All talks & lectures

Workshop ·

Prognosis and monitoring

A practical survey of prognostic modelling in multiple sclerosis and the steps needed to move prediction into clinical use.

EAN 2022 — AI & MS Workshop

A web sketch of the deck’s central idea
Delivered
Event
EAN 2022 — AI & MS Workshop
Format
Workshop
Source deck
31 slides

From the slides

Deck notes

This workshop reviewed how clinical, imaging and biomarker data have been used to predict disease activity and disability in multiple sclerosis. It also examined reporting quality, bias and the path to clinical translation.

The question

What must a prognostic model do before it can support monitoring and decisions for an individual patient?

Deck outline

  • Prognostic models for disease activity and progression
  • Predicting phenotypic conversion and long-term disability
  • Imaging and fluid biomarkers
  • Reporting standards and risk of bias
  • Biomarker qualification and regulatory pathways
  • Robustness, recovery endpoints and future directions

Take-away

Prediction alone is not enough: useful models need clear outcomes, transparent reporting and evidence that they remain reliable in the setting where they will be used.

Original deck

Slide by slide

Read the 31-slide presentation in order, or use the index to jump.

  1. Slide 01

    AI & MS Workshop

    Slide 1: AI & MS WorkshopOpen full-size slide

    Prognosis and monitoring

    Arman Eshaghi, MD, PhD

    Senior Research Fellow University College London

    &

    Queen Square Analytics

    UK

  2. Slide 02

    Conflicts of interest

    Slide 2: Conflicts of interestOpen full-size slide
    • Received travel support from the National Multiple Sclerosis Society and honorarium from the Journal of Neurology, Neurosurgery and Psychiatry for Editorial Commentaries.
    • Received research grants from Biogen, Merck and Roche through his institutions.
    • Founder and equity stake holder in Queen Square Analytics Limited.
    • Serves on the editorial board of Neurology (American Academy of Neurology).

  3. Slide 03

    Outline

    Slide 3: OutlineOpen full-size slide
    • Prognosis and monitoring
    • Literature highlights
    • Path to clinical translation
    • Limitations and future perspectives

  4. Slide 04

    Time to event

    Slide 4: Time to eventOpen full-size slide

    Prognostic models

    Input variables

    Risk

    Clinical

    Outcomes

    (events)

    Statistical models

    Blood biomarkers

    Imaging

    Machine learning

    Genetics

  5. Slide 05

    Outcomes in

    Slide 5: Outcomes inOpen full-size slide

    Prognostic modelling

    Phenotypic:

    CIS to RR

    RR to SP

    NEDA-3, 4 and NEPAD*

    Disease activity

    Treatment response scores (Rio, etc.)

    Progression

    Confirmed disability progression

    Progression independent of relapses

    (PIRA)

    Disability accrual/worsening

    *No evidence of progression or active disease

    Speaker notes

    https://www.uptodate.com/contents/indications-for-switching-or-stopping-disease-modifying-therapy-for-multiple-sclerosis

  10. Slide 10

    Kalincik et al, Towards personalized therapy for multiple sclerosis, Brain, 2017

    Slide 10: Kalincik et al, Towards personalized therapy for multiple sclerosis, Brain, 2017Open full-size slide
    Speaker notes

    An example of individual prediction of response to interferon β-1a and natalizumab in moderately advanced, active multiple sclerosis. Six treatment outcomes are predicted for a 43-year-old female with relapsing-remitting multiple sclerosis, with an EDSS score of 5 (with sensory functional score of 4, and pyramidal, cerebellar and ambulatory functional scores of 3) and increase in the EDSS score within the previous year, who first presented with spinal cord symptoms at the age of 33. The patient had previously experienced 10 multiple sclerosis relapses, mostly with pyramidal symptoms, six of these of high severity and one while treated with DMTs, with the most recent relapse having occurred 51 days before the date of the prediction. She was previously treated with two DMTs, her most aggressive DMT was natalizumab and she discontinued interferon β-1a subcutaneous 51 days before the prediction. Her brain and spinal cord MRI showed abnormal findings in keeping with the diagnosis of multiple sclerosis and her CSF showed oligoclonal bands that were not present in the serum. Information about 74–93% of the variables informing the predictive models was available. The curves represent the most probable number of outcome events or area under EDSS-time curve (±95% prediction interval) over the next 4 years if recommencing natalizumab or interferon β-1a subcutaneous. The shading of the curves illustrates the robustness of the prediction (quantified as the product of the size of the training cohort and the accuracy of the prediction in the testing cohort).

    References

  11. Slide 11

    Source: Eshaghi et al, Predicting personalized risk of disability progression, under revision

    Slide 11: Source: Eshaghi et al, Predicting personalized risk of disability progression, under revisionOpen full-size slide

  21. Slide 21

    Shortcomings

    Slide 21: ShortcomingsOpen full-size slide

    External validation and generalizability

    Real-world, prospective evaluation

  30. Slide 30

    Summary

    Slide 30: SummaryOpen full-size slide

    Outcome measures

    Prognostic modelling research in neurology is in its infancy

    Lengthy path to translation: patient benefit and cost-effectiveness

    Future: more robust research, emerging outcomes, double blind randomized controlled trials

Continue through the archive

Related talks

Nearby ideas from a different year, event or audience.