- PREDYSTIC® Infliximab RA Kit (PREDYSTIC®) is a validated algorithm to predict infliximab response in rheumatoid arthritis (RA) patients.
- PREDYSTIC® helps physicians to pre-emptively identify patients who would respond best to infliximab treatment. Those classified as low or moderate responders may benefit from alternative therapies sooner, thereby saving healthcare costs.
- Key learnings from diagnostic development: 1. Use machine learning techniques to detect patterns, remember to validate algorithm 2. Ensure algorithm is medically valid to establish trust with physicians 3. Check compliance with the new EU In vitro Diagnostics Regulation (IVDR).
- Egis is seeking partners to commercialise the technology for clinical practice.
Centivis: What is PREDYSTIC® and why is it needed?
Zsolt Hollo: PREDYSTIC® is a blood test combined with an algorithm, which predicts how likely a patient will respond to the RA biologic treatment infliximab. Based on clinical data, about 65% of patients are low or mid-responders to infliximab. PREDYSTIC® identifies these patients upfront so that physicians can consider prescribing alternative therapies (e.g. anti-CD20 drugs, IL6 treatments) which may be more effective for them. Not only might these patients benefit from alternative therapies sooner, PREDYSTIC® saves them 6 months of pain and healthcare costs associated with unsuitable treatment.
With PREDYSTIC®, physicians can be 88% accurate in classifying patients into low, moderate, or good responders to infliximab versus 35% accuracy in current practice. So, this diagnostic is a game changer. When we speak about personalised treatment, we often associate it with oncology. However, PREDYSTIC® makes personalised treatment a reality for RA patients as well.
Centivis: It seems like diagnostics are moving towards combining health information and different biomarkers into algorithms for more accurate predictions. PREDYSTIC® is a great example. How does it work?
Zsolt Hollo: PREDYSTIC® should be used in RA patients who have not used any biologic drugs but have responded inadequately to csDMARDs (conventional synthetic disease-modifying antirheumatic drugs) such as methotrexate.
It combines gene expression, lab parameters (e.g., inflammatory marker status) and health information in a proprietary algorithm that predicts how well patients would respond to infliximab after 6 months.
Specifically, the algorithm indicates how severe a patient’s condition might be after 6 months of infliximab treatment, as measured by the commonly used DAS28-CRP scale (Disease Activity Score for Rheumatoid Arthritis).
Those with a score of above 3.2 are classified as non-responders as their disease activity would be high or moderate 6 months after starting infliximab. These patients could benefit from non-infliximab therapies instead.
Centivis: What were the key learnings from developing such an algorithm?
Zsolt Hollo: I would highlight the following three key learnings:
1. Use machine learning-based techniques to detect patterns, remember to validate the algorithm
In our clinical trials, we recruited 217 RA patients and took data at different timepoints including RA status and genetic data. We had huge amounts of transcriptome-wide gene data, so the challenge was to identify the appropriate genes in combination with relevant medical factors, which would predict infliximab response. Thanks to machine learning-based techniques, we narrowed the data down to a pool of 48 genes which showed significant differences between responder and non-responder groups. To ensure that our findings were meaningful and not random, we also clinically validated the algorithm in an independent cohort of 51 patients.
2. Ensure algorithm makes medical sense to establish physician trust
Solid evidence is required for physicians to consider using this diagnostic in their daily practice. Thus, we invested in a clinical trial to arm ourselves with essential evidence when speaking with them. Furthermore, it is not merely about finding random patterns, but doing so in a way which squares with medical knowledge. To gain trust with physicians, we had to ensure that the genes identified, in combination with relevant medical factors, were explainable given the current medical understanding of the disease (e.g., genes identified are associated with inflammatory response etc.).
Physicians have responded enthusiastically so far as they find the tool clinically valid, and they see the ultimate need for such a diagnostic tool.
3. Check compliance with the new EU IVDR
The current regulatory rules for laboratory diagnostics are being phased out and Europe is moving towards the new EU IVDR (in vitro diagnostics regulation).
We are aware of this and have been keeping track of the changing regulatory landscape. The biggest change is that clinical trials and validation are required under the new regulation. We have checked that we have the corresponding data to be IVDR-compliant.
Centivis: What are the next steps for PREDYSTIC®?
Zsolt Hollo: First and foremost, we are seeking partners (e.g., pharmaceutical companies, lab partners) to commercialise the technology – we would ultimately like to launch this service into clinical practice so it can make a difference to patients.
Secondly, we could expand the test to other indications such as Crohn’s disease and explore it with respect to other drugs (e.g., other TNF-alpha blockers, CD20 drugs etc.). There are many potential routes to explore, and the future is bright for such a diagnostic.
Original interview available at CENTIVIS - PREDYSTIC interview