Abstract Number: PB0993
Meeting: ISTH 2020 Congress
Background: Patients enrolled into clinical trials are a well characterised cohort. Typical safety/efficacy endpoints are critical for product licensing, which requires guidance on dosing in routine clinical care. However, there is no requirement to review data to identify predictors of good/poor outcome that can influence routine clinical practice.
Aims: Investigate the feasibility of post hoc analyses of pathfinder 2 cohorts using a new hybrid clinical-machine learning (CML) technology to identify treatment and patient characteristics, which predict good and poor patient outcomes, the latter particularly pertinent as it allows for treatment intensification.
Methods: Patients who participated in pathfinder 2 were included. Treatment and patient characteristics that influence outcomes were identified. Clinical measures relevant for predicting patient outcomes (baseline demographics, laboratory/clinical measurements and pharmacokinetics) were assessed by supervised CML using proprietary software.
Results: Preliminary insights indicate that the data are sufficient to support supervised CML. Joint status and target joints were re-evaluated and the resolution was observed during the main trial phase. Distinct patient profiles emerged, particularly around their previous treatment regimen. Utilising CML, adult/adolescent subgroups were defined by segmentation of baseline annualised bleeding rate (ABR) for 150 patients who received 50 IU/kg N8-GP every 4 days and evaluated over 18 months. A distinct group was patients with a baseline ABR ≥20 who achieved a positive ABR outcome. Interestingly, patients with a baseline ABR < 20 showed more heterogenous ABR outcomes. The heterogenous outcomes need to be investigated through evaluation of impact of other covariates, including nature of the bleeds, trough levels, joint pain, and medical comorbidities potentially to enable the development of prediction scores.
Conclusions: This post-hoc analysis can help identify disease-modifying parameters that may influence treatment decisions. Preliminary results of the CML technology appear promising in identifying relevant criteria for predicting treatment outcomes beyond the standard analysis in a large clinical trial data set.
To cite this abstract in AMA style:Chowdary P, Benchikh El Fegoun S, Geybels MS, Tiede A. Data Mining: An Innovative Approach to Optimise Post-Hoc Analyses of Large Trial Data Sets into Clinically Relevant Results [abstract]. Res Pract Thromb Haemost. 2020; 4 (Suppl 1). https://abstracts.isth.org/abstract/data-mining-an-innovative-approach-to-optimise-post-hoc-analyses-of-large-trial-data-sets-into-clinically-relevant-results/. Accessed February 27, 2024.
« Back to ISTH 2020 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/data-mining-an-innovative-approach-to-optimise-post-hoc-analyses-of-large-trial-data-sets-into-clinically-relevant-results/