Re: The relation between total joint arthroplasty and risk for serious cardiovascular events in patients with moderate-severe osteoarthritis: propensity score matched landmark analysis
To the Editor:
We read with interest the online responses to our recent publication “The relationship between total joint arthroplasty and risk for serious cardiovascular events in patients with moderate-severe osteoarthritis: a propensity score matched landmark analysis.” We appreciate the thoughtful comments and wish to address some of the issues raised.
In her letter to the editor, Dr. Judy Stone correctly points out that we did not consider the impact of peri-operative complications on total joint arthroplasty (TJA) recipients in our study. These complications include venous thrombo-embolic events(1, 2), joint infection(3-5), and early revision arthroplasty(6). While the occurrence of these complications is a source of considerable morbidity, they have a relative low incidence. A recent report from our group on the outcomes of TJA performed in Ontario, Canada between 2002-2009 estimated the incidence of these complications at <1.4%(7).
In their letter, Drs. Mnatzaganian and Hiller used data from the Health in Men Study (HIMS) to demonstrate that persons with osteoarthritis (OA) who receive a TJA have an improved mortality rate relative to persons with OA that did not receive a TJA, or persons without OA. It is their contention that this demonstrates that our findings may actually be a reflection of confounding by indication, or a healthy-user bias, and do not reflect a true beneficial effect of TJA. While we agree that propensity score techniques do not account for unobserved bias, we believe that their analysis was flawed for several reasons, outlined below.
First, and perhaps most importantly, by including control (non-TJA) subjects who did not have OA in their study sample, Mnatzaganian and Hiller risk introducing substantial bias into their estimates, with the likely direction to be in favour of TJA recipients. An estimate of treatment effect using propensity score methods requires that all of the subjects in the sample be potential candidates for the treatment under question (Rosenbaum & Rubin, Biometrika 1983). By including people without OA, who are therefore unlikely to be candidates for TJA, this assumption is violated (indeed, when using any statistical technique to estimate the effect of an intervention, it is advisable to exclude subjects who are not indicated for the intervention). Further, in their study, OA was defined solely based on patient self-report. In contrast, our study used a true population cohort of potential candidates for TJA (9); all had moderate to severe hip or knee OA confirmed both clinically and radiographically.
Second, their analysis employed stratification on the propensity score, which has been shown to introduce a greater degree of bias compared to propensity-score matching which was used in our study(8).
Third, they used proportional hazards regression to estimate the propensity score, rather than using a regression method for binary outcomes (e.g. logistic regression), as is more typically done. It is not entirely clear what affect the use of a proportional hazards model to estimate the propensity score would have had on their analyses. However, it may be that they are estimating the probability of receipt of TJA within an unspecified duration of follow-up time. Greater details on the actual analysis conducted are required for a closer evaluation of their results.
Fourth, while we acknowledge that propensity score techniques only account for observed confounders, the expansive and thorough nature of our baseline questionnaire, as well as its linkage with a number of health administrative databases, allowed us to incorporate a multitude of factors into our propensity score. These factors included: socio-demographics (age, sex, body-mass-index, living arrangements, education, annual household income), health status (number of co-morbidities, SF-36 general health score, pre-existing cardiovascular disease, pre-existing depression, smoking status, use of NSAIDs), and arthritis severity (WOMAC summary score, presence of troublesome hips and knees). All of these factors are important considerations for receipt of a TJA, and are also determinants for our outcome of interest. However, the vast majority of these measures were not available to Drs. Mnatzaganian and Hiller, including living arrangements, body-mass index, smoking status, and arthritis severity. Furthermore, we excluded persons who died or had a serious cardiovascular event before the landmark date. As these persons were less likely to be offered TJA, their inclusion would have biased the results in favour of arthroplasty; by excluding them, our estimates of the benefits of TJA are conservative. In their study, Drs. Mnatzaganian and Hiller did not perform a landmark analysis, and as such are more likely to find a beneficial effect from TJA.
Nevertheless, we considered the possibility that our results reflected an unmeasured confounder that was not balanced between groups. As described in our report, we determined that an unmeasured confounder would have to have had a prevalence of at least 75% in one group, while being completely absent from the other group, with a relative risk ratio for the outcome of at least 0.65 (if found only among TJA recipients) or 1.50 (if found only among non-TJA recipients) to account for the observed TJA effect(16). In addition, this confounder would have had to be non-collinear with the other variables included in our propensity score. We believe it is unlikely that a confounder that satisfies these extreme properties is not known and was not measured.
We understand that the results of our analysis are provocative, and we believe that in addition to the need for confirmation in further studies, they need to be interpreted and acted on with caution. The potential cardio-protective benefit must be weighed against the very real risk for peri-operative morbidity and mortality in this patient population.
1. Merli G. Venous thromboembolism after total hip arthroplasty and total knee arthroplasty: current and future perspectives. American Journal of Managed Care. [Introductory]. 2011 Feb;17(1 Suppl):S4-5.
2. Ollendorf DA, Vera-Llonch M, Oster G. Cost of venous thromboembolism following major orthopedic surgery in hospitalized patients. American Journal of Health-System Pharmacy. [Comparative Study Research Support, Non-U.S. Gov't]. 2002 Sep 15;59(18):1750-4.
3. Clyburn TA CQ. Antibiotic laden cement: Current state of the art. AAOS Now. 2013;7(6)(Available online: http://www.aaos.org/news/bulletin/may07/clinical7.asp).
4. Cummins JS, Tomek IM, Kantor SR, Furnes O, Engesaeter LB, Finlayson SRG. Cost-effectiveness of antibiotic-impregnated bone cement used in primary total hip arthroplasty. Journal of Bone & Joint Surgery - American Volume. [Comparative Study]. 2009 Mar 1;91(3):634-41.
5. Joseph TN, Chen AL, Di Cesare PE. Use of antibiotic-impregnated cement in total joint arthroplasty. Journal of the American Academy of Orthopaedic Surgeons. [Review]. 2003 Jan-Feb;11(1):38-47.
6. Bozic KJ, Katz P, Cisternas M, Ono L, Ries MD, Showstack J. Hospital resource utilization for primary and revision total hip arthroplasty. Journal of Bone & Joint Surgery - American Volume. [Comparative Study Research Support, Non-U.S. Gov't]. 2005 Mar;87(3):570-6.
7. Ravi B, Croxford R, Hollands S, Bogoch E, Paterson M, Kreder H, et al. Patients with rheumatoid arthritis are at increased risk for complications following total joint arthroplasty. Arthritis & Rheumatism. 2013 (in press).
8. Austin PC. The performance of different propensity-score methods for estimating marginal hazard ratios. Statistics in Medicine. 2013;32(16):2837-2849. DOI: 10.1002/sim.5705.
9. Hawker GA, Wright JG, Coyte PC, Williams JI, Harvey B, Glazier R, et al. Differences between men and women in the rate of use of hip and knee arthroplasty. N Engl J Med. [Comparative Study Research Support, Non-U.S. Gov't]. 2000 Apr 6;342(14):1016-22.
10. Hawker GA, Guan J, Croxford R, Coyte PC, Glazier RH, Harvey BJ, et al. A prospective population-based study of the predictors of undergoing total joint arthroplasty. Arthritis Rheum. [Research Support, Non-U.S. Gov't]. 2006 Oct;54(10):3212-20.
11. Hawker GA, Wright JG, Badley EM, Coyte PC. Perceptions of, and willingness to consider, total joint arthroplasty in a population-based cohort of individuals with disabling hip and knee arthritis. Arthritis & Rheumatism. [Research Support, Non-U.S. Gov't]. 2004 Aug 15;51(4):635-41.
12. Hawker GA, Wright JG, Coyte PC, Williams JI, Harvey B, Glazier R, et al. Determining the need for hip and knee arthroplasty: the role of clinical severity and patients' preferences. Medical Care. [Research Support, Non-U.S. Gov't]. 2001 Mar;39(3):206-16.
13. Nuesch E, Dieppe P, Reichenbach S, Williams S, Iff S, Juni P. All cause and disease specific mortality in patients with knee or hip osteoarthritis: population based cohort study. Bmj. [Research Support, Non-U.S. Gov't]. 2011;342:d1165.
14. Hofler M. The effect of misclassification on the estimation of association: a review. International Journal of Methods in Psychiatric Research. [Review]. 2005;14(2):92-101.
15. Kristensen P. Bias from nondifferential but dependent misclassification of exposure and outcome. Epidemiology. 1992 May;3(3):210-5.
16. Schneeweiss S. Sensitivity analysis and external adjustment for unmeasured confounders in epidemiologic database studies of therapeutics. Pharmacoepidemiology & Drug Safety. [Research Support, N.I.H., Extramural]. 2006 May;15(5):291-303.
Competing interests: No competing interests