The factors that affect human health occur at several scales, from gene and protein interactions, to the cellular, organism, and social levels. Understanding and controlling human health is especially complex due to the inter-level interactions that cannot be integrated away and thus form true control hierarchies [1]. Yet, the fact that multiple levels participate in the co-regulation of human health, enables us to measure and study it from other levels beyond the molecular. Indeed, social media, electronic health records (EHR) and mobile application data enable population-level observation tools with the potential to speed translational research. For instance, disease-specific (sub)populations can be identified via machine learning and complex network analysis of EHR, discovering personalized disease trajectories [2]. Social media data can then be used to identify disease phase transitions by complementing clinical information from EHR with a temporally richer characterization of symptoms and behavioral attitudes, providing preliminary evidence for hypothesis-driven clinical and molecular studies. We will present our recent research exemplifying a complex systems approach to the study of multi-level human health. This includes big data studies demonstrating that: 1) social media data (Twitter & Instagram) cohort studies (depression, epilepsy & opioids) enable identification of previously unknown adverse drug reactions (ADR) and drug-drug interactions (DDI) [3, 4]; 2) web searches and lexical sentiment analysis of social media allow us to uncover collective social behavior in phenomena of interest to public health, such as human reproductive behavior at a planetary level [5]; 3) the longitudinal analysis of EHR reveals significant gender and age biases in the prevalence of known---and thus preventable---DDI [6]; 4) a study of Facebook timelines to identify behavioral markers in deceased patients from Sudden Death in Epilepsy (SUDEP). Together, this work shows how complex systems techniques can be used to test novel hypotheses of public-health relevance, and even invalidate long-standing ones---such as our recent results showing that the reigning biological hypothesis for observed human reproduction cycles is incompatible with newly available planetary data of online behavior, and that a cultural explanation of the phenomenon is much more likely [5].
