Α Markov model for longitudinal studies with incomplete dichotomous outcomes

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Longitudinal Data Analysis

Background:

• Missing outcome data in randomized controlled trials (RCTs) compromise study validity and precision.
• Incomplete individual patient data (IPD) is a common challenge in longitudinal studies.
• Accurate analysis of longitudinal binary data is crucial for reliable clinical trial conclusions.

Purpose of the Study:

• To propose and validate a multistate Markov model for analyzing incomplete longitudinal binary data from RCTs.
• To address the threat posed by missing outcome data to the validity of inferences.
• To provide a robust method for handling dropouts and relapses in time-to-event data.

Main Methods:

• Utilized a multistate Markov model to analyze dichotomous outcomes over time.
• Accounted for patient dropouts and relapses within the study period.
• Incorporated time of observation and allowed for time-dependent relative treatment effects.

Main Results:

• The multistate Markov model successfully analyzed incomplete individual patient data for a dichotomous outcome.
• The model jointly estimated relative treatment efficacy and dropout rates.
• Demonstrated the ability to make clinically relevant inferences, including time-dependent treatment effects.

Conclusions:

• The proposed multistate Markov model is a viable method for analyzing longitudinal, incomplete binary data.
• This approach enhances the precision and validity of inferences from randomized controlled trials with missing data.
• The model offers flexibility in incorporating assumptions about missingness mechanisms and unobserved outcomes.