Prognostic Modeling to Predict Outcomes in Older Patients with AML

Researchers used the Moffitt Cancer Center AML database to identify variables predictive of poorer outcomes in a large cohort of older patients with untreated acute myeloid leukemia (AML).

Using the Kaplan-Meier method to measure overall survival (OS) and a Cox regression model to determine the association between OS and predictive variables, a model for 12-month OS was developed using multiple logistic regression techniques. A total of 980 patients were identified for the analysis. Two-thirds of patients (66%) were male; median age at diagnosis was 76 years (range, 70-96). A total of 43.1% of patients had newly diagnosed AML, and 31% had high-risk cytogenetics. Other variables included Eastern Cooperative Oncology Group (ECOG) status, prior hypomethylating agent initial therapy, and blood/platelet counts. Median OS was 7.1 months for the entire cohort.

When evaluating the effect of all factors in the multivariate model, independent negative predictors for OS included secondary AML status, adverse-risk karyotype, ECOG ≥2, nonhypomethylating agent initial therapy, older age, increased white blood cell count, decreased platelets, and decreased hemoglobin.

A model to predict OS at 12 months was developed in a subset of 446 patients. In this subgroup analysis, independent predictive variables included karyotype, ECOG performance status, AML type (de novo vs secondary), age, and white blood cell count. There was no significant difference in OS based on whether patients received initial treatment in a clinical trial.

In this analysis of older patients, prognostic modeling was used to discern differences in longer-term survival with conventional therapies. Predictive techniques permitted the identification of high-risk subsets with strong discriminatory capacity. Research efforts continue, and decision modeling is being used to further inform the choice of optimal therapies for these patients. Statistically valid prognostic modeling techniques that help identify optimal therapeutic approaches could have positive implications for patients as well as payers of cancer care.

Lancet JE, et al. ASCO Abstract 7031.

Related Items