Final Risk Prediction Model for Neutropenic Complications in Patients Receiving Cancer Chemotherapy

Introduction: Neutropenic complications including severe neutropenia (SN) and febrile neutropenia (FN) represent major dose-limiting toxicities of cancer chemotherapy. A prospective study was undertaken to develop and validate a predictive model for neutropenic events in patients receiving cancer chemotherapy. The final risk model based on mature data is presented.

Methods: Between 2002 and 2006, 4458 consenting patients starting a new chemotherapy regimen at 115 randomly selected community oncology practices throughout the United States were enrolled including 3760 with cancers of breast, lung, colorectum, ovary and malignant lymphoma receiving at least one cycle of treatment. Using a 2:1 random split sample methodlogy, a risk model for first-cycle SN or FN was derived and validated based on multivariate logistic regression analysis incorporating pretreatment variable information. The cumulative risk of events over the initial 120 days of treatment was estimated by the method of Kaplan and Meier. High and low risk groups were defined on the basis of the median predicted risk and model test performance characteristics were estimated.

Results: Following adjustment for cancer type, important predictive factors included: older age, prior chemotherapy, abnormal hepatic or renal function, low pretreatment white blood count, immunosuppressive medications and planned relative dose intensity >85% as well as use of several specific chemotherapeutic agents including anthracyclines, taxanes, alkylating agents, topoisomerase inhibitors, gemcitabine or vinorelbine. Lower risk of neutropenic complications were associated with primary prophylaxis with a colony-stimulating factor (CSF). Individual risk estimates based on the model ranged from 0–89% with mean and median of 19.2% and 10.1%, respectively. The model was associated with an R² of 0.34 and demonstrated excellent discrimination with a c-statistic of 0.833 [95% CI: 0.813–0.852, P<.001]. The model predicted risk of cycle 1 SN or FN in high and low risk groups was of 34% and 4%, respectively. The cumulative risk of FN over the initial 120 days was 20% in high risk patients and 5% in low risk patients. Model performance included sensitivity and specificity of 90% and 59%, respectively, with a model diagnostic odds ratio of 12.8 [95% CI: 9.3, 17.7]. Application of the model to the validation data set was associated with similar excellent discrimination and test performance characteristics. CSF prophylaxis applied to high risk patients was associated with significantly lower risk of FN over repeated cycles of chemotherapy [HR = 0.51; 95% CI: 0.35 – 0.75; P <.0001]. Nearly two-thirds of patients classified as high risk but who did not receive primary CSF prophylaxis went on to receive secondary use during subsequent cycles.

Discussion: Based on excellent test performance characteristics, the risk model identified patients with a cumulative incidence of FN of at least 20% who are candidates for targeted prophylaxis with a CSF. Further validation of this model in actual clinical practice is currently underway.