Cortex Health
Cortex Health is a healthcare technology startup that builds software to assist post-acute care facilities, such as managing and tracking nurse checkup calls and inpatient hospital readmission alerts to improve customer satisfaction and reduce future hospital readmissions.
Problem
Hospital readmission is one of the primary ways that the flow of patients into the system is disrupted and a backlog builds up. Before a patient might check back into a hospital, however, they are often followed up about their care facilities and experience via a phone call with a nurse. Cortex Health works to reduce this pain point for its consumers and improve patient flow management for an improved post-care experience. As a result, Cortex Health aimed to predict how likely a patient is to be emergency re-admitted to a hospital.
Methodology
Our goal was to use predictive modeling to create patient emergency hospital readmission risk scores to help hospitals identify high-risk patients and better allocate resources. Two significant challenges we faced were manual, non-standardized data and widely varying quality across records. We performed extensive data cleaning and feature engineering in collaboration with the Cortex team to better understand the context of the healthcare data, and created timestamp-based data points for modeling. We built and tuned logistic regression, random forest, and CatBoost models, and evaluated the models based on the f-beta(1.5) score to weigh recall more heavily than precision, based on the client's preferences.
Results
We delivered a tuned CatBoost model with modified class weights to account for data imbalance. The final model will be implemented in their software to provide improved readmission indicators over the traditional point-based LACE index, which used only 4 variables.
Semester
Fall 2021