A Novel Approach of Analyzing BiAffect Data and Self-report Questionnaires through the Use of Personas

Mood is a central concept when dealing with an individual’s mental state, since mood changes reflect changes in the mental process. The most used method to assess the mental heath status of a patient is represented by self-report questionnaires, but their main issue is their instantaneous and intermittent nature, that is therefore not able to capture intraindividual variability, which indeed represents a further significant clinical information. In order to obtain a better and more accurate assessment of the patient’s mental health status, it is therefore paramount to acquire data that are as granular as possible and collected, not in a controlled environment, but during the patient’s daily life. In that regard smartphone, with their widespread and ubiquitous adoption, could be exploited to combine passive sensing to evaluate mental health and active querying through Eco- logical Momentary Assessments (EMAs). Thus, the aim of this study is to integrate data collected through the use of self-report questionnaires, to data relative to the smartphone’s keyboard dynamic, collected passively by the BiAffect application, and use these data to better understand and stratify mood instability. In order to do so we proposed a novel approach based on the use of Personas. This is a new emerging method in the healthcare, clinical health IT or consumer health IT sectors, which consists in creating a generic participant, called Persona, inside a cluster of users that share similar aspects and embody the same archetype of user. It is oriented toward achieving personalized digital healthcare solutions. The study has been conducted over a small sample of 6 females with a diagnosis of mood instability. Every day we extracted 14 features from the BiAffect’s application and 72 self- report questionnaires variables, composed of some already validated questionnaires along with some other questions regarding medication adherence, stressor exposure, medication, drug and alcohol use, sleep, pain, illness, and exercise. After having preprocessed both the self-report questionnaires and the BiAffect data, we merged the two datasets together in order to create a unique input dataset. For the creation of Personas, we have decided to use the code ”PersonaCreator” generated by Emanuele Tauro from Polytechnic of Milan. The two Personas created have 68 out of 86 variables significantly (pvalue < 0.001) different between the two clusters. To provide an output that would have been much more user-friendly, we constructed a person card for each cluster. Lastly, we validated the Personas developed. The results have shown a great capability to distinguish between the clusters, suggesting that the clustering was well performed and the obtained Personas are valid and usable. Given the goodness and validity of the obtained results, we can say that with our work we have paved the way for the use of Personas on BiAffect data to stratify mood instability.