PEBP 5.4 Medication Exposure Episodes

Definition of Exposure

• Broadly defined as any factor associated with an outcome.
• Can have different forms, including behavior, treatment (e.g., medication), trait, social condition, or comorbid disease.
• Pharmacoepidemiology often focuses on pharmacologically active substances.

Medication Exposure Episodes

• Steps to consider when constructing:
  • Develop a research question and objective.
  • Choose an appropriate exposure data source.
  • Check data quality and curate data.
  • Define and build treatment episodes.
• Focus is on how exposure data is constructed

Data Source Types

• Many exist, but mostly administrative healthcare claims databases and electronic health records are used.
• Exposure data can also be found in registries and surveys.
• Healthcare claims and health records are most frequently used in Pharmacoepidemiology studies

Measuring Exposure

• Exposure can be determined through self-reported data, prescription claims using national drug codes and drug classification systems, and EHR data like MAR records.
• Prescribing and dispensing data do not measure whether the patient took the medication.
• Errors in determining correct exposure lead to misclassification.

Recall Bias

• This is caused by participants not remembering previous events or experiences.
• This is a problem in studies that use self-reporting, like case-control studies.
• Patients with the outcome may be more likely to remember exposure, artificially inflating effect estimates.

Exposure Data in Healthcare Claims Databases

• Typically have information for outpatient pharmacy dispensing and outpatient (+ sometimes inpatient) medication purchased/administered by healthcare providers.
• Information typically includes the National Drug Code, days of supply, quantity, and dispensing/administration date.
• Advantages include no recall issue, precise information like agent name, strength and supply, and many well-performing/validated algorithms.
• Disadvantages include uncertainty about the intake, leading to possible exposure misclassification, and not capturing agents not dispensed in pharmacies.

Exposure Data in Electronic Health Records

• Contains structured data (patient characteristics) and unstructured data (free text describing patient history).
• Can be in unstructured/structured data and includes prescription data for outpatient and dispensing data for inpatient settings.
• Advantages include no recall issue, precise product information, potential capture of OTC and illicit drugs in unstructured data, and more granular information in inpatient settings.
• Disadvantages include uncertainty about intake/dispensing in outpatient settings and potential missing information if restricted to one healthcare system.

Self-Reported Exposure Data

• Includes surveys and registries.
• Data is highly dependent on survey design.
• Can be very precise with linked terminology or imprecise with free text entries.
• The advantage is that it can capture non-prescription drugs.
• The disadvantage is the susceptibility to recall issues, the need for additional labor to map exposure data, and potential variation depending on the study setting

Identifying Exposure in Observational Studies

• Need to identify treatment episode, i.e., the time someone took a medication to construct exposure windows.
• Exposure time window is the time we expect the medication to have an effect.
• While the patient is taking the medication, the exposure time window is some time after e.g., 6 months after

Induction & Latency Period

• Induction period: time from completion of causal effects of exposure to the start of the event or outcome.
• Latency period: time from start of the outcome until its identification.

Exposure Misclassification

• Can significantly impact effect estimates.
• If medication exposure increases the risk for a certain adverse outcome where 10% are among groups with exposure and 5% are among groups without exposure, the true relative risk is 2.0.
• If only 50% of patients assigned to the exposure group were actual medication users, the outcome incidence for the exposure group would be 7.5%.
• The result of the misclassification will lead to an observed biased relative risk of 1.5
• The impact depends on the type and scope of the misclassification.
• Differential misclassification biases either towards or away from the null.
• Non-differential misclassification usually biases toward the null (exemption, non-binary exposure).
• Data regarding single prescriber, pharmacy dispensing, and claims databases affects potential exposure misclassification

Sources of Misclassification

• Prescriptions may not be filled
• Medical charts in institutional settings
• Pharmacy databases
• Patient non-adherence
• Surveys / patient interviews
• Prospective ascertainment
• Unavailable variables

Outcomes in Observational Studies

• Need to revisit internal validity, clinical endpoints, surrogate outcomes, composite outcomes, patient-reported outcomes, and subjective vs. objective outcomes.
• Errors in determining correct outcomes leads to Misclassification
• Important to define data collection approach (Prospective or Retrospective)
• If retrospective, consider data source specific
• In general, in Observational studies, there is an increased level of uncertainty regarding outcome ascertainment, so it must be validated
• Outcome misclassification can lead to seriously biased effect estimates.

Measuring Outcomes

• This is done using self-reported data, medical claims data and electronic health records. and vital statistics.
• Errors in determining the correct outcome leads to misclassification.

Sources of Outcome Misclassification

• Inaccuracy of diagnostic procedures / decisions
• Severity of outcome may vary by quality of care
• Medical records
• Diagnostic bias
• Administrative databases
• Unavailable variables

Reducing Misclassification

• This is done using validated outcomes with high specificity and sensitivity
• Make sure outcomes are validated in the same database and patient population.

Key Considerations for Outcome Selection

• Two hypothetical examples are provided to showcase approaches
• Follow-up is insufficient
• The mechanism is not reflective
• Outcome is reliably ascertainable
• Variables have to be clearly defined

Key Takeaways

• Exposure & Outcome measurement in observational studies is research question-specific.
• Database (including the data generation process) and consideration of exposure-outcome mechanisms require special consideration when constructing exposure time windows.
• Misclassifying exposure & outcome can result in significant bias.
• Critical to be cautious when assessing the appropriateness of exposure and outcome measures in observational studies