Big data: HPV vaccines work.

In 2006, the CDC started recommending human papillomavirus (HPV) vaccinations for females ages 9-26. Because certain types of HPV cause cervical cancer, it was hoped that the program would reduce rates of the disease and, later, mortality.

Since 2012, cervical cancer rates have fallen steadily. A new report in the Journal of the American Medical Association out this week shows a dramatic decrease in cervical cancer mortality, which occurred after 2016. That makes perfect sense because it would generally take around 10 years for the effects of the HPV vaccines to be detectable in mortality data—that is, long enough for cancers to have been prevented (which we saw starting in 2012) and for that to translate to a mortality benefit.

Mortality from cervical cancer in US women under age 25 is now less than 1 per million people annually. That’s half the rate we say 10 years ago, and one-quarter of the late 20th century rate. Take a look at the stunning data in JAMA. It shows what was “expected” to happen (the yellow line) based on the pre-vaccine effect trends and what actually happened (the dashed blue line).

Dorali et al JAMA. Data: CDC. Annotations: Inside Medicine.

Is it still happening?

The mortality data in the JAMA paper go through 2021. But I checked, and it looks like 2022-2024 will also have low rates. So, this looks real and sustained.

Are observational data enough?

In the hierarchy of medical evidence, observational data like those presented in the new JAMA paper are considered inferior to data from randomized controlled trials. The main reason is that no matter how hard we try to find good “controls” for patients with a particular disease when conducting research on observational data, there are often confounding variables.

For example, what if rich people (who get better healthcare) happen to be the ones to who have gotten HPV vaccines? If that were the case, then the vaccinated cohort would appear to have a sudden benefit from vaccines, even though vaccinations just teased out those who have access to healthcare from those who do not. Even if you tried to control for that, it would actually be a tough critique to overcome—albeit, it could be possible, given the right data.

But here, it’s not even a study of vaccinated versus unvaccinated. It’s a study of observed versus “expected,” had the pre-vaccine rates/trends continued, as they reliably had for decades. So, the sudden decrease in mortality is quite convincing.

Would randomized trial data help here?

In general, randomized trials (controlled with placebo and blinding) are considered the most definitive (gold standard) type of medical evidence. They are the least confounded data possible because of the randomization and blinding. But in this case, a randomized controlled trial would not help, in any practical reality. That’s because such a trial would not be possible (or insightful). Here’s why:

1. Previous randomized controlled trial data already tell us that HPV vaccines decrease rates of cancerous lesions. It would be unethical to randomize people to get vaccine or placebo at this point, because you’d be depriving the control group of vaccines that we know reduce dangerous lesions.

2. It would be virtually impossible to conduct a randomized controlled trial large enough to capture these effects. The number of deaths from cervical cancer in this population each year is (mercifully) pretty small. There were around 20 deaths per year in females under age 25 in 1995-1997; the number dropped to around 4 deaths per year by 2019-2021. That means you’d have to randomize hundreds of thousands (if not millions) of people and follow them for a decade or two to get results like the ones in the new paper. The largest randomized trial ever conducted, I think, was the RECOVERY trial, which managed to randomize 40,000 people during the Covid era.

3. This is a corollary to the 2nd point, but the effect size deserves special attention here. When the outcome you’re measuring is very common, a small randomized controlled trial can be definitive. Imagine a disease that 50% of the population gets. If a vaccine prevents 75% of these compared to placebo, you might only need 50 people! But if you want to do a trial measuring mortality for a disease like cervical cancer, the trial would not be feasible. Around 1-2 out of 100,000 young women get cervical cancer, and even then, 90% survive. As above, a randomized trial would have to enroll hundreds of thousands of people and follow them for years. That ain’t happening.

Wait, are you saying randomized trials are not as good as observational studies?

I’m not saying that randomized trials are inferior to observational data. In fact, randomized controlled trials are the gold standard in ideal circumstances. What I am saying is that in some situations, we are never going to get a randomized controlled trial—either because of ethics, logistics, or both. Factors to consider include how common the condition is, how often bad outcomes occur, and how long a study would have to last to obtain the answers.

Now, not all observational studies are created equally. Some are great and some are terrible. The new JAMA paper has obvious strengths (mainly its size and crystal-clear modeling). But maybe we’d like to see observational data showing how HPV vaccine recipients fared compared to those who were not vaccinated. That would be “emulating” a trial, albeit it’s very difficult to factor out all the possible confounding variables, of which I can immediately think of several, including family history (direct or distant), social determinants, and local prevalence.

Overall, I’m just glad to see a nice study showing compelling evidence that we have made progress against cervical cancer since the vaccines were introduced, especially since this saves the lives of young people.

Questions? Feedback? Join the conversation in the Comments section!

Thanks to Dr. Cedric Dark for comments and Benjy Renton for alerting me to this study.