Background: Tolerability is a key consideration when selecting a PARP inhibitor (PARPi) for ovarian cancer (OC). Here we expand on earlier work (Arend et al 2021) to further characterize real-world tolerability and dose modifications in US patients (pts) with OC receiving PARPi therapy. Methods: A retrospective cohort of OC pts starting olaparib (ola), niraparib (nir) or rucaparib (ruc) between Jan 2017 and Dec 2020 was identified from MarketScan^® Commercial/Medicare Supplemental databases, increasing the period covered and number of pts included vs our previous analysis. Pts were followed up from first PARPi prescription (index) for ≥30 days until end of study period, disenrollment or death; baseline was 6 months pre-index. Clinical events of interest (CEIs; acute myeloid leukemia/myelodysplastic syndromes, anemia, leukopenia/neutropenia, thrombocytopenia, acute kidney injury, arthralgia, constipation, diarrhea, nausea, vomiting, dermatitis/rash/photosensitivity, fatigue, hypertension, infection, insomnia, pneumonitis, transaminitis) were identified via ICD-9/10 codes. Multivariable Cox regression compared the likelihood of CEIs, dose modifications and hospitalizations between PARPis, adjusting for baseline CEI, Charlson Comorbidity Index score, prior bevacizumab and cancer-related surgery. Persistence was defined as no PARPi treatment gaps of >90 days in pts with ≥6 months’ continuous enrollment. Results: Overall, 637, 538 and 227 pts received ola, nir and ruc, respectively (median [IQR] follow-up 10.5 [13.4] months). Baseline characteristics were similar across groups. The proportion of pts initiating PARPi at the highest indicated dose was 89.2%, 57.6% and 89.9% for ola, nir and ruc, respectively; 22.6%, 34.8% and 28.6%, respectively, required dose decreases. The likelihood of experiencing CEIs varied across the PARPis after adjusting for a priori confounders as shown in the table. Persistence with index PARPi was higher with ola (83.4%) vs nir (73.3%; P<0.001) and similar vs ruc (80.2%; P>0.05). Among all pts, mean time to non-persistence was shorter with nir vs ola and ruc (6.4 vs 7.9 and 7.6 months, respectively; both P<0.05). CEIs by PARPi dose and calendar year will also be presented. Conclusions: This is the largest real-world comparison of PARPi use in OC pts reported to date. It supports differences between PARPis in persistence with therapy and risk of experiencing a CEI, even after adjusting for confounders.

*P<0.001; ^†P<0.01; ^‡P<0.05.