Both the B-LONG and PROLONG-9FP trials were non-randomized, open-label studies that enrolled male subjects aged ≥ 12 years with severe or moderately severe HB (endogenous FIX ≤ 2 IU/dL) (Table 1) [7, 8]. Both studies were conducted in accordance with the Declaration of Helsinki and local regulations, the protocols were approved by the authorities and the institutional review board/ethics committee at each participating center, and signed informed consent was obtained from all patients. The primary efficacy endpoint of B-LONG was ABR. The primary efficacy endpoint of PROLONG-9FP was annualized spontaneous bleeding rate (AsBR) for bleeding episodes treated on-demand versus routine prophylaxis.

A qualitative ITC feasibility assessment was conducted on the study design, eligibility criteria, baseline characteristics, and outcomes and their definitions from the trials. Overall, the studies were comparable by matching and adjusting the IPD of PROLONG-9FP to the SLD of B-LONG. Subjects that were on weekly prophylaxis treatment were selected as the population of interest given that it was the only dosing regimen comparable between the two trials. From B-LONG, only subjects on weekly prophylaxis (B-LONG group 1) were selected for this MAIC (Table 1). From PROLONG-9FP, IPD was leveraged to combine subjects (1) who received weekly prophylaxis (PROLONG-9FP group 1, excluding data on prophylaxis every 10 or 14 days from subjects who switched after 26 weeks of weekly prophylaxis; Table 1) and (2) who received weekly prophylaxis after 26 weeks of on-demand treatment (PROLONG-9FP group 2; Table 1) [8]. This created a dataset that included complete data on weekly prophylaxis (i.e., excluding data not on weekly prophylaxis) and also included subjects who received either prior prophylaxis or prior on-demand treatment, from both PROLONG-9FP and B-LONG.

Individual patient data from the PROLONG-9FP trial were used to match and adjust subjects to the population of the B-LONG trial. No subjects were removed from PROLONG-9FP during matching given that the inclusion criteria of B-LONG was either comparable or broader. After matching, subjects from PROLONG-9FP were weighted using a method-of-moments propensity score algorithm to adjust for disease severity (endogenous plasma FIX level of < 1 IU/dL or 1–2 IU/dL), age, prior FIX regimen, and BMI so that the means and standard deviations (SDs), and proportions, were comparable to those in the B-LONG trial. The baseline characteristics selected for adjustment were based on previous literature, clinical input, and statistical performance assessments. The performance and suitability of each MAIC model were assessed on the basis of the effective sample size (ESS), which represents the number of non-weighted patients that would produce a treatment effect estimate with the same precision as the weighted sample estimate and is derived as the sum of patient weights squared divided by the sum of squared patient weights [11], and the distribution of patient weights with a particular focus on the presence of extreme values. A low ESS compared to the original sample size indicated large differences in patient weights due to large imbalances in patient populations prior to weighting. Analyses adjusting for prior ABR were explored, as this was also identified as a key baseline characteristic; however, the analyses could not be conducted because of substantial differences between trials causing extreme weights and a reduction in ESS to insufficient levels. In lieu of adjustment for prior ABR, prior treatment regimen may be a proxy for this factor.

Six outcomes were assessed in this study: ABR, AsBR, and annualized joint bleeding rate (AjBR); proportion of subjects without bleeds (overall), spontaneous bleeds, and joint bleeds. Both trials aligned on the definition for each outcome and only bleeding episodes that were treated were included in the analyses [12]. Annualized bleeding rates for all three bleeding types were derived as

$$\frac}}\times 365.25.$$

Analyses for each efficacy outcome included a naïve comparison and multivariable analyses which adjusted for 2–4 of the prognostic factors: disease severity, age, prior FIX regimen, and BMI. The analysis which adjusted for all four baseline characteristics was considered the full analysis.

Estimates of the comparative efficacy of rIX-FP versus rFIXFc were based on the difference between (a) an estimate of the outcome of interest for subjects in the comparator study (B-LONG) had they received rIX-FP and (b) the estimated outcome based on published SLD from the B-LONG trial. After subjects from the comparator trial were matched, a weighted estimate of the outcome with the PROLONG-9FP data was derived using a weighted, intercept-only generalized linear model. Specifically, a negative binomial distribution with log link was used for the ABR, AsBR, and AjBR outcomes, and a logistic distribution with logit link was used for binominal outcomes (i.e., proportion of patients without bleeding events, spontaneous bleeding events, and joint bleeding events). The intercept represents an estimate of the outcome of interest had patients from the comparator trial received rIX-FP. Robust standard errors (SEs) were estimated using the sandwich estimator with the R package “sandwich”, and relative treatment effects on the linear predictor scale (i.e., log-rate ratios [log-RR] and log-odds ratios [log-OR]) were derived by taking the difference between this estimated outcome based on SLD from the B-LONG trial [7, 13]. The variance of the relative treatment effect between rIX-FP and rFIXFc was estimated as the sum of the variance of the individual estimators for each treatment included in the comparison. The SEs were used to construct two-sided 95% Wald CIs based on normal approximation. Relative treatment effects (i.e., RR and OR) and CIs were transformed to the natural scale after estimation.

All analyses were conducted using R® version 3.6.1.

This article is based on two previously published phase 3 trials (PROLONG-9FP and B-LONG), does not contain any new studies with human participants or animals performed by any of the authors, and thus did not require ethics approval. Both PROLONG-9FP and B-LONG studies were conducted in accordance with the Declaration of Helsinki and local regulations, the protocols were approved by the authorities and the institutional review board/ethics committee at each participating center, and signed informed consent was obtained from all patients. Informed consent was not required for this analysis given the deidentified nature of the PROLONG-9FP individualized patient-level data and the use of anonymized, previously published data for the B-LONG study.