FaisalLab

THANKS: This would have not been possible as a team effort, with fantastic PostDoc, PhDs, and close collaborators in neurology, paediatrics, physiotherapy, genetics. Not fundable without our funders (BRCs, DRF, UKRI, MRC, EPSRC) and last but not least our wonderful patients!

BONUS 2/4: Our approach works both with natural behaviour data, e.g. when patients were eating or in playroom as well as during clinical assessments. This may be the start of a universal approach to disease monitoring in general, especially outside the clinic in our daily lives.

BONUS 1/3: 1. Our full-body behaviour fingerprints are not affected by patients loosing the ability to walk (unlike the need for 2 “gold”standard markers before/after). 2. Our is not confused by aging (DMD “gold”standard goes up&down as child grows paralysed), our’s only progress

What does this mean: First of all,we basically showed how poor established gold-standard clinical behavioural/observational assessments operate when compared to AI derived biomarkers. The gold-standard ain’t so golden, yet they are used as primary endpoints in drug development.

We used much larger clinical trial data (big thanks EFACTS & Gemelli) to boost prediction of clinical scores from clinical scores. Still our method performed far better at predicting future clinical scores & used at equal precision far fewer patients: FA 7 vs 160+ & DMD 9 vs 50+

Predicting gold-standard clinical scales is ceiling limited by the quality of the traditional “gold”-standard itself. So, we showed that we could predict disease progression 6-12 months into the future for every individual patients, unlike the gold-standard itself … 8/10

7/10 In both diseases we were able to reconstruct cross-sectionally the whole range of “gold”-standard clinical scales, i.e. movement data from the same day as the assessment predicted the clinical scale without requiring an expert clinical observer rating the patient by-eye…

6/10 From our previous work on capturing real-world motor behaviour in 100s of volunteers we identified sets of digital behaviour fingerprints that characterise human movements. Thus, one pipeline analysed clinical assessment & natural human behaviour in 2 very different diseases

5/10 In both papers we did longitudinal natural history studies with patients drawn from range of disease progression states of FA & DMD. Among other things we measured full-body movement data & clinical assessments (ie “gold”-standard measures) @ trial start, middle & end.

We focus on two very different rare, degenerative, genetic diseases that affect movement & eventually lead to paralysis & death. There are currently no cures for either disease, the practical benefit is to lower risk, cost, patients needed & duration of drug trials. 2/10