An intriguing report has just appeared in the ‘International Journal of Machine Learning and Computing’ – an unusual home for a scientific paper on ME/CFS. At first reading, it seems very complicated – the researchers from the Jason group at DePaul University in Chicago used ‘unsupervised machine learning’ (a series of statistical techniques for finding hidden structures in data) to examine information on ME/CFS patients collected using the DePaul Symptom Questionnaire (DSQ).
They brought together information collected at 3 different sites: data collected during an ME Research UK-funded study on the use of the DSQ in patients at Newcastle University, UK; data collected by DePaul University itself; and BioBank sample data from the CFIDS Association of America. All the patients (515, with 176 controls) were classified according to their fulfilment of 3 case definitions developed by consensus but not by experimentation (Fukuda 1994 CFS, Canadian 2003 ME/CFS, and ICC 2012 ME), and the raw data was then put through ‘machine learning’ techniques.
The upshot was that an ‘empirical’ definition based on the selection of 11 specific symptoms seemed to provide greater diagnostic accuracy than any of the 3 consensus-based criteria. These symptoms (in order of predictive accuracy) were:
1. Fatigue/extreme tiredness
2. Next day soreness or fatigue after non-strenuous, everyday activities
3. Minimum exercise causing physical tiredness
4. Physically drained or sick after mild activity
5. Dead, heavy feeling after starting to exercise
6 Feeling unrefreshed after waking up in the morning
7 Problems remembering things
8 Muscle weakness
9 Difficulty finding the right word to say or expressing thoughts
10 Only able to focus on one thing at a time
11 Pain or aching in muscles
If this intriguing attempt to establish an empirical (experimental) basis for a diagnosis is found to be valid, it could – at a stroke – end arguments over the relative merits of the many consensus definitions (of CFS, ME, ME/CFS, CFS/ME) that currently exist, and simplify diagnosis at the bedside or clinic.
Reference: Identifying defining aspects of chronic fatigue syndrome via unsupervised machine learning and feature selection. Watson SP, et al. International Journal of Machine Learning and Computing, 2014 April. Read more (full text).