Questions: Pharmacoepidemiology: Drug Safety and Adverse Event Surveillance

Question 1

A new antidepressant is associated with a doubled risk of type 2 diabetes in a large insurance claims cohort compared to non-users of antidepressants. Which methodological threat most plausibly explains this finding even if the drug has no real effect?

• A. Immortal time bias — the period before first prescription is misclassified as exposed
• B. Confounding by indication — depression itself increases diabetes risk, and depressed patients are more likely to receive the drug
• C. Disproportionality inflation — the database is too large to detect true null associations
• D. Channeling bias — the drug is preferentially prescribed to younger, healthier patients

Answer: (B)

Confounding by indication is the central threat in pharmacoepidemiology: the disease being treated is itself a risk factor for the outcome, making the drug appear harmful even if it is not. Depression independently raises diabetes risk via metabolic and behavioral mechanisms. The comparison group (non-users) is systematically different from drug users in a way that predicts the outcome. An active comparator design — comparing the drug to another antidepressant rather than to non-users — is the standard methodological fix.

Question 2

A spontaneous reporting analysis finds that Drug X has a proportional reporting ratio (PRR) of 8.5 for a specific cardiac arrhythmia, far exceeding the threshold used for signal detection. What does this finding establish?

• A. Drug X causally increases the risk of this arrhythmia in the general population
• B. Drug X should be immediately withdrawn from the market
• C. The drug-event pair is reported more often than expected by chance, warranting further investigation
• D. The absolute risk of arrhythmia in Drug X users is 8.5 times higher than in non-users

Answer: (C)

Disproportionality analysis in spontaneous reporting detects statistical signals — drug-event pairs reported more often than chance would predict — but cannot establish causation. The database has no denominator (you don't know how many patients took the drug without incident), voluntary reporting biases the sample, and confounding is uncontrolled. A strong PRR is a hypothesis-generator that triggers regulatory investigation, not a causal verdict. Interpreting it as a relative risk (option D) confuses signal strength with incidence ratio.

Question 3

Spontaneous adverse event reporting systems (like FDA's FAERS) cannot calculate true incidence rates of adverse drug reactions.

• T. True
• F. False

Answer: True

Spontaneous reporting systems lack a denominator: they record the number of adverse event reports filed, but not the number of people who took the drug without incident. Without knowing the total exposure, you cannot compute incidence (events per exposed person-time). This is a fundamental structural limitation, not a data-quality problem. It is why disproportionality analyses compare reporting rates within the database rather than computing absolute risks.

Question 4

In a large pharmacoepidemiology database study, a very high odds ratio (e.g., OR = 15) provides stronger evidence of causation than a modest odds ratio (e.g., OR = 1.5) because statistical association implies causation at sufficient magnitude.

• T. True
• F. False

Answer: False

In large administrative databases, statistical association is almost guaranteed to be achievable for any drug-outcome pair if you search long enough — the challenge is not detecting association but distinguishing true causal effects from confounding, selection bias, and multiple comparisons. The Bradford Hill criterion of 'strength of association' does play a role in causal assessment, but it cannot substitute for ruling out confounding. A very large association could still be entirely explained by confounding by indication. The discipline's goal is causal inference under observational constraints, not effect-size maximization.

Question 5

Why does confounding by indication pose a particular challenge in pharmacoepidemiology, and how does an active comparator design address it?

Model answer: Confounding by indication occurs because the disease for which a drug is prescribed is itself often a risk factor for the outcomes being studied — so sicker patients receive certain drugs, making the drug appear harmful even if it is neutral or protective. Simply comparing drug users to non-users conflates drug effect with disease effect. An active comparator design addresses this by comparing the drug of interest to another drug prescribed for the same indication — so both groups have the same underlying disease. This balances the confounding variable (disease severity) across exposure groups, isolating the drug's effect more cleanly.

This is the defining methodological challenge of pharmacoepidemiology because the very purpose of a drug (treating a disease) makes non-users a systematically invalid comparison group. Active comparator designs have become standard practice in the field because they eliminate the most severe form of indication confounding, though they do not solve all confounding problems — residual confounding from drug selection within the same indication class can remain.