Table 2.
Sources of Bias.
| First Author, Year | Possible Sources of Bias | Reported AUC | |||
|---|---|---|---|---|---|
| Resubstitution bias | Few events | Fit many models* | Multiple centers | ||
| Vaidya, 2008 | No | No (nev = 102, nm = 4) | Yes (up to 29) | No | 0.75–0.78 |
| Han, 2009 | Yes | No (nev = 36, nm = 3) | No (5) | No | 0.75–0.78 |
| Liangos, 2009 | Yes | Yes (nev = 13, nm = 3) | No (7) | Yes | 0.78 |
| Che, 2010 | Yes | Yes (nev = 14, nm = 5) | Yes (29) | No | 0.98 |
| de Geus, 2011 | Yes | No (nev = 47, nm = 2)‡ | No (7) | No | 0.83 |
| Parikh, 2011 | No^ | No (nev = 60, nm = 3) | Yes (165) | Yes | 0.77 |
| Katagiri, 2012 | Yes | No (nev = 28, nm = 2) | No (9) | No | 0.81 |
| Kokkoris, 2012 | Yes | No (nev = 36, nm = 2–3) | Yes (11) | No | 0.823–0.835 |
| Cho, 2013 | Yes | No (nev = 54, nm = 2) | No (1) | No | 0.800 |
| Kambhampati, 2013 | Yes | Yes (nev = 27, nm = 5) | No (1) | No | 0.80 |
| Liu, 2013 | Yes | No (nev = 26, nm = 2) | No (2) | No | 0.911–0.927 |
| Luo, 2013 | Yes | Yes (nev = 12, nm = 3) | No (2) | No | 0.99 |
| Parikh, 2013 | No^ | No (nev = 60, nm = 3) | Yes (455) | Yes | 0.78 |
| Siew, 2013 | Yes | No (nev = 127, nm = 2) | No (1) | No | 0.59 |
| Zeng, 2014 | Yes | No (nev = 37, nm = 2) | Yes (64) | No | 0.91–0.94 |
Dark red indicates a high likelihood of bias, dark pink indicates possible bias, and pale pink indicates low likelihood of bias. In the “Resubstitution bias” column, “no” indicated some attempt was made to address resubstitution bias. Under “Few events”, nev is the number of events and nm is the number of markers in the main reported combination; we considered nev:nm < 10 to be few events. The number of events reported may include individuals with missing marker values. We considered more than 10 models to be “many”. We calculated the number of models by hand (i.e., these were not explicitly reported by the authors). Under “Multiple centers”, “yes” indicates that multiple centers were involved and possible differences were not considered, though this does not mean there was bias due to center differences. Abbreviations: AUC – area under the receiver operating characteristic curve.
Includes search over univariate models (if best individual markers were chosen for combination).
Resubstitution bias was addressed, but may not be totally removed.
Reported 10 individuals with sustained AKI also had missing marker values.