Article: Rethinking ME/CFS Diagnostic Reference Intervals via Machine Learning, and the Utility of Activin B for Defining Symptom Severity
Brett A. Lidbury 1,* , Badia Kita 2 , Alice M. Richardson 1 , Donald P. Lewis 3 , Edwina Privitera 3 , Susan Hayward 4 , David de Kretser 2,5 and Mark Hedger
National Centre for Epidemiology and Population Health, RSPH, College of Health and Medicine, The Australian National University, Canberra, ACT 2601, Australia 2 Paranta Biosciences Limited, Suite 549, 1 Queens Rd, Melbourne, VIC 3004, Australia 3 CFS Discovery, Donvale Specialist Medical Centre, Donvale, VIC 3111, Australia 4 Centre for Reproductive Health, Hudson Institute of Medical Research, Clayton, VIC 3168, Australia 5 Department of Anatomy and Developmental Biology, School of Biomedical Sciences, Monash University, Clayton, VIC 3800, Australia
This research was funded by the Judith. J. Mason and Harold S. Williams Memorial Foundation (The Mason Foundation),
Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) – a complex and disabling disease that affects many parts of the body, including the brain and muscles, as well as the digestive, immune and cardiac systems, among others. ME is classified as a neurological disorder by the World Health Organization.
Activins –activins belong to the TGF-beta superfamily of proteins, and have diverse biological functions including in reproduction, apoptosis, cell proliferation and immune activity
Cytokine – small proteins released by many different cells in the body, including those of the immune system where they coordinate the body’s response against infection and trigger inflammation
Random forest (RF) – a type of supervised machine learning algorithm based on ensemble learning
Weighted standing time– measure of illness severity, by assessing orthostatic intolerance
Biomarker- measurable indicator of the severity or presence of some disease state
There is no validated diagnostic test for ME/CFS. Building on previous research, this research aims to evaluate activin B as a potential serum marker for MS/CFS.
All the participants were recruited via CFS Discovery, either via direct invitation to existing patients, or responses to advertising locally, and via social networking sites. Only participants with a previous ME/CFS diagnosis were recruited. Eighty-five (85) participants were initially recruited for the ME/CFS cohort, with five eventually excluded due to comorbidities and/or difficulties attending the required appointments. Seventeen (17) healthy control (HC) participants were recruited too and underwent the same assessment as the ME/CFS cohort, giving a total study cohort size of 97 participants. To provide a proxy for ME/CFS severity, the weighted standing time test was performed.
Research methods included pathology, clinical testing and pattern recognition algorithm random forest (RF), to identify wider marker patterns that separate ME/CFS cases from healthy controls.
Statistical analysis and machine learning were applied to analyse the data.
Activin B was significantly lower (p = 0.013) for the ME/CFS cohort compared to results from the HC participant cohort. This is an inversion of the previous results, which found that activin B was significantly elevated in ME/CFS participants. However, further analysis revealed that activin B is most useful for separating healthy individuals from patients experiencing moderate ME/CFS symptoms.
The re-calibration of the activin B assay, due to sensitivity variation across the range of detection, improved the accuracy of the assay at lower serum concentrations, thus enhancing activin B detection capacity and broadening the reference interval range, which may explain the differences in activin B results found for this study when compared to the previous results.
The potential to develop activin B as a general serum marker of ME/CFS needs multi-centre studies with large participant cohorts. Further investigation also hopes to identify additional biomarker patterns.
The longer-term aim is to develop simpler diagnostic tools from routine data to assist health professionals diagnose ME/CFS. Reference intervals would provide simpler and accurate guidance to clinicians supporting ME/CFS patient, which can be facilitated by random forest prediction.
There is no available animal model properly reflecting the clinical features of CFS. This promising research highlights the value of using data from ME/CFS patients and the possibilities of machine learning to develop diagnostic tools.