GUIDE-AI

For most chronic conditions, medical associations and experts draw on the available evidence to prepare treatment guidelines that detail what medicines patients should receive and when to have the best chance of treating the patient successfully and improving their quality of life. The guidelines, which are often hundreds of pages long, can be accessed by healthcare providers. However, it is increasingly difficult for many doctors to keep up to date with all the guidelines, and studies show that just a fraction of chronic disease patients are treated according to the guidelines for their condition.

The ultimate goal of GUIDE-AI is to harness the power of generative artificial intelligence (AI) to boost the proportion of chronically ill patients receiving the treatments as set out in the guidelines. To do this, the project will develop ‘guideline navigators’. These will use large language models (LLMs) to allow doctors to quickly and easily identify the right treatment for their patients as recommended by guidelines. In parallel, the project will use the same LLM backbone to deliver resources for patients explaining in lay language why their treatments have been selected for them.

The project will focus on four chronic conditions that affect a large proportion of people in Europe and where studies indicate that only a minority of patients are treated according to the guidelines: heart failure with reduced left ventricular ejection fraction (HFrEF); chronic kidney disease (CKD); chronic obstructive pulmonary disease (COPD); and asthma.

As a first step, the project team will analyse current barriers to greater use of guidelines in treatment through discussions with doctors and patients as well as healthcare administrators and payers. They will also retrieve relevant guidelines from, and collaborate with, the respective medical specialist organisations both at the international and national/local level. On the LLM side, the team will explore a range of models, and is committed to remaining independent from any single specific vendor or LLM.

Next, computer scientists and clinicians will jointly define the technical set-up of the guideline navigators. This will including setting out the medical information required by the guidelines to offer treatment recommendations, ensuring privacy, and regulatory aspects.

The guideline navigators will be tested on existing data before undergoing evaluation in a prospective, randomised study. Here, the team will assess the proportion of patients receiving guideline-directed therapies, and the number of patients whose treatment is changed following LLM analysis..

Although the project focuses on four common diseases, it should be possible to adapt the resulting blueprint to other diseases with guidelines.

Ultimately, by increasing the proportion of patients treated according to guidelines, GUIDE-AI will improve care and patients’ quality of life and boost the efficient use of resources in healthcare as patients should receive the right treatment sooner.