Reducing the question burden of patient reported outcome measures using Bayesian networks

Area of Science:

• Health Informatics
• Medical Statistics
• Psychometrics

Background:

• Patient Reported Outcome Measures (PROMs) are crucial for assessing patient health status in learning health systems.
• High volume of PROM questions can lead to patient burden.
• Existing methods use item response theory for question selection in computerized adaptive testing (CAT).

Purpose of the Study:

• To extend CAT methods for efficient and accurate question selection in PROMs using Bayesian Networks (BNs).
• To leverage BNs for more comprehensive probabilistic modeling of PROM question relationships.

Main Methods:

• Utilized Bayesian Networks (BNs) to model probabilistic relationships between PROM questions.
• Applied information theoretic techniques for dynamic question selection via CAT.
• Validated the approach on five clinical PROM datasets.

Main Results:

• CAT using BNs achieved high accuracy (AUC 0.92) in predicting measured constructs with a subset of questions (30%-75%).
• BN-based CAT demonstrated comparable predictive performance to answering all PROM questions.
• BNs improved prediction accuracy for unanswered items by an average of 5% compared to alternative CAT methods.

Conclusions:

• Bayesian Networks provide an effective framework for Computerized Adaptive Testing (CAT) in Patient Reported Outcome Measures (PROMs).
• BN-based CAT significantly reduces patient burden by selecting fewer, more informative questions.
• This approach enhances the efficiency and accuracy of health status assessment using PROMs.