Can administrative data improve the performance of clinical trial economic analyses?

Abstract

2 Background: Economic analyses of trials often rely on trial-collected health resource utilization data, which is expensive and may be incompletely recorded. We investigated whether routinely collected health administrative data (RCD) can be utilized to improve trial economic analysis performance. Methods: Health administrative data was probabilistically linked to Ontario patient data from the Canadian Cancer Trials Group CO.17 trial (n = 572), evaluating cetuximab plus best supportive care (n = 75 linked Ontario patients) versus best supportive care alone (n = 73). Completeness of trial data was compared to RCD. Cost-effectiveness in 2007 Canadian dollars was determined using RCD up to trial date of last contact (DOLC), and up to RCD DOLC. Incremental cost effectiveness ratio (ICER) confidence intervals (CI) were determined using bootstrapping with 5000 iterations. Cost-effectiveness acceptability curves were determined. Sensitivity analyses were performed. Results: Among 148 Ontario patients, up to trial DOLC, RCD vital status was concordant in > 96%. 29 deaths occurred after trial DOLC. Up to trial DOLC there were 34 net additional hospitalizations in RCD, and 28 net additional emergency room visits. Using RCD, total cetuximab group costs were $3,023,034, and $1,191,118 for best supportive care alone up to trial DOLC. Cost difference was driven by cetuximab drug costs ($1,531,370). Using RCD, the ICER was $211,128 per life-year gained (90% CI: $101,396, $694,950) when data was limited to trial DOLC, and $164,378 (90% CI: -$138,260, $644,555) using routinely collected data DOLC. ICER estimates were similar to the original economic analysis using trial-collected data ($199,742 (95% CI $125,973, $652,492)). Estimates were robust in sensitivity analysis. Conclusions: Administrative data were more complete than trial-collected utilization data, even under optimal conditions. There was also longer follow-up. We found that cost differences were robust to varying costing assumptions. Our findings demonstrate the potential of administrative data sources to relieve institutions, sponsors and patients from the burden of collecting key utilization information which requires considerable effort and cost.