Figure 4.
Assessing Generalizability of Individual Representations. We compared the generalizability of each of our representations and assessed performance using sensitivity, specificity, and AUC. For each representation, we plotted a boxplot to represent the distribution of the 26 findings for each test performance metric across healthcare systems. CV, Controlled Vocabulary; CVF, Controlled Vocabulary Filter Only; DM, Document MIMIC; DL, Document LIRE; N, N-grams. 1 = Kaiser Permanente of Washington, 2 = Kaiser Permanente of Northern California, 3 = Henry Ford Health System, and 4 = Mayo Clinic Health System.