The case-time-control design: further assumptions and conditions

Area of Science:

• Epidemiology
• Biostatistics
• Pharmacovigilance

Background:

• Confounding by indication poses challenges in nonexperimental drug effect assessment.
• The case-time-control design aims to mitigate this bias by using subjects as their own controls.
• Unmeasured confounding factors can bias results when exposure varies over time.

Purpose of the Study:

• To review the assumptions underlying the validity of the case-time-control design.
• To address criticisms questioning the design's validity.
• To clarify the conditions for correct application of the case-time-control approach.

Main Methods:

• Analysis of the logistic model explicitly underlying the case-time-control design.
• Evaluation of Greenland's "counterexamples" against the established model.
• Modification of example data to conform to the case-time-control model.

Main Results:

• Greenland's counterexamples do not invalidate the case-time-control design as they do not adhere to its underlying logistic model.
• These counterexamples stem from an alternative model including a confounder by period interaction, which is excluded in the case-time-control model.
• Adjusted case-time-control estimates accurately reflect the true treatment effect (odds ratio of 2) when data conforms to the correct model.

Conclusions:

• The case-time-control design is valid when its underlying assumptions, particularly the avoidance of confounder by period interaction, are met.
• Clarification of matching necessity and discussion of conditional independence and carryover effects are provided.
• The study reaffirms the utility of the case-time-control design for unbiased assessment of drug effects.