•  
  •  
 

Article Title

Predicting the Risk of Emergency Department Visits in Medicaid Members: Development and Temporal Validation of a Model

Publication Date

8-15-2016

Keywords

prediction modeling, emergency department

Abstract

Background/Aims: We developed and validated a model to predict the risk of emergency department (ED) visits in adult Medicaid members so that case-managers can identify the highest-risk patients and intervene. We then validated the model on newly enrolled Medicaid patients.

Methods: To develop the prediction model, we assembled a retrospective cohort of adult Medicaid members (18–64 years old) enrolled at Kaiser Permanente Northwest between 2010 and 2013. We measured patient characteristics during the 90 days before the start of follow-up that might predict ED visits. We followed patients for up to 180 days to identify the first ED visit. To validate the model, we assembled a distinct cohort of adult Medicaid members who joined Kaiser Permanente Northwest in the first quarter of 2014. We developed and validated separate models for men and women using Cox regression.

Results: We observed 2,587 patients who visited the ED during the 180-day follow-up. The overall 180-day risk of an ED visit was 13.9 per 100 (men) and 17.4 per 100 (women). The models discriminated the high- and low-risk patients adequately: concordance or c-statistic was 0.72 (men) and 0.71 (women), respectively. The model’s 10 predictor characteristics explained 35.2% of the variation in ED visits in men and 29.6% of the variation in women. Model calibration (agreement between observed and predicted) revealed that the mean predicted risks in the highest-risk patients underestimated the observed risks of an ED visit by approximately 11 per 100. The model for women validated adequately in the newly enrolled cohort because the c-statistic remained constant while the model for men disappointed because the c-statistic dropped by 0.05. For both men and women, the models continued to underestimate the absolute risk of an ED visit in the highest-risk patients.

Conclusion: The models identified the highest-risk patients with only 90 days of clinical history, and the models validated on new Medicaid patients. Time invested in managing the highest-risk patients may offer a superior return on investment compared with a strategy that does not stratify because the highest-risk patients suffer a disproportionate excess risk. The return on time invested may be even higher if recurrent ED visits are considered.

Share

COinS