Skip to main content
. 2025 Jun 4;13:e68138. doi: 10.2196/68138

Table 3.

Average AUCa values (%) and SD for different methods for the HFb prediction, ADc prediction, and PLSd prediction downstream tasks on the test datasets.

Model or dataset HF prediction (MDCe) AD prediction (MDC) HF prediction (MIMIC-IVf) PLS prediction (MIMIC-IV)
Logistic regression 62.4 (1.1) 56.4 (1.1) 83.8 (1.1) 54.2 (0.4)
Random forest 60.7 (0.5) 51.8 (0.3) 78.6 (1.6) 51.1 (0.3)
MLP 67.9 (3.0) 68.0 (1.5) 86.0 (0.5) 59.3 (1.9)
Bi-GRU 62.3 (1.2) 60.4 (1.1) 85.0 (1.3) 55.9 (1.0)
MLMg 67.7 (2.6) 69.5 (1.6) 86.2 (0.9) 60.2 (1.2)
MLM+TOOhRCSi 65.1 (1.2) 65.6 (0.7) 88.1 (0.7) 58.4 (0.9)
MLM+TOOCCSj 64.6 (2.7) 67.2 (1.3) 89.8 (0.8) 60.4 (1.2)
MLM+TOORVSk 72.8 (3.1) 70.4 (0.2) 87.9 (1.7) 57.3 (0.8)
MLM+TOOCVSl 73.9 (1.9) 71.9 (1.6) 87.2 (1.8) 58.8 (1.6)

aAUC: area under the receiver operating characteristic curve.

bHF: heart failure.

cAD: Alzheimer disease.

dPLS: prolonged length of stay.

eMDC: Malmo Diet and Cancer Cohort.

fMIMIC-IV: Medical Information Mart for Intensive Care IV.

fgLM: masked language modelling.

hTOO: trajectory-order objective.

iRCS: random code swapping.

jCCS: code swapping function.

jRVS: random visit swapping.

lCVS: conditional visit swapping.