Comparative efficacy, safety, and acceptability of single-agent poly (ADP-ribose) polymerase (PARP) inhibitors in BRCA-mutated HER2-negative metastatic or advanced breast cancer: a network meta-analysis

Search results and study characteristics

Overall, the database search identified 15,831 studies; 4,548 studies were excluded due to duplicate records and 11,238 studies were excluded based on the selection criteria after reading the title and abstract. Subsequently, 45 potentially relevant full-text articles were reviewed. After applying all the eligibility criteria, 43 studies were excluded because 25 were reviews, 14 were non-RCTs, three were case-reports, and one were not single-agent PARPi studies. Therefore, two RCTs, the EMBRACA and OlympiAD, were included in our NMA (Figure 1).

The trials involved a total of 733 patients: 492 received single-agent PARPi (OlympiAD, olaparib; EMBRACA, talazoparib) and 241 received TPC. The mean sample size was 366.5, ranging between 302 and 431. Regarding study quality, two domains had unclear risk of bias and five domains had low risk of bias; therefore, both RCTs were considered to have unclear risk of bias (Figure 2).

Tables 1 and 2 show the main characteristics of the RCTs. The two studies were randomized, controlled, open-label, multicenter international phase III trials with participation from 16–18 countries. The mean trial duration was 28.5 months (range, 15–42 months). In both RCTs, about half of the patients had TNBC; <30% of the participants had received platinum-based chemotherapy before participating in the RCTs.

Here, we included two RCTs fulfilling all the eligibility criteria, and there were no head-to-head studies assessing the differences between single-agent PARPi. Due to the absence of a loop connecting three arms, the node-splitting method was not calculated for all outcomes.

Efficacy outcomes

Regarding efficacy, olaparib was not associated with improved PFS (HR = 1.08, 95% CrI = 0.34–3.45) when compared with talazoparib, and TPC conferred worse PFS (HR = 1.85, 95% CrI = 0.82–4.15). However, this was not statistically significant (Figure 3).

In subgroup analysis according to hormone receptor status, both olaparib and TPC conferred worse PFS in the hormone receptor–positive subgroup compared with talazoparib, but statistical significance was not obtained (olaparib: HR = 1.74, 95% CrI = 0.43–6.69; TPC: HR = 2.10, 95% CrI = 0.79–5.58) (Figure 4A). In analyses of patients with TNBC, olaparib was associated with improved PFS (HR = 0.72, 95% CrI = 0.15–3.50), while PFS was worst with TPC (HR = 1.68, 95% CrI = 0.55–5.05) (Figure 4B). In the subgroup analysis according to prior exposure to platinum-based chemotherapy, olaparib was not associated with PFS (olaparib: HR = 0.90, 95% CrI = 0.32–2.49) compared with talazoparib, and PFS was worst with TPC (HR = 1.31, 95% CrI = 0.60–2.84) (Figure 4C). When analyzing patients in the no prior platinum subgroup, PFS was worse with both olaparib and TPC (olaparib: HR = 1.14, 95% CrI = 0.35–3.82; TPC: HR = 1.90, 95% CrI = 0.81–4.40) (Figure 4D).

In the subgroup analysis according to BRCA mutation type, olaparib was not associated with PFS, and PFS was worse with TPC in the BRCA1 mutation subgroup (olaparib: HR = 0.91, 95% CrI = 0.28–3.01; TPC: HR = 1.68, 95% CrI = 0.72–3.88) compared with talazoparib (Figure 4E), both results did not reach statistical significance. When exploring the difference between subjects with BRCA1 mutation, PFS was worse with both olaparib and TPC (olaparib: HR = 1.46, 95% CrI = 0.35–6.03; TPC: HR = 2.11, 95% CrI = 0.77–5.67), although significant associations were not recognized (Figure 4F). Similar results were obtained when we analyzed only the subjects who had not received prior chemotherapy (olaparib: HR = 1.01, 95% CrI = 0.30–3.41; TPC: HR = 1.76, 95% CrI = 0.75–4.04) (Figure 4G).

We defined OS and ORR as secondary efficacy outcomes: when compared with talazoparib, the NMA of OS showed that PFS might be worse with olaparib and TPC, but the differences were short of statistical significance (olaparib: HR = 1.18, 95% CrI = 0.61–2.31; TPC: HR = 1.31, 95% CrI = 0.83–2.07) (Figure 5A). Regarding ORR, compared with talazoparib, both olaparib and TPC were associated with non-significantly improved ORR (olaparib: OR = 0.83, 95% CrI = 0.05–12.64; TPC: OR = 0.22, 95% CrI = 0.03-1.50) (Figure 5B).

Safety

Here, major hematologic AEs were defined as primary safety outcome, and included anemia, neutropenia, and decreased white cell count. In the olaparib trial, grade 3–4 anemia was reported in 82 of 205 patients in the olaparib group and in 24 of 91 patients in the TPC group. In the talazoparib trial, grade 3–4 anemia was reported in 151 of 286 patients in the talazoparib group and in 23 of 126 patients in the control group. Compared with talazoparib, olaparib and TPC reduced the risk of developing grade 3–4 anemia (olaparib: OR = 0.34, 95% CrI = 0.003–34.94; TPC: OR = 0.07, 95% CrI = 0.003–1.86) (Figure 6A) and any-grade anemia (olaparib: OR = 0.37, 95% CrI = 0.02–6.81; TPC: OR = 0.20, 95% CrI = 0.03–1.52) (Figure 6B). Regarding neutropenia, olaparib showed grade 3–4 neutropenia compared with talazoparib (OR = 0.57, 95% CrI = 0.06–5.75), while TPC showed worse results (OR = 2.00, 95% CrI = 0.40–9.99) (Figure 6C). Similar results were obtained for any-grade neutropenia (olaparib: OR = 0.54, 95% CrI = 0.09–3.27; TPC: OR = 1.42, 95% CrI = 0.41–4.95) (Figure 6D). Similarly, olaparib had better decreased white cell count (grade 3–4) compared with talazoparib (OR = 0.42, 95% CrI = 0.04–4.19), while TPC had the worst result (OR = 1.32, 95% CrI = 0.28–6.19) (Figure 6E). The results were altered when we reanalyzed any-grade decreased white cell counts: Olaparib and TPC had better decreased white cell counts (olaparib: OR = 0.55, 95% CrI = 0.20–1.52; TPC: OR = 0.75, 95% CrI = 0.37–1.49) (Figure 6F).

The secondary safety outcome mainly included fatigue and headache. Compared with talazoparib, olaparib and TPC were associated with increased risk of grade 3–4 fatigue (olaparib: OR = 6.79, 95% CrI = 0.44-262.48; TPC: OR = 1.83, 95% CrI = 0.35–8.64) (Figure 7A). Regarding any-grade fatigue, olaparib and TPC did not show any significant difference compared with talazoparib (olaparib: OR = 1.01, 95% CrI = 0.42–2.40; TPC: OR = 0.74, 95% CrI = 0.43–1.28) (Figure 7B). Regarding headache, olaparib and TPC had better grade 3–4 headache compared with talazoparib (olaparib: OR = 0.14, 95% CrI = 0.003–4.17; TPC: OR = 0.35, 95% CrI = 0.01–3.18) (Figure 7C). Similar results were obtained for any-grade headache (olaparib: OR = 0.82, 95% CrI = 0.25–2.74; TPC: OR = 0.59, 95% CrI = 0.27–1.28) (Figure 7D).

Acceptability

Regarding acceptability, the primary and secondary outcome was treatment discontinuation rate and time to QoL deterioration, respectively. Treatment was discontinued due to any-grade AEs by 10 and 7 patients in the olaparib and TPC group, respectively, and by 17 and 11 patients in the talazoparib and TPC group, respectively. Similarly, compared with talazoparib, olaparib did not show any significant difference for treatment discontinuation (OR = 0.93, 95% CrI = 0.20–4.37), and TPC conferred higher risk, while no significant differences were found (OR = 1.49, 95% CrI = 0.55–3.93) (Figure 8A).

Regarding time to QoL deterioration, olaparib and TPC were associated with short time to clinically meaningful QoL deterioration compared with talazoparib (olaparib: HR = 1.16, 95% CrI = 0.19–7.17; TPC: HR = 2.64, 95% CrI = 0.75–9.39), while no significant difference was obtained (Figure 8B).

Search strategy and selection criteria

This study is registered with PROSPERO (CRD42019138939) and is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) extension statement for NMA [16].

For this NMA, we searched Web of Science, Embase, PubMed, Medline, ClinicalTrials.gov, the Cochrane Central Register of Controlled Trials, and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) from the date of inception to July 20, 2019, with no language restrictions. References derived from full text review were screened to identify potential randomized controlled trials (RCTs) that had not been indexed in the above databases.

We used a prespecified search strategy using terms applicable to the population of interest. The search strategy was designed using a combination of medical subject headings (MeSH) and keywords. The MeSH and keywords included “breast”, “ mammary”, “cancer”, “cancers”, “tumor”, “neoplasm”, “carcinoma”, “poly(ADP-ribose) polymerase inhibitors”, “poly(ADP-ribose) polymerases”, “PARPi”, PARPi names (olaparib, veliparib, niraparib, rucaparib, talazoparib), and synonymous words. The reference lists from relevant studies, meta-analyses and reviews were manually screened for potentially eligible publications.

The electronic database searches were supplemented with manual searches for published, unpublished, and ongoing RCTs in international trial registers, drug approval agency websites, and key scientific journals in the field [13], and important professional conferences such as that of the American Society of Clinical Oncology (ASCO).

Study selection

Four investigators (JW, YZ, LY, LR) selected the studies independently. The titles and abstracts of identified publications were screened; potentially eligible articles were retrieved for full-text review. Disagreements were resolved by discussion with a fifth author (XQ).

Eligible studies were phase II or III RCTs for which efficacy and/or safety outcomes were available. Studies that compared any single-agent PARPi with standard mono-chemotherapy in patients with BRCA-mutated HER2-negative metastatic or advanced breast cancer were included. Eligible patients were at least 18 years of age and had HER2-negative breast cancer that was hormone receptor–positive or was triple negative. Furthermore, patients with confirmed deleterious gBRCA mutation and previous neoadjuvant or adjuvant treatment with platinum were included. Study subjects were limited to those with metastatic or local advanced breast cancer. We only included double-blind trials, and included placebo in the NMA, and we increased the methodological rigor of the study design by minimizing performance and ascertainment biases [17].

Data extraction and quality assessment

Four investigators (JW, YZ, LY, LR) independently extracted and summarized the data from all included studies using a standardized data abstraction form and assessed the risk of bias. The standardized data extraction forms included the trial characteristics (e.g., publication year, first author, journal), subject characteristics (e.g., the number of patients, patient age, previous treatment), intervention details (e.g., PARPi type, dose, duration of treatment), and outcome measures (efficacy, safety, acceptability).

The quality of the included RCTs was assessed using the Cochrane risk of bias tool [18] on selection, attrition, performance, detection, and reporting bias. Each RCT was deemed to be at low, high, or unclear risk of bias based on its adequacy of random sequence generation; allocation concealment; participant and personnel blinding; outcome assessor blinding; and method of addressing incomplete data, selective reporting, and other potential sources of bias. Trials with a rating of high risk of bias in one or more domains were considered high risk, trials with low risk of bias in all domains were low risk, and trials with unclear risk of bias in one or more domains were unclear risk [19]. Four investigators (JW, YZ, LY, LR) were separately and independently assigned for each RCT. Any discrepancies were resolved by consensus or by a fifth investigator (XQ).

Outcomes

The outcomes of interest were efficacy, safety, and acceptability. The primary efficacy outcome of interest was progression-free survival (PFS), the primary safety outcome of interest was major all-grade and high-grade (grade 3–4) treatment-related adverse events (AEs), and the primary acceptability outcome of interest was discontinued treatment due to AEs. Major AEs were defined as hematologic adverse events (we only focused on anemia, neutropenia, and decreased white cell count), assessed by concomitant medications and clinically relevant changes in laboratory values. The secondary efficacy outcomes included overall survival (OS) and objective response rate (ORR). The secondary safety outcome was defined as all-grade and high-grade (grade 3–4) non-hematologic AEs (fatigue, headache); the secondary acceptability outcome was impact on quality of life (QoL). Decreased QoL was defined as patients with clinically meaningful deterioration of QoL, as evaluated by the global health status/QoL scale on the EORTC QLQ-C30 (European Organization for Research and Treatment of Cancer QoL of cancer patients questionnaire).

Statistical analysis

NMA was carried out using a random-effect model within a Bayesian framework and executed by R software using the gemtc package (version 3.5.1; R Foundation, Vienna, Austria) [20, 21], which recalls JAGS in R for Markov chain Monte Carlo (MCMC) sampling. MCMC methods were applied to calculate the hazard ratios (HRs) or odds ratios (ORs) with 95% credible intervals (CrIs). For each analysis, the function mtc.run was used to generate samples through MCMC sampler. Initially, 10,000 simulations were set up for each chain as the burn-in period, yielding 100,000 iterations to obtain the HRs or ORs of model parameters, while four Markov chains were run simultaneously. The Brooks-Gelman-Rubin statistic and trace plots were assessed to check whether the model convergence was satisfactory. The number of iterations would not increase if the model convergence was sufficient [22]. If closed loops existed in the network of comparisons in the present study, estimating incoherence and network-plot were carried out [23]. Furthermore, the inconsistency of the model was calculated using the node-splitting method, which separates evidence on a particular comparison into direct and indirect evidence [24, 25]. Inconsistency between indirect and direct sources of evidence was statistically assessed globally and locally [26]. The mtc.anohe command of the gemtc package was applied to evaluate global heterogeneity via the heterogeneity variance parameter I². If at least 10 legal studies were available, the Egger test and funnel plot were used to detect publication bias [27].

The pooled estimates of HRs with 95% CrIs were calculated for PFS, OS, and the time to QoL deterioration. HR < 1 indicated lower probability of developing PFS and OS events, and delayed time to QoL deterioration. The pooled estimates of OR with 95% CrIs were calculated for the effect for ORR, AEs, and treatment discontinuation. OR > 1 indicated lower ORR, AEs, and treatment discontinuation rates.

To determine whether the results were affected by study characteristics, we performed subgroup analyses for primary efficacy outcomes according to the following variables: previous chemotherapy, previous platinum-based therapy, hormone receptor status, and BRCA mutation type. Two-sided P-values of <0.05 were considered statistically significant.

Ethical statement

All authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.