The utility of using area under the curve to analyze symptom burden during radiation/chemoradiation for head and neck cancer.

Abstract

5525 Background: Patients undergoing radiation therapy alone (RT) or with chemotherapy (CRT) for head & neck cancer (HNC) experience acute toxicity. Using patient-reported outcomes to report symptom burden is widely accepted, but analytical techniques that utilize such information are lacking. Summary measures like area under the curve (AUC) are potential quality indicators and have been used as primary endpoints in symptom intervention trials. We compared 3 methods of analyzing patient-reported symptom burden data in a prospective longitudinal study. Methods: HNC patients completed the M. D. Anderson Symptom Inventory Head & Neck module (MDASI-HN) before and weekly for up to 7 weeks during RT/CRT. We calculated the AUC and the aggregated mean for selected symptoms, and we compared them by treatment group (RT/CRT) using independent t-tests with Bonferroni correction (p=.006). We fit linear mixed models for each selected symptom, with time, treatment group and group-by-time interaction as covariates. Results: Of 152 enrolled patients, 118 provided 7 weeks of complete data. We found that the treatment groups differed significantly on fatigue AUC (122.7 vs 173.4, p<.002) and aggregated mean (2.8 vs 3.8, p<.004). The AUC of 173.4 represents an average daily rating of 3.5 on a 0–10 scale. The groups also differed significantly on drowsiness AUC (90.1 vs 136.6, p<.003) and aggregated mean (2.1 vs 3.0, p<.005). The groups differed significantly on lack of appetite AUC (104.2 vs 150.9, p<.005), but not on aggregated mean (2.4 vs 3.4, p<.007). In all cases, CRT resulted in higher AUCs and aggregated means than did RT. The group-by-time interaction terms in the linear models were not significant for drowsiness (p<.18) or fatigue (p<.12). The group-effect terms were not significant for drowsiness (p<.04) or fatigue (p<.02) when compared to the adjusted p-value. Conclusions: The AUC is more sensitive than aggregated mean in detecting group differences and may require smaller sample sizes to show group differences. Although linear mixed models did not detect significant differences in treatment groups, the additional information provided by longitudinal analysis supplements those provided by the AUC.