Nadia Alam
Slide 1: Title Slide
Hello everyone. My name is Nadia Alam, and I am a PhD researcher at Warwick Medical School. Today, I will be presenting our study on Digital Phenotyping and Serious Mental Disorders, Predicting Symptom Re-Emergence and Relapse Among Slum Residents in Dhaka, Bangladesh: A Machine Learning Study. This research explores how smartphone data and advanced AI applications can help bridge the gap in mental health care for vulnerable populations.
Slide 2: Mental Health Crisis in LMIC Slums
Mental and behavioural disorders account for 12% of the global disease burden, with over 70% of this impact affecting low- and middle-income countries. Nearly 80% of individuals affected by serious mental disorders live in LMICs, where access to treatment remains severely limited. Factors such as rapid urbanisation, extreme poverty, and frequent exposure to trauma exacerbate mental health issues in these settings. One of the most vulnerable populations within LMICs is slum dwellers. In Dhaka’s Korail slum, which houses approximately 200,000 people, poor living conditions, overcrowding, and inadequate access to healthcare worsen the impact of mental illness. The lack of formal healthcare infrastructure forces many residents to rely on informal providers, traditional healers, or NGOs for medical care, which often leads to delayed or inappropriate treatment for severe mental disorders. A particularly alarming challenge in slum communities is the high rate of relapse in psychotic disorders, such as schizophrenia and bipolar disorder. Relapse often results in hospitalisation, worsening social and economic outcomes, and increased stigma. However, traditional relapse monitoring relies on face-to-face clinical assessments, which are inaccessible, costly, and unsustainable in low resource settings. Furthermore, Bangladesh has fewer than 0.5 psychiatrists per 100,000 people, making early intervention and continuous care practically impossible for most slum residents. Unfortunately, traditional clinical monitoring methods are expensive, require regular healthcare access, and are unsustainable for slum communities. This highlights the urgent need for scalable and accessible alternatives.
Slide 3 Digital Phenotyping, LMICs and perceptions
Digital phenotyping (DP) may solve this problem OR offers a solution, as it refers to the use of smartphone and wearable sensor data to monitor and predict health outcomes. It passively collects data such as movement patterns, screen activity, and communication behaviours, while also incorporating active data like self-reported questionnaires. Once this data is gathered, we can use advanced AI techniques, such as deep learning models to identify subtle behavioural shifts that may precede relapse. In high-income settings, ML models have successfully predicted schizophrenia relapse by analysing digital behaviour changes weeks before a clinical episode. However, these models are developed in contexts with continuous smartphone access and individualised device use.
To understand how slum residents perceive digital phenotyping, we conducted eight focus group discussions in Korail, Dhaka. Our findings revealed several key insights:
- Participants were cautious about sharing personal communication data such as the content of messages, but more comfortable with non-intrusive data, app usage patterns and location tracking.
- Given that smartphones are often shared among family members, participants suggested a family-centered approach to ensure ethical implementation.
- Many emphasised the importance of educational programs to build trust in DP technologies, indicating that awareness and engagement are critical for adoption.
These findings highlight the potential of DP to bridge mental health care gaps while emphasising the need for culturally sensitive implementation strategies.
Slide 4
Building on the insights from our qualitative work, our next step is to conduct a longitudinal cohort study to assess the feasibility and effectiveness of digital phenotyping (DP) in predicting relapse among individuals with serious mental disorders (SMDs) in Dhaka’s Korail slum. This study will integrate passive smartphone data and active clinical assessments over six months to identify early warning signs of symptom deterioration.
We will use a bespoke app developed in house, called DataDoc, which will be used to collect both active and passive data from participants:
Passive data includes mobility patterns (GPS tracking), digital interactions (call/message frequency), phone usage (screen time, app activity), and sleep behaviors—all of which have been associated with changes in mental health status.
Active data includes self-reported symptom scales, stress levels, and quality-of-life assessments collected via the app and in-person follow-ups.
Machine learning algorithms will then analyse these data streams to detect subtle changes in behaviour that could indicate impending relapse. By leveraging predictive models trained on historical relapse cases, we aim to develop an early warning system that can alert healthcare providers, caregivers, or the individual themselves before a relapse occurs.
This real-time, continuous monitoring approach could help mental healthcare in low-resource settings, offering a cost-effective and scalable alternative to traditional clinical monitoring. If successful, this could reduce hospitalisation rates, prevent severe symptom escalation, and improve long-term health outcomes for marginalised populations.
Ultimately, this study will contribute to understanding how digital phenotyping can be adapted for LMIC contexts while addressing key challenges such as shared phone usage, digital literacy, and ethical considerations.
