Time to rename target trials, 'observational studies'

It's easier to predict the winner of yesterday's world cup game than tomorrow's.

I kept an open mind with target trials. I thought the logic made sense and their design was elegant. Yet recent high profile failures should make us all reconsider target trials.

On GLP 1s and Alzheimer's: target trials claimed a benefit. But massive 3000+ person randomized data is negative.

On Tamiflu: in a target trial, among hospitalized and severely ill adults with flu, there was a huge mortality benefit. Yet, a randomized study failed to find benefit in this population. It trends towards increased deaths. (For this essay no need to debate useless vs harmful— it is sufficient that it doesn't help)

On paxlovid. The same thing. It works in a target trial but not in an actual trial.

On GLP1s and early cancer, same pattern. Early benefits in a target trial not present in randomized data. Strange results for thyroid cancer.

On covid boosters, a target trial showed preposterous results. That all cause mortality was reduced even before covid associated CV events were. The booster traveled back in time to cure cancer and diabetes and heart disease. #TheScience

Then there are the results of RCT duplicate- a pre specified comparison of target to randomized trials. Note that this is different than the more recent examples. In this case the authors already knew the randomized results they were aiming for. It's easier to predict the winner of yesterday's world cup game than tomorrow's.

Let me be clear: there are many clinical questions where you don't need a randomized trial. If I make a cure for late stage Alzheimer's, we would accept an uncontrolled trial of 3 people before licensure. A drug that takes a fatal disease’s well understood biomarker and normalizes it, will get approval without a randomized trial. Probably n of 12 people.

But there are situations where you want a target trial. All of these fit. Do glp-1 reduce Alzheimer's? This is about averting or delaying a bad outcome and many people aren't destined to experience the event. Does Tamiflu lower mortality from 10 to 9%. Here you want a controlled comparison. The effect size, if present, is too small for historical or external benchmarks.

And recent studies show that if you want a target trial, you need a randomized trial. The former makes false promises.

Why are we seeing so many target trials lately? Because researchers want to pad their CV. Because people doing them are ignorant to their shortcomings— proof is imagine publishing a covid booster target trial and posting that they reduce all cause mortality in 10 days and not thinking that invalidates your paper.

Because it's the cheap and lazy thing to do. Rcts take years and money. Target trials take a statistician and middle authors who didn’t help.

Will AI be able to fix target trials? Maybe— I have some ideas for how to do so. But proof will require predicting the results of ongoing trials.

What if the target trials were right and the rcts were wrong? And what if this is all just a dream? What if we are really living in the Matrix? What if…

Science magazine has a story lamenting that medical students are pumping out low quality observational studies. How dare they! You have to spend a decade in training to become a faculty member working hard to pump out low quality target trials.