Re: The relation between total joint arthroplasty and risk for serious cardiovascular events in patients with moderate-severe osteoarthritis: propensity score matched landmark analysis
Selection bias and confounding in total joint arthroplasty
George Mnatzaganian(1), Janet E Hiller(1,2)
1.Faculty of Health Sciences, Australian Catholic University, East Melbourne, Victoria, Australia
2.Discipline of Public Health, School of Population Health, University of Adelaide, South Australia, Australia
To the Editor:
We read with interest the propensity score matched landmark study by Ravi et al(1) showing that elective hip or knee arthroplasty was associated with a significant reduction in subsequent cardiovascular events in a relatively small sample of patients with osteoarthritis. The authors argued that these findings could be explained by the improved physical activity in patients undergoing arthroplasty and also by the anticipated reduced use of potentially cardio-toxic non-steroidal anti-inflammatory drugs. We argue that these findings may have other explanations.
Matching by a propensity score (PS) may provide some degree of balance in baseline characteristics between those who have and do not have an elective arthroplasty. Nonetheless, such propensity scores only account for observed covariates. Unobserved factors that influence selection of patients for this procedure cannot be accounted for.(2) Concerns with matching on observed variables have been raised by Judea Pearl, who has argued that “hidden” bias may actually increase because of unobserved confounders.(3) In clinical settings, various factors, not accounted for in the propensity score, may influence how a clinical decision is made or how a treatment strategy is planned.
Ravi’s results may be explained by a healthy cohort effect; that is, healthy patients are initially selected for this elective procedure which in turn ensures better future outcomes. Factors associated with this selection process are not accounted for in the propensity score.
Also Ravi et al findings were based on a relatively small sample – 153 matched pairs of participants – which is a limitation in such study designs since propensity score analyses require large samples.(2,3)
Informed by Ravi et al methods, we constructed a propensity score that predicted the risk of undergoing a total joint replacement (TJR) in a large population-based cohort of 12,000 men belonging to the Health In Men Study.(4,5) In a retrospective cohort analysis, we used Cox proportional hazards regression to construct the score that was based on: age, body weight, height, duration of smoking, physical exercise, socioeconomic disadvantage index, and Elixhauser’s 30 co-morbidities.(4) Instead of assessing cardiovascular events, we investigated all-cause ten-year mortality. We hypothesised that patients with osteoarthritis with TJR would not only significantly die less than those with osteoarthritis without TJR but also than those without osteoarthritis and without this procedure – indicating that these patients were initially healthier. When the study participants were stratified according to quintiles of the distribution of this score, the differences in patient characteristics between those who had and did not have TJR decreased (results not shown). Nonetheless, patients with TJR were significantly less likely to die than those without a TJR in both those with and without osteoarthritis (Table 1). This confirms our hypothesis that these patients were initially healthier than the rest of the cohort and propensity score stratification and matching cannot account for the hidden selection biases and confounding by unknown factors as discussed above. We conclude that more research is required to better understand the pathways for selection of patients for total joint arthroplasty.
Dr. George Mnatzaganian, Prof Janet E Hiller
1. Ravi B, Croxford R, Austin PC, Lipscombe L, Bierman AS, Harvet PJ, et al. The relation between total joint arthroplasty and risk for serious cardiovascular events in patients with moderate-severe osteoarthritis: propensity score matched landmark analysis. BMJ 2013; 347:f6187.
2. Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc 1984; 79:516-24.
3. Pearl, J. Understanding propensity scores. In Causality: Models, Reasoning, and Inference, Cambridge University Press, Second Edition, 2009.
4. Mnatzaganian G, Ryan P, Norman PE, Davidson DC, Hiller JE. Smoking, body weight, physical exercise, and risk of lower limb total joint replacement in a population-based cohort of men. Arthritis Rheum 2011; 63:2523-2530.
5. Mnatzaganian G, Ryan P, Norman PE, Davidson DC, Hiller JE. A propensity score for predicting major adverse outcomes after total joint replacement in men. J Epidemiol Community Health 2011; 65 (Suppl 1), A275-A275.
Competing interests: No competing interests