PERFORMANCE OF TIME TO DISCONTINUATION AND TIME TO NEXT TREATMENT AS PROXY MEASURES COMPARED WITH DIRECT OBSERVATION OF PROGRESSION-FREE SURVIVAL
Walker MS, Herms L, Miller PJE
Concerto HealthAI, Boston, MA, USA
OBJECTIVES Real-world data sources have historically provided limited access to information about the occurrence of disease progression. As a result, progression-free survival (PFS), a standard effectiveness outcome measure in oncology, has been calculated using proxies such as time-to-discontinuation (TTD) or time-to-next-treatment (TTNT). However, electronic medical record (EMR) data sources that allow detailed curation of unstructured content make it possible to calculate PFS based on directly observed progression. The aim of this research was to compare PFS as estimated from proxy TTD and TTNT values with PFS determined directly from curated progression data. This report cites findings from the metastatic breast cancer (mBC) setting. METHODS Previously curated EMR data were collected from Concerto’s Definitive Oncology Dataset for adult, female, hormone receptor-positive/human epidermal growth factor receptor 2-negative mBC patients with a diagnosis in 2008 or later who received systemic therapy. Kaplan-Meier methods were used to estimate PFS, TTD and TTNT based on direct observation of events in structured and unstructured data. RESULTS The study included a total of 378 patients, of which 138 (36.51%) were de novo stage IV or metastatic. The mean patient age was 60.30 (SD 13.30), 57.14% were Caucasian, and 68.25% were postmenopausal. The median PFS in the sample was 9.11 months [95% CI 7.89, 10.68], compared to a median TTD of 5.56 months [5.00, 6.48] and TTNT of 5.95 months [5.26, 6.94]. Censoring of endpoints occurred in a minority of patients, with 307 (81.22%) PFS events, 342 (90.48%) TTD events, and 336 (88.99%) TTNT events. CONCLUSIONS Estimates of TTD and TTNT aligned closely, suggesting that nearly all patients who discontinued initial therapy received subsequent treatment. However, PFS, measured directly in this study, was at least 50% longer than TTD and TTNT. This suggests that use of TTD and TTNT may significantly underestimate PFS when used as proxy measures of that endpoint.
Conference/Value in Health Info
2020-05, ISPOR 2020, Orlando, FL, USA