Social media AI analysis can help support diabetes management

In Singapore, more than 400,000 people manage diabetes daily – and this number is expected to rise to a million by 2050. As lifestyles change and more go for early health screenings, doctors are beginning to see more young people diagnosed with the lifelong condition, according to CNA.

Managing diabetes means making countless micro-decisions every day on what to eat, adhering to medicine regimens, and managing sleep and lifestyle patterns. As healthcare costs increase and Singapore’s population ages, it is critical to help people with diabetes and their caregivers navigate complex information and help them make the right decisions.

This is why knowing what’s on the minds of people with diabetes and their caregivers is important for doctors and policymakers, so they can deliver personalised care and collaborate for better outcomes.

At the recent Public Sector Day Singapore event, organised by AWS and GovInsider, Accenture’s Managing Director of Growth and Innovation, Sonia Gupta, and SingHealth Polyclinics’ Dr Mabel Leow, a research fellow, discussed the implications of a recent study analysing online trends around diabetes management.

To subscribe to the GovInsider bulletin click here. 

Listening to the voices of people with diabetes and their caregivers

“The hypothesis was, is there room to accelerate understanding what's contemporary and current in minds, or what's out there in the landscape that's not visible?” said Gupta.

The study analysed 130,000 documents from July 2023 to July 2024 from various sources, such as YouTube, Reddit, Facebook, news portals, and PubMed. These sources were verified by Accenture’s responsible artificial intelligence (AI) governance framework to help confirm they were in line with applicable data privacy requirements around personal data, said Gupta.

Using AI, the study performed unsupervised topic clustering to identify 52 distinct themes emerging from the perspectives of persons and caregivers with diabetes, healthcare professionals, and government policymakers. Then, they used a human supervised model to cluster these into 18 key themes, grouped under 3 macro themes.

These themes were: how people with diabetes and caregivers actively inform themselves on diabetes management trends; the emotional and financial burden on caregivers; and the difficulty of adhering to medical regimens.

“The emotional and financial burden for the caregiver was something very much present in these documents but was not fully recognised and engaged with by healthcare physicians and the governments in terms of the design of interventions and proactive care management,” she said.

Next, they used generative AI (GenAI) to interpret the findings in natural language. One such micro-theme includes specific concerns around diet, such as the difference in impact of natural sugar vis-a-vis added sugar.

Moving from physical care to holistic care

The study’s findings could help shape the way physicians work with individuals with diabetes on managing diabetes from a focus on physical care to a more holistic approach, said Dr Leow.

“We are very good with our physical care… we now need to move to a more psychosocial perspective on diabetic care, caring for individuals with diabetes overall wellbeing and not just their physical health,” she explained during the session.

The research helped the team identify gaps in the current approach to healthcare, particularly around supporting caregivers – concerns that may not arise in a clinical setting, she shared.

While doctors may only touch base with them over a short period of time, online forums allow them to look for guidance, seek support, and share their thoughts 24/7, she noted.

“It’s a good platform for them because this is probably something that healthcare cannot provide. We can have a social worker talk to them for 20 minutes, but we cannot be there for them every day. It provides a community of support for them,” Dr Leow said.

This is why studying these online spaces can help doctors gain a fuller view of social attitudes and concerns around diabetes, she explained. Gupta similarly noted that such research can help physicians and policymakers be more responsive to contemporary trends and inform proactive care.

To subscribe to the GovInsider bulletin click here.

Changing nature of health research

Thanks to AI, the team was able to analyse the 130,000 documents across a one-year period in a matter of six weeks – an accelerated pace of work that would have been unthinkable before.

“Using AI, we were able to get to very specific and relevant themes in a short window of time that wonderfully complements traditional methods,” said Gupta. While this study had to do with diabetes, the methodology could be applied to other chronic conditions as well.

Dr Leow explained that in the past, such research projects would require a lengthy recruitment process for in-person interviews, which would often be limited to a self-selecting pool of participants.

While researchers could analyse social media to assess sentiments, they needed to comb through online spaces manually. However, AI can help to speed up this process and analyse data much faster.

Analysing online sentiment may also result in more representative results, as these are naturally occurring spaces where people with diabetes may express themselves more freely, she noted.

“As we move to becoming a healthier nation, proactively engaging in holistic ways, taking all that’s good in our traditional and rigorous methods, but complementing it with the advancements in AI and the new world of technology gives us, can provide a more holistic healthcare ecosystem that gets us to better outcomes for the population as a whole,” said Gupta.