Online Program

Real-Time Client Attrition Prediction in the Nurse Family Partnership Home Visiting Program (309717)

*Kaushik Mohan, Two Sigma 
Shannon Sainer, Nurse Family Partnership 
Dave Santin, Two Sigma 
Toby Schwed, Two Sigma 
Jordan Sciandra, Nurse Family Partnership 
Candice Yip, Two Sigma 

Keywords: machine learning, predictive modeling, attrition, retention, dropout

Nurse Family Partnership (NFP) is an evidence-based community health nursing intervention for low income, first time mothers consisting of in-home visits from Registered Nurses from pregnancy through the child’s second birthday. Studies show that clients who remain in the program longer receive the greatest benefit from NFP, making retention a critical determinant of participant outcomes. While randomized clinical trials of NFP have had relatively low client attrition rates; broader replication of the program has yielded attrition rates of around 60%. In seeking to increase program completion, NFP partnered with Data Clinic, the data for good initiative of Two Sigma, to assess current predictors of attrition and develop a real-time flag to help nurses identify clients at risk of dropping out. The study used program implementation data and de-identified client information for all clients enrolled in NFP from January 2002 to May 2020. To predict client attrition in real-time, the design matrix was expanded to include three types of predictors: static, dynamic, and lookback, for each client for each week in their program. For each of these weeks, the outcome variable is defined as whether the client’s last touchpoint with the program is within the next two months. A subset of predictors was identified for modeling by studying univariate correlations for each of the predictors with the outcome variable and a logistic regression model was built as a baseline. Preliminary results are consistent with other studies: mothers who are Black/African-American, younger, single, and have not yet completed high school are at higher risk of attrition, while mothers who complete more of their most recent five visits and those who have more social support at these visits tend to have lower risk of attrition. Next steps include further identification of these individual-level risk factors and delivering a final model to implement into NFP's systems.