Healthy-user bias: the people who take the pill are already different

Suppose an observational study finds that people who take a certain vitamin, or eat a certain food, or keep up with a preventive medicine, have fewer heart attacks. Tempting conclusion: the vitamin works. But there is a quieter explanation. The kind of person who reliably takes a daily vitamin is also, on average, the kind of person who exercises, sees a doctor, sleeps enough, and does not smoke. This is healthy-user bias.

The people who adhere to any health behaviour tend to differ from those who do not in dozens of ways, many of them hard to measure and all of them tied to better health. So when the adherent group turns out healthier, you genuinely cannot tell how much of that was the treatment and how much was simply the sort of person who takes it. The intervention gets credit that partly belongs to the person.

There is an even sharper version, the healthy-adherer effect. In trials, people who take their assigned pills reliably do better than people who skip them, even when the pill is a placebo. Sticking to a routine is itself a marker of the habits and circumstances that keep people healthy. So comparing people who took the drug with people who did not is not comparing like with like.

This is one of the main reasons observational studies of preventive habits should be read with caution, and why randomized trials matter so much. Random assignment scatters those hidden differences evenly between the groups, so a difference in outcome can be pinned on the treatment. When a headline says people who do a healthy thing live longer, ask whether they were randomly assigned to do it, or simply chose to, because choosing it usually comes bundled with a hundred other advantages.