New research published in the journal Schizophrenia suggests that tracking how individuals feel during everyday social interactions using a smartphone app can help forecast the return of psychotic symptoms. The study indicates that subjective feelings about social experiences offer a reliable warning sign of relapse, often performing better than simply measuring whether someone is alone or with others. These findings provide evidence that brief digital check-ins might support early intervention strategies for people recovering from their first psychotic episode.
Psychotic disorders are severe mental health conditions that affect a person’s perception of reality. Up to 80 percent of individuals who experience an initial episode of psychosis will face a relapse within five years. A relapse is the re-emergence or worsening of symptoms like hallucinations or delusions, which often require clinical intervention. Early identification of the risk for relapse is an important part of improving long-term health outcomes.
Prior research indicates that changes in social behavior and emotional functioning are often early indicators of symptom exacerbation. Social withdrawal, loneliness, and altered emotional responses to other people tend to occur before the onset of overt psychotic symptoms. Routine psychiatric care usually involves periodic clinical visits, which capture symptom severity at a single point in time. These traditional assessments struggle to monitor the rapid emotional and social changes that occur between appointments.
“We were interested in whether subtle changes in everyday social experience could help identify relapse risk after a first episode of psychosis,” said study author Ryan Hammoud, a postdoctoral research associate in the Department of Psychosis Studies at the Institute of Psychiatry, Psychology and Neuroscience at King’s College London. “Although experiences such as withdrawal, loneliness, and how someone feels around others may precede worsening symptoms, these changes often occur between routine clinical appointments and can be difficult to detect. Smartphones offer an opportunity to capture these experiences as they occur in everyday life, rather than relying solely on someone recalling them during a later appointment.”
The scientists recruited 256 participants from ten early intervention clinics across England and Wales between 2020 and 2023. The sample had an average age of 25.7 years, and 44.5 percent of the participants were female. All participants had experienced a first episode of psychosis within the previous 24 months. The researchers required participants to own a smartphone and have access to mobile data.
Following a baseline clinical assessment, participants downloaded a custom smartphone application called Social Mind. For the first 30 days of the study, the app prompted participants once a day to complete a brief survey. These entries asked about the participants’ social contact, specifically whether they had been alone or with others over the past 24 hours, and whether they were alone at the exact moment of the prompt.
The app also assessed subjective social experience. Participants rated five items on a five-point scale, indicating how relaxing, enjoyable, stressful, or upsetting their current or recent social contexts were, and whether they preferred to be alone. The researchers combined these responses to create a social experience score ranging from five to 25, where higher scores indicated a more positive experience.
Over the following year, clinical staff determined whether participants had relapsed at four, eight, and twelve months using structured interviews and electronic health records. The authors controlled for baseline variables, including age, gender, ethnicity, education, and employment status.
Among the 256 participants analyzed, the baseline relapse rate grew progressively over the course of the study. Specifically, 8.2 percent of the individuals relapsed within four months, 13.5 percent relapsed within eight months, and 20.1 percent experienced a relapse by the twelve-month mark.
The researchers found inconsistent relationships between the sheer presence of other people and the likelihood of relapse. Reporting being alone over the past 24 hours was associated with a 133 percent relative increase in the odds of relapse within four months compared to those who were with others. However, this association was not statistically significant at the eight-month or twelve-month follow-ups. Momentary social contact, meaning being alone at the exact time of the app prompt, showed little association with early relapse but was linked to a 68 percent relative increase in the odds of relapse at the twelve-month mark.
Hammoud noted that this discrepancy between objective contact and subjective feeling stood out during the analysis. “I was surprised by how much more consistent social experience was compared with social contact itself,” he said. “We might intuitively assume that being alone is a clear warning sign, but our findings suggest a more nuanced relationship. Being alone does not necessarily mean someone is having a negative experience, and being around other people does not necessarily mean that experience is supportive or positive. It seems that the quality and emotional experience of someone’s social world may be more informative than the presence or absence of other people.”
“I would also emphasize that these findings should not be interpreted as suggesting people at risk of psychosis simply need to ’socialize more’,” Hammoud added. “Our results actually suggest that the subjective quality of social experiences appears to be more important than social contact alone.”
Subjective social experience demonstrated much more consistent associations with future clinical outcomes. Higher daily social experience scores, reflecting more positive feelings about social interactions over the past 24 hours, predicted reduced relapse risk across all time points. Each one-point increase in the daily positive social experience score was associated with a 16 percent relative reduction in the odds of relapse within four months, and a 9 percent reduction within twelve months.
Momentary social experience scores yielded similar patterns. Positive emotional responses to the social context at the exact time of the assessment were consistently associated with lower odds of relapse. A one-point increase in momentary positive social experience was linked to an 18 percent relative decrease in the odds of relapse within four months, an 11 percent decrease within eight months, and a 10 percent decrease within twelve months.
To test if this smartphone data could forecast individual risk, the authors built predictive models using machine learning. Machine learning is a type of data analysis where computer algorithms learn to identify patterns and make decisions based on complex data sets. They trained these models on the social experience and contact data to predict whether an individual would relapse within twelve months.
The predictive performance was relatively strong and improved with more data points. A model trained on just five smartphone assessments completed within a 30-day window achieved a validation accuracy of 79 percent. Even with very limited data, the algorithm performed well, as a model relying on just a single smartphone assessment collected within the first seven days achieved a validation accuracy of 71.5 percent.
“One of the clearest messages is that how someone feels in their social environment may matter more than simply whether they are alone or with other people,” Hammoud said. “More positive experiences were consistently associated with a lower risk of relapse, whereas simply being alone was a much less consistent predictor. We also found that even a small number of very brief smartphone assessments contained useful information about future relapse risk, suggesting that simple self-reports could eventually help clinicians identify changes in vulnerability between appointments.”
A strict reading of the data requires distinguishing between monitoring technology and therapeutic interventions. The smartphone application in this study functions purely as an assessment tool to signal clinical vulnerability, rather than a treatment that prevents relapse. Although the data indicates a strong association between negative social experiences and future symptom exacerbation, this relationship does not imply that poor social experiences directly cause the psychotic symptoms to return.
“An important point is that we are not proposing that smartphones or these predictions should replace clinical judgment,” Hammoud explained. “Their potential value is as an additional source of information between appointments. In our study, completing an assessment only took around 2-3 minutes, and can offer a relatively low-burden way of helping services understand changes in a person’s everyday experience that might otherwise be difficult to see.”
Participant dropout and missing data present challenges in interpreting the predictive models. Individuals who were experiencing the early signs of a relapse might have stopped engaging with the smartphone application. This non-random missing data could bias the predictive models, potentially inflating the accuracy estimates because the most adherent participants remained in the analysis. Also, relying on participants who own smartphones and have regular mobile data access might restrict the findings to a more digitally engaged population.
The researchers did not collect data regarding the living arrangements of the participants. Whether an individual lives with family, a partner, or alone can heavily influence both their daily social contact and their emotional responses to their social environment. Future studies should account for housing status to better understand the nuances of social support and relapse risk.
In addition, the predictive models were only tested internally on the study’s own data set. “I think readers should keep in mind that this is an important first step, rather than a tool that is ready to make clinical decisions,” Hammoud said. “The prediction models were internally evaluated but have not yet been externally validated in an independent population.”
Looking ahead, the researchers hope to refine these digital tools for broader application. “The next priority is to test these findings in independent populations and determine whether this kind of monitoring adds useful information to existing clinical care,” Hammoud said. “We also need to understand how these tools could be integrated into Early Intervention Services in a way that is helpful to both patients and clinicians, without creating unnecessary burden.”
“Longer term, the goal would be to move towards personalized monitoring,” he continued. “Rather than waiting until someone is clearly becoming unwell, brief measures of changes in their everyday experiences could help identify periods of increased vulnerability earlier and create an opportunity for more timely support.”
The study, “Using smartphone-based assessment of social experience to predict relapse in first-episode psychosis,” was authored by Ryan Hammoud, Anna Georgiades, Maria Chiara Del Piccolo, Shangqing Liu, Aljawharah Almuqrin, Robert Stewart, Ioannis Bakolis, Natasha Vorontsova, Stefania Tognin, and Andrea Mechelli.
Leave a comment
You must be logged in to post a comment.