Predictive model of multiple emergency department visits among adults: analysis of the data from the National Survey of Drug Use and Health (NSDUH)

Table 2 Model fitting and predictive performance: ≥3 emergency department visits/year

Analysis sample	Training Dataset: 2017	Validation Dataset: 2018	Additional Validation Dataset: 2015	Additional Validation Dataset: 2016
Model	AUC	AUC	AUC	AUC
All variables regression	0.80	0.79	0.79	0.79
LASSO	0.80	0.79	0.79	0.79
Stepwise regression	0.80	0.79	0.79	0.79
Single tree	0.76	0.75	0.75	0.76
Stepwise regression with complex interactions	0.80	0.80	0.79	0.79
Random forest	0.88	0.79	0.79	0.79

Note: All bootstrap-estimated standard deviations of the AUCs were < 0.01. Because the model based on 2017 data has shown stability over the years, we have combined data from 2015 to 2018 to produce more stable estimates of odds ratios associated with multiple visits to the ED. In Table 3, we present the odds ratios and variable ranking from stepwise and forest models for 2015–2018 data
AUC Area under the curve, LASSO Least absolute shrinkage and selection operator, ED Emergency department

ISSN: 1472-6963