Anemia represents the most common hematological toxicity in cancer patients receiving chemotherapy and is associated with considerable morbidity and cost. ASH/ASCO guidelines call for intervention at a hemoglobin (hgb)<10 g/dL. A meta-analysis has demonstrated the clinical value of early (hgb≥10 g/dL) versus late (hgb<10 g/dL) intervention with an erythroid stimulating protein (ESP). An anemia predictive model may help guide intervention sufficiently early in the course of chemotherapy when it can be most effective. A prospective, nationwide study was undertaken to develop and validate risk models for hematologic toxicities of chemotherapy. The analysis presented here is based on 3,640 patients with cancer of the breast, lung, colon and ovary or malignant lymphoma receiving a new regimen prospectively registered at 117 randomly selected U.S. practices. A logistic regression model for hgb<10 g/dL was developed and validated using a 2:1 random selection split sample methodology. Predictive performance characteristics were estimated [±95% CL]. Nadir hgb over 4 cycles of chemotherapy was <8 g/dL in 113 (3%), 8–10 in 959 (26%), 10–12 in 1,847 (51%), and ≥12 in 721 (20%). No significant differences were observed between the two populations. Independent risk factors for nadir hgb<10 g/dL (ORs) were: female gender (1.66); ECOG >1 (1.70); CHF (1.54); history of vascular disease (2.66); ulcer disease (2.58); COPD (1.29); connective tissue disease (1.84); advanced cancer stage (1.19); cancer type and chemotherapy based on anthracyclines (2.15), carboplatin (2.40), gemcitabine (2.48), cyclophosphamide (1.60), etoposide (2.84), topotecan (4.21), or trastuzumab (1.43), planned cycle length >1 week (2.0), while normal baseline hemoglobin, platelet count and GFR were associated with a reduced risk. Model fit was good (P<.001), R2 = 0.35 and c-statistic = 0.81 [.79–.83, P<.0001]. Mean and median predicted risk for hgb<10 g/dL were 0.29 and 0.22, respectively. An increasing risk cutpoint was associated with lower sensitivity and higher specificity. In the highest risk half, quarter and quintile of patients, hgb<10 g/dL was experienced by 47% [45–50], 64% [60–68], and 70% [65–74], respectively. Model performance characteristics at the median risk included: sensitivity: 82% [78–84]; specificity: 64% [62–66]; and diagnostic odds ratio: 7.80 [6.28–9.68]. Most covariates significant in the derivation model remained significant in the validation population. Model fit was good [P<.001] with an R2=.40 and a c-statistic of 0.83 [.81–.86; P<.001]. In the highest risk half, quarter and quintile of patients, hgb<10 g/dL was experienced by 50% [46–54], 67% [62–71], and 70% [65–75], respectively. Test performance of the validation model at the median risk included: sensitivity of 83% [79–86], specificity of 62% [59–66], and a diagnostic odds ratio of 7.90 [5.84–10.69]. Based on good performance characteristics, this validated prediction model identified chemotherapy patients at increased risk for developing clinically significant anemia who may be candidates for early targeted intervention with an ESP. A conditional risk model for subsequent risk of hgb<10 g/dL which includes changes during cycle 1 of chemotherapy has also been developed and will be presented.

Disclosures: Unrestricted research grant from Amgen Inc; other funding: Sunesis, Ortho Biotech.; Honorarium for speaking for Amgen; other honoraria from Astra Zeneca, Bristol-Myers Squibb, Cephalon, Lilly.; Speaker’s bureau and advisory committee for Amgen; advisory board for Astra Zeneca, Bristol-Myers Squibb, Cephalon, Lilly.

Author notes


Corresponding author