Methods of the systematic review of the cost-effectiveness of enhanced recovery following hip and knee replacement
The methods of the systematic review were registered with the International Prospective Register of Systematic Reviews number CRD42017059473. We defined the search strategies and database selection with assistance from an information specialist, and by comparing our search terms with those from previous reviews and review protocols of economic evaluations of joint replacement. We searched Ovid MEDLINE, EMBASE, the NHS Economic Evaluation Database (via The Cochrane Library) and the American Economic Association’s electronic bibliography (EconLit) (via ProQuest) for English-language peer-reviewed papers published between 1 January 2000 and 1 March 2017 that included a cost–utility analysis of an ERP or components of one, compared with usual care, in patients having hip or knee replacement (the complete search strategy is presented in Report Supplementary Material 7). Abstracts or conference presentations were excluded, as results are not presented in sufficient detail to allow critical appraisal of the economic evaluations.
We included studies with participants undergoing joint replacement for osteoarthritis and excluded populations with other indications for surgery, such as avascular necrosis, inflammatory arthropathy, previous/failed surgery, cancer, congenital conditions or infection, as well as studies looking at emergency procedures (e.g. due to trauma). We assumed that studies without details of the indication for surgery were representative of a population with osteoarthritis, and therefore included these studies. We also excluded evaluations of surgical technique or choice of implant.
Economic evaluations of any preoperative, perioperative or postoperative intervention within the joint replacement ERP were included, in addition to studies considering ERPs as a whole. Interventions had to be those that form part of the usual pathway of care (with or without enhanced recovery) for joint replacement. We included model-based evaluation and randomised controlled trials/cohort-based economic evaluation, and restricted the analysis to cost–utility analyses [i.e. reporting costs per quality-adjusted life-year (QALY) gained], as this is the preferred approach to inform decisions.90
The search strategy and inclusion and exclusion criteria were piloted by two reviewers. For the latter, the search was run and inclusion and exclusion criteria were applied to 10% of the search results to check consistency between reviewers. Studies were then independently screened based on their titles and abstracts by three reviewers. EndNote X7 (Clarivate Analytics, Philadelphia, PA, USA) was used to manage the references. Full texts were obtained for studies chosen for inclusion by any reviewer. As an amendment to the protocol, evaluations of thromboprophylaxis were excluded at the full-text stage because of a recent comprehensive systematic review in this area.88 Data extraction was performed by three reviewers, with disagreements resolved by a fourth reviewer, using a standardised form (presented in Report Supplementary Material 8).
The quality of reporting and risk of bias of the economic evaluations were assessed using the Consensus on Health Economic Criteria (CHEC) list,91 the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) questionnaire for modelling studies92 and the Assessment of the Validation Status of Health-Economic decision models (AdViSHE) tool.93 Items in the checklists were marked as yes, no, unknown or not applicable for each study, and a final assessment of the risk of bias was made by the reviewer. We added the Cochrane Collaboration’s tool96 to the original protocol to assess risk of bias in trial-based studies, referring to the original reports of trial outcomes when necessary.
The principal outcomes reported in this review were a point estimate of cost-effectiveness, in terms of incremental cost per QALY gained, and the probability of an intervention being cost-effective according to the willingness-to-pay threshold used by the authors of each study.
Results of the systematic review
We identified 11,060 studies and one additional study was found from other sources. After excluding duplicates, we screened 6221 titles and abstracts. We excluded 6072 papers based on their abstracts and 136 following review of their full texts (reasons for exclusion are given in ). We therefore included 13 papers in this review.86,97–106,109,110
Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of studies included in this review. EconLit, American Economic Association’s electronic bibliography; NHS EED, NHS Economic Evaluation Database.
summarises the characteristics of the 13 studies. Five studies included both hip and knee replacement patients,97–99,101,109 five included only hip replacement patients100,102–104,106 and three included only knee replacement patients.86,105,110 Two papers evaluated an entire ERP98,109 and the remainder evaluated components of a pathway, such as optimisation of comorbidities (specifically morbid obesity),110 measures to reduce allogenic blood transfusion,99,100 local infiltration of anaesthetic,101 prophylactic antibiotics and other infection prevention measures,97,102–104 and physical therapy.86,105 The final study106 concerned the optimal timing of follow-up post surgery.
Summary of studies included in the review
The results of the 13 studies are summarised in . The types of costs, populations and tools used for eliciting utilities and additional results for each study are reported in Report Supplementary Material 7.
Summary of findings from studies included in this analysis
Whole enhanced recovery pathway
Larsen et al.109 report an economic analysis alongside a randomised trial, with 56 hip replacement and 31 knee replacement participants in Denmark. The pathway used in the treatment arm (‘accelerated care’) was not different to that of the control group (‘conventional rehabilitation’) in terms of intraoperative management, analgesia, nausea control or bowel regulation. Differences in the treatment protocols between the two arms involved patient education, nutrition, admission times, staffing and mobilisation (described in Report Supplementary Material 7). The accelerated care pathway was less costly and more effective than the control, both overall and in the subgroup of hip replacement participants. For knee replacement patients, the authors reported the accelerated care pathway to be cost-saving but less effective than the control, albeit the difference was not statistically significant. They reported a cost saving of 618,075 Danish krone per QALY lost with accelerated care compared with conventional rehabilitation, which made it cost-effective (threshold of 160,000 krone per QALY in Denmark).
Brunenberg et al.98 report an economic evaluation of a before-and-after trial with 98 hip and 62 knee replacement participants in the Netherlands. The intervention (‘joint recovery programme’) consisted of a 20-minute pre-assessment screening 6 weeks before the operation, for physical assessment and analysis of the home situation to aid discharge planning, patient education sessions 1–2 weeks before surgery, group rehabilitation sessions and supervision by physical therapists and nurses (described in Report Supplementary Material 7). Patients in the ‘usual care’ group underwent conventional physiotherapy for 1 hour daily, did not receive pre-assessment screening or information sessions, and discharge arrangements were addressed during admission to hospital. The joint recovery programme intervention was less costly and more effective than the control for both hip and knee replacement, resulting in a cost saving of US$1261 per patient for hip replacement and US$3336 per patient for knee replacement, albeit with no statistically significant difference in QALYs. The probability that the joint recovery programme was the most cost-effective option was > 80% for willingness-to-pay thresholds up to US$45,000.
Preoperative components of the enhanced recovery pathway
McLawhorn et al.110 used a Markov model to assess the cost-effectiveness of bariatric surgery 2 years before total knee arthroplasty for morbidly obese (BMI ≥ 35 kg/m2) patients who were candidates for both operations because of end-stage knee osteoarthritis and failed non-operative weight loss interventions. The strategy including bariatric surgery was cost-effective at the stated willingness-to-pay threshold of US$100,000 per QALY in 98.8% of probabilistic simulations.
Courville et al.97 compared three strategies of screening for and treating S. aureus colonisation to prevent deep surgical site infections following total knee arthroplasty and total hip arthroplasty. Courville et al.97 found that decolonisation of all preoperative patients with mupirocin, without testing for S. aureus, was the dominant strategy (cheaper and more effective) when compared with treating only patients testing positive for S. aureus, or no screening or decolonisation for S. aureus.
Intraoperative components of the enhanced recovery pathway
Two studies investigated strategies to reduce allogenic blood transfusions: collecting autologous blood prior to surgery100 or aseptic collection of wound drainage.99 Autologous blood prior to surgery was found to be cost-effective, at US$2750 per QALY gained, whereas aseptic collection of wound drainage was not, at a cost of US$5.7M per QALY gained.
Marques et al.101 conducted economic evaluations alongside two randomised controlled trials of adding local wound infiltration with bupivacaine to usual anaesthetic care for total hip arthroplasty and total knee arthroplasty.101 The infiltration of local anaesthetic was found to be less costly and more effective than standard anaesthesia in both hip and knee replacement patients. Three studies used Markov models to investigate measures to reduce surgical site infection in the USA,103 UK102 and Australia.104 The cheapest and most effective measure in all three included use of antibiotic-impregnated cement. The two studies that looked at other factors102,104 each found the use of prophylactic systemic antibiotics to be less costly and more effective than non-use, and the use of conventional ventilation in operating theatres to be less costly and more effective than laminar airflow ventilation. Graves et al.102 considered use of body exhaust suits and found them to be dominated by strategies that did not include use of these suits.
Postoperative components of the enhanced recovery pathway
Fusco and Turchetti105 used a Markov model to evaluate a strategy of 10 face-to-face rehabilitation sessions followed by 10 telerehabilitation sessions after knee replacement, compared with 20 face-to-face sessions. They found the strategy including telerehabilitation to be cost-saving and to improve range of movement (knee flexion). However, they found no utility data for patients following a telerehabilitation programme, so for their base case assumed it to be non-inferior to face-to-face rehabilitation. In a sensitivity analysis, if telerehabilitation conferred an improvement in quality of life of at least 2.5%, the strategy’s probability of being cost-effective was 1 (at a willingness-to-pay threshold of €30,000/QALY).
Kauppila et al.86 performed an economic evaluation of a 10-day outpatient rehabilitation course between 2 and 4 months after knee replacement, which included clinical assessments, physical activity, sessions with a psychologist, and lectures from an orthopaedic surgeon, nutritionist and social worker.107 They found that patients who completed this course had higher costs and slightly worse quality-of-life outcomes over the 1 year of follow-up (albeit not statistically significant) than those receiving conventional orthopaedic care.
Bolz et al.106 compared three follow-up strategies: 2-yearly routine follow-up, follow-up twice (at 3 months, and between 1 and 2 years after surgery) or no follow-up. The model assumed that no revisions would be delayed in either strategy that included follow-up and the outcomes for these two strategies were identical in each analysis. The no follow-up strategy was dominant for any assumed rate of delayed revision between 1% and 50%.
Assessment of study quality and reporting quality
Using the CHEC list, the quality of the studies was generally good (). However, the time horizons in the four trial-based studies (1 year86,98,101,109) and two model-based studies (1 year97 and 7 years106) were too short to capture all relevant outcomes or costs. Furthermore, most studies did not report whether or not they had conflicts of interest.
Assessments of study quality based on tools from (a) CHEC; (b) ISPOR questionnaire; (c) AdViSHE tool; and (d) the Cochrane Collaboration.
Using the Cochrane Collaboration’s tool (see ), the risk of bias of the trial-based studies was rated as being low for items such as incomplete outcome data and selective reporting, but high or uncertain for the remainder. Three trial-based studies stated that participants were allocated at random;79,101,107 the fourth was a before-and-after trial whereby patients were recruited consecutively from a waiting list for hip and knee replacement.98 Larsen et al.79 discussed stratification and Kauppila et al.107 reported a computer-generated sequence. The risk of bias could not be assessed accurately, as the random sequence generation was not described precisely in any of the studies. The staff administering the interventions were not blinded to allocation in any trial, and in three trials98,101,107 outcomes were assessed by researchers aware of the treatment allocation.
Using the ISPOR questionnaire, the quality of the model-based studies was generally good (see ). However, none of the model-based studies reported a detailed process for internal and external validation. Three studies99,100,105 were based on previously published models.111–113 Of these, only Briggs et al.111 (the basis for Fusco and Turchetti’s model105) provided details of model validation. Further limitations in model validation were highlighted with the AdViSHE tool (see ).93
Methods of estimating costs of joint replacement
Setting and data sources
We used an incidence-based approach to estimate the primary and hospital care costs. Individuals undergoing an elective admission for joint replacement were identified and followed up retrospectively using data from three sources: the NJR, HES and CPRD GOLD (see Chapter 2).
Study participants
To estimate hospitalisation costs, we included only individuals identified in the NJR, HES and PROMs linked data set with a planned surgery for joint replacement between April 2008 and January 2017 (observation period). Patients without a concordant date of replacement between NJR and HES databases were excluded from the analysis. If an individual had a primary joint replacement and a contralateral joint replacement during the observation period, we counted the costs from the primary joint replacement and included the patient only once in the analysis. To estimate outpatient and primary care costs, we included only patients in the CPRD GOLD data set with a first ever clinical or referral record of planned joint replacement occurring from 1 April 2008 until 31 December 2016. For further detail on cost methodology, see Report Supplementary Material 9.
Death, complications and revisions at 1 year
All-cause mortality was estimated at 30 days and 1 year from the day of planned admission for joint replacement and using the date of death from the ONS mortality database.
For hospitalisation costs, we defined postoperative complications as one or more events happening up to 1 year after joint replacement. We also identified revisions occurring up to 1 year following joint replacement from revisions declared to the NJR registry by the surgeons and revisions reported to HES.
Costs
Hospitalisation costs were estimated by converting the diagnosis and operational codes for each finished consultant episode in a hospital admission into a HRG via the 2016/17 case mix grouper software (HRG4+ 2016/17 Local Payment Grouper, NHS Digital, UK).36 HRGs are standard groups of clinically similar treatments that consume a common set of health-care resources. HRGs for each finished consultant episode were then valued using NHS Reference Costs 2016–17114 and appropriate methodology,115 and summed to produce the total cost per hospital admission.
Primary care contacts included GP consultations in clinic or surgery, telephone contacts and out-of-office visits. These also included nurse face-to-face and non-face-to-face contacts and contacts with other community health-care professionals (e.g. health visitor or physiotherapist). Primary care contacts and tests were costed using 2016/17 unit costs from national cost databases.116 Pharmaceuticals were costed by matching each prescribed medication to a British National Formulary code and valuing these using 2016/17 cost data from NHS Digital Prescription Cost Analysis.117 See Report Supplementary Material 9 for more details on the cost methodology. Total costs per patient were aggregated into monthly and annual amounts for the purposes of the analysis.
Statistical analysis
We report total hospital inpatient costs for patients with complete follow-up data at years 1 and 2 following joint replacement, and for the whole sample after adjusting for censoring using the methodology developed by Lin et al.118 Costs are reported as means together with their 95% CIs, obtained from 1000 bootstrap estimates. We estimated the total annual joint replacement costs in the UK by multiplying the NHS primary and hospital costs in the year of surgery by the number of primary joint replacements in 2017 (96,717 hip replacements and 106,334 knee replacements).
Predictors of hospitalisation costs of joint replacement were estimated using a generalised linear model (GLM). After reviewing the literature, we examined the following predictors of costs in the year of the joint replacement: age, sex, EuroQol-5 Dimensions, three-level version/OHS/OKS before surgery and change at 6 months, complications and revision up to 1 year after surgery, social deprivation index; Charlson Comorbidity Index score, BMI prior to surgery, type of joint replacement (unicondylar, total and patellofemoral for knee; resurfacing or total for hip), surgical variables, ASA grade before surgery, thrombolysis agents used (low-molecular-weight heparin, none, aspirin and other), type of anaesthesia (general, epidural, spinal and nerve block), death and year of surgery.
We used t-test and Pearson’s chi-squared test to evaluate the missingness for the potential predictors of costs (e.g. BMI, EQ-5D/OHS/OKS before surgery and change at 6 months) in terms of age, sex, hospitalisation costs, LOS, Charlson Comorbidity Index score and type of joint replacement. We also performed multiple imputation of the missing data using a chained model with 20 iterations regressed on complete variables to inform the prediction models (see Report Supplementary Material 9 for more details).
The choice of the GLM family and link functions was informed by the modified Park test and the Box–Cox test, respectively. We applied stepwise backward selection (at a p-value of < 0.05) per 300 bootstrap samples to identify variables that were consistently selected for at least 50% of the analyses and to inform the final models. A two-tailed t-test with α = 0.01 (to account for the large sample size) was used to determine whether or not each coefficient was statistically significantly different from zero, and their selection as predictors of costs was informed using Akaike’s information criterion, mean square error and likelihood test. All analyses were performed using Stata.
Results of costs of joint replacement
We identified 397,119 and 457,747 patients as having had a primary THR or TKR, respectively, in the NJR, HES and PROMs linked data set. reports the baseline characteristics of the two cohorts. There was a lower proportion of women in the THR cohort than in the TKR cohort (40.4% vs. 57.0%). More younger individuals (69.1 years vs. 69.5 years) with lower BMI (28.8 kg/m2 vs. 30.7 kg/m2) underwent THR than TKR. Furthermore, the Oxford and EQ-5D scores were slightly lower in the THR cohort than in the TKR cohort (17.4 vs. 18.2 for OHS/OKS and 0.33 vs. 0.37 for EQ-5D).
Patient characteristics of study cohorts at primary joint replacement
Hospitalisation costs
The follow-up for the THR and TKR cohorts was 3.9 years (SD 2.5 years) (). In the first year after the joint replacement, 1.2% (4071/344,721) and 0.8% (2965/394,118) of individuals died in the hip and knee cohorts, respectively. An improvement at 6 months was reported for both cohorts using the EQ-5D and OHS/OKS instruments.
Patient outcomes and hospitalisation costs
Hospitalisation costs associated with index admission were £6208 (median £5824, SD £969) for THR compared with £6122 (median £5692, SD £967) for TKR. The mean LOS in the index admission was 4.82 [median 4, SD 3.8, interquartile range (IQR) 3–6] days and 4.77 (median 4, SD 3.5, IQR 3–5) days for THR and TKR, respectively. These costs and LOS represent averages across the whole observation period. However, we found no clear trend in costs over the observation period when resource use was valued using the recommended NHS Reference Costs 2016–17114 ().
Hospitalisation costs between 2008 and 2016
The mean 1-year hospitalisation costs were estimated to be £7817 (median £6258, SD £4618) and £7784 (median £6226, SD £4520) for THR and TKR, respectively, of which the index admission accounted for 79.4% and 78.5% of the total. Hospitalisation costs and LOS within 1 year were highly correlated for both types of joint replacement (Spearman’s correlation coefficient 0.84). Comparing 1-year costs over the observation period, we found that joint replacements decreased in each year relative to the previous year (see ). In 2015, TKR and THR 1-year costs were, respectively, £194 (95% CI £136 to £252) and £253 (95% CI £191 to £315) lower than in 2008, adjusting for other covariates.
Musculoskeletal diagnosis (ICD-10 chapter 1335) accounted for 32–35% of readmission costs within the first year of joint replacement, followed by injury diagnosis (ICD-10 chapter 19:35 21%) and circulatory system diagnosis (ICD-10 chapter 9:35 8–9%) ().
Top five diagnosis groups for hospitalisation costs following joint replacement by ICD-10 code and year
Individuals undergoing partial knee replacement had, on average, lower 1-year costs and LOS [mean £6897 (SD £3263) and 4.4 (SD 5.9) days, n = 29,066] than individuals undergoing patellofemoral joint replacement [£7381 (SD £3892) and 5.4 (SD 7.7) days, n = 4370] and TKR [£7860 (SD £4606) and 7.5 (SD 11.5) days, n = 360,682]. Individuals undergoing metal-on-metal (MoM) resurfacing had, on average, lower 1-year costs and LOS [mean £6553 (SD £2900) and 4.8 (SD 5.8) days, n = 7000] than individuals undergoing THR [£7842 (SD £4643) and 7.4 (SD 11.3) days, n = 337,721].
After adjusting for censoring, the mean 1-year costs were similar to the complete-case analysis, at £7827 (95% CI £7813 to £7842) and £7805 (95% CI £7790 to £7818) for THR and TKR, respectively. For THR, the mean cost in the first 2 years following joint replacement (2-year) adjusted for censoring was £9258 (95% CI £9233 to £9280), compared with £9277 (SD £6927) using only the complete cases (n = 293,618). For TKR, the cost in the first 2 years following joint replacement (2-year) adjusted for censoring was £9452 (95% CI £9430 to £9475), similar to £9446 using only complete cases (n = 333,123). Comparing 2-year costs over the observation period, we found that joint replacements decreased each year relative to the previous year (see ). In 2014, TKR and THR costs were £373 (95% CI £283 to £463) and £323 (95% CI £233 to £413) lower than in 2008, respectively, adjusting for other covariates.
Predictors of hospitalisation costs in the first year following joint replacement
Data were missing for 50% and 70% of patients concerning Oxford and EQ-5D scores (before surgery and at 6 months), BMI or other predefined variables to inform the prediction of hospitalisation costs for hip and knee replacement, respectively. The cohorts with incomplete data had a lower proportion of total joint replacements, higher mortality rates at 1 year, higher hospitalisation costs and higher LOS than complete cases. Following multiple imputation, the predictors of hospitalisation costs for THR and TKR are shown in and , respectively. A GLM with gamma family and identity link function was appropriate and had a good fit.
Predictors of 1-year hospitalisation costs following hip replacement (n = 330,765)
Predictors of 1-year hospitalisation costs following knee replacement (n = 391,691)
Holding all else constant, conventional THR was more expensive, on average, than MoM resurfacing (£451; p < 0.001). Women had higher mean hospitalisation costs than men (£167; p < 0.001) and increasing age was associated with higher costs (£28 per additional year). Individuals with higher quality-of-life values (EQ-5D and OHS) at baseline and reporting improvements at 6 months were associated with lower hospitalisation costs. Larger femur head size was associated with higher costs, with additional costs of up to £226 for head sizes of > 53 mm compared with the cost for head sizes of ≤ 28 mm. We found significant variation in hospitalisation costs concerning bearing surfaces with strong evidence (p < 0.01) that ceramic-on-ceramic, ceramic-on-metal and metal-on-ceramic bearings were associated with lower mean 1-year hospitalisation costs than metal-on-polythene bearings (the most common bearing type in the cohort). Costs were lower in recent years (–£31/year; p < 0.001), holding all else constant. Complications and revisions within the year were significantly associated with higher mean costs, with an additional £6601 (p < 0.001) and £11,255 (p < 0.001), respectively.
Holding all else constant, TKR was significantly associated with higher 1-year hospitalisation costs than unicondylar knee replacement (£391; p < 0.001). Women had higher mean hospitalisation costs than men (£248; p < 0.001) and costs were positively associated with age (£31 per additional year; p < 0.001) and higher deprivation, with individuals living in more deprived areas having higher costs. Individuals with higher quality-of-life values (EQ-5D and OKS) at baseline and those reporting improvements at 6 months had lower hospitalisation costs. Higher deformity and lower range of flexion were also significantly associated with higher costs. Costs were lower in recent years (–£10/year; p < 0.001), holding all else constant. Complications and revisions within the first year were significantly associated with higher costs, with an additional £6075 (p < 0.001) and £7753 (p < 0.001), respectively.
The mean 1-year hospitalisation costs were higher for patients who died (an additional £4682 for THR and £4597 for TKR), mostly as a result of higher LOS than those who survived the first year (mean of 30.4 days vs. 7.0 days for hip patients who did/did not die and 30.0 days vs. 7.0 days for knee patients who did/did not die).
Costs before and after joint replacement
Adding primary, outpatient and inpatient hospitalisation costs, the mean NHS costs associated with THR amounted to £9295 in the year of surgery () compared with £9483 following TKR (). Using the annual number of UK primary joint replacements in 2017, the NHS primary and hospital care costs were estimated at £897M (n = 96,717) and £1007M (n = 106,334) in the year of the THR and TKR, respectively.
Costs in the years before and after THR.
Costs in the years before and after TKR.
Hospitalisation costs accounted for the highest proportion of the total 1-year cost for both THR and TKR (82–84% of total costs in the year of joint replacement), followed by primary care (10–11% of total costs) and outpatient care (6–7% of total costs). In the second year after joint replacement, total costs decreased to £3095 for knee and £2692 for hip cohorts, with inpatient costs again being the largest component (53% for both knee and hip).
reports the hospitalisation costs in the months before and after joint replacement. The annual hospitalisation costs in the year of joint replacement were £6753 (95% CI £6732 to £6774) and £6563 (95% CI £6544 to £6583) higher for THR and TKR, respectively, than those of the previous year (). However, hospitalisation costs decreased in the 5 months prior to surgery, reflecting lower hospital admissions leading up to the elective admission. In the year of joint replacement, about 81–83% of hospitalisation costs occurred in the first month post joint replacement. Costs in the second year after joint replacement were £389 (95% CI £370 to £407) and £349 (95% CI £329 to £368) higher than costs in the year prior to surgery for knee and hip replacement, respectively.
Hospitalisation costs in the months before and after joint replacement.
Hospitalisation costs in the years before and after joint replacement
A similar pattern was observed with outpatient and primary care costs ( and ). Primary care costs in the year of the surgery were higher than in the previous year, at £64 (95% CI £37 to £91) and £5 (95% CI –£17 to £26) for TKR and THR, respectively (). The highest costs occurred in the first month after joint replacement. Outpatient costs were also higher in the year of surgery than in the previous year, with an additional £80 (95% CI £53 to £107) and £15 (95% CI –£31 to £61) for TKR and THR, respectively, with the highest costs occurring in the second month (). However, outpatient costs in the second year after surgery were significantly lower than in the year preceding the surgery for both types of joint replacement [–£105 (95% CI –£78 to –£133) and –£126 (95% CI –£109 to –£143) for knee and hip, respectively]. In contrast, primary care costs were lower in the second year after surgery for THR (by –£53), but higher for TKR (by £37) than in the year preceding surgery.
Primary care costs in the months before and after joint replacement.
Outpatient care costs in the months before and after joint replacement
Primary care costs in the years before and after joint replacement
Outpatient care costs in the years before and after joint replacement
Discussion
The systematic review of cost-effectiveness evidence of enhanced recovery identified two studies evaluating an entire pathway.98,109 This is consistent with reviews of cost-effectiveness of enhanced recovery programmes for other surgical sites,119–121 which have found few studies reporting the effect on quality of life and none presenting cost-effectiveness results using QALYs. The ERP was found to be associated with reduced costs for all patients and the incremental cost-effectiveness estimate favoured the enhanced recovery protocol, with a high probability of it being the most cost-effective option. Furthermore, we were able to examine the hospitalisation costs of different types of joint replacement through their identification in the NJR data set and linkage to hospital records. We found partial knee replacement to have lower 1-year costs than TKR, even after adjusting for potential confounders. Previous studies have examined the costs of joint replacement, but consisted of smaller samples and without linkage to NJR data.74,122
Cost-effectiveness of enhanced recovery
We identified 11 studies presenting cost–utility data for components of an ERP.86,97,99–106,110 However, these cover only a few of the potential ERP components and the available evidence was mostly based on single studies. These studies covered a variety of interventions, across different health-care systems, and used different cost perspectives. The studies identified in this review were generally of good quality, according to the CHEC list,91 with a short time horizon identified as a key limitation in six studies.86,97,98,101,106,109 When the models were assessed against the ISPOR questionnaire,123 there were concerns about the lack of model validation work, potentially questioning the reliability of 9 of the 13 studies identified.97,99,100,102–106,110 The trials were generally of good quality.
Our review had some limitations. We may have missed relevant evidence by limiting our search to reports published in the English language and excluding studies that did not report QALYs. Of the studies excluded because they did not report QALYs, one investigated a complete ERP.124 This was a retrospective review of total knee arthroplasty patients in Turkey who followed either a rapid-recovery protocol or a standard care protocol. The rapid-recovery protocol was found to be cost-saving and associated with statistically significant differences in knee flexion and extension at 6 months.
Costs of joint replacement
Consistent with Chapter 5, we found 1- and 2-year hospitalisation costs to decrease over recent years. However, we found no clear trend concerning the costs of the elective surgery itself, despite a significant decrease in LOS between 2008 and 2016. This can be explained by the nature of costing in the NHS, which is based on HRGs for hospital episodes rather than days in hospital. Therefore, it is key to note that such a costing approach was shown here not to be sensitive to capture what we would expect to be a decrease in costs.
We identified a reduction in hospitalisation costs in the 5 months prior to surgery for both types of joint replacement, reflecting fewer hospitalisations leading up to the elective admission. This suggests that the selection of patients for joint replacement takes into account their recent history of hospital admissions. Furthermore, primary care costs were slightly lower in the second year after surgery for THR, but slightly higher for TKR, than in the year preceding surgery, possibly reflecting differences in recovery times between the two procedures.
The main predictors of costs were similar for THR and TKR. As in previous work,122 we found preoperative quality of life to be associated with hospitalisation costs: 1-year costs were higher for individuals with worse preoperative quality of life, even after adjusting for other covariates. In addition, 1-year costs were lower for individuals reporting improvements in quality of life at 6 months.
Having a revision or complication was found to be associated with very high additional costs in the first year after surgery, after adjusting for other covariates. Given the high annual primary care and hospital costs of £897M and £1007M for THR and TKR, respectively, there is a considerable economic incentive to fund research aimed at identifying cost-effective ways of improving the quality of life of patients following joint replacement and reducing the risk of revisions and complications.
Our study had some limitations. NJR data were obtained for individuals undergoing joint replacement, with osteoarthritis as an indication for surgery. Hence, individuals without osteoarthritis as one of the indications were not available for analysis (e.g. rheumatoid arthritis or hip fracture). A large proportion of the cohort had missing data for one or more key covariates of the hospitalisation costs, in particular EQ-5D or OHS responses and BMI, which necessitated the use of missing data methods, specifically multiple imputation. A key assumption whenever multiple imputation is utilised is that the missing data may be classified as missing at random. This assumption is always untestable, but owing to the large number of relevant covariates in our linked data we judged it to be reasonable in this case.
