What is target trial emulation?
Writing down the randomised trial you wish you could run, then making an observational analysis copy it as closely as possible: who is eligible, exactly what the treatment is, and a time zero when everyone’s clock starts. Analysed this way, the Nurses’ Health Study data gave answers on hormone therapy much closer to the trial’s.
Last time's quick check: the non-Bt fields benefited from their neighbours' Bt maize, so they weren't a clean untreated comparison. In the 1980s, observational studies kept finding the same thing: people who ate more fruit and vegetables rich in beta-carotene, or had more of it in their blood, got less lung cancer.
It looked like a cheap way to prevent cancer. So two large trials gave beta-carotene supplements to people at high risk: one in Finland, with about 29,000 male smokers, reported in 1994; one in the United States, reported in 1996. Both found more lung cancer in the supplement groups, not less: about 18 per cent more in Finland, and 28 per cent more in the US trial, which was stopped early.
What went wrong? People who eat plenty of vegetables differ in many ways from people who don't. And a nutrient inside a whole food isn't the same as a high-dose pill. Remember Lind's question: what exactly is "it"? Hormone replacement therapy tells a similar story. For years, observational studies, including the large Nurses' Health Study, suggested that women taking it had less heart disease. Then, in 2002, the Women's Health Initiative trial found a higher risk of heart disease, especially in the first years.
Part of the explanation was healthy-user bias: women who chose hormone therapy tended to be healthier, wealthier and more health-conscious in ways the studies hadn't fully captured. Part was timing: the observational studies mostly saw women who had already been taking it for years, and missed the early risk.
Then something remarkable happened. Miguel Hernán and colleagues went back to the Nurses' Health Study data and analysed it as if it were a trial. They defined who would have been eligible, and the moment each woman started therapy or didn't, called time zero. Then they compared new users with non-users from that moment, just as the trial had.
The observational data then gave answers much closer to the trial's. The approach is called target trial emulation: write down the trial you wish you could run, then make the observational analysis copy it as closely as possible. Most questions about diet, farming practice and health will never get a long randomised trial.
Target trial emulation is how careful researchers use the data they have without fooling themselves: define the treatment exactly, start the clock at the same moment for everyone, and compare like with like from there. That completes the toolbox. Next, in Module 4: the machine-learning methods, in depth.
People who eat plenty of vegetables differ in many ways from people who don’t, and a nutrient inside a whole food isn’t the same as a high-dose pill. When two large trials gave beta-carotene supplements to people at high risk, both found more lung cancer in the supplement groups, not less.
Bias that arises when the people who choose a treatment are healthier to begin with. Women who chose hormone therapy tended to be healthier, wealthier and more health-conscious in ways the studies hadn’t fully captured, so observational data suggested less heart disease. In 2002 the Women’s Health Initiative trial found a higher risk.
Causal evidence is changing how farming is measured and rewarded. Leave your email and we will tell you when new films, courses and findings are published.
Prefer to read online? Vitagri also writes on Substack. See our privacy policy for how we use your email.