Table 2.
Characteristic | Variable | Results |
---|---|---|
Study Design | Retrospective cohort | 66 (82.5)5,28–92 |
Prospective cohort | 5 (6.3)93–97 | |
Case-control | 3 (3.8)98–100 | |
Qualitative | 3 (3.8)12,101,102 | |
Retrospective + prospective cohort | 3 (3.8)103–105 | |
Locationa | United States | 55 (68.8)5,28,29,31,32,34,36,40–44,47–50,52,54–60,62,63,65–70,73–80,82,83,85,90–93,95,97–100,102,104,105 |
United Kingdom | 7 (8.8)39,53,55,61,63,67,77 | |
Taiwan | 5 (6.3)38,39,45,67,88 | |
Canada | 4 (5)12,33,81,87 | |
China | 4 (5)37,39,71,103 | |
Australia | 3 (3.8)86,94,96 | |
Germany | 2 (2.5)51,101 | |
India | 2 (2.5)64,89 | |
Switzerland | 2 (2.5)35,72 | |
South Korea | 1 (1.3)30 | |
Saudi Arabia | 1 (1.3)46 | |
Spain | 1 (1.3)84 | |
France | 1 (1.3)76 | |
Portugal | 1 (1.3)75 | |
Year | 2019 | 23 (28.8)12,42,43,60,64,65,67,68,70,72,79,82,83,92,95–99,101–103,105 |
2018 | 18 (22.5)30,36,39,41,46,61,63,66,69,71,74,78,81,88,89,91,94,104 | |
2017 | 15 (18.8)5,29,31,44,51–53,57,58,75,77,80,85,87,100 | |
2016 | 12 (15)28,32,37,40,47,49,50,62,73,84,86,93 | |
2015 | 12 (15)33–35,38,45,48,54–56,59,76,90 | |
Machine learning methodsa | Neural networks | 34 (42.5)28–33,35,39,41,43,45,47–51,54,60,66,68,70,72,76,79–84,89,94,96,99,100 |
Random forests | 32 (40)30,35–38,40,42–44,49–51,54,57,63–66,71,76,79,83,88,90,91,94,95,97,98,103–105 | |
Regressionb | 27 (33.8)30,32,35,39–41,43,47,48,50,54,56,61,62,65,70,76,79,83,84,87,88,90–92,94,95 | |
Support vector machines | 24 (30)32,34,35,37–39,44,46–48,50,51,55–57,59,61,62,66,75,79,94,98,103 | |
Boosting | 17 (21.3)32,47,49,50,52,53,65–67,69,76,78,90,91,94,98,103 | |
Decision trees | 15 (18.8)32,35,38–40,47,48,50,52,53,56,62,76,78,88 | |
Penalized regressionc | 13 (16.3)5,35,37,44,55,57,65,66,91,95,98,100,103 | |
Bayesian | 13 (16.3)5,35,37,39,44,57,62,66,76,79,86,93,98 | |
Topic modeling | 9 (11.3)55,57,59,66,79,80,85,91,99 | |
Ensemble | 7 (8.8)52,53,62,76,78,81,95 | |
Nearest neighbors | 6 (7.5)35,50,54,72,98,103 | |
Gaussian process | 5 (6.3)5,29,31,55,98 | |
Clustering | 5 (6.3)51,61,87,89,92 | |
Reinforcement learning | 4 (5)34,63,73,77 | |
Generalized additive models | 3 (3.8)76,90,95 | |
Bagging | 3 (3.8)35,50,76 | |
Discriminant analysis | 2 (2.5)35,61 | |
Word vectorization | 2 (2.5)64,89 | |
Other methodsd | 10 (12.5)32,35,37,44,54,58,72,74–76 | |
Sample size | Qualitative | 10-19 clinicians |
Model development | 127 patients to 296,724 hospital admissions | |
Clinical specialtya | Intensive care (adult) | 37 (46.3)5,12,28,30,34,43,44,51–53,55,56,58–61,63,64,66–69,71–77,79,80,83,86,87,89,91,101 |
In-hospital, acute care/not further specified | 14 (17.5)30,31,34,46,50,54,69,82,85,90,97,103–105 | |
Emergency medicine | 14 (17.5 )12,35,36,40,42,57,69,84,86,88,90,93,95,99 | |
Cardiology | 7 (8.8)37,71,81,86,88,92,102 | |
Pediatric (acute and intensive care) | 6 (7.5)32,58,70,78,81,100 | |
Nephrology | 5 (6.3)62,66,75,79,83 | |
Neonatal intensive care | 4 (5)33,94,96,98 | |
Stroke | 4 (5)47–49,86 | |
Surgery | 3 (3.8)38,41,86 | |
Diabetes | 2 (2.5)39,65 | |
Pulmonology/respiratory | 2 (2.5)42,90 | |
Nursing | 1 (1.3)45 | |
Trauma | 1 (1.3)95 | |
In-hospital, specifically: emergency medicine, intensive care, cardiothoracic surgery/transplant, neurology/vascular/stroke, gastrointestinal, oncology/hematology/immunology/pharmacology | 1 (1.3)86 |
Values are n (%) or range.
Studies may fall into more than 1 category.
Does not meet our definition of machine learning but included for purposes of reporting all methods authors compared.
Including LASSO, Ridge, and Elastic Net.
Generalized linear models, conditional inference recursive partition, conditional random fields, Weibull-Cox proportional hazards, lazy learner, piecewise-constant conditional intensity model, analysis of covariance (see footnote b), analysis of variance (see footnote b), fuzzy modeling (see footnote b), switching-state autoregressive model, mimic learning, nearest shrunken centroids, J48 algorithm, PART rule.