Methodological challenges in the evidence synthesis of health outcomes of digital health technologies

Medical devices and pharmaceuticals are worlds apart, but healthcare would be impossible without them. Digital biomarkers are the subject of this thesis defined as objective, measurable, physiological, and behavioural parameters collected using wearable, portable, implantable, or digestible digital devices. Since the 1970s, systematic reviews and meta-analyses have dominated medical evidence synthesis. They provide medical decision-making evidence. To avoid biases and maintain methodological quality, the Cochrane Handbook recommends systematic reviews follow certain procedures during study stages. This thesis comprises six hypotheses related to digital biomarkers. The first hypothesis aimed to evaluate the suitability of using tools provided by the World Health Organization (WHO), including ICD-11 (International Classification of Diseases, 11th Revision), ICHI (International Classification of Health Interventions), and ICF (International Classification of Functioning, Disability and Health), for categorizing populations, interventions, outcomes, and behavioral/physiological data in studies involving digital biomarkers. The results indicated that these tools were not applicable for categorizing digital biomarker studies as a whole. However, further analysis revealed that these tools were suitable for categorizing digital biomarker studies involving non-general populations or populations with specific diseases. The second hypothesis focused on comparing the statistical power of direct and indirect digital biomarkers. The results indicated that there was no significant difference in power between these two types of digital biomarkers (p-value > 0.05). The next three hypotheses compared the characteristics of systematic reviews and meta-analyses of digital biomarker-based interventions with those of non-digital biomarkers or pharmaceuticals. The comparisons were made in terms of methodological quality, quality of evidence, and publication bias. Although all these hypotheses revealed non-significant differences ...