Quantitative Effect of Age In Predicting Empirically-Defined Treatment-Related Mortality and Resistance In Newly Diagnosed AML: Case Against Age Alone as Primary Determinant of Treatment Assignment

Abstract

Abstract Abstract 2191 Background: Treatment protocols for newly diagnosed AML typically use age (often 60 years) alone to restrict eligibility to either younger or older patients. Implied in this practice is the assumption that age is the principal predictor of therapeutic failure in AML due to either early treatment-related mortality (TRM) or resistance to therapy in patients who do not incur TRM. Yet, clinical observation and previous studies indicate that other prognostic factors modulate the effect of age on TRM and resistance, suggesting that age as sole or primary criterion for treatment allocation may be suboptimal. Methods: To test this hypothesis in newly-diagnosed non-APL AML, we quantified the relative effects of age and other covariates using 1,127 patients (median age: 57 years) treated on Southwest Oncology Group (SWOG) trials from 1986–2009 and 1,604 patients (median age: 61 years) treated on various protocols at M.D. Anderson Cancer Center (MDA) from 2000–2008. We calculated weekly hazard rates (the probability of death in week × given that the patient was alive at the beginning of the week) for both cohorts overall and in various age subgroups. We used the area under the receiver operator characteristic curve (AUC) to quantify the effects of covariates for prediction of TRM and resistance (no TRM but patient does not enter CR or relapses within 1 year of CR), where an AUC of 1 indicates that a covariate is perfect at prediction while an AUC of 0.5 indicates no prediction (i.e. it is no better than flipping a coin). Results: Despite the use of different treatment protocols, survival in the SWOG and MDA cohorts was virtually superimposable. In both cohorts, the maximum weekly hazard occurred at weeks 3 and 4 from start of treatment, after which it decreased. The maximum hazard was relatively independent of age and remained between weeks 3 and 5 in patients age <60 years, age 60–70 years, and age >70 years, respectively. The existence of such a discrete cut-point suggested that patients who die early are qualitatively distinct and prompted us to examine the relative effect of age and other covariates in patients who (a) died within the first 30 days of treatment (our empirically-based definition of TRM, 9% of MDA and 12 % of SWOG patients, respectively) and (b) survived at least 30 days but did not enter complete remission or relapsed within 1 year (“resistant”, 43% of MDA and 59% of SWOG patients, respectively). A model including age alone to predict early mortality had an AUC of 0.67, while a model including performance status (PS) alone had an AUC of 0.72. By comparison, a more refined model hat included PS, age, platelet count, cytogenetics, secondary AML, albumin, white blood cell count, peripheral blood blast count, and LDH yielded an AUC of 0.86. Elimination of age resulted in a model with an AUC that was only minimally lower (0.85). Prediction of resistance was more difficult with a model including age, secondary AML, cytogenetics, peripheral blood blasts, race, hemoglobin, and marrow neutrophils giving an AUC of only 0.70. Elimination of age had little effect (AUC 0.67) while age alone gave an AUC of 0.64. Conclusion: Age alone appears inadequate in predicting resistance and particularly TRM. The use of models based on several covariates improves predictive ability, but the ability to predict resistance is still limited, suggesting the value of randomized trials to assess treatment designed to reduce resistance. The observation that elimination of age has little effect on the predictive ability of such models, suggests that age is primarily a surrogate for other covariates. Disclosures: No relevant conflicts of interest to declare.