Case Study: Machine Learning Model Using Insurance Claims Data to Help Predict ALS Diagnosis

The Situation

Amyotrophic lateral sclerosis (ALS) often faces significant diagnostic delays. Utilizing machine learning methods on large healthcare databases has the potential to create models that can expedite ALS diagnosis, leading to improved quality of life and potentially an improved survival rate for ALS patients.

A small pharmaceutical company wanted to find a way to shorten the time to diagnosis to get treatment to patients faster. It approached EVERSANA to develop a predictive disease model that could identify future ALS diagnoses by analyzing the pre-diagnosis medical histories of ALS patients within a large insurance claims database.

The Solution

Diagnosing ALS is inherently complex and may be inconsistently documented in medical claims data. Designing a high-quality predictive model for ALS requires a blend of clinical and technical expertise to handle data processing and feature engineering effectively.

A retrospective analysis was performed on EVERSANA’s claims database for patients diagnosed with ALS. Using mutual information (MI), the study compared unique clinical features of the ALS cohort with those of demographically matched control patients. The top differentiating features were then used to train a classifier to predict ALS diagnoses. This model was subsequently applied to identify patients in the database who had clinical profiles similar to those in the ALS cohort but had not yet been diagnosed.

The Results

A model comprised of diagnoses, procedures and drugs applied to a medical administrative claims database predicted ALS diagnosis three to 12 months prior to actual diagnosis with positive predictive values (PPVs) that exceed the actual prevalence of ALS. The model was able to provide the client with the tools to predict ALS with higher PPVs than could be obtained by predicting ALS based solely on any individual precursor diagnoses.