Unsupervised learning of disease subtypes from continuous time Hidden Markov Models of disease progression

摘要: Healthcare providers worldwide are faced with the formidable task of improving efficiency and quality of care while also lowering costs. This is especially needed in the US healthcare system, which suffers from exceptionally high per capita costs and underperforms in comparison to other developed countries on accessibility, equity, efficiency, care quality and health outcomes [10](Figure 1). To begin addressing this problem, the US Congress passed the HITECH (Health Information Technology for Economic and Clinical Health) Act in 2009 to promote the adoption and “meaningful use” of electronic health records, paving the way for improved health information exchange and data management. Since then, the volume of healthcare data has risen dramatically, already reaching 150 exabytes (150⇥ 1018 bytes) in 2011 [40]. In order to leverage this vast resource, big data analytics tools are being developed to address several clinical tasks: screening, diagnosis, treatment, prognosis, monitoring and management [14].Data analytics in healthcare can be implemented using either a batch processing approach or an online approach. The batch processing approach can be applied to study population health management by deriving actionable information from largescale health data. For instance, it could allow researchers to assess the effect of different genetic or environmental risk factors on disease prevalence, or to monitor the effectiveness of drugs or interventions. At the same time, such insights can be applied at the individual level for precision medicine. By profiling patients to identify individual genetic, environmental and lifestyle characteristics, it …