Opioid use, post-operative complications, and implant survival after unicompartmental 1 versus total knee replacement: a population-based network study

The aim of this study was to compare unicompartmental and total knee replacement (UKR and TKR), 31 emulating the design of the Total or Partial Knee Arthroplasty Trial (TOPKAT) using routinely- 32 collected data. The primary outcome in TOPKAT was patient-reported outcomes, with secondary 33 outcomes including post-operative complications and implant survival. Five US and UK healthcare databases, part of the Observational Health Data Sciences and 36 Informatics (OHDSI) network, were analysed. Opioid use from 91 to 365 days after surgery, as a 37 proxy for persistent pain, was assessed. Post-operative complications (venous thromboembolism, 38 infection, readmission, and mortality) were considered over 60 days following surgery and implant 39 survival over five years following surgery. Propensity score matched Cox proportional hazards 40 models were fitted for each outcome. Calibrated hazard ratios (cHRs) were generated for each 41 database to account for observed differences in control outcomes and these were combined using 42 meta-analysis.


Abstract 29
Background 30 The aim of this study was to compare unicompartmental and total knee replacement (UKR and TKR), 31 emulating the design of the Total or Partial Knee Arthroplasty Trial (TOPKAT) using routinely-32 collected data. The primary outcome in TOPKAT was patient-reported outcomes, with secondary 33 outcomes including post-operative complications and implant survival. 34

Methods 35
Five US and UK healthcare databases, part of the Observational Health Data Sciences and 36 Informatics (OHDSI) network, were analysed. Opioid use from 91 to 365 days after surgery, as a 37 proxy for persistent pain, was assessed. Post-operative complications (venous thromboembolism, 38 infection, readmission, and mortality) were considered over 60 days following surgery and implant 39 survival over five years following surgery. Propensity score matched Cox proportional hazards 40 models were fitted for each outcome. Calibrated hazard ratios (cHRs) were generated for each 41 database to account for observed differences in control outcomes and these were combined using 42 meta-analysis. 43

Findings 44
In total, 32,379 and 250,377 individuals who received UKR and TKR were matched and included in 45 the analysis. UKR was associated with a reduced risk of post-operative opioid use (cHR from meta-46 analysis: 0.81 (95% CI: 0.73 to 0.90)). UKR was also associated with a reduced risk of venous 47 thromboembolism (cHR: 0.62 (0.36 to 0.95)), but little difference was seen for infection (cHR: 0.85 48 (0.51 to 1.37)) and readmission (cHR: 0.79 (0.47 to 1.25)). There was insufficient evidence to 49 conclude there was a reduction in risk of mortality. UKR was also associated with an increased risk of 50 revision (cHR: 1.64 (1.40 to 1.94)). 51 Interpretation 52 UKR was associated with a reduced risk of opioid use compared to TKR, which may indicate a 53 reduced risk of persistent pain after surgery. UKR was associated with a lower risk of venous 54 thromboembolism. UKR was also, however, associated with an increased risk of revision compared 55 to TKR This study emulates the TOPKAT design using routinely-collected data. Where possible, similar 76 eligibility criteria were specified and outcomes assessed in a similar manner. Patient-reported 77 outcomes (the primary outcome in TOPKAT) were not available, and so opioid prescriptions were 78 used as a proxy for persistent pain following surgery. Post-operative complications and implant 79 survival were also assessed. The findings from this study will provide further evidence to inform 80 considerations of the relative merits of UKR and TKR. 81 Implications of all the available evidence 82 In this study, UKR was associated with a reduced risk of post-operative opioid use between 91 to 365 83 days after surgery relative to TKR, and this may indicate a reduced risk of persistent pain after UKR. 84 As seen in this study and in previous research, UKR also appears to have a lower risk of venous 85 thromboembolism compared to TKR. However, while revision rates were similar for UKR and TKR in 86 TOPKAT, the findings from this study support that of previous observational research showing UKR 87 Introduction 92 Knee replacement is one of the most common surgical procedures and typically leads to substantial 93 improvements in pain, function and quality of life. 1 However, there is variation in how knee 94 replacements are performed. One area of particular uncertainty is around whether to use 95 unicompartmental or total knee replacement (UKR or TKR) for those individuals with osteoarthritis 96 confined to a single compartment of the knee. While all the compartments of the joint are replaced 97 in TKR, only the affected part of the joint is replaced in UKR. 98 With patient-reported pain and function key indications for knee and hip replacement, it follows that 99 they should also be considered as a key measure of the effectiveness of surgery. Previous research 100 has generally found UKR and TKR to result in broadly similar gains in patient-reported outcomes after 101 surgery. 2 Both UKR and TKR are major orthopaedic procedures and so are accompanied by a risk of 102 post-operative complications. Findings from previous research suggests that UKR, which is a quicker 103 and less-invasive procedure relative to TKR, may have a lower risk of some post-operative 104 complications, notably venous thromboembolism, infection, and mortality. 2 As well as the short-term 105 risk of post-operative complications, patients who have had a knee and hip replacement have a long-106 term risk of revision surgery, in which implant components are removed, added or exchanged. 107 Revision procedures are associated with significant morbidity for individuals, with those undergoing 108 revision surgery generally reporting worse patient-reported outcomes before and after revision 109 procedures compared with those undergoing primary procedures. 3 Observational research has 110 consistently found UKR to have a higher risk of revision procedures compared to TKR, with the 111 increased risk maintained over 25 years after the primary procedure. 2,4 112 In a recently published randomised controlled trial comparing UKR and TKR, the Total or Partial Knee 113 Arthroplasty Trial (TOPKAT), 264 patients were randomly assigned UKR with another 264 assigned 114 TKR, with 245 and 269 going on to receive UKR and TKR, respectively. Surgeons performing the 115 procedures were either 'equipoise' surgeons who performed both surgeries, or 'expertise' surgeons 116 who performed only one of the procedures while another 'expertise' surgeon in the same centre 117 performed the other. To perform a given procedure surgeons needed to have been practising it for 118 at a year and to have performed it at least ten times in the previous year. 5 The trial was powered to 119 assess the primary outcome which was self-reported pain and function, as measured by the Oxford 120 Knee Score (OKS). 5 Both groups achieved substantial improvements in OKS relative to baseline 121 scores, with the gains broadly similar across the two comparator groups. Post-operative 122 complications and implant survival were also assessed in TOPKAT as secondary outcomes. Fewer 123 individuals had a post-operative complication after UKR compared to TKR. In contrast to the 124 previous observational research, UKR and TKR were also seen to have similar rates of revision after 5 125 years in the trial. 6 126 The aim of this study was to emulate the TOPKAT trial design using routinely-collected data, so as to 127 answer the same causal question. A study which uses routinely-collected data to emulate the 'target 128 trial' should be harmonised, with similar study designs applied to allow for meaningful 129 comparisons. 7,8 The primary outcome was patient-reported pain and function. As this was not 130 possible, the effect of type of procedure (UKR or TKR) on persistent pain after surgery was 131 considered. Secondary outcomes in the target trial included post-operative complications and 132 implant survival, and these were also assessed in this study. 133

134
A network cohort study was conducted across 5 observational health care databases from the US 135 and the UK. Exposure cohorts 155 Individuals who underwent either a UKR or TKR were identified. Study participants were required to 156 have data captured over at least the year prior to surgery. We excluded patients using published 157 exclusion criteria of TOPKAT, 5 with individuals required to be aged 40 or over at surgery, and have no 158 prior evidence of knee arthroplasty, knee fracture, knee surgery except for diagnostic procedures, 159 rheumatoid arthritis, inflammatory arthropathies, or septic arthritis. In addition, patients with spine, 160 hip, or foot pathology in the year prior to surgery were also excluded. These criteria were intended 161 to identify patients who were eligible for either type of knee replacement, and exclude patients who 162 were not indicated for either UKR or TKR. 163 Outcome definitions 164 Relating to patient-reported outcomes which were the primary outcome in the target trial, 165 persistent pain after surgery was assessed using opioid use (identified by a written or dispensed 166 opioid prescription) as a proxy, with a time-at-risk 91 days after surgery to 1 year after surgery. were calibrated based on the estimated residual error from negative control outcomes and synthetic 199 positive control outcomes. 15,16 Empirical calibration is a process whereby the residual error of an 200 estimator is quantified and incorporated into a calibrated version of the estimator. The calibrated HR 201 (cHR), in this case, reflects the distribution of estimates on the negative control outcomes. For 202 example, if the negative control estimates are on average greater than the null, an increased risk for 203 the outcome of interested will be attenuated following calibration. The cHRs were only estimated if 204 a sufficient number of control outcomes were observed during a given time-at-risk window. Each 205 analysis was conducted separately in each database. 206 Findings across databases were combined using meta-analysis, with the inverse variance random-207 effects approach used. 17 At the request of peer review, results were meta-analysed for each of the 208 outcomes. An I 2 above 40% can, however, be taken to indicate substantial heterogeneity across 209 databases. 18 Estimates for negative and positive controls were pooled before performing empirical 210 calibration on the pooled estimates. 211

Sensitivity analyses 212
Pre-specified sensitivity analyses were run for each of the outcomes of interest, with variations of 213 cohort definitions, time-at-risk, and approaches to matching (Appendix Table 2). 214 Role of the funding source 215 The funder of the study had no role in study design, data collection, data analysis, data 216 interpretation, or writing of the report 217 Results 218 32,379 individuals who had UKR and 250,377 who had TKR were matched using propensity scores 219 (see Appendix Figure A1 for study flowcharts). Prior to matching, individuals undergoing UKR were 220 younger and healthier than those undergoing TKR (Appendix Table A3). Diagnostics for propensity 221 score matching and control outcome findings are summarised in Appendix Figure A2. After 222 matching, both cohorts appeared largely comparable in terms of observed characteristics (Table 1  223 and Appendix Figure 2). Individuals in the matched CCAE, Optum, and PharMetrics cohorts were 224 generally younger and had fewer comorbidities compared to THIN and, in particular, MDCR. THIN 225 covered the broadest age range of individuals. Pre-operative opioid use was well balanced for the 226 comparator groups, with between 30% to 45% of individuals classified as an opioid user before 227 surgery. 228 UKR was consistently associated with a reduced risk of opioid use after surgery relative to TKR, with 229 cHRs for the use of opioids in the 3 to 12 months post-surgery ranging from 0.70 (0.57 to 0.90) for 230 THIN to 0.86 (0.78 to 0.96) for Optum. The estimate from meta-analysis was 0.81 (0.73 to 0.90). The 231 cumulative incidence of opioid use in the 3 to 12 months post-surgery was around 35% to 40% for 232 UKR and about 5 percentage points higher for TKR in the 4 databases from the US. Opioid use was 233 around 20% for UKR and 25% for TKR in the database from the UK (Appendix Figure A2). These 234 findings were generally similar across sensitivity analyses. When considered up to 5 years, UKR was 235 still associated with a reduced risk of opioid use, but the estimated effects were slightly attenuated 236 with cHRs ranging from 0.86 (0.78 to 0.96) for CCAE to 0.90 (0.82 to 1.02) for MDCR, with no meta-237 analysis performed for these outcomes. 238 UKR was consistently associated with a lower risk of venous thromboembolism compared to TKR. 239 The Optum (Figure 1). Estimates from meta-analysis were 0.85 (0.51 to 1.37) and 0.79 (0.47 to 1.25) for 245 infection and readmission, respectively, although in both cases I 2 was above 0.5. Finally, there was 246 little evidence of an association between procedure and mortality, with a cHR of 1.26 (0.55 to 3.09) 247 in Optum and a HR of 0.51 (0.03 to 2.51) in THIN. Findings were broadly similar across sensitivity 248 analyses. There was stronger evidence, however, that UKR was associated with a reduced risk of 249 readmission when considered over the year following surgery rather than 60 days in CCAE and 250 MDCR, cHRs 0.75 (0.66 to 0.86) and 0.76 (0.64 to 0.93), respectively. 251 UKR was consistently associated with an increased risk of revision compared to TKR over the five 252 years following surgery (Figure 1), with cHRs ranging from 1.48 (1.25 to 1.83) for PharMetrics to 2.16 253 (1.63 to 3.15) for MDCR. The estimate from meta-analysis was 1.64 (1.40 to 1.94), although I 2 was 254 0.5. After 5 years, implant survival was generally around 97.5% to 95% for TKR and 95% to 92.5% 255 following UKR (Appendix Figure A3). These findings were similar across the various sensitivity 256 analyses considered. 257 Results for the primary analysis and each sensitivity analysis are detailed in Appendix Table A4.  258 These can also be viewed, along with study flow charts, characteristics of study participants before 259 and after matching, and propensity score distributions, using the interactive web-based application 260 at http://data.ohdsi.org/UkaTkaSafetyEffectiveness. 261 Discussion 262 In summary, compared to TKR, UKR was associated with a reduced risk of post-operative opioid use. 263 This may indicate that UKR has a lower risk of post-operative persistent pain. UKR was also 264 associated with a decreased risk of post-operative venous thromboembolism. There was insufficient evidence to conclude there UKR led to a reduction in risk of infection, readmission, or mortality. TKR 266 was associated with a lower risk of revision. 267 The primary outcome in TOPKAT was patient-reported pain and function, as measured by OKS. 268 Outcome scores were broadly similar for the two comparator groups. Indeed, while risk of revision over 5 years after UKR is currently around 6% in the UK, risk for TKR is 296 approximately 2.5%. 26 The incidence of revision for study participants from the UK included in this 297 study are in line with these previous findings, with revision risks seen to be slightly higher for study 298 participants from the US. As with previous observational studies, UKR was also consistently 299 associated with an increased risk of revision in this study. UKR can therefore be expected to have a 300 higher risk of revision than TKR. 301 This analysis has been informed by data from 280,000 patients across 5 databases in 2 countries. 302 This retrospective analysis relied though on data captured in electronic health records and 303 administrative claims, and therefore our ability to emulate the inclusion criteria used the TOPKAT 304 trial was limited. In particular, these data did not have radiographic information and so it was not 305 possible to assess whether an individual's osteoarthritis was confined to one compartment of the 306 knee. Patient-reported outcomes, the primary outcome in TOPKAT, was also not captured in the 307 databases used and so opioid use was used as a proxy for persistent pain. This has limitations, 308 however, as opioids may not necessarily have been taken even if dispensed. We used large-scale 309 propensity score matching to balance the two cohorts using more than 10,000 candidate baseline 310 characteristics. However, as with all observational studies, there remains the potential risk of 311 confounding due to unmeasured factors. We employed a large panel of negative control outcomes 312 to mitigate the threat of systematic error. While the definitions for exposures and outcomes were 313 clinically reviewed and relied on codes used in prior published studies, 20,27-30 individual cases were 314 not validated and may be subject to misclassification. There may be measurement errors, for 315 example with baseline characteristics, such as comorbidities, and outcomes, such as revision, 316 potentially not recorded within the databases, in which case they would also be missed in the 317 analysis. As patients in this study were selected on the basis of their inclusion criteria for the TOPKAT 318 trial, the results from this study may also not necessarily generalise to those patients excluded from 319 the trial but are eligible for both procedures. In addition, while meta-analysis was used to combine 320 findings across databases, in a number of cases substantial heterogeneity was present and so the 321 resulting estimates should be interpreted with caution. 322 In conclusion, with a lower risk of post-operative opioid use, UKR may be associated with a reduced 323 risk of persistent pain compared to TKR. UKR is also associated with a lower risk of venous 324 thromboembolism. UKR is also, however, associated with an increased risk of revision. The merit of 325 using real-world data for assessing the effectiveness of treatments is still debated, 31,32 and 326 randomised controlled trials remain the 'gold standard' for establishing efficacy. This study has 327 demonstrated the value of real-world evidence for complementing the evidence produced from 328 randomised trials. 329 330 331 332