Data, Science & Technology

The DST team uses data science, analytics and AI, together with clinical and core teams to develop deployable technologies for our pilots with the view to scale.

MOHT16.JPG

Professor Robert Morris

Chief Technology Strategist

Mental health and the role of data science
Digital Phenotyping refers to the continual collection of a person’s health signals, such as sleep and exercise patterns, heart-rate, and social activity (how many calls and text messages sent) using personal devices such as a smartphone and wrist-wearables. This technology is expected to have transformative potential for healthcare, where it may be used as a tool to passively measure, in real-time, various social, environmental, and behavioral determinants of health, which together comprise an estimated 55% of the causes of premature death.

 

The management of mental illnesses has emerged as the pioneering use-case for digital phenotyping. In 2020, the DST team has worked closely with the Institute of Mental Health on the HOPE-S (Health Outcomes via Positive Engagement in Schizophrenia) clinical trial involving patients with Schizophrenia that were recently discharged from hospital. 

 

Schizophrenia is a serious mental disorder in which people interpret reality abnormally. Schizophrenia may result in some combination of hallucinations, delusions, and extremely disordered thinking and behavior that impairs daily functioning, and can be disabling. Approximately 1% of people are diagnosed with Schizophrenia worldwide.

In the HOPE-S study, our goal is to predict the occurrence of clinical relapses (e.g. rehospitalisation) through changes or abnormalities in a patient’s digital bio and behavioural markers (collected through digital phenotyping). With carefully calibrated models, we aim to preempt and avoid these relapses through early intervention, for example, by calling the patient into the clinic for therapy. 

The requirements of the HOPE-S clinical trial drove us to develop the HOPES (Health Outcomes through Positive Engagement and Self-empowerment) digital phenotyping platform, which systematically collects and analyses data from smartphones and wearable devices. We plan to expand the use of the HOPES platform to broader user bases serving various burdens of disease, for example, we are currently embarking on initiatives to use HOPES in the management of mental wellness in healthy populations.

Management of patients with multiple chronic diseases
There has been little research on treating multi-morbid (MM) patients (those with multiple chronic diseases), because current care models are designed for single diseases. Unfortunately, MM patients cannot be effectively treated by simply combining the treatments of constituent chronic disease, due to complex drug interactions and side effects from each drug, and the varying health conditions of different patients.

 

As a result, multi-morbid patients often receive fragmented yet more expensive care, resulting in poorer quality of life and accelerated demise.

 

The DST team has developed a machine learning model to simultaneously predict the onset and progression of several co-occurring clinical conditions on 20,000 MM patients who visited the Singapore healthcare system. Using input features such as demographics, medical history, and lab test results, we can estimate the effect of individual features on the MM outcome. For example, we can predict how soon a patient will start having complications related to heart or kidney in 3-5 years if they already have diabetes.

 

We can compute the risk factors associated with each treatment for individual MM patients, so as to derive a highly personalized and effective treatment plan. For example, a patient may be overwhelmed by generic medical advice about multiple things such as choice of food, exercise and medication adherence. It may be useful to prioritise these interventions by the relative impact of these risk factors in a personalised manner. Electronic health records (EHR) could be used to understand the risks and needs of these MM patients better, through the use of personalized clinical goals.

 

With the knowledge of how much each personalised risk factor contributes to a patient’s chronic disease, we can address them more effectively by treating symptoms of higher importance for individual patients. We can also also look at population MM statistics and derive nationwide initiatives to counter the disease. For example, we could run interventions such as telehealth programs for management of obesity to indirectly reduce diabetes.