Table 1 Overview of cohorts used for the prediction models
From: Attention-based neural networks for clinical prediction modelling on electronic health records
| Â | Mortality | Readmission | Dementia | |
|---|---|---|---|---|
Observations | 3,836,184 visits | 206,995 visits | 287,208 patients | |
Outcome (%) | 36,952 (1%) | 20,202 (9.8%) | 4674 (1.6%) | |
Sex: # female (%) | 2,232,244 (58.2%) | 109,684 (53%) | 148,660 (51.8%) | |
Age mean (std) | 73.8 (9.0) | 62.7 (18.0) | Â | 61.3 (8.5) |
Index event | GP visit after 60 | Inpatient visit of adults | First GP visit of patients aged 50–79 | |
Time-at-risk1 | 30 days | 30 days | 5 years | |
Observation window2 | 1 year prior to index | 1 year prior to index | 1 year prior to index | |
Features | Age, gender, conditions, drug exposure and procedure codes | Age, gender, conditions, drug exposure and procedure codes | Age, gender, conditions, drug exposure and procedure codes | |
Number of features | 2056 | 2303 | 1748 | |
Average features per observation / feature matrix density | 49.5 / 2.3% | 45.2 / 1.9% | 21.8 / 1.2% | |