Table 1.
Assessment of risk of bias in included studies
| Source of bias | Macpherson and Lofgren7 | Auerbach et al 8 | Katz et al 9 | Vazirani et al 10 |
| Selection bias | ||||
| Randomisation (RCT) | Permuted blocks of size 2 and 4, stratified by quartiles of anticipated LoS | – | – | – |
| Allocation concealment (RCT) | Not reported | – | – | – |
| Control for confounders (NRS) | – | Patients sampled at random; propensity score weighting | Consecutive patients | Regression methods |
| External validity (RCT and NRS) | Uncertain as study setting was more restrictive than a non-experimental setting | Potential unobserved confounding | Potential unobserved confounding | Potential unobserved confounding |
| Performance bias | ||||
| Blinding of participants and/or investigators (RCT) | No | – | – | – |
| Measurement of exposure (NRS) | – | No blinding but 5% of medical record abstractions were reviewed for data validity | No blinding | No blinding |
| Detection bias | ||||
| Blinded outcome assessment (RCT and NRS) | No blinding but 100% interobserver and intraobserver agreement* | Not reported | Not reported | Not reported |
| Attrition bias | ||||
| Completeness of follow-up (RCT and NRS) | Yes | Yes | 35 (8.3%) patients with missing medical records | Yes |
Ellipses indicate not applicable.
*A researcher re-abstracted length of stay data on 10 randomly selected records, and a physician not associated with the study abstracted length of stay from the same 10 records.
LoS, length of stay; NRS, non-randomised studies; RCT, randomised controlled trials.