The study design and primary analysis results have been previously published for PRIMA/ENGOT-OV26/GOG-3012 (NCT02655016).6 Briefly, PRIMA was a phase III trial that enrolled patients with newly diagnosed advanced (International Federation of Gynecology and Obstetrics [FIGO] stage III or IV) high-grade serous or endometrioid ovarian, primary peritoneal, or fallopian tube cancer (collectively referred to as ovarian cancer) with complete or partial response to first-line platinum-based chemotherapy at high risk for disease progression. After first-line chemotherapy, patients were randomized 2:1 to receive oral niraparib or placebo once daily.

At study initiation, all patients received a fixed starting dose of 300 mg once daily; however, the protocol was amended to incorporate an individualized starting dose based on baseline body weight and platelet count. Randomization stratification factors were clinical response to first-line chemotherapy (complete/partial response), receipt of neoadjuvant chemotherapy (yes/no), and tumor homologous recombination-deficiency status (homologous recombination-deficient vs homologous recombination-proficient or not determined per the MyChoice CDx [Myriad Genetics, Salt Lake City, Utah, USA]). PRIMA was conducted according to the principles of the Declaration of Helsinki, good clinical practice, and all local laws under the auspices of an independent data safety and monitoring committee (Institutional Review Board list; see Online Supplemental Table 1).

All patients provided written informed consent.6 The primary endpoint, progression-free survival (defined as the time from randomization after completion of platinum-based chemotherapy to the earliest date of objective disease progression on imaging [according to the Response Evaluation Criteria in Solid Tumors, version 1.1]20 or death from any cause) was assessed by blinded independent central review in the homologous recombination-deficient and overall (intent-to-treat) populations per prespecified hierarchical testing.6 Beyond the primary analysis, the protocol specified investigator assessment of progression-free survival, which was shown to be concordant with blinded independent central review progression-free survival in the primary analysis.6 7

This post hoc analysis analyzed factors associated with long-term progression-free survival (defined as ≥2 years) in patients randomized to niraparib in the intent-to-treat population. The 2-year cut-off was selected based on conditional progression-free survival analyses in the intent-to-treat population which showed that about two-thirds of patients treated with niraparib who were alive and progression-free at the 2-year landmark were expected to remain so at the 4-year landmark.8 Patients treated with niraparib were grouped based on their investigator-assessed progression-free survival status (progressive disease/censoring <2 years, hereafter referred to as progression-free survival <2 years, vs progressive disease/censoring ≥2 years after randomization, hereafter referred to as progression-free survival ≥2 years).

Patients without evaluable baseline/post-baseline radiological assessments were censored at the date of randomization (unless death, disease progression, or subsequent anticancer therapy initiation occurred within 2 visits of randomization). Patients who had not died, experienced disease progression, or initiated subsequent anticancer therapy and those who had died, experienced disease progression, or initiated subsequent anticancer therapy after ≥2 missed radiological assessments were censored at the date of the latest evaluable radiological assessment.

Multivariable logistic regression modeling using backward elimination (significance level=0.15) was used to identify baseline characteristics associated with long-term progression-free survival (Figure 1). As previously described, the least significant variable that did not meet the significance level for remaining in the model was removed. This process was repeated until all variables in the model that did not meet the specified significance level had been removed.21

Multivariable logistic regression modeling method using backward elimination to identify characteristics for further assessment.

Based on prior association with ovarian cancer prognosis, the authors selected the following baseline variables for inclusion in the logistic regression model: Eastern Cooperative Oncology Group (ECOG) performance status scores (0 vs 1), FIGO stage at diagnosis (III vs IV), clinical response to platinum-based chemotherapy (complete vs partial response), number of prior chemotherapy cycles (6, 7–9, or missing), primary tumor location (ovarian, primary peritoneal, or fallopian tube), body mass index (<18 kg/m2, 18–<25 kg/m2, 25–<30 kg/m2, or ≥30 kg/m2), categorical age (<65 years vs ≥65 years), type of debulking surgery (primary debulking surgery, interval debulking surgery, or no surgery), number of baseline target lesions (0 vs ≥1), number of baseline non-target lesions (0, 1, or ≥2), BRCA/homologous recombination-deficiency status (BRCA1/homologous recombination-deficient, BRCA2/homologous recombination-deficient, BRCA wild-type/not determined/homologous recombination-deficient, or BRCA wild-type/not determined/homologous recombination-proficient/not determined), residual disease status (no visible residual disease after neoadjuvant chemotherapy/interval debulking surgery, visible residual disease after primary debulking surgery, or missing/no surgery), and duration from end of chemotherapy to date of randomization in weeks. Categories for target (0 and ≥1) and non-target (0, 1, and ≥2) lesions at baseline were determined based on sample size. Potential characteristics of interest including race, ethnicity, and location of baseline target/non-target lesions, although important, were excluded from the multivariate logistic regression model because of low sample sizes (see Online Supplemental Table 2).

Correlations between the variables were evaluated to check for multicollinearity (linear dependence between multiple independent variables),22 and all correlation coefficients were confirmed to be <0.70, indicating that the variables were not linearly dependent. As a sensitivity analysis, an alternative definition of categorical progression-free survival was used in which patients who were censored <2 years were excluded because their exact progression date was unavailable; this definition was less conservative because it did not assume that patients who were censored <2 years experienced disease progression <2 years after randomization. Using the less conservative definition, 45 patients were excluded from the progression-free survival <2 years category. All analyses were conducted using data from the updated ad hoc analysis (clinical cut-off date November 17, 2021) and performed using SAS version 9.4. Associations with p values <0.05 were considered significant.