The balanced survivor average causal effect.

Area of Science:

• Biostatistics
• Clinical Trials
• Longitudinal Data Analysis

Background:

• Longitudinal data analysis is complicated by patient death before scheduled measurements, a phenomenon known as truncation by death.
• The survivor average causal effect (SACE) is a common estimand but relies on untestable assumptions, such as treatment monotonicity.

Purpose of the Study:

• To introduce a novel estimand, the balanced-SACE, for longitudinal outcomes in the presence of death.
• To propose a new estimator for the balanced-SACE that does not require the monotonicity assumption.

Main Methods:

• The balanced-SACE is defined on a subset of always-survivors balanced for potential survival times.
• A simple estimator is proposed by comparing outcomes in equivalent fractions of longest-surviving patients between treatment and control groups.
• Bias expressions, sensitivity analyses, and bootstrap inference are discussed.

Main Results:

• The balanced-SACE provides an alternative to SACE without relying on monotonicity.
• The proposed estimator offers a practical approach to analyzing data truncated by death.
• Sensitivity analyses help assess the robustness of the findings.

Conclusions:

• The balanced-SACE and its estimator offer a valuable tool for causal inference in longitudinal studies with informative censoring.
• This approach relaxes restrictive assumptions, enhancing the applicability of causal inference methods in clinical research.