Purpose of Review
In pharmacoepidemiological studies, the utilization of extensive databases is indispensable for assessing the efficacy and safety profiles of drug exposures across diverse and large populations. Given that treatments are not randomly assigned in these settings, the careful selection of a relevant comparison group is paramount. This comparator group can be composed of individuals initiating: (1) a treatment with a similar indication (active comparator), (2) a treatment indicated for a different condition (inactive comparator), or (3) no specific treatment (non-initiators). This review examines recent literature, outlining key considerations and implications in comparator selection for pharmacoepidemiological studies.
Recent Findings
The choice of comparator is contingent upon the specific scientific question being addressed and practical feasibility. Pharmacoepidemiological studies, by their nature, rely on the decision to initiate or not initiate a particular treatment, making them susceptible to confounding, particularly related to the comparator choice. This includes confounding by indication, disease severity, and frailty. We delineate different forms of confounding inherent in pharmacoepidemiological research and discuss each type of comparator along with illustrative examples and a detailed case study. Our commentary emphasizes potential challenges associated with comparator selection in each study, highlighting the critical need to understand the population receiving treatment and the interplay between patient characteristics and outcomes.
Summary
Advanced statistical methodologies alone may not suffice to fully mitigate confounding in observational studies. Rigorously evaluating the extent to which comparator selection can either minimize or introduce systematic bias is a cornerstone of robust pharmacoepidemiological research.
Keywords: pharmacoepidemiology, comparator selection, new user design, confounding, detection bias
Introduction
Randomized controlled trials (RCTs) are rightly considered the gold standard for evaluating treatment effects on specific outcomes, as randomization and blinding minimize potential biases. However, observational studies often offer a more pragmatic approach to assess drug effectiveness and safety, especially for rare outcomes (e.g., cancer), because they can encompass large, heterogeneous populations with extended follow-up periods. They are also vital for understanding drug effects in real-world scenarios, contrasting with RCTs that typically involve highly selected populations under stringent monitoring. Large administrative databases, enriched with drug reimbursement and dispensing data, are particularly valuable, capturing longitudinal exposure information across diverse healthcare settings.
Study design robustness is crucial to minimize biased treatment effect estimations. In observational studies incorporating the concept of a hypothetical intervention, commonly leading to a new user study design[1], the comparator selection stands out as a critical design element. Three primary comparator options exist: the active comparator[2], the inactive comparator[3], and the non-initiator comparator. The choice among these depends on the research question and feasibility considerations. An active comparator is defined as a specific drug or drug class with an indication and formulation similar to the treatment of interest. The selection of an active comparator hinges on the research question’s focus, whether class-level or drug-specific effects are of primary interest[4, 5]. Conversely, an inactive comparator is a drug or drug class not indicated for the same condition as the treatment of interest. Despite this difference, inactive comparators can be instrumental in synchronizing study cohorts based on various factors, including healthcare utilization and the commencement of study follow-up. The seemingly simplest comparator option, yet often complex in execution, is the non-initiator comparator. In this scenario, the comparator group comprises individuals who do not initiate the treatment under investigation. Non-initiator comparisons are frequently employed when suitable active or inactive comparators are not available.
Observational studies of drug effects are susceptible to several biases, including: (1) time-related biases (e.g., immortal time bias, time-window bias, immeasurable time bias)[6–9], (2) confounding by frailty, (3) confounding by indication and its variations, and (4) outcome detection bias. The likelihood of these biases arising is significantly influenced by the chosen comparator. The subsequent sections will provide an overview of each bias type and discuss how comparator selection (active, inactive, or non-initiator) affects the probability of encountering these biases. To contextualize these design choices, we will draw upon contemporary examples from pharmacoepidemiology literature, highlighting crucial considerations for comparator selection (summarized in Table 1). We will conclude with a case study demonstrating a structured decision-making process for selecting a comparator in a pharmacoepidemiological study.
Table 1. Review of Comparator Selection Considerations and Implications for Selected Pharmacoepidemiologic Studies.
| # | Topic | Author, Year | Considerations for comparator selection | Comparator selected | Rationale/Potential for bias remaining | Approach used to address remaining bias |
|---|---|---|---|---|---|---|
| 1 | Glargine and the risk of cancer among diabetics | Stürmer, 2013 | High body mass index (BMI) is the main driver of (indication for) the need to add insulin in patients with type 2 diabetes and a risk factor for several cancers; no information on BMI in claims data. | Active comparator: long-acting human (NPH) insulin | By comparing to a medication with the same indication, the authors hope to reduce the risk of unmeasured confounders (e.g. BMI). If the choice to initiate a specific drug was associated with BMI, then confounding by BMI status could exist. | Examined the association between BMI and choice of insulin using 2 external electronic medical record databases; result: no effect of BMI on choice of insulin. |
| 2 | The effect of nicotine replacement therapy (NRT) on cardiovascular disease in smokers | Dollerup, 2017 | Smoking is a strong risk factor for heart disease, so the authors wanted a comparison group consisting of smokers. This would reduce the risk of confounding by smoking status. | Active comparator: smoking cessation counseling | By comparing to NRT to smoking cessation treatment, study is restricted to smokers who want to quit smoking. It is possible that the prescribing physician preferentially referred heavy smokers with substantial existing heart damage to NRT. Therefore, confounding by disease severity could exist. | The authors acknowledged limitations in the discussion. |
| 3 | The effect of postoperative chemotherapy on mortality among stage II-III rectal cancer patients | Lund, 2016 | Individuals receiving postoperative chemotherapy may be healthier than those not receiving postoperative chemotherapy. The authors wanted to identify a group of patients who were similarly “healthy” to those receiving postoperative chemotherapy. | Non-initiator: compared postoperative 5-fluorouracil (5-FU) or capecitabine to no chemotherapy. Active comparator: compared individuals receiving 5-FU or capecitabine to individuals receiving 5-FU/capecitabine + oxaliplatin | They restricted their study population to non-metastatic rectal cancer patients who had received preoperative chemoradiation or radiotherapy. Non-user comparison: physician may preferentially give healthier patients chemotherapy post-surgery (confounding by frailty). Active comparison: residual confounding by disease severity and frailty. | The authors stratified into clinically meaningful age groups. They acknowledged limitations and estimated the direction of bias. They used active and non-initiator comparisons. |
| 4 | Benzodiazepines and mortality | Patorno, 2017 | Non-initiators may have lower disease burden and therefore lower mortality or perhaps less access to care/surveillance and higher mortality. | Primary analysis: non-initiators. Sensitivity analysis: inactive comparator (SSRIs) | Non-user comparison: By requiring the non-users and users to have filled 1+ non-benzodiazepine prescription in the 0–90 and 91–180 days prior to index date, they restricted to individuals utilizing the healthcare system. inactive comparison: The inactive comparator were SSRI users. This medication class is used for long-term treatment of chronic conditions. It takes several weeks to notice symptom abatement. There is little potential for abuse. Benzodiazepines are frequently given in an as needed way and they work for the acute management of certain conditions. They have abuse potential. Habitual benzodiazepine users may be predisposed to higher mortality compared with SSRI users. | The authors used high dimensional propensity score models with many variables. They stratified into clinically meaningful age groups. They performed a sensitivity analysis with a comparator with overlapping indications. |
| 5 | Influenza vaccine and mortality | Jackson, 2005 | Individuals close to death and not expected to live to flu season may have had the vaccine withheld. individuals receiving the vaccine may be healthier and at lower risk of death | Non-user comparison in time periods where influenza vaccine should have no effect on mortality | The authors examined the effect of vaccine receipt on mortality in the time before, during and after influenza. They examined patterns of relative mortality risk over the three intervals to try and disentangle the true vaccine from bias attributable to health differences. | The authors incorporated many variables associated with health status into the models. They acknowledged that despite comprehensive variable selection, confounding by health status still exists. |
| 6 | Influenza vaccine and mortality | Zhang, 2017 | Individuals close to death and not expected to live to flu season may have had the vaccine withheld. individuals receiving the vaccine may be healthier and at lower risk of death | Non-user comparison in time period where influenza vaccine should have no effect on mortality | Examining the effect of vaccine receipt on mortality in a non-influenza time period aids in estimating the amount of confounding by frailty that exists. | The authors added variables related to independent living to their propensity score model. They acknowledged that residual confounding likely still exists. |
| 7 | Antibiotic use and recurrent breast cancer | Wirtz, 2017 | Individuals with many antibiotic events may be sicker than non-users and therefore differentially screened compared with non-users | Non-user comparison | The authors examined the association between antibiotic use and surveillance mammography. | The authors adjusted for screening in overall analysis. They acknowledged that ongoing surveillance is difficult to model and there may be residual confounding. |
| 8 | Statin use and recurrent breast cancer | Wirtz, 2017 | Adherent statin users may be healthier than non-users and therefore differentially screened compared with non-users | Non-user comparison | The authors examined the association between statin use and surveillance mammography. | The authors adjusted for screening in overall analysis. They acknowledged that residual confounding may exist. |
| 9 | Lithium and fetal outcomes | Patorno, 2017 | Individuals with bipolar disorder are much more likely to engage in unhealthy behaviors and to have more comorbidities. | Primary analysis: non-initiator Sensitivity analysis: active comparator (lamotrigine) | Non-user analysis: The non-user comparison group had a small percentage of individuals with a bipolar disorder diagnosis. They may have been untreated and as such may have distorted the non-user group. Sensitivity analysis: By restricting to an active comparator in individuals with a bipolar diagnosis, they the authors reduced confounding by bipolar behaviors. | The authors used rich propensity score model with many variables. They performed multiple sensitivity analyses including one with an active comparator. |
| 10 | Case study: Antidepressants and colorectal cancer (CRC) | Antidepressants, SSRIs in particular are commonly prescribed by a primary care physician. The prescribing physician generally wants to see the patient more frequently shortly after initiation to evaluate drug effects and titrate dosage. SSRIs are given for the long-term management of many diseases. It frequently takes a couple of weeks to observe symptom abatement. They are commonly used drugs and we expected a large number of initiators. | Primary analysis: inactive comparator – antihypertensive initiators excluding beta-blockers | The authors chose a comparison group 1) with little known association with the outcome 2) that must be engaged with the healthcare system 3) that regularly takes a medication given for the long-term management of chronic disease and that is commonly prescribed by a primary care physician 5) with a large number of anticipated initiators. | The authors had a strong understanding of the pathogenesis of CRC and used a wide set of clinically important covariates (including screening/diagnostic events) associated with CRC in the propensity score model. It was well-balanced. They performed several sensitivity analyses where they varied latency and lag assumptions. They acknowledged that there could still be some residual confounding. |
Overview of Biases Relevant to Comparator Selection
Several factors can influence treatment decisions and are also potentially linked to the outcome, thereby confounding the observed treatment effects. In the simplest scenario, where the comparison is against no treatment (a non-initiator comparator), time-related biases, such as immortal time bias, can arise from the requirement of no drug use post-cohort entry[6, 7]. This bias emerges because a natural synchronization between treatment initiation and non-initiation is absent. Given the extensive literature on time-related biases in pharmacoepidemiology[6, 7, 10, 8, 9], we will not delve further into examples here.
Confounding by frailty[11] occurs when specific treatments are less likely to be given to individuals in poor health, as potential benefits may be minimal or non-existent. This bias can lead to an overestimation of beneficial treatment effects[12]. Confounding by indication[13–15] arises when the reason for treatment (the indication) is also a risk factor for the outcome. A variant of this is confounding by disease severity, where disease severity influences both treatment choice and outcome risk, potentially making a treatment appear more harmful than it is. Another derivative of confounding by indication occurs when behaviors or characteristics associated with the primary indication are also outcome risk factors. Studies of psychiatric medications are particularly susceptible to this type of confounding, as individuals with certain conditions, especially if untreated, may be more prone to unhealthy behaviors compared to the general population[16–21].
Outcome detection bias occurs when outcome ascertainment differs between treatment groups[22, 23]. This bias can arise if overall health status or health-seeking behavior influences the likelihood and timing of outcome diagnosis. While possible for many outcomes, cancers with screening programs (e.g., prostate, colorectal, breast) are especially vulnerable. Cancer diagnosis likelihood is influenced by: 1) being healthy enough for screening (routine or diagnostic), 2) adherence to screening guidelines, and 3) engagement with the healthcare system. These same factors can also affect physician visits and medication initiation. This bias can operate in both directions; for example, statins might appear to increase cancer risk[24], while Alzheimer’s medications might seem to reduce it.
Active Comparators
Active comparator studies, by comparing treatment to medications with similar indications, inherently reduce confounding compared to inactive or non-initiator comparator designs. However, they are still not immune to confounding by disease severity and frailty.
Active Comparators: Glargine and Cancer Risk
Concerns about the long-acting insulin analog glargine potentially increasing cancer risk prompted Stürmer et al[25] to conduct an active comparator, new user study. They compared glargine initiation to human NPH insulin initiation within the Medical Outcomes Research for Effectiveness and Economics registry (2003-2010).
Body mass index (BMI), a risk factor for both cancer and diabetes, was a potential confounder unmeasured in the study dataset. To address this, the researchers examined the BMI-treatment choice association in two external datasets. After covariate adjustment, no association was found (adjusted odds ratio (aOR) 1.00, 95%CI 0.98–1.02; 0.99 0.96–1.03). They concluded no short-term association between glargine and overall cancer incidence (aHR 1.11, 0.95–1.32) or breast, prostate, or colon cancer incidence. This study exemplifies the use of external validation datasets to assess potential residual bias when crucial confounders are unavailable in the primary observational data.
Active Comparators: Nicotine Replacement Therapy and Cardiovascular Disease (CVD)
Dollerup et al[26] investigated the association between nicotine replacement therapy (NRT) and cardiovascular disease, using smoking cessation counseling (SCC) alone as an active comparator. While no association with CVD was found at 4 weeks, NRT was associated with increased ischemic heart disease (aHR 1.35, 95%CI: 1.03–1.77) or cerebrovascular disease (aHR 1.54, 1.08–2.19) at 52 weeks compared to SCC. Smoking intensity, a potential confounder, might explain this, as physicians may preferentially prescribe NRT to heavier, longer-term smokers who might have more pre-existing heart damage. However, smoking intensity data was unavailable. Other smoking cessation medications could have served as comparators[27], potentially reducing confounding by smoking severity, though this relies on accurate identification of the smoking cessation indication.
Active Comparators: Postoperative Chemotherapy in Older Adults with Rectal Cancer
Lund et al[28] examined postoperative chemotherapy and rectal cancer survival in older patients post-preoperative chemoradiation or chemotherapy. Using cancer registry data linked to Medicare claims, they compared mortality among patients receiving postoperative 5-fluorouracil (5-FU) or capecitabine, 5-FU/capecitabine plus oxaliplatin, or no chemotherapy. Postoperative 5-FU/capecitabine alone was associated with reduced mortality (aHR 0.46, 95%CI: 0.30–0.72) in patients aged 66–74, compared to no postoperative chemotherapy. No effect was seen in those over 74. Despite propensity score weighting for measured confounders, the authors suggested potential overestimation of postoperative chemotherapy benefits. Even with restriction to patients healthy enough for preoperative therapy and surgery, those receiving postoperative chemotherapy might still be healthier than those not, a bias potentially impossible to fully eliminate, as seen in influenza studies[29–31].
Inactive Comparators
When aiming for a causal contrast between treatment and no treatment[3], an inactive comparator, with no known association to the outcome, offers an alternative to non-users. Inactive comparators help mitigate time-related biases by synchronizing follow-up start at medication initiation. However, identifying a suitable inactive comparator is challenging, as it has its own indications and potential associations with the outcome. This approach has been termed active comparator[32–35], negative exposure control[36], and inactive comparator[3].
Key considerations for inactive comparator selection include: First, the inactive comparator‘s association with the outcome should be well-established. Ideally, for a use vs. non-use contrast, the inactive comparator should have evidence of no association with the outcome. Understanding the active comparator-outcome association is also crucial for proper inferences. For instance, in evaluating a new anti-hypertensive drug and angioedema with angiotensin-converting enzyme inhibitors (ACEIs) as an active comparator, the known ACEI-angioedema risk[37] must be considered, or a different comparator with no such association identified.
The inactive treatment should also be used similarly to the active treatment, e.g., for chronic disease management, not just acute symptom relief. Sufficient anticipated users are needed for precise estimation, considering potential concomitant users who would be ineligible. Finally, outcome risk factors should be well-known and directly measurable in the data.
Inactive Comparators: Benzodiazepines and Mortality
A study on benzodiazepines and mortality[35] primarily compared benzodiazepine initiators to matched non-users who had recently visited a physician. To minimize healthcare access differences, both groups were required to have filled non-benzodiazepine prescriptions prior to the index date. No association was found in this primary analysis. A sensitivity analysis used selective serotonin reuptake inhibitors (SSRIs) as an inactive comparator, as SSRIs, except for a small suicide risk increase in young individuals[38], are not linked to increased mortality. This represents an inactive comparison due to differing indications. Furthermore, usage patterns differ significantly: SSRIs are for long-term management of conditions like depression and anxiety[39, 40], taken daily for months, with symptom reduction taking weeks. Benzodiazepines, however, are for acute anxiety, panic, and sleep disorders, are controlled substances with abuse potential, and are often prescribed “as-needed,” contrasting with daily SSRI use. These population differences may explain the slightly increased mortality risk seen with benzodiazepine initiators (aHR 1.09, 95%CI 1.03–1.16). Regular benzodiazepine users, compared to SSRI users, may have a higher predisposition to substance abuse and other unmeasured factors increasing mortality.
Non-user Comparators
When a clear treatment alternative is absent, many studies use non-initiator comparators, comparing treatment initiators to those not initiating treatment. While prone to immortal time bias, strategies exist to mitigate this[41]. Advanced adjustment techniques like propensity scores, especially high-dimensional propensity scores[42], can reduce measured confounding (see example 4 above). However, unmeasured confounding, particularly regarding drug indication and frailty, remains a challenge due to measurement difficulties.
Non-user Comparators: Influenza Vaccine and Mortality
Numerous studies[43–48] have shown a strong inverse association between flu vaccine and mortality, but substantial evidence suggests this is largely due to underlying population differences. Individuals near death, unlikely to survive the flu season, may be less likely to be vaccinated, while vaccinated individuals are perceived as healthier and more likely to benefit. Jackson et al[29] addressed this by examining the association between vaccination and death before flu season, a negative control outcome unaffected by vaccination. Vaccine receipt was still associated with substantial mortality reduction (risk ratio (RR) = 0.39, 0.33–0.47). Adjustment for diagnoses did not significantly alter this, highlighting the challenge of controlling for frailty with administrative data variables. A contemporary study[31] attempted to reduce frailty confounding by adjusting for independent living markers as health status proxies, but while the mortality association attenuated (HR 0.68, 0.67–0.70), substantial residual confounding remained.
Non-user Comparators: Statins, Antibiotics, and Breast Cancer Outcomes
Wirtz et al[49] investigated why prior studies reported increased second breast cancer risk with antibiotics[50, 51] and decreased recurrent breast cancer risk with statins[52], all using non-user comparators. They examined screening practices of antibiotic and statin users. Adherent statin use was linked to more surveillance mammography (odds ratio (OR): 1.11, 95%CI 1.01–1.25) versus non-users, while heavy antibiotic use was associated with less mammography (OR: 0.90, 95% CI 0.82–0.99). Although screening behavior adjustment did not qualitatively change inferences, this study highlights how screening practices can vary among initiators of different medication classes, introducing potential detection bias when using non-user comparators.
Non-user Comparators: Lithium and Pregnancy Outcomes
Lithium, a primary bipolar disorder treatment (affecting ∼1–2.5% of the population[53]), presents challenges in pregnancy research. Untreated bipolar disorder can lead to risky behaviors (e.g., substance abuse)[54, 19, 18, 53] and potentially impact fetal development. Concerns about lithium safety in early pregnancy[55, 56] led Patorno et al[57] to study lithium use in the first trimester and fetal cardiac malformations. Disentangling bipolar-associated behaviors, medications, and fetal outcomes is complex. Untreated women at conception might engage in behaviors negatively affecting fetal development.
In their primary analysis, using Medicaid data, they compared first-trimester lithium users (no bipolar diagnosis required) to non-users, controlling for measured confounding with propensity score matching. They reported increased risk with lithium (adjusted hazard ratio (aHR) 1.65, 95%CI 1.02–2.68). However, untreated bipolar women in the non-user group might bias the association towards the null. Though the proportion of bipolar diagnoses was small in non-users, untreated women could obscure risk in this group.
To address confounding by indication, they conducted an analysis restricted to women with bipolar disorder, comparing lithium users to lamotrigine users (an active comparator), also indicated for bipolar disorder and not associated with cardiac malformations. This active comparator approach, and restricting to bipolar women, aimed to reduce confounding by unhealthy behaviors. Characteristics of lamotrigine and lithium users were similar regarding mental health diagnoses and comorbidities before propensity score weighting. The association in this active comparator analysis was similar, albeit higher (aHR 2.25, 95%CI 1.17–4.34).
Case Study: Antidepressants and Colorectal Cancer
As a case study of comparator selection, we explore designing a study on SSRI use and incident colorectal cancer (CRC), also considering serotonin norepinephrine reuptake inhibitors (SNRIs) and tricyclic antidepressants (TCAs). We will outline the comparator selection process and relevant considerations, such as covariate availability (Table 1).
Comparator Choice Considerations
Ideally, an active comparator minimizes bias. Psychiatric drugs are particularly prone to confounding due to indication-related behaviors. However, we found no similarly indicated comparator without potential CRC risk associations[58–61]. Other psychotropic drugs were considered, but definitive assurance of no CRC risk association was lacking. Some anti-psychotics, given to dementia patients for behavior management[62, 63] even without psychosis history, might lead to outcome detection bias due to lower screening and diagnosis rates in this population.
Lacking a suitable active comparator, we preferred a group initiating daily medication for chronic disease management, commonly prescribed by primary care physicians who also often prescribe SSRIs (except in severe mental illness cases). Non-initiator selection was thus ruled out. Starting follow-up at medication initiation is also more straightforward.
Anti-glaucoma drugs were briefly considered as an inactive comparator, as used in prior literature[33, 34, 64, 32]. However, these are typically prescribed by ophthalmologists, who may be less aware of patients’ overall health compared to primary care physicians. Their use might also be more for symptom abatement than long-term prophylaxis. We also questioned sufficient initiator numbers. Statins were excluded due to ongoing debate about their cancer risk association[65, 66]. Anti-hypertensives (AHTs) have substantial evidence suggesting no dramatic CRC risk alteration, with some evidence suggesting potential risk reduction[67], except for beta-blockers which may reduce cancer risk[68]. Hypertension is also commonly managed by primary care physicians.
Identifying Risk Factors in Administrative Data
We proceeded with a new user design[69] using AHTs as an inactive comparator in a Medicare beneficiary population[70]. We hypothesized “late-acting” effects of all three antidepressant classes on CRC risk, acting near the adenoma-carcinoma transition, suggesting detectability within a few years of follow-up. CRC natural history is well-documented[71–74], and major risk factors are generally known[75]. Measurable risk factors included age, male sex, inflammatory conditions history, and black race. Family history and genetics[75], strong risk factors, were unmeasurable. We used diagnoses for all non-CRC conditions prior to initiation as a proxy for general genetic cancer predisposition. We had data or proxies for hormone replacement therapy, non-steroidal anti-inflammatory use, diabetes, alcohol consumption, smoking, and obesity, all modestly affecting CRC risk[75]. Diet, exercise, BMI, or aspirin use (generally unavailable) were lacking.
Potential for Confounding by Indication Derivative
Limited evidence suggests depressed individuals, for whom antidepressants are indicated, may be less likely to adhere to screening guidelines[76, 77] and less likely to be screened or visit physicians, potentially leading to lower CRC diagnosis rates and outcome detection bias if screening behavior data was absent. However, administrative data contains CRC screening and diagnostic event information as Medicare covers these services[78]. We could therefore control for recent screening behavior potentially influenced by depression status. Furthermore, if antidepressants improve depressive symptoms, screening might become less of a concern after a few months of follow-up.
Antidepressants and CRC Results
We identified 530,304 SSRI, SNRI, TCA, or AHT initiators meeting age, enrollment, and CRC-free criteria. Substantially more individuals initiated AHTs (n=417,491) than antidepressants (SSRI: n=87,401; SNRI: n=12,211; TCA: n=13,201). Median continuous medication class use post-second prescription (overall=332 days) varied across classes [AHT=363; SSRI= 252; TCA=172; SNRI=238 days].
We observed 1,728 CRC events in 631,920 person-years (PY), with incidence ranging from 214 per 100,000 PY for TCA initiators to 281 cases per 100,000 PY for AHT initiators. SSRI initiators showed a reduced CRC rate compared to AHT initiators (aHR 0.85 95%CI 0.71–1.00). TCA and SNRI initiators also had lower adjusted CRC rates compared to AHT initiators [0.83, 0.52–1.31; 0.91, 0.59, 1.41], respectively. Sensitivity analyses showed a 5%−20% reduced CRC rate among SSRI users compared to AHT initiators. These associations aligned with previously reported estimates[79–84].
Our AHT comparator, while not perfect, had limitations including sample size reduction due to concomitant AD and AHT users, and follow-up time differences. Lacking absolute proof of no CRC association with AHTs, sensitivity analyses with additional inactive comparators hypothesized to have no outcome association could have been informative, as used in a study of immune-related conditions and keratinocyte cancer risk[85]. However, our data provided key variables, proxies for others, and enabled excellent covariate balance. We also compared drug classes requiring somewhat regular physician interaction. Our analysis suggests SSRIs do not increase CRC risk compared to AHTs in a Medicare population.
Conclusions
Comparator selection in observational pharmacoepidemiological studies is often complex. Systematic bias is a constant concern due to potential unmeasured confounding. However, careful design and comparator selection can minimize bias, considering the factors outlined in Table 1. Understanding (1) the treated population, (2) patient characteristic-outcome associations, and (3) outcome natural history and risk factors is crucial. Advanced statistical techniques may not fully eliminate confounding in observational studies. Complementary strategies, such as active comparators or addressing potential bias impact on treatment effect estimates (e.g., via validation studies and multiple bias modeling), offer promising avenues for future research in comparator methodology.
Footnotes
Compliance with Ethical Standards
Conflict of Interest
Monica D’Arcy declares no conflicts of interest; Til Stürmer reports grants from the National Institute on Aging, during the conduct of the study, grants from Astrazeneca and Amgen, outside the submitted work, membership (Center for Pharmacoepidemiology) of GlaxoSmithKline, UCB BioSciences, Merck, and Shire, outside the submitted work, and stock in Novartis, Roche, BASF, AstraZeneca, and NovoNordisk; Jennifer L. Lund reports grants from PhRMA Foundation, outside the submitted work; Dr. Lund’s husband is a full-time, paid employee of GlaxoSmithKline.
Human and Animal Rights and Informed Consent
This article does not contain any studies with human or animal subjects performed by any of the authors.
