Predicting emergency department visits and hospitalizations for patients with heart failure in home healthcare using a time series risk model

Abstract

Objectives: Little is known about proactive risk assessment concerning emergency department (ED) visits and hospitalizations in patients with heart failure (HF) who receive home healthcare (HHC) services. This study developed a time series risk model for predicting ED visits and hospitalizations in patients with HF using longitudinal electronic health record data. We also explored which data sources yield the best-performing models over various time windows.

Materials and methods: We used data collected from 9362 patients from a large HHC agency. We iteratively developed risk models using both structured (eg, standard assessment tools, vital signs, visit characteristics) and unstructured data (eg, clinical notes). Seven specific sets of variables included: (1) the Outcome and Assessment Information Set, (2) vital signs, (3) visit characteristics, (4) rule-based natural language processing-derived variables, (5) term frequency-inverse document frequency variables, (6) Bio-Clinical Bidirectional Encoder Representations from Transformers variables, and (7) topic modeling. Risk models were developed for 18 time windows (1-15, 30, 45, and 60 days) before an ED visit or hospitalization. Risk prediction performances were compared using recall, precision, accuracy, F1, and area under the receiver operating curve (AUC).

Results: The best-performing model was built using a combination of all 7 sets of variables and the time window of 4 days before an ED visit or hospitalization (AUC = 0.89 and F1 = 0.69).

Discussion and conclusion: This prediction model suggests that HHC clinicians can identify patients with HF at risk for visiting the ED or hospitalization within 4 days before the event, allowing for earlier targeted interventions.

Keywords: electronic health records; heart failure; home care services; machine learning; natural language processing; nursing informatics.

Figure 1.

Overview of study methods.

Figure 2.

F1 score of risk prediction models when adding the different sets of variables. Set 1: OASIS only. Set 2: OASIS+vital signs. Set 3: OASIS+vital signs+visit characteristics. Set 4: OASIS+vital signs+visit characteristics+NLP variables. Set 5: OASIS+vital signs+visit characteristics+NLP variables+TF-IDF variables. Set 6: OASIS+vital signs+visit characteristics+NLP variables+TF-IDF variables+Bio-Clinical BERT variables. Set 7: OASIS+vital signs+visit characteristics+NLP variables+TF-IDF variables+Bio-Clinical BERT variables+topic modeling variables.

Figure 3.

F1 score of ED visit and hospitalization risk prediction for different time windows.

Figure 4.

Performance of prediction model to predict emergency department visits and hospitalizations within 4 days. (Left) Receiver-operating characteristic curves. (Right) precision-recall curves.

Figure 5.

Twenty variables associated with risk for ED visits and hospitalizations using LASSO. The x axis represents the log of the L1 penalty parameter (alpha), and the y axis represents the coefficient values of the predictors in the model. The L1 penalty parameter shows the strength of the regularization, and as the value of alpha increases, the coefficients shrink toward zero. Each line in the plot represents a different predictor in the model, and the slope of the line represents the change in the magnitude of the coefficient as alpha increases. Predictors with nonzero coefficients at high values of alpha are considered more important, while predictors with zero coefficients are less important. We generated several variables of TF-IDF, describing lexical features of the text, including the “ratio of nonalphanumeric symbols to text length” and “ratio of numeric digits to text length.”