Combination of longitudinal biomarkers in predicting binary events

Area of Science:

• Biostatistics
• Medical Diagnostics
• Longitudinal Data Analysis

Background:

• Single biomarkers have limited diagnostic accuracy in disease screening.
• Longitudinal biomarkers, tracking changes over time, offer potential for improved diagnostic performance.
• Combining multiple biomarkers can enhance predictive power compared to individual markers.

Purpose of the Study:

• To propose a novel pattern mixture model (PMM) framework for predicting binary disease status using longitudinal biomarker data.
• To evaluate the performance and robustness of the PMM compared to existing models.
• To apply the PMM to predict fetal macrosomia using longitudinal ultrasound measurements.

Main Methods:

• Developed a pattern mixture model (PMM) framework for binary disease prediction from longitudinal biomarker sequences.
• Estimated marker distribution using a linear mixed effects model.
• Utilized a likelihood ratio statistic for biomarker combination, optimizing the receiver operating characteristic (ROC) curve.
• Derived individual disease risk scores and 95% confidence intervals via Bayes' theorem.
• Compared PMM performance against the shared random effects (SRE) model.

Main Results:

• The PMM framework provides an approximation to the shared random effects (SRE) model.
• Extensive simulations demonstrated that the PMM is more robust than the SRE model across various model specifications.
• The PMM approach was successfully applied to a fetal growth study for predicting macrosomia.

Conclusions:

• The proposed pattern mixture model (PMM) offers a robust and accurate method for disease prediction using longitudinal biomarkers.
• PMM enhances diagnostic accuracy by effectively combining multiple longitudinal markers.
• This framework has practical applications in clinical settings, such as predicting fetal growth complications.