Background

From 2014 to 2030 in the United States, primary total knee arthroplasty (TKA) is projected to increase by 85% to 1.26 million surgical procedures [1]. TKA is widely regarded as a cost-effective intervention for end-stage knee osteoarthritis, improving both a patient’s functional status and overall health quality [2]. However, with the societal burden of cost for TKA increasing, there is a need to evaluate the economic efficiency of current models of care [3, 4].

Total Joint Arthroplasty (TJA) is reportedly the most frequent procedure leading to post-acute admission to an Inpatient Rehabilitation Facility (IRF), representing one of the most significant costs associated with TKA [4, 5]. Acute post-discharge care following primary TKA can account for up to 37% of the total procedure cost, with home discharge reported as costing $16,000 less than discharge to an IRF in the United States [6, 7]. While IRF patients receive multidisciplinary input such as physical therapy and occupational therapy, this has yet to translate into evidence for improved functional outcomes when compared with TKA patients discharged directly to home [8, 9]. Moreover, retrospective studies relying on administrative datasets have shown TKA patients with an IRF discharge have a significantly higher adverse event and 30-day readmission rate compared to those discharged home [10, 11].

The decision to discharge a patient to an IRF is dependent on many variables and identifying pre-operative factors that increase the likelihood of IRF discharge will better facilitate pre-operative discharge planning and resource allocation. Given an estimated 1.82 billion is spent on IRF discharge after lower-extremity arthroplasty it is imperative to determine which factors create a higher risk for non-home discharge [12]. To date, only individual studies have investigated TKA patient characteristics predictive of discharge destination, therefore, the aim of this review was to systematically review the literature and conduct a meta-analysis on reported intrinsic patient factors predictive of IRF discharge.

Methods

This systematic review and meta-analysis was prospectively registered on PROSPERO (International prospective register of systematic reviews), registration CRD42019134422 and is reported in accordance with the guidelines from the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement [13].

Search strategy

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.

Table 1 Critical review databases and search terms
Full size table

Study eligibility

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.

Study selection

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.

Data extraction

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.

Quality assessment

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

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.

Results

Literature search and study characteristics

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 [19,20,21,22,23,24,25,26,27]. Of those, 4 articles [22,23,24, 27] met the criteria to undergo meta-analysis.

Fig. 1
figure 1

Prisma Flow Diagram of systematic search, screening and selection process

Full size image

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.

Table 2 Study Characteristics
Full size table

Methodical quality

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.

Table 3 Results of risk of bias assessment using the Quality in Prognosis Studies (QUIPS) tool for included studies
Full size table

Intrinsic patient factors analysed

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.

Table 4 Intrinsic Patient Factors Predictive of Inpatient Rehabilitation Discharge
Full size table

Demographic factors

The patient factor gender was able to undergo meta-analysis in 4 studies [22,23,24, 27] to provide results of a combined effect on predictability of discharge destination. Being of female gender increased the likelihood of IRF discharge by 78% when compared to male gender (OR = 1.78; 95% CI = 1.43–2.20) (Fig. 2). The association between age and discharge destination was reported in 6 studies [20, 22,23,24, 26, 27]. Older age was predictive of IRF discharge in all included studies, with the greatest effect for those aged 75 years and older (Fig. 3). Increased BMI was also able to be included for review in 3 studies [22,23,24], with those in the severely obese (≥40 kg/m2) category having the highest likelihood of IRF discharge (Fig. 4).

Fig. 2
figure 2

Meta-analysis showing the adjusted odds ratios and 95%CI of a random effects (RE) model for likelihood of discharge to IRF for females compared to males. Unadjusted odds ratios with 95%CI are also reported

Full size image
Fig. 3
figure 3

Adjusted odds ratios and 95%CI showing likelihood of discharge to IRF with older age. Unadjusted odds ratios with 95%CI are also reported

Full size image
Fig. 4
figure 4

Adjusted odds ratios and 95%CI showing likelihood of discharge to IRF with increase in BMI. Unadjusted odds ratios with 95%CI are also reported

Full size image

Clinical factors

Studies varied in their reporting of patient comorbidities. CCI was reported in 3 studies but due to methodological heterogenicity could not be meta-analysed [22,23,24]. The CCI quantifies an individual’s burden of disease and corresponding 1-year mortality risk, with a lower score equalling a lower risk [28]. Figure 5 shows a trend was towards a higher CCI being more predictive of IRF discharge.

Fig. 5
figure 5

Adjusted odds ratios and 95%CI showing likelihood of discharge to IRF with higher CCI. Unadjusted odds ratios with 95%CI are also reported

Full size image

Behavioural factors

Smoking status (non-smoker or currently smoking) was reported on in 3 studies and was included in meta-analysis [22, 23, 27]. Smoking showed an overall decreased likelihood of IRF discharge (OR = 0.80; 95% CI 0.42–1.50), however, heterogeneity of the studies was moderate (I2 = 68.5%) (Fig. 6).

Fig. 6
figure 6

Meta-analysis showing the adjusted odds ratios and 95%CI of a random effects (RE) model for the effect of smoking on discharge to IRF. Unadjusted odds ratios with 95%CI are also reported

Full size image

Patient reported outcome measures

A self-reported measure of physical function was assessed in 3 [22,23,24] of the 9 included studies, and due to the similarity of the design and scoring systems of the SF-12 and VR-12 it was decided to combine the results of these tools (Fig. 7) [29, 30]. Only 1 study, Prohaska et al., demonstrated a consistent association with IRF discharge related to lower self-reported physical function, with the other included studies showing the relationship to be more unclear.

Fig. 7
figure 7

Adjusted odds ratios and 95%CI showing likelihood of discharge to IRF with lower VR-12 score. Unadjusted odds ratios with 95%CI are also reported. ★The SF-12 Health Survey was used.

Full size image

Discussion

This systematic review and meta-analysis demonstrates that it is difficult to develop predictive models for intrinsic patient factors associated with IRF discharge based on the current existing literature. The large degree of heterogeneity and wide variation of statistical analysis and reporting across the included studies precluded meta-analysis for all but 2 variables. Two included factors underwent meta-analysis, gender and smoking (Figs. 2 and 6). There was a clear association between female gender and likelihood of IRF discharge, however, the relationship between smoking status and IRF discharge was less certain due to the heterogeneity of the studies included in the smoking meta-analysis (I2 = 68.5%). For the other intrinsic patient factors included in this review, the strongest trends for discharge to IRF were older age, greater comorbidity or the severely obese. A worse self-reported physical function was not consistently associated with an increased risk for IRF discharge.

Although to our knowledge this is the first systematic literature review and meta-analysis where the primary outcome of interest is discharge destination, systematic reviews on patient factors predictive of increased length of hospital stay following TJA have been published [31, 32]. The patient risk factors found in this review that trended towards an increased likelihood of discharge to IRF, are similar to those reported for increased length of stay, including female gender, older age, increased comorbidities and higher BMI [31, 32].

A limitation of this review is that the studies included were mostly of retrospective design and all utilised large medical databases as the source of patient predictors. The use of large databases for analysis has been reported as having limitations such as coding bias [33]. Additionally, using a retrospective cohort design limits the investigation to only variables recorded at that time. It is likely that other surgical, psychological and sociological variables could be predictive of discharge destination, however, if this information is not available at the time of retrospective data collection then it will not be included in analysis. In terms of sociological factors, insurance status has been shown to be predictive of IRF discharge [34], however, due to the international variability of insurance models this was excluded from analysis.

One psychological variable that has been shown to be predictive of discharge destination following TJA is patient expectation [35]. Halawi et al. found that a patient’s pre-operative expectation of their discharge destination was the strongest predictor of actual discharge destination even when adjusted for other variables such as age and caregiver assistance. As this study did not separate TKA from total hip arthroplasty it was not eligible for inclusion in this review, but it does highlight the need for more studies incorporating patient belief systems into predictive modelling for IRF discharge.

With a growing body of evidence suggesting that IRF discharge following routine primary TKA is not superior to home discharge, further prospective high-quality studies investigating the patient factors that are predictive of discharge destination are needed. Previous studies have assessed patient factors by retrospectively accessing medical records, however, many of these are non-modifiable such as age and gender. With significantly increased costs associated with IRF when compared with home discharge after TKA, modifiable patient factors such as BMI and patient expectation should be given priority in future investigations.

Conclusion

This systematic review and meta-analysis illustrates that although literature exists on investigating which intrinsic patient factors are predictive of IRF discharge, there is large variation in statistical methods and reporting. Female gender and smoking were two patient factors able to be included in this meta-analysis, with female gender shown to be predictive of IRF discharge, however, the relationship between smoking and discharge destination was less certain. There was also a trend for those of older age, increased comorbidity or in a severely obese category to have an increased likelihood of IRF discharge.