Example of current #scicomm around observational vs prospective research

treatment comparison observational
1

This segment from yesterday’s PBS News Hour makes an interesting study in science communication around issues of observational data research and causal mediation of effects.

William Brangham opened with a caution about the observational nature of the research, phrased in a way that [it seems to me] would be easily understood by the public:

The results in these cases come from what are known as observational studies, not more rigorous clinical trials, and researchers say there are still many open questions.

Interestingly he then immediately broaches the topic of effect mediation, again in what seems like an accessible way:

That said, this was the hot topic at a recent conference of America’s top cancer doctors, where a number of observations all pointed in the same direction, that GLP-1s appeared to help fight cancer above and beyond the benefits that you would expect from weight loss alone.

In response to this probe,

… Again, do you believe that this is principally a function of weight loss being the real actor here?

Dr. Neil Iyengar of Winship Emory responds,

…
And it does seem that it’s primarily through the weight loss function. We also know that GLP-1s have some anti-inflammatory effects as well. And we’re learning about some possible immune-related effects as well. But I think it’s really driven through the large amounts of weight loss that these drugs can induce, as opposed to prior or other diabetes drugs.

A couple of small gripes are that Iyengar fell back on “associated with” language where more forthright causal statements might have been preferable, and also (perhaps relatedly?) that absolute magnitudes were mostly left out. Observe how vague his statement is here:

We know that one in seven male cancer-related deaths and one in six female cancer-related deaths are related to obesity. If we can reduce the obesity problem, which we know we can do with the GLP-1s, this really stands to remarkably shift the global burden of obesity-related cancers.

If “related to” here really means attributable to, then the latter language would have been preferable, IMO. (What do you think, @Pavlos_Msaouel ?)

The segment closes with Iyengar very helpfully exposing his clinical reasoning in the face of uncertainty from the lack of prospective studies.

3 Likes
2

Strongly believe this is a topic where causal understanding by carefully dissecting biology will come in handy. On it :microscope:

1 Like
3

The timelines are too long between intervention and cancer outcomes for a randomized trial so the next best option is to go for observational data that are likely retrospective (for the same timeline reason). Now we need to decide what is to be done and if a target trial emulation from retrospective observational data will help as Frank has discussed here. Even if one decides that a target trial emulation is to be done (we attempted this for bariatric surgery and cancer) there is still a debate on what is the best way forwards to convert emulation failure to emulation success (See our response to Hernan’s methodological guidance here).

2 Likes
4

Great interview, David- thanks for flagging it! To Suhail’s point above- no single GLP-1 agonist RCT is going to be long enough or large enough to accrue enough cancer events for a definitive conclusion re causality. But this drug class is being studied so intensively and this is such a popular class of drugs (way more popular than bariatric surgery ever was or ever will be) that a future meta-analysis of GLP-1 agonist trials (perhaps with longterm followup of subjects) would be a great option. Hopefully sponsors are designing these trials in such a way that they will be amenable to future meta-analyses. It would be particularly nice if different sponsors could work with each other to start planning this now.

3 Likes
5

Thanks for highlighting that post of Frank’s, Suhail. (It seems to me John Snow used Frank’s Design 4 to good effect BTW, but maybe that’s for a different thread.)

1 Like
6

I wonder what this would involve. Would it require mainly a platform trial with uniform protocol? Or would there be other considerations?

7

An answer to this question is beyond my pay grade. But lots of RCT meta-analyses have been done for other commonly-used medications e.g., statins, SGLT-2 inhibitors, so there is likely a fair precedent…The main difference in the case of a GLP-1 agonist meta-analysis would be that the outcomes that are being meta-analyzed (incident cancer diagnoses) would not be the primary outcome the trials were designed to assess.

The interviewee in the video you linked made a good point about use of these medications in patients who already have a diagnosis of cancer. Recommending use of a GLP-1 agonist during therapy for cancer wouldn’t be a good idea, since we don’t know how these medications might interact with cancer therapies. But, as he mentioned, perhaps there’s a role for studying them in patients with a high BMI who have completed their treatments, to compare future recurrence rates among those treated and not treated with a GLP-1 agonist (?) RCTs involving cancer survivors would effectively be enriched as these patients would (unfortunately) be at higher risk than those in the non-cancer-survivor population to experience the outcome of interest (perhaps generating an answer more quickly). But again, I’m really not qualified to speculate on all this, since I’m not an oncologist. One important caveat I can think of: clinicians would need to remain vigilant so as to not confuse GLP-1 agonist-induced weight loss with cancer recurrence-related weight loss…

2 Likes