Real-world drug studies reveal how chronic disease care diverges from clinical trials

Patients treated in routine care rarely resemble the volunteers who enroll in the trials that support regulatory approval. They are often older, more likely to live with several diseases at once, take many medicines together, and stay on treatment far longer than any trial follows them. A new editorial in Frontiers in Epidemiology argues that this gap is exactly where pharmacoepidemiology earns its keep: by describing how medicines are actually used in practice and estimating their real-world effectiveness and safety, the field informs prescribing, regulatory labeling, and health policy. Minji Sohn of Sahmyook University in Seoul frames a collection of 27 studies spanning observational research, pharmacovigilance, and health-policy analysis across the continuum of chronic disease care.

The scope of the collection is striking. The contributing studies draw on national insurance, administrative, and vital statistics databases from Taiwan, Sweden, Italy, China, and the United States; spontaneous reporting systems from the United States and Japan; hospital-based cohorts from Qatar, China, and Spain; and survey and dispensing data from Jordan, Colombia, and Uruguay. The call for papers solicited work on medicine effectiveness and safety, drug utilization, patient safety, post-marketing surveillance, pharmacoeconomics, and methodology, and the published articles capture each of these domains. Several also tackle questions that remain difficult to examine in routine care, including drug interactions, exposure–response relationships, genetic determinants of treatment response, and nonadherence. Sohn organizes the findings into five themes, each carrying lessons about how real-world evidence should be interpreted.

The first theme is safety signal detection in long-term pharmacotherapy, and it is dominated by a single data source. Nine articles used spontaneous adverse event reporting databases, all of them drawing on the US Food and Drug Administration Adverse Event Reporting System, known as FAERS, with one study extending the analysis to Japan’s Adverse Drug Event Report database, JADER. Sohn attributes this concentration to the scale and open accessibility of FAERS, but she also flags the cost: most of the signals described are conditioned on reporting behavior in a single regulatory setting. Because a disproportionality estimate reflects the reporting process as much as the drug itself, replication in JADER, EudraVigilance, or the World Health Organization’s VigiBase is one of the clearest ways to distinguish drug-related patterns from reporting artifacts.

Several of these studies examined immunomodulatory therapies that patients often take for years rather than weeks. Xiong and colleagues analyzed malignancy reports associated with disease-modifying anti-rheumatic drugs in rheumatoid arthritis and detected signals for a broader range of malignancies with targeted synthetic agents than with other DMARDs. Drug-specific median times to onset ranged from 0.83 to 1.67 years for targeted synthetic agents and from 3.58 to 7.08 years for conventional synthetic agents. Yu and colleagues focused on older adults receiving tumor necrosis factor-α inhibitors and found substantial heterogeneity among agents: etanercept showed the weakest and certolizumab pegol the strongest overall infection signals, while golimumab showed the strongest signal for tuberculosis and the highest reporting odds among patients with ankylosing spondylitis. Liu and colleagues assessed the interleukin-1 blockers anakinra and canakinumab and found profiles broadly consistent with existing labeling, with injection-site reactions more frequent for anakinra and gastrointestinal and respiratory events reported more often with canakinumab, occurring predominantly in children and adolescents. Neither drug showed a significant signal for tuberculosis.

Other studies applied the same approach to newer or less well-studied exposures. Zhang and colleagues characterized the post-marketing profile of onasemnogene abeparvovec in spinal muscular atrophy, finding that most events were reported early, particularly within the first 8 days after administration, and that thrombocytopenia and thrombotic microangiopathy were more frequently reported in patients weighing at least 8.5 kilograms. A second Zhang-led study combined FAERS with JADER to quantify reports of serotonin syndrome associated with the concomitant use of opioids and serotonergic antidepressants, a combination that matters in perioperative care. Zhu and colleagues paired a disproportionality analysis of drug-induced gingival overgrowth with network pharmacology and identified reporting signals involving bisphosphonates, in addition to the three drug classes traditionally implicated: calcium channel blockers, immunosuppressants, and antiepileptics. Chen and colleagues prespecified analyses of cardiovascular events reported with tirzepatide, stratified by reporter type, because consumer submissions accounted for 92.3 percent of all reports. In analyses restricted to reports from healthcare professionals, tirzepatide had lower reporting odds for heart failure than all other drugs in the concurrent FAERS comparator, with a reporting odds ratio of 0.18 and a 95 percent confidence interval of 0.07 to 0.43, and no reporting signal for atrial fibrillation or tachycardia. Elevated reporting odds for tachycardia appeared only in weight-stratified analyses of the full dataset.

Sohn is candid about what disproportionality analysis can and cannot do. It is rapid, draws on reports from broad post-marketing populations, and can detect signals for events too rare to characterize adequately in pre-approval trials. But it lacks a denominator, cannot estimate incidence, and remains highly sensitive to reporting behavior, notoriety bias, confounding by indication, and channeling bias. Studies based on a single spontaneous reporting system rarely provide definitive evidence on their own, yet they remain valuable when they address clinically important questions, probe rare or delayed events that are difficult to capture in trials, or incorporate design features that sharpen interpretation. Three articles showed how such limitations can be addressed through design rather than merely acknowledged. Chen and colleagues stratified by reporter type. Spadotto and colleagues tackled the long-standing question of whether ivabradine causes atrial fibrillation by combining a meta-analysis of randomized trials with a disproportionality analysis using beta-blockers as active comparators. In the randomized trials, the ratio of ischemic cerebrovascular events to atrial fibrillation was lower with ivabradine than with placebo, a pattern consistent with enhanced detection of atrial fibrillation, although a direct effect on incidence cannot be excluded; the authors present this detection explanation as a hypothesis requiring prospective evaluation. Tan and colleagues used a comparable dual strategy for calcitonin analogs in osteoporosis in older adults, pairing signal detection with a meta-analysis of randomized controlled trials that found no significant effect on fracture prevention.

The second thematic group relied on population-based observational data to examine associations between medicine exposure and clinical outcomes, where the definition and timing of exposure relative to follow-up can profoundly shape effect estimates. Li and colleagues drew on Taiwan’s National Health Insurance Database covering 2008 to 2022 to examine dipeptidyl peptidase-4 inhibitor use and diabetic retinopathy, finding a graded association across categories of cumulative defined daily doses relative to propensity score-matched sulfonylurea users. Yu and colleagues addressed the long-contested question of antihypertensive medicine use and cancer in 270,320 patients with newly diagnosed hypertension and found no clinically meaningful overall association; a marginal signal for calcium channel blockers did not persist in analyses that excluded early follow-up periods, illustrating how strongly such estimates depend on latency assumptions. Bérard and colleagues examined pregnancy outcomes associated with exposure to dihydroergotamine or triptans in 767,994 pregnancies and found that, after adjusting for maternal migraines, drug exposure was not significantly associated with preterm birth, low birth weight, major congenital malformations, or spontaneous abortion. The estimates based on 189 dihydroergotamine-exposed pregnancies were highly imprecise, and the authors regarded the observed association with septal defects as a chance finding requiring replication.

The third theme moved from individual medicines to prescription patterns and treatment trajectories. Amrouch and colleagues, on behalf of the AFFIRMO project, applied a reduced, operationalized set of the STOPP/START version 2 criteria to 203,042 older adults in Sweden with atrial fibrillation and at least one comorbidity, extending the evaluation beyond oral anticoagulants to the full medication regimen and examining associations with mortality, hospitalization, stroke, bleeding, and falls. Sukik and colleagues evaluated polypharmacy phenogroups across heart failure phenotypes in 4,902 patients in Qatar and found that major, but not excessive, polypharmacy was associated with lower mortality in patients with preserved or mildly reduced ejection fraction, with hazard ratios of 0.62 and 0.70 respectively; in these groups, major polypharmacy largely reflects guideline-directed medical therapy. Lin and colleagues found low uptake of several guideline-recommended treatments among 5,356 hospitalized patients with heart failure: angiotensin receptor-neprilysin inhibitors were used in 12.7 percent of patients, sodium-glucose cotransporter 2 inhibitors in 8.1 percent, and quadruple therapy in 3.75 percent, with several of these agents also associated with lower odds of readmission. Shen and colleagues traced stroke prevention prescriptions among 13,072 patients with coexisting atrial fibrillation and chronic obstructive pulmonary disease in Taiwan from 2013 to 2022, documenting a marked shift from warfarin, which fell from 24.0 percent to 6.1 percent, to non-vitamin K antagonist oral anticoagulants, which rose from 15.4 percent to 65.4 percent; the 1-year incidence rates of ischemic stroke and major bleeding also declined over the same period, though the study reports concurrent trends rather than an effect of the prescription shift on either outcome. Casula and colleagues found that 41.14 percent of 16,025 patients in Lombardy received no lipid-lowering therapy within 3 months of an atherosclerotic cardiovascular event, with older age, female sex, and polypharmacy associated with non-initiation, and Campos and colleagues evaluated ixekizumab persistence in axial spondyloarthritis at ten Spanish hospitals, finding that 56.3 percent of patients remained on treatment at 52 weeks, with primary nonresponse the most frequently reported reason for discontinuation, followed by adverse events and secondary loss of response.

The fourth theme shifted attention to the structural conditions that shape medicine use across health systems. Tang and colleagues used an interrupted time-series analysis to evaluate China’s National Reimbursement Drug List negotiation for evolocumab and alirocumab, documenting improved availability, changes in defined daily doses, lower per-dose costs, and a narrower gap between tertiary and secondary hospitals. Cabral and colleagues described antipsychotic dispensing among children and adolescents in Montevideo from 2018 to 2022, reporting increasing use over time with higher utilization among providers serving populations of lower socioeconomic status. Abed and colleagues conducted a mixed-methods study of 405 Jordanian outpatients with multimorbidity and identified cost, polypharmacy, forgetfulness, and rural residence as barriers to adherence, together with six qualitative themes including economic strain, religious coping, and caregiving burden. Cardona and colleagues analyzed 42,601 pharmacotherapeutic follow-up records of older adults in Colombia and found that inappropriate medicine use and administrative inefficiencies were prominent correlates of nonadherence, framing adherence at the system level rather than as individual responsibility. Sohn argues that medication-taking behavior should not be simplified into an individual patient trait but conceptualized across initiation, implementation, and discontinuation phases, with nonadherence understood as a product of the structural conditions under which treatment is delivered.

The collection closes with three articles extending beyond conventional designs: Zhao and colleagues investigated pharmacogenomic and epigenetic determinants of leflunomide response in 240 patients with rheumatoid arthritis, reporting an interaction between a PON1 promoter methylation site and the rs705379 variant that may provide a basis for investigating heterogeneity in treatment response; Yang and colleagues evaluated the cost-effectiveness of commercial Chinese polyherbal preparations for primary insomnia through a network meta-analysis; and Liang and colleagues conducted a bibliometric mapping of two decades of research on traditional Chinese medicine, gut microbiota, and diabetes. What unifies these disparate contributions, Sohn writes, is that each depends on a transparent account of how the data were generated. Her broader conclusion is that the value of real-world evidence does not depend on study scale alone but on whether the design accurately captures the treatment trajectory and the data-generating process. Triangulation enhances causal inference only when the underlying data sources are subject to biases operating in orthogonal or opposing directions, and design choices such as excluding early follow-up or stratifying by reporter type can alter estimates rather than simply refine them. With reporting guidance now available for both disproportionality analyses and studies using routinely collected health data, and with tools such as target trial emulation, negative control outcomes, and preregistered analysis plans increasingly accessible, she argues that explicit design specification should be regarded as part of the expected baseline standard of evidence in chronic disease pharmacoepidemiology.

Subject of Research: Pharmacoepidemiology of medicine use, safety, and treatment trajectories in chronic diseases across real-world data sources

Article Title: Editorial: Pharmacoepidemiology in chronic diseases

Article References: Sohn, M. (2026). Editorial: Pharmacoepidemiology in chronic diseases. Frontiers in Epidemiology, 6, 1963125. https://doi.org/10.3389/fepid.2026.1963125

Image Credits: AI Generated

DOI: 10.3389/fepid.2026.1963125

Keywords: pharmacoepidemiology, chronic disease, drug safety, pharmacovigilance, FAERS, real-world evidence, polypharmacy, drug utilization, medication adherence, observational studies, health policy, disproportionality analysis