The paper

Methods

Everything below was fixed in a plan registered before any comparison between arms was computed; the plan, its codebooks, both codings and the code as registered are under plan/, and the one addition since is listed in deviations.json.

Question. The VA atlas compared 215,970 veterans with diabetes who started a GLP-1 receptor agonist with 1,203,097 who continued their other treatment, over 175 outcomes, and reported associations with many outcomes beyond the drugs' established effects (Xie et al. (2025)). Placebo-controlled trials that posted results to ClinicalTrials.gov list every serious adverse event (one that results in death, is life-threatening, requires or prolongs hospitalization, or causes lasting disability) by arm and by MedDRA term, whether or not anyone published them; their non-serious events are listed only above a frequency threshold, so we used serious events alone. We asked what these trials show for each association the atlas named in its results (family A), for other questions asked of these drugs (family B), and, without hypotheses, for every organ system (family C) and preferred term (family D).

Data. We searched ClinicalTrials.gov's API (version 2.0.5, data as of 2026-10-09) on 2026-10-12 for interventional studies with posted results naming, among their interventions, any of 21 GLP-1 receptor agonists (data/molecules.json) or cagrilintide, which is tested in combination with semaglutide (data/search.json). The searches found 539 studies, whose records are in data/ctgov/ as the API returned them. The National Library of Medicine asserts no proprietary rights in the database.

Eligibility. The code keeps studies with randomized allocation, a parallel or factorial design, and at least two adverse-event groups (432 of 539). Two coders then read each of these trials' titles, conditions, arms, interventions and adverse-event groups, without any counts, and decided under a codebook which adverse-event groups compare a GLP-1 receptor agonist with placebo with nothing else different. Doses were pooled; a factorial or multi-background trial gave one comparison per background. Arms adding another active drug, active comparators, open-label untreated controls, and groups that were not randomized arms over the randomized period were left out. The coders also recorded the molecules and the population randomized. They agreed on eligibility for 426 trials (Cohen's kappa 0.97) and on identical comparisons for 227 of the 232 both accepted; a third reader decided the 14 trials on which they differed, with reasons. Tirzepatide, retatrutide, survodutide, cotadutide, mazdutide and efinopegdutide, which act on other receptors too, are pooled with the selective agonists and analysed apart in a sensitivity analysis.

Outcomes. Family A holds the 49 associations the atlas reported against usual care in its results, outside the cardiometabolic outcomes (myocardial infarction, cardiac arrest, heart failure, stroke and chronic kidney disease) that the large trials adjudicate as end points and often leave out of their adverse-event tables. Family B holds gallbladder and biliary disease, intestinal obstruction or ileus, ischaemic optic neuropathy, diabetic retinopathy complications, thyroid, pancreatic, obesity-associated and any cancer, depression, fractures, falls, death from any cause and any serious adverse event. Two coders assigned each of the 4,644 distinct serious event terms to the family A and B outcomes it meets, under a second codebook that defines each outcome, and to its MedDRA preferred term; they saw only the terms and the organ systems they were reported under. They gave identical outcome sets to 4,569 terms (kappa 0.976 over all term–outcome pairs), and a third reader decided 163 terms. All coders and readers were separate instances of Claude Opus 5.5, each working alone.

Counting. For each comparison and arm, the count for an outcome is the sum over its terms of participants affected, capped at the participants at risk; a participant with two of an outcome's terms is counted twice. Deaths come from the groups' death counts where every group in a comparison reports them; older records often don't. Organ systems are the reported MedDRA system organ classes.

Estimates. All intervals are 95% confidence intervals. The primary estimate is the Mantel–Haenszel risk ratio of the drug arm against placebo, stratified by comparison, with the variance of Greenland and Robins (1985) and a two-sided Wald test. Outcomes with fewer than 10 participants affected in all were not tested. Within each of families A and B, p-values were corrected by the method of Benjamini and Hochberg (1995). Each tested outcome was also estimated by random effects (DerSimonian and Laird (1986)), by Peto's odds ratio (Yusuf et al. (1985)), from its most frequent preferred term alone, with every single trial left out, among selective agonists only, and separately in type-2 diabetes and in overweight or obesity without diabetes. A conservative variance multiplies the Greenland–Robins variance by how far an outcome's count could overstate participants, computed blind to arm: the count over all comparisons divided by the sum over comparisons of its most frequent preferred term, capped at the factor by which summing all terms overstates participants with any serious event (2.06). The registered rule calls an outcome robust when q < 0.05 and p < 0.05 with the conservative variance, under random effects, and with every single trial left out. Trials were paired by size and one of each pair drawn into a first half by a seeded random generator.

Against the atlas. Each family A outcome is confirmed (q < 0.05 in the atlas's direction), opposite (q < 0.05 against it), excludes the atlas's estimate (not significant, but the 95% interval excludes the atlas's hazard ratio), inconclusive, or has too few events. An outcome is informative when a two-sided 5% test had 80% power to detect the atlas's hazard ratio, computed before registration from the pooled count of participants affected, blind to arm.

Scans. Families C and D hold every organ system with at least 10 participants affected and every preferred term with at least 20. Each item was estimated in the first half of the trials with Benjamini–Hochberg correction over its family (organ systems with the conservative variance), and replicates when q < 0.05 there and p < 0.05 in the second half in the same direction.

Checks of the data. Gallbladder and biliary disease and nausea and vomiting were registered as positive controls, effects the trials' own publications established.

Software. The analysis runs in Python 3.12 with NumPy 2.5.3 and SciPy 1.18.1 in under a minute, and a second script rounds its outputs for the paper. results/outcomes.csv lists every family A and B estimate, results/outcome_counts.csv every comparison's counts, and results/R1.json to results/R6.json everything the plan names.

Results

Trials. The search found 539 studies, of which 432 passed the structured filter and 234 compared a GLP-1 receptor agonist with placebo, in 257 comparisons with 161309 participants: 93765 on the drugs and 67544 on placebo, since many trials randomized several doses against one placebo arm. 109631 participants had type-2 diabetes and 42927 overweight or obesity without diabetes. The halves held 117 and 117 trials.

Positive controls. Serious gallbladder and biliary disease affected 967 participants on the drugs and 589 on placebo (risk ratio 1.38, 1.24 to 1.53, q < 0.001), and serious nausea and vomiting 200 and 83 (2.02, 1.54 to 2.64, q < 0.001). Both are robust by the registered rule.

Infections and the lung. Serious bacterial infections affected 2409 participants on the drugs and 2532 on placebo, in 115 comparisons with events (risk ratio 0.88, 0.83 to 0.93, q < 0.001). The estimate was 0.87 under random effects (p < 0.001, I-squared 0), had p 0.0011 with the conservative variance, and ranged from 0.87 to 0.89 with each trial left out, the largest p being < 0.001. It was 0.89 in the first half (p 0.0014) and 0.86 in the second (p < 0.001), 0.87 in type-2 diabetes (0.82 to 0.93) and 0.92 without diabetes (0.8 to 1.06; interaction p 0.50). Its most frequent preferred term, Urinary tract infection, gave 0.92 alone (p 0.25).

Serious pneumonia affected 1442 and 1568 participants (risk ratio 0.87, 0.81 to 0.93, q 0.0011): 0.86 under random effects (p < 0.001), p 0.0019 with the conservative variance, 0.85 to 0.9 with each trial left out (largest p 0.0077), 0.85 and 0.91 in the halves (p < 0.001 and 0.096), and 0.86 for the preferred term Pneumonia alone (p 0.0027). It was 0.89 in type-2 diabetes and 0.77 without diabetes (interaction p 0.11). Sepsis (0.87, 0.78 to 0.98, q 0.10), respiratory failure (0.83, 0.72 to 0.96, q 0.10) and chronic obstructive pulmonary disease (0.83, 0.71 to 0.98, q 0.12) pointed the same way without reaching significance after correction.

The scan. Of 27 organ systems, 5 were significant in the discovery half and 4 replicated, all lower on the drugs: cardiac disorders; infections and infestations ({{S.organ_systems.items.infections and infestations.discovery.rr}} in discovery, q {{S.organ_systems.items.infections and infestations.discovery_q}}; {{S.organ_systems.items.infections and infestations.replication.rr}} in replication, p {{S.organ_systems.items.infections and infestations.replication.p}}); metabolism and nutrition disorders; and respiratory, thoracic and mediastinal disorders. Hepatobiliary disorders, higher on the drugs in discovery ({{S.organ_systems.items.hepatobiliary disorders.discovery.rr}}), did not reach the threshold in replication (p {{S.organ_systems.items.hepatobiliary disorders.replication.p}}). Of 518 preferred terms, 2 were significant in discovery and both replicated: acute myocardial infarction ({{S.preferred_terms.items.Acute myocardial infarction.discovery.rr}} and {{S.preferred_terms.items.Acute myocardial infarction.replication.rr}}) and cellulitis (0.64, q 0.016; 0.75, p 0.037).

Against the atlas. Of the atlas's 49 associations, 5 had too few events in the trials (cannabis, stimulant and opioid use disorders, bulimia, and interstitial nephritis). Of the 44 tested, 3 were confirmed, 0 were opposite, 4 excluded the atlas's estimate, and 37 were inconclusive. Table 1 shows the informative outcomes and those whose interval excludes the atlas's estimate; results/outcomes.csv lists all.

Table 1. Family A outcomes with the power to test the atlas's estimate, and those whose interval excludes it: participants affected, risk ratio (confidence interval), Benjamini–Hochberg q, the atlas's hazard ratio, and the category.

Among the outcomes with the power to test the atlas, 2 were confirmed, 1 excluded and 2 inconclusive. Acute pancreatitis affected 229 participants on the drugs and 173 on placebo (p 0.24); under random effects the estimate was 1.04 (0.85 to 1.27). The associations with alcohol use disorder (0.74, 0.41 to 1.33), suicidality (0.86, 0.6 to 1.23), seizures (1.06, 0.81 to 1.39) and dementia (0.97, 0.64 to 1.47) had intervals that hold both no effect and the atlas's estimate. Kidney stones (1.21, 1.03 to 1.43) and hypotension (1.3, 1.04 to 1.62) were higher on the drugs, as the atlas found, without reaching significance after correction (q 0.10 and 0.10).

Death and serious events. Death from any cause affected 2556 of 69064 participants on the drugs and 2848 of 50491 on placebo (risk ratio 0.87, 0.83 to 0.92), and any serious adverse event 18190 and 16841 (0.96, 0.94 to 0.98); both are robust.

Other questions. Table 2 shows the other family B outcomes.

Table 2. Family B outcomes other than the positive control, death and any serious event: participants affected, risk ratio (confidence interval), and Benjamini–Hochberg q.

Thyroid cancer had p 0.023 before correction; under random effects it was 1.16 (p 0.46), and 1.16 and 2.55 in the halves. No family B outcome other than the positive control, death and any serious event reached significance after correction.

Discussion

The lower risk of serious infection agrees with what meta-analyses of published trials have found. Han et al. (2025) pooled 136 trials with 164,322 participants and reported a risk ratio of 0.89 (0.86 to 0.93) for serious infections, with lower risks for respiratory and skin infections; an umbrella review of 60 meta-analyses graded the association with serious infections as high-quality evidence, with an odds ratio of 0.89, and still called it a suggestion that requires further confirmation (Yang et al. (2026)); and Yu et al. (2023) found respiratory disease lower (0.86) across 28 trials. In SELECT, infection was the most frequent cause of death other than cardiovascular, and less frequent on semaglutide (hazard ratio 0.71) (Scirica et al. (2024)). Our estimates are close to all of these. What this study adds is the source and the safeguards: the complete serious-event tables of every eligible trial on the registry, which publications often omit or report differently (Tang et al. (2015)); trials and terms coded blind to the results and fixed in a registered plan; and a replication across random halves of the trials, in which serious infections as an organ system and cellulitis reappeared. The atlas's estimate for bacterial infections, from veterans on usual care, is the same as the trials', which suggests that the association in the cohort was not mainly confounding.

Why the risk is lower is not settled. Weight loss and better glycaemic control each lower infection risk, and Han et al. found larger reductions with more weight loss and HbA1c reduction; the drugs also have anti-inflammatory effects that could act directly. Infection outcomes that track obesity, such as cellulitis, fell most, which fits weight loss but does not separate it from the rest. A meta-regression on each trial's weight and HbA1c change, or individual participant data from the large outcome trials, would tell these apart. The reduction appeared in type-2 diabetes and, for pneumonia, at least as strongly without diabetes, where glycaemia changes little.

Most of the atlas's associations could not be tested. Substance use, psychosis, suicidality, seizures and dementia rarely reach the hospital in trials of a few years, and serious-event tables do not record outpatient diagnoses, so the trials' silence is not evidence against the atlas. The associations the trials could test fell into two groups. Infections and nausea and vomiting matched the atlas closely. Acute pancreatitis, arthritis and gastritis did not: the atlas reported a hazard ratio of 2.46 for drug-induced acute pancreatitis, while the trials exclude a risk ratio above 1.38 for acute pancreatitis of any cause. A diagnosis of drug-induced pancreatitis needs a clinician to attribute it to a drug, which is likelier in a patient known to take one carrying a pancreatitis warning; trials count every case on both arms alike. A meta-analysis of 62 published trials found a risk ratio of 1.44 (1.09 to 1.89) for pancreatitis, not significant once stratified by background medication (Wen et al. (2025)); our interval includes a modest increase but not one of the atlas's size. Arthritis and gastritis, slightly more frequent on the drugs in the atlas, were not more frequent in the trials. For deep vein thrombosis the interval lay above the atlas's estimate, but pulmonary embolism moved the other way and venous thromboembolism as a whole was 1.01, so we read it as chance among many tests.

For thyroid cancer, a French case-control study found a hazard ratio of 1.58 (1.27 to 1.95) after one to three years of use (Bezin et al. (2023)). The trials' estimate points the same way but rests on few cases, is not significant after correction or under random effects, and differs between the halves, so they neither confirm nor exclude an increase of that size. More thyroid imaging or examination on the drugs could raise detection in trials and cohorts alike.

For practice, a lower risk of serious infection, consistent across trials, populations and analyses, is a benefit to weigh beside the drugs' known harms, most of all for people at high risk of serious infection; and the trials find no increase in ileus, pancreatic cancer or cancer overall. For research, the registry already holds the randomized evidence to test observational findings about these drugs at scale, cheaply and without selection by publication; what it lacks is power for rare and outpatient outcomes. Settling those needs trials that record them, such as dementia or substance use, as adjudicated outcomes, and individual participant data that would allow time-to-event analysis.

Limitations

• Adverse-event terms are reported by investigators and not adjudicated, and trials differ in how they classify an event as serious, though both arms of a trial are classified alike.
• Some trials collected adverse events only while participants took the study drug. If more participants stopped the drug than placebo, as nausea leads them to, the drug arm would be observed for less time and show fewer events of every kind. Outcomes with no expected effect (cancer, falls, ileus, anaemia) sit near one, which argues against a large deficit of this kind but cannot exclude a small one, and some of the infection reduction could come from it.
• Counts sum terms, so a participant with two infection terms counts twice. The conservative variance allows for this, and the result for bacterial infections survived it, but its single most frequent term, urinary tract infection, was not significant alone; the finding rests on the composite. Pneumonia was significant on its own preferred term.
• Most participants had type-2 diabetes, and the trials enrolled people fit for trials, followed mostly for one to five years.
• The search covered ClinicalTrials.gov only, by drug name, and trials that never posted results, including many outside the United States and many older trials, are missing; results posting is required for many trials but not all, so selection by outcome is reduced, not removed.
• A cohort's hazard ratio and a trial's risk ratio are compared as if alike, which holds for rare events over similar follow-up. The atlas studied veterans with diabetes against usual care, so an interval that excludes its estimate does not show the atlas wrong for its population, only that the trials do not reproduce it.
• The coders, the third reader and the author are models of one family. They worked blind to the counts, but share whatever priors that family holds.
• Multi-receptor agonists are pooled with selective ones; analysing selective agonists alone gave the same infection estimates. Doses, molecules, durations and time to event were not analysed.

Provenance

A person asked for a new study outside mathematics, on a question that matters. The agent, a Claude-family model (claude-opus-5-5), proposed several, including this one, which the person chose; the person also approved registering the plan. The agent designed the study, searched ClinicalTrials.gov, wrote both codebooks, the plan and the code, and registered the plan. Separate instances of the same model screened the trials and coded the terms, two working independently and a third deciding where they differed. The agent then ran the analysis, wrote the claims and this paper, and applied the hazard screen. The data are the ClinicalTrials.gov records retrieved on 2026-10-12, carried unchanged in data/ctgov/, and the atlas's hazard ratios as its text states them. No person wrote, coded, analysed or reviewed any part of the work.