Healthy user bias

Healthy user bias, also known as the healthy user effect, is a form of selection bias in observational studies, particularly in pharmacoepidemiology, where individuals who receive preventive therapies or medications are more likely to engage in other health-promoting behaviors, such as regular exercise, healthy diet, smoking cessation, and seeking routine medical care, thereby confounding results and often leading to an overestimation of the therapy's benefits.^[1]^[2] This bias arises because healthier or more health-conscious individuals are systematically more inclined to initiate and adhere to preventive treatments, creating systematic differences between treated and untreated groups that are not accounted for by measured confounders.^[1] A related phenomenon, the healthy adherer effect, further exacerbates this issue by showing that patients who adhere to one preventive regimen are prone to adopting additional healthy practices compared to non-adherers, skewing adherence-outcome associations.^[2] Notable examples include observational studies on hormone replacement therapy (HRT), which initially suggested a 33% reduction in coronary heart disease risk among users due to their overall healthier lifestyles, a finding later contradicted by randomized controlled trials like the Women's Health Initiative, which reported a 29% increased risk.^[1] Similarly, studies on statin use have demonstrated that users receive more preventive services, such as vaccinations and cancer screenings, contributing to apparent protective effects against unrelated outcomes.^[3] Influenza vaccination research has also been affected, with observational data indicating 40–50% mortality reductions that may partly reflect pre-existing health behaviors rather than vaccine efficacy alone.^[1] The implications of healthy user bias are significant in real-world evidence generation, as it can mislead policy decisions and clinical guidelines by inflating the perceived efficacy of interventions, particularly for chronic disease prevention.^[2] To mitigate this bias, researchers employ strategies such as the new user design, which compares incident users to non-users to reduce differences in baseline health; active comparators, using similar preventive therapies as references; and advanced statistical methods like high-dimensional propensity score adjustment to balance unmeasured confounders.^[1]^[2] Despite these approaches, completely eliminating the bias remains challenging, underscoring the need for caution when interpreting non-randomized data on preventive interventions.^[2]

Definition and Characteristics

Core Definition

Healthy user bias, also referred to as the healthy user effect, is a form of selection bias in observational studies, particularly in pharmacoepidemiology, where individuals who initiate and adhere to preventive therapies or interventions systematically differ from non-users in ways that promote better health outcomes, such as engaging in additional health-promoting behaviors or seeking more preventive care.^[1] This bias leads to an overestimation of the true benefits of the intervention, as positive outcomes may be attributed to the therapy itself rather than the underlying characteristics of the users.^[4] The phenomenon is most pronounced in studies comparing long-term users to non-users, where unadjusted analyses fail to account for these confounding health behaviors.^[3] At its core, healthy user bias encompasses two interrelated components: the healthy initiator effect, which involves baseline confounding where healthier individuals are more likely to start the therapy, and the healthy adherer effect, a form of selection bias that occurs over time as only those who maintain healthier lifestyles continue adherence.^[4] These effects arise because users of preventive treatments, such as statins or hormone replacement therapy, often exhibit broader patterns of health consciousness, including better diet, exercise, and compliance with screenings or vaccinations, which independently contribute to improved health metrics.^[1] Failure to adjust for these factors can distort effect estimates, making interventions appear more efficacious than they are in the general population.^[3] This bias is distinct from other selection biases, such as the healthy worker effect in occupational studies, but shares similarities in how it systematically excludes or underrepresents sicker or less health-oriented individuals from the exposed group.^[1] Recognizing and mitigating healthy user bias typically requires study designs like the new-user approach or restriction to comparable cohorts based on health status and indication, though complete adjustment remains challenging due to the difficulty in measuring unobservable behaviors from administrative data.^[4]

Distinguishing Features

Healthy user bias is characterized by the systematic tendency for individuals who adopt preventive health interventions to exhibit broader patterns of health-conscious behaviors and access to care, distinct from the specific effects of the intervention itself. This bias primarily manifests in observational studies evaluating preventive therapies, where users of one intervention are more likely to engage in complementary healthy practices, such as regular exercise, balanced diets, smoking cessation, or adherence to other screenings and vaccinations. Unlike general selection bias, which may stem from arbitrary participant enrollment, healthy user bias specifically arises from pre-existing differences in health-seeking propensity, leading to overestimation of intervention benefits.^[1] A key distinguishing feature is its focus on preventive rather than curative treatments, where non-users often represent a sicker or less proactive population. For instance, in studies of hormone replacement therapy (HRT), observational data initially suggested cardiovascular protection, but randomized controlled trials (RCTs) refuted this, attributing the discrepancy to users' overall healthier lifestyles, including lower rates of obesity and higher physical activity. This contrasts with confounding by indication, where treatment assignment correlates with disease severity (e.g., sicker patients receiving aggressive therapies), potentially biasing toward harm rather than benefit. Healthy user bias, by contrast, inflates apparent efficacy through unmeasured confounders like socioeconomic status and motivation for self-care.^[1]^[2] Another hallmark is its overlap yet distinction from the healthy adherer effect, which pertains to better outcomes among those who adhere to therapy due to inherent healthiness, rather than initiation of use. Healthy user bias operates at the cohort level upon treatment uptake, often requiring analytical adjustments like propensity score matching or active comparators (e.g., comparing statin users to users of other chronic medications) to approximate comparable health behaviors. In vaccine studies, such as influenza vaccination among pneumonia patients, vaccinated individuals showed lower mortality (48% reduction) and hospitalization (27% reduction), largely attributable to baseline health advantages like greater physical independence (95% vs. 84% in unvaccinated). This bias is particularly prevalent in pharmacoepidemiology, where it challenges causal inference without rigorous design elements like new-user cohorts to minimize survival or exposure time artifacts.^[2]^[5]

Historical Context

Origins in Epidemiology

The concept of healthy user bias traces its origins to the broader phenomenon of selection bias in epidemiology, particularly the "healthy worker effect" first described in occupational studies. In 1885, William Ogle, a medical statistician, observed in his analysis of English and Welsh mortality data that workers in physically demanding occupations exhibited lower overall mortality rates than the general population, attributing this to a selection process where only healthier individuals could enter and remain in such roles.^[6] This effect highlighted how employment acts as a filter for health, leading to underestimation of occupational hazards in cohort studies compared to the broader populace.^[7] The healthy worker effect laid the groundwork for recognizing similar biases in non-occupational settings, evolving into the "healthy user bias" within pharmacoepidemiology during the late 20th century. This shift became evident in observational studies of preventive therapies, where users of interventions like hormone replacement therapy (HRT) were systematically healthier than non-users even before treatment initiation. A seminal observational study from the Nurses' Health Study in 1985 reported that postmenopausal women using HRT had approximately 50% lower risk of coronary heart disease compared to non-users, suggesting protective effects that later randomized trials contradicted.^[8] This discrepancy prompted scrutiny of selection mechanisms, with early discussions framing it as "prevention bias" in estrogen use.^[9] The term "healthy user bias" gained traction in the 1990s as researchers quantified pre-treatment health differences among therapy adherents. In a 1996 analysis of the Multiple Risk Factor Intervention Trial data, Matthews et al. demonstrated that women who later initiated estrogen replacement therapy were already healthier—exhibiting lower rates of hypertension, diabetes, and overweight—than non-initiators at baseline, formalizing the bias as a form of confounding by health-seeking behavior.^[10] By the late 1990s, this concept was explicitly invoked to explain inflated benefits in observational data on postmenopausal hormones, marking its establishment as a key concern in evaluating preventive interventions.^[11]

Evolution and Recognition

The concept of healthy user bias emerged in the late 1980s and early 1990s within pharmacoepidemiology, particularly in response to discrepancies between observational studies and randomized controlled trials evaluating preventive therapies. Early recognition stemmed from analyses of hormone replacement therapy (HRT) in postmenopausal women, where observational data suggested substantial cardiovascular benefits that were later refuted. For instance, the 1985 Nurses' Health Study reported a 50% reduction in coronary heart disease risk among estrogen users compared to non-users, attributing this to the therapy itself.^[12] However, subsequent scrutiny revealed that users were systematically healthier at baseline, engaging more in preventive health behaviors, which confounded the apparent protective effects.^[1] By the early 1990s, epidemiologists began formalizing this phenomenon as a distinct source of selection bias. In 1991, Elizabeth Barrett-Connor coined the term "prevention bias" to describe how healthier individuals are more likely to initiate and adhere to preventive interventions, such as estrogen therapy, leading to overestimation of benefits in non-randomized studies.^[13] This idea was further developed in 1996 by Matthews et al., who demonstrated that women using estrogen had a more favorable cardiovascular risk profile—lower cholesterol, better blood pressure, and healthier lifestyles—prior to therapy initiation, highlighting the role of pre-existing health behaviors in biasing results. These contributions shifted focus from therapy efficacy to underlying patient selection dynamics, marking the evolution from anecdotal observations to a named bias in epidemiological literature. The bias gained broader recognition in the 2000s as similar patterns appeared in studies of other preventive measures, including statins, vitamins, and vaccinations. The 2002 Women's Health Initiative trial provided pivotal evidence, showing no cardiovascular benefit—and potential harm—from combined estrogen-progestin therapy, directly contradicting earlier observational findings and underscoring healthy user bias as a key confounder.^[14] In 2007, Brookhart et al. extended the concept to the "healthy adherer effect," observing that statin adherents not only used more preventive services like cancer screenings but also exhibited lower mortality from unrelated causes, such as hip fractures, independent of the drug's pharmacological action.^[15] Comprehensive reviews, such as Shrank et al. in 2011, synthesized these insights, emphasizing the bias's prevalence across observational pharmacoepidemiology and calling for adjusted analytical methods to mitigate its impact.^[1] This period solidified healthy user bias as a cornerstone of bias assessment in non-experimental research, influencing study design and interpretation in public health.

Mechanisms and Causes

Selection Processes

Selection processes in healthy user bias primarily involve the non-random allocation of individuals into treatment or study groups, where healthier or more health-conscious participants are disproportionately represented among users of preventive therapies. This occurs through patient self-selection, where individuals with proactive health behaviors—such as regular exercise, balanced diets, or routine screenings—are more likely to initiate and adhere to preventive medications like statins or hormone replacement therapy (HRT). For instance, in observational studies of HRT, women who chose to use the therapy often exhibited lower baseline risks for cardiovascular events due to their overall healthier lifestyles, leading to overestimated treatment benefits.^[1] Physician-driven selection further exacerbates this bias, as clinicians tend to prescribe preventive interventions preferentially to patients perceived as lower risk or more compliant, excluding those with comorbidities or frailty. This selective prescribing creates systematic differences between treated and untreated groups, confounding associations in pharmacoepidemiologic research. A classic example is seen in studies of lipid-lowering therapies, where statin initiators were more likely to receive concomitant preventive services, such as influenza vaccinations, reflecting underlying health-seeking differences rather than the drug's direct effects.^[15]^[1] In cohort or database studies, selection into the analytic sample can amplify these issues if enrollment criteria or follow-up favor adherent, healthier participants, resulting in depleted comparator groups. For preventive therapies like metformin, users often differ from non-users in unmeasured factors such as socioeconomic status and access to care, which drive both treatment uptake and better outcomes independently of the intervention. This selection dynamic underscores the need for designs like new-user cohorts to minimize prevalent user distortions, though it does not fully eliminate baseline differences arising from health consciousness.^[2]^[16]

Behavioral and Socioeconomic Factors

Behavioral factors play a significant role in healthy user bias, as individuals who initiate or adhere to preventive therapies often exhibit a broader pattern of health-conscious behaviors that independently improve outcomes. For instance, users of preventive medications like statins are more likely to engage in activities such as regular exercise, maintaining a healthy diet, and avoiding smoking, which can confound estimates of treatment efficacy in observational studies.^[1] This self-selection into healthier lifestyles leads to overestimation of benefits, as seen in early observational data on hormone replacement therapy (HRT), where users demonstrated reduced cardiovascular risk partly due to their higher engagement in preventive screenings and vaccinations, rather than the therapy itself.^[1] Adherence to therapy further amplifies this bias, with adherent patients showing greater utilization of services like influenza vaccinations (hazard ratio [HR] 1.21) and cancer screenings, such as prostate-specific antigen tests (HR 1.57), compared to non-adherent individuals.^[15] Socioeconomic factors contribute to healthy user bias by influencing access to care and the propensity for health-promoting behaviors, with higher socioeconomic status (SES) individuals more likely to receive preventive interventions. Those with elevated education and income levels tend to have better health literacy and resources, enabling them to seek out therapies and maintain adherence, while also benefiting from fewer comorbidities.^[3] In a cross-sectional analysis of U.S. adults using National Health and Nutrition Examination Survey data, users of antihypertensives and lipid-lowering drugs exhibited higher odds of high education (odds ratio [OR] 1.2) and non-impoverished status (OR 1.3), alongside reduced smoking rates (OR 1.2), illustrating how SES shapes user profiles.^[17] This disparity results in biased comparisons, as lower-SES non-users may face barriers like limited healthcare access, exacerbating the apparent superiority of therapy outcomes among users.^[3] The interplay between behavioral and socioeconomic factors often compounds the bias, as higher SES correlates with both the adoption of healthy behaviors and selective prescribing by providers, who may avoid recommending therapies to socioeconomically disadvantaged or frailer patients. For example, frail individuals with functional limitations have 26-33% lower odds of statin initiation, partly due to socioeconomic constraints on access and behavioral engagement.^[1] Such mechanisms underscore the need for adjustment in observational research to isolate true treatment effects from these confounding influences.^[17]

Examples in Observational Studies

Preventive Therapies

Healthy user bias manifests prominently in observational studies of preventive therapies, where individuals prescribed such interventions, such as statins or hormone replacement therapy (HRT), often exhibit unmeasured healthy behaviors that confound estimates of treatment efficacy.^[1] These users are more likely to engage in additional health-promoting activities, including regular exercise, balanced diets, smoking cessation, and routine screenings, leading to inflated apparent benefits compared to non-users.^[2] For instance, in studies of lipid-lowering agents like statins for primary prevention, adherent patients demonstrate higher utilization of other preventive services, such as influenza vaccinations (hazard ratio of 1.21) or prostate-specific antigen testing (hazard ratio of 1.57), which correlates with lower overall mortality risks that may not be attributable to the drug itself.^[1] A classic example involves statins, where observational data initially suggested a 23% reduction in hip fracture risk among users, attributed partly to their healthier lifestyles rather than the medication's direct effect.^[1] This bias extends to the "healthy adherer effect," a related phenomenon where patients who adhere to statin regimens are more prone to other positive health practices, resulting in outcomes like a 25% lower rate of motor vehicle accidents (hazard ratio of 0.75), independent of cardiovascular protection.^[1] Similarly, in analyses of metformin for diabetes management—a preventive therapy against complications—observational associations with reduced cancer incidence may reflect users' propensity for healthier behaviors, such as increased physical activity and better dietary adherence, rather than the drug's pharmacological action alone.^[2] Hormone replacement therapy provides another illustrative case, with the Nurses' Health Study reporting a one-third reduction in coronary heart disease risk among users based on observational follow-up.^[1] However, randomized controlled trials, such as the Women's Health Initiative, later revealed a 29% increased risk of coronary events, highlighting how healthy user bias in non-randomized designs systematically overestimates benefits by selecting for women with superior baseline health profiles and preventive care engagement.^[1] To address this in statin studies, researchers have developed tools like the Preventive Services Index, which aggregates utilization of services such as mammograms and colon cancer screenings (scoring 0-8 over two years) to adjust for health-seeking behavior in older adults, demonstrating correlations as high as 0.844 with influenza vaccination rates and improving the validity of cost-effectiveness estimates.^[18] These examples underscore the challenge in isolating true therapeutic effects from the confounding influence of users' broader wellness-oriented lifestyles in observational research.^[2]

Occupational and Cohort Studies

In occupational epidemiology, healthy user bias often appears as the "healthy worker effect" (HWE), a form of selection bias where employed individuals exhibit lower mortality and morbidity rates compared to the general population due to the preferential hiring of healthier candidates and the tendency for less healthy workers to leave employment.^[7] This bias arises from two primary mechanisms: the healthy hire effect, in which only physically and mentally capable individuals are selected for jobs (e.g., pre-employment medical screenings for firefighters or manual laborers), and the healthy survivor effect, where healthier workers remain employed longer while those with health issues exit the workforce, reducing their exposure to occupational hazards.^[7] Seminal work by McMichael formalized HWE in 1976, highlighting its distortion of exposure-outcome associations in cohort designs.^[19] A classic example is observed in early cohort studies of gas workers in the UK, where Doll et al. (1965) reported standardized mortality ratios (SMRs) below 100 for overall causes, masking potential risks from coal gas exposure; this underestimation was later attributed to HWE, as the cohort was inherently healthier than the national population.^[20] Similarly, in a large retrospective cohort of 38,672 United Autoworkers-General Motors employees followed from 1938 to 1994, truncating follow-up at employment termination introduced collider-stratification bias, yielding biased hazard ratios (HRs) for metalworking fluid exposure and mortality—e.g., an HR of 0.92 for all-cause mortality with truncation versus 1.09 with full follow-up—demonstrating how restricting analysis to active workers amplifies HWE and underestimates risks.^[21] In medical radiation worker cohorts, such as a study of over 100,000 Korean workers, HWE components varied by sex: prior exposure was positively associated with continued employment in males (HR 1.06 for mortality) but inversely in females (HR 0.82), leading to differential bias in estimating radiation-related cancer incidence and underscoring the need for sex-stratified analyses to mitigate distortion. Broader cohort studies of preventive interventions, like statin use among employed populations, have shown healthy adherer effects, where adherent workers had a 23% lower rate of workplace accidents (HR 0.77), reflecting underlying health behaviors rather than drug efficacy alone.^[22] These examples illustrate how HWE in occupational cohorts can confound associations, often requiring internal comparisons or adjustments for employment duration to approximate true effects.^[7]

Implications for Research Validity

Bias in Effect Estimation

Healthy user bias systematically distorts effect estimates in observational studies of preventive interventions by confounding treatment exposure with unmeasured or residual healthy behaviors among users. This selection bias occurs when healthier individuals—those with better adherence to lifestyle recommendations, higher socioeconomic status, or greater access to care—are more likely to initiate and continue therapies, leading to an apparent protective association that exaggerates true drug effects. As a result, unadjusted analyses often overestimate benefits, such as reduced risk of cardiovascular events or mortality, while underestimating potential harms.^[1]^[2] The bias arises primarily through noncomparability between treated and untreated groups, where users exhibit systematic differences in prognostic factors not fully captured by available covariates. For instance, in pharmacoepidemiologic studies, healthy user bias manifests as a spurious inverse association between exposure and adverse outcomes due to concurrent healthy practices like exercise or smoking cessation, which independently improve health. Adjustments for observable confounders, such as age or comorbidities, may attenuate but not eliminate this distortion if key behaviors remain unmeasured, as administrative databases often lack detailed lifestyle data. In extreme cases, the bias can invert true null or harmful effects into protective ones, misleading causal inferences.^[1]^[2]^[23] Illustrative examples highlight the magnitude of this bias in effect estimation. Observational studies of hormone replacement therapy (HRT) initially reported a 40-50% reduction in coronary heart disease risk among users, attributed partly to healthy user bias, but randomized controlled trials later revealed a 29% increased risk, underscoring the overestimation.^[24] Similarly, for statin initiators in cohort studies, hazard ratios for myocardial infarction appeared overly protective (e.g., HR 0.55), far exceeding randomized trial estimates (e.g., HR 0.81), due to residual confounding from health-promoting behaviors among new users. Influenza vaccination studies showed 40-50% mortality reductions before adjustment, which dropped to 29% after accounting for functional status, demonstrating how bias inflates preventive effect sizes. These discrepancies emphasize the need for bias-aware designs to align observational estimates with experimental evidence.^[1]^[23]

Consequences for Public Health

Healthy user bias in observational studies often results in the overestimation of benefits from preventive interventions, leading to potentially misguided public health policies and clinical guidelines that promote treatments with inflated apparent efficacy.^[1] For instance, early observational data suggested that hormone replacement therapy (HRT) reduced coronary heart disease risk by 40-50% in postmenopausal women, influencing widespread prescribing practices—reaching an estimated 15 million users in the United States by 2001—and shaping guidelines that encouraged its use for cardiovascular protection.^[24] However, subsequent randomized controlled trials (RCTs), such as the Women's Health Initiative, revealed no such benefit and instead demonstrated a 29% increased incidence of coronary heart disease, highlighting how bias-driven conclusions can expose populations to unnecessary risks and resource misallocation.^[1] This bias extends to other preventive therapies, where healthier users' behaviors confound outcomes, potentially undermining trust in evidence-based medicine and diverting attention from truly effective strategies. In studies of statin use, observational evidence linked the drugs not only to cardiovascular benefits but also to spurious reductions in hip fractures (by 23%) and Alzheimer's disease risk, prompting off-label recommendations that were later unsupported by RCTs.^[1] Similarly, for antihypertensives and lipid-lowering drugs, users often exhibit higher socioeconomic status, education, and healthier lifestyles—such as non-smoking and physical activity—resulting in overestimated cardiovascular event reductions compared to trial data, which may lead to over-reliance on pharmacotherapy at the expense of addressing modifiable risk factors like diet and exercise.^[17] The public health ramifications include delayed identification of ineffective or harmful interventions, as seen with influenza vaccination in the elderly, where observational studies reported 40–50% mortality reductions that were questioned by pre-season data indicating selection of healthier recipients rather than vaccine efficacy.^[1] Such distortions can perpetuate inequities, as biased studies may overlook vulnerable populations less likely to engage in preventive care, ultimately hindering accurate policy formulation and resource prioritization for broader health improvements.^[3]

Mitigation and Analytical Strategies

Study Design Modifications

To mitigate healthy user bias in observational studies, researchers often employ the new-user (or incident user) design, which restricts the study population to individuals who initiate treatment during the observation period, excluding prevalent users who have been on therapy for an extended time. This approach helps address the bias arising from long-term users who may appear healthier due to surviving initial adverse effects or demonstrating better adherence, particularly in studies of preventive interventions. For instance, in pharmacoepidemiologic research on statins, the new-user design has been used to compare outcomes like myocardial infarction risk between new initiators and non-initiators, reducing confounding from prior treatment tolerance.^[1] Another key modification is the active-comparator design, which compares initiators of the study treatment to initiators of an alternative active treatment for the same indication, rather than to untreated individuals. This strategy minimizes differences in health-seeking behaviors and unmeasured confounders, such as frailty or lifestyle factors, that healthier users might exhibit when seeking any preventive therapy. An example is seen in studies of biologic disease-modifying antirheumatic drugs (bDMARDs) for rheumatoid arthritis, where new users of bDMARDs are compared to new users of synthetic DMARDs to assess infection risks more equitably, showing reduced bias in effect estimates compared to non-user controls.^[25] Combining the new-user and active-comparator designs further enhances robustness, as it ensures pretreatment comparability and avoids immortal time bias while focusing on incident exposures. This hybrid approach has been particularly valuable in evaluating preventive therapies like tumor necrosis factor-alpha inhibitors, where early treatment hazards (e.g., infections within the first 90 days) can be more accurately assessed without the distortion from healthier long-term survivors.^[26]

Statistical Adjustment Techniques

Statistical adjustment techniques aim to control for the confounding introduced by healthy user bias in observational studies, where users of preventive interventions often exhibit unmeasured or hard-to-measure healthier behaviors that independently affect outcomes. These methods typically involve modeling the probability of treatment receipt or using external variation to isolate causal effects, thereby reducing bias from factors like health-seeking behavior or adherence patterns. Common approaches include regression-based adjustments, propensity score methods, instrumental variable analyses, and specialized index scores derived from proxies of healthy user characteristics.^[1] Regression adjustment represents a foundational technique, wherein multivariable models incorporate covariates that proxy for healthy user behaviors, such as utilization of preventive services (e.g., vaccinations or screenings) or markers of overall health status like comorbidity indices. By including these variables, the model estimates treatment effects conditional on observed confounders, potentially mitigating bias if the proxies adequately capture the underlying healthy user effect. For instance, adjusting for influenza vaccination rates as a marker of health-seeking has been shown to attenuate exaggerated benefits in studies of preventive therapies. However, this method assumes correct model specification and may perform poorly with high-dimensional data or unmeasured confounders.^[3]^[1] Propensity score methods offer a more flexible approach by estimating the probability of treatment assignment based on observed covariates and then using this score for adjustment via matching, stratification, weighting, or inclusion in outcome models. High-dimensional propensity score (hd-PS) calibration extends this by empirically selecting and adjusting for hundreds of potential confounders from administrative data, such as prior healthcare utilization patterns that reflect healthy user tendencies. In studies of statin initiation, hd-PS adjustment reduced bias from unmeasured health behaviors, yielding estimates closer to randomized trial results compared to standard regression. This technique is particularly valuable in claims databases where proxies for healthy user bias, like frequency of physician visits, are abundant.^[1] Instrumental variable (IV) analysis addresses unmeasured confounding inherent in healthy user bias by exploiting exogenous variation in treatment receipt that does not directly affect the outcome except through the treatment itself. A valid instrument, such as physician-specific prescribing preferences for preventive drugs, must be associated with treatment probability but independent of patient health behaviors or outcomes. In pharmacoepidemiologic research, IV methods using regional variation in statin prescribing attenuated healthy adherer effects, providing bounds on causal effects that traditional adjustments could not achieve. Despite its strengths, IV estimation requires strong instruments and can produce wider confidence intervals, limiting precision in smaller studies.^[1] Specialized index scores have emerged to quantify and adjust for healthy user bias more directly, often constructed from weighted combinations of preventive service utilization as proxies for overall health engagement. One such index, developed in a cohort of over 900,000 type 2 diabetes patients, assigns points based on factors like age, sex, and receipt of services such as mammograms or lipid screenings, achieving moderate predictive ability (C-statistic 0.605) for behaviors like influenza vaccination. When incorporated into regression models, this score improved adjustment for bias in estimates of preventive therapy effects, though external validation across diverse populations remains essential for broader applicability.^[27]