Manni Mohyuddin: About the importance of intent to treat analysis

Manni Mohyuddin, Assistant Professor at the Huntsman Cancer Institute, University of Utah, shared a post on X/Twitter:

“Inspired by a recent discussion on twitter/X about a trial- here’s a thread about the importance of intent to treat analysis, and the caveats/shortcomings of per-protocol or analysis per “compliance/adherence”. Buckle yourself up for an educational

This was a trial of Mediterranean diet in reducing breast cancer recurrence. We know the Mediterranean diet is good for you- in a rigorous randomized trial, it has actually reduced cardiovascular morbidity.

But this is a different question altogether- can it reduce breast cancer recurrence? This trial builds upon some previous large well-done randomized trials that showed zilch effect of dietary interventions in cancer.

Eating more veggies did not reduce prostate cancer progression.Article website

Low fat high fiber diet did not reduce recurrence of colorectal adenomas.Article website

A diet high in vegetables, fruit and fiber did not reduce breast cancer recurrence or death. Additional details

Here’s the link to this trial:
Mediterranean diet to reduce risk of breast cancer recurrence:

So essentially, this trial was all set to follow a legacy of failure of dietary interventions to change the natural history of cancer. And on an intent to treat analysis, this trial failed. At 5 yrs f/up, virtually identical number of events, HR=0.99, as null as it gets.

Why is an intent to treat analysis so important? One of the key benefits of randomization is that all measured and unmeasured confounders are adjusted for. This is the problem with observational studies (especially dietary ones)

If you look at dietary epi studies- let’s say organic food- those who eat organic food have better outcomes. But that’s because only rich people with access and in good underlying health eat organic food. It isn’t eating organic food that makes them healthy.

In any retrospective dietary epi study that compares eaters of organic food to non-eaters, there is no possible way all confounders can be adjusted for. Even if the investigators say they did. You can only adjust for the confounders you know of and you can measure!

So many confounders simply cannot be measured. Can you truly and fully measure health literacy, access to healthcare, socioeconomic status Hence nutritional epidemiology research is often fatally flawed- no matter what fancy stats they do to adjust Going back to intent to treat

On an intent to treat analysis, you analyze all patients randomized to an intervention. Regardless of whether they actually followed the intervention per protocol or not. This obviously allows for best generalizability of results- but also-

If you start focusing only on those who followed the interventions religiously and you discard those who didn’t- you essentially now are undoing the benefit of randomization in removing confounders.

For example- in this study- they did an analysis of compliance to dietary interventions and compared outcomes according to compliance levels. To assume that the only difference between these groups is how they eat is false.

People who are more compliant with dietary interventions within a trial are those who have better access/better health literacy/more health seeking behaviors/better health to allow changes to lifestyle etc.
By doing such a comparison, you’ve undone the key benefit of a RCT

Which is adjusting for all confounders (measured and unmeasured)
So such an analysis does indeed render a randomized controlled trial into a confounded observational study!

Another key flaw in this figure pointed out by David Russler-Germain is the y axis (take a look) They’ve magnified it so much between 0.95 to 1.00 just to magnify the difference in effect size. This is a common technique to amplify marginal differences and should not be allowed!

My intention in writing this thread is to educate; not criticize people or individuals.
There are compelling reasons to eat healthy- cardiovascular reasons etc. Eating more plants is better for the world too.
But important to adhere to rigorous scientific principles!

There is no rigorous data to support any specific diet for patients with cancer.
As people live longer due to better cancer therapies, prevention from other causes of death becomes even more important.
Diet is important.
But interrogating the evidence like this isn’t the way”

Read further.

Soure: Manni Mohyuddin/X