Identifying optimal cutpoints for quality of life function in patients with advanced cancer

Abstract

8139 Background: Substantial resources have been spent to develop validated quality of life (QoL) tools in cancer. However, few hospitals in the US routinely utilize QoL measures to develop treatment plans in patient care. Indeed, to our knowledge, our center is the only hospital-based oncology program in the US that routinely collects QoL data on patients not participating in clinical trials. We used survival analysis to determine if we could convert continuous scales of QoL data into categorical prognostic outcomes. Methods: Baseline QoL data were collected from 310 patients treated at our center between 04/01 and 10/03. 2 QoL tools were used: Ferrans and Powers Quality of Life Index (QLI), which has 4 functional domains, and EORTC QLQ-C30, which has 5 functional domains and 9 symptom items. Using the Wilcoxon test, differences in survival were measured in serial increments of 10 points for EORTC and 3 points for QLI (these levels are associated with significant improvement in QoL). Results from Wilcoxon test were plotted as a function of QoL cutoff points. Linear and polynomial solutions were fitted and Wald statistic identified the best fit. Results: Of 310 patients, 180 were females and 130 males, with a median age of 55 years (range 21 - 82). 64.2% had failed prior treatment. Most common cancers were breast (25%), colorectal (23.3%), and lung (17.7%). Statistical analysis indicated that Health and Function Domain (Wald Test P = 0.0194) had two cutpoints at values 10 and 20. We found a statistically significant difference in survival between patients with scores < 10, 10 to 20, and > 20, the median survival being 169, 349, and 692 days respectively (p < 0.0001). Similarly, EORTC fatigue item (Wald Test P = 0.022) has two cutpoints at values 35 and 80. The median survival of patients with scores < 35, 35 to 80, and > 80 were 450, 413, and 129 days respectively (p < 0.0001). Conclusions: The transformation of a QoL research tool into a clinical management tool is a challenge. As an initial step, we showed that we could produce categorical survival outcome categories from QoL data in 310 patients with advanced cancer. The next step is to refine this analysis to produce discrete outcome categories that will facilitate physician thinking on treatment planning. No significant financial relationships to disclose.