Numerous studies on topics ranging from molecular to environmental determinants of health have shown that most humans tend to share key characteristics (e.g., comorbidities or genes) that form distinct patient subgroups. A primary goal of precision medicine is to identify such patient subgroups and infer their underlying disease processes in order to design interventions that are targeted to those processes. Because such targeted treatments can have a profound impact on patient outcomes, the identification and comprehension of patient subgroups is critical to patient-centered outcomes research (PCOR).
However, current approaches to identify patient subgroups are not designed to reveal relationships within and across patient subgroups, which constrains the ability of clinician and patient stakeholders to fully comprehend the processes underlying heterogeneity in a patient population. This project focuses on using a visual analytical method to 1) quantitatively identify the number, size, and statistical significance of patient subgroups and their most highly co-occurring characteristics; and 2) visualize that information through a network to reveal the relationships within and across patient subgroups. This approach is designed to enable stakeholders to infer the disease processes underlying each patient subgroup, with the goal of iteratively refining the variables to predict those subgroups.
We will demonstrate the utility of this method to address the urgent problem of predicting hospital readmission in the elderly. Over 2.3 million older Americans are readmitted within 30 days of being discharged from the hospital; of these readmissions, 75 percent are preventable, imposing a significant burden in terms of mortality, morbidity, and resource consumption nationwide. However, because the predictive power (measured by the C statistic) of current models to predict readmission is in the range of 0.60 to 0.65, there is considerable room for improvement.
Our project will 1) develop a computational method to automatically identify and visualize patient subgroups and their characteristics in datasets, such as from Medicare, and electronic medical records; 2) use the approach to identify patient subgroups in three index conditions—chronic obstructive pulmonary disease, congestive heart failure, and hip/knee arthroplasty—common in the elderly obtained from the Medicare database, and engage stakeholders to infer the disease processes underlying each patient subgroup, with the goal of refining the variables included in the analysis; 3) develop, validate, and test the improvement of regression models that incorporate patient subgroup information compared with the existing models in all three index conditions; and (4) use feedback from PCOR researcher stakeholders to operationalize the method in a software application with training, and to disseminate the method to other PCOR researchers.
- 3 chronic obstructive pulmonary disease (COPD) patients
- 3 congestive heart failure (CHF) patients
- 3 hip/knee arthroplasty readmitted patients
Other Stakeholder Partners
- Mukaila Raji, MD, MS, FACP, University of Texas Medical Branch at Galveston
- Sharma Gulshan, MD, MPH, University of Texas Medical Branch at Galveston
- Ravi Varadhan, PhD, John Hopkins University