Table 1.

Summary of available methodologies^a

Data source	Experimental strategies	Statistical methodologies	Strengths	Weaknesses
Spontaneous reports	Passive Reporting System (FDA)	Proportional Reporting Ratio	Large enough numbers to measure rare adverse events.	Confounding by indication
	Active Reporting System (VA)	Bayesian Neural Networks		Systematic underreporting
		Empirical Bayes screening		Questionable representativeness
		Multi-Item Gamma Poisson Shrinker		Publicity bias
		Cumulative Sum		Extreme duplication
		Random-Effects Poisson Regression		Unknown population at risk
Ecological methods	National rates	Time series methods	Large samples or entire populations can be studied.	Do not know if person experiencing the AE actually took the drug.
	Natural experiments	Change-point analyses	Permits between stratum comparisons	Subject to ecological fallacy
	Small area estimation	Mixed-Effects Poisson Regression	Hypothesis generation	Geographic variability in reporting
Meta-analysis	Synthesis of randomized controlled trials	Fixed-Effects model (MH model)	Randomization	Limited generalizability
	Synthesis of observational studies	Random-Effects model (DL model)	Person-level	Exclusion of zero-event studies
		Mixed-Effects Logistic Regression		Heterogeniety
		Multilevel mixture models		Publication bias
				Ascertainment bias
Medical claims	Case-control studies	Fixed-Effects Logistic and Poisson	Large samples	Confounding by indication
	Cohort studies	Mixed-Effects Logistic and Poisson	Person-level	Confounding by time of treatment
	Within-subject designs	Person-time logistic models	Concomitant medications	Unmeasured confounders
	Between-subject designs	Propensity scores	Comorbid diagnoses	Unsystematic diagnostic criteria
	Matching		Prior history of relevant AEs	Based on filled prescriptions only
	Differential effects		Generalizable	Limited dosage and duration data
	Coherence

^aAbbreviations: AE, adverse events; DL, DerSimonian and Laird; MH, Mantel-Haenszel; RCT, randomized controlled trial.