Show up at any clinic on a Tuesday morning. Just look at the waiting room. You’ll see a woman in her seventies juggling diabetes, hypertension, and a shingles flare that won’t quite settle. Next to her, a man in his fifties—second stent, metabolic syndrome, kidney function already on the slide. Over in the corner, a young adult whose asthma never calmed down after a viral illness two winters ago. Three medications deep, still reaching for a rescue inhaler more often than any guideline thinks reasonable.
Now flip open the landmark trial that supposedly justifies each of those treatments. The typical study cohort was younger. Leaner. Diagnostically spotless. The exclusion list reads like a fantasy: no more than one chronic condition, liver enzymes strictly inside the reference range, no more than three concurrent medications. This isn’t some minor methodological asterisk. It’s the uncomfortable, grinding truth of modern evidence-based medicine. The populations we study look almost nothing like the patients sitting in front of us.
The Architecture of Exclusion
Clinical trials chase a single, clean question. Less noise means a narrower population. Every eligibility criterion that tightens the sample—no recent hospitalizations, no cognitive impairment, BMI capped at 35, creatinine no higher than 1.5 mg/dL—slices away variability. It boosts the odds of finding a statistically sweet treatment effect. Methodologically, it’s tidy. Clinically, it’s a slow-motion train wreck.
Take the standard phase III cardiovascular outcomes trial. Average age hovers around 63. Meanwhile, the median age for a first myocardial infarction keeps climbing, and the real slog of post-infarction care lands hardest on people over 75. Excluding older adults is so baked into the process that a 2019 JAMA Internal Medicine analysis found more than half of cardiovascular trials explicitly shut out patients on age alone. Another big slice did it indirectly, using comorbidity criteria as a proxy. The evidence base tells you how a drug behaves in a 60-year-old with isolated hypertension. It says next to nothing about the 82-year-old with hypertension, atrial fibrillation, stage 3a chronic kidney disease, and intermittent confusion driven by a fistful of pills.
Comorbidity as a Contaminant
The logic of exclusion gets aggressive around comorbidity. A trial for a new biologic in rheumatoid arthritis will block anyone with a history of malignancy, serious infection, or significant cardiac disease. But the actual rheumatoid arthritis population is swimming in exactly those problems. Chronic inflammation speeds up atherosclerosis. Decades of corticosteroids push infection risk higher. By the time a patient has cycled through two disease-modifying antirheumatic drugs, the chance she also carries hypertension, depression, or COPD is not marginal. It’s the default.
When the drug hits the market, the label mirrors the trial population, not the people who will actually take it. The clinician is left squinting into the unknown. Will the biologic wake up that latent hepatitis B? Will it tip the heart failure that was an exclusion criterion in every phase II and III study? The package insert stays quiet. Post-marketing surveillance eventually patches some holes, but the process is sluggish, patchy, and leans on voluntary reporting systems that catch only a sliver of adverse events.
The Renal Function Blind Spot
Kidney function is a perfect little case study in how trial exclusions warp clinical practice. A 2020 review in Clinical Pharmacology & Therapeutics sifted through a decade of new drug approvals. Nearly two-thirds of trials explicitly excluded patients with moderate or severe renal impairment. Plenty excluded even mild impairment. Yet renal dysfunction isn’t some rare outlier in the populations that will swallow these drugs. It shows up in roughly 10% of the global population, and a much fatter slice of those with diabetes, heart failure, or hypertension.
The result? A pharmacopeia that, for an enormous chunk of patients, is basically experimental. Dosing recommendations for reduced renal function are often missing at launch. When they exist, they’re frequently built on small, single-dose pharmacokinetic studies in healthy volunteers, not on clinical outcomes. The nephrologist managing a patient with a GFR of 35 mL/min and a fresh indication for an anticoagulant is making calls with a level of uncertainty the original trial investigators never had to breathe.
Polypharmacy and the Interaction Void
Trials treat polypharmacy as a confounding variable to be scrubbed out. Typical protocols restrict concomitant medications with a heavy hand: no strong CYP3A4 inhibitors, no drugs that stretch the QT interval, no more than two antihypertensives, no recent corticosteroid bursts. The safety database captures drug–drug interactions poorly. Often not at all.
Real patients, especially older ones, regularly swallow five, ten, fifteen medications. The median number of chronic meds in a seventy-year-old with multiple conditions is not three. It’s nine. Each added drug multiplies interaction risks—not adds, multiplies. A statin that caused zero myopathy in the trial might do so very predictably when combined with a calcium channel blocker and a proton pump inhibitor that fiddle with its metabolism. An SSRI that seemed fine in isolation can produce serotonin syndrome when layered onto a triptan and an antiemetic with serotonergic teeth.
The absence of polypharmacy data isn’t a side note. It’s a systematic failure that dumps the burden of pharmacovigilance onto the prescriber, and then straight onto the patient.
The Demographic Distortion
Trial populations aren’t just medically narrower. They’re demographically skewed. Racial and ethnic minorities stay underrepresented across nearly every therapeutic area. A 2022 analysis of FDA drug approvals showed Black participants made up just 8% of trial populations, despite representing over 13% of the U.S. population and carrying a disproportionate share of many diseases under study. Hispanic and Indigenous representation was even thinner.
The pharmacogenomic fallout is real. Genetic polymorphisms that shape drug metabolism—CYP2D6 variants, HLA-B*5701, G6PD deficiency—shift sharply across ancestral populations. A drug that sails through a predominantly white European trial cohort might trigger Stevens-Johnson syndrome in Han Chinese patients, or hemolytic anemia in people of African or Mediterranean descent. The trials don’t flag these risks because they were never designed to look.
The Generalizability Gap in Mental Health Trials
Psychopharmacology trials serve up some of the worst offenders. The typical major depressive disorder trial excludes patients with any comorbid anxiety disorder, substance use disorder, or personality disorder. It tosses out anyone with suicidal ideation requiring hospitalization, anyone who’s already failed more than one antidepressant, and anyone with an unstable medical condition. Translation: it excludes almost every human who actually walks into a psychiatry clinic.
What’s left is a cohort of moderately depressed, medically healthy, highly motivated people with no complicating psychosocial mess. The remission rates reported—often 30–40%—have almost nothing to do with real-world practice, where sequential treatment failures, co-occurring anxiety, and the grinding demoralization of chronic illness reshape the whole picture. The chasm between trial efficacy and real-world effectiveness isn’t a footnote. It’s built into the walls of the evidence base.
Pragmatic Trials and the Fantasy of the Real World
The research community hasn’t been completely asleep. The rise of pragmatic trials, comparative effectiveness research, and real-world evidence initiatives is a grudging nod to the fact that the traditional RCT, for all its internal validity, bleeds external validity by the gallon. Pragmatic trials loosen the eligibility screws, allow flexible dosing, and measure outcomes that matter to patients and health systems rather than surrogate endpoints. They pull patients from community practices, not just academic centers with dedicated research coordinators and the luxury of frequent follow-up.
Still, pragmatic trials remain a sliver of total research output. They’re harder to fund, harder to place in high-impact journals addicted to the clean signal of explanatory trials, and harder to squeeze into a regulatory framework that still genuflects to the double-blind, placebo-controlled, single-disease model. The incentives are backwards: a sponsor chasing regulatory approval has every reason to design a trial that maxes out the chance of a positive result. That means cherry-picking a population where the drug is most likely to work and least likely to cause trouble.
What the Clinician Must Do
So what’s a thoughtful clinician supposed to do? First, read the supplementary appendix. The inclusion and exclusion criteria aren’t bureaucratic boilerplate. They’re a map of what the trial does not know. When the appendix shows patients with an ejection fraction below 30% were shut out, understand that the drug’s safety in advanced heart failure is a blank page. When the trial capped BMI at 30, recognize that the 40% of your patients with obesity never entered the evidence base.
Second, treat every new prescription in a complex patient as an n-of-1 experiment. Start low, go slow, watch like a hawk. The recommended starting dose in the package insert came from a population that looked nothing like the person in front of you.
Third, push for better from the research machinery. Support patient registries, feed data into post-marketing surveillance systems when you can, and accept that the most important safety data for a drug often surfaces years after launch—once a sufficiently broad, messy population has actually used it. The randomized trial is not the final word. It’s the opening statement.
Frequently Asked Questions
Why don’t researchers simply enroll more representative patients in clinical trials?
The incentives are stacked hard the other way. Heterogeneous populations pump up variability, which drains statistical power and makes a positive result harder to snag. Regulators, who worship internal validity, don’t demand representative enrollment as a condition of approval. Sponsors face brutal pressure to get drugs to market fast and cheap; recruiting older, sicker, more tangled patients is slower and costs more. Until regulatory standards shift to treat external validity as a core piece of trial quality, the gap will sit there, unmoved.
How can I tell if a trial’s results apply to my specific patient?
Start with the baseline characteristics table. Hold the trial population’s mean age, comorbidity profile, and medication load up against your patient. If your patient would have been blocked by even one major criterion, the trial’s estimates of benefit and harm don’t transfer cleanly. Subgroup analyses done after the fact? Handle with care—they’re usually underpowered and hypothesis-generating at best. Search for pragmatic trials or observational studies in populations closer to your patient’s profile, even if they come with a higher risk of confounding.
Are there any therapeutic areas where trial populations are more representative?
Oncology has nudged forward a bit, partly because the disease is so brutal and the history of effective treatments so thin that pressure built to enroll a broader range of patients. Some infectious disease trials, especially those in low-resource settings, have included more representative populations out of sheer necessity. But even here, excluding patients with organ dysfunction, poor performance status, or concurrent cancers stays common. No specialty has fixed this. Some have just admitted the problem more plainly.