality, so it is possible that a small improvement in all-cause mortality might go undetected due to low event rates. I tried to account for this by using a meta-analysis for all-cause mortality but also by considering the composite cardiovascular outcome in my internal calculus.

ASCOT-LLA; Sever et al., 2005

• Population: The study enrolled patients with and without diabetes, but this publication is a pre-specified subgroup analysis specifically on the patients with diabetes.

  • The average patient: 63 years old, white, BP 165/92, and PREVENT score 24%.

  • Inclusion/exclusion criteria: 40-79 years old with hypertension and at least 3 of: type 2 diabetes, male, age >55, microalbuminuria or proteinuria, smoking, total cholesterol:HDL >6, premature family history of cardiovascular disease, left ventricular hypertrophy, history of stroke or TIA, peripheral artery disease.

• Treatment: Atorvastatin 10 mg (n = 1,258 for patients with diabetes) vs placebo (n = 1,274).

• Follow-up: 3 years.

• Change in LDL: 126 to 93 at 3 years.

• Change in A1c: A1c wasn’t reported but fasting glucose 8.6 mM which is roughly A1c 7-7.5. This number was unchanged at 3 years.

• Change in SBP: 165 to 138 at 3 years.

• Outcome: All-cause mortality was not reported. 23% reduction and NNT 37 for cardiovascular events (MI, stroke, revascularization, unstable angina, chronic stable angina, life-threatening arrhythmia, heart failure, peripheral artery disease, retinal vascular thrombosis, TIA). The event rate for the placebo group was 11.9%.

• Safety: No statistically significant adverse events.

ASPEN; Knopp et al., 2006

• Population: Everyone in this study had diabetes but it included both primary and secondary prevention patients.

  • The average patient: 60 years old, white, BP 133/77, A1c 7.8, and PREVENT score 16%.

  • Inclusion/exclusion criteria: 40-75 years old with type 2 diabetes, LDL <140 if MI, unstable angina, or an interventional procedure within 3 months and LDL <160 otherwise.

• Treatment: Atorvastatin 10 mg (n = 959 for the primary prevention subgroup) vs placebo (n = 946 for the primary prevention subgroup).

• Follow-up: 4 years.

• Change in LDL: 114 to 79 at 4 years.

• Change in A1c: 7.6 to 7.8 at 4 years.

• Outcome: There was no difference in all-cause mortality or cardiovascular events (MI, stroke, revascularization, resuscitated cardiac arrest, or unstable angina). The event rate for the placebo group was 10.8%.

• Safety: No statistically significant adverse events.

Beishuizen et al., 2004

• Population: This was one of three studies that enrolled all patients with diabetes for primary prevention.

  • The average patient: 58 years old, white, 55% of patients were on insulin, A1c 7.6, and the PREVENT score was not calculable as the study did not list BP or eGFR but considering that half the patients were on insulin, I would consider this a high-risk patient population.

  • Inclusion/exclusion criteria: 30-80 years old with type 2 diabetes, and total cholesterol 156-266.

• Treatment: Cerivastatin 0.4 mg (n = 125) vs placebo (n = 125). If you’re thinking that you’ve never heard of cerivastatin it’s because it was recalled during the study for unclear safety concerns and partway through the study the cerivastatin patients were switched to simvastatin 20 mg.

• Follow-up: 2 years.

• Change in LDL: 131 to 108 at 2 years.

• Outcome: No difference in all-cause mortality. 83%!!! reduction and NNT 7 for cardiovascular events (not defined). The event rate for the placebo group was 15%.

• Safety: No statistically significant adverse events.

CARDS; Colhoun et al., 2004

• Population: This was one of three studies that enrolled all patients with diabetes for primary prevention.

  • The average patient: 61 years old, white, BP 144/83, 15% of patients were on insulin, A1c 7.8, PREVENT score 16%.

  • Inclusion/exclusion criteria: 40-75 years old with type 2 diabetes, LDL <170, and at least one of: microalbuminuria, smoking, hypertension, or retinopathy.

• Treatment: Atorvastatin 10 mg (n = 1,428) vs placebo (n = 1,410).

• Follow-up: 4 years.

• Change in LDL: 116 to 77 at 4 years.

• Change in A1c: 7.8 to 8.2 at 4 years.

• Outcome: There was a statistically insignificant reduction (p = 0.059) in all-cause mortality of 27% and NNT 66. There was also a 37% reduction and NNT 31 for cardiovascular events (cardiovascular death, MI, resuscitated cardiac arrest, coronary revascularization, stroke, and unstable angina). The event rate for the placebo group was 9.0%.

• Safety: No statistically significant adverse events.

Heljic et al., 2009

• Population: This was one of three studies that enrolled all patients with diabetes for primary prevention. This trial had a lot of missing information.

  • The average patient: 60 years old, A1c 8.8, and the PREVENT score could not be calculated due to missing eGFR, smoking status, and BP.

  • Inclusion/exclusion criteria: Patients with type 2 diabetes, A1c <10, BP <140/90.

• Treatment: Simvastatin 40 mg (n = 45) vs placebo (n = 50).

• Follow-up: 1 year.

• Change in LDL: 158 to 131 at 6 months.

• Change in A1c: 8.8 to 7.8 at 6 months.

• Outcome: All-cause mortality was not reported. There was a 53% reduction and NNT 13 for cardiovascular events (not defined). The event rate for the placebo group was 14.0%.

• Safety: No statistically significant adverse events.

HPS; Collins et al., 2003

• Population: This study enrolled patients with and without diabetes for primary and secondary prevention. I focused on their pre-specified subgroup analysis for patients with diabetes.

  • The average patient: 62 years old, white, BP 148/82, 19% of patients were on insulin, A1c 7.8, PREVENT score 22%.

  • Inclusion/exclusion criteria: 40-80 years old with total cholesterol >135, and at least one of: diabetes, coronary disease, occlusive disease of a noncoronary artery, or treated hypertension + male + age >65.

• Treatment: Simvastatin 40 mg vs placebo (n = 5,963 total patients with diabetes).

• Follow-up: 4.8 years.

• Change in LDL: 123 to 85 at 4.8 years.

• Change in A1c: 7.0 to 7.1 at 4 years.

• Outcome: All-cause mortality was not reported. There was a 19% reduction and NNT 20 for major vascular events (MI, stroke, and coronary and noncoronary revascularization). The event rate for the placebo group was 25.1%. In a further subgroup analysis of patients with diabetes and primary prevention there was a 31% reduction and NNT 23.

• Safety: No statistically significant adverse events.

Before we move on to the meta-analysis, I want to highlight CARDS as the only trial that both: 1) Enrolled patients with diabetes for primary prevention and was 2) Powered for a clinical outcome. It was also the study that came closest to finding a statistically significant effect for all-cause mortality. Trial recruitment was stopped early due to the magnitude of the effect that atorvastatin had on the primary outcome (a composite of cardiovascular events). They speculate that if they had been able to reach the initially planned sample size then the finding for all-cause mortality would have reached statistical significance. This is not the most evidence-based conclusion, and if a trial is stopped prematurely for efficacy then that endpoint is usually falsely elevated, but the effect for all-cause mortality is quite large (NNT 66) and I buy it.

A caveat to CARDS is diabetes management in 2026 vs 2004:

1. The patients at enrollment had an A1c of 7.8 which increased to 8.2 by the end of the study. On one hand this is good because it really isolates the benefit of a statin. However, it really does not fit into real life modern diabetes management. I can confidently say that A1c <8 is good, and many would argue that an even stricter A1c goal is appropriate. Would statins add benefit on top of A1c <8?

2. This study was pre-SGLT2i and GLP-1 agonists, both of which we know to improve all-cause mortality. Would statins add benefit on top of SGLT2’s and GLP-1’s?

Aspirin used to be the go-to medication for primary prevention of cardiovascular disease, but as other cardiovascular technology evolved, aspirin moved out of vogue. I wonder if this is a similar situation. If patients on maximally tolerated metformin, GLP-1, and SGLT2i with A1c <8 were enrolled, would statins help?

Efficacy from meta-analyses

Cholesterol Treatment Trialist (CTT) collaboration, 2008

• The CTT is a group of investigators that share their data to conduct individual patient-level meta-analyses. Having the individual patient-level data gives them more ability to converge on a precise outcome and also conduct more sophisticated subgroup analyses.

• Population: 18,686 patients with diabetes from 14 RCTs which included patient with and without diabetes for both primary and secondary prevention. The average patient was 63 years old, white, BP 148/83, and 64% did not have a history of cardiovascular disease. PREVENT score 21%.

• Follow-up: 4.3 years.

• Outcome: They found that statin treatment for patients with diabetes had a statistically insignificant trend towards improvement in all-cause mortality of 9% with NNT 111 in patients with diabetes (p = 0.02; due to multiple testing they set the p value for significance at 0.01). There was a statistically significant improvement in major vascular events (MI, stroke, revascularization, coronary death) of 21% and NNT 27. They did do a subgroup analysis of primary prevention patients and found a similar improvement in major vascular events of 27% and NNT 38. All-cause mortality was not reported.

Conclusions

As I was trying to synthesize the data into a nice tidy conclusion I started to think about how I draw conclusions from the data. People complain that they cannot trust AI because it is a black box. For instance, I could ask OpenEvidence if a patient with diabetes should take a stain, and it would spit out a bunch of information that may or may not be correct. The black box part is the process connecting the question to the response. Nobody knows how OpenEvidence came up with that answer and that should be concerning.

I have similar complaints about guidelines. They offer citations for RCTs and meta-analyses but rarely explain the thought process of meshing these various results. When the results are consistent and the thought process is straightforward that is fine. But in most cases, including this one, there are different ways to draw conclusions. Maybe one author would conclude from the CTT meta-analysis that there is no benefit to all-cause mortality and therefore not recommend statins. Perhaps another would only consider CARDS because it was the only trial that directly answered the question of interest.

In primary care, there are almost always multiple acceptable answers to a clinical scenario, and these answers can be reached in different acceptable ways. It’s impossible for an individual clinician to question the epistemology of every decision they make in a day, but a crucial part of critical appraisal is to explain how we jump from the studies to the clinical conclusion.

Here, I’ve read what I consider to be the relevant RCTs. I summarized each RCT and meta-analysis. From there I thought about three things:

1. The most important result was CARDS because it was designed to answer my exact question. It found a near significant effect on all-cause mortality and robust effect on cardiovascular outcomes. The trial was terminated early, so in my head I am thinking that if there is a true effect it’s not as robust as the 27% reduction and NNT 66 they reported.

2. The CTT meta-analysis was also important because the outcome I was most interested in, all-cause mortality, was not the primary outcome for any of the RCTs and they were not powered to address that outcome. It did not find a statistically significant effect on all-cause mortality but they did see a 9% reduction and NNT 111, which roughly fits with the CARDS result.

3. I came into this analysis with the preconceived notion that statins were good for patients with diabetes and it would have taken striking results to change my mind.

Based on the results my conclusion is that statins are good for primary prevention of cardiovascular disease and mortality in patients with diabetes. I hope the jump from the studies to this conclusion is clear, but to some extent it’s a black box even to me. I prioritized CARDS, but did I weigh it as 90% of my decision-making? 50%? I had this preconceived notion that statins were good, but how negative of a result would have swayed me? I came in with an interest in all-cause mortality and no RCT or meta-analysis found a statistically significant effect on all-cause mortality. Some would argue that is a pretty damning negative result. I still stand by my conclusion.

I find the black box aspect of AI to be scary, and it frustrates me that guidelines writers delve deeper into the rationale, but some black box is inevitable. I hope that I’ve increased the transparency as much as possible and would love to hear different conclusions on the data.

References

1. American Diabetes Association Professional Practice Committee for D. 10. Cardiovascular Disease and Risk Management: Standards of Care in Diabetes-2026. Diabetes Care. Jan 1 2026;49(Supplement_1):S216–S245. doi:10.2337/dc26-S010Blumenthal RS, Morris PB, Gaudino M, et al. 2026

2. ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the Management of Dyslipidemia: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol. Mar 13 2026;doi:10.1016/j.jacc.2025.11.016

3. Beishuizen ED, van de Ree MA, Jukema JW, et al. Two-year statin therapy does not alter the progression of intima-media thickness in patients with type 2 diabetes without manifest cardiovascular disease. Diabetes Care. Dec 2004;27(12):2887–92. doi:10.2337/diacare.27.12.2887

4. Cholesterol Treatment Trialists C, Kearney PM, Blackwell L, et al. Efficacy of cholesterol-lowering therapy in 18,686 people with diabetes in 14 randomised trials of statins: a meta-analysis. Lancet. Jan 12 2008;371(9607):117–25. doi:10.1016/S0140-6736(08)60104-X

5. Colhoun HM, Betteridge DJ, Durrington PN, et al. Primary prevention of cardiovascular disease with atorvastatin in type 2 diabetes in the Collaborative Atorvastatin Diabetes Study (CARDS): multicentre randomised placebo-controlled trial. Lancet. Aug 21–27 2004;364(9435):685–96. doi:10.1016/S0140-6736(04)16895-5

6. Collins R, Armitage J, Parish S, Sleigh P, Peto R, Heart Protection Study Collaborative G. MRC/BHF Heart Protection Study of cholesterol-lowering with simvastatin in 5963 people with diabetes: a randomised placebo-controlled trial. Lancet. Jun 14 2003;361(9374):2005–16. doi:10.1016/s0140-6736(03)13636-7

7. Knopp RH, d’Emden M, Smilde JG, Pocock SJ. Efficacy and safety of atorvastatin in the prevention of cardiovascular end points in subjects with type 2 diabetes: the Atorvastatin Study for Prevention of Coronary Heart Disease Endpoints in non-insulin-dependent diabetes mellitus (ASPEN). Diabetes Care. Jul 2006;29(7):1478–85. doi:10.2337/dc05-2415

8. Heljic B, Velija-Asimi Z, Kulic M. The statins in prevention of coronary heart diseases in type 2 diabetics. Bosn J Basic Med Sci. Feb 2009;9(1):71–6. doi:10.17305/bjbms.2009.2860

9. Sever PS, Poulter NR, Dahlof B, et al. Reduction in cardiovascular events with atorvastatin in 2,532 patients with type 2 diabetes: Anglo-Scandinavian Cardiac Outcomes Trial--lipid-lowering arm (ASCOT-LLA). Diabetes Care. May 2005;28(5):1151–7. doi:10.2337/diacare.28.5.1151

Some would say that statins should be in the water. Others would say they are a muscle-wasting poison. The truth is probably somewhere in between.

The ACC AHA 2026 Dyslipidemia guideline recommends using the PREVENT score to risk-stratify patients and offer the following flowchart to decide who should be on a statin (Blumenthal et al., 2026):

I wrote about the PREVENT score previously. It updates the PCE score by adding heart failure as an outcome and recalibrating risk using a contemporary cohort as the previous cohort overestimated the risk. The PREVENT score is roughly 5% lower than the PCE score for the same patient.

I played around with the PREVENT score a bit. If I set all the variables like blood pressure and cholesterol at the lowest end of normal and only adjusted the age, every man 55-79 years old and woman 59-79 years old would be recommended a statin. This is people with a systolic BP of 100, no diabetes, non-smokers, with a BMI of 20. That feels aggressive, which led me to this deep dive into the evidence underlying statins in low-risk patients. Based on the RCTs, which I explain below, I define low risk as: no diabetes, no history of cardiovascular disease, LDL <190, and normal CRP.

tldr

For a low-risk patient, I would not recommend a statin. I do not find convincing evidence for a benefit to all-cause mortality.

Efficacy from RCTs

For this deep dive I selected RCTs that evaluated statins for primary prevention in patients without diabetes and I focused on all-cause mortality as the patient-oriented outcome.

On one hand, avoiding a heart attack or stroke sounds good. On the other hand, we do not have clear evidence that avoiding cardiovascular events is a patient-oriented outcome. Thirty years ago, an MI was a death sentence and clearly a patient-oriented outcome. Today, with advances in medical technology and interventional cardiology, that is no longer the case. Modern meta-analyses have found that nonfatal MIs are not a good surrogate outcome for either cardiovascular or all-cause mortality (O’Fee et al., 2021). The authors speculate that this is due to the ability of high-sensitivity troponins to detect MIs with minimal myocardial injury and PCI technology to reverse the larger one’s.

Ultimately, I feel that any intervention for primary prevention needs to demonstrate an effect on all-cause mortality because there should be a high bar to take a person and turn them into a patient who takes a daily medication and needs lab monitoring.

Importantly, none of the individual RCTs were powered for all-cause mortality, so it is possible that a small improvement in all-cause mortality might go undetected due to low event rates. Accordingly, I spent more time thinking about meta-analyses for this question than I typically would.

ACAPS; Furberg et al., 1994

• Population: 40-79 years old and mild cardiovascular disease defined by mild carotid artery stenosis. The average patient was 62 years old, white, and BP 131/77. 12% were smokers and 2% had diabetes. PREVENT score 8%.

• Treatment: Lovastatin 20-40 mg (n = 460) vs placebo (n = 459) with goal LDL 90-110.

• Follow-up: 2.8 years.

• Change in LDL: 156 to 113 at 6 months.

• Outcome: 88%!!! improvement in all-cause mortality and NNT 65. The mortality rate for the placebo group was 1.7%. There was also a 65% improvement and NNT 50 for cardiovascular events (MI, stroke, cardiovascular death). The cardiovascular event rate for the placebo group was 3.1%.

• Safety: No statistically significant adverse events.

AFCAPS/TexCaps; Downs et al., 1998

• Population: Men 45-73 years old and postmenopausal women 55-73 years old with LDL 130-190. The average patient was 58 years old, white, and BP 138/78. 12% were smokers and 2.5% had diabetes. PREVENT 7%.

• Treatment: Lovastatin 20-40 mg (n = 3,304) vs placebo (n = 3,301) with goal LDL <110.

• Follow-up: 5.2 years.

• Change in LDL: 150 to 112 at 1 year.

• Outcome: No difference in all-cause mortality. There was a 37% reduction and NNT 49 for acute major coronary events (fatal and nonfatal MI, unstable angina, sudden cardiac death). The event rate for the placebo group was 5.5%.

• Safety: No statistically significant adverse events.

ALL-HAT, 2002

• Population: >55 years old, hypertension, LDL 120-190. The average patient was 66 years old, white, and BP 145/84. 23% were smokers and 35% had diabetes. PREVENT 12%.

• Treatment: Pravastatin 20-40 mg (n = 5,170) vs usual cares (n = 5,180) with goal LDL decrease of 25%. Of note, 25% of the usual cares patients were taking statins by the end of the study.

• Follow-up: 4.8 years.

• Change in LDL: 145 to 104 at end of study.

• Outcome: No difference in all-cause mortality or fatal + nonfatal MI. The event rate for the placebo group was 8.1%.

• Safety: No statistically significant adverse events.

ASCOT; Sever et al., 2003

• Population: 40-79 years old, hypertension, total cholesterol <251. The average patient was 63 years old, white, and BP 164/95. 33% were smokers and 24% had diabetes. PREVENT 12%.

• Treatment: Atorvastatin 10 mg (n = 5,168) vs usual cares (n = 5,137).

• Follow-up: 3.3 years.

• Change in LDL: 130 to 89 at end of study.

• Outcome: No difference in all-cause mortality. There was a 36% reduction and NNT 90 for fatal + nonfatal MI. The event rate for the placebo group was 2.9%.

• Safety: No statistically significant adverse events.

HOPE-3; Yusuf et al., 2016

• Population: Men >55 years old, women > 65 years old, with one risk factor for cardiovascular disease. The average patient was 65 years old, Chinese, and BP 138/81. 27% were smokers and 5% had diabetes. PREVENT 10%.

• Treatment: Rosuvastatin 10 mg (n = 6,361) vs usual cares (n = 6,344).

• Follow-up: 5.6 years.

• Change in LDL: 127 to 98 at end of study.

• Outcome: No difference in all-cause mortality. There was a 24% reduction and NNT 90 for cardiovascular death and nonfatal MI or stroke. The event rate for the control group was 4.7%.

• Safety: Increased muscle pain/weakness with rosuvastatin (5.8 vs 4.7%), increased cataract surgery with rosuvastatin (3.8 vs 3.1%).

JUPITER; Ridker et al., 2008

• Population: Men >50 years old, women >60 years old, LDL <130, and CRP >2. The average patient was 66 years old, white, and BP 134/80. 15% were smokers and 0% had diabetes. PREVENT 9%.

• Treatment: Rosuvastatin 20 mg (n = 8,901) vs placebo (n = 8,901).

• Follow-up: 1.9 years.

• Change in LDL: 108 to 55 at end of study.

• Outcome: 20% reduction in all-cause mortality and NNT 181. There was a 44% reduction and NNT 81 for a composite cardiovascular outcome including fatal and nonfatal MI, fatal and nonfatal stroke, hospitalization for unstable angina, arterial revascularization, cardiovascular death. The event rate for the control group was 2.8%.

• Safety: There was a 25% increase in new diabetes with NNH 166.

MEGA; Nakamura et al., 2006

• Population: Men and postmenopausal women 40-70 years old, cholesterol 220-280. The average patient was 58 years old, Japanese, and BP 132/78. 15% were smokers and 0% had diabetes. PREVENT 9%.

• Treatment: Pravastatin 10-20 mg (n = 3,966) vs usual cares (n = 3,866) with goal cholesterol <220.

• Follow-up: 5.3 years.

• Change in LDL: 157 to 122 at end of study.

• Outcome: The p-value for all-cause mortality was 0.055 with a 28% risk reduction and NNT 90. There was a 33% reduction and NNT 119 for a composite cardiovascular outcome including fatal and nonfatal MI, angina, cardiac and sudden death, coronary revascularization. The event rate for the control group was 2.6%.

• Safety: No statistically significant adverse events.

PROSPER; Shepherd et al., 2002

• Population: 70-82 years old, hypertension, total cholesterol 154-348. The average patient was 75 years old, white, and BP 154/83. 27% were smokers and 11% had diabetes. PREVENT 20%.

• Treatment: Pravastatin 40 mg (n = 2,891) vs placebo (n = 2,913).

• Follow-up: 3.2 years.

• Change in LDL: 147 to 107 at 2 years.

• Outcome: No difference in all-cause mortality. There was a 15% reduction and NNT 47 for fatal + nonfatal MI and fatal + nonfatal stroke. The event rate for the placebo group was 16.2%.

• Safety: There was a 25% increase in new cancer diagnoses in the pravastatin group with NNH 75. There was no difference when broken down to cancer subtypes, but the greatest effect was seen for GI cancers.

WOSCOPS; Shepherd et al., 1995

• Population: Men 45-64 years old with cholesterol >252 and LDL >155 on two occasions with one occasion >174 and one occasion <232. The average patient was 55 years old, white, and BP 136/84. 44% were smokers and 1% had diabetes. PREVENT 6% but the average WOSCOPS patient would not qualify for PREVENT due to their baseline LDL of 192.

• Treatment: Pravastatin 40 mg (n = 3,302) vs placebo (n = 3,293) with goal LDL <110.

• Follow-up: 4.9 years.

• Change in LDL: 192 to 142 at 6 months.

• Outcome: The p-value for all-cause mortality was 0.051 with a 22% reduction and NNT 111. There was a 31% reduction and NNT 41 for fatal and nonfatal MI. The event rate for the placebo group was 7.5%.

• Safety: No statistically significant adverse events.

Before we move on to the meta-analyses I want to highlight three trials: ACAPS and JUPITER as the studies that found a statistically significant effect on all-cause mortality, and WOSCOPS as the study that was the closest to statistical significance.

1. For ACAPS, which found a striking 88% reduction in all-cause mortality and NNT 65, I tend to consider this as somewhat of a statistical anomaly. First, it was a small trial with a total of 919 patients. Second, AFCAPS enrolled a very similar population, used the same intervention, and found no effect on all-cause mortality.

2. For JUPITER, which found a 20% reduction in all-cause mortality and NNT 181, I do not consider this representative of a typical low-risk population. The premise of JUPITER was that inflammation is an independent risk marker for cardiovascular disease and it specifically studies a population of patients with normal LDL and elevated CRP.

   1. The study that prompted JUPITER was a posthoc analysis of AFCAPS that found that lovastatin improved coronary event rates in patients with an above median LDL regardless of CRP, but that lovastatin only improved coronary event rates in patients with a below median LDL if they had an elevated CRP (Ridker et al., 2001). This would suggest that CRP could dictate a differential response to statins.

   2. However, other studies like a subgroup analysis of HOPE-3 found no difference in effect size when comparing patients with normal vs elevated CRP. Similarly, a post-hoc analysis of JUPITER itself found that patients with higher CRP had higher rates of cardiovascular events but did not have a differential benefit to rosuvastatin when compared to lower CRP (Ridker et al., 2010).

   3. It remains unclear whether CRP identifies a subgroup with a differential response to statins but considering that JUPITER only enrolled patients with an elevated CRP, I do not feel that it represents a typical low-risk patients.

   4. The other caveat to JUPITER is that it was the only study that used a high-intensity statin. There is some evidence that high-intensity statins outperform moderate-intensity and my other conclusion from JUPITER is that we need an RCT of high-intensity statin for primary prevention in patients with a normal CRP (CTT, 2010).

3. WOSCOPS was the closest to finding a statistically significant effect on all-cause mortality (p for all-cause mortality was 0.051). However, the average patient in this trial had an LDL of 192, which would put patients in the severe hypercholesterolemia range. I do not think there is debate as to whether these patients should be on a statin and I do not find it reasonable to include this in an analysis of low-risk patients.

Efficacy from meta-analyses

Cholesterol Treatment Trialist (CTT) collaboration, 2012

• The CTT is a group of investigators that share their data to conduct individual patient-level meta-analyses. Having the individual patient-level data gives them more ability to converge on a precise outcome and also conduct more sophisticated subgroup analyses.

• Sample: 27 RCTs which included both primary and secondary prevention studies.

• Findings: They found that patients at lower risk had a greater relative risk reduction for major coronary events:

Interestingly, they did not find this trend for all-cause mortality and they did not observe a statistically significant effect on all-cause mortality until they reached patients with a >10% risk for having a vascular event in 5 years. I suppose this is because the younger patients that avoid a vascular event would not have died from it. In the whole population they found a 9% reduction in all-cause mortality with NNT 500.

USPSTF; Chou et al., 2022

• Sample: 26 RCTs of primary prevention.

• Findings: 8% reduction in all-cause mortality with NNT 286 at 1.6 years and a 28% reduction in composite cardiovascular events with NNT 78).

Ray et al., 2010

• Sample: 9 RCTs of primary prevention in patients without diabetes.

• Findings: No effect on all-cause mortality.

Meta-analyses can lead to different conclusions depending on how you determine the inclusion criteria. For this reason, I’m cautious about treating meta-analyses as the undisputed king of the evidence hierarchy. Of these three trials, I would consider Ray et al.’s study to be most representative of a low-risk patient, and this trial was negative for all-cause mortality.

Safety from meta-analyses

For safety-related findings, I find that meta-analyses are more useful because RCTs are almost never powered for safety outcomes. I tend to think of medication benefits as condition-specific, whereas medications harms are more generalizable. Thus, it’s not always reasonable to pool results to look for the evidence across heterogeneous patients but it’s more reasonable to pool results to look at safety.

The USPSTF meta-analysis found that statins were not associated with an increased risk of cancer, transaminitis, myalgia, rhabdomyolysis, or diabetes (Chou et al., 2022). They did note that one study on high-intensity statin did find statins caused diabetes. The CTT meta-analyses found that statins were associated with an increase in transaminitis (RR 1.41, NNH 1111; CTT, 2026) and diabetes (RR 1.10 for low/moderate-intensity and 1.36 for high-intensity, NNH 1000 and 76; CTT, 2024). They did not find an effect on myalgia.

Conclusions

Let’s start with statin safety. In the context of safety, statins are probably the most studied medication in all of human history. I don’t necessarily think this means they are safe, but I do think it means we have a really good idea of what side effects are caused by statins. For instance, there is lots of discussion about the muscle-related problems, but these are rare and much more likely to be due to the nocebo effect. There is a real effect on diabetes but mainly for high-intensity statins which would not typically be recommended for primary prevention.

Next, let’s use the USPSTF meta-analysis to think about efficacy (Chou et al., 2022). I want to start by highlighting the four studies that contributed the most to the all-cause mortality analysis.

One study (JUPITER) enrolled a specific population of patients with normal LDL and elevated CRP and I do not think this study should be generalized to the typical low-risk patient.

The other four studies (ALLHAT, ASCOT, HOPE-3) enrolled similar patients with high-normal to slightly elevated LDL that were treated with low or moderate-intensity statins. These three studies did not find an effect on all-cause mortality. In fact, the only primary prevention RCT powered for all-cause mortality was ALLHAT, which was stone cold negative with a risk ratio of 1.00.

Based on these studies, I do not think that statins improve all-cause mortality for the low-risk patient, which I would define as no diabetes, no history of cardiovascular disease, normal CRP, and LDL <190. Given the positive findings from JUPITER, I do think it is reasonable to check a CRP and consider statin treatment if the CRP is elevated.

A couple questions that this deep dive left me with:

1. Who should be on a statin? Second prevention? Diabetes?

2. Where is the line that a patient crosses over from low- to high-risk for primary prevention?

References

1. ALLHAT Officers and Coordinators for the ALLHAT Collaborative Research Group. The Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial. Major outcomes in moderately hypercholesterolemic, hypertensive patients randomized to pravastatin vs usual care: The Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT-LLT). JAMA. Dec 18 2002;288(23):2998–3007. doi:10.1001/jama.288.23.2998

2. Blumenthal RS, Morris PB, Gaudino M, et al. 2026 ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the Management of Dyslipidemia: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol. Mar 13 2026;doi:10.1016/j.jacc.2025.11.016

3. Cholesterol Treatment Trialists Collaboration. Assessment of adverse effects attributed to statin therapy in product labels: a meta-analysis of double-blind randomised controlled trials. Lancet. Feb 14 2026;407(10529):689–703. doi:10.1016/S0140-6736(25)01578-8

4. Cholesterol Treatment Trialists’ Collaboration. Effects of statin therapy on diagnoses of new-onset diabetes and worsening glycaemia in large-scale randomised blinded statin trials: an individual participant data meta-analysis. Lancet Diabetes Endocrinol. May 2024;12(5):306–319. doi:10.1016/S2213-8587(24)00040-8

5. Cholesterol Treatment Trialists C, Baigent C, Blackwell L, et al. Efficacy and safety of more intensive lowering of LDL cholesterol: a meta-analysis of data from 170,000 participants in 26 randomised trials. Lancet. Nov 13 2010;376(9753):1670–81. doi:10.1016/S0140-6736(10)61350-5

6. Cholesterol Treatment Trialists Collaboration, Mihaylova B, Emberson J, et al. The effects of lowering LDL cholesterol with statin therapy in people at low risk of vascular disease: meta-analysis of individual data from 27 randomised trials. Lancet. Aug 11 2012;380(9841):581–90. doi:10.1016/S0140-6736(12)60367-5

7. Chou R, Cantor A, Dana T, et al. Statin Use for the Primary Prevention of Cardiovascular Disease in Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA. Aug 23 2022;328(8):754–771. doi:10.1001/jama.2022.12138

8. Downs JR, Clearfield M, Weis S, et al. Primary prevention of acute coronary events with lovastatin in men and women with average cholesterol levels: results of AFCAPS/TexCAPS. Air Force/Texas Coronary Atherosclerosis Prevention Study. JAMA. May 27 1998;279(20):1615–22. doi:10.1001/jama.279.20.1615

9. Furberg CD, Adams HP, Jr., Applegate WB, et al. Effect of lovastatin on early carotid atherosclerosis and cardiovascular events. Asymptomatic Carotid Artery Progression Study (ACAPS) Research Group. Circulation. Oct 1994;90(4):1679–87. doi:10.1161/01.cir.90.4.1679

10. Nakamura H, Arakawa K, Itakura H, et al. Primary prevention of cardiovascular disease with pravastatin in Japan (MEGA Study): a prospective randomised controlled trial. Lancet. Sep 30 2006;368(9542):1155–63. doi:10.1016/S0140-6736(06)69472-5

11. O’Fee K, Deych E, Ciani O, Brown DL. Assessment of Nonfatal Myocardial Infarction as a Surrogate for All-Cause and Cardiovascular Mortality in Treatment or Prevention of Coronary Artery Disease: A Meta-analysis of Randomized Clinical Trials. JAMA Intern Med. Dec 1 2021;181(12):1575–1587. doi:10.1001/jamainternmed.2021.5726

12. Ray KK, Seshasai SR, Erqou S, et al. Statins and all-cause mortality in high-risk primary prevention: a meta-analysis of 11 randomized controlled trials involving 65,229 participants. Arch Intern Med. Jun 28 2010;170(12):1024–31. doi:10.1001/archinternmed.2010.182

13. Ridker PM, Danielson E, Fonseca FA, et al. Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein. N Engl J Med. Nov 20 2008;359(21):2195–207. doi:10.1056/NEJMoa0807646

14. Ridker PM, MacFadyen J, Libby P, Glynn RJ. Relation of baseline high-sensitivity C-reactive protein level to cardiovascular outcomes with rosuvastatin in the Justification for Use of statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER). Am J Cardiol. Jul 15 2010;106(2):204–9. doi:10.1016/j.amjcard.2010.03.018

15. Ridker PM, Rifai N, Clearfield M, et al. Measurement of C-reactive protein for the targeting of statin therapy in the primary prevention of acute coronary events. N Engl J Med. Jun 28 2001;344(26):1959–65. doi:10.1056/NEJM200106283442601

16. Sever PS, Dahlof B, Poulter NR, et al. Prevention of coronary and stroke events with atorvastatin in hypertensive patients who have average or lower-than-average cholesterol concentrations, in the Anglo-Scandinavian Cardiac Outcomes Trial--Lipid Lowering Arm (ASCOT-LLA): a multicentre randomised controlled trial. Lancet. Apr 5 2003;361(9364):1149–58. doi:10.1016/S0140-6736(03)12948-0

17. Shepherd J, Blauw GJ, Murphy MB, et al. Pravastatin in elderly individuals at risk of vascular disease (PROSPER): a randomised controlled trial. Lancet. Nov 23 2002;360(9346):1623–30. doi:10.1016/s0140-6736(02)11600-x

18. Shepherd J, Cobbe SM, Ford I, et al. Prevention of coronary heart disease with pravastatin in men with hypercholesterolemia. West of Scotland Coronary Prevention Study Group. N Engl J Med. Nov 16 1995;333(20):1301–7. doi:10.1056/NEJM199511163332001

19. Yusuf S, Bosch J, Dagenais G, et al. Cholesterol Lowering in Intermediate-Risk Persons without Cardiovascular Disease. N Engl J Med. May 26 2016;374(21):2021–31. doi:10.1056/NEJMoa1600176

I cannot even hazard a guess for the cardiologists but hopefully it is close to 100%. For practicing primary care doctors, I can only speak for myself, but when I was a full-time clinical doctor I exclusively practiced from point-of-care resources like UpToDate. The only time I read a guideline in its entirety was if I was writing a substack post about it. There may have been a time when I read part of a guideline to inform my clinical care but I cannot remember a specific example.

Is this a reflection of me? Of primary care doctors? Or the guidelines themselves? The purpose of a guideline is to distill clinical research into a useable format for practicing clinicians and a 123-page document with 130 recommendations is hardly accessible. I want to explain how I approach this guideline with my part-time practicing-primary-care-physician hat, and part-time academic-critical-appraisal hat. I hope to provide some perspective on sorting out what information is and is not useful to gain from a guideline. I won’t discuss the specific content of the guideline here, but I am working on evidence deep-dives for: statins for primary prevention, risk stratification tests beyond the lipid panel, and dietary interventions. Let me know if there is anything else that you think might be interesting. Here we go:

Conclusions

After paring this guideline down, I was basically left with reading the recommendations, and only reading the recommendation-specific supportive text for recommendations that I found especially pertinent to my day-to-day practice. Thinking about this pared down version of the guideline, it sparked another question: Is this pared down reading of a guideline different from reading UpToDate?

To compare, I read the UpToDate page on LDL therapy for primary prevention of cardiovascular disease. The biggest discrepancy was that UpToDate recommended using the PCE to risk stratify patients rather than using the PREVENT score, but they were overall pretty similar. I found that the guideline authors provided deeper rationale for their recommendations and accordingly was much shorter than the 123-page ACC AHA guideline.

For my practicing-primary-care-physician hat, UpToDate provides perfectly adequate point-of-care guidance, and I do not find much value-add to reading the guideline. I think the reason to reach for the guideline is if you question the mainstream medical recommendations. Like: Does everyone with a PREVENT score of 3% really need a statin? Or, who really needs an Lp(a)? In this case, the guideline can help you see what the medical authority figures are thinking, and you can see for yourself if you agree with their reasoning.

Friday, 4:17 PM: My last patient is still being roomed. I’m running a bit behind because the medical assistant (MA) helping me is also rooming for another provider. There’s a hiring freeze at our healthcare system and we can’t replace the MA who recently quit to be a prep cook at Epic. “Better benefits,” she said.

I try to make use of this time and knock out some inbox messages. I open the first message: patient JA calling to check on her Oxycontin refill. I haven’t met JA yet. I feel a little weird sending an opioid for a patient I’ve never met. One of the other physicians at our primary care clinic retired early because he couldn’t deal with the electronic medical record (EMR), and I absorbed his patients into my panel. I’ve asked JA to schedule an appointment with me so we could discuss an integrative approach to her chronic pain. She’s interested, but could she just wait a couple months because she saw one of the APPs a couple months ago and it cost her $400 for a 10-minute appointment.

I feel bad for her. It’s only 5 mg twice a day. I sign the order.

But first, I need to check the PDMP. The patient has three locations in the PDMP. Click through each. Looks fine. I hit sign.

Please confirm with your fingerprint.

Please confirm with your phone.

My phone is in my office. Walk to office, get phone, but now the confirmation has timed out and I have to reconfirm my fingerprint. I finally sign the prescription.

4:22 PM. My patient is still not ready, so I open the next message: an ER visit for my patient HF. Last month she was diagnosed with heart failure with reduced ejection fraction and started on guideline-directed medical therapy in the hospital. She was told to follow-up with me in one week. Our schedule has dedicated hospital follow-up slots, but I filled them with other chronic patients, so she scheduled with me in a month. Before she made it to her hospital follow-up appointment, she was readmitted with another heart failure exacerbation. Her home lasix dose wasn’t quite right.

4:22 PM. Another message from JA. The Walgreens she usually goes to is out of Oxycontin and she really needs her script. She’s been out for three days, and her pain is really bad. She works for a home health care agency, and two of her co-workers recently quit, so she’s had to take on more patients while they try to hire replacements. She can’t take a day off because she’s a single mom. She knows she’s only supposed to use one pharmacy according to her controlled substance agreement, but couldn’t I make an exception just this one time because the CVS has Oxycontin?

I sign the order.

Confirm with fingerprint.

Confirm with phone.

4:24 PM: I see my 4:10 patient with four chief complaints. I’ll bill 99215 based on time, but that visit could have been four 99214’s.

4:55 PM. I recheck my inbox to try and clear it out for the weekend. JA called again. Now CVS is also out of Oxycontin, but they do have Vicodin so could she try that? I try to send the e-prescription but when I check the PDMP, for some reason she doesn’t have an account. Does this matter? I sign the order.

Confirm with fingerprint.

Confirm with phone.

4:56 PM: I wrap up my notes from the day.

5:15 PM: Another message from JA. Insurance won’t approve Vicodin so could she try switching to Norco? I sign the order and log off before she can call back.

Monday, 12:17 PM: Another message from JA. It’s in ALL CAPS. Her insurance company denied Norco. She’s been out of the medication for six days now. I don’t know what to do, but it’s like she’s yelling at me. I just wanted to help.

12:31 PM: I message our clinic pharmacist, who checks the prior auth status of other opioids. None of them are covered. I considering ignoring the message and hoping it resolves itself, but I build up the courage to message JA back. I tell her we have to wait until her pharmacy restocks Oxycontin and maybe she could try flexeril in the meantime.

She doesn’t message back.

Tuesday, 11:21 AM. JA calls to request a new primary care physician. Nobody in our system has openings.

Thursday, 9:07 AM. JA gets her Oxycontin from Costco.

This is a small problem. It got resolved in a week. But I see variations of this every day in clinic. Patients can’t get medications because of prior authorizations. Care becomes fragmented through patient portal messages. Inboxes keep growing.

It’s easy to brush these off as isolated issues: a staffing issue here, a workflow issue there. It’s tempting to think we can fix this with technology, or good old fashioned hard work. But are these problems an annoying coincidence, or the sign of something deeper?

The more I encounter these situations in different settings with different patients, the more I start to see patterns: a system reshaped by consolidation, a war of attrition between physicians and other healthcare stakeholders, providers leaving medicine, and a growing access desert. If we’re in the midst of a gradual healthcare apocalypse, these patterns are the horsemen.

The white horseman is conquest by the false savior. He represents the false promise of efficiency through healthcare system consolidation.

Excessive spending on healthcare is getting a lot of attention today, but policy analysts and politicians have been working diligently on this problem for years. Two decades ago, the Affordable Care Act (ACA) was designed to address the incentives related to fee-for-service. This payment system rewards more testing and more procedures, whether or not they were good for patients. The goal was to transition to a value-based care payment system through models like accountable care organizations where groups of providers would be accountable for both the quality and costs of taking care of a population. These models were well-intentioned, but they favored large organizations with infrastructure and administrative support, which in turn prompted consolidation.

It’s not clear if this consolidation improved efficiency, but it clearly increased the market power of healthcare systems to negotiate higher payments from insurers. These costs are ultimately passed on to patients and providers. Patients pay these prices with their wallet, and studies show that Americans are spending more than ever on out-of-pocket healthcare dollars (Amin et al., 2024). Meanwhile, physicians pay with their labor.

The red horseman is the war of attrition. He represents the small recurring conflicts between physicians and other healthcare stakeholders.

One battle is between physicians and insurers in the form of prior authorizations. Clinical decisions increasingly require approval, as insurance companies use prior authorizations to steer care towards more cost-effective options. Medicare Advantage plans, which are notorious for utilizing prior authorizations, unleashed 53 million prior authorizations in 2024 (Fuglesten Biniek et al., 2026).

Another battle is between physicians and the systems they work within. The transition from fee-for-service to value-based care changed how our productivity is measured. Instead of visit numbers, the emphasis shifted to panel size. This incentivized larger panel sizes where access was increased by allowing more clinical work to migrate from the patient room to the EMR (Arndt et al., 2024).

Then, there are the patients who just want to access care. A recent survey found that the average wait time for a physician appointment is over a month (AMN Healthcare, 2025). Titrating diabetes medications could probably wait a month, but titrating lasix in a patient with heart failure cannot.

Each individual conflict is trivial, but that’s the exact strategy in a war of attrition.

The green horseman is death. He represents the physicians that are lost to this war of attrition.

The cumulative burden of administrative work, inbox demands, and the resulting sensation of helplessness lead to burnout (Agarwal et al., 2020). Some physicians respond by leaving medicine entirely. More often, the loss is subtle as providers cut back their FTE. A minority are seeking a third path: alternative models like direct primary care (DPC).

By cutting out insurance companies entirely, DPC facilitates a direct contract between physician and patient. For a flat monthly fee, the patient receives unlimited primary care services. The goal is to restore the physician-patient relationship through longer appointments and transparent pricing. While this model can work beautifully for some patients, it’s not accessible to everyone.

The black horseman is famine. He represents the growing inequality produced by worsening access.

As more physicians quit, it becomes harder for patients to access care. Gone are the olden days when patients had “my doctor.” Fewer patients today have a source of regular healthcare and with fewer physicians taking new patients, there is no light at the end of the tunnel (Jabbapour et al., 2024; Morgan et al., 2026).

Patients with money are turning to the alternative models like concierge medicine or DPC. At a concierge practice, patients still pay all the typical insurance fees like co-pays for office visits, but they also pay a $2,000-10,000 annual subscription fee to be a member of the clinic. DPC creates a more affordable model by cutting out the insurance companies, but it still costs $500-2,000 per year.

Nearly 50 years ago, Dr. Arnold Relman, then editor of the New England Journal of Medicine, prophetically described the medical-industrial complex (Relman, 1980). He highlighted the tension between healthcare as a basic right and the power of the free market to improve the quality of healthcare. While he acknowledged the potential benefits of private investments in healthcare, he also warned that:

“We should not allow the medical-industrial complex to distort our health-care system to its own entrepreneurial ends. It should not market useless, marginal, or unduly expensive services, nor should it encourage unnecessary use of services.”

Since then, the health care system has expanded from $250 billion to the $4 trillion beast it is today. Although the spending keeps growing, the outcomes haven’t matched. Over the past 15 years, life expectancy in the US has plateaued, while it continues to rise in comparable countries.

On one hand, it’s hopeful that healthcare has survived the same problems for 50 years. On the other hand, there are major policy changes looming over the next decade that signal the end of infinite growth. The One Big Beautiful Bill Act plans to cut $1 trillion from federal healthcare funds including Medicaid and Medicare. When a system built for expansion encounters constraint, something has to give.

Conquest reshaped the system.

War defined it.

Death reduced it

Famine determines who is left with access.

JA eventually got her Oxycontin, but the horsemen are here to stay.

References:

1. AMN Healthcare. 2025 Survey of Physician Appointment Wait Times and Medicare and Medicaid Acceptance Rates. 2025.

2. Agarwal SD, Pabo E, Rozenblum R, Sherritt KM. Professional Dissonance and Burnout in Primary Care: A Qualitative Study. JAMA Intern Med. Mar 1 2020;180(3):395–401. doi:10.1001/jamainternmed.2019.6326

3. Amin K, Cox, C., Ortaliza, J. & Wager, E. Health Care Costs and Affordability. 2025. Health Policy 101.

4. Arndt BG, Micek MA, Rule A, Shafer CM, Baltus JJ, Sinsky CA. More Tethered to the EHR: EHR Workload Trends Among Academic Primary Care Physicians, 2019-2023. The Annals of Family Medicine. 2024–01–01;22(1)doi:10.1370/afm.3047

5. Jeannie Fuglesten Biniek NS, Meredith Freed, and Tricia Neuman. Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024. 2026.

6. Jabbarpour Y JA, Byun H, Siddiqi A, Petterson S, Park J. The Health of US Primary Care: 2024 Scorecard Report — No One Can See You Now. 2024.

7. Morgan KM, Karadakic R, Barnett ML. Primary Care Clinicians Available for New Patient Visits. JAMA Intern Med. Mar 1 2026;186(3):378–379. doi:10.1001/jamainternmed.2025.7465

8. Relman AS. The new medical-industrial complex. N Engl J Med. Oct 23 1980;303(17):963–70. doi:10.1056/NEJM198010233031703