Predicting unplanned medical visits among patients with diabetes: translation from machine learning to clinical implementation

Table 5 Sensitivity analysis showing the disruption of balanced accuracy when adding normally-distributed noise (0.3 × standard deviation) to each variable

Variable range	New balanced accuracy (%)	Change in balanced accuracy (vs. 65.8% on original sample) (%)
A1C	65.7	− 0.1
BMI	64.7	− 1.1
BP	64.5	− 1.3
HDL	64.4	− 1.4
LDL	65.8	− 0.0
Tobacco use	65.0	− 0.8

Balanced accuracy is the average of the sensitivity and specificity rates (see text), based on test sets across 25 cross-validation tests using repeated-hold-20%-out subsampling. Change in balanced accuracy is relative to the optimized classification results using the original data sample in Table 2 (65.8%)
A1C glycohemoglobin. BMI body mass index. BP blood pressure. HDL high-density lipoprotein. LDL low-density lipoprotein

ISSN: 1472-6947