ECTRIMS 2021 | Developing a normative database to improve sNfL performance in MS

At the end of the day, it is important to accurately determine what is a normal serum NfL level in control persons to be able to depict subtle increases in neurofilament levels. We performed a study involving about 10,000 control persons’ serum samples where we looked at the already more or less known association between NfL and age. We also confirmed the relatively strong association between BMI and NfL. We generated a normative database by producing Z scores, or percentiles, which are relatively interchangeable to each other, which then produce Z scores in MS patient samples. So by having a Z score in an MS patient sample, you know how many standard deviations the MS patient’s NfL deviates from the control’s mean of relatively same age. So leading to a highly precise estimate of the deviation of serum NfL compared to controls.

Regarding the fixed cutoff, it is known that CSF NfL, and there’s plenty of studies showing that serum NfL is actually age-dependent. This we’ve shown in the Swiss MS Cohort Study where we included approximately 8,000 longitudinal serum samples from 1,300 patients that have been followed for a median of five years. We see that the risk is that younger patients fall below these fixed cut-offs, because obviously the age effect is less strong in these younger patients. By defining their NfL by Z scores or percentiles we could clearly see that, especially in the young patients, values below a fixed cut-off are already pathological. On the contrary, older patients, maybe also more relevant for progressive MS, showed more false positive results by using a fixed cut-off. We were able to show that patients that are wrongly negative or wrongly positive classified, do indeed show more disease activity in the previous year and also in the subsequent year.

We’ve looked into 1,300 MS patients with more than 7,000 longitudinal samples. We clearly saw that Z scores predict disease activity in the next year. So patients with a Z score above two had about 3.5 fold risk of disease activity, meaning MRI or clinical disease activity in the next year. When we compared the prognostic value of Z scores in direct comparison in a multivariate model to currently used standard measures in clinical practice, like new T2 gadolinium lesions, relapses in the last year, or EDSS worsening, we saw that serum NfL Z scores add independent prognostic value.

We saw the same also in patients with NEDA-3 status in the last year. So patients being NEDA-3, but having increased NfL values had more disease activity in the next year. So this is a relatively strong argument that NfL is able to likely depict subclinical or sub-MRI disease activity in these patients. We are developing an NfL shiny app, accessible to investigators, where people after entering BMI and age or only age can calculate Z scores or percentiles for their serum NfL levels based on this normative database comprising about 4,500 healthy controls, because we believe it is important to use the optimal adjusted NfL for future research and to fully be able to make use of the signal we are capturing by this biomarker.