APPROACH

Osteoarthritis presents itself across a spectrum. While one person with OA might experience a small degree of pain moving their knee, another might be unable to walk. It is clear that there should not be a “one size fits all” treatment for OA. Yet within traditional clinical trial settings, it has been common practice to accept all osteoarthritis patients for the same treatment within a clinical trial even though they may present with wildly different symptoms.

This is especially important to remedy because some osteoarthritis treatments – if given to the wrong group – can actually accelerate the progress of this disease. For instance, injecting corticosteroids directly into the knee is a common treatment for OA, but it has been shown to rapidly worsen the condition of some patients. For those patients, treatment with hyaluronic acid (a therapeutic class referred to as ‘viscosupplements’) may be preferable. Although viscosupplements do not tend to tackle pain as effectively as corticosteroids, they work differently by lubricating and cushioning the joint space which may impart a protective effect..

The APPROACH project sought to improve the success rate in osteoarthritis clinical trials by categorising patient profiles into phenotypes. If investigations examine just one phenotype of patients per clinical trial, there might be better results and more targeted, effective therapies may emerge. They collected a wide range of data on more than 10 000 patients and then used bioinformatics and machine learning to analyse it.

Identifying endotypes

The APPROACH project successfully identified three endotypes (i.e. three subgroups with similar molecular characteristics): one which had low tissue turnover and low levels of repair, one which showed more structural damage, and a third which featured systemic inflammation including joint tissue and cartilage degradation. In the future, trials could be conducted that look at just one of these subtypes in order to find a more specific and accurate treatment. This could also impact on the success of clinical trials for osteoarthritis treatment more generally.

Collecting data on more than 10 000 patients

The project amassed a wide range of data thanks to the enthusiastic participation of osteoarthritis patients. Although the project specifically looked at the knee, data was also collected that may be relevant for future studies on hand and hip joint arthritis.

The data collected by the project is available via an integrated bioinformatics platform that functions as a repository of clinical data, biomarker data, images, as well as storage of bio-samples from a broad spectrum of OA patients. The platform is fully accessible via FairPLUS, highlighting a synergy between two IMI projects.

The bioinformatics team that took part in the project were not knowledgeable about osteoarthritis in the beginning, which was an advantage as it meant that there was an internal control preventing bias in the data. For instance, at the start of the project the bioinformatics team were given the data and the machine learning tools rendered predictions. The team working on the machine learning portion of the project had no way of knowing whether the predictions were realistic or not, which meant that they were “double-blind” in a way that the osteoarthritis researchers could not be.

The algorithms developed as part of the APPROACH project were successful in identifying osteoarthritis patients based on a minimum amount of clinical parameters aimed at predicting the likelihood of disease progression within two years. These algorithms combined information on clinical factors, gait, biochemical data, omic and image-based analysis and will be further refined going forwards.

How the data were gathered

Most patients spent at least one full day with the researchers, where they underwent a battery of tests, from gait and motion analyses to urine and blood tests to MRIs and CTs. While the biochemical parameters were gleaned from the urine and blood samples offered by the participants, imaging techniques were very important for these analyses. While an X-ray will only tell a researcher about the bone, MRI and CT scans show damage in the tissues surrounding the bones which can give clues as to which subgroup the patient belongs to.

The gait analyses were carried out by a SME called Dynamic Metrics. Usually such analyses have to be carried out in large labs with specialist equipment, but Dynamic Metrics developed an app which could compute a wide range of data about a person’s gait from just a few small sensors strapped with Velcro around a person’s legs. The data is not as complete as more formal analyses, but it provided the researchers with a wide breadth of information which could be included in the bioinformatics platform without a high cost and without taking up too much of the patients’ time.

What the future holds

The APPROACH consortium partners continue to analyse the data and publications arising from the project are being released regularly. Although the bulk of the work is done, it’s estimated that there is about ten years’ worth of data within the APPROACH datasets still waiting to be analysed. Some of the project’s results will be utilised in another new Horizon project, which is currently awaiting funding, that will examine how to make existing therapies better using stratified patient data. Importantly, the foundation and findings provided by the APPROACH consortium are also expected to foster successful development of the next wave of therapeutics for OA patients.