Grant: Project Grant

Patient-reported outcome measures (PROMs) are designed to be used by patients to describe their health, including their quality of life, pain, fatigue, and functional abilities. PROMs can also be used by clinicians, health professionals, and healthcare decision makers to describe the outcomes of care from the patient's point of view. In order to contribute to improving patient care, PROMs must be valid (i.e., measure what they are supposed to measure) and reliable (i.e., measure the same thing over time and for patients with different characteristics). Sometimes, patient responses on PROMs are not consistent with what was expected. These inconsistencies may occur because patients with different characteristics may not understand or interpret questions about their health in the same way (differential item functioning), or their interpretations may change over time (response shift). While these differential interpretations and response shifts in PROMs data could be a result of positive adaptations to health challenges, such as a chronic illness or a healthcare treatment, they are often difficult to detect and, if ignored, could lead to incorrect conclusions about the outcomes of care. Our research focuses on data-driven machine-learning methods to efficiently cluster (i.e., group) patients with similar patterns of differential item functioning or response shifts. We will develop and apply machine-learning item response methods and compare them to model-based item response methods using computer simulation. We will apply the proposed machine-based methods to existing data for patients with specific health conditions to predict healthcare use. The research will be shared with healthcare providers, PROMs developers, and researchers. The overall goal of our study is to improve the ability to accurately interpret patient-reported information when measuring the effectiveness and responsiveness of the healthcare system.