Matching-adjusted indirect comparison of ribociclib + nonsteroidal aromatase inhibitor versus abemaciclib + endocrine therapy in hormone receptor-positive/HER2-negative early breast cancer

A systematic literature review (SLR) was conducted by searching Embase, MEDLINE and CENTRAL databases (from database inception to 25 June 2024) to identify all relevant publications assessing the relative efficacy of adjuvant treatments for HR+/HER2- EBC. The search strings used to identify the relevant literature are described in Supplementary Table 1. The SLR was performed and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.

Forty-four unique studies were included in the clinical SLR. While patient-level data for the Breast International Group (BIG) 1-98 study investigating tamoxifen and letrozole in the adjuvant setting were available, because this study was from 2005 and had primary end points that were defined differently from those currently used, it was not considered further in this analysis [19]. Of all studies analyzed initially, six were identified that included a population at high risk of recurrence. Different parameters were used across these studies for segregating risk categories into low, high and mixed, which included nodal status, tumor size, histological grade, Ki-67 index, age and HR status. Of the six studies that included patients with high risk of recurrence, two studies were excluded from the indirect treatment comparison (ITC) as the risk classifications were not aligned with high-risk characteristics in the target population (Supplementary Table 2). Specifically, patients in SUCCESS-C were classified as having high risk disease with tumors ≥2 cm and no requirements regarding nodal involvement [20]. In a study analyzing patients who met the inclusion criteria of the TAILORx trial, the high-risk subpopulation was characterized by presence of metastases, which was an exclusion criterion for the ITC analyses [21]. The PENELOPE-B trial of palbociclib + ET and the SWOG S1207 and UNIRAD trials of everolimus + ET were excluded because these trials failed to demonstrate clinical benefit of the tested drugs, and these drugs were thus not approved for use in the adjuvant setting [22–24].

Finally, cohort 1 of monarchE met all eligibility criteria (patients with ≥4 positive lymph nodes or 1 to 3 positive lymph nodes with either tumor size ≥5 cm or histological grade 3 disease) for the present study and had reported positive results for the comparator of interest and was, therefore, included in the ITC. In addition, access to patient-level data from the NATALEE trial allowed for selection of patients who met the cohort 1 criteria. Thus, the NATALEE trial, assessing ribociclib + NSAI versus NSAI alone, and the monarchE trial, assessing abemaciclib + ET versus ET alone, were finally considered suitable for inclusion in this ITC.

This ITC used individual patient data (IPD) from the NATALEE trial (data cutoff, 29 April 2024; median follow-up, 49.6 months) and aggregate data reported for cohort 1 of the monarchE trial (data cutoff, 3 July 2023; median follow-up, 54.0 months) [12,13,17,25,26]. For context, the sample sizes and number of iDFS and DRFS events per outcome in the primary publications of these clinical trials for these data cutoffs are described in Supplementary Table 3.

The intent-to-treat (ITT) population of NATALEE included adults with high-risk node-positive and node-negative HR+/HER2- EBC. Specifically, those included in the NATALEE trial were patients with anatomic stage II or III disease regardless of nodal status; patients with stage IIA N0 (T2 N0) disease were required to have grade 3 disease, or grade 2 disease with evidence of additional high-risk features (i.e., Ki-67 of ≥20% or determination of high risk by genomic assay: Oncotype DX Breast Recurrence Score of ≥26 or high-risk scores by Prosigna/PAM50, MammaPrint or EndoPredict) [9,27]. In contrast, monarchE was composed exclusively of adult patients with high-risk node-positive HR+/HER2- EBC. Cohort 1 of monarchE included patients with ≥4 positive lymph nodes or 1 to 3 positive lymph nodes with at least one of the following features: tumor size ≥5 cm or histological grade 3 disease regardless of Ki-67 status. Cohort 2 of monarchE included patients with 1 to 3 positive lymph nodes, with Ki-67 of ≥20%, tumor grade <3, and tumor size <5 cm [10,13,28]. Table 1 provides the key eligibility criteria for the two trials from which the target populations (monarchE cohort 1-eligible patients) were compared in this MAIC.

Cohort 1 of monarchE corresponds to the patient population in which abemaciclib + ET has received positive recommendations from the National Institute for Health and Care Excellence (NICE) and the European Medicines Agency [16,29]. Therefore, abemaciclib was approved to only be available to these patients in clinical practice, and thus cohort 1 was the focus of this indirect comparison. Patients in NATALEE who met the key eligibility criteria of monarchE cohort 1 were selected (selected NATALEE population).

Following the SLR, the two trials (NATALEE and monarchE) considered suitable for inclusion in this ITC were compared regarding the efficacy and safety with ribociclib + NSAI versus abemaciclib + ET as adjuvant treatment for patients with high-risk HR+/HER2- EBC. A feasibility assessment, as described below, determined that an unanchored MAIC was the most appropriate approach for comparing relative efficacy (iDFS and DRFS) and safety between ribociclib + NSAI and abemaciclib + ET in the target population (monarchE cohort 1-eligible patients), which served as the primary analysis. Sensitivity analyses were performed to test the robustness of the assumptions made in the primary analysis.

Prior to performing the ITC analyses, a feasibility assessment was conducted. The feasibility assessment considered differences in the control ET arm and differences in efficacy and safety follow-up durations between NATALEE and monarchE. The following components of ITCs were considered: pairwise meta-analysis versus network meta-analysis, anchored versus unanchored ITC approaches and application of population-adjustment methods. A pairwise meta-analysis was deemed necessary as only two trials, NATALEE and monarchE, could be included; no other trials identified by the SLR reported outcomes of interest for the target population of the ITC.

The possibility of using an anchored approach was also assessed. The NATALEE control arm comprised NSAI only (i.e., letrozole and anastrozole) while the monarchE control arm comprised NSAIs, exemestane and tamoxifen. There is evidence that the efficacy of NSAIs and exemestane are similar in EBC [30,31]. However, the results of previous meta-analyses have shown that tamoxifen has worse efficacy compared with NSAIs as adjuvant treatment for HR+/HER2- EBC [32,33]. As such, an anchored MAIC would have required restricting analysis to abemaciclib + AI, excluding patients who received tamoxifen, who comprised approximately 30% of the monarchE trial population [10]. This restriction would not reflect the fully approved indication of abemaciclib in clinical practice, thus limiting generalizability [14,17].

Finally, a comparison of the NATALEE and monarchE study eligibility criteria and baseline characteristics revealed between-study heterogeneity that would potentially bias the ITC. Moreover, the monarchE trial had narrower eligibility criteria than NATALEE and only had aggregate data published; therefore, it was necessary to select NATALEE patients who met the monarchE cohort 1 eligibility criteria. After matching NATALEE patients to monarchE (the selected NATALEE population), residual differences in many baseline characteristics were observed. These differences could bias the ITC if the characteristics were treatment effect modifiers or important prognostic factors for EBC. Unanchored ITCs assume all prognostic factors and effect modifiers are accounted for and are therefore subject to bias or confounding from differences in such characteristics between study populations [34]. Population adjustment methods such as an unanchored MAIC reduce the risk of bias due to differences in characteristics between studies and thus were chosen for this analysis. Based on these findings from the feasibility assessment, an unanchored MAIC was considered the most appropriate approach for comparing ribociclib + NSAI versus abemaciclib + ET in the target population (monarchE cohort 1-eligible patients).

Characteristics that were feasible to be included in the MAIC were identified from published studies of the monarchE trial. The importance of baseline characteristics as effect modifiers or prognostic factors for efficacy was determined qualitatively and quantitatively. External clinicians were interviewed to determine the baseline characteristics that are known to be prognostic factors for efficacy in EBC, by varying degrees of influence. Each of the three clinicians interviewed were able to identify a number of factors that they considered to be relevant as prognostic factors based on their expertise and experience. The clinicians interviewed were selected based on their clinical expertise in treating BC patients and covering different regions as well as for their prior experience with health technology assessments in BC. Supplementary Table 4 shows the clinician ranking of prognostic factors by levels of ‘high importance’, ‘low importance,’ and not important (i.e., ‘—’) for a list of variables used in primary and sensitivity analyses. The identified factors generally corresponded to prognostic factors determined to be important in a prior analysis from NATALEE [35]. Conversely, confounding factors for safety outcomes are generally less established than those for efficacy. All prognostic factors identified for efficacy outcomes were thus used for weighting the IPD from the NATALEE trial for safety outcomes as well.

Using IPD from NATALEE, p-values were also generated as a tool to distinguish between treatment effect modifiers and prognostic factors using IPD from NATALEE. Treatment effect modification was assessed for each clinical characteristic based on p-value for the interaction effects with treatment using Cox proportional hazards regression of patient-level failure time data for the iDFS outcome. Cox models were specified with covariates for treatment, a given clinical characteristic, and the interaction of treatment arm with that characteristic. The p-values for the interaction terms were then used to assess whether the treatment effect with ribociclib + NSAI was modified based on an α of 0.05. A p-value > 0.05 for a characteristic suggested that there was no adequate evidence to treat it as an effect modifier on iDFS (Supplementary Figure 1). To help inform selection of effect modifiers and prognostic factors, p-values from interaction and log-rank tests were used as descriptive tools to help inform selection of effect modifiers and prognostic factors. However, these should not be interpreted as confirmatory due to the exploratory nature of the analysis.

Statistical assessments for prognostic factors were conducted based on p-values obtained from log-rank tests for iDFS outcomes in the selected NATALEE patient population, irrespective of treatment. The log-rank p-value was used to determine whether there was supporting evidence for a difference in iDFS events among the Kaplan–Meier (KM) curves with respect to the characteristic. A p-value < 0.05 suggested that the characteristic was prognostic. All quantitative assessments for treatment effect modification and prognostic factors are summarized in Supplementary Table 5.

To enable MAIC, patients were weighted by inverse odds of propensity scores (which measure the probability of the patient being included in NATALEE) using available patient-level data (from NATALEE) to adjust the distribution of patient characteristics (prognostic factors or effect modifiers) to match with the aggregate monarchE data (Figure 1) [36,37]. To elaborate, the weight assigned to each patient in the selected NATALEE population was proportional to the inverse of the probability or relative likelihood of their covariates occurring in the selected NATALEE population relative to monarchE cohort 1. Weights were assigned to individual patients in the ribociclib + NSAI treatment arm and NSAI alone treatment arm such that the mean baseline characteristics in the NATALEE trial matched with the abemaciclib + ET and ET alone arms of monarchE cohort 1, respectively. Given that only aggregate data were available for monarchE, it was possible to adjust only for characteristics that were available at baseline for cohort 1 of the monarchE trial in the literature.

The primary analysis for the unanchored MAIC was conducted using all the available characteristics shown in Table 2, consistent with guidance for unanchored MAICs from NICE Decision Support Unit Technical Support Documents (DSU TSDs) 18 [34,36,37]. According to the guidance, adjusting for factors that are not prognostic does not bias the unanchored comparison, but it may reduce effective sample size (ESS) unnecessarily. The primary analysis adjusted for all variables that were indicated as having prognostic influence by clinicians, all variables that modified treatment effect based on interaction tests, all variables that were prognostic based on stratified log-rank tests and available variables without evidence of prognostic influence (Table 2).

Additionally, sensitivity analyses were conducted to assess the range of plausible estimates for the relative treatment effects (Table 2). Two alternative sets of MAIC weights were estimated to explore the impact of adjusting for fewer variables (i.e., baseline disease and demographic characteristics). Variables included in these sensitivity analyses were based on clinical input and statistical tests for treatment effect modification and prognostic factors. Sensitivity analysis 1 included characteristics rated highly prognostic by two or more clinicians. Sensitivity analysis 2 included characteristics that had at least some prognostic value according to clinicians (i.e., low importance or high importance by at least 1 clinician) and all variables with statistically significant findings for interaction tests or tests of prognostic factors with evidence of prognostic influence based on statistical tests. Weights estimated in these sensitivity analyses were then used in the comparison of outcomes, following the same procedures as the primary analysis.

Results were estimated separately for MAIC with primary analysis weights, unweighted MAIC and sensitivity analysis weights applied for the selected NATALEE population. Covariate balance was assessed using standardized mean differences (SMDs) before and after weighting (Supplementary Tables 6 & 7). An SMD of <0.1 was considered a negligible difference, an SMD of 0.1 to <0.2 was considered a small difference (which is acceptable if it is not an effect modifier), an SMD of 0.2 to <0.5 was considered a moderate difference, and an SMD of ≥0.5 was considered a large difference.

Calculation of MAIC weights was conducted in R (R Core Team; 2013) using an adaptation of sample code from the NICE DSU TSD 18. Standardized MAIC weights (mean value equal to 1.0) were generated using the method of moments, with active (ribociclib + NSAI and abemaciclib + ET) treatment and control (NSAI only and ET only) arms matched separately to ensure that any residual imbalances between active and control arms in monarchE were replicated in the matched NATALEE sample.

A Cox regression model (semiparametric) was chosen to compare ribociclib + NSAI with abemaciclib + ET for the time-to-event analyses as this is a commonly used approach that can be specified as a proportional hazard model and allowed the estimation of hazard ratios. This univariate Cox proportional hazards model was fitted to MAIC-weighted IPD from the selected NATALEE patient population and reconstructed data from monarchE cohort 1. Outcomes for the ITCs were iDFS, DRFS and safety events that were classified as grade ≥3 treatment-emergent adverse events (TEAEs).

iDFS, in both the NATALEE and monarchE trial, was evaluated based on Standardized Definitions for Efficacy End Points (STEEP) criteria (definition consistent with STEEP 2.0) as the time from randomization until local/regional, contralateral, distant recurrence, second primary nonbreast malignancy, or death from any cause, whichever occurred first [38,39]. Likewise, DRFS in both the NATALEE and monarchE trial, also per STEEP(definition consistent with STEEP 2.0), was defined as time from randomization to distant recurrence or death from any cause, whichever occurred first [38,39]. For time-to-event outcomes, patients with no event by the last visit date were censored. Time-to-event data for iDFS and DRFS outcomes in patients in the monarchE trial were obtained by digitizing published KM curves. Pseudo-IPD from monarchE cohort 1 were reconstructed via digitalization of published KM curves using the Guyot algorithm. The reconstructed pseudo failure time data for abemaciclib + ET and ET only arms from cohort 1 of monarchE were assessed by comparing the numbers of patients at risk over time and landmark survival estimates at 24, 36, 48 and 60 months against those reported in the primary publications for monarchE. The number of patients at risk in the reconstructed KM curves for abemaciclib arms matched with the numbers reported in the monarchE publication and the landmark survival estimates matched within +/- 0.001 of the reported survival rates at 24, 36, 48 and 60 months validating the reconstructed data [12]. While visually similar to published survival curves, reconstructed data may not fully replicate original IPD. KM curve estimation and Cox regression were conducted using SAS (SAS Institute; Cary NC) [40].

For safety analysis, grade ≥3 TEAEs with an incidence of ≥5% were calculated using logistic regression, and ORs for these TEAEs between each treatment were estimated.

The validity of the proportional hazards assumption between ribociclib + NSAI and abemaciclib + ET, before and after weighting, was tested by a visual inspection of the log-cumulative hazard plots as well as by the Schoenfeld global test of proportionality. Schoenfeld residuals were generated to assess the proportional hazards assumption for iDFS and DRFS with ribociclib + NSAI versus abemaciclib + ET before and after weighting. If the proportional hazards assumption was violated, survival curves were still presented descriptively to support visual comparisons of time-to-event trends; however, interpretation of the hazard ratios may be limited.

Among the 5101 patients from the NATALEE trial ITT population, 3307 patients (ribociclib + NSAI, n = 1658; NSAI alone, n = 1649) met the key eligibility criteria for cohort 1 of monarchE (n = 5120; abemaciclib + ET, n = 2555; ET alone arm, n = 2565). MAIC weights were estimated adjusting for all baseline characteristics, and most weights for both treatment arms were below 5.0 in the primary analysis (Supplementary Figure 2). There were 22 patients in the selected NATALEE population with an estimated weight of 0.0, suggesting that they did not have sufficient overlap with those in monarchE cohort 1 to be meaningfully reweighted. Most of the estimated weights for both treatment arms were below 5.0. The maximum weight values in the ribociclib plus NSAI arm and the NSAI only arms were 16.2 and 24.0, respectively. For the primary analysis, after weighting, the ESS for the ribociclib + NSAI and NSAI only arms were 448 and 453, respectively, constituting an estimated reduction of 72.5%–73% from the selected sample size in each arm. Given that the ESS remained above 400 per treatment arm after weighting, use of the MAIC with full covariate adjustment was deemed appropriate. Table 3 is an attrition table for the selected NATALEE population included in this ITC to compare with the monarchE cohort 1 population.

The ESS for sensitivity analysis 1, which only included highly prognostic variables, was 821 patients for the ribociclib + NSAI arm and 767 for the NSAI alone arm in the selected NATALEE population, constituting an estimated reduction of 50.5% and 53.5% from the selected sample size in each arm, respectively. The ESS for sensitivity analysis 2, which included any variables with some prognostic value, was 617 patients for the ribociclib + NSAI arm and 620 patients for the NSAI alone arm, constituting an estimated reduction of 62.8% and 62.4% from the selected sample size in each arm, respectively.

The two trials were well balanced for many of the reported characteristics at baseline before adjustment for the selected NATALEE population, with SMDs from the monarchE cohort 1 patient population below 50% for most characteristics (Supplementary Table 6). For two characteristics (region and missing Ki-67 status), the SMDs exceeded 50%, which were adjusted. The SMDs exceeded 20% for the proportions of patients with a Ki-67 index of ≥20%, those residing in geographic region reported as ‘other’ (i.e., not North America, Europe or Asia), those whose ethnicity was not Hispanic or Latino, those with stage IIA or IIIA disease, and those with prior neoadjuvant chemotherapy. An SMD between 20% and 50% suggests differences between studies (i.e., not insignificant), which were adjusted [40]. After weighting, the SMDs between the patient characteristics of the weighted NATALEE population and the monarchE cohort 1 population were relatively small or zero (Supplementary Table 7).

Based on the MAIC analyses, no statistically significant difference was observed in the iDFS between the ribociclib + NSAI versus abemaciclib + ET arm in the target populations (monarchE cohort 1-eligible patients). After applying the MAIC weights, Cox regression yielded similar iDFS with the ribociclib + NSAI arm in the selected NATALEE population versus the abemaciclib + ET arm in the monarchE cohort 1 subset (hazard ratio: 0.901; 95% CI: 0.678–1.197; nominal p = 0.470) (Figures 2 & 3). Likewise, unweighted iDFS with ribociclib + NSAI was similar in the ribociclib + NSAI arm and the abemaciclib + ET arm in the target population (hazard ratio: 1.055; 95% CI: 0.887–1.255; nominal p = 0.545) (Figure 3). Comparisons of the NSAI only arm in the selected NATALEE population versus the ET only arm in cohort 1 of monarchE yielded similar iDFS for the weighted comparison (hazard ratio: 0.890; 95% CI: 0.703–1.126; nominal p = 0.332) and unweighted analysis (hazard ratio: 0.987; 95% CI: 0.849–1.147; nominal p = 0.862) (Figure 3). Schoenfeld residuals of weighted and unweighted iDFS in the ribociclib + NSAI arm versus the abemaciclib + ET arm (weighted: p = 0.667; unweighted: p = 0.516) for the population analyzed suggested that the proportional hazards assumption was not violated (Supplementary Figure 3).

Findings from sensitivity analyses for iDFS were generally consistent with the primary analysis (Figure 2 & Supplementary Table 8). For sensitivity analysis 1, Cox regression yielded a hazard ratio of 0.937 (95% CI: 0.743–1.182; nominal p = 0.545) for weighted iDFS between the ribociclib + NSAI arm of the selected NATALEE population and the abemaciclib + ET arm of monarchE cohort 1. For sensitivity analysis 2, Cox regression yielded a hazard ratio of 0.933 (95% CI: 0.725–1.201; nominal p = 0.591) for weighted iDFS between the ribociclib + NSAI arm and the abemaciclib + ET arm.

Based on the MAIC analyses, there was insufficient evidence to suggest differences in DRFS between the ribociclib + NSAI versus abemaciclib + ET analysis in the target populations (monarchE cohort 1-eligible patients). Weighted DRFS was similar in the ribociclib + NSAI selected NATALEE population compared with abemaciclib + ET in the monarchE cohort 1 subset (hazard ratio: 0.946; 95% CI: 0.686–1.305; nominal p = 0.734) (Figures 4 & 5). Unweighted DRFS with ribociclib + NSAI in the selected NATALEE population was similar to that with abemaciclib + ET in the monarchE cohort 1 subset (hazard ratio: 0.980; 95% CI: 0.811–1.184; nominal p = 0.832) (Figures 4 & 5). Comparisons of the NSAI alone arm in the selected NATALEE patient population versus the ET alone arm in monarchE cohort 1 before and after weighting yielded similar DRFS for unweighted (hazard ratio: 0.920; 95% CI: 0.782–1.083; nominal p = 0.315) and weighted comparison (hazard ratio: 0.828; 95% CI: 0.649–1.058; nominal p = 0.131) (Figure 5). Schoenfeld residuals of unweighted and weighted DRFS in the ribociclib + NSAI arm versus the abemaciclib + ET arm (unweighted: p = 0.014; weighted: p < 0.001) for the trial populations analyzed suggested that the proportional hazards assumption may have been violated (Supplementary Figure 4).

Findings of sensitivity analyses for DRFS were generally consistent with the primary analysis (Figure 4 & Supplementary Table 9). For sensitivity analysis 1, weighted analysis yielded a hazard ratio of 0.892 (95% CI: 0.689–1.155; nominal p = 0.385) for DRFS between the ribociclib + ET arm of the selected NATALEE population and the abemaciclib + ET arm of monarchE cohort 1. For sensitivity analysis 2, weighted analysis yielded a hazard ratio of 0.879 (95% CI: 0.662–1.166; nominal p = 0.370) for DRFS between the ribociclib + NSAI arm of NATALEE and the abemaciclib + ET arm.

Comparisons of grade ≥3 TEAEs suggested lower odds (OR <1) for diarrhea, leukopenia and lymphopenia before and after MAIC weighting in patients treated with ribociclib + NSAI in the selected NATALEE population compared with abemaciclib + ET in monarchE cohort 1 (Figure 6 & Supplementary Table 10). For increased alanine aminotransferase (ALT) and neutropenia, the analysis suggested significantly greater odds (OR >1) in the ribociclib + NSAI arm compared with the abemaciclib + ET arm.

Results of the sensitivity analyses for the safety outcomes were also consistent with those observed in the primary analyses (Supplementary Table 11). For sensitivity analysis 1, as seen in the primary analysis, increased ALT and neutropenia were the only TEAEs for which ribociclib + ET had significantly greater odds than abemaciclib + ET in the population analyzed. For sensitivity analysis 2, consistent with primary analysis, ribociclib + ET had statistically significantly greater odds for increased ALT and neutropenia than abemaciclib + ET.

This analysis encountered several limitations related to differences in these adjuvant CDK4/6i trials. First, unlike anchored ITCs, where differences in unknown prognostic factors between trials do not affect the validity of the comparison, the current unanchored ITC analysis – while using MAIC weights to adjust for all available baseline characteristics reported for monarchE – was limited by the lack of access to IPD from monarchE. Thus, even though an extensive number of reported patient baseline characteristics from monarchE were taken into account, it is possible that there were unreported prognostic factors that were not possible to include in the estimation of MAIC weights. This poses a risk of residual confounding, which is an inherent limitation of all unanchored comparisons [51]. Second, the SMD for missing Ki-67 test results exceeded 50%, while for Ki-67 index of ≥20%, it was 32.1%. As the Ki-67 index was one of the characteristics with clinician consensus on prognostic importance, the imbalance could introduce bias if not fully adjusted despite weighting. Third, the findings of this study could be biased to the extent that there is any residual confounding after MAIC weighting. Fourth, patients from NATALEE who would be eligible for inclusion in the monarchE trial constituted a subset of the total ITT NATALEE patient population. After selection and weighting the MAIC, a reduction in the ESS from the original trial sample size may have led to loss of statistical power and a break in randomization from the original NATALEE ITT population. Fifth, due to its exploratory nature, unmeasured confounders and differences in safety reporting, this cross-trial safety comparison should be interpreted with caution. Also, a longer AE follow-up period in NATALEE than monarchE may have led to overestimation of ORs for AEs with longer follow-up. However, for early-onset AEs such as diarrhea, the shorter follow-up in monarchE may still suffice, supporting the validity of OR comparisons. Potential instability of estimates for rare AEs was also a limitation. Sixth, the proportional hazards assumption may have been violated for DRFS. As a result, hazard ratio estimates should be interpreted alongside survival curve shapes and absolute risk differences, rather than as constant relative risk over time. Finally, another limitation is the differential follow-up time between the NATALEE and monarchE trials, which was a modest 4 month difference in follow-up that is unlikely to bias efficacy outcome comparisons.

Some of the other limitations of this ITC are commonly noted in such comparisons. The KM curves from the monarchE trial used in this analysis were based on reconstructed IPD. While the KM curves estimated from the reconstructed data were similar to those reported in the monarchE publications, the pseudo-failure time data may not have matched the patient-level failure time data with complete accuracy. However, as the differences between the reported hazard ratios and the hazard ratios estimated from pseudo-failure time data were relatively small, they were unlikely to bias the findings reported here.