Abstract

Predictive Modelling: Identifies Response To Fixed-Dose N8-Gp Prophylaxis In The First Weeks Of Treatment As A Predictor Of Long-Term Outcomes

Author
Pratima Chowdary, Soraya Benchikh El Fegoun, Milan S. Geybels, Andreas Tiede
Citation

(2021), Abstract. Haemophilia, 27: 18-181. https://doi.org/10.1111/hae.14236

Date of Publication
Abstract

Introduction: The aim of this study was to demonstrate the value of fixed regimen prophylaxis with N8?GP (turoctocog alfa pegol; Esperoct®, Novo Nordisk A/S, Bagsværd, Denmark), using predictive modelling with pathfinder 2 trial data.

Methods: Predictive models were designed to use patient characteristics to predict outcome at the end of the main phase. Predictions after ~1.5 years under prophylactic treatment were made using data collected prior to as well as up to 12 weeks after treatment initiation.

We included patients who received N8?GP prophylaxis (50 IU/kg every 4 days) with sufficient treatment follow?up. Models classified main phase outcome of interest per patient (annualized bleeding rate [ABR] ?1 or > 1). Supervised classification models (random forests, ridge regression, gradient boosted trees) were developed to discriminate between these groups based on patient data.

Contribution of variables to model performance were interpreted at the group?level to describe overall model behaviour, and at patient?level to explain individual predictions using Shapley additive explanations. Models were developed across five windows, incrementally increasing from baseline and screening visit (pre?treatment), to visit 3 through 5 (post?treatment). Area under the receiver operator characteristic curve (AUC?ROC) was used to assess model performance.

Results: Predictive models were developed on a cohort of 161 to 166 (dependent on modeling time period) previously treated adult patients (>12 years of age; median treatment period 14 months). The most performant model was observed at visit 5; i.e. 12 weeks after treatment initiation with an AUC of 0.79. The cumulative bleed count observed during the initial 12 weeks period post?initiation was the most informative variable, with this variable alone achieving an AUC of 0.75. Baseline Von Willebrand factor and mean FVIII at 30 mins across visits were also highly ranked a strong driver for predictions for predicting ABR after ~14 months of treatment. Pharmacokinetic data did not have a strong impact on the predictive performance.

Discussion/Conclusion: Predictive modelling provides insights into the potential to identify long?term clinical outcome from trial data and evidence for the potential of N8?GP to provide simplified prophylaxis in the treatment of patients with haemophilia A.

This website is intended
for healthcare professionals (HCPs)
outside of the U.S.

Take me to www.novonordisk.com