Prediction of Treatment-Related Mortality in Patients with Relapsed and Refractory Acute Myeloid Leukemia

Introduction

Despite improvements in supportive care, treatment-related mortality (TRM) remains a significant problem for patients with acute myeloid leukemia (AML). In order to quantify fitness for intensive AML therapies, we recently developed a multivariate model for predicting TRM, defined as death within 28 days of treatment initiation in patients undergoing intensive induction chemotherapy for newly-diagnosed AML. Here, we examine the performance of this TRM model in patients with relapsed or refractory AML and assess whether the model can be improved for this population.

Methods

Using a database of patients treated for AML at our institution since 2003, we identified patients with relapsed or refractory disease treated with intensive chemotherapy, defined as regimens that were at least as intense as typical “7+3” regimens; patients who received treatment protocols with cytarabine <100 mg/m²/day or demethylating agents alone were excluded. We collected the following variables required to calculate the TRM score: ECOG performance status, age, platelet count, albumin, presence or absence of secondary AML, WBC, peripheral blood blast percentage, and creatinine. TRM scores were then calculated as described in J Clin Oncol 2011;29:4417. For a subset of these patients, the following additional variables were collected: duration of prior complete remission, number of prior chemotherapy regimens, number of prior hematopoietic cell transplants, and the number of antibiotics prescribed for treatment of presumed or documented infections at the time of chemotherapy initiation. For this smaller subset, we used a logistic regression model to create a preliminary updated TRM model that included these clinical variables in addition to the variables in the original TRM score. Finally, we used the area under the receiver operator characteristic curve (AUC) to quantify the ability of a model to predict TRM; in this approach, an AUC of 1 indicates perfect prediction of TRM while an AUC of 0.5 indicates no prediction; AUC values of 0.6-0.7, 0.7-0.8, and 0.8-0.9 are commonly considered as poor, fair, and good, respectively.

Results

A total of 270 patients met our study inclusion criteria. Fifteen (5.6%) died within 28 days of starting chemotherapy, i.e. experienced TRM. The AUC for the previously published TRM score in predicting early death in this population was 0.66. For comparison, in our original study of 2,238 patients with newly-diagnosed AML, we obtained an AUC of 0.82. Expanding the definition of TRM to include those patients who died within 60 (N = 47, 17.4%) or 90 days (N = 75, 27.8%) of starting chemotherapy did not improve the original model’s performance in our cohort (AUCs of 0.62 and 0.59, respectively). The additional variables described above were available for 133 of the 270 patients in our cohort. Within this subset, the AUC for the original TRM score was 0.65 in predicting death within 28 days. Incorporation of these new covariates in a preliminary TRM model yielded an improved AUC of 0.81 for this smaller subset.

Conclusions

Our data indicate that the TRM score, originally developed for patients with newly-diagnosed AML, only has a fair ability to predict TRM in patients with relapsed and refractory AML. However, with inclusion of additional covariates, such as prior CR duration and number of prior chemotherapies, TRM can be predicted quite well, to a degree similar to the accuracy demonstrated by the original model in newly-diagnosed patients. While our findings need to be confirmed in larger, independent cohorts of patients, they suggest the possibility of developing an objective measure to determine fitness for intensive salvage chemotherapy in AML.