Response: Confounding by indication is unlikely to explain the higher inhibitor incidence in boys treated with a recombinant FVIII product

In January 2013, the Research of Determinants of Inhibitor Development (RODIN) study unexpectedly showed a higher inhibitor incidence in previously untreated patients (PUPs) with severe hemophilia A treated with Kogenate FS (Bayer) (also named Helixate NexGen; Product D in this letter) than in those treated with Advate (Product E).¹  Two other groups in charge of national hemophilia cohorts in the United Kingdom and France recently published similar findings.^2,3  Given the lack of an obvious pathophysiologic mechanism, possible biases have been raised.^1-5  One of the most plausible is confounding by indication (CbI),⁶  whereby Product D might have been preferred for initial treatment of PUPs with a high a priori inhibitor risk, after publication of 2 articles reporting a low inhibitor rate with this product.^7,8 

Berntorp and Iorio further discuss this possible bias,⁹  postulating that (1) this bias might have existed in only a few hemophilia treatment centers (HTCs), and (2) prescribers might have selected “at-risk” patients based on “subtle nuances” not recorded in cohort studies and therefore not considered in multivariate analyses. They analyzed our tabulated data, comparing inhibitor rates between Products D and E, first in 3 selected HTCs (in which at least 10 patients were treated first with Product D or E and where an inhibitor rate of at least 50% was observed with Product D), and then in the remaining 30 HTCs. We performed survival analyses in these 2 HTC groups, adjusted for the same risk factors as in our article.³  Adjusted hazard ratios (aHRs) for Product D compared with Product E (D/E) were 3.20 (95% confidence interval [CI], 0.93-11.0) for the first 3 HTCs and 1.23 (95% CI, 0.68-2.23) for the remaining 30 HTCs.

In the early 2000s, the only well-known risk factors for inhibitor development were genetic, namely, the F8 gene defect, a family history of inhibitors, and ethnic origin. If some prescribers had indeed preferred Product D for at-risk PUPs, an association should have emerged between these genetic risk factors (if known at the first factor VIII [FVIII] infusion) and the chosen product. However, no consistent trend has been found, either in the entire sample of the 3 published studies^1-3  or in the aforementioned subgroups of French HTCs (Table 1). Furthermore, when a product is deemed less immunogenic (eg, plasma-derived FVIII products), its preferential prescription to at-risk patients is apparent (see RODIN results, Table 1).^1,5  Some subtle nuances in socioeconomic conditions or practical modalities of initial treatment, as highlighted by Berntorp and Iorio,⁹  might correlate with inhibitor risk factors, but their independent association with inhibitor development remains to be demonstrated. Nevertheless, it would be rather odd for preferential product prescription to be based on such characteristics and not on acknowledged genetic risk factors.

The first article showing a low inhibitor rate with Product D included limited numbers of PUPs of white European origin and minimally treated patients (n = 31), followed until the 20th exposure day.⁷  Only 3 French HTCs (6 patients in total) participated in this study. The second article reported 30 additional American PUPs and had similar limitations.⁸  Because of these limitations, it could be considered that the observed inhibitor proportion (9/60) underestimated the real-life risk with Product D. The clinicians in charge of the 3 aforementioned French HTCs (J.G., P.L., and C.R.) certify that they never preferentially prescribed Product D to at-risk PUPs. Nevertheless, these clinicians and other French clinicians we interviewed stated that their product choice for PUPs was influenced by other factors, such as FVIII product shortages (eg, Product D production was reduced in 2001 after an inspection by the US Food and Drug Administration), their willingness to use various brands of FVIII products in their HTC, and practical considerations (eg, some clinicians preferentially chose Product D for children owing to its lower injected volume compared with Product E). The numbers of PUPs first treated with Products D and E per HTC show no discernible temporal pattern (supplemental Data, available on the Blood Web site).

Although the CbI hypothesis is unlikely, it cannot be formally excluded. To explore it further, we repeated our primary analysis after incorporating propensity scores (PS)^6,10  based on inhibitor risk factors known at the first FVIII infusion (see characteristics shown for the FranceCoag Network, Table 1). Estimated D/E HRs obtained with a Cox model adjusted on PS were slightly lower than our published results³ : crude HR 1.59 (95% CI, 1.02-2.48) and aHR 1.46 (95% CI, 0.88-2.41). Although this statistical technique cannot consider unmeasured confounders, these preliminary results do not support a major CbI.

Given the heterogeneity of inhibitor rates observed with each product in the different HTCs (mainly because of limited patient numbers), a very broad spectrum of relative risk estimates can be obtained by selecting HTC subgroups. Without a clear demonstration of CbI, any calculation based on arbitrary HTC subgroups remains unconvincing.

We agree that a properly conducted randomized trial comparing Products D and E would provide stronger evidence of a difference in immunogenicity. Unfortunately, implementation of such ambitious comparative trials in PUPs appears difficult in Western countries.¹¹  Another approach would be to identify a pathophysiologic mechanism for the suspected difference, and this is why we called for nonclinical studies.³  If a difference in immunogenicity exists, the 2 main issues would be (1) to assess whether the immunogenicity of Product D has been stable since market release, and (2) to determine whether production in baby hamster kidney cells is involved.