Data-driven approaches to maximize clinical impact in spinal cord injury research

Abstract

Spinal cord injury (SCI) is a medical condition resulting from damage to the spinal cord. As the spinal cord represents the primary connection between the brain and peripheral organ systems, a disruption leads to numerous impairments in locomotion, sensation, and organ functions. A SCI therefore undermines the overall quality of life and independence of the individuals affected and their families. This realization is particularly relevant since the field still lacks an intervention, pharmacological or otherwise, to promote the restoration of functions and/or regeneration of the damaged spinal cord. While clinical trials conducted to date did not find any promising intervention, they support the field by thoroughly collecting large amounts of data. The surge of data science, including statistical and machine learning (ML) methods, holds the promise to uncover new insights in better defining and enhancing recovery following SCI by extensively investigating retrospective data. This thesis aimed to leverage the potential of data science to maximize clinical impact in SCI research. This effort was pursued around three pillars: (i) enlarging the surveillance within clinical studies, (ii) promoting best methodological practices from data science applied to SCI research, and (iii) highlighting the importance of effective research dissemination. Firstly, the general context in which this thesis fits is outlined in Part A. Then, Part B sets benchmarks through the secondary analyses of major datasets collected in the field, namely the Sygen clinical trial, the European multicenter study on human spinal cord injury (EMSCI), the Murnau center, and SCIRehab cohorts. Chapter 1 studies how recovery following SCI evolved over the last two decades and showed that, despite an evolving standard of care, neurological recovery has remained largely unchanged in this period. This observation paves the way to using historical patient data to enrich placebo arms in future clinical trials, therefore maximizing the exposure to the intervention of interest. Chapter 2 describes the natural progression of serological markers following SCI, providing an additional surveillance tool when testing pharmacological interventions that might affect individuals beyond the primary injury targeted. Similarly, studies testing the effect of new pharmacological interventions may be affected by interactions with medications that are prescribed following injury. We therefore exhaustively report in Chapter 3 what constitutes the current pharmacological standard of care. We reveal an extensive polypharmacy that individuals with SCI are subject to. To characterize the effects of this polypharmacy on SCI recovery, we systematically review the literature in Chapter 4 and describe both clinical and pre-clinical evidence supporting beneficial or detrimental effects in neurological recovery following SCI. Secondly, Part C adapts known methods from data science to be translated to SCI research applications. We initially investigate the potential of serological biomarkers as predictors of motor recovery in Chapter 5. This analysis shows that accounting for clinical characteristics specific to the condition improved predictions, while still being limited by factors such as missing data leading to small cohorts to be studied. We therefore further characterise missing data in the context of SCI in Chapter 6. Here we develop guidelines on how to handle missing information based on simulation studies. We demonstrate that last observation carried forward imputation is a viable approach for imputing missing neurological outcomes after SCI, owing to the distinctive plateau in recovery starting around six months after initial trauma. Finally, Chapter 7 explores the concept of positive deviance to detect individuals recovering beyond clinical expectations. While data extracted from such individuals may impair the performance of ML prediction models, understanding the mechanisms underlying their phenomenal recovery holds the potential to uncover patterns leading to improved recovery. Lastly, Part D underlines the importance of science communication to effectively link research from bench to bedside. Chapter 8 particularly promotes the use of new tools such as interactive data visualization to elevate the presentation of research outcomes while leaning towards more transparent and accessible research not only for the scientific and clinical communities but also the individuals affected, their families and society. Overall, this thesis contributes to the in-depth benchmarking of decisive elements guiding clinical studies in SCI, such as neurological recovery, the evolution of serological biomarkers, and medications commonly prescribed as part of the standard of care. This work leads the path towards improved data analyses and recovery prediction following SCI by integrating known characteristics from the condition. In the context of the SCI research field, this thesis participates in revising the approaches employed to discover interventions to improve recovery following SCI.