Using AI to identify feature importance for no-show appointments in substance abuse treatment settings

Title : Using AI to identify feature importance for no-show appointments in substance abuse treatment settings

Abstract:

In substance abuse settings no-show rates are correlated with higher mortality rates.  The reasons behind no rates are multifactored and complex, limiting insights gained from traditional statistical models.  Machine learning models can handle more features but are difficult to explain.  Therefore, we propose using explainable artificial intelligence to take advantage of machine learning’s ability to integrate more features but also allow for explainability through a 10-step process.   We demonstrate this 10-step process with EHR data from outpatient substance abuse clinics evaluating factors influencing no-show rates.

Audience Take Away Notes:

• How explainable artificial intelligence (xAI) can be used to identify directional and interpretable impact from features in machine learning models.
• How xAI can help them identify the directional impact of features from machine learning models.
• How relapsing patients are most at risk for no-show appointments.

Biography:

Dr. Hanauer studied Research Methods at Indiana University – Bloomington.  He worked for Centerstone, one of the nation’s largest non-profit mental health providers, for 3 years as a Biostatistician, first authoring several research articles and presenting at dozens of national conferences nationwide.  He now works as a Senior Data Scientist at CleanSlate Centers, where he builds prediction models for no-show rates, patient relapse, and NLP models to support call center analytics.  Additionally, he provides statistical and machine learning consulting services for universities and other healthcare clients.