OBJECTIVES: Cost-effectiveness analyses are increasingly informed by observational data but observational data comparisons are likely biased due to confounding by indication. Statistical methods, like genetic matching (GenMatch), have been developed to overcome this by optimising covariates balance (i.e. similarity between baseline patient characteristics) between intervention and comparator groups. However, poor covariates balance may be achieved, and covariate variance may be underestimated when the comparator group is small because the number of potential matches is limited. In oncology, multiple treatment lines are often administered to patients, and the same treatment can be used in different lines. Additionally, potential matches are selected at the moment they become eligible for the intervention, while patients who have received the intervention may not have received it directly when they became eligible. Matching based on treatment line (instead of traditional patient level matching), would increase the number of potential matches. This study compares the cost-effectiveness results of 1) patient and 2) treatment line level matching.

METHODS: GenMatch was applied on the patient and treatment line levels. Resource use and comparative effectiveness inputs were determined for the unmatched, patient level matched, and treatment line level matched comparator groups. The literature informed utilities. A cost-effectiveness analysis was performed based on the unmatched, patient level matched, and treatment line level matched comparisons.

RESULTS: Covariates were more balanced under treatment line level matching compared to patient level matching. Incremental quality-adjusted life years (QALYs) were -0.188, 0.030, and 0.021 based on the unmatched, patient level matched, and treatment line matched comparisons, respectively; incremental costs were -€43,024, €7,258, -€7,657, respectively. Incremental net monetary benefits (at a willingness-to-pay of €80,000 per QALY) were €27,958, -€4,818, €9,298, respectively.

CONCLUSIONS: Treatment line level matching improved covariates balance the most. Matching (patient or treatment line level) dramatically influenced (cost-)effectiveness results and, most likely, the reimbursement decision.