Purpose: Clinical guidelines recommend the use of prophylactic granulocyte colony stimulating factors (G-CSF) to prevent febrile neutropenia (a common side effect of cancer chemotherapy) when its risk is greater than 20%. In the absence of a clinical prediction tool, this risk assessment is left in the hands of the treating oncologist. The purpose of this study was to identify factors influencing oncologists' decisions to use primary prophylaxis G-CSF at the start of chemotherapy. Methods: Data for this study were obtained from a prospective observational multi-center study of cancer patients starting chemotherapy. Patients with solid cancer tumors treated with myelosupressive chemotherapy regimens were included. The primary outcome variable was G-CSF use planned at the beginning of the first chemotherapy cycle or prior to a neutropenic event within the first cycle. Three categories of covariates were considered: patient's socio-demographic characteristics, clinical indicators, and treating institution characteristics. The association between outcome and covariates was investigated using univariate followed by multivariate logistic regression analyses. Model selection was conducted using a modified stepwise procedure. To account for the clustered nature of the data, the models included a random intercept. The data were split randomly in a 2:1 ratio to training and validation samples, respectively. Significance of the random intercept was tested using a likelihood ratio test. Results: 4466 patients were registered in the study. 3213 (71.9%) diagnosed with solid tumor were included in the analysis. 611 (19%) of those included were treated with primary prophylaxis G-CSF. Of the initial 146 covariates tested in the univariate analysis, 32 were significant and considered in the multivariate analysis. The final multivariate model includes 12 covariates, of which 5 were found significantly associated with outcome in the validation step: year of diagnosis (p<0.0001), employment status (p=0.0095), treating institution's geographic region (p=0.0025), cancer type (p=0.0123), and chemotherapy regimen (p=0.0137). Differences in decisions across clinics was found to be significant (p<0.001). Conclusions: While risk factors associated with neutropenia as indicated by clinical studies play a significant role in the decision process, the use of G-CSF varies across geographic regions and treating institutions. Several clinically established neutropenia risk factors do not appear to be part of the decision to use G-CSF, emphasizing the need for a quantitative risk assessment tool.