OBJECTIVES: To estimate the cost-effectiveness of niraparib, a poly (ADP-ribose) polymerase (PARP) inhibitor, compared with routine surveillance (RS) and olaparib for the maintenance treatment of patients with recurrent ovarian cancer. METHODS: A decision-analytic model estimated the cost per quality-adjusted life-year (QALY) gained for niraparib versus RS and olaparib from a US payer perspective. Recurrent ovarian cancer patients with or without germline BRCA mutation, who were responsive to their last platinum-based chemotherapy regimen, entered the model (non-gBRCAmut, gBRCAmut). The model had three health states; progression-free disease, progressed disease and dead. For both non-gBRCAmut and gBRCAmut populations, mean progression-free survival (PFS) was estimated for niraparib and RS using parametric survival distributions based on ENGOT-OV16/NOVA (Niraparib Phase III trial). For non-gBRCAmut, olaparib PFS was estimated from Study 19 (Olaparib Phase II trial). For gBRCAmut, a cost-minimisation analysis was conducted for niraparib versus olaparib. Due to the immaturity of OS data in ENGOT-OV16/NOVA, mean OS benefit was estimated from the Study 19 RS arm, as double the mean PFS benefit for both niraparib and olaparib. Costs included; drug, subsequent chemotherapy, monitoring, adverse events, and terminal care. Quality-of-life was captured using EQ-5D. A 3.00% discount rate per annum was used. RESULTS: Treatment with niraparib increased costs and QALYs compared to RS, resulting in an incremental cost-effectiveness ratio of $94,186 and $58,804, for non-gBRCAmut and gBRCAmut, respectively. Niraparib dominated olaparib in both populations, with a cost difference of -$60,400 and similar QALYS in gBRCAmut, and a cost difference of -$55,858 and incremental QALYs of 1.437 in non-gBRCAmut. This cost decrease was driven by lower monthly drug costs for niraparib compared with olaparib based on average dose. CONCLUSIONS: These estimates indicate that niraparib dominates olaparib and falls within an acceptable cost-effectiveness range versus RS. Mature OS data is required to validate these findings.