Assessing Clinical Readiness for Sepsis Prediction Using Vital Signs

1

Vital-sign completeness filtering reduced 40,336 PhysioNet records to 3,766 eligible sequences, favoring records with more complete temporal data.
2

The report does not clearly describe residual-missing-value handling or compare included and excluded records and performance across missingness levels.
3

The relationship between the 4:1 train/test split, 5-fold cross-validation, and test-set use for model selection or epoch choice is unclear.
4

The reported normalized MAE of 0.11% may represent 0.11, or 11% of sequence length, and does not indicate prediction timing in hours.
5

The authors are asked to report absolute timing error, clinically interpretable warning time, and the probability threshold used for sensitivity and specificity.

1