Performance of a clinical risk prediction model for inhibitor formation in severe haemophilia A. Issue 4 (14th May 2021)

• Record Type: Journal Article
• Authors: Hassan, Shermarke; Palla, Roberta; Valsecchi, Carla; Garagiola, Isabella; El‐Beshlawy, Amal; Elalfy, Mohsen; Ramanan, Vijay; Eshghi, Peyman; Karimi, Mehran; Gouw, Samantha Claudia; Mannucci, Pier Mannuccio; Rosendaal, Frits Richard; Peyvandi, Flora
• Is Part Of: Haemophilia. Volume 27:Issue 4(2021)
• Page Start: e441
• Page End: e449
• Publication Date: 2021-05-14
• Subjects: factor VIII -- haemophilia A -- immunogenicity -- inhibitors -- prediction; Hemophilia -- Periodicals; 616.1572005
• DOI: 10.1111/hae.14325
• Languages: English
• ISSNs: 1351-8216

Abstract

Abstract: Background: There is a need to identify patients with haemophilia who have a very low or high risk of developing inhibitors. These patients could be candidates for personalized treatment strategies. Aims: The aim of this study was to externally validate a previously published prediction model for inhibitor development and to develop a new prediction model that incorporates novel predictors. Methods: The population consisted of 251 previously untreated or minimally treated patients with severe haemophilia A enrolled in the SIPPET study. The outcome was inhibitor formation. Model discrimination was measured using the C‐statistic, and model calibration was assessed with a calibration plot. The new model was internally validated using bootstrap resampling. Results: Firstly, the previously published prediction model was validated. It consisted of three variables: family history of inhibitor development, F8 gene mutation and intensity of first treatment with factor VIII (FVIII). The C‐statistic was 0.53 (95% CI: 0.46–0.60), and calibration was limited. Furthermore, a new prediction model was developed that consisted of four predictors: F8 gene mutation, intensity of first treatment with FVIII, the presence of factor VIII non‐neutralizing antibodies before treatment initiation and lastly FVIII product type (recombinant vs. plasma‐derived). The C‐statistic was 0.66 (95 CI: 0.57–0.75), and calibration was moderate. Using a model cut‐off point of 10%, positive‐ and negative predictive values were 0.22 and 0.95, respectively. Conclusion: Performance of all prediction models was limited. However, the new model with all predictors may be useful for identifying a small number of patients with a low risk of inhibitor formation.