This systematic review protocol will be reported according to the “Preferred Reporting Items for Systematic Review and Meta-analysis – Protocols” (PRISMA-P) [22] and the “Meta-Analysis of Observational Studies in Epidemiology” (MOOSE) [23] guidelines. In addition, this review has been registered prospectively with Open Science Framework. Any conflict between reviewers throughout the review process will be resolved through discussion; and if a consensus is not able to be reached, a third author will be consulted. The full search strategy for the primary databases can be found in Multimedia Appendix 1.

To ensure that a Cochrane review has not been published on the topic of impact of reducing LOS on post-acute care, the Cochrane library has been searched using the MeSH term and keyword “Arthroplasty.” This search has confirmed a Cochrane review has not been previously published on this topic.

Database searches were devised with recommendations from an external research librarian and previous literature suggesting optimal database combinations [24]. A comprehensive literature search will be performed of MEDLINE (Ovid), Web of Science, CINAHL, Emcare (Ovid), Embase (Ovid), National Health Service Economic Evaluations Database, and the Cochrane library from their inception to January 2021. In addition, Google scholar will be utilized to supplement the primary database search and reduce the risk of publication bias. Keywords and MeSH were combined with the operator “OR” for the population and intervention. The population and intervention queries were combined with the “AND” operator.

Following independent screening of the first 10% of studies, the agreed eligibility criteria will be refined by discussion between the 2 reviewers (SR and CSh), as per PRISMA-P guidelines [22]. This final eligibility criteria will then be utilized to independently screen titles and abstract of identified citations by 2 reviewers. Any arising discrepancies will be resolved through discussion between reviewers.

Forward and backward citation tracking of studies for full-text screening will be used to identify any additional studies. Following this, the same 2 reviewers will undertake screening of full-text studies to determine final eligibility for inclusion. A PRISMA flow diagram will be used to report the findings of this study selection process [25]. A further full-text review will be undertaken of potentially eligible studies. Interrater reliability will be assessed and will be reported as Cohen kappa statistic [26] to inform the agreement between reviewers.

Given that retrospective studies may report on the same population within the same timeframe, care will be taken to determine if these studies are independent. In the event that multiple studies are determined to be assessing the same population, all of these studies will be treated as one, and reference to all will be made, to reduce the risk of overestimation [27]. A further critical analysis of these studies will be undertaken to highlight any discrepancies, and authors will be contacted if such discrepancies are identified.

The first round of deduplication will be undertaken through EndNote X9. After this, Covidence will be used to perform further deduplication, screening, and data extraction. STATA (version 16) software will be used to conduct statistical analyses.

Two reviewers will use a standardized data extraction form to extract data from each study independently. A further comparison of this, completed, data extraction form to assess consistency and accuracy will also be undertaken. Inconsistencies will be resolved with discussion with other authors. Study authors will be contacted via email if full-text copies of included studies are unattainable or clarification of methods or results is required.

The form for data extraction will follow items 1 to 16 in the “Reporting on ERAS Compliance, Outcomes, and Elements Research (RECOvER)” checklist [28] to inform details of the article, characteristics of patients and the details of the enhanced recovery program, and compliance within the program. The details of the employed enhanced recovery strategies will also be scored against the ERAS society’s recommendation.

In addition, a results summary will be recorded with total number of patients and total number of patients with outcomes; any measure of prognostic effect (including odds ratio, relative risk and hazard ratios); standardized mean difference or mean difference and 95% confidence intervals for each outcome measure as appropriate; use of univariate/multivariate analysis; and variables included in multivariate analysis. Patient-reported outcome measures will be recorded in means (SD), and any measure of economic evaluation will be recorded.

Data extraction for health economic outcomes will additionally include the following:

Any missing values will be calculated provided sufficient data are available. Study authors will be contacted to obtain missing data.

Risk of bias will be assessed by 2 reviewers through the Cochrane Risk-of-Bias 2 (RoB 2) tool [29] for randomized trials and Newcastle-Ottawa Scale (NOS) [30] for nonrandomized studies. Health economic outcomes will be assessed with the Consensus on Health Economic Criteria (CHEC) list [31], and model-based studies with the questionnaire produced by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) [32]. A summary table presenting all articles included in the review will be used to present the results of the critical appraisal. Meta-analysis will be performed if articles have adequate data for quantitative synthesis, and those with a high risk of bias will be excluded in the form of a sensitivity analysis to determine their influence on the meta-analysis [33].

All studies will be presented in both tabular and narrative form. A quantitative analysis of the findings from the included studies is planned for all outcomes. A random effects model will be utilized as heterogeneity is anticipated between studies [34]. Meta-analyses will be reported with a measure of effect (for dichotomous outcomes) and mean difference (for continuous outcomes). When different outcome measures are utilized for the same outcome, a standardized mean difference will be reported [35]. 95% confidence intervals will be reported for all. Care will be taken to report continuous variables separately from categorical variables.

Heterogeneity will be assessed by the Cochran Q test and I² statistic [36]. Subgroup analyses with random-effects meta-analysis will be undertaken when significant heterogeneity is encountered. If heterogeneity is not resolved through this process, exclusion of any outlying studies in the form of sensitivity analysis will be considered if a clear explanation for conflicting results is found upon review of the papers in question. Results of meta-analysis, both including and excluding papers in question, will be presented and interpreted. If variation in the results cannot be accounted for, a meta-analysis will not be conducted, and results will be presented as per the “Synthesis Without Meta-analysis” (SWiM) guidelines [37]. Pooled estimates will be reported with a forest plot and 95% confidence interval.

Several subgroup analyses are planned if data permits. Studies will be grouped according to their ERAS protocol to understand the impact of different ERAS programs on outcomes. Additionally, each ERAS protocol will be scored according to the number of ERAS society recommendations implemented to understand the cumulative impact of the ERAS recommendations [14, 15]. The impact of varying levels of compliance within ERAS programs will also be evaluated through a subgroup analysis. A separate subgroup analysis for programs using a same-day discharge or outpatient surgery will also be conducted. Finally, the impact of patient-related factors, within ERAS surgery, on outcomes will also be explored.

Furthermore, it is also possible that a single comparator may not be determinable, as a significant portion of the included studies may present LOS as a continuous variable. As such, analyses will be undertaken with LOS as a continuous and categorical variable (categories for LOS will be formulated depending on the included studies). If significant differences in specific components of ERAS programs, such as the composition of spinal anesthesia, method of preoperative education, or duration of postoperative regional analgesia, are encountered between studies, sensitivity analyses will be used to assess the impact of these differences.

Strategies employed to reduce risk of publication bias include a search strategy that does not restrict gray literature, and data from unpublished work included in meta-analysis [33,38]. Furthermore, nonreporting biases will be attempted to be identified from other factors through contour-enhanced funnel plots with an appropriate test of asymmetry for meta-analyses including at least ten studies [38,39]. Any potential sources of asymmetry will be explored; and a trim and fill method will be utilized to account for the possibility of publication bias if deemed appropriate [40].

In addition, outcomes mentioned in the “Methods” section of studies will be compared with those in the results during the data extraction process. In randomized trials, the final reported outcomes will be compared with a published protocol, if available. Both reviewers will assess the completed data extraction table to determine whether certain outcomes were included by most included studies but not addressed by a few. If reporting bias is found, studies will be investigated for possible explanations, and authors may be contacted to provide clarity [33].

The Grades of Recommendation, Assessment, Development and Evaluation (GRADE) approach will be used by 2 reviewers, to independently assess the quality of cumulative evidence [41,42]. The quality of evidence will be downgraded if significant limitations are present, as evidenced from the risk of bias. In addition, evidence will be downgraded if wide confidence intervals are encountered and if statistical heterogeneity, as assessed through the I² statistic, is over 25% [43].