Abstract
We compared thromboembolic events, major haemorrhage and death after total hip replacement in patients receiving either aspirin or low-molecular-weight heparin (LMWH). We analysed data from the National Joint Registry for England and Wales linked to an administrative database of hospital admissions in the English National Health Service. A total of 108 584 patients operated on between April 2003 and September 2008 were included and followed up for 90 days. Multivariable risk modelling and propensity score matching were used to estimate odds ratios (OR) adjusted for baseline risk factors. An OR < 1 indicates that rates are lower with LMWH than with aspirin. In all, 21.1% of patients were prescribed aspirin and 78.9% LMWH. Without adjustment, we found no statistically significant differences. The rate of pulmonary embolism was 0.68% in both groups and 90-day mortality was 0.65% with aspirin and 0.61% with LMWH (OR 0.93; 95% CI 0.77 to 1.11). With risk adjustment, the difference in mortality increased (OR 0.84; 95% CI 0.69 to 1.01). With propensity score matching the mortality difference increased even further to 0.65% with aspirin and 0.51% with LMWH (OR 0.77; 95% CI 0.61 to 0.98). These results should be considered when the conflicting recommendations of existing guidelines for thromboprophylaxis after hip replacement are being addressed.
Venous thromboembolism is a rare but significant complication following total hip replacement (THR). A wide range of pharmaceutical agents are available for thromboprophylaxis, many of which have been shown in randomised controlled trials (RCTs)1 to reduce the risk of deep-vein thrombosis (DVT) and pulmonary embolism (PE). However, these drugs may increase other risks, such as bleeding, wound infection and thrombocytopenia.2-4
A recent review comparing 11 national and international guidelines on thromboprophylaxis in orthopaedic surgery concluded that different thromboprophylactic drugs were recommended based on the same available evidence and reported that the recommendation whether to use aspirin or not remained a significant source of conflict.5 For example, the United Kingdom-based National Institute for Health and Clinical Excellence (NICE) and the American College of Chest Physicians (ACCP) recommend low-molecular-weight heparin (LMWH) or fondaparinux in combination with mechanical prophylaxis as the most cost-effective option, but aspirin was not recommended.1,6 The American Academy of Orthopaedic Surgeons (AAOS), on the other hand, included the use of aspirin in its recommendations.7 Aspirin is an appealing method of thromboprophylaxis as it is inexpensive and can be administered orally without requiring laboratory monitoring. Not surprisingly, it is provided to about 20% of patients undergoing THR in England and Wales.8
There are only four small RCTs, each including 300 patients or less, that directly compare aspirin with LMWH.9 These suggest that LMWH is better than aspirin for the prevention of DVT. However, the advocates of aspirin argue that there is no direct evidence that LMWH is more effective than aspirin in preventing PE and reducing post-operative mortality.9
We directly compared the rate of PE and death in the first 90 days after joint replacement with either aspirin or LMWH as the only pharmacological method of thromboprophylaxis, as well as the rates of DVT, major haemorrhage and return to theatre, in more than 100 000 patients who had a primary THR in England between 2003 and 2008. Information about the method of thromboprophylaxis and outcome events was derived from the National Joint Registry (NJR) for England and Wales10 and the Hospital Episode Statistics (HES), the administrative database for all hospital admissions to the English National Health Service (NHS).11
We used multivariable logistic regression and propensity score matching to adjust for differences in the baseline risk of the patients who received LMWH or aspirin. Propensity score methods are increasingly being used to eliminate potential bias when observational studies are used to compare the effects of different treatments.12 Although in many cases regression and propensity score methods will produce similar results, propensity score methods are being advocated as more robust and less sensitive to model misspecification, especially when outcome events are rare.13
Patients and Methods
We used records from the NJR linked at patient level to records from the HES database. The NJR aims to collect data prospectively on all patients who undergo a hip or knee replacement in England and Wales.10 The HES database includes all patients admitted to NHS hospitals in England, or to private hospitals but funded by the NHS.11 HES records contain patient details and diagnostic information coded using the International Classification of Diseases, 10th revision (ICD-10),14 and operative procedure codes using the United Kingdom Office for Population Censuses and Surveys classification, 4th revision (OPCS-4).15 The HES database also includes information about date of death from the Office for National Statistics.
All 316 495 patients who were identified in the HES database as having undergone a primary THR in the English NHS between 1 April 2003 and 30 September 2008 were eligible for inclusion (Fig. 1). Of these patients, 170 274 (53.8%) could be linked to a record in the NJR. The linkage was carried out based on a hierarchy of deterministic linkage criteria, including NHS number, year of birth and gender. Details of these criteria have been published elsewhere.16 Of the 149 048 patients who had undergone a first primary THR and could be followed up for at least 90 days, we included only those 108 584 who had been given aspirin or LMWH as the only pharmacological thromboprophylaxis, irrespective of whether mechanical prophylaxis methods had also been used. Patients receiving more than one type of pharmacological prophylaxis were excluded.
Fig. 1
Flowchart describing patient inclusion and linkage between the National Joint Registry (NJR) and Hospital Episode Statistics (HES) data (LMWH, low-molecular-weight heparin).
Information regarding the method of thromboprophylaxis was extracted from NJR records along with relevant details of the operation and patient characteristics that were considered risk factors for thromboembolic events or haemorrhage. We used the Royal College of Surgeons Charlson Score17 to search the HES database for comorbid conditions.
HES data were also used to identify thromboembolic events (PE, DVT) and death within 90 days of surgery, major haemorrhage (cerebrovascular accident or gastrointestinal haemorrhage) and return to theatre for wound complications within 30 days of surgery. Minor haemorrhage was not considered as it is unlikely to be accurately coded in the HES database.
Multivariable risk model
Logistic regression was used to assess the effect of treatment on outcomes, with adjustment for patient characteristics that are risk factors for thromboembolism or haemorrhage. Results are presented as odds ratios (OR) with 95% confidence intervals (CI); ORs < 1 indicate that rates are lower with LMWH than with aspirin. P-values were based on the likelihood ratio test and a p-value < 0.05 was considered to indicate statistical significance. All risk models included the type of pharmacological prophylaxis, age (grouped into three equal-sized categories with cut-off values rounded to the nearest multiple of five), gender, American Society of Anaesthesiologists (ASA) grade,18 number of comorbid conditions according to the Charlson score, indication for surgery (osteoarthritis or other), whether regional anaesthesia (epidural or spinal procedures) was used, prosthesis type (cemented, cementless, hybrid or resurfacing), approach (anterior, anterolateral, lateral or posterior), use of mechanical thromboprophylaxis, and provider type (NHS hospital, NHS treatment centre, independent hospital or independent sector treatment centre).
Use of regional anaesthesia was the only variable with missing values. In order to address this we used multiple imputation by chained equations.19 Five datasets were created containing imputed values for missing values. Rubin’s rules were used to combine the estimation results based on each of these datasets.
Propensity score matching
Logistic regression was also used to estimate a propensity score, which represents the probability that patients were given aspirin or LMWH.20 The same variables that were included in the multivariable risk model were also candidates for inclusion in the propensity score model. In this case, however, we used information on the 14 individual comorbidities that are included in the Charlson score, rather than a simple count of the comorbid conditions. Instead of using multiple imputation, we included patients for whom information on the use of regional anaesthesia method was missing by grouping them into a separate category.
Following the approach recommended by Austin,20 we first used a stepwise process to select the variables for inclusion in the propensity score model based on statistical entry (p < 0.1) and removal (p > 0.2) criteria. Secondly, we added possible two-way interactions of the selected variables. The propensity score distributions in the two treatment groups were compared graphically and interactions were removed or re-entered in order to achieve similar distributions. Thirdly, for each of the 22 942 patients in the aspirin group we selected a patient from the LMWH group by matching on the logarithm of the estimated propensity score, using a nearest-neighbour matching algorithm with a maximum calliper width of 0.2 standard deviations (sd). All but two patients in the aspirin group could be matched. The balance in the variables across the treatment groups was assessed by calculating for each variable a standardised difference (i.e., difference between the groups divided by the sd among all patients). Balance was considered to be achieved if these differences were < 10%. Fourthly, logistic regression was performed on the set of matched pairs to estimate the effect of treatment on the outcomes. A robust standard error was used to allow for the clustering within the pairs.
Results
Of the 108 584 patients who could be included, 22 942 (21.1%) had been prescribed aspirin and 85 642 (78.9%) LMWH as the only pharmacological thromboprophylaxis. Patients in the aspirin group were more likely than those in the LMWH group to have a posterior approach and to receive mechanical prophylaxis (Table I). The two groups were otherwise very similar.
Table I
Characteristics of the patients and their treatment according to pharmacological prophylaxis with aspirin or low-molecular-weight heparin (LMWH)
| Characteristic | Aspirin (n = 22 942) | LMWH (n = 85 642) | LMWH (n = 22 940) matched based on propensity score |
|---|---|---|---|
| Age group (years) (n, %) | |||
| < 65 | 8236 (36) | 28 738 (34) | 8095 (35) |
| 65 to 74 | 8104 (35) | 31 028 (36) | 8151 (36) |
| ≥ 75 | 6602 (29) | 25 876 (30) | 6694 (29) |
| Gender (n, %) | |||
| Male | 9430 (41) | 34 174 (40) | 9356 (41) |
| Female | 13 512 (59) | 51 468 (60) | 13 584 (59) |
| ASA* grade (n, %) | |||
| 1 (healthy) | 5275 (23) | 16 956 (20) | 5265 (23) |
| 2 (mild systemic disease) | 14 666 (64) | 55 517 (65) | 14 607 (64) |
| 3 to 5 (severe systemic disease or worse) | 3001 (13) | 13 169 (15) | 3068 (13) |
| Number of comorbidities | |||
| 0 | 17 681 (77) | 66 216 (77) | 17 670 (77) |
| 1 | 4388 (19) | 15 975 (19) | 4473 (19) |
| ≥ 2 | 873 (4) | 3451 (4) | 797 (4) |
| Indication for surgery | |||
| Osteoarthritis | 21 414 (93) | 80 224 (94) | 21 459 (94) |
| Other | 1528 (7) | 5418 (6) | 1481 (6) |
| Regional anaesthesia | |||
| Neither epidural or spinal | 7786 (34) | 26 263 (31) | 7778 (34) |
| Epidural or spinal | 14 271 (62) | 53 380 (62) | 14 299 (62) |
| Missing | 885 (4) | 5999 (6) | 863 (4) |
| Prosthesis type | |||
| Cemented | 11 043 (48) | 42 664 (50) | 11 050 (48) |
| Uncemented | 6048 (26) | 26 807 (31) | 6046 (26) |
| Hybrid | 4014 (18) | 10 683 (13) | 4025 (18) |
| Resurfacing | 1837 (8) | 5488 (6) | 1819 (8) |
| Approach | |||
| Anterior/anterolateral/lateral | 10 523 (46) | 51 083 (60) | 10 538 (46) |
| Posterior | 12 419 (54) | 34 559 (40) | 12 402 (54) |
| Mechanical thromboprophylaxis | |||
| No | 4122 (18) | 24 033 (28) | 4106 (18) |
| Yes | 18 820 (82) | 61 609 (72) | 18 834 (82) |
| Provider type | |||
| NHS hospital/treatment centre | 21 545 (92) | 77 427 (90) | 21 576 (94) |
| Independent hospital/treatment centre | 1397 (8) | 8215 (10) | 1364 (6) |
-
* ASA, American Society of Anesthesiologists
Without adjustment for potential risk factors we did not find statistically significant differences in the rate of thromboembolic events, major haemorrhage or return to theatre between the aspirin and the LMWH group (Table II). The PE rate was the same in both groups (0.68%) and the 90-day mortality was 0.65% with aspirin and 0.61% with LMWH (unadjusted ORs 0.93). The differences in the other outcomes were also small, and all the unadjusted ORs varied around 1.
Table II
Effect of aspirin and low-molecular-weight heparin (LMWH) on outcomes with adjustment based on the multivariable risk model. An odds ratio < 1 indicates that rates are lower with LMWH than with aspirin (CI, confidence interval)
| Treatment group | Unadjusted odds ratio (95% CI) | Adjusted odds ratio (95% CI) | ||||
|---|---|---|---|---|---|---|
| Outcome (n, %) | Aspirin (n = 22 942) | LMWH (n = 85 642) | p-value | |||
| Pulmonary embolism | 156 (0.68) | 583 (0.68) | 1.00 (0.84 to 1.20) | 0.97 (0.81 to 1.17) | 0.78 | |
| Deep-vein thrombosis | 227 (0.99) | 806 (0.94) | 0.95 (0.82 to 1.10) | 0.91 (0.79 to 1.06) | 0.23 | |
| Death | 150 (0.65) | 520 (0.61) | 0.93 (0.77 to 1.11) | 0.84 (0.69 to 1.01) | 0.06 | |
| Cerebrovascular accident/ gastrointestinal haemorrhage | 176 (0.77) | 620 (0.72) | 0.94 (0.80 to 1.11) | 0.92 (0.77 to 1.09) | 0.34 | |
| Return to theatre | 71 (0.31) | 312 (0.36) | 1.17 (0.91 to 1.52) | 1.15 (0.88 to 1.50) | 0.29 | |
We found that with risk adjustment based on the multivariable risk model, the difference in 90-day mortality increased but did not quite reach statistical significance (adjusted OR 0.84; p = 0.06). The impact of the risk adjustment on the differences in the other outcomes was small.
There was close agreement between the characteristics of the 22 942 patients in the aspirin group and those in the equally sized LMWH group who were matched on their propensity score (Table I). All standardised differences were < 2%, which confirms that the matched groups were well balanced.
A comparison of the outcomes in these matched groups produced results that were comparable to those obtained with the multivariable risk model (Table III). However, the difference in the 90-day mortality (0.65% in the aspirin group and 0.51% in the LMWH group) was now slightly larger (matched OR 0.77; p = 0.04).
Table III
Effect of aspirin and low-molecular-weight heparin (LMWH) on outcomes with adjustment based on propensity score matching. An odds ratio < 1 indicates that rates are lower with LMWH than with aspirin (CI, confidence interval)
| Treatment group | Adjusted odds ratio (95% CI) | |||||
|---|---|---|---|---|---|---|
| Outcome (n, %) | Aspirin (n = 22 940) | LMWH (n = 22 940) | p-value | |||
| Pulmonary embolism | 156 (0.68) | 146 (0.64) | 0.94 (0.75 to 1.17) | 0.56 | ||
| Deep-vein thrombosis | 227 (0.99) | 193 (0.84) | 0.84 (0.70 to 1.03) | 0.10 | ||
| Death | 150 (0.65) | 116 (0.51) | 0.77 (0.61 to 0.98) | 0.04 | ||
| Cerebrovascular accident/ gastrointestinal haemorrhage | 176 (0.77) | 167 (0.73) | 0.95 (0.77 to 1.17) | 0.63 | ||
| Return to theatre | 71 (0.31) | 75 (0.33) | 1.17 (0.76 to 1.46) | 0.74 | ||
Discussion
This is the first published study large enough to compare meaningfully the effects of aspirin and LMWH on PE, major bleeding and mortality in patients undergoing THR. However, the study design was non-randomised, with its inherent risk of producing biased results. Two fundamentally different methods were used to adjust for baseline differences in patient characteristics. Both provided evidence, albeit weak, that mortality within 90 days of surgery may be lower with LMWH. The observed differences in the rates of PE, major bleeding and return to theatre were small and not statistically significant.
Methodological considerations
The potential for bias in the results of non-randomised comparisons is well recognised. A ‘review of reviews’ comparing the results of experimental and observational studies concluded that there are differences between the results of studies with randomised and non-randomised designs, but no consistent pattern indicating systematic bias.21 In order to explore the robustness of our risk-adjusted findings we contrasted multivariable risk modelling with propensity score matching.
The multivariable model included all available risk factors that were thought to be associated with the outcome, which allows the estimation of a treatment effect conditional on those risk factors. Propensity score matching, on the other hand, aims to balance the available patient characteristics that are associated with the assignment of treatment in order to replicate the design of an RCT, which produces a marginal treatment effect that should be interpreted as the mean difference at group level.20 The results of both approaches were equivalent, which provides further support for our results.
An additional argument supporting the validity of our results is that it is generally accepted that the choice of thromboprophylaxis depends on the standard policy of the hospital or the preference of the surgeon, rather than on the characteristics of the patients. This notion is in line with the observed similarity of the characteristics of the patients in the aspirin and the LMWH groups (Table I). However, a number of relevant risk factors, such as obesity and smoking, were not available to us and could not therefore be included in the risk adjustment. If, for example, orthopaedic surgeons were more likely to prescribe LMWH to patients with a higher body mass index or a history of smoking or other relevant risk factors, our results would underestimate the advantage of LMWH.
The outcomes were identified through linkage of the NJR data with the HES database. We were only able to link 54% of the patients who had undergone a THR in the NHS according to the HES database to an NJR record. Furthermore, the limitations of the ICD-10 and OPCS-4 coding systems are well recognised.22 In combination, the linkage and coding issues will have caused errors in the assessment of the outcomes in a number of patients. These errors will have led to an underestimation of the event rates (given that missing an event is more likely than erroneously identifying one) as well as an underestimation of the size of the difference in the effect of aspirin and LMWH (given that the degree of misclassification is expected to be the same in the LMWH and aspirin groups). The effects of linkage and coding issues are especially relevant for DVT, which is difficult to diagnose without the systematic use of imaging techniques, as well as for PE and haemorrhage. Mortality, on the other hand, is a well-defined outcome and therefore least likely to be affected by coding inaccuracies.
As an aside, given these coding errors and their probable different impacts on DVT, PE and death, we do not feel that our results can help to determine whether reduction in the incidence of DVT is linked to a reduction in PE and death. This is a fundamental question that lies at the centre of the conflict of the guidelines that recommend aspirin and those that do not.5,23
Comparison with other studies
Two questions need to be addressed: how do the rates of thromboembolic events and haemorrhage observed in our study compare to those found by others and how do the effects of aspirin and LMWH reported by us correspond to the available evidence? Irrespective of whether aspirin or LMWH was used, the DVT rates presented in our study are much lower than those reported in RCTs. For example, the systematic review of RCTs of thromboprophylaxis treatment carried out by the developers of the NICE guidelines found that DVT rates varied from 34% to 50% with aspirin and from 5% to 38% with LMWH.1 This is because we could only identify DVTs that were coded as a diagnosis in the HES database, and not the asymptomatic DVTs that are used as endpoints in the RCTs. However, PE rates according to the systematic review carried out by NICE are similar to those in our study, ranging from 0% to 1% with aspirin and from 0% to 7.8% with LMWH.1 Furthermore, the combined rates of DVT and PE presented in our study are comparable to those reported by the Scottish Arthroplasty Project,24 which found that between 2003 and 2009 this rate varied between 1.3% and 1.8% for all primary THRs.
The only four RCTs so far that have directly compared the effects of aspirin and LMWH suggest that DVT rates are lower with LMWH: three of the four studies show a statistically significant reduction.9 As there were no RCTs directly comparing the effects of aspirin and LMWH on PE and death, the developers of the NICE guidelines carried out network meta-analyses, which allowed an indirect assessment of the relative effectiveness of aspirin and LWMH, as both have been compared against placebo.25 The results of these meta-analyses are not easy to interpret. For example, the results are not always consistent over possible combinations with mechanical prophylaxis. Furthermore, the differences between the relative risks of aspirin and LMWH seem small and not statistically significant: the relative risk of PE with high-dose aspirin compared with no prophylaxis is 0.16 (95% CI 0.02 to 1.12) and the relative risk with LMWH is 0.12 (95% CI 0.02 to 0.59). On the other hand, the network meta-analysis suggests that aspirin does not increase the risk of haemorrhage, whereas LMWH would.
Implications
Given our finding that mortality is lower with LMWH than with aspirin, the results of our study could simply be interpreted as further evidence that LMWH should be used as pharmacological thromboprophylaxis after THR. We feel that this would not acknowledge the complexity of the decision that needs to take into account the full range of possible outcomes. For example, we could not investigate the frequency of minor bleeding and wound problems as they cannot be identified in HES data unless they lead to return to theatre or readmission.
We recommend that the new evidence provided by our study on the effects of aspirin and LMWH is used by the developers of the existing guidelines when they revisit their recommendations. We expect that the low observed rates of PE and death irrespective of whether aspirin or LMWH was used will have an impact on the relative cost-effectiveness of the various options.
Our results are important for those who are contemplating setting up an RCT to compare aspirin with LMWH with PE and death as primary outcomes after THR. For example, an RCT that aims to detect a 20% reduction in the PE rate (from 1.0% to 0.8%) with a power of 80% and a two-sided significance level of 0.05 needs to include about 36 000 patients in each group. If death were to be used as outcome the size of the RCT to detect a similar reduction would need to be even larger. These parameters could be included in a formal expected-value-of-information analysis that should be carried out before embarking on such a large RCT to ensure that the costs of the trial do not exceed the benefits that can be expected from the information that it might provide.26
Our study also highlights the potential of the linkage between the NJR and HES data. The baseline risks of venous thromboembolic events and haemorrhage are important factors determining the balance between risks and benefits of the different prophylaxis strategies. The linked data would allow the development of a series of risk models that could quantify these baseline risks according to patient characteristics, including body mass and comorbidity. Linkage with other databases, including those containing information about primary care as well as the large database which is currently being built up in England and Wales containing pre- and post-operative measures reported by patients, would allow a further refinement of these risk models.27
1 No authors listed. National Institute for Health and Clinical Excellence. Venous thromboembolism: reducing the risk, 2010. http://www.nice.org.uk/nicemedia/live/12695/47920/47920.pdf (date last accessed 27 July 2011). Google Scholar
2 Patel VP , WalshM, SehgalB, et al.Factors associated with prolonged wound drainage after primary total hip and knee arthroplasty. J Bone Joint Surg [Am]2007;89-A:33–38. Google Scholar
3 Saleh K , OlsonM, ResigS, et al.Predictors of wound infection in hip and knee joint replacement: results from a 20 year surveillance program. J Orthop Res2002;20:506–515. Google Scholar
4 Streiff MB , HautER. The CMS ruling on venous thromboembolism after total knee or hip arthroplasty: weighing risks and benefits. JAMA2009;301:1063–1065. Google Scholar
5 Struijk-Mulder MC , EttemaHB, VerheyenCC, BüllerHR. Comparing consensus guidelines on thromboprophylaxis in orthopaedic surgery. J Thromb Haemost2010;8:678–683. Google Scholar
6 Geerts WH , BergqvistD, PineoGF, et al.Prevention of venous thromboembolism: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (8th Edition). Chest2008;133(Suppl):381–453. Google Scholar
7 No authors listed. American Academy of Orthopaedic Surgeons. Clinical guideline on prevention of symptomatic pulmonary embolism in patients undergoing total hip or knee arthroplasty, 2007. www.aaos.org/research/guidelines/PE_guideline.pdf (date last accessed 27 July 2011). Google Scholar
8 Emsley D, Martin J, Newell C, et al. National Joint Registry for England and Wales: 5th annual report, 2008. http://www.njrcentre.org.uk/NjrCentre/LinkClick.aspx?fileticket=Da4%2b2sUVa%2fI%3d& tabid=86& mid=523 (date last accessed 27 July 2011). Google Scholar
9 Karthikeyan G , EikelboomJW, TurpieAGG, HirshJ. Does acetyl salicylic acid (ASA) have a role in the prevention of venous thromboembolism?Br J Haematol2009;146:142–149. Google Scholar
10 No authors listed. National Joint Registry for England and Wales. http://www.njrcentre.org.uk (date last accessed 27 July 2011). Google Scholar
11 No authors listed. The Health and Social Care Information Centre. Hospital episode statistics, 2011. http://www.hesonline.nhs.uk/Ease/servlet/ContentServer?siteID=1937 (date last accessed 27 July 2011). Google Scholar
12 Rosenbaum PR , RubinDB. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc1984;79:516–524. Google Scholar
13 Joffe MM , RosenbaumPR. Invited commentary: propensity scores. Am J Epidemiol1999;150:327–333. Google Scholar
14 World Health Organization. International Classification of Diseases (10th revision). WHO: Geneva, 1994. http://www.who.int/classifications/icd/en/ (date last accessed 13 September 2011). Google Scholar
15 No authors listed. NHS Connecting for Health. Office of Population, Censuses and Surveys Classification of Surgical Operations and Procedures (4th revision). NHS: London, 1987. http://www.connectingforhealth.nhs.uk/systemsandservices/data/clinicalcoding/codingstandards/opcs4 (date last accessed 13 September 2011). Google Scholar
16 Ellams D, Forsyth O, Mistry A, et al. National Joint Registry for England and Wales: 7th annual report, 2010. http://www.njrcentre.org.uk/NjrCentre/LinkClick.aspx?fileticket=QkPI7kk6B2E%3d& tabid=86& mid=523 (date last accessed 27 July 2011). Google Scholar
17 Armitage JN , van der MeulenJH; RCS Comorbidity Consensus Group. Identifying co-morbidity in surgical patients using administrative data with the Royal College of Surgeons Charlson Score. Br J Surg2010;97:772–781. Google Scholar
18 Dripps RD . New classification of physical status. Anesthesiology1963;24:111. Google Scholar
19 Royston P , CarlinJB, WhiteIR. Multiple imputation of missing values: new features for mim. Stata J2009;9:252–264. Google Scholar
20 PC Austin. The performance of different propensity score methods for estimating the marginal odds ratio Stat Med2007;26:3078–3094. Google Scholar
21 Deeks JJ , DinnesJ, D’AmicoR, et al.Evaluating non-randomised intervention studies. Health Technol Assess2003;7:1–173. Google Scholar
22 Raferty J , RoderickP, StevensA. Potential use of routine databases in health technology assessment. Health Technol Assess2005;9:1–92. Google Scholar
23 Eikelboom JW , KarthikeyanG, FagelN, HirshJ. American Association of Orthopedic Surgeons and American College of Chest Physicians guidelines for venous thromboembolism prevention in hip and knee arthroplasty differ: what are the implications for clinicians and patients?Chest2009;135:513–520. Google Scholar
24 No authors listed. Scottish Arthroplasty Project: annual report, 2009. http://www.arthro.scot.nhs.uk/Reports/Scottish_Arthroplasty_Project_Report_2009.pdf (date last accessed 27 July 2011). Google Scholar
25 Lumley T . Network meta-analysis for indirect treatment comparisons. Stat Med2002;21:2313–2324. Google Scholar
26 Griffin S , WeltonNJ, ClaxtonK. Exploring the Research Decision Space: the expected value of information for sequential research designs. Med Decis Making2010;30:155–162. Google Scholar
27 No authors listed. Guidance on the routine collection of Patient Reported Outcome Measures, 2009. http://www.dh.gov.uk/en/Publicationsandstatistics/Publications/PublicationsPolicyAndGuidance/DH_092647 (date last accessed 27 July 2011). Google Scholar
No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.
We thank the staff of all the hospitals in England and Wales who contributed data to the National Joint Registry. We also thank the staff of the NJR Centre for the collection of the data and for their support of the data linkage and analysis.
The National Joint Registry for England and Wales is funded through a levy raised on the sale of hip and knee replacement implants. The cost of the levy is set by the NJR Steering Committee. The NJR Steering Committee is responsible for study design and data collection. The authors are responsible for the data analysis reported in this paper, the preparation of the manuscript, and the decision to publish.
No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.
Supplementary material. A table showing the International Classification of Diseases (10th edition; ICD-10) and Office of Population, Censuses and Surveys Classification of Surgical Operations and Procedures (4th revision; OPCS-4) codes used for the analysis of Hospital Episode Statistics data is available with the electronic version of this paper on our website at www.jbjs.org.uk
