Table 2.

Performance metrics for predicting prospective social needs for patients at Johns Hopkins health system using electronic health record data between 2016 and 2021: logistic regression models.

	Precision	Recall	AUC	PRC
Logistic regression model I— Using EHR structured data
Overall population	0.508 (0.499:0.518)	0.038 (0.036:0.041)	0.703 (0.701:0.705)	0.255 (0.254:0.256)
65+ years old patients	0.447 (0.412:0.482)	0.018 (0.017:0.020)	0.699 (0.694:0.705)	0.228 (0.223:0.232)
Racial group—White	0.491 (0.469:0.513)	0.024 (0.023:0.025)	0.690 (0.688:0.692)	0.228 (0.224:0.232)
Racial group—Black	0.524 (0.508:0.541)	0.067 (0.064:0.069)	0.712 (0.708:0.716)	0.312 (0.306:0.318)
Neighborhood characteristics—Most disadvantaged	0.509 (0.483:0.535)	0.051 (0.048:0.054)	0.714 (0.709:0.718)	0.272 (0.264:0.279)
Neighborhood characteristics—Least disadvantaged	0.469 (0.402:0.536)	0.003 (0.002:0.003)	0.666 (0.653:0.679)	0.180 (0.172:0.189)
Logistic regression model II— Using EHR structured and unstructured data
Overall population	0.580 (0.574:0.586)	0.124 (0.124:0.124)	0.701 (0.699:0.703)	0.378 (0.375:0.38)
65+ years old patients	0.558 (0.552:0.563)	0.107 (0.104:0.111)	0.701 (0.696:0.705)	0.370 (0.364:0.375)
Racial group—White	0.550 (0.539:0.560)	0.084 (0.082:0.087)	0.686 (0.685:0.688)	0.335 (0.330:0.341)
Racial group—Black	0.606 (0.588:0.624)	0.221 (0.216:0.225)	0.715 (0.707:0.723)	0.466 (0.452:0.481)
Neighborhood characteristics—Most disadvantaged	0.582 (0.573:0.591)	0.149 (0.146:0.153)	0.710 (0.708:0.712)	0.400 (0.396:0.403)
Neighborhood characteristics—Least disadvantaged	0.500 (0.462:0.538)	0.027 (0.022:0.031)	0.664 (0.661:0.668)	0.271 (0.260:0.282)

Abbreviations: AUC = area under the receiver operating characteristic (ROC) curve, PRC = precision-recall curve.