This study is mostly Greek to me. I prefer your smouldering explanation. Personally I prefer a Doctor not doing a lumbar puncture and repeating MRI’s, especially as I am now 67 and deteriorating physically. A thorough physical examination, subjective history, questionnaire several weeks before an MS clinic appointment could give you an idea of where I am in the spectrum. Yes, and one for my husband or wife, caregivers, therapists. I’ll go back to gadolinium used for MRI’s. Ever see retained gadolinium nanoparticles or their effects? I came today to see how I can get prior authorization for an electric standing wheelchair. I believe your approach and consideration of studies helps me understand Greek translation. For prior authorization of an electric standing wheelchair, smouldering is a picture I want to paint. NFL levels not in the criteria.
You always have a choice to say No. If your doctor doesn’t agree, you have a choice to change doctors. When my doctors didn’t believe me and I was admitted comatose to ER, I realized, listing “gadolinium and gadolinium derivatives” as an allergy.
Gladolinium did not react well with me when I had it injected during an MRI Exam. All I can say is that I will never have it again. I was incredibly ill with migraines. Xx
Do I understand correctly that this study used AI? If so, it feels like a desperate attempt to do domething with AI, just because we can. I can't see the point of making any kind of conclusion when training and testing cohorts were so different. We looked at young people without recent relapse and not on treatment and compared them to old people with recent relapse established on medication for a prolonged period. Guess what, they are different!
Yes, they used machine learning, which is a form of AI. They didn't really compare the groups; they used one group to train the model and the other to validate it. The problem is that neither group is really representative of MS; both are highly selected populations. I think the model's findings are not new and confirm what we already know about MS.
My engineer daughter was saying, the other day, that they have been given a load of AI /machine learning tools to help them. Her problem is that she hasn't been able to replicate the output that her colleagues have achieved, even when shes tried to input the exact same training data etc as they did. She also said that repeating her own attempts produced wildly different results. She thinks that the problem might be that the way she works is different because of her dyslexia. At the moment she is not a fan. "A new graduate with no experience of our business can learn to do this in a week, I can do it in a day. I've been told I'm not booking enough hours to the project. Where's the time / money for me to try to teach the machine, and myself how drive it, to reap the benefit every subsequent time this repetitive task comes round.
Thank you for this very clear and critical commentary. Your points on continuum versus “distinct” subtypes, therapeutic lag, and the risk of over-interpreting cross-sectional biomarker models are particularly well taken.
Building on similar concerns, we recently published a small, monocentric, cross-sectional study at MS diagnosis using multimodal biomarkers (serum/CSF NfL, OCT retinal measures, cognition and clinical variables) combined with unsupervised methods. Importantly, we deliberately avoided framing our findings as new MS subtypes, but rather as early phenotypic patterns that may reflect different temporal dynamics within a single disease continuum.
The work is clearly exploratory and shares many of the limitations you highlight, but we thought it might be of interest in the context of your discussion on smouldering MS and biological lag.
Zanghì A et al. A multimodal approach to distinguish multiple sclerosis phenotypes at diagnosis using biomarker profiles. Therapeutic Advances in Neurological Disorders, 2025.
Thanks. Your findings are similar to those reported in the BRAIN paper. The first group with raised NFL levels, more activity (relapses & MRI) with relative preservation of the end organ (retinal nerve fibre layer), and a second group with an opposite phenotype. I note there were no differences in age or age of onset between the two groups. It would be great to look for a biomarker of resilience or reserve to see if it adds anything.
The question I have is whether identifying or categorising patients in either subgroup changes treatment strategies? I suspect not.
To complicated for me , I do know I will never has a LP again as for weeks after my hearing was greatly affected to the point of not thinking I could take it any longer
The critique around temporal lag and regression to the mean is really strong. The fact that 23% of patients switched subtypes in testing kinda undermines the entire premise of "distinct" categories. I worked on a project years ago involving biomarker sequencing for another condition, and the biggest challenge was always accounting for how different metrics evolve at asynchronous rates. It's frustrating when a well-intentioned model gets overinterpreted without acknowleging these fundamental dynamics.
Thank you for explaining this study and why it has so many flaws. It takes a scientist/doctor to really analyze a study to make sure it proves /disproves the hypothesis. It brings me back to grad school days but I can only admit to understanding by the skin of my teeth some of the explanation. I do have a somewhat related MS question. I recently had a couple of labs done for another medical problem which revealed Serum light chain markers which could be a sign of multiple myeloma. There were no other indications of MM. could these just be the result of 8/9 years on Ocrevus ( which I stopped in April 24?). My hematologist has been repeating the tests at each visit to make sure he’s not missing something. Thank you.
The test for MM in the blood is usually a serum protein electrophoresis to detect a monoclonal gammopathy. Light chains are generally detected in the urine (Bence-Jones proteins) as they are rapidly cleared from the blood by the kidneys. MM is a malignancy of plasma cells and, to the best of my knowledge, is not necessarily linked to anti-CD20 therapy. Not all monoclonal gammopathies become myeloma. This is why they are often initially diagnosed as an MGUS (monoclonal gammopathy of undetermined significance) and followed serially. If you see a haematologist, you will need further tests, including a bone marrow aspirate and biopsy.
what do you think about the failure of Tolebrutinib recently? It may appear that Clene Nanomedicines CNM-AU8 may be a safer bet at getting approval because it is much safer than BTK inhibitors and addressing Mitochondria in the brain might reduce or halt smouldering MS because the Mitochondria is compromised in MS. NAD+ and boosting ATP seems critical in MS? Any thoughts?
This study is mostly Greek to me. I prefer your smouldering explanation. Personally I prefer a Doctor not doing a lumbar puncture and repeating MRI’s, especially as I am now 67 and deteriorating physically. A thorough physical examination, subjective history, questionnaire several weeks before an MS clinic appointment could give you an idea of where I am in the spectrum. Yes, and one for my husband or wife, caregivers, therapists. I’ll go back to gadolinium used for MRI’s. Ever see retained gadolinium nanoparticles or their effects? I came today to see how I can get prior authorization for an electric standing wheelchair. I believe your approach and consideration of studies helps me understand Greek translation. For prior authorization of an electric standing wheelchair, smouldering is a picture I want to paint. NFL levels not in the criteria.
I know that gadulinium does to my body, and at 59 feel that I can make a choice as to whether I must include the graduation....
You always have a choice to say No. If your doctor doesn’t agree, you have a choice to change doctors. When my doctors didn’t believe me and I was admitted comatose to ER, I realized, listing “gadolinium and gadolinium derivatives” as an allergy.
You are right of course. I will deny the next time I am going into the tube.
Gladolinium did not react well with me when I had it injected during an MRI Exam. All I can say is that I will never have it again. I was incredibly ill with migraines. Xx
Do I understand correctly that this study used AI? If so, it feels like a desperate attempt to do domething with AI, just because we can. I can't see the point of making any kind of conclusion when training and testing cohorts were so different. We looked at young people without recent relapse and not on treatment and compared them to old people with recent relapse established on medication for a prolonged period. Guess what, they are different!
Yes, they used machine learning, which is a form of AI. They didn't really compare the groups; they used one group to train the model and the other to validate it. The problem is that neither group is really representative of MS; both are highly selected populations. I think the model's findings are not new and confirm what we already know about MS.
My engineer daughter was saying, the other day, that they have been given a load of AI /machine learning tools to help them. Her problem is that she hasn't been able to replicate the output that her colleagues have achieved, even when shes tried to input the exact same training data etc as they did. She also said that repeating her own attempts produced wildly different results. She thinks that the problem might be that the way she works is different because of her dyslexia. At the moment she is not a fan. "A new graduate with no experience of our business can learn to do this in a week, I can do it in a day. I've been told I'm not booking enough hours to the project. Where's the time / money for me to try to teach the machine, and myself how drive it, to reap the benefit every subsequent time this repetitive task comes round.
Thank you for this very clear and critical commentary. Your points on continuum versus “distinct” subtypes, therapeutic lag, and the risk of over-interpreting cross-sectional biomarker models are particularly well taken.
Building on similar concerns, we recently published a small, monocentric, cross-sectional study at MS diagnosis using multimodal biomarkers (serum/CSF NfL, OCT retinal measures, cognition and clinical variables) combined with unsupervised methods. Importantly, we deliberately avoided framing our findings as new MS subtypes, but rather as early phenotypic patterns that may reflect different temporal dynamics within a single disease continuum.
The work is clearly exploratory and shares many of the limitations you highlight, but we thought it might be of interest in the context of your discussion on smouldering MS and biological lag.
Zanghì A et al. A multimodal approach to distinguish multiple sclerosis phenotypes at diagnosis using biomarker profiles. Therapeutic Advances in Neurological Disorders, 2025.
DOI: https://doi.org/10.1177/17562864251369747
Thanks. Your findings are similar to those reported in the BRAIN paper. The first group with raised NFL levels, more activity (relapses & MRI) with relative preservation of the end organ (retinal nerve fibre layer), and a second group with an opposite phenotype. I note there were no differences in age or age of onset between the two groups. It would be great to look for a biomarker of resilience or reserve to see if it adds anything.
The question I have is whether identifying or categorising patients in either subgroup changes treatment strategies? I suspect not.
To complicated for me , I do know I will never has a LP again as for weeks after my hearing was greatly affected to the point of not thinking I could take it any longer
Yes it creates a vacuum inside your skull cap because your brain rests in it. I just have horrible headaches until my body can produce more...
The critique around temporal lag and regression to the mean is really strong. The fact that 23% of patients switched subtypes in testing kinda undermines the entire premise of "distinct" categories. I worked on a project years ago involving biomarker sequencing for another condition, and the biggest challenge was always accounting for how different metrics evolve at asynchronous rates. It's frustrating when a well-intentioned model gets overinterpreted without acknowleging these fundamental dynamics.
Thank you for explaining this study and why it has so many flaws. It takes a scientist/doctor to really analyze a study to make sure it proves /disproves the hypothesis. It brings me back to grad school days but I can only admit to understanding by the skin of my teeth some of the explanation. I do have a somewhat related MS question. I recently had a couple of labs done for another medical problem which revealed Serum light chain markers which could be a sign of multiple myeloma. There were no other indications of MM. could these just be the result of 8/9 years on Ocrevus ( which I stopped in April 24?). My hematologist has been repeating the tests at each visit to make sure he’s not missing something. Thank you.
The test for MM in the blood is usually a serum protein electrophoresis to detect a monoclonal gammopathy. Light chains are generally detected in the urine (Bence-Jones proteins) as they are rapidly cleared from the blood by the kidneys. MM is a malignancy of plasma cells and, to the best of my knowledge, is not necessarily linked to anti-CD20 therapy. Not all monoclonal gammopathies become myeloma. This is why they are often initially diagnosed as an MGUS (monoclonal gammopathy of undetermined significance) and followed serially. If you see a haematologist, you will need further tests, including a bone marrow aspirate and biopsy.
what do you think about the failure of Tolebrutinib recently? It may appear that Clene Nanomedicines CNM-AU8 may be a safer bet at getting approval because it is much safer than BTK inhibitors and addressing Mitochondria in the brain might reduce or halt smouldering MS because the Mitochondria is compromised in MS. NAD+ and boosting ATP seems critical in MS? Any thoughts?