Effect of Age Adjustment in Predicting Outcome
To the Editor:
We read with interest the report of Ling et al1 that examined long-term outcome in a cohort of patients in whom pure mitral regurgitation due to flail leaflet was first diagnosed by echocardiography between 1980 and 1989.
In the introduction, the authors correctly state that it is uncertain whether early surgery or conservative management should be the preferred approach in these patients, irrespective of symptoms. Accordingly, they analyzed their cohort on an “intention-to-treat” basis, comparing follow-up events among patients who underwent surgery within 1 month after the diagnosis (the early-surgery group) with those who did not (the conservative-treatment group). Because this was not a randomized trial, the validity of the study is threatened by incomparability of prognosis of both treatment groups at baseline (ie, confounding).
Age is always an extremely important potential confounder. When the baseline characteristics of the patients at the time of diagnosis are compared (Table 1 in the article), there is an important difference in mean age between the early-surgery group and the conservative-treatment group (61.1 versus 66.5 years), which is called “slight” by the authors. The estimated probability of 10-year survival for 61- and 67-year-old men (at approximately 1990 in The Netherlands) was 77.7% and 63.7% respectively, a difference of 14%. The same difference (14% in 10-year survival) was found between early surgery and conservative treatment in the present study. The horizontal distance between the 2 groups in Figure 1 in the article is ≈4 years; this means that patients in the early-surgery group lived ≈4 years longer. If we look again at the Dutch general population in 1990, 61-year-olds lived 4 years longer than 67-year-olds. Thus, the expected impact of the age factor alone is of the same magnitude as the survival difference presented here as a result of treatment. It is unlikely that all age-related confounding is eliminated by including age as a linear covariate in a multivariate Cox model, if the data originate from an elderly population followed up over a long period of time. After all, the use of a linear component in the Cox model fails to correct for the extremely nonlinear (almost exponential) increase of “background mortality” (ie, the expected mortality in the source population) in higher-age groups. Thus, the baseline characteristic of age was biased in favor of survival in the early-surgery group, and it is doubtful whether this bias has been sufficiently removed.
Although there are more effective ways to deal with differences in age and associated expected mortality between 2 groups of patients,2 we agree with the authors that the only way to solve the issue of timing of surgery in asymptomatic patients with pure mitral regurgitation due to flail leaflets and no signs of left ventricular dysfunction is to perform a randomized, controlled trial.
- Copyright © 1998 by American Heart Association
We appreciate the interest of R.B.A. van den Brink et al in our recent publication on early surgery in patients with mitral regurgitation due to flail leaflets.R1 They raise an important question regarding the adequacy of our age adjustment and point to expected survival data relative to a 5-year age difference, suggesting that our “linear” age term in the Cox model does not adequately account for the increasing mortality at older ages. Several points need to be made in response to this concern:
First, it is not at all clear how relevant expected survival data for entire living populations are to the question of how to adjust survival data for a specific disease. Although comparison with expected survival data is useful for large populations, in particular of young patients, it is more difficult to interpret in older patients for several reasons.
The symptoms in the general “elderly” population of comparison are unknown, but it is a common observation that a notable percentage of this population expresses limitation in exercise capacity. Physical activity and exercise capacity are associated with long-term mortality,R2 so that the exact expected survival of patients may vary according to their symptoms. Therefore, it is possible that the expected survival of mostly asymptomatic and elderly patients, such as those in the conservative treatment group, may well be far better than the “standard” expected survival.
Comorbidity is an important determinant of survivalR3 but is indeterminable in the general population. Comorbidity increases with age but cannot be adjusted for in the comparison to expected survival. Indeed, a population with comorbidity lower than the general population may have an expected survival that is better than “standard.” Conversely, the results of our study held true after adjustment in multivariate analysis for comorbidity.
The subset of 50-year-olds who have a disease may be a very different subset from the subset of 70-year olds who have the same disease. Indeed, if we had large enough populations of patients with a disease and without surgical intervention, we would almost certainly want to use these populations as the reference populations for calculating the appropriate age adjustments. Thus, internal age adjustment is not to be lightly dismissed in favor of the use of general external populations.
Second, the point may have been missed that, although age was biased in favor of the early-surgery group, the very important prognostic factor of symptoms was weighted against the early-surgery group. Because we appropriately controlled for age but also for the baseline predictors of outcome,R4 a very sizable narrowing due to age adjustment was strongly counteracted by the adjustment for these predictors of outcome.
Third, it is incorrect to say that the proportional hazards model with a linear age term makes a linear adjustment for age. Despite the term “linear,” age enters into the Cox model in an exponential way. In our fitted model, every year of age was associated with a multiplicative factor of 1.1 in the hazard function. This factor is compounded, so that at a 5.4-year age difference, the factor is 1.67, or 67% higher, and at 10 years, it is 2.8, ie, a 3-fold higher risk. If we apply the 2 modes of correction (Cox model and expected survival based on the 1980 US white population) to the early-surgery group, a 5.4-year age difference decreases the expected 10-year survival from 79% to 67% using our Cox model and to 65% using the expected survival, demonstrating the magnitude of the age adjustment used in our model and the fact that these 2 modes of correction produce very similar results. Therefore, the association of outcome with treatment strategy cannot be attributed to our approach to age adjustment.
van den Brink et al correctly echo our statement that only a randomized trial can give a completely reliable answer to the question of the impact of early surgery on the outcome of severe mitral regurgitation. However, we believe that the results of our observational study suggest strongly that early surgery provides an improved outcome, and these results represent a first step in the consideration of this option for patients with severe mitral regurgitation. The considerable recent improvements in surgical resultsR5 R6 R7 make this option clinically viable until a randomized trial is completed.
Ling LH, Enriquez-Sarano M, Seward JB, Orszulak TA, Schaff HV, Bailey KR, Tajik AJ, Frye RL. Early surgery in patients with mitral regurgitation due to flail leaflets: a long-term outcome study. Circulation. 1997;96:1819–1825.
Enriquez-Sarano M, Tajik A, Schaff H, Orszulak T, Bailey K, Frye R. Echocardiographic prediction of survival after surgical correction of organic mitral regurgitation. Circulation. 1994;90:830–837.
Enriquez-Sarano M, Schaff H, Orszulak T, Tajik A, Bailey K, Frye R. Valve repair improves the outcome of surgery for mitral regurgitation. Circulation. 1995;91:1264–1265.