image

4 Data Scientists

image

Using medical records of 60 patients

image

Develop and improve ‘diabetic nurse’ app

Client

Watch Out Diabetes

Purpose

Watch Out Diabetes is a research-oriented digital health programme focused on developing novel solutions for women at risk of type 2 diabetes. Research shows that 10% of women develop gestational diabetes during pregnancy. These women do receive exceptional care during pregnancy and 6-weeks after giving birth they are screened for type 2 diabetes. A significant proportion of women do not have type 2 diabetes at screening stage but due to receiving very little after care, as many as 7 out of 10 of these women go on to develop type 2 diabetes within 5 years of giving birth. Type 2 diabetes is a life-long condition but its onset could be delayed or prevented with lifestyle interventions, weight loss and good cardiovascular health. Watch Out Diabetes is plugging this hole in health delivery by providing advice and support to women outside hospitals and GPs centres through a targeted, digital diabetes prevention programme designed specifically for women. Watch Out Diabetes has launched a beta version of an app for testing. Their aim is to develop a ‘Diabetic Nurse’ that could provide personalised lifestyle and nutritional advice and interact with women on the programme in order to optimise the care in the community. Therefore, the company needed a strong academic team to develop the intelligence of these features. The S2DS team focused on the generation of a weekly food shopping lists and the smart food recommendation engine.

Approach

Over a five-week period, a team of four PhD scientists implemented the project and used machine learning to understand patients’ preferences. The allowed calories requirement and other information were extracted from National Institute of Health and Care Excellence (NICE) and American Diabetes Association (ADA) guidelines using NLP.

The aim of the study was to develop and validate an intelligent recommendation engine which can provide shopping lists for healthy food recipes based on various factors including age, activity level and body mass index for women previously diagnosed with gestational diabetes. Moreover, the engine encompasses a combination of a learning and predictive process which register woman's habits, previous choices and food preferences. Consequently, the system automatically adapts the recommendations and predicts a very personalised recipe and shopping list.

By using basic health information, age, activity level, weight, etc., supplied by the user, the app determines the user’s nutritional requirements and creates recommendations based on their individual user preferences. In doing so, the app will generate a list of suitable meals that a user can choose from. The user can then select a list of meals, and using the list of meals selected, the app pulls information from its database and provides a list of recipes. By deconstructing those recipes, the app is able to compile a shopping list and match it to the nearby supermarket stores.

A user can create this shopping list for multiple meals for a week or two depending on shopping habits and preferences. The user makes their choices from the recipes which are recorded and stored, the app learns from those preferences and makes future recommendations adapted to that user.

I had a great honour to be a mentor to the brilliant group of students on the S2DS 2016 summer course and work on the development of an intelligent food recommendation engine. I have to admit I was amazed by what can be achieved in only five weeks when such a diverse, talented and motivated team is brought together to work on a specific problem. Every session was an inspiring and illuminating interplay of ideas and the entire project was an invaluable learning experience for the mentors, students and the companies

Dina Radenkovic, CEO, Watch out Diabetes
The Outcome

Over the five weeks the team produced a prototype algorithm which is now being integrated into the Watch Out Diabetes app. This will be able to show individual users a bespoke meal plan with corresponding shopping list.

Testing of the app with 60 independent users of different ages, activity level, BMI's, and food preferences, showed that the recipe recommendation engine returned very accurate results. The Recommendation Engine (RE) is currently being optimised for additional factors such as the phase of the menstrual cycle and additional nutritional requirements and will be tested once again with a larger cohort. The development process will be summarised in a paper submitted to the Computer Methods and Programs in Biomedicine.

Connect with the Data Science Community

For Data Scientists & Businesses

TOP