Developing a decision support tool to enable precision treatment of type 2 diabetes

Lead Research Organisation: University of Exeter
Department Name: Institute of Biomed & Clinical Science


Type 2 diabetes is very common throughout the world. Severe health problems can occur for people living with diabetes including blindness, kidney failure, amputations, heart attacks and strokes. These problems can be prevented if the blood glucose (sugar) is prevented from going too high. Most people with type 2 diabetes need medication to lower blood glucose: over 3 million people in the UK need these medicines. Although it is clear the first medicine to be used should be a tablet called Metformin, it is not clear what type of medicine should be used after this. Many medicines are available which on average lower blood glucose to a similar extent, but it is unclear which is the best to use for individual patients, who may respond differently to different medicines.

Our research group, in work funded by the MRC, has analysed the response to blood glucose lowering medicines in many thousands of people with diabetes both from their family doctor records and clinical trials. We have shown that simple features like a patient's sex, how overweight they are, and the results of blood tests are linked to how well a specific type of treatment lowers the blood glucose. This means we can now work out from simple, routinely collected, information which type of medicine is likely to lower blood glucose the most. This exciting work means that it will now be possible to choose the right type of medicine to best lower the glucose for each patient.

The aim of this project is to build on this information to develop a computer program, called a decision support tool, to work out which treatments would be best for a person with type 2 diabetes.

We have already developed a simple tool that combines different routinely available information to accurately predict which glucose lowering medicines will be most effective in lowering a person's blood glucose. However this has not been tested for all diabetes medicines and for people of different ethnicities, and there are many other aspects to consider when choosing treatment, for example whether side effects are likely, or whether a person has other conditions that affect treatment choice.

In the first part of this research project we will address these gaps. We will use information from over a million people with type 2 diabetes from healthcare records and clinical trials to:

1) Expand the tool so that it works for all diabetes medicines, and in people of all ethnicities

2) Expand the tool to predict whether people are likely to need to stop treatments quickly due to side effects.

3) Adapt the tool so that it recommends certain medicines in people with medical conditions that affect what medicine can or should be taken. For example, certain medicines have been shown to be better in people who have heart disease.

In the second part of the project we will test whether the tool can be improved by adding in additional information likely to be more available in the future, for example a person's genetic information, or blood tests that are not routinely tested. We will develop a process so that future changes - for example new features that predict response, or new medicines - can be made rapidly, keeping the tool up to date.

We will then work with people with diabetes, doctors, and nurses to determine the best way to present the results, so that they are easy to understand and helpful for informing discussions on which treatments to try. We will work with a computer programmer to make this tool into an online calculator or app.

This research is really important as it will provide a way to help people with diabetes receive the treatment they are most likely to benefit from, and avoid treatments that are likely to give them side effects. This could have important benefits for people with diabetes and the NHS, by reducing the complications of high blood glucose, and reducing the use of medicines which do not work well or cause unpleasant side effects.

Technical Summary

In Type 2 diabetes (T2D) the many different classes of glucose-lowering therapies have an approximately similar efficacy overall and there is huge variation in prescribing patterns between doctors. We have shown in our previous MRC funded collaboration that the clinical characteristics of people with T2D are associated with altered glucose response, which allowed us to build a rational "precision" approach to match an individual to the most effective drug, by integrating routine features in prediction models. Drug decisions are complex and involve tolerability, safety and the impact of co-morbidities as well as efficacy in lowering glucose. This project will use routine clinical and individual trial data from over 1 million people with T2D to create validated integrated models that inform all these aspects of prescribing.

WP1: Best drug for glycaemia. We will extend and test our validated model for glucose lowering response so that it predicts glycaemic response to all non-insulin therapies used after Metformin. The model will then be optimised to work in all major UK ethnic groups.

WP2: Beyond glycaemia. We will create a model that predicts likely tolerability for a person with T2D for all therapies. We will integrate limitations to the therapeutic choice based on safety (e.g. contra-indications) and co-morbidities (e.g. if cardiovascular disease, suggest drugs with non-glycaemic cardiovascular benefit).

WP3: Looking ahead. Utilising genetics and novel biomarkers. We will examine the improvement in prediction by incorporating cutting-edge technologies such as genetics, novel biomarkers and data-driven sub-classification models.

WP4: Integrated tool ready for clinical use. All the inputs and outputs of the model will be integrated and put into an appropriate format for intended users. Working with people with T2D, nurses, and GPs, we will work up visualisations of outputs and create an online calculator/app version of the treatment support tool.


10 25 50