Supplement
Neurofilament Light and Heterogeneity of Disease Progression in Amyotrophic Lateral Sclerosis: Development and Validation of a Prediction Model to Improve Interventional Trials
Exploratory analysis of blood Neurofilament Light’s (NfL) impact on multivariate regression models predicting survival
Introduction:
Precise predictive models for surviva l could be a complementa ry tool for future interventiona l tria ls in
Amyotrophic La tera l Sclerosis (ALS). They could be used to increa se sta tistica l power in studies with surviva l as the prima ry endpoint or for rea sons of stra tified ra ndomiza tion a nd a nticipa tion of dropouts . Although our study focuses a nd hence wa s designed to develop a nd va lida te a model predicting disea se progre ssion, we a imed to explore the benefit blood NfL could provide for the prediction of surviva l.
Methods:
For surviva l a na lyses, we pooled the da ta from the three cohorts used in our study. Overa ll, 32 pa tients (26%) of a ll pa rticipa nts were followed up to dea th. This group consists of 10 pa tients with spina l onset ALS from the development cohort (DC), 14 pa tients with spina l onset a s well a s 8 pa tients with bulba r onset from va lida tion cohort 1 (V1).
We fitted a multiva ria te linea r regression model with the sa me ca ndida te predictors used for the disea se
progression model: ba seline ln(NfL), sex, a ge, site of onset (bulba r or spina l), body ma ss index (BMI), monthly ALSFRS-R decrea se between disea se onset a nd ba seline (𝚫FRS), a nd ALSFRS-R score a t ba seline. We a ga in used sequentia l F-tests to elimina te non-significa nt predictors one by one, only this time we used overa ll surviva l (in months since dia gnosis) instea d of the ALSFRS-R slopes.
Additiona lly, we included a ll pa tients of the pooled cohort in a Ka pla n Meier a na lysis for surviva l since disea se onset with subgroups defined a s NfL ba seline a bove the media n (>86 pg/ml) or below the media n (≤86 pg/ml).
The log-ra nk test wa s used to test for significa nce. Pa tients not decea sed throughout follow-up were censored a t the time of the la st visit.
Results and Discussion:
Eva lua ting the impa ct NfL could ha ve on multiva ria te regression models for surviva l, we found tha t including ln(NfL) consistently improved model performa nce (p<0.05, F-test) (Supplement Table 1). In contra st to the prediction model for ALSFRS-R slopes, overa ll surviva l wa s significa ntly correla ted with 𝚫FRS. 𝚫FRS a nd ln(NfL) were sta tistica lly significa nt in a ll models we tested. Given the sma ll number of decea sed pa tients (n=32), the a na lysis presented here is intended a s explora tion. Still, we find surviva l is significa ntly correla ted with NfL blood levels a t ba seline.
Supplement Table 1. Impact of NfL on Survival Models
Predictors and p-values
( ˙ P < 0.1, * P < 0.05, ** P < 0.01)
Correlation
(squa re root of a djusted R²)
Correlation for same model without ln(NfL)
Including all candidate predictors ln(NfL) 0.01 **
site of onset 0.08 ˙ ALSFRS-R score 0.02 * age at onset 0.16 𝚫FRS 0.01 **
sex 0.52 BMI 0.69
ln(NfL):site of onset 0.54 ln(NfL):sex 0.05 *
0.66 0.59
After removing BMI ln(NfL) 0.01 **
site of onset 0.06 ˙ sex 0.56
ALSFRS-R score 0.05 * age at onset 0.25 𝚫FRS 0.01 **
ln(NfL):site of onset 0.47 ln(NfL):sex 0.05 *
0.65 0.58
After removing age at onset and NfL:site of onset
ln(NfL) 0.01 **
site of onset 0.06 ˙ sex 0.56
ALSFRS-R score 0.05 * 𝚫FRS 0.01 **
ln(NfL):sex 0.04 *
0.66 0.58
After removing sex ln(NfL) 0.01 **
site of onset 0.07 ˙ ALSFRS-R score 0.05 * 𝚫FRS 0.01 **
0.61 0.59
After removing site of onset ln(NfL) 0.01 **
ALSFRS-R score 0.07 ˙ 𝚫FRS 0.01 **
0.58 0.53
After removing ALSFRS-R score ln(NfL) 0.02 *
𝚫FRS 0.05 *
0.45 0.41
Table legend: : = interaction between candidate predictors
A Ka pla n Meier a na lysis for surviva l since disea se onset with subgroups defined a s NfL ba seline a bove media n (>86 pg/ml) or below median (≤86 pg/ml) showed significantly better survival in patients with low ba seline NfL levels (P < 0.001, Supplement Figure 1). The Ka pla n Meier results fit the results of previous studies performing simila r a na lyses (1-4), confirming tha t NfL is a strong predictor for surviva l in pa tients with ALS.
Supplement Figure 1. Kaplan Meier
Ka pla n Meier plot for surviva l since disea se onset for pa tients split a ccording to their ba seline NfL levels. If not decea sed, pa tients were censored a t la st visit. Pa tients with NfL ba seline levels below media n (≤ 86 pg/ml; blue) showed significa ntly longer surviva l (P < 0.001, log-ra nk) tha n pa tients with ba seline NfL levels a bove media n (> 86 pg/ml, red).
Conclusion:
Our explora tion points towa rd blood NfL a t ba seline being a promising bioma rker to improve models predicting surviva l in ALS. Due to the low number of decea sed ca ses in our study, results a re intended only a s explora tion for future resea rch.
Supplement References
(1) Verde F, Steina cker P, Weisha upt JH, Ka ssubek J, Oeckl P, Ha lbgeba uer S, et a l. Neurofila ment light cha in in serum for the dia gnosis of a myotrophic la tera l sclerosis. J Neurol Neurosurg Psychia try 2019;90(2):157 -164.
(2) Thouvenot E, Dema ttei C, Lehma nn S, Ma ceski‐Ma leska A, Hirtz C, Junta s‐Mora les R, et a l. Serum neurofila ment light cha in a t time of dia gnosis is a n independent prognostic fa ctor of surviva l in a myotrophic la tera l sclerosis. Europea n Journa l of Neurology 2020;27(2):251 -257.
(3) Lu C, Ma cdonald-Wa llis C, Gra y E, Pea rce N, Petzold A, Norgren N, et a l. Neurofila ment light cha in.
Neurology 2015 Ja n 1,;84(22):2247-2257.
(4) Steina cker P, Huss A, Ma yer B, Grehl T, Grosskreutz J, Borck G, et a l. Dia gnostic a nd prognostic significa nce of neurofila ment light cha in NF-L, but not progra nulin a nd S100B, in the course of a myotrophic la tera l sclerosis: Da ta from the Germa n MND-net. Amyotroph La tera l Scler Frontotempora l Degener 2017 Feb;18(1-2):112-119.