Table 4 The information extracted using the specific form, and references for each criterion

The databases used were sorted into six categories, as recommended by Levac et al. [26]: electronic health records, administrative data, claims data, patient or disease registers, health surveys, or clinical trials data

The use (or not) of International Classification of Diseases (ICD) [30] to classify the diseases studied was recorded.

Outcomes were extracted and classified according to the categories typically observed in the literature: mortality, cost, length of stay, readmission, and others [4, 29,30,31]. The methods used to explore outcomes were detailed and then sorted using an inductive approach.

The scales or scores used to stratify analyses for the initial severity of the patient’s condition were classified as a known score (e.g. the Charlson Comorbidity Index (CCI), as adapted by Deyo [32, 33], or the Elixhauser score [34]) or another scale/score. All the covariates used to adjust the statistical models were listed and then sorted using an inductive approach.

The types of graphic representation of the results were classified as graphs, tables, or both. Graphs were subdivided into scatter plots, line graphs, bar graphs, histograms, pie charts, box plots or other graphs, as described by Slutsky et al. [28]

The variable used to assess hospital volume was classified as continuous, categorical, or both. If volume was considered as a categorical variable, we extracted more detailed information on the categories: quantiles, statistically defined cut-off points, arbitrary thresholds or those based on the literature, and other types.

Lastly, each study conclusion was noted and classified as having found (or not) a statistically significant positive (or negative) relationship between volume and outcome.