Eigenroutines: modelo computacional para identificação de padrões comportamentais em pessoas com transtornos neuropsiquiátricos utilizando histórico de contextos

Abstract

Neuropsychiatric disorders have been neglected by public health attention during a large period of time, often due to low mortality that was directly related to mental health problems. This behavior translated into low investments in research and medical practices. As consequence neuropsychiatric disorders are between the diseases with largest economic and social burden. Considering that a good part of these disorders can be identified based on patient behavior, and that smartphones and wearables are every day more available in the overall society, it is fair to imagine that the usage of these devices could help in the diagnosis process, in treatment and in a better understanding of these disorders and how they affect the society through a set of epidemiological profiles. Ubiquitous computing, more specifically in the application and usage of context history, is a way of taking advantage of mobile devices sensors as a data source of human behavior. In this computational model, the device is frequently recording information about the user’s context, such as accelerometer, incoming and outgoing calls, messages, ambient luminosity and location, so that it can derive a set of contexts in a timeline, as if they were snapshots of each moment. This set of snapshots can identify behavior changes, that can be used in the management of neuropsychiatric disorders, and if applied in a large scale, could be enough to build epidemiological profiles of the managed disorders. This thesis proposes a model to identify behavioral patterns in people with neuropsychiatric disorders based on context history so that the treatment of a given patient can be used as basis for the treatment of other patients, and that when used in scale, this model could be capable to building epidemiological profiles of disorders being managed and treated in the platform. The model is based in the hypothesis that the ubiquitous availability of mobile devices and the massive usage of them have created an opportunity to use these devices as a way of identifying behavior, and that it could be used to support the management of neuropsychiatric disorders. And that beyond the management of a single person, this approach could: (1) compare and adapt the treatment of a given person based on another person with similar symptoms that is or have been under treatment, and (2) build epidemiological profiles of mental disorders using the social and demographic data as well as the diagnosis and treatment data available in the platform. This thesis describes the model, architecture, a prototype and well as the usage of the model, studies the limits and the necessity of a large-scale usage so that the profiles can be built. The major contribution are: (1) diagnosis support based on clinical judgement using context history of patients in advanced treatment stages with the history of patients in the initial phase, (2) the inclusion of mental health professional as a key resource in the definition of interventions and in considering the professionals clinical judgment as part of the algorithm definition in the diagnosis process and (3) the building of psychiatric epidemiological profiles using context history and mental health professionals diagnosis.