PEPPER Overview

Diabetes is a widespread chronic health condition that lasts for life. Many people with insulin-treated diabetes rely on complex calculations and human memory to estimate their insulin doses, which they must do several times a day. As well as being an enormous mental burden, poor management can lead to additional health problems; a large dose can even be fatal.

PEPPER, short for Patient Empowerment through Predictive PERsonalised decision support, tries to tackle this problem by utilising portable technology, together with artificial intelligence (AI) and mathematical modelling, to give people freedom from daily decision-making. The project has created a tool that makes predictions based on real-time data, gathered from unobtrusive, wearable devices, in order to empower individuals to manage their condition more easily. The resulting application has potential benefit to society by improving health outcomes and thereby reducing costs.

Currently there is no decision support system for insulin dosing on the market that adapts itself based on real-time activity data and blood glucose data. The PEPPER project has addressed this by providing personalised decision support on two alternative mobile platforms: one based on a smartphone, and another via the handset of a minimally obtrusive patch pump, which is about the size of a tic-tac box and was manufactured by Cellnovo Ltd (Figure 1). The company went into administration on 26 March 2019 however. Therefore the API delivered here is not a single integrated solution, but rather a collection of individual APIs that comprise the constituent parts of PEPPER.

Image cell
Figure 1: The Cellnovo System

Users of the system also wear a fitness band and a continuous glucose monitor, which is around the size of a small USB stick. Additional information, such as carbohydrate consumption and alcohol intake, can be added manually on the handset (Figure 2). The PEPPER system is also designed to offer improvement in interactions between individuals and health professionals via a secure, cloud-based server.

Image man
Figure 2: The PEPPER System Architecture

The system design process involved users at every stage to ensure that it meets patient needs and raises clinical outcomes as well as improving lifestyle, monitoring and quality of life. Prototypes of the system were demonstrated at various stages to a community of stakeholders including individuals with Type 1 Diabetes, the Vice-President of Diabetes UK, the Director of Research Partnerships from the Juvenile Diabetes Research Foundation (JDRF) and a representative of the Sociedad Española de Endocrinología y Nutrición. Tim Omer, representing the Nightscout patient community, said "as we capture higher quality and quantity of data about our condition, it is refreshing to finally see progress in assisting the patient with analysing this data to provide actionable feedback to reduce the burden of Type 1 Diabetes".

PEPPER draws together computer scientists, clinicians and industry leaders to create and clinically validate this ground-breaking tool. The project, funded by the Horizon 2020 programme, runs from 1 February 2016 until 31 January 2020 and originally included six partners from three countries: Oxford Brookes University, Imperial College London, University de Girona, Institut d'Investigació Biomèdica de Girona Dr. Josep Trueta, Romsoft SRL and Cellnovo Ltd. Following the liquidation of the latter company, the consortium was reduced to the remaining five members however. The approach used and resulting system architecture provide a generic framework for providing adaptive decision support anytime, anywhere, which could be applied to other health conditions that are monitored by wearable technology.

Individual API structure

Each of the individual APIs is produced by a different partner within the PEPPER consortium. As such, the licensing and contact information of each component corresponds to the organisation responsible.

The presentation of each API comprises four pages:

  1. Overview: This page explains what the interface does and how it can be used. It may also include some examples of use.
  2. API: Each API is generated from the Java source code using Javadoc, which is the industry standard for documenting Java classes.
  3. Licensing: Some of the collections are open source and free to use, while others are distributed by the authoring institutions which retain the intellectual property rights. The arrangements for each are described on the respective licensing page, which may also include links to the code repositories.
  4. Contact: This is the contact information for API.

Individual API components

There are four separate code bases within the PEPPER API. All except for the web interface were written in Java or Javascript. Figure 3 is a representation of how the components of the original PEPPER system interoperated and shows how the individual APIs fit within it. The central component of the PEPPER system is the insulin recommender system, which calculates the bolus dose from the various sources of physiological, dietary, physical activity and environmental data. The resulting dose is constrained by the safety system, which keeps the recommendation within a safe range, as well as providing additional safety-critical features. Data is synchronised with a cloud-based server, which provides the web interface to clinicians. The mobile handset contains the logic that draws together each of these component parts. Figure 3 is split into two sections:

  1. The PEPPER Server Application (PSA). This comprises the web-based frontend and PEPPER remote database, together with the program logic to connect to the insulin recommender and safety system. The corresponding API is the PEPPER Web Interface (PWI).
  2. The PEPPER Mobile Application (PMA). This comprises a mobile graphical user interface (GUI), together with the program logic to combine the insulin recommender and safety system. This logic component is not provided as an implemented API, since it belonged to Cellnovo Ltd. and was therefore lost when the company was liquidated. Instead we provide its interface component as a GUI API, which facilitates data input as well as visualisations of the output data. In addition, this architecture document offers a detailed description of the objects, communication and processes that are intended connect the PSA and PMA.
An overview of each of the constituent APIs are given below.
PEPPER SERVER APPLICATION (PSA) CASE-BASED REASONING ENGINE Revise & Retain Web-based Clinical Interface Data Analytics Data Export PEPPER MOBILE APPLICATION (PMA) INPUTS Glucose Measurements (CGM) Activity Data (Activity Band) User Manual Inputs (Graphical interface) User Demographic Data BOLUS INSULIN RECOMMENDER CASE-BASED REASONING ENGINE Retrieve & Reuse SAFETY SYSTEM MODULE 1 Predictive Glucose Alerts and Alarms MODULE 2 Predictive Low- glucose Basal Insulin Suspend MODULE 3 Adaptive Carbohydrate Recommender MODULE 4 Dynamic Bolus Insulin Constraint OUTPUTS Handheld alerts/ alarms and carer alarms (SMS) Insulin suspension pump command Rescue carbohydrate recommendation Insulin bolus recommendation Key Safety System Insulin Recommender Web Interface GUI
Figure 3: The PEPPER Detailed System Architecture
  1. Safety System: This API has multiple modules. It uses continuous glucose monitoring data to provide predictive glucose alarms and and low-glucose insulin suspend and carbohydrate recommendations. It also constrains the bolus suggested by the insulin recommender system.
  2. Insulin recommender: This API uses Case-Based Reasoning (CBR) to provide personalised insulin recommendations. The CBR cycle is divided into two parts: the local and remote. The local part runs on the handheld unit and the remote part on a server. Both parts contain a case-base and periodically the local case-base is synchronised with the remote case-base. The evaluation step of the CBR cycle occurs on the server and requires aproval by an expert clinician before a new case is incorporated to the case-base. The CBR parameters include CGM and capillary glucose data, physical activity, time, basal insulin, hormone cycle, stress, alcohol, meal composition, and sleep. Most of these parameters are automatically collected (or calculated) by the handset unit. Exceptions include alcohol consumption, meal composition and hormone cycles, which need to be manually input.
  3. Web interface: This API provides the logic and interface for all of the patient data used within the system. It has four components. The PEPPER Backend API represents the PEPPER remote database. The PEPPER TOOLS API provides the logic and processes to connect the database to the insulin recommender and safety system APIs. The PEPPER Frontend API represents the user interface for this application, and is primarily intended for use by clinicians. Finally the Custom Applications API gives developers from the world the opportunity to create their own applications that connect to Pepper database and then use the PWI instruments
  4. GUI: The handset Graphical User Interface (GUI) and data visualisation API are provided to replicate those implemented within the PEPPER mobile application.

Acknowledgement

This project has received funding from the EU Horizon 2020 programme, grant agreement No. 689810.