PHILADELPHIA (October 4, 2019) – Researchers at Fox Chase Cancer Center have developed a statistical method that produces better cancer treatment cost comparisons when measured or unmeasured confounders are present.
“When considering two treatments, we want to take into account cost and effect of therapy. We were interested in estimating cost effectiveness by looking at claims data from SEER-Medicare,” said lead researcher Elizabeth Handorf, PhD, assistant professor in the Biostatistics and Bioinformatics Facility at Fox Chase.
Handorf and her colleagues used net monetary benefit (NMB) to quantify cost effectiveness. They used a statistical method to account for the correlation between cost and survival. Lastly, they conducted a sensitivity analysis to determine potential bias in the standard estimates.
Unmeasured confounders are hidden data researchers don’t have access to that can change the outcome of a cost-effectiveness analysis.
“We want to know if small differences in the unmeasured data could easily change the conclusion regarding cost effectiveness,” Handorf said.
To test the performance of their statistical method, the researchers conducted a simulation of 400 individuals, 200 given an intervention and 200 controls. Their method produced accurate estimates of treatment effects on cost, survival, and NMB.
Handorf and her team then demonstrated their method’s accuracy at providing cost, survival, and NMB estimates in an actual cost-effectiveness comparison of two treatments for Stage II/III bladder cancer. They compared costs of radical cystectomy (RC), a procedure where the entire bladder and surrounding tissue are removed versus bladder preserving therapy (BPT), a combination of radiotherapy and chemotherapy treatments.
The researchers drew Medicare payment information on 1,860 patients. The data collected included costs for inpatient, outpatient, and nursing home care. Patients underwent either RC (1,440) or BPT (420).
Handorf and her colleagues found BPT had a decrease in mean survival of 1.74 years and decrease in costs of $16,271. When adjusting for measured confounders, BPT was not cost effective.
When a cost-effectiveness analysis has unmeasured confounding variables, it can introduce bias to an analysis. In this comparison, smoking status and personal income may have biased the results.
The researchers found their result could change if there were differences in smoking rates between the two patient groups. Differences in patient incomes could also change the conclusions.
“It’s important to take into account whether biases are changing the conclusions of cost effectiveness analyses,” Handorf said. “This method could be applied to any cost-effectiveness analysis using any observational data resource.”
The study, “Estimating Cost-Effectiveness From Claims and Registry Data With Measured and Unmeasured Confounders,” was published in Statistical Methods in Medical Research. The study was supported by grants from the U.S. Public Health Service.