Chapter 10 — Further Reading

Tier 1 — Verified canonical

ClinicalTrials.gov — the single most useful resource for this chapter. Search "semaglutide" and sort by condition. The breadth of what is being tested is the chapter's opening observation, and the status column — recruiting, active, completed, results posted — is the discipline. A great many of the indications generating coverage correspond to trials that are ongoing rather than reported.

DailyMed and Drugs@FDA — approved indications for the semaglutide and tirzepatide products, including the cardiovascular risk reduction indication and the obstructive sleep apnea indication. An approved indication is the clearest possible statement of what a regulator concluded the evidence supported, and comparing the approved list to the list of things being discussed is instructive.

PubMed — published reports of SELECT, FLOW, the MASH trials, the HFpEF trials, and the sleep apnea trials.

On early stopping. The methodological literature on truncated trials — comparing effect estimates from trials stopped early for benefit against completed trials and subsequent meta-analyses — is well established and not controversial among trialists. CONSORT requires reporting of interim analyses and stopping rules, and the DAMOCLES guidance addresses data monitoring committee practice. Both are public.

On MASLD/MASH nomenclature. The change from NAFLD/NASH to MASLD/MASH was a formal multi-society consensus. Both terminologies appear in the literature and will for some years.

On observational study biases. Standard epidemiology texts cover confounding by indication, the healthy adherer effect, protopathic bias, and immortal time bias. Rothman's Epidemiology: An Introduction is short and unusually clear. The STROBE statement is the reporting standard for observational studies and its checklist is a good way to see what such a study should disclose.

Tier 2 — Attributed, specifics unverified

On FLOW. A randomized, double-blind, placebo-controlled trial of semaglutide in adults with type 2 diabetes and chronic kidney disease, with a composite kidney endpoint, stopped early for efficacy on the recommendation of its independent data monitoring committee. Manufacturer-sponsored. This book states that it was stopped early and does not quote a hazard ratio, deliberately — the effect magnitude from a truncated trial should be held loosely.

On MASH trials. Randomized trials of semaglutide in MASH have reported improvement in histological measures. Long-term outcome data — progression to cirrhosis, liver-related mortality — has not been demonstrated for this class as of this writing.

On HFpEF. Trials of semaglutide in adults with HFpEF and obesity have reported improvement in symptom and physical-limitation scores and in exercise capacity. Primary endpoints were symptom and function measures rather than hospitalization or mortality.

On obstructive sleep apnea. Trials of tirzepatide in adults with moderate-to-severe obstructive sleep apnea and obesity reported substantial reductions in the apnea-hypopnea index, supporting an approval for that indication.

On the dementia epidemiology. Analyses of large healthcare databases have reported lower dementia incidence among people with type 2 diabetes prescribed GLP-1 receptor agonists compared with other diabetes therapies. These are observational and subject to confounding by indication, healthy adherer effects, protopathic bias, surveillance effects, and potentially immortal time bias. Trials in early Alzheimer's disease have been conducted; results should be checked against current sources.

On addiction. Patient reports of reduced alcohol consumption and smoking on GLP-1 receptor agonists are consistent and unprompted. Randomized trials in alcohol use disorder and smoking cessation are ongoing as of this writing; some smaller studies have reported signals and some have not.

On inflammation. GLP-1 receptor agonists reduce inflammatory markers including C-reactive protein. Whether this mediates any of the clinical benefits is not established, and weight loss itself reduces inflammation, which makes the mediation question difficult.

On the historical record of observational drug-benefit findings. That observational analyses have repeatedly supported benefits which subsequent randomized trials did not confirm — and in some prominent cases contradicted — is well documented across hormone therapy, vitamin supplementation, and several drug classes. Specific episodes are not itemized here.

Tier 3 — Illustrative and constructed

• The four-explanations diagram, the biopsy-endpoint comparison, the ten-claim rating table, the compression-error framing, and the five-bias diagram in Case Study 2 are this book's own teaching devices.
• The worked selection mechanism in Case Study 1 (the drug with a true 20% effect whose interim looks scatter to 31%) is a constructed illustration of a real statistical phenomenon. The numbers are invented to demonstrate the logic and describe no actual trial.
• Figure 10.3 renders a real published trial in this book's six-field format. Figure 10.CS1 renders a body of methodological literature rather than a single study, and says so.
• The dossier's split-by-indication worksheet is this book's own construction.

If you only do one thing

Search ClinicalTrials.gov for "semaglutide" and read the status column.

Sort or filter by condition and note how many distinct conditions appear. Then, for the three that most surprise you, check whether the trial is recruiting, active, completed, or has results posted.

The gap between "there is a trial" and "there is a result" is the entire content of §10.7 and §10.8, and seeing it as a column in a database is more convincing than any argument in this chapter.

Looking ahead

Chapter 11 covers insulin — the drug against which everything in this chapter should be measured, and the one with a century of evidence behind it.

Useful preparation: none required. But if you want a contrast worth holding, note that insulin's evidence base was built without any of the apparatus this chapter relies on. There were no randomized trials in 1922, no data monitoring committees, no registries, and no confidence intervals. The drug worked so unmistakably that none of it was needed — which is a situation that essentially never recurs, and understanding why is most of Chapter 11.