Case Study
Testing
the hypothesis
that estimates of cost-effectiveness (CE) lose validity soon after drug launch due to changes in relative effectiveness that accrue from improvement in execution and selection effects.
The Question
Standard CE models rely on an assumption of time-stopping at launch even though many models are expected to be ‘lifetime’ models. Numerous studies have highlighted how ICERs change over time, but these have largely been limited to looking at how ‘prices’ change over time (so called dynamic pricing models).
Yet there is still a strong acceptance of the assumption that relative effectiveness does not (or that it likely falls over time). This ignores basic economic theory, that states that once a new technology is embedded in a system, over time, practitioners will optimize the way (and the amount) that technology is used to maximize its efficiency.
The Approach
To test this hypothesis, we studied a cohort of cancer patients retrospectively from treatment initiation to death over a period of over ten years, using SEER-Medicare linked sample data.
We followed patients from the year at which a new drug was launched, and compared the relative effectiveness of the drug in practice to its comparator drug from the RCT which was used to estimate the relative effectiveness (and the resulting ICER) in launch CE models.
Cohorts for each treatment were match-adjusted for comparability. We replicated this analysis for three separate drugs: FOLFOX for 1L colorectal cancer, FOLFOX for 3L colorectal cancer and gemcitabine for 1L pancreatic cancer.
The Results
We calculated hazard ratios (HR) of mortality by year for each regimen and line of therapy. There were consistent downward trends in the HRs across regimens, suggesting that clinical effectiveness was improving over time. Comparisons of HRs using both the IPTW-ATT weights and HRs based on the unweighted but covariate-adjusted Cox models were used to cross-validate results.
Although there were some differences in the HRs for second line FOLFOX, the HRs were generally similar between the unweighted and IPTW-ATT weighted models. Overall, these results suggest that the downward trend in the HRs is not sensitive to weighting. This translates directly into ICER estimates, for example for 1L FOLFOX for colorectal cancer, the implied incremental ICER based on the observational data fell from $610,000 per life year gained in year 1 to $27,000 per life year gained in year 7.
The Long And
Short Of it
The study showed that the impact of the change in price improved ICERs marginally over time, but that change in relative effectiveness made these drugs more than 3 times as cost-effective after embedded in practice for 2-3 years.
This is consistent with the theory of ‘learning-by-doing’ that new technologies take time for practitioners to optimize when, how and to whom to deliver it in practice.