Charlie – An app for predicting diabetes makes progress in public health innovation
Type 2 diabetes is a chronic condition in which an individual’s body does not produce or use insulin well, resulting in elevated blood glucose (or blood sugar). Over time, high blood glucose levels can lead to heart disease, stroke, blindness, amputations and other serious medical issues. Diabetes currently affects 30.3 million people in the United States (or 9.4% of the population), but it is unequally distributed across demographic groups.1 Due to a number of sociopolitical factors, it occurs most frequently among low-income individuals, American Indians, Blacks and Hispanics. And while type 2 diabetes is manageable with proper care, treatment can be expensive and onerous. Beyond the mandate to eat well and exercise, individuals with diabetes must test their blood several times per day and self-administer insulin injections based on complex calculations of their current blood glucose levels, the amount (and type) of carbohydrates they expect to consume and projected physical activity. Many diabetics eventually become delinquent in their testing and insulin shots, especially those who also suffer from mental health issues or who lack adequate access to healthcare resources. And because the calculations are complicated and imprecise, even compliant patients often miscalculate their optimal insulin dosages.
Concerned about high rates of type 2 diabetes complications and their concentration among certain socio-economic and racial groups, medical researchers from St. Marcarius-Alexandria University Hospital pledged to make a change. Teaming up with a group of computer scientists, they developed a multi- platform application, named Charlie, which utilizes artificial intelligence technologies to make diabetic care easier, more holistic and more accessible. Taking advantage of smartwatches’ biosensors to test blood glucose through the skin, the app’s algorithms calculate the optimal level and type of insulin for each user. Individuals still administer their own insulin injections, but Charlie makes the process more efficient and effective. It provides them with a precise dosage calculation, which accounts for current blood glucose levels, as well as lifestyle data (e.g. predicted carbohydrate consumption, exercise levels) and other medical data (e.g. preexisting healthcare conditions).
Charlie distinguishes itself from similar medical devices that use biosensors to test blood glucose in two respects. First, it incorporates a data collection platform. This means that user data—along with anonymized datasets collected by the University Hospital in other experiments—is fed into Charlie’s algorithms, enabling them to constantly improve and provide individuals with increasingly more accurate, individualized insulin dosage recommendations. Charlie also uses this data to provide subjects with personalized reminders to exercise, eat well and check their blood glucose levels.
Second, unlike other AI-enabled medical devices, Charlie contains a forum for information sharing and social networking. The developers hoped the forum would serve dual functions. First, by providing a space where users could post, view and discuss emerging and ongoing scientific developments in diabetes research, they wished to counteract what they saw as misinformation about type 2 diabetes— and healthcare in general—on the internet. The forum would be a neutral aggregator of information, and users would be expected to self-regulate. However, to ensure users were kept up-to-date on the latest research, Charlie’s developers conceded the use of automated content moderation algorithms (ACMAs) to privilege the most recent information in results. Second, noting the high incidence of depression, anxiety and feelings of isolation among people with diabetes, Charlie’s developers thought to design an informal space where users could communicate with one another, thus organically building a network of support. As a bonus, Charlie could use natural language processing techniques to analyze the emerging discourse in order to add contextual data points to individuals’ profiles. These analyses could then be used to improve customized treatments.
Charlie passed through the university IRB process with little trouble, as it promised substantial improvement in diabetes care with minimal risk to the individual research subjects. Best practices in diabetes care are already well-established in the medical community, and Charlie didn’t aim to disrupt them. The technology merely uses existing frameworks to provide more efficient, accurate and personalized care. The review board did note that there was some precedent for assessing the social media platform separately from the medical device itself, but ultimately decided that the forum would not cause undue harm to individual users.
Upon IRB approval, Charlie was rolled out in a clinical trial at the University Hospital to generally positive results. In a survey administered at the end of the trial, users reported liking the regular, individualized, scientifically grounded analyses of their health. Data analysis revealed that those who used the system had higher rates of medical compliance, lower blood glucose levels and improvements in mental health over those in the control group. However, these results were not equally strong for all users. In particular, racial minorities did not experience the same positive results as white users.
Results for Charlie’s social networking forum were also mixed. Discussion was frequently lively, but not all users found it productive. Conflicting reports and comments abounded, and would sometimes devolve into hostile arguments, in which users would gang up and call each other ‘pseudoscientists’ - or worse. Complaints that the negative tone and tenor of conversation was leading some users to disengage with the platform were accompanied by concerns over the emergence of homogenous mini-publics, or echo chambers, that had begun to form around shared scientific and/or philosophical beliefs. This kind of fractured discourse was in direct conflict with the developers’ intentions to create a positive, productive space to encourage healthy living amongst those living with diabetes.
1 Centers for Disease Control and Prevention (CDC), The National Diabetes Statistics Report, 2017. https://www.cdc. gov/diabetes/pdfs/data/statistics/national-diabetes-statistics-report.pdf. Note that these figures do not differentiate between type 1 and type 2 diabetes; however, type 2 diabetes accounts for 90-90% of all diabetes cases.
Publication: Forthcoming in Q2/2026
