Lower limb arthroplasty: can we produce a tool to predict outcome and failure, and is it cost-effective? An epidemiological study

Temporal trends in hip and knee replacement in the UK67

Total joint arthroplasty is a successful surgical intervention and is considered reliable and effective for pain relief and improved function and quality of life, with 90% prosthesis survival at 10 years. 3,8 In total, 160,000 hip and knee replacements were carried out in England and Wales in the 12 months prior to April 2010. 70 This number is expected to rise, and studies from the USA predict an increase in hip replacements (174%) and knee replacements (673%) to nearly 3.5 million per annum by 2030. 66 More than 650,000 knee replacements alone were carried out in the USA in 2008,100 and almost 80,000 in the UK in 2009. 17 The USA saw the number of knee surgeries increase from 31.2 per 100,000 person-years [95% confidence interval (CI) 25.3 to 37.1 per 100,000 person-years] in 1971–1976, to 220.9 per 100,000 person-years (95% CI 206.7 to 235.0 per 100,000 person-years) in 2008. 101

For the analysis we selected the patient data with a medical diagnosis code for THR (n = 27,113) or TKR (n = 23,843) between 1991 and the end of 2006. Patients were included if they were aged > 18 years at the time of operation. Evidence suggests that 20% of THR/TKRs in England are carried out in private institutions102 and the impact of this on arthroplasty provision needs further investigation. However, as the CPRD database had not been validated at this time, private practice codes were not included in the analysis. The NHS rates were validated using the HES (2005–6)103 and NJR. 104

Directly age- and sex-standardised replacement rates for calendar years were calculated using 10-year age groups with the mid-year population estimates for 2003 as the reference standard, as published by the ONS,105,106 the General Register Office for Scotland and the Northern Ireland Statistics and Research Agency. The 95% CI was computed using the Poisson model appropriate for directly standardised rates. The mean age at total replacement was calculated for the hip and knee for each calendar year and 95% CIs computed. To investigate patterns over time we calculated age distribution at operation by sex for three consecutive 5-year periods for the hip and knee.

The estimated age-standardised rate of THR increased from 60.3 per 100,000 person-years (95% CI 53.7 to 67.0 per 100,000 person-years) to 144.6 per 100,000 person-years (95% CI 138.1 to 151.1 per 100,000 person-years) for women (Figure 1a), and from 35.8 per 100,000 person-years (95% CI 30.4 to 41.3 per 100,000 person-years) to 88.6 per 100,000 person-years (95% CI 83.4 to 93.7 per 100,000 person-years) for men (Figure 1b). The increase in rates over time for THRs were steady between 1993 and 2005. The rate of TKR increased from 42.5 per 100,000 person-years (95% CI 37.0 to 48.0 per 100,000 person-years) to 138.7 per 100,000 person-years (95% CI 132.3 to 145.0 per 100,000 person-years) for women (see Figure 1a), and from 28.7 per 100,000 person-years (95% CI 23.9 to 33.6 per 100,000 person-years) to 99.4 per 100,000 person-years (95% CI 93.9 to 104.8 per 100,000 person-years) for men (see Figure 1b). The temporal trend for knees showed a marked plateau from the mid-1990s, followed by a sharp rise from 2000.

FIGURE 1.

Trends in primary THR and TKR rates in (a) women; and (b) men, based on data from Culliford et al. 67

The mean age at operation was significantly higher for women than for men for all years after 1991: the mean age at THR was 70.3 years (95% CI 69.8 to 70.8 years) for women and 67.6 years (95% CI 66.9 to 68.2 years) for men; and the mean age at TKR was 70.1 years (95% CI 69.6 to 70.5 years) for women and 69.2 years (95% CI 68.6 to 69.7 years) for men. The highest rates of THRs and TKRs were for women aged between 70 and 79 years, with a mean rate of THR of 541.8 per 100,000 person-years (95% CI 501.0 to 582.5 per 100,000 person-years) and of TKR of 555.3 per 100,000 person-years (95% CI 514.1 to 596 per 100,000 person-years).

The final results showed that the rates of hip and knee arthroplasty continued to increase, but that the rise was more marked for knees than for hips. Women were 67% more likely than men to undergo THR, and 45% were more likely to undergo TKR, but sex ratios have been consistent over time, as demonstrated in Figure 2.

FIGURE 2.

The sex ratio for the number of hip and knee replacements vs. year of procedure based on data from Culliford et al. 67

Women were, on average, 3 years older than men at THR, but the age difference between men and women undergoing knee replacements was only half as great. BMI was significantly higher for patients undergoing TKR than for those undergoing THR (p < 0.0001) and was higher for women than for men. There was little sex difference in the number of replacements carried out in patients between the ages of 60 and 79 years, who made up almost two-thirds of the total number of patients undergoing arthroplasty during 1991–2006.

Unicompartmental knee replacements are also becoming more popular, and the rates have increased over the last decade. We, again, used CPRD data and Read/OXMIS codes to identify all patients who underwent primary TKR or UKR between 1986 and 2006. However, the final analysis was restricted to the use of data between 1999 and 2006, as very few UKRs were carried out before this time.

The results of the statistical analysis give the number of TKRs and UKRs performed in each year; the mean age [and standard deviation (SD)] of patients of each sex and undergoing each type of operation was calculated to explore the profile for each intervention. The total numbers of TKRs and UKRs performed in the UK in 2006 were estimated by applying the CPRD rates to the population of the UK in that year.

There were substantially more TKRs (n = 18,450) than UKRs (n = 266) in 2006. The rate of TKRs increased from 55.4 per 100,000 person-years in 1999 to 123.5 per 100,000 person-years in 2006. The rate of UKRs increased from 0.25 per 100,000 person-years in 1999 to 3.0 per 100,000 person-years in 2006. Both men and women undergoing UKR were, on average, younger than those undergoing TKR (p < 0.0001). Men who underwent TKR were, on average, younger than women undergoing TKR (p < 0.0001). There was no statistically significant difference between the mean age of men and women undergoing UKRs (p = 0.74). TKR was performed more often in women (n = 10,836) than in men (n = 7614), but UKR was performed less often in women (n = 126) than in men (n = 140). The ratio of TKRs to UKRs fell from 250 : 1 in 1999 to 40 : 1 in 2006. The estimated numbers of operations performed in 2006 were 74,800 TKRs and 1800 UKRs.

The results showed a 12-fold increase in UKRs since 1999, and that this was still significantly less than TKRs, and UKRs are performed on a younger age group than TKRs.

The lifetime risk of total hip and knee arthroplasty: results from the UK General Practice Research Database99

Lifetime risk is a patient-centred measure of risk for the onset of disease or the occurrence of specific events. The concept is easily understood by clinicians and policy-makers, and can be made even more informative by also calculating interval risks (e.g. 10 years) at different ages to establish the periods of greatest lifetime risk. Population-based estimates are needed for effective and efficient health-care planning and resource allocation. No lifetime risk estimates were available in the literature for patients who were undergoing these surgical procedures, but published incidence rates existed for hip and knee replacement in the UK67,116 and internationally. 119,123,124

The primary aim of this analysis was to use the CPRD database combined with the ONS mortality data to provide estimates for the lifetime risk of undergoing a primary THR or TKR in the UK. OXMIS/Read codes were used to identify THRs and TKRs for the period 1991–2006 in the CPRD database. Patients were included if aged ≥ 50 years at the time of replacement. Sex-specific all-cause mortality data from the ONS were obtained for the same period. 125

The analysis was done with CPRD data that were aggregated into single-year age intervals, with the age label defined as age at last birthday at the end of a calendar year, starting at the age of 50 years. We used data for the time period 1991–2006 and identified 49,105 patients who had undergone a THR (n = 25,845) or TKR (n = 23,260). Consistent definitions were applied to death data and exposure to risk. Incidence rates for joint replacement were computed by dividing the count of primary THRs and TKRs in the CPRD data by the corresponding amount of person-time spent by those in the entire CPRD population who matched the age band, sex and time interval of interest. This was achieved by a life table method similar to that described by Kim et al. 39 CIs at the 95% level were estimated under a Poisson model. 126 Risks were estimated separately for sex and hip/knee. This was repeated with 60, 70 and 80 years of age as the starting point for the risk of replacement. We further computed 10-year risk percentages from age 50 years up to the age of 80 years. All estimates for single calendar years used mortality data matched to the same calendar years, but for the estimates based on the entire study period we used 2006 mortality rates with a sensitivity analysis. Lifetime risks of THR and TKR, stratified by sex for individual calendar years, were estimated in order to compare temporal trends.

The results, using rates from 2005, showed that the estimated mortality-adjusted lifetime risk of THR at age 50 years was 11.6% for women and 7.1% for men. For the aggregated data over the period 1991–2006, the mortality-adjusted lifetime risk of THR at age 50 years was estimated at 8.3% for women and 5.2% for men. The lifetime risk of THR at age 50 years rose from 4.0% (95% CI 3.0% to 5.0%) to 11.1% (95% CI 9.9% to 12.2%) for women and from 2.2% (95% CI 1.4% to 3.0%) to 6.6% (95% CI 5.7% to 7.5%) for men. Therefore, our findings estimated that, between 1991 and 2006, the lifetime risk of THR at age 50 years rose from 4.0% (95% CI 3.0% to 5.0%) to 11.1% (95% CI 9.9% to 12.2%) for women and from 2.2% (95% CI 1.4% to 3.0%) to 6.6% (95% CI 5.7% to 7.5%) for men.

Again, using the rates from 2005, we estimated that the mortality-adjusted lifetime risk of TKR at age 50 years was 10.8% for women and 8.1% for men. The aggregated data for the period 1991–2006 estimated the mortality-adjusted lifetime risk for TKR at age 50 years at 7.0% for women and at 5.2% for men. The same time period for TKR saw an increased risk for women from 2.9% (95% CI 2.1% to 3.8%) to 10.6% (95% CI 9.5% to 11.7%) and for men from 1.8% (95% CI 1.1% to 2.6%) to 7.7% (95% CI 6.8% to 8.7%). As with hips, TKR estimates of risk also increased, for women from 2.9% (95% CI 2.1% to 3.8%) to 10.6% (95% CI 9.5% to 11.7%) and for men from 1.8% (95% CI 1.1% to 2.6%) to 7.7% (95% CI 6.8% to 8.7%).

As a sensitivity analysis these estimates were recalculated using 1991 mortality data, but this resulted in only small reductions in the lifetime risk estimates of between 0.6 and 0.8 percentage points at age 50 years and of 0.2 and 0.3 percentage points at age 80 years. These reductions were seen for both THR and TKR, and for men and women.

The lifetime risk decreases with increasing age for both THR and TKR in men and women. At age 80 years, the sex gap in risk of THR reduced to 40% higher for women than for men (22% higher for TKR). Estimated risk percentages at ages 50, 60, 70 and 80 years are presented in Table 7.

The sex gaps in the estimates obtained for the whole study period were similar to those for the 2005 estimates.

Our results showed that between 1991 and 2006 the lifetime risk of THR at age 50 years increased from 4.0% (95% CI 3.0% to 5.0%) to 11.1% (95% CI 9.9% to 12.2%) for women and from 2.2% (95% CI 1.4% to 3.0%) to 6.6% (95% CI 5.7% to 7.5%) for men. For TKR, the risk increased for women from 2.9% (95% CI 2.1% to 3.8%) to 10.6% (95% CI 9.5% to 11.7%) and for men from 1.8% (95% CI 1.1% to 2.6%) to 7.7% (95% CI 6.8% to 8.7%) (Figure 8).

FIGURE 8.

Estimated lifetime risk at age 50 years of undergoing (a) THR; or (b) TKR based on age- and sex-specific incidence adjusted for mortality. Data from the GPRD. 99

The lifetime risks of hip and knee replacements are estimated to be between 5% and 10%, which is substantially below the estimated lifetime risk of hip and knee OA. Our estimates based on UK GPRD data from 2005 suggest a lifetime risk of THR and TKR for women or men aged 50 years living in the UK of 10–11% and 6–7%, respectively.

Future projections of total hip and knee arthroplasty in the UK: results from the UK Clinical Practice Research Datalink98

Future predictions for lower limb arthroplasty in the UK are limited,116,123 and international predictions are mainly concentrated around the USA and Europe. 34,68,127,128 With the steady increase in rates of hip and knee surgery, up-to-date future predicted rates are necessary as part of our understanding of this treatment intervention. The most recent published future projections of the UK covered England only,116 were based on a 10-year period of HES data and did not account for BMI changes or other important risk factors for arthroplasty.

Deciding on the correct method for forecasting is important and dependent on high-quality research data. More sophisticated modelling approaches require at least one population-based cohort or survey data set with long-term follow-up. 129

We produced and published the national future projections for THR and TKR. 98 We used three national data sets that were representative of age, sex and BMI in the UK population. Using the CPRD database, in combination with national population forecasts from the ONS, we aimed to calculate age- and sex-specific forecasts for the number of THR and TKR operations per year in the UK between 2010 and 2035. Secondary analysis aimed to produce forecasts that reflect the changing distribution of BMI during the same period. To project estimated THR and TKR rates, HSE data were used. We constructed a denominator to estimate BMI-specific rates and obtained sex-specific population projections from the ONS for the period 2011–35. 106 The methods of Kurtz et al. 66 were further extended to incorporate the inclusion of BMI, in addition to age and sex.

European Collaborative Database of Cost and Practice Patterns of Total Hip Replacement137,138

The European Collaborative Database of Cost and Practice Patterns of Total Hip Replacement (EUROHIP) study consists of 327 patients receiving THR treatment across 20 orthopaedic centres in Europe. Patients completed self-administered questionnaires on demographic variables and baseline of pain, stiffness, mobility and quality of life. In addition, they were asked about their expectations of surgery 1 year after the operation. We also collected preoperative radiographs and data on operative procedure, including the prosthesis type used. Patients undergoing primary hip replacement in whom the indication for surgery was OA were included, but those with hip disease other than OA, severe mental conditions and/or dementia were excluded.

Exeter Primary Outcomes Study139

In the Exeter Primary Outcomes Study (EPOS), participants were recruited between January 1999 and January 2002 from seven centres in England and Scotland. Patients received primary THR using a cemented Exeter femoral stem component (Stryker Howmedica Osteonics, Mahwah, NJ, USA). 140 A variety of cemented and uncemented acetabular components were used in included patients. Ethics approval was obtained from the North Western Multiple Centre Research Ethics Committee and the local research ethics committees. The cohort was representative of a wider orthopaedic cohort, as the participating hospitals (both teaching and district general hospitals) covered a wide geographical area including urban and rural locations; thus, it covered both affluent and somewhat deprived inner-city suburban areas with an overall population of 1 million.

Elective Orthopaedic Centre141

The Elective Orthopaedic Centre (EOC), also known as the South West London Elective Orthopaedic Centre (SWLEOC), is a purpose-built orthopaedic treatment centre that was opened in 2004. The centre serves a population of 1.5 million people in south-west London and it performs TKR surgeries across four acute NHS trusts: Kingston, St George’s, Mayday, and Epsom and St Helier. In this work package we included patients who received either primary UKR or TKR.

Knee Arthroplasty Trial142

Participants in the Knee Arthroplasty Trial (KAT) received primary TKR across 34 centres in the UK between July 1999 and January 2003. The primary outcome was the Oxford Knee Score (OKS) at 12 months after surgery. For this work package we used a wide range of preoperative predictors such as patient characteristics, as well as clinical and surgical factors.

Portsmouth and North Staffordshire143

We used the data on patients undergoing THR from two districts in England, Portsmouth and North Staffordshire, with a population of approximately 1 million. The districts were selected for our work because, first, they were specialised in the assessment and treatment of hip OA; second, there was much support from the local orthopaedic surgeons; and, third, the district had a diverse socioeconomic profile. The orthopaedic surgeons recorded all men and women aged > 45 years who were listed for primary THA between 1993 and 1995. For our work we excluded the patients who had sustained a hip fracture within the past year, with a diagnosis of rheumatoid arthritis (RA) or ankylosing spondylitis, and those with secondary OA.

Clinical Practice Research Datalink77,94

The CPRD, formerly known as the GPRD, is an English NHS observational data and interventional research service. It is jointly funded by the NHS National Institute for Health Research (NIHR) and the Medicines and Healthcare products Regulatory Agency. The database is designed to facilitate the linkage of anonymised patients’ clinical data, thus enabling a number of epidemiological studies which would subsequently be beneficial for improved health-care services. The CPRD has become a gold standard in utilising data by observational researches. Currently over 890 clinical reviews and publications have benefited from the service. More information about the database has been previously detailed in Chapter 2.

A summary of the cohorts and data sets used in work package 2 is provided in Table 13.

Interpretation of patient-reported outcomes for hip and knee replacement surgery155

Having identified a cut-off score for outcome of hip arthroplasty at 12 months, we then used a second cohort, the EOC, to produce a PASS score for the OHS but also to validate the OHS. This purpose-built EOC performs hip and knee replacement surgeries for four acute NHS trusts (Mayday, Kingston, St George’s, and Epsom and St Helier) serving 1.5 million people. The database routinely collected OHS and OKS preoperatively and 6 months after surgery.

We obtained baseline and 6-month postoperative Oxford scores from 1523 patients undergoing hip replacement and 1784 patients undergoing knee replacement. Six months after the surgery, patients were asked to complete their overall satisfaction with surgery using a VAS. On a VAS, 0 depicts no satisfaction and 100 shows a complete satisfaction (very satisfied). We identified a threshold value of ≥ 50. This cut-off point was observed with 93% patients who had hip arthroplasty and with 89% who had knee arthroplasty.

A ROC curve was used to identify PASS score thresholds for absolute changes in OHS and OKS. For OHS this cut-off point was 14 when 97.6% patients declared satisfaction with surgery, and for OKS the cut-off point was 11 when 95.4% said that they were satisfied with surgery.

Table 15 shows these results at the baseline and follow-up at 6 months scores in tabular format.

The mean improvement (the absolute change) in OHS was 19 units (10.5 units) and in OKS was 14.5 units (9.8 units). Of interest, 80 patients undergoing THR had no change or worsening of OHS at 6 months, of whom 56.3% still declared themselves satisfied with surgery. Of the 143 patients whose OKS remained unchanged or deteriorated, 54.6% reported that they were satisfied with surgery. Of those whose pain score improved, 94.9% of hip replacement patients and 92.2% of knee replacement patients were satisfied. Using ROCs, the OHS associated with satisfaction at 6 months was 35 units (95% CI 32.9 to 37.1 units) and, of the patients achieving this score, 98% were satisfied with their surgery compared with 78.6% of those not meeting the threshold. The same figures for OKS were 96.7% and 70.1%, respectively. Figure 11 shows the thresholds according to baseline Oxford scores. Overall, it can be seen that patients starting with a higher baseline score, not surprisingly, have higher thresholds for satisfaction. Patients with worst pain/function baseline scores require higher change in Oxford scores to achieve the highest level of satisfaction, in contrast to patients with the better preoperative scores. However, patients with low preoperative scores (severe symptoms) require lower 6 months’ postoperative Oxford scores to achieve a higher state of satisfaction.

FIGURE 11.

Bar charts illustrating the cut-off points of (a) OHS; and (b) OKS in relation to satisfaction 6 months post surgery. Reproduced with permission and copyright © of the British Editorial Society of Bone and Joint Surgery. 155

A PROMs score on its own does not translate into a clinically meaningful outcome for either clinicians or patients. We aimed to describe how absolute changes in Oxford scores can relate to patients’ satisfaction with surgery. We used PASS156 to identify the cut-off points for the difference between the scores at baseline and 6 months after surgery. PASS depicts the value of OHS/OKS beyond which a patient’s consideration of his or her own health status is good. These cut-off points for the change, as well as 6-month scores, are useful for both patients and clinicians as they improve the understanding of representation of a ‘very good outcome’ as opposed to a ‘good outcome’ and of the expectations of the operation.

Overall, this study demonstrated that thresholds for satisfaction could be identified for use in this research programme. The value obtained for the knee was not dissimilar from the previous study and a new value for the OKS was identified. This was substantially, and significantly, lower than the hip score.

Novel methodological approach for measuring symptomatic change following total joint arthroplasty157

There has been confusion in the orthopaedic literature owing to the different methods used to define patient-reported outcomes. Some papers have used the score as the main outcome with or without adjustments for baseline score. Others have used the changes in score. This has often caused discrepant results in predictors of outcome, most notably when using baseline functional score and pain score as predictors. Harmonising the outcome measure is needed in current research. Both of these functional and pain scores have had limitations. We propose a new score – percentage of potential change (PoPC). PoPC is computed as the actual change divided by the potential improvement multiplied by 100. Thus, PoPC is a measure to express relativity of an actual change in PROMs in relation to a potential change, that is what could have been attained (Figure 12).

FIGURE 12.

Percentage of potential changes. Reprinted from The Journal of Arthroplasty, vol. 29, Kiran A, Hunter DJ, Judge A, Field RE, Javaid MK, Cooper C, Arden NK. A novel methodological approach for measuring symptomatic change following total joint arthroplasty, pp. 2140–5, 2014, with permission from Elsevier. 157

We have used the data from the EOC of patients who underwent THA and TKA between 2004 and 2009. Patients had completed OHS and OKS questionnaires both preoperatively and 6 months postoperatively. For the analysis, 1523 OHS and 1784 OKS completed questionnaires were used. In addition to Oxford scores, patients were also asked to complete a short questionnaire about their satisfaction with surgery on a VAS. A threshold of ≥ 50 units was used to generate a binary variable to identify whether or not patients were satisfied with surgery. Patients with potential improvement have a PoPC of > 0 units, with a potential worsening PoPC of < 0 units and patients with no actual change have a PoPC value of 0 units. PoPC allows the expression of how much a patient’s symptoms have improved or worsened.

Kiran et al. ,157 again, demonstrated excellent improvements following knee and hip replacement surgeries. Correlation (Spearman’s rank-order correlation) of patient satisfaction score with each of the outcome measures was greatest for PoPC and lowest for the relative change (Table 16).

The results showed that the ROC analysis, anchored on satisfaction as the outcome area under the curve (AUC) for hip replacement were follow-up score of 0.86 units, PoPC of 0.86 units, actual change of 0.83 units, relative change of 0.75 units, and for knee replacement were 0.85, 0.85, 0.8 and 0.72 units, respectively.

Kiran et al. 157 have demonstrated the importance of defining outcome measures. Different outcome measures identified different numbers of patients as being responders and non-responders, which has major implications in terms of health economics and health-care planning. It also demonstrates that outcomes for individual subjects differ depending on the outcome measures. This is critically important when trying to define outcome in an individual patient, as in this programme.

In summary, so far we have demonstrated that the OHSs and OKSs can be used to define patient-reported outcome following hip and knee replacement surgery. We have defined cut-off points for the OHSs and OKSs that equate to patient satisfaction with the procedure for use in the following sections to identify predictors of outcome.

Assessing patients for joint replacement: can preoperative Oxford Hip and Knee Scores be used to predict patient satisfaction following joint replacement surgery and to guide patient selection?50

In response to increasing moves to use the Oxford scores to ration access to lower limb joint replacement, we investigated the predictive nature of preoperative OHS and OKS in determining patient satisfaction postoperatively. We used the database from the SWLEOC, in which OHS or OKS were routinely collected with the addition of a satisfaction questionnaire both preoperatively and 6 months postoperatively. A total of 1523 THR patients and 1784 TKR patients were selected. Patients were asked routinely to complete the questionnaires with OHSs and OKSs at baseline, preoperatively and 6 months after their surgery. They were also asked to measure their overall satisfaction with surgery using a VAS, with scores from 0 to 100.

We used scatterplots to identify, and describe, the associations between participants’ preoperative Oxford scores and satisfaction 6 months postoperatively. We found no such association, as shown in the scatterplots (Figure 13).

FIGURE 13.

Scatterplots showing relationship between (a) baseline OKS; and (b) baseline OHS compared with satisfaction at 6 months after surgery. Reproduced with permission and copyright © of the British Editorial Society of Bone and Joint Surgery. 50

Spearman’s rank-correlation coefficients between Oxford scores and satisfaction at 6 months postoperatively were –0.04 units for the OHS (95% CI –0.09 to 0.01 units) and 0.04 for the OKS (95% CI –0.01 to 0.08 units). We also found no differences in median satisfaction scores by baseline OHS (Table 17) or OKS (Table 18) with the Kruskal–Wallis test. Interestingly, TKR results suggested that TKR patients with the lowest preoperative scores were most dissatisfied with their surgery at 6 months after their operation; conversely, scores were not predictive in patients undergoing THR.

This study suggests that, based on 6-month outcomes, preoperative Oxford scores should not be used to predict patient satisfaction after surgery. We conclude that it is unlikely that PROMs can be used on their own to predict patient satisfaction, but it is likely that a combination of multifactorial elements would play an important factor. Such a multidimensional instrument would incorporate a wide range of risk factors in the assessment of pain, function, satisfaction and health-related quality of life (HRQoL).

Introduction

In the UK, within the NHS, PROMs have been routinely used as the outcome measure. PROMs comprise the OHS, the OKS, the EQ-5D and several questions asking patients about their satisfaction with surgery, services and expectations. The OHS was developed in 1996 mainly for use in clinical trials. 146 The OHS and OKS are joint-specific measures,145,158 whereas EQ-5D is non-specific, asking patients about their general health state, mobility, self-care, usual activities, pain and anxiety/depression. 148

As previously described, the OHS and OKR score patients’ satisfaction between 0 and 48 units, with 0 describing the worst possible pain and function, and 48 indicating the best possible outcomes. These scales have been used to identify patients with surgery failure; however, scores alone are not sufficient to inform clinicians and patients of the outcome. For this reason, we introduced a dichotomous variable, the PASS score. 156 The PASS score is a necessary measure to translate the continuous Oxford score variable into a binary variable, that is, to calculate a cut-off point as an indicative score for surgery satisfaction or dissatisfaction. It is a useful variable that provides clinicians and patients with more meaningful information about outcomes of surgery. In addition to PROMs, we have also used the WOMAC, which assesses pain, stiffness and function. 159

Findings

Knee

We investigated the baseline OKS from the EOC database. We included patients undergoing total and UKR surgeries, but excluded those with previous knee surgeries and bilateral operations. Judge et al. 138 investigated the OKS at baseline and 6 months after surgery, as well as the absolute change in OKS. 138 The histogram shows a relatively normal distribution of the baseline OKS (Figure 14a). The OKS distribution after 6 months is skewed to the right, indicating that pain and function improved in the majority of patients (Figure 14b).

FIGURE 14.

Distributions of OKS at (a) baseline; and (b) 6 months after surgery. Reproduced from Judge A, Arden NK, Cooper C, Kassim Javaid M, Carr AJ, Field RE, Dieppe PA. Predictors of outcomes of total knee replacement surgery. Rheumatology, 2012, vol. 51, pp. 1804–13, with permission of Oxford University Press. 141

We established a PASS score using the ROC curve in order to produce the outcome for this study. A score of 30 units was used as a PASS score, which identified 71.7% of the patients (OKS ≥ 30 units) as satisfied at 6 months postoperatively. A higher baseline OKS predicts better outcome, defined by the PASS score, at the 6-month follow-up [odds ratio (OR) 1.52, 95% CI 1.40 to 1.66]. It also predicted a higher follow-up OKS (β = 1.70, 95% CI 1.43 to 1.96).

In another study, Sánchez-Santos et al. 142 used the data from 1967 patients from the KAT across 34 UK centres in the UK. 142 The results showed a baseline OKS mean of 18.2 units (SD 7.5 units) among patients who completed both pre- and postoperative questionnaires. The study found an association between the OKS at baseline and the OKS 12 months after surgery; patients with a better preoperative OKS achieved better pain and functional outcome following TKR (β = 0.35, 95% CI 0.29 to 0.42).

Hip

Our work confirmed that a higher PROMs score was associated with a better outcome. 139,160 The mean baseline OHS was 16.5 units (SD 7.6 units) in patients from the EPOS cohort139 and 16.49 units (SD 7.7 units), 15.67 units (SD 8.61 units), 19.51 units (SD 8.77 units) and 17.52 units (SD 8.30 units) in a meta-analysis combining responders at 12 months of the four cohorts. 160 Baseline SF-36 physical function score, collected from responders of two England health districts, was 20 units (SD 5.35 units).

Primary hip replacement for OA was assessed by using the EPOS prospective cohort. 139 The study showed that better preoperative pain/function was associated with a higher postoperative score. Figure 15a shows the left-skewed histogram of baseline OHS. Figure 15b depicts the 1-year OKS distribution, indicating that the majority of patients exhibited better outcomes at 1–5 years using a repeated measures linear regression model (baseline OHS 10 units; β = 2.68, 95% CI 2.16 to 3.21).

FIGURE 15.

The distribution of OHSs at (a) baseline; and (b) 1 year postoperatively. Reproduced from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

More importantly, this study found that, regardless of the preoperative OHS, participants attained a statistically significant improvement in pain and function after THR. 139 We observed that the patients with the worst baseline scores attained the greatest improvements in pain and function (patients with a preoperative OHS of < 5 achieved a 28.8-point change in the score). However, the patients with the best baseline scores still attained a substantial improvement (patients with a baseline OHS of > 30 achieved a 10.6-point change) (Figure 16).

FIGURE 16.

Change in OHS over time, stratified by baseline score, from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

We have investigated the association of the baseline OHS and BMI. 160 We observed that the baseline OHS decreased as BMI increased; that is, patients of normal weight (BMI 18.5–25 kg/m²) had an OHS of 17.02 units (IQR 15.69–18.34 units), whereas patients with the highest weight (BMI ≥ 40 kg/m²) had an OHS of 12.25 units (IQR 9.02–15.49 units). Underweight patients (< 17.02 kg/m²) also had a low OHS (14.01 units, IQR 9.54–18.48 units). A higher baseline OHS was associated with a better outcome.

Interestingly, a further study investigating patients from two England public health districts (Portsmouth and North Staffordshire districts) demonstrated that patients reporting better baseline physical function were at higher risk of a bad postoperative outcome. 143 This can be explained by the fact that the study used a different measure from the OHS, that is, the SF-36. Functional improvement was classified as change in SF-36 physical function score in the upper quartile. The cut-off point for improved outcome on the SF-36 was defined as ≥ 30 units, which falls in the upper quartile. Patients with better preoperative scores had also improved outcomes, although they achieved a lesser change between pre- and postoperative scores. Finally, we observed that the increased baseline OHS was associated with a greater OHS. 138 This study emphasises the importance of classifying outcome similarly across all studies.

Conclusion

Preoperative pain/function was one of the strongest predictors of outcome. A high baseline preoperative score was related to a better postoperative score. It has been established that patients with lower preoperative pain and better preoperative function attain the highest postoperative pain and function, whereas patients with the worst baseline scores achieve the highest change between baseline and follow-up. 32,44–47 We also need to consider floor and ceiling effects when we estimate change because the level of satisfaction attained differs according to the baseline degree of functional disability. It is also known that patients with the lower preoperative scores never obtain the original functional and pain levels, although this outcome is possible in patients with better preoperative scores. Table 19 summarises our findings.

Introduction

Previous research on the association of age with the outcome of TJR has been heterogeneous. Some authors have found an association between age and outcome, but others have found no such evidence. 44–49,51,162 Literature reviews suggest that age is not a strong predictor of outcome. 43

Findings

Knee

Data from the EOC database showed no association with age overall but a minor effect on OKS function, with younger patients having a better outcome. 141 However, despite the statistical significance, the effect size observed was small. We further explored this association using the KAT data and found that patients aged < 60 years and ≥ 80 years presented a worse pain and functional status at 12 months after knee surgery. 142 We observed that younger women (aged < 60 years) had a better outcome than men, whereas in the oldest age group (aged ≥ 80 years) women had a worse outcome than men.

Age at TJR was also a significant predictor of revision for knee. Importantly, Culliford et al. 97 found that the patients undergoing TKR had a 4.3% reduction in revision rates for each extra year of age. 97 Figure 17 illustrates that patients aged between 55 and 65 years had up to 10% revision rates at 15 years after primary knee replacement; patients aged > 85 years had a < 2% revision rate; interestingly, the youngest group of patients had the highest revision rate.

FIGURE 17.

Cumulative incidence estimate for revision of TKR by age. Reproduced from Culliford et al. 97 This is an Open Access article distributed in accordance with the Creative Commons Attribution-Noncommercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial.

Hip

Judge et al. 30 investigated the association between age and patient outcomes following THR surgery in the EUROHIP cohort. Age was grouped at baseline as < 50, 50–69 or ≥ 70 years. The results showed that there were no statistically significant differences across the age groups, although the youngest responders presented better outcomes. 30 In another study, Judge et al. 143 collected data from 282 patients from two England health districts: Portsmouth and North Staffordshire. The primary outcome was the long-term functional improvement after THR. Patients included in the study were aged ≥ 45 years and were listed for THR for primary OA. To identify the risk factors to predict functional improvement in the long term (≈8 years), we used the logistic regression modelling. The results showed that older patients were less likely to have an improvement in physical function; however, the association found in the study was weak. 143 Furthermore, data analysis from EPOS and EUROHIP cohorts revealed a non-linear association between age and outcome in patients undergoing THR. 138 In those patients aged ≥ 75 years, increasing age was associated with worse outcomes. In addition, a worse outcome was found in those patients aged 50–60 years than in those who are younger (aged < 50 years) and older (aged > 60 years),139 although there was a small, statistically significant change in achieved postoperative outcome associated with patient age (Figure 18).

FIGURE 18.

The change in OHS over time, layered by age, from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution-Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and the use is non-commercial and is otherwise in compliance with the licence.

Among patients aged > 65 years, revision rates 15 years after the primary hip replacement were up to 10%97 (Figure 19). Among THR patients, the revision rate fell by 3% for each extra year of age, after adjusting for the competing risk of death.

FIGURE 19.

Cumulative incidence estimate for revision of THR by age. Reproduced from Culliford et al. 97 This is an Open Access article distributed in accordance with the Creative Commons Attribution-Noncommercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial.

Conclusion

We have demonstrated that increasing age reduces the risk of revision surgery. 97 Different, and heterogeneous, results were seen when using PROMs as the primary outcome. Some of our work showed that the effect of age on joint replacement surgery outcome is not significant,30,141,160 whereas other work showed the effect of age to be non-linear, with the youngest and oldest patients having the worst outcomes. 138,139,142 Age was associated with functional outcome following TKR and THR, although size effect of the association was small.

Overall, the small statistically significant differences relating to patient age at the time of surgery were greatly outweighed by the substantial change in PROMs achieved by these patients. 139 Table 20 summarises associations found between age and outcomes.

Introduction

The work carried out in work package 1 showed that the lifetime risk of THR in the UK is higher among women than among men. 98,99 It was observed that, at the age of 50 years in 2005, the risk of THR was 11.6% (95% CI 11.1% to 12.1%) for women and 7.1% (95% CI 6.7% to 7.5%) for men. Similarly, the risk of TKR was also greater among women (10.8%, 95% CI 10.3% to 11.3%) than among men (8.1%, 95% CI 7.6% to 8.5%). 99 In work package 2 we aimed to confirm sex, as previously described in the literature, as a predictor of surgery outcome following THR and TKR.

Findings

Conclusion

The results from the knee and hip studies show that sex may have small effects in knee arthroplasty patients with little effect on hip arthroplasty when using PROMs as the outcome. In general, women had a worse PROMs outcome (of minor clinical significance),141,142 whereas men were at a higher risk for prosthesis failure resulting in the revision arthroplasties. 97 Our results are summarised in Table 21.

Introduction

The World Health Organization recommends that BMI is classified into four categories: underweight (< 18.5 kg/m²), normal (between 18 and 25 kg/m²), overweight (> 25 to 30 kg/m²) and obese (class I, > 30 to 35 kg/m²; class II, > 35 to 40 kg/m²; and class III, > 40 kg/m²). BMI is widely recognised as an important predictor for many conditions, including OA, and, as such, a number of patients referred for lower limb arthroplasty will have a raised BMI. Although some studies identified a positive relationship between increased BMI and susceptibility to knee and hip OA, and the need for replacement surgeries,163–165 there is increasing concern that obesity could be seen as an obstacle to accessing replacement surgeries. We aimed to confirm an association between BMI and patient-reported outcomes of THR surgeries. A number of detailed studies investigating the BMI effects on surgery outcomes have been carried out. This section will detail the results from several publications.

Knee

We used the GPRD data to describe the association of BMI with revision rates. 97 The data were collected from patients who had undergone hip and knee replacement surgeries between 1998 and 2011. From this cohort we then identified those with subsequent revision surgeries. We investigated the effects of BMI on the time of revision surgery. We estimated cumulative incidences of TKR revisions at 1, 5, 10 and 15 years. The results showed that at 5 years the estimated cumulative revision rate for TKR was 1.9% (95% CI 1.8% to 2.1%). The cumulative incidences across all BMI groups are shown in Figure 20. For each increased unit of BMI the estimated risk of TKR revision increased by 1.015 (95% CI 1.002 to 1.028).

FIGURE 20.

Estimated cumulative incidence for TKRs by BMI. Reproduced from Culliford et al. 97 This is an Open Access article distributed in accordance with the Creative Commons Attribution-Noncommercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial.

Our results suggested that BMI appears to be a significant risk factor of time to revision of TKR. The risk of revision for morbidly obese TKR patients was found as high as 6% after 10 years following surgery. Up to approximately 7 years there was a more even distribution across all BMI categories; however, there was a much higher risk for the morbidly obese patients between 7 and 10 years.

In Judge et al. ,141 a high BMI was related to a worse outcome at 6 months after TKR (coefficient total OKS –0.44 units, 95% CI –0.86 to –0.01 units), although BMI was not found to be a clinically important predictor of outcome taking into account the PASS score binary variable (OR total OKS 0.90 units, 95% CI 0.80 to 1.01 units) (Table 22). Patients who had a higher BMI showed worse functional outcomes (coefficient function OKS –0.33 units, 95% CI –0.57 to –0.09 units). Cohort analysis by Sánchez-Santos et al. 142 showed that a high BMI was a determinant factor associated with a decreased OKS at 12 months’ follow-up. Although statistically significant, the effect sizes are small and below the minimal clinically important difference of the OKS.

TABLE 22 - Associations found between BMI and postoperative knee score

DVT, deep-vein thrombosis; NS, not significant; PE, pulmonary embolism; UTI, urinary tract infection.

Cohort	Hip/knee	PROMs/revision	Outcome	Association found	For example	Reference
EOC	Knee	PROMs	6-month PASS score	NS	Adjusted ORtotal OKS: 0.90 (95% CI 0.80 to 1.01)	Judge et al.141
EOC	Knee	PROMs	6-month OKS	High BMI, worse outcome	Linear model coefficienttotal OKS: –0.44 (95% CI –0.86 to –0.01)	Judge et al.141
KAT	Knee	PROMs	12-month OKS	High BMI, worse outcome	Linear model coefficientBMI (10 units): –1.5 (95% CI –2.4 to –0.6)	Sánchez-Santos et al.142
GPRD	Knee	Revision	15 years	Increasing BMI increases risk (small)	Adjusted ORTKR: 1.015 (95% CI 1.002 to 1.028)	Culliford et al.97
CPRD	Knee	PROMs	DVT/PE 6 months after surgery	Increasing BMI increases risk	Adjusted ORTKR (obese): 1.59 (95% CI 1.26 to 1.99)	Wallace et al.77
CPRD	Knee	PROMs	Anaemia 6 months after surgery	Obesity decreases TKR risk	Adjusted ORTKR (obese): 0.74 (95% CI 0.58 to 0.94)	Wallace et al.77
CPRD	Knee	PROMs	A wound infection 6 months after surgery	Increasing BMI increases risk	Adjusted ORTKR (obese): 1.23 (95% CI 1.01 to 1.50)	Wallace et al.77
CPRD	Knee	PROMs	A UTI 6 months after surgery	TKR NS	Adjusted ORTKR (obese): 0.93 (95% CI 0.74 to 1.17)	Wallace et al.77
CPRD	Knee	PROMs	Death 6 months after surgery	Underweight increases risk	Adjusted ORTKR (underweight): 4.61 (95% CI 1.64 to 13.01)	Wallace et al.77

Wallace et al. 77 further investigated the association between BMI and the risks of complications 6 months following TKR. We used the CPRD to collect baseline BMI measurements, as recorded by GP practices, on patients who had undergone primary TKR between 1995 and 2011. We selected 32,485 TKR patients (including those who died within 6 months of their surgery) and, of those, < 1% were underweight, 17% were of normal weight, 38% were overweight, 29% were obese class I, 12% were obese class II and 4% were obese class III. The following outcomes were recorded following their surgeries: myocardial infarction, stroke, deep-vein thrombosis (DVT) or pulmonary embolism (PE), respiratory infection, anaemia, urinary tract infection, wound infection and death.

The study analysis showed that a higher BMI was associated with a significantly higher risk of developing wound infections, up from 3% to 4.1% (adjusted p < 0.05), in TKR patients. An association was also found between increased BMI and DVT/PE risk, up from 2.0% to 3.3% (adjusted p < 0.01), in TKR patients. Interestingly, no association was found between BMI and other confounders, particularly myocardial infarction, stroke and mortality.

Hip

We have published several studies investigating associations between BMI and THR outcomes. 97,160,166 Table 23 shows these associations. The results of the study by Culliford et al. 97 show that at 5 years the estimated cumulative rate of revision surgery for THR was 2% (95% CI 1.8% to 2.1%). Figure 21 depicts cumulative incidences across all BMI groups for patients undergoing THR. BMI was a significant predictor of revision. Severely obese THR patients seem to have an increased risk of revision surgery in the first year. For each additional unit of BMI, the estimated risk of THR revision rises by 1.02 units (95% CI 1.009 to 1.032 units).

FIGURE 21.

Figure showing estimated cumulative incidence for THRs by BMI. Reproduced from Culliford et al. 97 This is an Open Access article distributed in accordance with the Creative Commons Attribution-Noncommercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial.

Batra et al. 166 used the analysis from four prospective cohort studies of patients who had undergone THR for OA: EPOS, SWLEOC, St Helier and EUROHIP. We determined the relationship between BMI and OHS at 1 year following THR. This was adjusted for the baseline OHS. We used a meta-analysis to combine the results from separately built models in all four cohorts. All models were adjusted for common variables such as sex and age. The analysis showed that, for every 5-unit rise in BMI, the 1-year OHS fell by 0.81 units (95% CI 0.55 to 1.08 units)166 (Figure 22).

FIGURE 22.

Fixed-effects meta-analysis from Batra et al. 166 Osteoarthritis and Cartilage, vol. 20, Batra RN, Judge A, Javaid MK, Thomas GE, Beard D, Murray D, et al. Pre-operative BMI as a predictor of patient reported outcomes of primary hip replacement surgery: a combined analysis of 4 prospective cohort study, pp. S152–3, 2012, with permission from Elsevier.

We then combined the data from all cohorts and used multiple imputations for missing data. With this analysis, when the sex and age variables were adjusted, every 5-unit rise in BMI was associated with a 1-year fall in OHS of 0.72 units (95% CI 0.46 to 0.99 units); when adjusting for all potential confounders, this reduction was decreased to 0.51 units (95% CI 0.09 to 0.92 units). The results suggest that, for each 5-unit increase in BMI, the difference in 1-year OHS becomes more significant.

Following these preliminary results, Judge et al. 160 further expanded the investigation and used the same cohorts, exposure, primary outcome OHS and confounding variables to reinvestigate whether or not BMI is a clinically significant predictor of patient-reported outcomes in patients with THR. Tables 24 and 25 show that patients achieved significant improvement in their OHS, regardless of their baseline BMI value.

The results confirmed the following associations between BMI and OHS: for each 5-kg/m² increase in BMI, the OHS at 1 year decreased by 0.78 units (95% CI 0.27 to 1.28 units; p < 0.001). Obese class II patients would have a 1-year OHS 2.34 units lower than that of people with a normal BMI.

Our results confirmed that the effect of BMI on 1-year postoperative OHS is statistically significant, although the degree of significance is low: patients who are obese class II would have an OHS 2.34 units lower than patients with a normal BMI. In addition, patients who are rated as obese class II achieved a 22.2-unit change in their OHS following THR. As suggested by Murray et al. ,145 the smallest change in OHS that can be regarded as clinically important is approximately 5 units. Therefore, although a difference of 2.34 units is statistically significant, a difference of this magnitude in OHS at 1 year across all categories of BMI will have clinical significance only in patients who are rated as obese class II or III. This effect is greatly outweighed by a significant change in OHS in obese class II people (22.2-unit change), indicating substantial improvement in outcomes in this patients over the year. Thus, we conclude that BMI should not be indicative to deny patients access to hip replacement surgery.

Judge et al. 138 analysed the data from the EPOS cohort preoperatively and every year up to 5 years after the patients’ THR surgery. The analysis showed that a 10-kg/m² increase in BMI was associated with decrease in OHS of 1.54 units (95% CI 0.64 to 2.45 units) averaged between the 1- and 5-year follow-ups, although these differences were small compared with the overall benefit of the operation (Figure 23). 139 Similarly, patients from the EUROHIP cohort presented similar results at 1 year after their THR [OR return to normal (obese) 0.8 units, 95% CI 0.5 to 1.3 units]. 30 In contrast, there was no association found between BMI and functional outcomes in THA patients from Portsmouth and North Staffordshire districts. 143

FIGURE 23.

Change in OHS over time, stratified by BMI from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution-Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

Wallace et al. 77 investigated the association between BMI and the risks of complications 6 months after THR surgery using the CPRD. 77 We collected the baseline BMI measurements from the CPRD for THR patients between 1995 and 2011. For THR we selected 31,817 patients. From this cohort, BMI distribution was as follows: 1.5% underweight, 28% normal weight, 40% overweight, 21% obese class I, 7% obese class II and 2% obese class III. The results showed that in THR patients increased BMI was associated with a significantly higher risk of wound infections, ranging from 1.6% to 3.5% (adjusted p < 0.01). The association between increased BMI and DVT/PE risk was significant, with increased BMI from 2.2% to 3.3% (adjusted p < 0.01) in THR patients.

Conclusion

We found weak associations between increasing BMI and worse PROMs outcome. In accordance with other studies,167 however, the effect sizes were small and often below the minimal clinically important difference for the Oxford scores. BMI was found not to be a strong predictor of functional outcomes. Therefore, high preoperative BMI should not be a deterrent to knee or hip replacement surgeries. 26,43

Obese patients have a high risk of developing of DVT, PE, wound infection and urinary tract infection following knee and hip replacement surgeries.

We cannot advocate selecting patients for joint replacement surgeries without consideration of BMI, but we do suggest that denial of surgery based on high BMI is unwarranted.

Introduction

Historically, deprivation was measured using the Index of Multiple Deprivation (IMD) 2004. 53 This index was extracted from the residence area where the patients lived at the time of surgery. The index is a compound of seven deprivation indices, employing the indicated weightings: income (22.5%), employment (22.5%), health deprivation and disability (13.5%), education, skills and training (13.5%), barriers to housing and services (9.3%), crime (9.3%) and living environment (9.3%). Poorer areas attract a higher deprivation score and more prosperous areas have a lower score. In addition, educational level was considered in some of our studies as a proxy of socioeconomic,30,161 as well as employment, status. 30 Table 26 shows postoperative scores in relation to deprivation indices.

Findings

Conclusion

Higher levels of deprivation were associated with worse patient outcomes following TKR. A higher attained educational level was associated with better postoperative reported outcomes following THR.

Introduction

The most frequent indication for THR and TKR in the UK is OA, which is the most common type of arthritis in developed countries and for which TJR is the only effective therapy in severe cases. 2 The total number of hip procedures reported in the UK in 2013 was 89,945. Of these, 80,194 were primary procedures and 9751 were revision replacements. 168 The indication for primary hip replacement was OA in 91% of cases. The total number of knee procedures reported in the UK in 2013 was 91,703. Of these, 85,920 were primary procedures and 5783 were revision replacements. The indication for primary knee replacement was OA in 97% of cases.

Another important disease for THR and TKR indication is RA. RA is a chronic autoimmune disease affecting joints. Severe RA also requires surgical intervention. RA indication for THR and TKR in 2013 was 1%. 168 Since the licensing of biological agents for the treatment of RA, the number of arthroplasties in patients with RA is declining. Association between the indication for lower limb joint replacement (OA or RA) and postoperative score is shown in Table 27.

Findings

Conclusion

Most of the patients undergoing lower limb joint replacements have been diagnosed with OA. In only one study did patients diagnosed with RA have better outcomes than those diagnosed with OA. 141 We adjusted the model by the confounding factor age, as the patient population diagnosed with RA is younger than that diagnosed with OA141,169 Similar results were also observed for OKS pain score. The other studies were related to OA patterns. These studies unveiled better TKR outcomes in patients with a superolateral pattern of OA or no reduction in joint space, any or a low number of other joints affected by OA, as well as any or a low number of previous interventions on other joints. In summary, fewer baseline OA complications were related to better outcomes.

Introduction

The American Society of Anesthesiologists (ASA) status classification system is a standard measure of fitness for surgery, scored from 1 (normal, healthy) to 6 (brain-dead patients whose organs are being removed for donor purposes). Our studies included patients up to ASA 4 (severe systemic disease that is a constant threat to life).

In addition, we considered some coexisting diseases that could affect surgery outcomes. We take into account DVT and PE, urinary tract infection, other musculoskeletal diseases and neurological, respiratory, cardiovascular, renal or hepatic disease. ASA status and number of comorbidities were analysed in several cohorts, listed in Table 28. Some of these coexistent diseases were also related to BMI (see Tables 18 and 20).

Findings

Hip

In the EUROHIP cohort, there was no statically significant association between ASA grade and change in WOMAC score at 12 months: the median OHS for ASA 1 patients was 40.6 (95% CI 35.2 to 46.0), for ASA 2 patients was 35.4 (95% CI 32.8 to 38.1), for ASA 3 patients was 33.3 (95% CI 27.8 to 38.9) and for ASA 4 patients was 38.5 (95% CI 30.9 to 46.2) (p = 0.13). 30 We did not find differences between preoperative ASA grades and the OHS 12 months after hip surgery. 160

In EPOS, the number of coexistent diseases was statistically significantly associated with outcome, that is, for each additional preoperative disease the OHS between 1 year and 5 years was reduced by 0.90 units (95% CI 0.54 to 1.27 units)139 (Figure 24).

FIGURE 24.

Change in OHS over time, stratified by number of coexisting diseases, from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

Conclusion

We can conclude that ASA grade is not an important predictor of outcomes, although healthier preoperative conditions could relate to better outcome. On the other hand, the number of other coexisting diseases is a predictor of worse outcome, although the effect size was small and it would be necessary to identify the weight for each illness in the final outcome.

Introduction

We employed several scoring systems to assess mental health. For instance, patients from EUROHIP, the EOC, St Helier and EPOS cohorts completed a SF-36 questionnaire. This tool measures quality of life through eight health concepts, from which we selected mental health (psychological distress and psychological well-being). 170,171 In the KAT cohort, quality of life was assessed using a subset of those in the SF-36 (SF-12). Both questionnaires score from 0 (the worst possible health) to 100 (best possible health). In addition, we estimated anxiety and depression among EOC participants using the EQ-5D. 148

Findings

Hip

In the EPOS cohort, lower preoperative SF-36 mental health scores were associated with reduced postoperative improvement in OHS between 1 and 5 years (Table 29). 139 The differences in achieved postoperative OHS among categorised mental health levels were not substantial, although they were statistically significant when we followed up between 1 and 5 years (Figure 25). We observed the same results when we followed up patients at 12 months and we used the same questionnaire (coefficient 0.76 units, 95% CI 0.18 to 1.33 units). 160 Finally, we obtained a similar increase in the OHS outcome (coefficient 0.59 units, 95% CI 0.24 to 0.94 units) in patients from the EPOS and EUROHIP cohorts followed up at 12 months. 161

TABLE 29 - Associations found between mental health and postoperative score

NS, not significant.

Cohort	Hip/knee	PROMs/revision	Outcome	Association found	For example	Reference
EOC	Knee	PROMs	6-month PASS score	NS. Anxiety/depression worse outcome	Adjusted ORtotal OKS (extremely anxious/depressed): 0.70 (95% CI 0.42 to 1.18)	Judge et al.141
EOC	Knee	PROMs	6-month OKS	Anxiety/depression worse outcome	Linear model coefficienttotal OKS (extremely anxious/depressed): –2.21 (95% CI –4.34 to –0.09)	Judge et al.141
KAT	Knee	PROMs	12-month OKS	Worse mental health score associated with poor outcome	Linear model coefficientSF-12 (10 units): 0.9 (95% CI 0.4 to 1.3)	Sánchez-Santos et al.142
EPOS	Hip	PROMs	1- to 5-year OHS	Lower SF-36 mental health score associated with poor outcome	Δ linear model coefficientSF-36 (10 units): 0.76 (95% CI 0.46 to 1.07)	Judge et al.139
EPOS, EUROHIP, EOC and St Helier	Hip	PROMs	12-month OHS	Lower SF-36 mental health score associated with poor outcome	Linear model coefficientSF-36 (10 units): 0.76 (95% CI 0.18 to 1.33)	Judge et al.160
EPOS and EUROHIP	Hip	PROMs	12-month OHS and WOMAC score	Lower SF-36 mental health score associated with poor outcome	Linear model coefficientSF-36 (10 units): 0.59 (95% CI 0.24 to 0.94)	Judge et al.161

FIGURE 25.

Change in OHS over time, stratified by SF-36 mental health categories from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

Conclusion

Our results show an association between preoperative mental health and the PROM. Therefore, patients having worse preoperative mental health were more likely to have worse postoperative outcome scores. This is concordant with other studies in the literature which used detailed measures of mental health28,29,172,173 or more specific ones, for example the Beck Depression Inventory. 174 However, there were no clinical important differences among mental health categories and the outcome found after surgery.

Introduction

Surgical variables are not usually included in models assessing lower limb joint replacement. We considered clinical examination findings, that is, the presence of fixed flexion deformity, joint deformity (valgus, varus, no deformity) and presence of anterior cruciate ligaments (ACLs) and posterior cruciate ligaments. We also considered variables extrinsic to the patient, that is grade of operating surgeon (consultant, associate specialist staff, registrar and senior house officer) and the grade of senior surgeon present at the operation (consultant, associate specialist staff and registrar). 142 Furthermore, information about the surgical approach (anterolateral and posterior) and patient position (supine and lateral) was also collected. 139

Findings

Hip

In EUROHIP, we observed increasing differences between the 12-month follow-up and baseline scores from K/L grade 2 to K/L grade 4, that is, the median WOMAC score for K/L grade 2 was 29.7 units (95% CI 22.6 to 36.8 units), K/L grade 3 was 34.4 units (95% CI 31.3 to 37.4 units) and K/L grade 4 was 38.5 units (95% CI 35.4 to 41.7 units); differences among groups were statistically significant (p = 0.03). Although this effect seems to increase also from grade 2 to grade 0, the absolute numbers of patients in groups 0 and 1 (six and five patients, respectively) did not let us affirm that K/L grade was a U-shaped curve, with group 2 being the worst. We also analysed K/L grade in patients from four cohorts, being K/L grades 1 and 2 versus K/L grade 4, and found this not to be statistically significant, that is 0.43 units (95% CI –2.75 to 3.62 units) and 0.86 units (95% CI –4.73 to 6.44 units), respectively (Table 30). We observed better outcomes in hip replacement associated with a larger offset size (offset of ≥ 44 mm).

TABLE 30 - Associations found between surgical and drug variables and postoperative score

HRT, hormone replacement therapy; NS, not significant.

Cohort	Hip/knee	PROMs/revision	Outcome	Association found	For example	Reference
KAT	Knee	PROMs	12-month OKS	Fixed flexion deformity, varus deformity, absent preoperative ACL were associated with better outcome	Linear model coefficientfixed flexion deformity: 1.5 (95% CI 0.6 to 2.4)	Sánchez-Santos et al.142
					Linear model coefficientno valgus/varus deformity: –1.6 (95% CI –3.0 to –0.3)
					Linear model coefficientpreoperative ACL absent: 1.5 (95% CI 0.0 to 3.0)
EUROHIP	Hip	PROMs	12-month WOMAC score	K/L grade change increases from 2 to 4	K/L grade 2 worse outcome; p = 0.03	Judge et al.30
EPOS, EUROHIP, EOC and St Helier	Hip	PROMs	12-month OHS	NS K/L grade	Linear model coefficient(K/L 1): 0.43 (95% CI to 2.75 to 3.62)	Judge et al.160
					Linear model coefficient(K/L 2): 0.86 (95% CI –4.73 to 6.44)
EPOS	Hip	PROMs	1- to 5-year OHS	Femoral offset size larger = better outcome	Δ linear model coefficient(offset): 0.17 (95% CI 0.06 to 0.28)	Judge et al.139
EPOS and EUROHIP	Hip	PROMs	12-month OHS and WOMAC score	Femoral offset size larger = better outcome	Linear model coefficient(offset): 0.18 (95% CI 0.03 to 0.32)	Judge et al.161
EPOS	Hip	PROMs	12-month OHS	Posterior approach = better outcome	Linear model coefficient: 2.2 (95% CI 1.1 to 3.3)	Judge et al.139
EPOS and EUROHIP	Hip	PROMs	12-month OHS and WOMAC score	Posterior approach = better outcome	Linear model coefficient: 2.42 (95% CI 0.44 to 4.39)	Judge et al.161
GPRD	Hip and knee	Revision	10-year implant survival	HRT reduces revision rates	HRHRT ≥ 6 months: 0.62 (95% CI 0.41 to 0.94)	Prieto-Alhambra et al.175

Change among femoral offset size categories had a small statistically significant difference in the postoperative OHS achieved (Figure 26). We found a significant interaction between offset size and sex, with the effect limited to women. 139

FIGURE 26.

Change in OHS over time, layered by femoral component offset (mm offset) categories, from Judge et al. 139 This is an open-access article distributed under the terms of the Creative Commons Attribution Noncommercial license, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the licence.

We also identified the effect of surgical approach as significant, with the posterior approach having better outcomes than anterolateral; that is, there is a difference in the OHS at 12 months of 2.2 units (95% CI 1.1 to 3.30 units). 139 This result was very similar to those presented by Judge et al. ;161 that is, the difference at 12 months for the OHS was 2.42 units (95% CI 0.44 to 4.39 units).

Finally, we found that patients from GPRD receiving hormone replacement therapy had a lower risk of joint (hip and knee replacements) revision surgery after 6 months (adjusted HR 0.62, 95% CI 0.41 to 0.94) and after 12 months (adjusted HR 0.48, 95% CI 0.29 to 0.78)175 (Figure 27 and see Table 30).

FIGURE 27.

Kaplan–Meier estimates of probability of revision surgery according to hormone replacement therapy use. Reproduced from Annals of the Rheumatic Diseases, Hormone replacement therapy and mid-term implant survival following knee or hip arthroplasty for osteoarthritis: a population-based cohort study, Prieto-Alhambra D, Javaid MK, Judge A, Maskell J, Cooper C, Arden NK, COASt Study Group. vol. 74, pp. 557–63, 2015, with permission from BMJ Publishing Group Ltd. 175 HRT, hormone replacement therapy.

Conclusion

Clinical factors have demonstrated their importance in TKR as predictors of OKS outcomes. 142 Among them, we identified fixed flexion deformity, preoperative valgus/varus deformity and preoperative damaged ACL. We unveiled an association between femoral offset size and THR, with femoral offsets > 44 mm related to better outcomes,139,161 although this finding may happen only in women. 139 In addition, we discovered that a posterior approach may have better outcomes than an anterolateral approach. 139,161 Furthermore, K/L status was not a good predictor of the outcome. 30,160

Patient-reported outcomes following primary hip replacement surgery: development and internal validation of a prognostic tool138

Following identification and confirmation of risk factors within this work package we aimed to develop a clinical risk prediction tool for both hip and knee arthroplasty patients. Such tools will be beneficial for clinical decision-making and patients’ expectation of their surgery.

The recent literature describes combining of data from multiple risk factors and, thus, including the broad range of predictors in a prognostic model. 80,81 This includes patient-related risk factors such as age, sex, education, obesity and mental health status; clinical predictors such as preoperative level of pain and function, indication for surgery, coexisting conditions and radiographic (grade) variables; and surgery-related risk factors, such as femoral component offset. In this work we aimed to develop a similar prognostic model allowing inclusion of a number of risk factors to predict pain and function following hip replacement surgery.

We collected data from prospective cohorts of patients receiving primary THR for OA: EUROHIP and EPOS. From EUROHIP we used the data from 845 patients and from EPOS we used the data from 1247 patients. By combining the data from these large cohorts we took into account a comprehensive range of both already-investigated risk factors and novel predictors.

As OHS was one of the predictors included for the analysis, we collected OHS questionnaires preoperatively and 1 year postoperatively. OHS was used as a primary outcome measure. In the EUROHIP cohort, outcome measures were predominantly collected via WOMAC, in contrast to EPOS, which collected OHS. To derive OHSs from WOMAC we used the truncated regression model for 110 patients in whom both OHS and WOMAC data are complete. Using this model we could predict WOMAC score at baseline (R² = 75.5%) and at 1 year (R² = 63.4%).

Predictor variables we included in this work are represented in Table 31.

In the EPOS study, patients completed the SF-36. The SF-36 is an instrument that measures quality of life in eight domains. One of these domains is the mental health component, which we have selected for this work. In SF-36 the lowest score, 0, is indicative of the worst possible health and the highest score, 100, of the best possible health. We obtained the fixed flexion range of motion variable in degrees from the Charnley Modification of D’Aubigné–Postel Grade questionnaire. 176 From comorbidities we collected information on coexistent diseases, such as DVT, PE, urinary tract infection, other musculoskeletal disorders, neurological, renal, cardiovascular, respiratory or hepatic disease or treatment for other medical conditions. We have also collected detailed intraoperative information: grade of operating surgeon and patient position. From the implant information we obtained and used data such as implant material (stainless steel or ceramic), femoral head size in millimetres and the femoral component size (offset).

From the EUROHIP cohort we obtained data on home circumstances, employment, education, duration of pain in the affected hip, the number of preoperative expectations of surgery162 and other joints affected by OA, as well as other surgeries in other joints. Medications (prior to surgery) that were considered relevant to the study included analgesic/non-steroidal anti-inflammatory drugs (NSAIDs), bisphosphonates, medications for heart disease, anticoagulants, antidepressants, bronchodilators and antidiabetic drugs. We obtained EQ-5D scores for the patient’s health state today, mobility, self-care, usual activities and pain and anxiety. We obtained the data collected from baseline radiographs of the pelvis taken in the anteroposterior view. Radiographic variables included the standard K/L grade (0–4)162,176–179 and the intra-articular pattern of disease distribution (superolateral, superomedial, medial or concentric). Osteophyte size in the superior–femoral, superior–acetabular, inferior–femoral and inferior–acetabular regions was recorded. For osteophytes with moderate sizes, we created an ordinal variable of the number of sites. From intraoperative data we also obtained records from surgical teams on patients’ height and weight, prosthesis type and ASA status, the last being a standard measure of fitness for surgery, scored from 1 (normal, healthy) to 4 (severe systemic disease that is a constant threat to life).

Methodology for combining variables

Because cohorts contained a different set of confounders, combining the two studies resulted in a high proportion of missing data. Appendix 3 shows the variables available in all cohorts.

Previously methodology has been developed to combine data from multiple sources. 80,81,178,179 For instance, Heymans et al. 178 described the method multivariate imputation by chained equations (MICE), based on a Bayesian approach,176 in which data from three controlled trials were combined to identify risk factors of chronic low back pain. Jackson et al. 80 used the multiple imputation methods within Bayesian graphical modelling and described the associations between low birthweight and air pollution, thus incorporating adjustments for crucial confounders that were not available within the individual data set. These two approaches have been further compared in both simulation180 and cases studies. 181 It was concluded that, even when the proportion of missing data is high, the performance of both approaches is similar. Moreover, they both correct most of the bias with a non-hierarchical (simple) data structure.

In our work we used MICE to combine the data sources as we were more familiar with this methodology, it is easier to implement in standard statistical software and it takes less time to run the models.

Use of the multiple imputation method allows the assumption that the data are missing at random. This is plausible, as the reason behind the missing data is that not all covariates were collected in each study. 178 At first, we created 25 copies of the data set. In each copy missing values were replaced by imputed values,176,182,183 with 20 cycles of regression switching. Imputations were made by drawing from the posterior predictive distribution of each variable that required imputation. We included all of the covariates together with the outcome variable (12-month postoperative OHS) in the imputation model, as this carries information about the missing values of predictors. 176 Prior to imputation we transformed continuous variables so that they were approximately normally distributed. For the imputation of continuous variables we used the linear regression method, whereas for categorical variables we used logistic, ordinal and multinomial regression. At the second stage we fitted a statistical model to each of the imputed data sets separately. The results were then averaged to obtain a single estimate of the association. We calculated standard errors using Rubin’s rule, as it accounts for the variability between results of imputed data sets and reflects the uncertainty associated with imputing missing data. 176

Although the methodology we used has been tested and used before and proved to be transparent, and it is relatively easy to implement within the standard statistical software, its use required us to make the assumption that data were missing at random. This was plausible in the context of this study because the reason for the missing data was that the variables were not collected. This was further supported by inclusion of a wide range of covariates to ensure that sufficient predictors were included to recover missing data for missing information. We are aware, however, that there were too many missed data in this study. Even if the missing at random assumption is valid, with increased proportion of missing variables the reliability of the regression coefficient may diminish. 182 However, we are reassured by the previous simulations and case studies in which the methods have been shown to be effective, even with extreme numbers (> 90%) of missing data. 180,181

Statistical methods used

For statistical analysis we used Stata version 12.1 (StataCorp LP, College Station, TX, USA). We used analysis of covariance (ANCOVA) to identify predictors of the 12-month follow-up OHS, adjusting for preoperative OHS. In order to model non-linear relationships for continuous variables we used linear splines. We use MICE in order to combine data from two studies and adjust for a wider range of variables. 80,81 In the imputation model we included all of the listed covariates together with the outcome variable (12-month postoperative OHS). Before imputation we log-transformed continuous variables so that they were approximately normally distributed. We created 25 imputed data sets for missing data by using MICE in Stata. The final regression model included all predictor variables and was fitted to each imputed data set. This was then averaged for overall estimated associations. We applied the automatic backward selection per 200 bootstrap samples of imputed data sets. The variables retained were those consistently selected for at least 70% of the time.

We assessed the performance of the tool by using the calibration and discrimination methods. 184,185 By calibration we could judge how close the predicted OHS was to the observed OHS for each tenth of predicted score in 10 equally sized groups. By discrimination we assessed the measure of variation and for this we used R². In total, 200 bootstrap samples with replacement and combined with multiple imputations were used to estimate overoptimism in the predictive ability of the model, and obtain bias-corrected estimates of R². We compared R² from models developed in 200 bootstrap samples with R² in the same models applied to the original sample. As previously described in this chapter, a change in OHS by fewer than 5 units is described as clinically not significant. We used this knowledge to finally test the ability of the predictive tool to identify patients with the worst possible scores after their THR (i.e. identify the patients with < 5-unit changes in OHS). We then used a logistic regression model and assessed the discriminatory ability of the tool by calculating the area under the ROC curve.

Results

From the selected patients, only 63.7% in EUROHIP and 87.1% in EPOS had completed both a baseline and 1-year follow-up OHS. We observed discrepancies between the patients who did not answer the 1-year follow-up questionnaire and those who responded (Table 32). In the EUROHIP cohort, responders had better baseline pain and function and EQ-5D mental health scores, lower educational level and higher levels of obesity as well as greater ASA scores. In the EPOS study we found differences in younger responders and also in those employed and with a better preoperative SF-36 score.

The mean age of patients in EUROHIP was 65.7 years (SD 10.6 years) and 68 years (SD 10.7 years) in EPOS. There were 55.2% and 62.8% women in EUROHIP and EPOS, respectively. BMI was similar across the two cohorts, with a mean BMI of 27.8 kg/m² (SD 4.4 kg/m²) in EUROHIP and 27.3 kg/m² (SD 4.9 kg/m²) in EPOS.

Table 33 displays the predictors that have been found statistically significant.

For example, higher baseline OHS (better pain and function) was associated with an increased follow-up OHS, which is indicative of better pain and function outcomes. The effect of age was non-linear and there was a threshold effect. Patients aged > 75 years had worse outcomes. Worse outcomes were also observed in patients with an increased BMI, lower levels of education and lower baseline SF-36 scores. The pattern of OA was found to be an important predictor of outcome as a radiographic variable. Patients with superomedial, medial or concentric OA had worse outcomes than those with a superolateral pattern of disease or with no reduction in joint space. Worse outcomes were also linked with having had previous surgeries in other joints and having had arthritis in other joints. A posterior surgical approach and femoral component offset of ≥ 44 mm were associated with significantly better outcomes.

Internal validation and model performance

To validate the model internally we used bootstrapping to the imputed data sets. To ensure that the risk factors are replicated in other external validation studies we ensured that all predictors identified were those consistently selected across the 200 bootstrap resamples at least 70% of the time. The performance of the model was assessed by calibration and discrimination (see Table 34). Calibration of the predicted 12-month postoperative OHS was good, except in the lowest deciles of OHS, in which case the model overestimated the predicted score. Calibration of this predicted change in OHS was very good across all deciles of change in OHS. The model showed a discriminatory ability with a bias-corrected R² of 23.1%. We assessed the performance of the model and found the calibration of the predicted 12-month OHS to be good. The exception was the lowest deciles of OHS, where we found that the developed model overestimated the predicted score. The performance of the model is described in Table 34.

We also calculated the predicted the absolute change in OHS by subtracting the predicted 12-month score from the observed preoperative score. It was observed that the calibration of this absolute predicted change was very good. Importantly, the model also showed good discriminatory ability, with a bias-corrected R² of 23.1%.

Development of a risk prediction tool

We have developed a web-based tool that can be used to inform patients of the likely outcomes following their surgery. This web-based tool (Figure 28) uses the predictive equation:

FIGURE 28.

Example of a web-based calculator displaying the predictive tool based on the patient’s individual preoperative characteristics.

We tested the ability of the tool to identify the patients with the worst possible outcomes of surgery. The worst outcomes are defined as those THR patients whose preoperative score does not improve for at least 5 units over the 12 months. For 89 patients who met the criteria of the worst possible outcome, the discriminatory ability of the tool was good, with an area under the ROC curve of 0.78 (95% CI 0.73 to 0.82) (Figure 29). The tool had a sensitivity of 72% and a specificity of 73%. The calibration was also reasonable, as indicated by the Hosmer–Lemeshow p-value of 0.13.

FIGURE 29.

The ROC curve showing discriminatory ability of the model to identify those patients with the worst outcome (change in OHS of 5 units or fewer).

Main findings

Our study found that predictors associated with worse outcomes were old age, worse preoperative mental health score, higher BMI and the lower education attainment. Better preoperative pain and function were associated with better postoperative pain and function, and vice versa, patients with worse preoperative pain and function were observed to obtain the greatest symptomatic improvement (greatest change) between baseline and follow-ups described in previous literature. 52,186 As for the surgical factors, much less is described previously in the literature of their association with patient-reported outcomes of THR. In this work we confirmed that the larger femoral component was associated with a better THR outcome.

A new finding was observed with radiographic variables. Specifically, the pattern of joint space narrowing was found to be a strong predictor: patients with a medial, superomedial or concentric pattern of OA, as identified on a radiograph, had worse outcomes, as opposed to those with a superolateral pattern or with no intra-articular space. This observation applies only to patients with OA, as patients with other types of arthritis (such as RA) were not included in the analysis. The results can be explained by the likelihood that the superolateral pattern of disease developed as a result of subtle local mechanical factors such as abnormalities of acetabulum or the femoral head, that is acetabular dysplasia and cam-type femoroacetabular impingement. 33 It was previously suggested that these types of mechanical risk factors are associated with superolateral disease. 27,48 Superomedial, medial and concentric patterns of OA are more commonly observed in women, with the presence of Heberden’s nodes47 and are generally bilateral. 49 Therefore, the observed relationship between the pattern of OA and the outcome may be because of patients with mechanical disease having better pain and functional outcomes.

No associations were found between the radiographic severity of disease and the surgery outcome using K/L grade. This may be explained by the existing of other confounding factors in the model, that is, other markers of radiographic severity of disease.

We conclude that the tool developed within this work package to predict THR outcomes had calibration and discriminatory ability. The tool has promising potential to inform patients and their clinicians of the outcomes and help them with making the decision regarding whether or not to select patients for THR. The tool predicts the outcomes 12 months postoperatively and provides the absolute change in OHS expressed as a percentage. The performance of the tool to identify patients with the worst outcome following THR was also good. We tested the model for internal validation but future work will be done to provide external validation in the prospective cohort of pragmatic setting of NHS patients.

A clinical tool for the prediction of patient-reported outcomes after knee replacement surgery: a prospective cohort study142

To combine previously described risk factors into a prognostic tool to predict outcomes 12 months after TKA we used the data from the KAT cohort. 187,188 KAT is a pragmatic, partial factorial, unblinded randomised controlled trial (RCT). In total, 2352 participants were recruited through a random sample to KAT between July 1999 and January 2003, and then stratified by surgeon by sex, age and site of disease. The trial contains information on patients receiving primary TKR surgeries across 34 centres in the UK. One hundred patients who received UKR withdrew from the trial before the procedure or died were excluded from the analysis. From the remaining 2252 patients with primary TKR, 1649 agreed to complete both baseline and postoperative questionnaires. All TKR types were included whether or not a metal-backed tibial component, patellar resurfacing and a mobile bearing were used. Patients had completed questionnaires pre- and postoperatively (at 3 months, 1 year and annually after their surgery). The primary outcome measure for our analysis was 12-month postoperative OKS. Predictor variables collected for this work are summarised in Table 35.

Statistical methods used

For statistical analysis we used Stata version 12.1. We compared patient characteristics in responders and non-responders with preoperative and 12-month postoperative questionnaires. This helped with determining selection (response) bias. We assessed the relationship between the potential predictors on two outcomes: continuous 12-month postoperative OKS and change between pre- and postoperative OKS.

For continuous 12-month postoperative OKS we used general linear models to identify risk factors. We assessed the linearity of continuous variables with the outcome by the fractional polynomials and the collinearity between variables by variance inflation factor (VIF). Owing to the heteroscedasticity of the variance of residuals, robust standard errors were used to estimate the sandwich variance estimator. We used the chained equations to generate 40 imputed data sets to investigate the effects of missing data. We used all potential predictors (including outcome) and combined estimated parameters by using Rubin’s rule. Afterwards we selected 200 bootstrap samples with replacement from these 40 imputed data sets. We applied automatic backward selection within each bootstrap sample (significance level of 0.157). Age and sex were force entered into all models. All variables appearing for > 70% of the time were then used in the final regression model.

For change in the OKS, for identified predictors we used the repeated measures linear regression in which pre- and postoperative OKSs were included as an outcome. In order to describe the change in the outcome, we included the interaction term between each potential risk factor and time point.

Results

Internal validation and model performance

For the internal validation we used 200 bootstrap samples with replacement in the combination of multiple imputations and assessed predictive ability for bias-corrected estimates. We examined the discrimination (R²-statistic) and calibration to assess predictive ability. In the final predictive model the R² was 20.2%. In total, 14.3% of the variability of outcome was explained by age, sex and preoperative risk factors. With the inclusion of other patient characteristics the variability was 16.6%, and with the inclusion of clinical and surgical variables the variance reached 18.9% and 20.2%, respectively. We found that the model calibration was good, with close agreement between predicted and observed values of 12-month postoperative OKS.

Main findings

Although the association between outcome of TKR and several potential determinants has been previously established,189,190 we aimed to identify surgical variables as novel predictors of pain and functional outcome at 12 months postoperatively following knee replacement surgery. From surgical factors it was found that the presence of fixed flexion deformity and preoperative varus deformity and absence of ACL were associated with better postoperative outcome. Preoperative deformity of fixed flexion leads to significant changes in the biomechanics of the knee, subsequently leading to a debilitating condition. The surgery aims to correct the knee malalignment and fixed flexion deformity, thus improving biomechanics. Hence, this may be the cause of the improvement in pain and function. Severe deformity and absence of the ACL may also result in more severe joint damage and may manifest as severe symptoms of OA. The discriminatory ability of the model was also improved by inclusion of these surgical predictors in the model compared with the model with only patient characteristics. From patient risk factors such as age and sex, we identified that younger women had a better outcome than men, but in the older age group men performed better. The worst outcomes were associated with the lowest baseline OKS, low socioeconomic determinants, high BMI, worse mental health (as indicated by SF-12 score), worse ASA grade, history of previous knee surgery and the presence of other conditions impacting on mobility. From radiographic variables, patients with more severe disease achieve a greater degree of improvement after the knee replacement.

This work offers clinicians and patients a tool to predict outcomes 12 months after TKA. The tool can be used to decide whether or not patients should be selected for surgery. Potentially modifiable risk factors (such as BMI and mental health) can be improved before surgery, whereas factors such as age and sex will simply help to inform the decision of whether or not an operation is sensible. We have validated the tool internally and the next step would be to develop a web-based calculator, similar to the hip predictive tool, which would enable clinicians and patients to predict outcome after their knee surgery. The external validation of the tool is also required. This would be carried out in the prospective cohort of patients in the pragmatic NHS setting.

Preoperative transitions

The first section of the model covering the states and transitions between the surgical consultation and primary THR rendered this model not only novel but also contingent on information not systematically collected before. This is because no data were found in the published literature that described referral decisions by orthopaedic surgeons. In order to obtain estimates for these probabilities, we conducted a systematic expert elicitation exercise to estimate mean referral rates as well as uncertainty around those values.

We used an individual, direct method of expert judgement elicitation, which was a mathematical approach that revealed experts’ answers as distributions. We presented experts with a set of questions about their referral decisions of hypothetical THR patients. The histogram technique was employed, whereby a frequency chart showing intervals for the range of answers were presented to experts, who were asked to specify their relative subjective probabilities for each interval by placing a finite number of crosses throughout the grid. We conducted a pilot exercise and then decided to run individual guided interviews with the seven selected expert surgeons from hospitals in Southampton, Bournemouth, Oxford and Portsmouth. Responses from these surgeons were analysed, with their responses aggregated. Appendix 6 details the individual responses.

The deterministic model was populated with the mean values obtained from the responses provided in the expert elicitation. For the probabilistic sensitivity analysis (PSA), we assigned the corresponding beta distribution if two conditions were met: (1) the resulting probability density function had to appropriately fit the corresponding pooled probability distribution from experts’ responses; and (2) there was no significant difference between the observed mean value and that generated by the inverse of the cumulative density function evaluated at 0.5 (a difference > 0.05 was considered excessive). Table 38 shows the mean and SD, as well as the distribution and its parameters, if applicable, used to populate preoperative transition probabilities for the deterministic and probabilistic economic model.

Postoperative transitions

Data for the breakdown of good and poor outcomes after primary THR were obtained from the NHS PROMs initiative. The COASt project obtained, from the Health and Social Care Information Centre, the non-identifiable HES and PROMs records of all patients who had a hip replacement operation and who consented to participate in the PROMs initiative. Of a total of 171,881 PROMs records for the three fiscal years between 2009 and 2012, only 128,084 were linkable with HES. After dropping records for other hip replacement procedures, removing records from patients aged < 45 years and keeping only those with non-missing postoperative OHS so that their outcome category could be determined, the data set of primary THR by age and sex group comprised 68,156 interventions.

The outcomes of operations could all be classified as good or poor based on the set criteria. Thus, an OHS below 38 units at 1 year after the primary surgery was considered a poor outcome. The percentage of patients in each outcome category after primary THR, based on postoperative OHS reported in the HES PROMs data from 2009 to 2012, allowed for an estimate of the probability of poor outcome by patient group. As the split between good and poor outcome can naturally be considered binomial data, we fitted a beta distribution to the probability of poor outcome immediately after a primary THR based on the counts of good and poor outcomes within each patient subgroup, as reported in Table 39. Given the large number of observations, uncertainty around these parameter values was quite low.

We used data collected preoperatively and annually during 5 years after a primary THR by the EPOS group to estimate the probabilities of moving from each of the outcome categories in the first year to each outcome category in year 2, and of moving between health states representing outcome categories during the second and subsequent years after the primary. The EPOS data available to us on primary THR patients included OHSs and other demographic information from a total of 1589 patients.

Table 40 reports the transition probabilities estimated from the sample for each patient subgroup as well as the distribution parameters. Only one transition from each outcome category is reported as the other will result from calculating one minus the probability of death, minus the probability of revision (reported in the next section) and minus the probability reported in Table 40.

Using data from the EPOS trial, we extrapolated proportions of good and poor outcomes up to 10 years after the primary THR by calculating the mean transition probability for remaining in each outcome category between years 2 and 5 and applying it to the transitions between years 5 and 10. Table 41 shows those mean probabilities and the parameters of the beta distributions calculated based on average counts. Finally, yearly mortality rates from both model states were assumed to be the same all-cause sex- and age-specific death rates used preoperatively. 125

Revision rates were estimated for our two categories of outcome after THR. These rates were based on the rates proposed by Kalairajah et al. 210 for four groups, using 27, 33 and 41 as cut-off points on postoperative OHS at 6 months after primary surgery based on a sample of over 15,000 THRs from the New Zealand Joint Registry. Based on the overall proportion of patients classified as having poor and good outcomes in the HES PROMs data set (45% and 55%, respectively), we chose to consider the three Kalairajah et al. ’s groups scoring up to 41 units as poor (42%) and those above 41 units as good (58%). We estimated instantaneous revision rates (assuming the rate was constant over the 2 years) and then probabilities of revision at 1 year for the two outcome categories. This is described in detail in Appendix 7. Table 42 shows the distribution parameters of each revision rate for the PSA, which will be applied to all patient subgroups as the data were not reported by sex and age groups.

The probabilities of good and poor outcome following a THR revision were estimated using HES PROMs data, analogously to the approach taken for the probabilities following primary THRs. As the study identifying cut-off points for outcome categories used data from primary THRs only,154 and it has not been replicated on revision operations, we used the threshold identified for the second year (OHS 33) to classify our HES PROMs patient records into good or poor outcomes. We chose the lower second year cut-off point as opposed to that for the first year after the operation because patients undergoing a revision THR would have had problems with their primary prosthesis and are less likely to perform well than the broader spectrum of patients undergoing a THR for the first time. The resulting transition probabilities are shown in Table 43.

As in the case of primary THRs, our economic model required two sets of transition probabilities as the cohort moves through the Markov model following a revision procedure. First, after their first year in good or poor outcome immediately following the revision, patients who do not die would transit into good or poor outcome at year 2, which are modelled as separate health states; and, second, patients in either outcome category at year 2 or onwards may remain in the health state they are in or move to the other one at each iteration.

Given that no other data were available with yearly follow-ups of revision THR patients, we used the same data from EPOS primary THR records to produce estimates for the transition probabilities between outcome categories after a revision, using 33 as the cut-off point for outcome classification at 1 year after revision THR. Table 44 shows the estimated probabilities of the transition between good or poor outcomes during the first year following the revision to the same outcome category the second year after the procedure. Table 45 shows the mean transition probabilities and distribution parameters entered in the model for the transition of patients between outcome categories after the second year following the revision procedure.

Preoperative costs

For the costs associated with all health states previous to a THR, records about both consultations and prescriptions were considered. Consultation and medication costs were then added together and, thus, the progression of estimated total costs attributable to hip pain used by patients during the 15 years prior to their THR was produced, showing an expected upwards trend, especially in the 2 years prior to the primary surgery. This growing estimated costs confirmed the increasing burden generated by unresolved hip pain and problems experienced by patients who are referred for a THR. We used the estimates for the year immediately prior to a THR to populate all preoperative states. Deterministic analysis was based on the above mean values for costs attributable to hip problems. For PSA, a normal distribution was used to model the uncertainty around the difference in resource use between the two groups. The process for estimating these preoperative costs is discussed in more detail in Appendix 8. Table 46 shows the estimated cost of all preoperative model health states by patient subgroup.

Primary total hip replacement costs

The cost of a primary THR was estimated separately for each patient subgroup based on Healthcare Resource Group (HRG) assignment. In order to do this, we obtained the table of relative frequencies of HRGs assigned by the payment by results system in the NHS to each THR reported in HES and eligible for PROMs for fiscal year 2011–12 (N Gutacker, Centre for Health Economics, University of York, 2013, personal communication). Based on these frequencies and the NHS reference costs by HRGs for the same year,214 we calculated the mean cost for a primary THR by patient subgroup (Table 47).

Postoperative costs

To estimate postoperative primary health-care costs, we followed the same process used for preoperative costs. However, the resulting costs pool together the costs associated with good and poor outcomes. To produce separate estimates for the latter, we developed a model to predict outcome categories after surgery based on health-care resource use during the first year after the primary surgery. This was done using data from the COASt cohort, described in Chapter 5, and Appendix 9 explains in detail the estimation and use of the model. This model proved to be a unique and a valuable instrument, as the COASt cohort appears to be the only data set with both postoperative outcome and resource use data available. The connection between these was then used to estimate the outcome category of patients reported in the CPRD who have a wealth of resource use data but no outcome measure available. The classification of patients by outcome group using the model led to an estimated mean of £280 for likely poor-outcome patients and £24 for likely good-outcome patients during the first year after the primary. The mean values and distribution of costs for the first year after the operation added to the mean costs of surgery to produce the total costs associated to the model states combining the primary THR and the first postoperative year in good or poor outcome. These are shown in Table 48 with parameter distributions set to normal. As THR costs were estimated regardless of surgery outcome, the slightly higher overall mean costs of poor outcomes are explained by their higher primary care costs.

Costs for the second and subsequent years in either outcome category were estimated based on CPRD records of resource use, as above, but with the application of an adjusted surgery outcome prediction model. Once CPRD records were labelled as likely poor and likely good outcomes, records for years 2 through 10 after the primary surgery were pooled together, and mean values and distributions estimated to represent the yearly cost of primary care for patients 2 years and onwards after a THR. These values are reported in Table 49 by patient subgroup.

Revision total hip replacement and postoperative costs

Parameters for the four revision and post-revision model states were estimated following the same process as that of post-primary health states. Table 50 reports the mean and SDs of the costs associated to the revision THR operation by patient subgroup.

For the costs of primary care provided to revision THR patients we used CPRD data. In order to distinguish between primary care costs provided to good and poor outcome patients after a revision THR, we employed the predictive model described previously using OHS 33 units as a threshold for outcome categories and resource use data from primary surgery. Given the low number of revision cases further divided after fitting the predictive model to the data, results for the primary care cost component could not be presented for each of the eight patient subgroups and were instead produced in aggregate form for all patients. Table 51 shows the mean cost associated to the model states combining group-specific revision THR and common primary care costs for the first postoperative year in which the differences by outcome categories are explained entirely by primary care costs associated to each group.

For the model states representing each outcome category after a revision THR, Table 52 indicates the mean and SD of primary care costs that, given the low number of CPRD observations, were applied to all patient subgroups as well. These were estimated by pooling together all CPRD records for the years 2 through to 8.

Mapping Oxford Hip Score to EuroQol-5 Dimensions

Since April 2009, NHS providers performing unilateral hip replacements have been required to collect both EQ-5D scores and the OHS, a condition-specific outcome measure. 144 Before this, however, EQ-5D was not routinely collected from THR patients, whereas the OHS questionnaire was commonly collected and regarded as an important indicator of the success of THR. 145 The OHS consists of 12 patient-completed statements covering pain, mobility and ability to carry out regular tasks. The current scoring system assigns values between 0 and 4 to each item, higher scores corresponding to better outcomes. Individual scores are summed, giving a total score ranging from 0 (worst) to 48 (best). 145 The EQ-5D is a widely used generic measure of health outcomes. It produces a summary index for each of the 243 descriptive health states by applying a preference-based valuation derived from a sample of the general population. 215

The ability to estimate EQ-5D scores based on the OHS would enable estimation of utility data for older data sets for which OHSs were collected but EQ-5D scores were not. Older data sets are of key importance given the need for long-term follow-up of hip replacement patients whose prostheses, in most cases, last for many years without the need for replacement.

In order to estimate the summary EQ-5D index from OHS responses, we employed two known conversion algorithm techniques: transfer-to-utility (TTU) regression and response mapping. The TTU regression approach uses regression equations to predict the values of one outcome measure, using scores from a second measure as regressor(s). 216 We used three different TTU regression methods in this modelling exercise: two are variations of the linear regression model, and the third is a two-part model combining a binary outcome and a linear regression model. The first model regressed total OHSs on the EQ-5D summary index using OLS. The second method employed responses to all 12 questions of the OHS questionnaire as categorical regressors. The third was a two-part logit OLS model, whereby a first logit model would estimate whether or not OHS responses are associated to an EQ-5D score of 1 unit, and if not a second OLS model would predict the value (< 1 units). Meanwhile, response mapping seeks to predict the responses to each of the five individual EQ-5D questions instead of predicting the summary score directly. 217 A multinomial logistic model was applied to do this. Each of these models is described in more detail in Appendix 10.

Data were obtained from the SWLEOC database. The full data comprised 3504 hip replacements, each with preoperative and/or 6-month postoperative responses to the OHS and EQ-5D questionnaires, plus basic demographic, socioeconomic and clinical information. All except two operations were performed between 2006 and 2008. All models were estimated on 1759 operations for which we had data on both pre- and postoperative OHS and EQ-5D scores, sex, age and deprivation. As we were interested in cross-sectional mapping, we pooled pre- and postoperative records together, providing 3518 outcome observations.

The simplest model, continuous OLS with total OHS as the only regressor, was statistically significant with residuals approximately normally distributed. The categorical version of the linear regression model including all OHS separate questions also reported residuals nearly normally distributed. However, it produced some coefficients that were both statistically non-significant and inconsistent with the positive relationship between OHS and EQ-5D; that is, they were either negative or did not follow an increasing progression within the same question. This was fixed by pooling response levels together. For the two-part approach, we estimated the first (logistic) part of the model and found that only 13 of the 48 regressors were statistically significant. For the second part (categorical OLS) we estimated the model on observed EQ-5D scores lower than 1 and included all OHS questions, as none could be excluded based on repetitive likelihood ratio tests. We collapsed response levels using the same methods as with the categorical OLS model. Again, we found that residuals were approximately normally distributed but with a high peak at zero from perfectly fitted cases of observed EQ-5D equal to 1. For the response mapping approach we found that all five multinomial models (one for each EQ-5D question) were statistically significant (p < 0.001); however, many of the individual regressors were not. These results are also detailed further in Appendix 10.

All selected variations of each model were internally validated and predictive power was highest for both OLS models. In addition to assessing the models’ ability to predict the observed mean EQ-5D score, we also calibrated them by recording prediction errors through the range of values of the dependent variable. All models reported a high predictive power of the aggregate mean, although the level of precision was not uniform across the full range of scores. Overall performance of the four models was within the range of other reported mapping studies based on their root-mean-square errors of around 0.20. 218

Preoperative quality-adjusted life-years

The HES PROMs data used to estimate the probabilities of good and poor outcome after a primary THR were the most appropriate to estimate the HRQoL of patients before they undergo the operation given its large number of observations and national representativeness. In order to fit a distribution to these health utilities, which range from –0.5 to 1 quality-adjusted life-years (QALYs), disutilities (the distance in QALYs between the recorded utility and its maximum value of 1) were calculated so that the range becomes positive and bound by zero. As disutilities, a gamma distribution was fitted using the mean and SDs to produce distribution parameters. Table 53 shows mean values and distribution parameters for disutilities associated to all preoperative model states for each patient subgroup.

Quality-adjusted life-years after primary total hip replacement

Patients undergoing a primary THR accrued health utilities depending on their outcome category. First, for health utilities associated to model states including the operation, we considered the EQ-5D postoperative summary scores by patient subgroups reported in the HES PROMs data set. We produced QALYs from the latter by considering the progression of scores following a primary THR observed in EPOS because the latter reported a measure at 3 months that helps better understand patients’ rate of improvement during the first year after the primary surgery. Figure 31 shows estimated EQ-5D scores before, 3 months and 1 year after the operation mapped from the summary OHSs using the mapping method reported above. As evidenced in this graph, poor outcomes not only improve less in the first 3 months than good outcomes, but they also do not improve any more after that, on average, whereas good outcomes still improve a bit more in the last 9 months of the first year after the THR. This progression was used to estimate QALYs during the first year after the primary surgery using the HES PROMs data.

FIGURE 31.

Mean EQ-5D summary scores mapped from OHSs reported by EPOS patients.

We applied the progression patterns to the data on pre- and postoperative EQ-5D summary scores reported in the HES PROMs data set, to estimate QALYs associated to the first year after primary THR for each patient subgroup by outcome category. We calculated disutilities and estimated the parameters for the corresponding gamma distributions. Table 54 shows the mean value, as well as the parameters for the gamma distributions, describing uncertainty around the mean value of the QALYs associated to the first year after a primary THR for both good and poor outcome patients.

For the health states representing the second and subsequent years after the primary surgery in either outcome category, we also benefited from the longer follow-up performed under EPOS and the representativeness in the HES PROMs data when estimating the required QALY values. We estimated EQ-5D scores among patients in the second and subsequent years after surgery based on the data in the EPOS and HES PROMs by combining the yearly relationships found in the former with the point estimates, variability and representativeness of the latter. Table 55 shows the mean values and distribution parameters for disutilities during the second and subsequent years after the primary THR for each outcome category and by patient subgroup.

Quality-adjusted life-years after revision total hip replacement

Health utility estimates were obtained following an analogous procedure as that used for after a primary THR. For the health states, including the revision procedure and the first year in either of the outcome categories, QALYs were estimated based on pre- and postoperative EQ-5D summary scores from patients undergoing a revision THR in HES PROMs combined with the estimated EQ-5D progression by EPOS primary THR patients used in the previous section. We used the progression of scores after a primary procedure because no data set was available containing follow-up HRQoL measures for revision THR patients before and 1 year after the operation, as well as at a third point in between. Table 56 shows the means and SDs of the EQ-5D summary scores of revision THR patients extracted from the HES PROMs data set.

In order to estimate the QALYs associated to this first year after the revision THR, we connected the start and end points reported in Table 56 using the differential progression by outcome category found for primary patients. We estimated QALYs based on this progression, converted them into disutilities and produced the mean values and gamma distribution parameters shown in Table 57.

The lack of follow-up data on sufficient revision THR patients meant that, for the model states representing the second and subsequent years after their revision operation, we used the patterns of progression observed in patients who underwent a primary surgery. Estimated EQ-5D summary scores, assumed to remain constant over the year, were converted into disutilities, and their mean values and gamma distribution parameters estimated. These values are shown in Table 58.

Modelling assumptions

As with all models, the one presented here attempts to reflect the true care pathway of patients as they are assessed for a THR, which most undergo, but it necessarily simplifies what in reality is a more complex process. Our model, as any other, simplifies reality so that we can produce estimates for the cost-effectiveness of the outcome prediction tool. This simplification is achieved by making a number of assumptions that can make the model feasible.

First, this model assumes that the outcome prediction tool is capable of identifying potential poor surgical outcomes before patients have the operation. In particular, we are assuming that the information in the EPOS and EUROHIP data sets is representative of the equivalent characteristics and outcomes in the wider population eligible to undergo THR in the UK. Second, we are assuming that all, or most, patients with an OHS of ≥ 38 units the first year after their primary surgery are free from pain and major mobility limitations as well as satisfied with the operation, and that the opposite is true for those who have a OHS of < 38 units. Third, we assume that all patients found to be candidates for surgery, but presenting a risk factor that should be dealt with before the operation, whether it is excessive weight, diabetes, high blood pressure or something else, can all be grouped together into the risk factor modification state (in which most patients would be expected to stay for only a short period). Similarly, we group a diverse set of patients into the health state of long-term medical management. Fourth, we assume that probabilities of good and poor outcomes are the same in the model whether the patient comes from the risk factor modification section or from that of long-term medical management. Fifth, we do not allow for multiple revisions. And, sixth, we assume that the tool would be used by orthopaedic surgeons.

Parameter assumptions

During the process of identifying parameter values to populate the economic model, a number of assumptions were made, whether because of the simplification forced by the fact that we were modelling a complex reality or because of limitations in the data available.

For transitions, first, although preoperative transition probabilities may vary between patient subgroups, the values extracted from the expert elicitation exercise were assumed to apply to all patients regardless of age or sex. Second, transitions between good and poor outcomes after year 2 post operation were estimated based on follow-up questionnaires only up to year 5. Third, we assume that surgery outcome after the primary has no bearing on surgery outcome after revision (as patients can move between outcome categories after a primary). Fourth, we used the cut-off point derived for primary THRs to categorise outcomes after revisions. Transition probabilities between outcome categories were also assumed to be equivalent after primaries and after revisions when those calculated for the former were applied to the latter. Fifth, and finally, we applied all-cause mortality rates from the general population.

For HRQoL values, first we assumed that the pattern of progression by outcome category during the first year after the operation in EPOS is generalisable to the wider population. We also assumed that the connection between OHSs in the first and second years is representative of the changes all or most patients would experience. Second, we did not consider a utility decrement when assigning health utility estimates to the good and poor outcome states where most patients would remain for long periods of time, until death in many cases. Third, we assumed that the progression of health utility estimated from primary THR patients in EPOS was also applicable to revision THR patients. Fourth, and finally, we did not consider any uncertainty around the time trade-off weights reported in the literature for the EQ-5D. 215

Regarding assumptions about cost parameters, the first significant assumption was that there was no cost to the risk factor modification state. Second, surgery costs were assumed to be the same regardless of outcome category 1 year after the operation. Third, the costs of complications were not explicitly included. And, fourth, in using preliminary results from the COASt cohort for sections of the cost estimation exercise, we assumed that the cohort is representative of clinical practice and, more generally, of patients in the UK.

The implications of these modelling and parameter assumptions, as well as related limitations, are discussed at great length in Appendix 11.