Estimating causal effects using prior information on nontrial treatments

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Epidemiology

Background:

• Randomized controlled trials (RCTs) face analytical challenges due to deviations from assigned treatments.
• Intention-to-treat (ITT) analysis estimates treatment allocation effects, but estimating actual treatment received effects is complex.
• Existing methods for estimating treatment received effects are limited, primarily to trials with switches between assigned treatments.

Purpose of the Study:

• To develop a novel method for comparing treatment effects in active-controlled trials with non-trial treatments and no treatment.
• To apply and illustrate this method using data from the PENTA 5 trial in HIV-infected children.

Main Methods:

• Combines the instrumental variables approach with prior information on treatment effects.
• Utilizes prior information on the distribution of effects for non-trial treatments and one trial treatment.
• Elicits prior information from investigators, as demonstrated in the PENTA 5 trial.

Main Results:

• Prior information in the PENTA 5 trial indicated benefits from all treatments, with some uncertainty in the magnitude.
• Incorporating prior information altered point estimates and increased standard errors compared to analyses ignoring non-trial treatments.
• The method allows for adjustment in complex departure patterns from randomized treatments.

Conclusions:

• The proposed method relies on correct specification of causal treatment effects, which requires sensitivity analyses for validation.
• Prior information, ideally from literature, enhances the analysis of RCTs with treatment deviations.
• This approach is valuable for RCTs involving non-trial treatments or where trial treatments are not universally adopted.