UCI Diabetes 130-US Hospitals
A publicly available diabetes inpatient dataset from UCI containing approximately 100,000 hospitalization records from 130 US hospitals. It is commonly used in research on readmission prediction and clinical risk modeling.
Timeline (last 90 days)
The UCI Diabetes 130-US Hospitals dataset contains 101,766 inpatient records from 130 US hospitals, corresponding to 71,518 patients.
In the UCI diabetes readmission dataset, repeat-admission patients contribute 46.2% of the rows.
If the UCI diabetes readmission dataset is split naively at random, 6,791 patients appear in both the training and test sets, and 41.6% of the visits in the test set come from patients the model has already seen.
In the UCI diabetes readmission dataset, admission type codes 5, 6, and 8 represent not available, NULL, and not mapped respectively, covering 10.2% of visits, and should be converted to missing values.
In the diabetes readmission dataset, the readmitted column has three values: '<30' (within 30 days), '>30' (later), and 'NO'.
In the UCI diabetes readmission dataset, the positive rate for any readmission is 47.1%, and the positive rate for 30-day readmission is 11.4%.
In the UCI diabetes readmission dataset, 2,423 visits end in death or hospice care, and all 1,652 death outcomes are labeled 'NO' (not readmitted).
In the UCI diabetes readmission prediction task, the majority-class baseline accuracy is about 52.9%, but it cannot identify a single readmitted patient.
On this dataset, the histogram gradient boosting model achieves a ROC-AUC of 0.691 (95% CI 0.683–0.700), with a training cost of about 10 seconds.
Shuffling the "number of inpatient visits in the past year" feature reduces ROC-AUC by 0.075, an impact about four times that of other features.
2 more timeline events
All Facts (12)
The UCI Diabetes 130-US Hospitals dataset contains 101,766 inpatient records from 130 US hospitals, corresponding to 71,518 patients.
50%UnverifiedIn the UCI diabetes readmission dataset, repeat-admission patients contribute 46.2% of the rows.
50%UnverifiedIf the UCI diabetes readmission dataset is split naively at random, 6,791 patients appear in both the training and test sets, and 41.6% of the visits in the test set come from patients the model has already seen.
50%UnverifiedIn the UCI diabetes readmission dataset, admission type codes 5, 6, and 8 represent not available, NULL, and not mapped respectively, covering 10.2% of visits, and should be converted to missing values.
50%UnverifiedIn the diabetes readmission dataset, the readmitted column has three values: '<30' (within 30 days), '>30' (later), and 'NO'.
50%UnverifiedIn the UCI diabetes readmission dataset, the positive rate for any readmission is 47.1%, and the positive rate for 30-day readmission is 11.4%.
50%UnverifiedIn the UCI diabetes readmission dataset, 2,423 visits end in death or hospice care, and all 1,652 death outcomes are labeled 'NO' (not readmitted).
50%UnverifiedIn the UCI diabetes readmission prediction task, the majority-class baseline accuracy is about 52.9%, but it cannot identify a single readmitted patient.
50%UnverifiedOn this dataset, the histogram gradient boosting model achieves a ROC-AUC of 0.691 (95% CI 0.683–0.700), with a training cost of about 10 seconds.
50%UnverifiedShuffling the "number of inpatient visits in the past year" feature reduces ROC-AUC by 0.075, an impact about four times that of other features.
50%UnverifiedSelecting the threshold under a constraint of at least 70% sensitivity yields 0.428, with sensitivity 0.697, specificity 0.547, and precision 0.579 on the test set, and would flag 56.8% of patients.
50%UnverifiedThe model is well calibrated: mean predicted risk 0.469, actual observed rate 0.471, calibration slope 1.05, and every decile within 5 percentage points of the observed rate.
50%