PPT10 Observational Studies
American College of Clinical Pharmacy (ACCP) Clinical Research Challenge
Organization and Structure of the Competition:
- The American College of Clinical Pharmacy (referred to in discussion as ECCP / ACCP) hosts an annual national competition known as the Clinical Research Challenge.
- The initial phase is a local examination administered at participating pharmacy schools to evaluate proficiency in interpreting published clinical literature.
- The examination format mirrors journal club exercises, presenting students with published literature (such as randomized controlled trials [RCTs] or case-control studies) and assessing their critical appraisal skills.
- Each participating institution utilizes the local exam results to select a team of students to represent the school in the national competition rounds.
- The national competition comprises sequential rounds:
- Round 1: A team-based journal club examination and the development of a research proposal based on a specified letter of intent (LOI) for a given clinical research question.
- Subsequent Rounds: Top-performing teams advance through proposal development, culminating in the top teams presenting their work nationally.
Course Integration, Logistics, and Scoring:
- Course topics directly align with the core competencies evaluated in the ACCP Clinical Research Challenge, specifically the critical appraisal of trial methodology and observational research.
- Participation serves as an extra credit opportunity for up to bonus points, with points scaled downward based on relative performance on the examination.
- Because the local exam is designated for extra credit, it is administered entirely outside of regularly scheduled class hours.
- Tentative scheduling is set for the final day of formal classes on Friday, December 4, during the open block following the PPE class from to .
- A student survey determines interest via a binary (yes/no) selection:
- Students seeking solely extra credit without interest in national team selection indicate "no" to opt out of team consideration.
- The top scoring individuals who respond "yes" receive an invitation to join the institutional team.
- If an invited student subsequently declines due to external conflicts, invitations roll down sequentially to the next highest-ranking exam performer.
- The ACCP competition schedule typically initiates in mid-November, though system outages may temporarily delay national website schedule updates.
Foundations and Hierarchy of Observational Study Designs
Hierarchy of Clinical Evidence:
- Top Tier: Systematic reviews and meta-analyses combined with randomized controlled trials (RCTs).
- Second Tier: Analytical observational studies, consisting primarily of prospective and retrospective cohort studies and case-control studies (including nested case-control studies).
- Third Tier: Descriptive observational studies, including case reports and case series.
- Base Tier: Expert opinions and clinical consensus statements.
Descriptive versus Analytical Observational Studies:
- Descriptive Studies: Focus solely on describing health events, distributions, or observations without comparing groups or evaluating statistical hypotheses regarding treatment differences.
- Analytical Studies: Explicitly formulate and test hypotheses to evaluate statistical associations or relationships between defined exposures and clinical outcomes.
- Absence of Intervention: Unlike RCTs, investigators in observational studies do not assign treatment or intervene in care; they passively observe and systematically collect clinical data as events occur naturally.
Clinical Roles of Observational Studies in Drug Therapy:
- Real-World Effectiveness: RCTs evaluate drug efficacy in highly protocolized, ideal, and tightly controlled settings that lack generalizability to routine clinical practice. Observational studies assess real-world effectiveness across broad patient populations with variable adherence, comorbid conditions, and polypharmacy.
- Evaluating Drug-Induced Harm and Safety Risks: RCTs are fundamentally limited by ethical constraints, statistical power, and follow-up duration. Randomizing human subjects to potentially toxic or lethal exposures (e.g., randomizing individuals to cigarette smoking to assess oncogenesis) is unethical. Safety signals, rare adverse drug events, and long-term toxicity must be investigated through observational methods.
Inherent Limitations of Observational Designs:
- Susceptibility to Bias: Susceptible to selection bias (systematic errors in how subjects are chosen for inclusion or allocated to cohorts) and information/measurement bias (systematic errors in capturing exposure or outcome data).
- Confounding: Lack of random treatment assignment introduces both measured and unmeasured confounding variables, where the apparent association between exposure and outcome is distorted by underlying prognostic factors.
- Inability to Prove Causality: Observational designs can demonstrate statistical associations, but they can never definitively establish direct cause-and-effect relationships due to residual, unmeasured confounding.
Comparative Study Overview: Clopidogrel and Proton Pump Inhibitor Interactions
Core Clinical Question:
- Does the concomitant administration of a proton pump inhibitor (PPI) with the antiplatelet agent clopidogrel attenuate clopidogrel's clinical efficacy and lead to an increased risk of adverse cardiovascular outcomes in patients post-acute coronary syndrome (ACS)?
Study 1: Retrospective Cohort Study (Cole et al.):
- Title Focus: Risk of adverse outcomes associated with concomitant use of clopidogrel and proton pump inhibitors following acute coronary syndrome (ACS).
- Data Source: Veterans Affairs (VA) Medical Centers across the United States, capturing data across individual VA hospitals.
- Study Population: Patients discharged from a VA facility with documented acute myocardial infarction (AMI) or unstable angina (ACS) who received clopidogrel.
- Exposure Group: Clopidogrel therapy prescribed concomitantly with a PPI at hospital discharge or at any time during follow-up.
- Comparator Group: Clopidogrel therapy prescribed without a concurrent prescription for a PPI.
- Clinical Outcomes: All-cause mortality or rehospitalization for recurrent acute coronary syndrome.
Study 2: Nested Case-Control Study (Jorrelink et al.):
- Title Focus: A population-based nested case-control study evaluating the potential drug interaction between proton pump inhibitors and clopidogrel.
- Data Source: Comprehensive administrative health databases maintained by the Ontario Health Administration (Ontario, Canada).
- Study Population: Patients aged years or older discharged alive from an acute care hospital following treatment for an acute myocardial infarction who initiated clopidogrel therapy.
- Case Definition: Patients who died or were readmitted with recurrent myocardial infarction within days of discharge.
- Control Definition: Matched individuals selected from the same cohort who remained event-free (alive and not readmitted for MI) during the equivalent follow-up period.
- Exposure Definition: Current PPI exposure (defined by prescription dispensing within days prior to the index date) compared against non-use (no PPI dispensing within the preceding year).
Domain 1: Selection of Participants and Selection Bias
General Methodological Criteria for Selection Validity:
- Observational Validity Framework: Evaluated across core domains: selection of participants, measurement of variables, confounding control, and follow-up/temporality (in contrast to the domains utilized for RCT appraisal).
- Cohort Designs: Require clinically sensible inclusion and exclusion criteria. The exposed and unexposed cohorts must originate from the identical target population and must be completely free of the outcome of interest at baseline to verify that exposure precedes outcome onset.
- Strengthening Cohort Validity: The implementation of an active comparator design and/or a new-user design substantially reduces selection bias by preventing the inclusion of prevalent users who have already demonstrated tolerance to the therapy.
- Case-Control Designs: Cases must be identified objectively through reliable clinical registries or validated diagnostic/claims codes. Controls must be sampled from the identical source population that produced the cases, remaining at risk for the outcome during the risk window, and sampled independently of exposure status.
- Nesting Advantage: Performing a nested case-control study within a well-defined cohort guarantees that controls and cases share the identical underlying source population.
Participant Selection Appraisal: Retrospective Cohort (Cole et al.):
- Source Population: Discharges from VA medical centers for acute myocardial infarction or unstable angina.
- Cohort Size and Filling Completeness: A total of patients were identified; approximately of eligible participants filled their clopidogrel prescriptions directly through the internal VA outpatient pharmacy system, ensuring high capture rates of pharmacy records.
- Population Origin: Both PPI users and non-users were drawn from the identical underlying VA ACS registry.
- Absence of New-User Design: PPI exposure was defined as receipt of a PPI at discharge or at any time during longitudinal follow-up. Because the investigators did not restrict entry to incident (new) users, prevalent PPI users were mixed with new initiators, presenting a vulnerability to immortal time and prevalent-user biases.
Participant Selection Appraisal: Nested Case-Control (Jorrelink et al.):
- Inclusion Criteria: Older adults aged years or older who received inpatient treatment for an acute MI and filled a prescription for clopidogrel within the post-discharge period.
- Exclusion Criteria: Specific clinical and administrative exclusions included:
- Age younger than years (to ensure comprehensive prescription capture under the provincial universal drug coverage plan starting at age , thereby guaranteeing a -year baseline look-back window).
- Receipt of clopidogrel or other antiplatelet agents (e.g., ticlopidine) during the year prior to the index hospital admission.
- Residence in a long-term care institution.
- Receipt of a PPI regimen intended for Helicobacter pylori eradication (triple or quadruple therapy regimens).
- Case Identification: Any cohort member readmitted with recurrent myocardial infarction or dying within days post-discharge.
- Control Selection and Matching Ratio: Event-free controls were sampled from the risk sets of the identical parent post-MI cohort. The matching algorithm targeted up to controls per individual case ( matching ratio), dropping to fewer controls via a predefined matching algorithm only when insufficient exact matches existed.
- Selection Bias Risk Assessment: Minimal selection bias risk; cases and controls derived from the identical post-MI cohort under universal health system surveillance.
Domain 2: Measurement of Exposure and Outcomes (Information and Detection Bias)
Principles of Measurement and Avoidance of Bias:
- Objective Standards: High validity requires objective, prospectively recorded data sources: automated pharmacy dispensing records, validated diagnostic codes (e.g., ICD codes), and adjudicated procedural records applied identically across study arms.
- Recall Bias: Almost exclusively affects traditional retrospective case-control studies relying on patient interviews or surveys, where cases search their memories more rigorously than controls to explain their disease. Controlled by using electronic records or blinding participants to study hypotheses.
- Interviewer Bias: Occurs when data collectors systematically probe cases or exposed individuals with greater depth. Mitigated by using blinding or standardized, structured interviewing protocols.
- Detection (Surveillance) Bias: Occurs predominantly in cohort studies when the exposed cohort is subjected to more frequent medical visits, diagnostic testing, or clinical monitoring than the unexposed group, artificially elevating outcome discovery. Mitigated by blinding outcome adjudicators to exposure status and utilizing uniform surveillance protocols.
Measurement Appraisal: Retrospective Cohort (Cole et al.):
- Exposure Measurement: Longitudinal dispensing records from the centralized VA pharmacy database. Exposure episodes were categorized based on continuous refills, with discontinuation defined when a refill gap exceeded days between sequential supply periods.
- Outcome Measurement: Mortality was objectively identified via the VA Vital Status Files. Recurrent ACS rehospitalization was determined via manual medical chart reviews conducted across VA medical centers.
- Methodological Vulnerabilities:
- Detection Bias: Chart reviews were performed by unblinded clinical adjudicators who had access to exposure history; knowledge of PPI co-prescription may have influenced the diagnostic interpretation of ambiguous clinical chart notes.
- Missing Outcome Events: Any hospitalization or cardiovascular death occurring outside the VA system was not systematically captured, potentially biasing event counts.
- Exposure Misclassification: Over-the-counter (OTC) PPI purchases were not captured in VA prescription databases. Patients classified as non-users who acquired OTC PPIs were misclassified, diluting the observed relative risk toward the null.
Measurement Appraisal: Nested Case-Control (Jorrelink et al.):
- Outcome Identification: Hospital readmission for acute recurrent MI was captured via validated administrative inpatient discharge diagnosis codes.
- Exposure Identification: Validated outpatient prescription dispensing files from the Ontario Drug Benefit database.
- Risk Assessment for Information Biases: Interviewer bias and recall bias were completely eliminated because all data were captured electronically and prospectively through administrative billing and dispensing without human recall or subjective interviews.
- Potential Measurement Errors: Intermittent adherence or compliance failure could not be confirmed beyond the physical dispensing of the drug. Exposure to non-prescription OTC PPI formulations was absent from the database, introducing potential minor exposure misclassification.
Domain 3: Confounding Assessment and Mitigation Strategies
Systematic Evaluation of Confounding:
- Definition: Confounding arises when an extraneous variable is independently associated with the exposure, serves as a known risk factor for the outcome, and does not reside on the causal biological pathway between exposure and outcome.
- Step 1 (Table 1 Inspection): Baseline characteristics must be scrutinized to detect structural imbalances between comparison groups. Large sample sizes (e.g., or ) yield statistically significant -values () even for clinically meaningless differences (e.g., a or imbalance). Analysts must assess magnitude rather than relying exclusively on -values.
- Standardized Differences: A standardized difference below indicates acceptable prognostic balance across groups.
- Absolute Gap Rule of Thumb: If standardized differences are not reported, an absolute percentage gap of or less between exposed and unexposed groups generally reflects adequate balance, provided the covariate is not an exceptionally dominant clinical risk factor.
- Step 2 (Unmeasured Factors): Analysts must identify critical prognostic variables entirely absent from the datasets (e.g., over-the-counter medication use, smoking status, left ventricular ejection fraction, or disease severity).
- Step 3 (Design Mitigation): Study design methods to control confounding include restriction (excluding patients presenting with the confounder) and matching (pairing subjects on specific covariates).
- Step 4 (Statistical Mitigation): Analytical methods to control confounding include stratification (e.g., Mantel-Haenszel pooling) and multivariable regression modeling (e.g., multivariable logistic regression, Cox proportional hazards regression).
- Crucial Methodological Reality: Only true random assignment in RCTs balances both measured and unmeasured confounders. Every observational method operates exclusively upon known, captured covariates.
Propensity Score Methodology:
- Definition: A propensity score represents the conditional probability that an individual receives a specific treatment/exposure given their observed baseline characteristics, derived via multivariable logistic regression:
* *Function:* Reduces a high-dimensional vector of observed baseline covariates into a single scalar score, enabling direct matching or stratification between exposed and unexposed individuals sharing identical scores.
* *Limitations:* Subjects lacking an overlapping score in the comparison group (extreme tails of the distribution) are pruned, substantially reducing the analyzed sample size and diminishing study power. Furthermore, propensity scores provide zero protection against unmeasured or unknown confounders.
* *Demonstration Example:* A hypothetical cohort of patients with marked imbalances in congestive heart failure ( in exposed vs. in unexposed; gap) and ischemic heart disease ( vs. ) was subjected to propensity score matching. Following matching, the cohort was pruned to , successfully equalizing the distribution of CHF ( vs. \) and ischemic heart disease ( vs. \), despite residual nominal statistical significance due to the large sample size.
Stratification and the Mantel-Haenszel Technique:
- Mechanism: Divides the entire cohort into homogenous subgroups (strata) based on categories of the confounder, recalculating exposure-outcome associations within each independent stratum.
- Stratification Example: In a hypothetical case-control study evaluating the association between alcohol consumption and lung cancer, the crude unadjusted odds ratio is elevated at . Because smoking is an established confounder, the sample is stratified by smoking status:
- Smokers Stratum:
- Non-Smokers Stratum:
- Mantel-Haenszel Pooled Adjusted Ratio: The calculated summary odds ratio drops to , demonstrating that the apparent crude association was entirely an artifact of confounding by tobacco smoking.
Multivariable Regression Analysis:
- Mechanism: Incorporates multiple measured confounders as independent predictor variables within a unified mathematical model to isolate the adjusted effect of the primary exposure on the clinical outcome.
- Atrial Fibrillation Example: A study evaluating a specific risk factor for the incidence of atrial fibrillation identified an unadjusted crude odds ratio of . Following multivariable regression adjustment for age, comorbidities, and medications, the adjusted odds ratio dropped to (representing a attenuation in effect size), demonstrating substantial baseline confounding that was partially removed by mathematical adjustment.
Confounding Appraisal: Retrospective Cohort (Cole et al.):
- Baseline Imbalances (Table 1): Marked prognostic differences were evident between PPI users and non-users:
- Mean Age: years in non-users versus years in PPI users ().
- Diabetes Prevalence: in non-users versus in PPI users (absolute gap of ).
- Prior Myocardial Infarction: in non-users versus in PPI users (absolute gap of ).
- Heart Failure: in non-users versus in PPI users (absolute gap of ).
- Liver Disease: in non-users versus in PPI users ( gap; statistically significant but clinically minor).
- Design Methods: No restriction or matching was utilized at study inception.
- Analytical Methods: Investigators constructed multivariable logistic regression models adjusting for all baseline clinical covariates recorded in Table 1. Additionally, they performed secondary sensitivity analyses restricted to patients receiving a PPI at discharge or follow-up, and executed a multivariable time-varying Cox proportional hazards model adjusting for exposure status changes over time.
- Unmeasured Confounders: Gastrointestinal bleed history, over-the-counter NSAID/aspirin use, and the specific clinical indication/severity necessitating PPI initiation were not captured, leaving notable vulnerability to confounding by indication.
- Baseline Imbalances (Table 1): Marked prognostic differences were evident between PPI users and non-users:
Confounding Appraisal: Nested Case-Control (Jorrelink et al.):
- Interpretation Caveat for Case-Control Table 1: Imbalances between cases and controls reflect differences according to outcome status (recurrent MI vs. event-free survival), not exposure status. Clinical expertise must determine whether an outcome-associated variable also correlates with PPI exposure to constitute a true confounder.
- Baseline Characteristics (Table 1):
- Congestive Heart Failure: Highly elevated among cases ().
- Diabetes Mellitus: Elevated among cases ().
- Renal Insufficiency: Elevated among cases ().
- Sex Distribution: Male sex was in cases versus in controls (, representing a non-significant, balanced factor).
- Design Methods: Implemented rigorous restriction (age , antiplatelet naive for year, community-dwelling) alongside direct individual matching of cases to controls.
- Analytical Methods: Implemented conditional multivariable logistic regression models adjusting for established cardiovascular comorbidities and prognostic covariates.
- Unmeasured Confounders: Smoking status, outpatient lipid panels, systemic blood pressure measurements, and concurrent self-administration of over-the-counter aspirin (which reduces cardiovascular events but causes dyspepsia leading to secondary PPI use) were unmeasured.
Domain 4: Follow-Up Duration, Temporality, and Attrition
Follow-Up and Censoring in Cohort Studies:
- Length of Follow-Up: Must be clinically sufficient to allow biological development and clinical manifestation of the primary outcome (e.g., acute stent thrombosis occurs in days, while cardiovascular death unfolds over months to years).
- Censoring Rules: Cohort subjects are censored when they experience the study outcome, die from competing causes, withdraw, are lost to follow-up, or reach the scheduled termination date of the study.
- Attrition Bias: Unequal, differential loss to follow-up between the exposed and unexposed cohorts introduces substantial attrition bias, skewing relative risk calculations.
Temporality in Case-Control Studies:
- Absence of Attrition: Because case-control studies initiate with identified outcomes and look backward to evaluate historical exposure, loss to follow-up is conceptually non-existent.
- Establishing Temporality: Requires establishing that drug exposure preceded outcome occurrence. This necessitates establishing a strict index date (date of case event) and defining a clear, historic exposure window entirely prior to the index date to eliminate protopathic bias or reverse causality.
Follow-Up Appraisal: Retrospective Cohort (Cole et al.):
- Surveillance Window: Patient enrollment spanned discharges between October 2003 and January 2006, with administrative surveillance concluding in September 2006.
- Follow-Up Duration: The median post-discharge follow-up was (; interquartile range [IQR] of to , representing to ).
- Adequacy of Follow-Up: A median of provides sufficient observation time to capture recurrent ACS hospitalizations and mortality in a high-risk post-ACS cohort, though individuals at the lower boundary () had brief surveillance windows.
- Attrition Risk: The investigators failed to measure or report exact loss to follow-up, drop-out rates, or care migration outside the VA system, generating moderate-to-high methodological concern regarding unmeasured attrition.
Temporality Appraisal: Nested Case-Control (Jorrelink et al.):
- Timing Protocol: Patients were followed longitudinally for up to following initial ACS discharge to identify readmission for acute MI (index date).
- Exposure Window: Current PPI exposure was strictly defined as prescription dispensing occurring within the directly preceding the index date, with secondary windows assessing previous exposure ( to ) and remote exposure ( to ).
- Temporality Assurance: Clear temporality was established. Exposure was verified prior to the recurrent ischemic event, precluding exposure acquisition secondary to the outcome.
Interpreting Results: Magnitude, Significance, and Biological Gradients
- Measures of Association Across Observational Designs:
- Incidence and Ratios in Cohort Studies: Cohort studies follow populations over time and measure incident disease, permitting direct calculation of cumulative incidence, incidence rate ratios, relative risks (RR), and hazard ratios (HR) derived via Cox models.
- Odds Ratios in Case-Control Studies: Because the total denominator of the source population is unknown in case-control studies, absolute incidence cannot be computed. Case-control analyses are mathematically restricted to reporting odds ratios (OR).
- Rare Disease Assumption: When the cumulative incidence of an outcome in the population is low (), the calculated odds ratio mathematically approximates the relative risk:
Statistical Significance Criteria:
- Alpha Threshold: Assessed via a standard threshold of .
- Confidence Intervals ( CI): For ratio metrics (HR, RR, OR), statistical significance requires that the confidence interval does not cross or encompass the null value of . For absolute risk differences, the interval must not cross the null value of .
Evaluating Effect Size and Residual Confounding:
- Rule of Thumb for Association Strength: Relative risk or odds ratio estimates exceeding (or protective ratios below ) reflect moderate-to-strong associations that are difficult to explain away entirely by minor residual confounding or unmeasured bias.
- Weak Associations: Ratios ranging between and represent weak statistical associations that can be introduced by unmeasured residual confounding, subtle exposure misclassification, or baseline channel bias.
- Comparative Study Findings:
- Nested Case-Control Study (Jorrelink et al.): Adjusted (). Statistically significant, but an effect size of is weak and vulnerable to residual confounding.
- Retrospective Cohort Study (Cole et al.): Adjusted risk estimate . Both independent studies converged on an almost identical effect estimate ( to relative elevation in adverse cardiovascular risk).
Biological Gradients (Dose, Duration, and Recency Responses):
- Concept: Demonstrating that progressive increases in exposure magnitude, cumulative duration, or proximity correlate with increased risk supports biological plausibility.
- Duration-Response (Cole et al.): While formal daily dose-response relationships were not quantified, a clear duration-response was identified: for every incremental increase in the duration of concomitant PPI treatment, there was a corresponding increase in the risk/odds of adverse cardiovascular events.
- Recency-Response (Jorrelink et al.): Investigators categorized PPI use into current ( to prior to index date), previous ( to ), and remote ( to ) use. The association was only present among current users (), whereas previous and remote use showed no increased risk, providing evidence of a recency response gradient.
- Specificity of Interaction: Jorrelink et al. evaluated patients receiving PPI therapy without clopidogrel and demonstrated that PPI mono-therapy was not associated with recurrent MI, confirming that the elevated risk was specific to the co-administration of both agents.
Clinical Impact Measures: Number Needed to Harm (NNH) and Observational Limitations
- Absolute Risk and NNH Calculation in Cohort Designs:
- Restriction: Calculation of absolute risk reduction (ARR), absolute risk increase (ARI), number needed to treat (NNT), and number needed to harm (NNH) requires incidence data; therefore, these metrics can be computed from cohort studies, but cannot be derived directly from case-control studies.
- Mathematical Formula:
- Cohort Study NNH Derivation (Cole et al.):
- Based on crude unadjusted event rates across the median follow-up period, the calculated absolute risk increase yields:
* *Clinical Translation:* One additional death or recurrent ACS rehospitalization is projected to occur for every post-ACS patients co-prescribed clopidogrel and a PPI over an average treatment span of .
- Critical Methodological Caveat Regarding Observational NNT/NNH:
- Standard calculations of NNT and NNH utilize raw, crude baseline incidence figures.
- In the study by Cole et al., the unadjusted crude odds ratio was approximately , whereas multivariable adjustment for measured baseline covariates attenuated the estimate down to .
- Because the crude event rate difference contains the distorting influence of baseline confounding, an NNH derived from unadjusted data substantially overestimates real-world clinical harm.
- If fully adjusted, the true NNH would be substantially larger (e.g., dozens or hundreds of patients), and if the remaining association of is an artifact of residual confounding by indication, the true biological NNH is infinite (signifying no actual drug-induced harm).
Questions & Classroom Discussion
ACCP Clinical Research Challenge Inquiries:
- Student Inquiry: Is this competition part of ECCP or something distinct?
- Response: It is an annual national pharmacy research competition organized by the American College of Clinical Pharmacy (ACCP), centered on journal-club style critical evaluations of published literature.
- Scheduling Discussion: Class consensus confirmed the exam date for the afternoon of Friday, December 4, from to , following the scheduled PPE course session, during which no competing final examinations take place.
Appraisal of Study Features (Interactive Prompts and Student Responses):
- Prompt: What specific data source did Cole et al. utilize in their cohort study?
- Student Response: Veterans Affairs (VA) hospital medical centers ( regional centers).
- Prompt: What population and data source were captured in the nested case-control study by Jorrelink et al.?
- Student Response: Administrative healthcare databases of the Ontario Health Administration, analyzing older adults aged years or older discharged post-myocardial infarction.
- Prompt: How were cases and controls identified in the Jorrelink study, and what was the matching ratio?
- Student Response: Cases were defined as patients readmitted for recurrent MI or dying within post-discharge. Event-free controls were sampled from the same post-MI risk sets, targeting an algorithm of up to controls per individual case ().
- Prompt: What exposure measurement sources were used, and did they introduce interviewer or recall bias in the Jorrelink study?
- Student Response: Dispensing records from computerized prescription databases were utilized. Because objective electronic dispensing records were evaluated without human recall or interviews, recall bias and interviewer bias were not present.
- Prompt: Did the study designs account for over-the-counter (OTC) PPI use?
- Student Response: Neither study had the capacity to track OTC purchases, leading to potential misclassification of unexposed controls.
- Student Commentary & Side Inquiries: Incidental discussions addressed non-prescription aspirin acquisition as an unmeasured confounder, as well as an aside regarding student laboratory blood draws.