Background
The lack of real world data on clinical trends, lifestyle, and the patient journey between consultations severely limits physicians on their ability to deliver evidence-backed personalized care for subjects with type 2 diabetes (T2DM) in developing countries.

Aim
This study aims to demonstrate the potential of a digital therapeutic, Wellthy Diabetes, delivered through a smartphone app, to engage patients to self-report lifestyle and clinical data in a low literacy and low health engagement setting like India. The objective is to capture relevant clinical and lifestyle data that can be made available to treating physicians and allied health professionals at the moment of care to enable augmented data-driven clinical decisions and better personalization of care without having health professionals make any additional investment in time, people or physical infrastructure.

Method
This study describes real world data of 64 subjects with T2DM, sampled from a pool of 165 participants, each using the Wellthy Diabetes smartphone App (WD), developed in scientific collaboration with the RSSDI. The inclusion criteria for subjects for this study included A.) Age >18 years, B.) Glycosylated Hemoglobin (A1c) levels in the range of 6.5%-13%, C.) Ability to use an Android smartphone, and D.) Active usage of WD for a minimum period of 7 days. WD provides users with features to self-report their meals, blood glucose (mg/dl), physical activity (minutes of active time), and weight (kgs), while engaging them in a lifestyle modification intervention coached and augmented by Artificial Intelligence (AI). WD tracked the duration of physical activity using inbuilt sensors on the smartphone. Information on readiness to change and diabetes self-efficacy (Problem Areas in Diabetes PAID-5-Short Form), clinical variables (body mass index and A1c), and medical prescriptions were collected at baseline.

Results
A total of 11,739 diabetes related clinical and lifestyle data points were captured from 64 patients over 2641 person days of usage of WD App. The mean duration of a subject on the app during this study was 41.2 days during which on average 183 clinical and lifestyle data points were tracked per user at a mean velocity of 4.7 data points per user per day.

Subjects logged 3121 instances of meals of which 38% were breakfast, 34% were lunch, and 27% were dinner. The average number of meal logs per patient was 48.76 (median = 39) at the rate of 1.18 meal logs per person day with 100% users recording a meal instance with breakfast, lunch, and dinner being reported by 100%, 98.65%, and 96.92% subjects respectively.

A total of 583 blood glucose (BG) readings were logged during the study period of which 49% were fasting (FBG), 31% post-prandial (PBG), and 20% random (RBG). Ninety-Two percent of users reported at least 1 blood glucose reading with 84%, 68% and 42% reporting at least one FBG, PBG and RBG reading respectively. The mean FBG and PBG readings were 136 mg/dl [CI: 102-172] and 180 mg/dl [CI:115 - 234] respectively.

Ninety-five percent users (n=61/64) reported a weight reading. A total of 302 weight readings were reported at an average of 4.95 weight readings per user.

A total of 59,859 minutes of physical activity was tracked for the participants through 7,733 instances of activity tracking, out of which 87% of the instances were tracked using inbuilt sensors on the smartphone and 13% were self-reported by users. On average 120 instances of activity were tracked per user.



Discussion
This study demonstrates that the Wellthy Diabetes Smartphone App engages participants in an effective way to enable tracking and promotes self-reporting of significant volumes of clinical, lifestyle, and patient journey data in real world settings. This data has the potential to deliver insights into patient’s clinical trends and lifestyle state between appointments. Availability of such insight coupled with data driven guideline based real time doctor decision support can help treating physicians augment their ability to translate evidence based guidelines into practice during every consultation, and deliver more personalized care to every patient, thereby improving both patient satisfaction and clinical outcomes. This model of care may be adopted by developing countries.

Evidence of an artificial intelligence tool to augment data driven decisions for diabetes care in developing countries