A data mining approach in home healthcare: outcomes and service use

The purpose of this research is to understand the performance of home healthcare practice in the US. The relationships between home healthcare patient factors and agency characteristics are not well understood. In particular, discharge destination and length of stay have not been studied using a data mining approach which may provide insights not obtained through traditional statistical analyses.

The data were obtained from the 2000 National Home and Hospice Care Survey data for three specific conditions (chronic obstructive pulmonary disease, heart failure and hip replacement), representing nearly 580 patients from across the US. The data mining approach used was CART (Classification and Regression Trees). Our aim was twofold: 1) determining the drivers of home healthcare service outcomes (discharge destination and length of stay) and 2) examining the applicability of induction through data mining to home healthcare data.

Patient age (85 and older) was a driving force in discharge destination and length of stay for all three conditions. There were also impacts from the type of agency, type of payment, and ethnicity.

Patients over 85 years of age experience differential outcomes depending on the condition. There are also differential effects related to agency type by condition although length of stay was generally lower for hospital-based agencies. The CART procedure was sufficiently accurate in correctly classifying patients in all three conditions which suggests continuing utility in home health care.