Predicting potentially avoidable hospitalizations.

Area of Science:

• Health Services Research
• Predictive Analytics
• Healthcare Management

Background:

• Ambulatory care sensitive conditions (ACSCs) hospitalizations indicate primary care access and effectiveness.
• Resource limitations make universal early intervention challenging.

Purpose of the Study:

• Develop a predictive model for identifying high-risk patients for early intervention.
• Assess the predictive power of various patient variables for ACSC hospitalizations.

Main Methods:

• Utilized hierarchical logistic regression with a random intercept on a large VA patient cohort (n=2,987,052) from FY2011-2012.
• Employed a random split-sample method for model development and validation to prevent overfitting.
• Assessed predictive contributions of demographic, socioeconomic, prior utilization, and Hierarchical Condition Category (HCC) data.

Main Results:

• The full model achieved high predictive accuracy, with c-statistics of 0.856 for 90-day and 0.835 for 1-year ACSC hospitalizations in validation samples.
• Demographic and socioeconomic variables improved prediction (c-statistic 0.721), while prior utilization and cost further enhanced it (c-statistic 0.826).
• Hierarchical Condition Categories (HCCs) were the final addition to the full predictive model.

Conclusions:

• Administrative data demonstrate significant effectiveness in predicting ACSC hospitalizations.
• The developed model possesses high predictive ability, supporting primary care providers in identifying at-risk patients.
• Early intervention guided by this model can effectively reduce ACSC hospitalizations.