Healthy User Bias

Healthy user bias—closely paired with the healthy adherer effect—is a systematic confounding distortion in observational epidemiology and pharmacoepidemiology whereby individuals who voluntarily initiate a preventive medication, dietary supplement, or screening regimen (or who maintain persistent adherence to a prescribed long-term therapy) systematically differ from non-users in baseline health consciousness, socioeconomic status, functional capacity, and unmeasured health-promoting behaviors. Characterized by Jerry Avorn, Sebastian Schneeweiss, M. Alan Brookhart, and Kenneth J. Rothman, healthy user bias acts in the opposite direction of confounding by indication: whereas curative treatments prescribed for acute symptomatic illness are channeled toward sicker patients (making the drug appear harmful), primary preventive interventions are preferentially sought out and sustained by health-conscious, functionally intact individuals (making the preventive exposure appear artificially protective against a wide spectrum of clinical outcomes).

In a randomized controlled trial (RCT), random allocation balances baseline health behaviors, socioeconomic resources, and physiological frailty between the active treatment arm and the control arm. In routine clinical practice and observational cohort studies, however, the initiation and continuation of a preventive therapy (A = 1)—such as postmenopausal hormone therapy, statin lipid-lowering therapy, multivitamin supplementation, or seasonal influenza vaccination—is governed by selective patient and clinician behaviors.

Within a Directed Acyclic Graph (DAG), an unmeasured or imperfectly measured latent construct termed health-seeking propensity and functional robustness (U) acts as a common parent cause of both preventive therapy uptake (A) and favorable clinical outcomes (Y, such as reduced myocardial infarction, fracture, sepsis, or all-cause mortality): A ← U → Y. Empirically, the latent construct U encompasses four domains that are poorly captured in administrative healthcare claims databases:

1. Health-Promoting Lifestyle Behaviors: Patients who proactively request or adhere to preventive prescriptions exhibit lower rates of cigarette smoking and heavy alcohol consumption, regular aerobic exercise, lower visceral adiposity, and nutrient-dense diets.
2. Preventive Healthcare Utilization: Adherent preventive users simultaneously undergo regular cancer screening (mammography, colonoscopy, cervical cytology), dental care, blood pressure monitoring, and adult immunizations.
3. Socioeconomic Advantage and Health Literacy: Higher educational attainment, household income, and prescription insurance coverage facilitate both medication adherence and lower baseline cardiovascular and all-cause mortality.
4. Selective Prescribing Away from Frailty ("Sick Stopper" Depletion): Clinicians routinely withhold or deprescribe primary preventive drugs (such as statins or bisphosphonates) in patients experiencing cognitive decline, cachexia, metastatic malignancy, or end-stage organ failure whose limited life expectancy precludes long-term preventive benefit.

When an observational study compares preventive users (A = 1) against non-users (A = 0) using multivariable regression adjusted only for recorded diagnostic codes and age, the open backdoor pathway A ← U → Y produces a spurious protective hazard ratio (HR < 1.0), often compounded by prevalent user bias when ongoing long-term users are enrolled.

The clearest empirical proof that medication-taking behavior is confounded with baseline mortality risk comes from analyzing adherence to inert placebo pills inside double-blind randomized controlled trials. Because a lactose placebo tablet possesses zero pharmacological activity, any mortality difference between trial participants who consistently take ≥80% of their assigned placebo tablets ("good placebo adherers") and those who take <80% ("poor placebo adherers") reflects behavioral and functional confounding:

• Coronary Drug Project (1980): In the randomized Coronary Drug Project evaluating clofibrate for secondary prevention of myocardial infarction, men in the active clofibrate arm who adhered to ≥80% of their capsules had a 5-year mortality rate of 15.0%, compared with 24.6% among poor adherers (p < 0.001). However, in the placebo arm, patients who adhered to ≥80% of their inert placebo capsules had an almost identical 5-year mortality rate of 15.1%, compared with 28.3% among poor placebo adherers (p < 10⁻¹⁵). Adjusting for 40 baseline clinical covariates left this apparent mortality reduction in placebo adherers unchanged.
• Beta-Blocker Heart Attack Trial (BHAT, 1990) and CHARM Trial (2005): In both trials, patients exhibiting high adherence to placebo capsules experienced a 35% to 50% lower risk of all-cause mortality than patients with poor placebo adherence, demonstrating that daily pill-taking regularity serves as a proxy for intact executive function, social support, and physiological reserve.

Postmenopausal Hormone Therapy, Statins, and Frailty Bias

Throughout the 1980s and 1990s, observational studies reported that postmenopausal women taking estrogen or combined estrogen–progestin therapy experienced a 35% to 50% reduction in coronary heart disease incidence (RR ≈ 0.50–0.65). In 2002, the Women's Health Initiative (WHI) randomized trial (N = 16,608) demonstrated that estrogen plus progestin actually increased coronary heart disease events by 29% (HR = 1.29), stroke by 41% (HR = 1.41), and pulmonary embolism by 113% (HR = 2.13). Reanalyses by Miguel A. Hernán and James M. Robins (2008) proved that the discrepancy resulted from healthy user confounding combined with prevalent user bias.

Similarly, to test whether observational claims that statin adherence lowered sepsis, fracture, and dementia by 40%–60% reflected healthy user bias, Colin R. Dormuth, M. Alan Brookhart, and Sebastian Schneeweiss (2009) analyzed 141,086 statin initiators in British Columbia using negative control outcomes. Statin-adherent patients were more likely to undergo screening mammography (HR = 1.22), PSA testing (HR = 1.57), and fecal occult blood testing (HR = 1.31), while experiencing a 25% lower rate of motor vehicle accidents (HR = 0.75) and a 22% lower rate of workplace trauma (HR = 0.78). Likewise, Lisa A. Jackson and colleagues (2006) demonstrated that the apparent 50% reduction in all-cause winter mortality attributed to influenza vaccination in elderly claims databases was strongest in autumn before any influenza virus circulated (RR = 0.39), reflecting unmeasured end-of-life frailty among unvaccinated seniors.

In contemporary biostatistics and public health, four strategies mitigate healthy user bias:

1. Active-Comparator New-User (ACNU) Design: Restricting the cohort to incident new initiators of the preventive drug and comparing them against incident new initiators of another preventive therapy prescribed for an equivalent stage balances health-seeking behaviors.
2. Negative Control Outcomes and Negative Control Time Periods: Evaluating outcomes that lack a biological link to the drug (such as accidental trauma or pre-exposure time windows) quantifies residual confounding.
3. Pre-Treatment Healthcare Utilization & Claims Frailty Indices: Adjusting for prior adherence to other chronic medications, prior preventive screenings, and validated frailty scores attenuates the open path A ← U → Y.
4. Instrumental Variables and Mendelian Randomization: Using genetic variants (such as HMGCR or PCSK9 alleles) or physician prescribing preferences separates biological drug effects from patient lifestyle behaviors.

How does healthy user bias differ from confounding by indication?

Confounding by indication occurs when a therapeutic drug is selectively prescribed to patients with greater disease severity, biasing observational estimates toward making the treatment appear harmful (RR > 1.0). Healthy user bias occurs with preventive interventions, supplements, and screening: health-conscious and functionally robust individuals selectively initiate and adhere to preventive regimens, biasing observational estimates toward making the intervention appear artificially protective (RR < 1.0).

What is the "sick stopper" effect?

The sick stopper effect occurs when patients who develop a life-limiting illness (such as metastatic cancer or advanced dementia) discontinue long-term preventive medications like statins because preventive care is no longer prioritized. Failing to account for time-varying confounding when patients stop therapy exaggerates the apparent survival benefit of ongoing adherence.

1. Shrank, W. H., Patrick, A. R., & Brookhart, M. A. (2011). "Healthy user and related biases in observational studies of preventive interventions: a primer for physicians." Journal of General Internal Medicine, 26(5), 546–550. https://doi.org/10.1007/s11606-010-1609-1
2. Coronary Drug Project Research Group. (1980). "Influence of adherence to treatment and response of cholesterol on mortality in the Coronary Drug Project." New England Journal of Medicine, 303(18), 1038–1041. https://doi.org/10.1056/NEJM198010303031804
3. Dormuth, C. R., Patrick, A. R., Shrank, W. H., Wright, J. M., Glynn, R. J., Sutherland, J., & Brookhart, M. A. (2009). "Statin adherence and risk of accidents: a cautionary tale." Circulation, 119(15), 2051–2057. https://doi.org/10.1161/CIRCULATIONAHA.108.824151
4. Jackson, L. A., Jackson, M. L., Nelson, J. C., Neuzil, K. M., & Weiss, N. S. (2006). "Evidence of bias in estimates of influenza vaccine effectiveness in seniors." International Journal of Epidemiology, 35(2), 337–344. https://doi.org/10.1093/ije/dyi274