Intrinsic patient factors predictive of inpatient rehabilitation facility discharge following primary total knee arthroplasty: a systematic review and meta-analysis

Relevant online databases PubMed, Cumulative Index to Nursing and Allied Health (CINAHL), Embase, Cochrane, and Pedro were systematically searched from database inception to October 3rd, 2019. Key terms were identified for the search, including knee arthroplasty, predictor, and discharge, as well as synonym words, utilising Medical Subject Heading and Boolean operator terms. The complete search strategy is reported in Table 1.

We included all articles of any study design investigating pre-operative patient factors for their level of predictivity of IRF discharge following primary TKA. Key inclusion criteria were that the article be available in full text English, a search term was required in the title or abstract, the population studied was primary unilateral knee replacement patients, a pre-operative intrinsic patient factor was a variable in the study and an outcome included discharge destination. Intrinsic factors were defined as those inherent to the individual, including demographic characteristics age, gender, ethnicity, socioeconomic status as well as clinical factors such as presence of co-morbidity. Behavioural factors such as smoking and alcohol consumption, and patient reported outcome measures that capture the patient’s perspective were also considered intrinsic factors and included. Non-intrinsic factors were excluded, where the patient was subject to an intervention such as participation in a pre-operative exercise class or education session. Additional exclusion criteria were post-operative variables including length of stay (LOS), readmission status and other post-operative complications. Studies that did not separate the reporting of primary TKA were also excluded, as were studies that took place after discharge from the post-acute hospital setting, such as outpatient or inpatient rehabilitation.

Based on the inclusion criteria, an initial screening of titles and abstracts was conducted, next, a screening of extracted full text papers was conducted for final review. Those studies that met the inclusion/exclusion criteria for this literature review were then screened for eligibility to be included for meta-analysis.

A modified form based on the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) Checklist was used for data extraction [14]. Data regarding author, year of publication, country, study design, exclusion criteria, patient factors investigated, sample size, participant age and gender as well as a description of statistical analysis undertaken were included.

For assessing the quality of individual studies, the Quality In Prognosis Studies (QUIPS) tool was applied [15]. The QUIPS tool is a validated tool for assessing risk of bias in prognostic factor studies and provides a qualitative assessment of six domains: (I) Study Participants, (II) Study Attrition, (III) Prognostic Factor Measurement, (IV) Outcome Measurement, (V) Study Confounding and (VI) Statistical Analysis and Reporting [16]. For each of these 6 domains, the responses `yes’, `partial’, `no’ or `unsure’ for three up to seven items within each domain are combined to assess the risk of bias. Two reviewers, following the guidelines of Hayden et al. 2013 [16] independently assessed each study, ranking the risk of bias as high, moderate or low. If the authors disagreed on the risk of bias rating, a consensus agreement was reached by joint discussion.

Statistical analysis and graphical reporting were conducted in R statistical software, version 3.5.3 [R-Core], using packages metafor and forestplot [17]. To assess the effect of predictors of discharge destination, odds ratios with the corresponding 95%CI were extracted from the results of univariate and multivariable analyses. When these effect sizes were not reported in univariate results, they were computed from the count data, if available.

Meta-analysis of a patient factor was considered where there was a minimum of three studies reporting an association between the predictor and discharge destination. Meta-analysis was only applied to a factor if the reference categories were similar. Forest plots without pooled effect were produced for those studies considered ineligible for meta-analysis to gain insight into the degree of predictability of the patient factor. Heterogeneity of included studies’ estimates were assessed by computing the I² statistic and was considered statistically significant at P < 0.10. I² values were used to describe the percentage of total variation across studies; an I² value of 25% was considered low, 50% moderate, and 75% high [18]. Pooling of the odds ratios across studies was carried out with a random-effects model using the inverse-variance method.

The results of the search strategy and screening process are shown as a flowchart in Fig. 1. After duplicates were removed, 1557 articles were screened for eligibility with reasons for exclusion listed. A total of 9 articles published between 2011 to 2018, with 218,151 TKA patients, were included in this review [19‐27]. Of those, 4 articles [22‐24, 27] met the criteria to undergo meta-analysis.

The individual studies and their characteristics can be found in Table 2. Patient demographics were similar across studies for mean age ranging from 61 to 70 years, however, female gender had greater variation, ranging from 56 to 83% of the study populations. Country of origin was the United States for 8 of the 9 papers, with 1 being from Australia. All studies were of observational cohort design, 7 were performed retrospectively.

QUIPS ranking did not vary by more than one category between raters for any criteria for each publication and consensus was achieved by discussion. Table 3 presents the risk of bias scores for all included studies. The risk of bias was ranked low across all studies for “study participation,” “prognostic factor measurement,” and “outcome measurement.” However, “study confounding,” was ranked moderate or high for risk of bias across 6 of the 9 studies. As 8 of the 9 studies did not report on perioperative factors such as surgical and anaesthetic technique or physical therapy protocols this could have a confounding effect on the other patient factors assessed.

Patient factors in the included studies that were analysed for their association with discharge destination were age, American society of anaesthesiology (ASA) score, body mass index (BMI), Charlson comorbidity index (CCI), diabetes, gender, ethnicity, haemoglobin (Hb), knee range of motion (ROM), socioeconomic status (SES), 12 item short form health survey or 12 item Veteran’s RAND health survey (SF-12/VR-12), smoking and venous thromboembolism (VTE) history. Table 4 details the predictability of each patient factor on discharge destination for the included studies. Of the 13 patient factors reported on, 6 factors met the criteria for comparison and a synthesised review, these were age, BMI, CCI, gender, SF-12/VR-12 survey, and smoking status.