SAN DIEGO – Even if a P value hints at statistical significance by dipping under .05, it might not tell you anything worthwhile. Effect sizes are hugely important – as long as accompanying P values measure up. And pharmaceutical companies often keep revealing numbers under wraps unless you know what – and whom – to ask.
Those lessons come courtesy of, who spoke to colleagues about study numbers at Psych Congress 2019.
Dr. Citrome, clinical professor of psychiatry and behavioral sciences at New York Medical College, Valhalla, offered several tips about interpreting medical statistics as you make clinical decisions.
Don’t get hung up on the P value.
The P value helps you understand how likely it is that a difference in a study is statistically significant. In medical research, P values under .05 are considered especially desirable. They suggest that an outcome – drug A performed better than drug B, for example – didn’t happen purely by chance.
Here’s the hitch: The P value might not matter at all. “Clinicians often assume that if the P value is less than.05, the result must be important. But even a P value of less than .05 is meaningless outside of the context of how big the treatment effect is,” Dr. Citrome said. “If a clinical trial result shows us a small effect size, then who cares?”
Understand what effect sizes tell you.
Effect size measurements evaluate clinical impact and include number needed to treat (NNT) and number needed to harm (NNH). NNT refers to the number of patients needed to treat with an intervention in order to get a positive effect in one additional patient; NNH is the reverse and examines negative effects that can range from the minor (mild dry mouth) to the devastating (death).
What’s a good size for an NNT? “I respect any NNT versus placebo of less than 10,” Dr. Citrome said. “It’s something I’ll probably consider in day-to-day practice.” Double-digit and triple-digit NNTs “are usually irrelevant unless we’re dealing with very specific outcomes that have long-term consequences.”
As for the opposite side of the picture – NNH – values higher than 10 are ideal.
He cautioned that NNT and NNH, like P values, cannot stand alone. In fact, they work together. In order to have value, NNT or NNH must be statistically significant, and P values provide this crucial insight.
Consider Dr. Citrome’s blood pressure.
As Dr. Citrome noted, research suggests that, among patients with diastolic BP from 90 to 109 mm Hg, 1 additional person will avoid death, stroke, or heart attack for every 141 people who take an antihypertensive medication, compared to those who do not, over a 5-year period. That’s a lot of people taking medication for a long time, with potential side effects, for a fairly small effect size. However, the outcomes are dire, so it is still worth it.
Then there’s Dr. Citrome himself, who has had diastolic BP in the range of 115 to 129 mm Hg. The NNT is 3. For every 3 people who take an antihypertensive vs. not over a 5-year period, 1 additional person will avoid a potentially catastrophic cardiovascular event.
“Guess who’s pretty adherent to taking his antihypertensive medication?” he asked. “I am.”
Ask for effect sizes if you don’t see them.
It’s not unusual for pharmaceutical representatives to avoid providing information about medication effect sizes. “The sales representatives as well as speakers at [Food and Drug Administration]–regulated promotional speaking events can only speak to basically what’s on the label. NNT and NNH are not currently found on product labels. But they are very relevant, and we need to know this information.”
What to do? “There is a workaround here,” he said. “Ask the sales rep to talk to a medical science liaison, who is free to come to your office and talk about all the data that they have.”
Dr. Citrome reported multiple disclosures, including various relationships with pharmaceutical companies.