IDENTIFY CHF AND COPD PAITIENTS AT HIGH RISK OF HOSPITALIZATION- USING PREDICTIVE ANALYTICS FOR PATIENT OUTREACH
Author(s)
Wang QC, Higgins SL, Chawla R, Nigam S
Independence Blue Cross, Philadelphia, PA, USA
Presentation Documents
OBJECTIVES: To develop predictive models to identify CHF and COPD patients at high risk of hospitalization within the next 6 months to be used for case management outreach. DATA SOURCES Data were extracted from several sources and included patient diagnoses, service utilization, lab data, and medication adherence from a large health insurance claims database, ZIP code level demographic data from the U.S. Census, patient level illness burden scores, medical episode groupers, Experian consumer and credit information, and call data between patients and customer service representatives. STUDY POPULATION All commercial and Medicare members who were identified with congestive heart failure (CHF) and chronic obstructive pulmonary disease (COPD) and who were continuously enrolled for at least 6 months during the model development period and 6 months during the predictive period were included. METHODS: Using three years of historical data from 2010 to 2012 and admissions between January and June 2013 as the target variable, the data were randomly split in half as training and validation data. The training data were used to build the predictive model. The validation data were used to evaluate model performance. Several algorithms were utilized to build predictive models: logistic regression, neural networks, and decision trees. The models were evaluated based on the lift chart and/or area under the ROC curve. The selected models were used to score data and predict future admissions. RESULTS: The key factors predicting admissions in the next 6 months included length of time identified with CHF and COPD, medication adherence, prior admissions, recent specialist visits, having had a customer call that mentioned ‘hospital bed/hospital stay’, and being on oxygen (for COPD). CONCLUSIONS: Four models of predicting patients at highest risk of admission have been developed, which were used to generate a list of patients with high probability of admission for case management outreach.
Conference/Value in Health Info
2014-05, ISPOR 2014, Palais des Congres de Montreal
Value in Health, Vol. 17, No. 3 (May 2014)
Code
PRM118
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Cardiovascular Disorders, Respiratory-Related Disorders