| Self-controlled, within patient methodology has become increasingly common in pharmacoepidemiologic studies, partially attributable to its implicit adjustment for time-invariant confounders. Pairing traditional study designs (e.g. case-control, cohort) with within patient analysis (e.g. case-time-control, self-controlled case series) can enhance insight into a drug safety question given the contrasting strengths and weaknesses of these designs.;The case-control and the case-time-control share a similar sampling strategy and can be conducted using the same cases and controls. We demonstrated pairing of these designs, where both designs observed an approximate 2-fold increased rate ratio for acute kidney injury with fluoroquinolones. The cohort and self-controlled case series also share a similar sampling strategy. We demonstrated pairing of these designs, whereby both techniques identified an approximate 2-fold increased rate ratio for hypotension during the eight weeks after initiating and restarting tamsulosin therapy. In these studies, concordance of results with two analytic techniques strengthened the robustness of study conclusions.;Discordant results however can be challenging to interpret and leave uncertainty for the true association. We provide insight and explore potential explanations for discordance when using between and within patient designs as co-analytic techniques. We also quantified risk window bias in a simulation study to emphasize the importance of basing the duration of selected risk windows on prior clinical knowledge of the safety issue.;Some exposure-outcome associations will be more amenable to between patient analysis, others will be more applicable to within patient analysis, and many questions will be answerable by both techniques. Within patient analysis should be considered when there is no acceptable active comparator, the drug exposure is transient, the outcome onsets acutely, and between patient confounding is of greater concern than within patient confounding. Conversely, the ideal design is a between patient analysis when an active comparator is available, the exposure is chronic, the outcome is immediate or delayed, and time-varying confounding within patients is present to a greater extent than confounding between patients. Selecting an ideal study design is vital to elucidating an accurate exposure outcome association, and when appropriate, pairing within and between patient designs can enhance insight into a drug safety question. |
