Using smartphone-based personal sensing to understand and predict risk of psychotic relapse at the individual level

Psychosis, which directly affects 2-3% of the population, is a severely debilitating psychiatric disorder characterised by a range of symptoms including false beliefs, false perceptions and disorganised thinking. Although antipsychotic medication can lead to rapid improvement in the acute-phase of the disorder, most patients subsequently relapse. This causes considerable distress to patients and their families, and has major economic implications both the NHS and society. A key challenge in the clinical management of psychosis is that we have limited understanding of the factors that lead to a relapse; this means that doctors are unable to optimise treatment of individual patients based on their level of risk. In other to address this challenge we have developed a smartphone app which allows the close monitoring of people across multiple contexts, time-points and locations in real time (Urban Mind; www.urbanmind.info). In the present project we will use an adapted version of the Urban Mind app to measure daily social stress - which is thought to be a strong predictor of the risk of relapse - in 225 patients with a first episode of the illness. We will examine whether this measure can predict the risk of relapse over a 12-month period. We will then test whether we get the same results when we use the app in a different group of 225 patients. A key aim of this project is to examine the extent to which "passive" data, which are collected via continuous background monitoring of phone usage and as such place minimal burden on patients, can be used as proxy for "active" data, which require patients to log their experiences into their phone. The results will be used to develop a predictive model linking real-time measures of daily social stress with risk of future psychotic relapse. This model could be integrated within routine clinical care, helping doctors optimise treatment of individual patients based on their level of risk. This could improve long-term clinical outcomes amongst people with first episode psychosis and have major economic benefits for society. The application of smartphone technologies in mental health care has the potential to fundamentally change the way in which patients are assessed, treated and monitored.

Despite the efficacy of antipsychotic medication in the acute phase of psychosis, around 80% of patients experience at least one psychotic relapse. Predicting risk of relapse, and using this information to stratify treatment, is a key clinical challenge in the management of the illness. In other to address this challenge, we have developed a smartphone app which allows close monitoring of daily social stress - a strong candidate mechanism for psychosis based on animal and human models - across multiple contexts, time-points and locations (Urban Mind; www.urbandmind.info). Firstly, we will use an adapted version of the Urban Mind app to monitor daily social stress in 225 patients with first episode psychosis recruited from the South London & Maudsley NHS Foundation Trust. Secondly, we will develop a predictive model which links daily social stress with risk of future psychotic relapse, by integrating the data collected via our app with clinical data over a 12-month follow-up period. Thirdly, we will validate our predictive model in an independent group of 225 patients with first episode psychosis recruited from a different NHS Trust (Barnet Enfield and Haringey Mental Health Trust). A key aim of the project will be to examine the extent to which, within our predictive model, "passive" data generated via continuous background monitoring of phone usage (i.e. personal sensing), can be used as a proxy for "active" data which require patients to log their responses into their phone (i.e. experience sampling). Our predictive model will generate an individualised report which includes a prognostic risk score for psychotic relapse. In the final stage of the project, we will pilot the integration of this report in the electronic database nested within routine clinical care, which we are currently developing as part of an existing MRC-funded Mental Health Data Pathfinder programme. This could improve clinical outcomes and have major health economic benefits for society.

Smartphone-based personal sensing, i.e. the moment-by-moment quantification of human phenotype in situ using people's personal smartphones, holds considerable potential for clinical as well as academic psychiatry. The proposed research will benefit a wide range of stakeholders including (i) clinicians; (ii) service users; (iii) NHS and other service-providers; (iv) policy-advisors and policy-makers; (v) academics.

(i) Clinicians. A key challenge in the clinical management of psychosis is that we have limited understanding of what causes psychotic relapses; this means that, at present, interventions designed to reduce the risk of relapse are offered to all patients, even though the level of risk varies greatly between individuals and within individual patients over time. Identifying a reliable predictor of relapse risk would allow the selective delivery of interventions when they are most needed. The present project is designed to lead to the development and validation of a prognostic risk score for psychotic relapse to be integrated within routine clinical care. This will help clinicians decide how to best allocate the limited resources for relapse prevention that are available within a clinical team.

(ii) Service users. The development of an individualised risk score for psychotic relapse and its integration within routine clinical care, will enable service users to receive targeted clinical support reflecting their level of risk. Critically, systematic reviews indicate that such targeted clinical support can reduce the likelihood of psychotic relapse in those who have suffered a first episode (Vigod et al 2013; British Journal of Psychiatry 202:187-94). The proposed research, therefore, will help improve clinical outcomes and reduce the personal and societal costs of the illness. In addition, a better understanding of how daily social stress affects future risk of suffering a psychotic relapse, could inform the development and validation of novel clinical interventions aimed at preventing the re-occurrence of psychotic symptoms.

(iii) NHS and other service-providers. The total economic cost of psychosis in the UK is thought to be over £1 billion, and the cost of caring for patients who suffer a psychotic relapse is four times higher than that of treating those with a stable course of illness. By generating an individualised risk score that could be used to predict risk of psychotic relapse, the present project could enable the NHS to make informed decisions on the most effective and cost-efficient use of clinical resources. Furthermore, because our individualised risk score will capture within-patient changes in risk of psychotic relapse over time, the present project could help reduce the need for frequent face-to-face assessments, which would further reduce health costs.

(iv) Policy-advisors and policy-makers. The results of the present project will inform policy-advising and policy-making in the emerging area of digital health. For example, despite the growing optimism regarding the use of personal smartphones to monitor and treat psychiatric patients in real time, at present there is a considerable lack of data on the clinical utility of this approach. If our hypothesis that smartphone-based personal sensing allows prediction of risk of future psychotic relapse with high levels of accuracy is confirmed, this will provide much-needed evidence base for policy-advisors and policy-makers.

(v) Academics. The benefits of our research to academics are discussed in the section on Academic Beneficiaries.