A predictive instrument for chemotherapy dose reductions would help optimize delivery of planned chemotherapy and rationalize the use of myeloid growth factors. We analyzed data on 833 women with breast cancer treated with cyclophosphamide, doxorubicin, and fluorouracil, for six cycles in two phase III clinical trials. From the first study ( n=323), we generated a logistic regression model that predicts an individual patient's probability of receiving significantly reduced chemotherapy, defined as less than 85% of the planned dose over cycles 2-6, using data generated from cycle 1. The model was validated on data from the second study ( n=510). The predictive model's variables include nadir absolute neutrophil count (ANC) in cycle 1 (OR: 0.14, 95% CI: 0.06-0.30, P<0.001) and percent drop of platelets between day 1 and the nadir in cycle 1 (OR: 1.04, 95% CI: 1.02-1.05, P<0.001). Both variables are dose adjusted based on the chemotherapy cycle 1 dose. The model's discriminatory performance was good (ROC area=0.82), as was the calibration of predicted with actual frequencies of dose reductions. In the validation dataset, model variables remained significant, with an ROC area of 0.78 and good calibration. In summary, we devised and validated a predictive instrument that uses data from a patient's first cycle of chemotherapy to compute the probability of requiring a significant chemotherapy dose reduction on subsequent cycles. This instrument could help clinicians select patients who will benefit from early administration of myeloid growth factors.