A new platform that measures the body’s immune-protein response, coupled with machine learning, can accurately distinguish between bacterial and viral infections within minutes – an effective tool in the fight against AMR.
The Challenge: Is it a bacterial or viral infection? To treat or not to treat with antibiotics?
When a patient presents with fever, in many cases, the question comes down to whether it is a bacterial or viral infection, and if to treat, or not to treat, with antibiotics.
Making this diagnosis can be challenging as bacterial and viral infections are frequently clinically indistinguishable.
Tools for detecting pathogens are used to aid the diagnosis, but actionability is often constrained by their inherent limitations:
- Reliance on pathogen sampling, which can be difficult when site of infection is inaccessible or unknown
- Inability to distinguish between detection of a pathogen versus coloniser
- Limited performance against emerging pathogens
As a result, the disease causing pathogen is not clearly identified in as many as two out of three patients with acute infection, even when applying cutting edge microbiological tools.1–3
New diagnostic paradigm: decoding the immune response with machine learning
A complementary diagnostic paradigm has emerged in recent years that overcomes the limitations of direct pathogen detection, namely harnessing the body’s immune-response to infection. It has several advantages:
- No requirement to access the infection site because the immune system circulates throughout the body
- Capability to distinguish between pathogen and coloniser, as the immune system is primarily triggered by disease-causing agents
- Robustness to evolving pathogens, as the immune system is triggered by multiple pathogen features
Currently, individual host-proteins – such as procalcitonin and C-reactive protein – are used to support infection management. However, as single biomarkers, they are often insufficiently accurate due to patient-to-patient variability.
Advancements in host-response profiling and machine learning algorithms have opened the way to a new generation of diagnostics that involve computational integration of multiple biomarkers.
Today, there are several such tools in development. Only one diagnostic has been cleared (currently in Europe) that combines the host-response with machine learning to distinguish between bacterial and viral infections.
Its development included screening over 100,000 biomarker combinations. The best performing combination computationally integrates three circulating host-proteins: TRAIL/IP-10/CRP.1
This signature has since been validated in multiple double-blind international studies enrolling thousands of patients and has shown superiority to routine tests.2–6
Reduction to practice: measuring the immune-protein signature within minutes to fight AMR
Having an accurate immune-signature is not enough. For example, although already used to guide treatment of >10,000 patients in Europe, the impact of the TRAIL/IP-10/CRP signature measured using an ELISA platform has been constrained by the laboratory burden and prolonged turnaround time (two hours).
Broad impact of immune-signatures requires platforms that can measure multiple biomarkers, across a wide dynamic range, quantitatively, rapidly (within 15 minutes), across different clinical settings, in an easy to use manner. Meeting these requirements is readily achievable for circulating host-proteins and will likely take longer for other host biomarker families (e.g. intracellular proteins, RNAs, metabolites).
Several platforms for measuring host-protein signatures are in development, including one to measure TRAIL/IP-10/CRP in under 15 minutes.
Regulatory clearance of these products will pave the way to reducing diagnostic uncertainty when and where needed, improving patient management and reducing antibiotic misuse, ultimately helping the global fight against AMR.
