Table 1.
Characteristics of included studies (n = 59)
| n = 59 | |
|---|---|
| Variables | Median (IQR) |
| Sample size | 4782 (616–23.264) |
| Predictors included in final modela | 10 (7–14) |
| % (n) | |
|---|---|
| Year of publication | |
| <2015 (Before TRIPOD guideline) | 17 (10) |
| >2016 | 83 (49) |
| Number of publications per journal | |
| <5 publications per journal | 69 (41) |
| >5 publications per journal | 31 (18) |
| Prospective database | 3 (5) |
| National/Registry databaseb | 51 (30) |
| Mention of using TRIPOD | 20 (12) |
| Predicted outcome | |
| Complications | 24 (14) |
| PROM | 20 (12) |
| Mortality | 19 (11) |
| Health management | 19 (11) |
| Other | 19 (11) |
Abbreviations: IQR, interquartile range; ML, machine learning; PROM, patient reported outcome measure; TRIPOD, transparent reporting of a multivariable prediction model for individual prognosis or diagnosis.
a
The amount of predictors that were included in the final, best performing ML algorithm. In 14% (8/59) this could not be extracted from the study or was unclear.
b
This includes databases, such as Surveillance, Epidemiology, and End Results (SEER) or American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP).