The Rise and Fall of T

Testosterone prescriptions were on the rise until worries about risk of heart attack and stroke emerged. But were those worries overblown?

Asset 11.png
 

By Daniel Oppenheimer
Editor, Texas Health Journal

 
 

During the first decade of the 21st century, prescription testosterone use among American men tripled, with many of the additional ‘scrips possibly being written for men who didn’t suffer from hypogonadism, for which testosterone is medically prescribed.

This testosterone boon was driven, in no small part, by direct-to-consumer marketing that touted the benefits of “T” and the deficits of “Low T,” while downplaying potential risks. Testosterone therapy, it was implied, would increase energy, sexual virility, confidence, and other good things even in men who weren’t medically deficient.

Use peaked in 2013, and then rather abruptly began declining. Between 2013 and 2016, according to a new study from researchers at The University of Texas Medical Branch at Galveston (UTMB), overall prescriptions for existing testosterone users decreased by almost half, and prescriptions for new users decreased by more than 60 percent.

The reason for the decline was straightforward. Researchers, alarmed by the dramatic increase in prescriptions, and concerned that many men who were receiving testosterone didn’t have adequate medical reasons, had begun devoting more time to studying the potential risks of prescription testosterone. Then they found some.

Two studies from 2013 and 2014 reported an association between prescription testosterone and an increased risk of heart attack and stroke. In 2014, informed by this research, the Food and Drug Administration issued an alert announcing that they were “investigating the risk of stroke, heart attack, and death in men taking FDA-approved testosterone products.”

The studies and FDA communication, publicized and amplified by the media, changed the orientation of many doctors, patients, and potential patients. As a result, less testosterone was prescribed. In 2017, prescriptions were back down to where they were in 2008.

Jacques Baillargeon 2.jpg

On the surface it seems like a story of science triumphing over hype and marketing. For Jacques Baillargeon, professor in preventive medicine and community health at UTMB, it’s more complicated. Prior to their study of testosterone prescriptions, he and his colleagues had done research that cut against the conclusions of those original, influential studies.

“We went in to the testosterone research with the hypothesis that we would see an increased risk of adverse cardiovascular events,” said Baillargeon.  “That was the literature at the time, and it was getting a lot of news.”

Using anonymized data from Medicare, they identified 6355 older patients treated with testosterone from 1997-2005, and matched this cohort to 19,065 non-users from the same time period. They followed the users and non-users until they experienced a heart attack, died, or left Medicare for another reason, looking to see whether the men who received testosterone were more likely to get heart attacks.

“We saw no association between heart attack and testosterone therapy,” said Baillargeon. “A later study we did found no association between testosterone therapy and venous thromboembolism.”

Since that 2014 FDA warning, research has accumulated on both sides of the debate. Baillargeon believes the balance of the evidence is on his side, but it is not by any means settled.

“I think everyone is operating in good faith,” he said, “but we are making different choices about the methodologies and data sources we use. The important thing is to be extremely transparent about every detail of our methodologies.”

He believes it is important, as well, to reckon with the consequences of these debates, and of the resulting narratives. In the case of prescription testosterone, one clear consequence of the initial wave of research has been the decline in prescriptions. If the suggested risks are real, this may be a good thing. There may have been men, and doctors, who were too quick to buy into the story being told by the advocates and manufacturers of testosterone products. The decline in prescriptions may be saving lives.

“What I would say, though, after this dramatic decrease, is who are the men who are discontinuing the use?” said Baillargeon. “Was it the men who did not need treatment in the first place, or those who had clinically significant hypogonadism. If a significant proportion of the latter group have discontinued or forgone testosterone therapy, it is possible that there will be some significant clinical risks in the long term.”[J1]

The increased risks to these men, from not using testosterone therapy, include osteoporosis, muscle wasting, metabolic syndromes, type 2 diabetes, and ultimately even the cardiovascular events that the warnings were meant to protect against.

“It would be nice to think that the guys who didn’t need testosterone in the first place were the only ones who were discontinuing, but it is unlikely that that is the case,” said Baillargeon. “Ironically, these men with clinically significant hypogonadism may be placing themselves at greater risk for adverse heart events by not taking testosterone.”

The Promise and Perils of Big Data

Jacques Baillargeon 1.jpg

It is an instructive case, for Baillargeon, when it comes to the promise and the perils of big data. When he is assimilating new graduate students and postdocs into his lab, which analyzes vast health databases to identify and understand important patterns and relationships, one of the key points he makes is to be careful.

“Keep the inherent biases in mind,” he says. “Most of these big data sources were not designed for research, but for billing. If something is not a billable claim or related to billing in some meaningful way, it may not be accurately and consistently recorded.”

The Medicare administrative claims database from the Centers for Medicare and Medicaid Services (CMS), for instance, is an extraordinary resource for researchers, enabling them to look at anonymized records from 5% of all Medicare beneficiaries over a 20 year period, which adds up to roughly 40,000,000 records. The records include reliable and nationally representative data on basic demographic information, health care procedures, disease diagnoses and outcomes, mortality, major adverse health events (like strokes and heart attacks), and medication prescriptions.

They lack, however, more detailed demographic information, the kinds of physician’s notes that would be provided in an electronic medical record, the detailed profiling that is done of participants in clinical trials, and good or consistent information on many health conditions that don’t necessitate a specific diagnosis, procedure, or prescription.

Another database used by Baillargeon and his colleagues, the Clinformatics Data Mart, has similar power and constraints. It includes records from 78,000,000 people, over 15 years, who used commercial insurance.

“We can do a lot with databases like this, but we have to ask the right questions,” says Baillargeon.

Baillargeon and his colleagues used the CMS data for their study of heart attack and testosterone therapy, and though he is confident in the analytical rigorousness of their work, he is cautious about its bearing on the larger truth of the matter. It should inform the debate, but can’t settle it. Their study of testosterone prescriptions, on the other hand, is more definitive.

“Tracking prescriptions is simply less complex. It is something for which these data are well suited. The comparative studies are more complex. We have to really think about selection bias.”

Unlike randomized clinical trials, in which test and control subjects can be chosen and profiled with study questions in mind, the big data sets were assembled with other objectives in mind, and they can’t be retroactively fleshed out. The people whose lives are reflected in the data can’t be further studied or profiled. The important information that isn’t captured in the data won’t suddenly materialize.

For certain narrow questions (e.g. the numbers of prescriptions over time), decisive answers can be found. For more complex questions, researchers like Baillargeon and his colleagues use a variety of statistical techniques to retrospectively control for various confounding factors, but there are limitations. The choices researchers make, in the best of faith, can lead to different findings. Conflicts may not be resolved for many years, by new data or more controlled studies, or perhaps ever. In some cases, studies simply have to be abandoned because the data isn’t good enough.

The complexity doesn’t deter Baillargeon from engaging in the work, but it makes him careful.

“People sometimes talk about big data in this fast and loose way,” said Baillargeon. ”We have these large administrative databases, with tens of millions of records. It is great in terms of statistical power, and it is enabling us to ask important questions in a whole host of realms, including continuity of care, post-acute care, end-of-life care, women’s health, drug prescribing patterns, comparative effectiveness of drugs, and others. Two members of our faculty just receive a grant to train MD PhD in health services research, using these large databases. It is exciting work, but I think we also have to have a healthy respect for the degree of bias that can exist in this data.”