QSP Modeler Sanofi Cambridge, Massachusetts, United States
Disclosure(s):
Kyeri Sung: No financial relationships to disclose
Objectives: Hidradenitis suppurativa (HS) is a chronic inflammatory skin disease characterized by the development of multiple types of painful lesions such as abscesses, inflammatory nodules, and draining tunnels. A predictive model was developed to translate biomarkers simulated by a HS QSP model into a clinical endpoint, the Hidradenitis Suppurativa Clinical Response (HiSCR), a commonly-used metric of therapeutic response [1]. This predictive model consists of three sub-models which estimate the counts of three specific types of HS lesions based on a biomarker profile. To facilitate drug development efforts in HS, biomarkers from a virtual population (VP) of HS patients generated from a HS QSP model was linked to this predictive model to bridge between mechanism of action and clinical response.
Methods: Separate ensemble models were developed for three types of HS lesions: abscesses, inflammatory nodules, and draining tunnels. Each lesion model was trained using individual patient-level data, with biomarker measurements selected based on population-level correlations with counts for each lesion type. Models were tested using k-fold cross-validation and further validated using independent clinical data. Biomarker profiles from a representative HS VP were simulated for regimens of approved HS treatments, and the lesion models predicted the temporal changes in abscesses, inflammatory nodules, and draining tunnels. Changes in lesion counts were used to calculate the VP HiSCR response as defined [2].
Results: The three individual lesion models displayed adequate predictive performance on validation data, as evaluated by standard metrics including area under the ROC curve. The accuracy of the lesion count predictions from biomarker treatment data was sufficient to recapitulate the observed HiSCR in the validation data. When treated biomarker profiles from baseline-matched VPs were used as input, the combined model matched clinical HiSCR observations associated with these treatments. This analysis supplements clinical observations by linking biomarker measurements with lesion response and the HiSCR.
Conclusions: By leveraging inflammation-modulated biomarker measurements that have been independently associated with HS, this model describes three types of HS lesions which are used to calculate the HiSCR. To our knowledge, this is the first predictive model for HS lesions. This model also has the capacity to be extended beyond the HiSCR to other clinical scores used in HS clinical practice that are calculated using lesion counts.
Citations: [1] Daoud et al. (2023). DOI: 10.3389/fmed.2023.11451522. [2] Kimball et al. (2014). DOI: 10.1111/bjd.13270.
