Article Text

Download PDFPDF

Research paper
What does the ALSFRS-R really measure? A longitudinal and survival analysis of functional dimension subscores in amyotrophic lateral sclerosis
  1. James Rooney1,
  2. Tom Burke1,2,
  3. Alice Vajda1,
  4. Mark Heverin1,
  5. Orla Hardiman1,2
  1. 1Academic Unit of Neurology, Trinity Biomedical Sciences Institute, Trinity College Dublin, Dublin, Ireland
  2. 2Beaumont Hospital, Dublin, Ireland
  1. Correspondence to Dr James Rooney, Academic Unit of Neurology, Trinity Biomedical Sciences Institute, Trinity College, 152-160 Pearse Street, Dublin 2, Dublin D02 R590, Ireland; jrooney{at}


Introduction ALS functional rating scale (revised) (ALSFRS-R) is the most widely used functional rating system in patients with amyotrophic lateral sclerosis (ALS). However, heterogeneity in ALSFRS-R progression renders analysis challenging. We have explored the characteristics of total ALSFRS-R, and ALSFRS-R subscores in longitudinal and survival models, to determine whether subscore analysis enhances the precision of the instrument.

Methods All cases with ALSFRS-R scores on the Irish ALS register were included. ALSFRS-R subscores were defined for bulbar, motor and respiratory domains. Longitudinal models were used to visualise fitted total ALSFRS-R and ALSFRS-R subscore progression. In addition, the prognostic value of convenience and computed ALSFRS-R slope and subscore slopes were compared.

Results 407 incident cases were identified with a complete ALSFRS-R measure. 233 (57%) patients were male, and 125 (31%) had bulbar-onset disease. ALSFRS-R bulbar and motor subscore slopes provided a better fit in prognostic models when combined over the total ALSFRS-R slope. Longitudinal analysis revealed that the ALSFRS-R motor subscore deteriorated earlier in spinal-onset disease over bulbar-onset disease, while in bulbar-onset disease the ALSFRS-R bulbar subscore deteriorated earlier and faster than in spinal-onset disease.

Discussion Our analysis builds on previous knowledge of ALSFRS-R subscores. Decline in ALSFRS-R motor subscores in patients with spinal-onset disease, and decline in ALSFRS-R bulbar subscores in patients with bulbar-onset disease, may predate reported disease onset dates. Respiratory subscores were not prognostically informative after adjustment for bulbar and motor subscores. These results provide robust evidence that the ALSFRS-R should not be reported as a single combined score, but rather as domain specific subscores.

View Full Text

Statistics from


Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder with a typically poor prognosis, although progression is recognised to be highly variable.1 In the absence of an established biomarker, the rate of decline on the ALS functional rating scale (revised) (ALSFRS-R) has become the most commonly used measure of progression by clinicians in research and in clinical trials. The ALSFRS was designed as a 10-item functional score across fine and gross motor, bulbar and respiratory functional domains.2 ,3 This was later extended to the ALSFRS-R by including extra respiratory scores to better capture the respiratory functional domain.4 The longitudinal analysis of change in ALSFRS-R, which is generally considered to be linear for the majority of the illness, is complicated both by heterogeneity5 and by longitudinally informative censoring due to dropout of more ill patients, and/or the mortality of patients over time.6 This problem has led to the development of alternative measures, such as the Combined Assessment of Function and Survival, a non-parametric ranked score combining functional and survival outcomes for use in clinical trials in ALS.7

The rate of change of the ALSFRS-R, or ‘ALSFRS-R slope’, has gained widespread acceptance as a prognostic indicator.8–10 However, a recent Rasch analysis of the ALSFRS-R has identified multidimensionality within the score and suggests that the use of subscores corresponding to bulbar, motor and respiratory domains may be superior to the use of a single combined score.11 Until now, the prognostic value of these subscores has not been evaluated.

Using data from the Irish ALS register, our aims were to explore the prognostic and longitudinal characteristics of ALSFRS-R functional subscores, defined by the Rasch analysis of Franchignoni et al.11 ,12


Case ascertainment

We identified all incident cases from the Irish population-based register13 ,14 for whom at least one ALSFRS-R measurement was available. A small number (n=8) of respiratory onset cases were excluded, as this group was too small for analysis. We then collated demographic, clinical, survival information and longitudinal ALSFRS-R measurements. Information included: gender, age and site of onset, diagnostic delay, revised El-Escorial category, attendance at the multidisciplinary team (MDT) clinic and Riluzole use. ALSFRS-R scores were generated by physiotherapy and medical assessors, all of whom had undergone standardised training using online tools developed by the NEALS and ENCALS groups for clinical trial purposes.

Definition of ALSFRS-R subscores

The Rasch analysis carried out by Franchignoni et al11 determined that the ALSFRS-R scale captured information corresponding to three latent domains corresponding to bulbar, motor and respiratory functions. The domains were defined as follows: Bulbar score=the sum of ALSFRS-R questions 1–3 (maximum score of 12), motor score=the sum of ALSFRS-R questions 4–9 (maximum score of 24) and finally respiratory score=the sum of ALSFRS-R questions 10–12 (maximum score of 12). We shall refer to these domain scores as ALSFRS-R_bulb, ALSFRS-R_motor and ALSFRS-R_resp, respectively, through the remainder of this manuscript. (Franchignoni et al further suggested reducing ALSFRS-R responses from five levels (score 0–4) to three levels (score 0–2). The supplementary information contains our analysis duplicated using these collapsed scores).

supplementary data

Calculation of ALSFRS-R slopes

Slopes were calculated using the common convenience method for calculating ALSFRS-R slope=(48—ALSFRS-R/time from onset) based on the first ALSFRS-R for each person, as this methodology has been reported in prognostic models8 ,9 and can be calculated from a single score. This method was extended to produce convenience estimates for ALSFRS-R_bulb, ALSFRS-R_motor and ALSFRS-R_resp slopes. These were calculated as ALSFRS-R_bulb slope=(12—Bulbar-score/time from onset), ALSFRS-R_motor slope=(24—motor-score/time from onset), and ALSFRS-R_resp slope=(12—respiratory-score/time from onset), based on the first ALSFRS-R for each person.

Survival model

For the survival analysis, time of entry to follow-up was date of onset. Using Cox regression, we constructed a survival model including the recognised prognostic variables of age at onset, site of onset, diagnostic delay and the revised El-Escorial criteria. Using this survival model as a comparison point, we first added the convenience estimate for ALSFRS-R_slope to this model (ie, based only on the first ALSFRS-R of each case and assuming a total score of 48 prior to illness onset). Next, we replaced ALSFRS-R slope by the convenience estimate of slope for each ALSFRS-R subscore individually. Finally subscores were added to the model in combination. Model fit was assessed using the Akaike Information Criterion.

Longitudinal model

A linear mixed effects multilevel model was used to account for repeated measures of ALSFRS-R within individuals. Models were estimated using maximum likelihood and compared using analysis of variance. The final model for total ALSFRS-R as outcome was then altered to consider ALSFRS-R subscores as the outcome variable; that is, a separate model was fit for the motor, bulbar and respiratory subscores. Graphs of model fit were plotted with separate curves for spinal and bulbar onset cases.


All data formatting and statistical analysis were carried out using R V.3.2.3,15 with additional packages (tableone, ggplot2, nlme, survival, rms, caret, classInt, gridExtra) (Kazuki Yoshida and Justin Bohn. tableone: Create “Table 1” to Describe Baseline Characteristics. R package version 0.7.3; Pinheiro J, Bates D, DebRoy S, Sarkar D and R Core Team. _nlme: Linear and Nonlinear Mixed Effects Models_. R package version 3.1-123; Therneau T (2015). _A Package for Survival Analysis in S_. version 2.38; Max Kuhn. caret. R package version 6.0-71; Baptiste Auguie. gridExtra: Miscellaneous Functions for “Grid” Graphics. R package version 2.2.1; Frank E Harrell Jr. rms: Regression Modeling Strategies. R package version 4.4-1. 2015; Bivand R. classInt: Choose Univariate Class Intervals. R package version 0.1-23. 2015).16–21


A total of 407 incident cases were identified with at least one complete ALSFRS-R 233 (57%) were male, 125 (30.7%) had bulbar-onset disease. The data included 1550 ALSFRS-R scores, with a median of 2 per patient (IQR: 1–5). Baseline demographics are shown in table 1.

Table 1

Baseline demographics

Survival analysis

Ninety-four per cent of the patients attended the MDT and 86% were prescribed riluzole. 20 cases with missing values were omitted from the survival analysis. Median follow-up of survival was 3.0 years (CI 2.8 to 3.3 years) with 294 (73%) dying during the follow-up period. HRs from the basic multivariate survival model, as well as models for ALSFRS-R slope and subscore slopes (all estimated via the convenience method) added to the basic model in isolation and in combination, are shown in online supplementary table S1. The best-fit model, model 6, included both ALSFRS-R_motor (HR 4.7 CI 3.7 to 6.0) and ALSFRS-R_bulb (HR 9.9, CI 5.6 to 17.6) ALSFRS-R slopes, but not the ALSFRS-R_resp slope. (Using collapsed scores (see online supplementary information table S2), the best fit model included the collapsed ALSFRS-R_resp slope).

Notably, in model 6, the site of onset lost its importance (bulbar onset HR 1.0; CI 0.74 to 1.45), indicating that the subscore slopes most likely capture the same variation in survival as captured by the site of onset. To confirm this, we tested the ability of ALSFRS-R subscore slopes to distinguish between spinal and bulbar onset cases. The cases were split into a training set of 204 cases and a test set of 203 cases, and a logistic model was generated with bulbar onset as the dependent variable. Convenience estimates of ALSFRS-R_bulb slope and ALSFRS-R_motor slope were entered as the independent variables. This model was then applied to the test set to estimate probabilities of bulbar-onset disease. Probabilities over 50% were categorised as bulbar, below 50% as spinal. The model correctly identified 47 of 61 bulbar cases in the test set with 133 of 142 correctly identified as spinal, leaving 9 spinal cases incorrectly labelled as bulbar and 14 bulbar cases incorrectly labelled as spinal. Thus, the model has a sensitivity of 90% and specificity of 84% for identifying bulbar onset cases.

Longitudinal analysis

All 407 cases and 1550 ALSFRS-R measurements were included in the longitudinal analysis.

Figure 1 displays the longitudinal linear trend in ALSFRS-R for patients with spinal and bulbar onset modelled by the linear mixed effects model. After inclusion of spline terms, further model comparisons indicated the inclusion of site of onset as a fixed effect modifier of total ALSFRS-R over time (p<0.0001). Using this final model, graphs of total fitted models of ALSFRS-R and ALSFRS-R subscores over time were then produced (figure 2). Total ALSFRS-R shows a crossover of spinal and bulbar scores. However, ALSFRS-R_motor and, in particular, ALSFRS-R_bulb show greater deviance over time. For ALSFRS-R_motor, patients with spinal and bulbar onset symptomatically decline at a similar rate in the first 4 years before converging; however, figure 2 indicates that this symptomatic decline may have occurred prior to the reported disease onset time in patients with spinal onset. Patients with bulbar onset deteriorate at a markedly increased rate, with patients with spinal onset rarely declining below 8/12 on the ALSFRS-R_bulb score. This also suggests that decline of the ALSFRS-R_bulb score may begin before reported onset in patients with bulbar onset. Conversely, ALSFRS-R_resp was relatively non-informative, showing only a mild separation between patients with bulbar and spinal onset over time and perhaps the most revealing feature is that respiratory subscores rarely drop lower than 8/12. (Longitudinal models of collapsed scores (see online supplementary information figure S1) did not exhibit longitudinal trends different from the longitudinal models of full scores).

Figure 1

Linear trend in total ALSFRS-R over time as modelled via linear mixed effects model. Although the fitted lines are in close agreement throughout with overlapping CIs, it is noteworthy that the patients with bulbar onset show a steeper slope of decline in overall ALSFRS score. ALSFRS-R, ALS functional rating scale (revised).

Figure 2

Graph of longitudinal total ALSFRS-R and ALSFRS-R subscores. Linear mixed models for total ALSFRS-R and ALSFRS-R subscores in the Irish cohort when fitted with a smoothing spline. ALSFRS-R, ALS functional rating scale (revised).


The ALSFRS-R slope is well established as a prognostic indicator in ALS.8–10 However, by compressing multidimensional domains into one combined score, information may be lost. ALSFRS-R subscores have previously been used to quantify regionality of disease, which was seen to be a prognostic factor.22 Furthermore, it has been suggested that cognitive impairment, particularly executive dysfunction at onset, predicts a bulbar subscore decline.23 Our analysis complements and expands on these previous ALSFRS-R subscore analyses.

Our results indicate that the use of ALSFRS-R_bulb and ALSFRS-R_motor slopes may offer improved prognostication. We have shown that ALSFRS-R_bulb slope and ALSFRS-R_motor slope can be used to confirm site of onset with very good sensitivity and specificity despite a very simple model and a moderate sample size, and observed that site of onset became unimportant in prognostic models when subscore slopes were included. We have further shown that longitudinal analysis of ALSFRS-R_bulb andALSFRS-R_motor revealed distinct characteristics of spinal versus bulbar onset cases. Taken together, these results indicate that ALSFRS-R_bulb and ALSFRS-R_motor capture the divergent clinical and prognostic characteristics of patients with spinal and bulbar onset. The observed longitudinal trend in subscores may also explain observations that increased regionality (estimated from ALSFRS-R subscores) of disease was prognostic,22 since typically the bulbar score in patients with bulbar declines very rapidly (figure 2).

In assessing ALSFRS-R_resp, we must recognise that the cohort did not, by design, include any patients with respiratory onset and that the use of non-invasive ventilation (NIV) was not taken into account. Although the leading cause of death in patients with ALS is respiratory failure,24 our finding that ALSFRS-R_resp slope was not prognostic in the final survival model, and the longitudinal characteristics of ALSFRS-R_resp shown in figure 2, most likely indicate that the ALSFRS-R_resp questions are not sensitive to the respiratory burden of the majority of patients, further reinforcing the findings of Franchignoni et al.11 The respiratory questions of the ALSFRS-R focus on chronic progressive symptoms (dyspnoea, orthopnoea and ventilation), and it may be that the absence of a question covering acute exacerbations or frequency of chest infections and questions relating to compliance in usage of NIV account for the absence of sensitivity. However, in our analysis of collapsed ALSFRS-R subscores, the slopes based on the collapsed respiratory scores suggested by Franchignoni et al did show prognostic value (see online supplementary information).

Our observation that the decline in ALSFRS-R_bulb in patients with bulbar onset, and decline in ALSFRS-R_motor in patients with spinal onset, may begin before reported symptom onset times contrasts with a recent analysis by Proudfoot et al,6 who reported a median lag of 5.8 months between reported symptom onset and the beginning of decline in ALSFRS from the maximum score of 48 in the PROACT data set. Proudfoot et al considered that a lack of sensitivity of the ALSFRS-R in early stages of the diseases may offer an explanation; however, our results here would suggest the opposite. Since PROACT constitutes data only from clinical trials with very specific entry criteria for patients, selection bias in favour of prevalent cohorts of patients may explain the discrepancies, as our cohort was based on mixed incident cases of MDT attenders and non-attenders. Alternatively, the different mathematical models used could be responsible. Nevertheless, our results imply that unrecognised minor symptomology amounting to a small ALSFRS-R decrease may be present before the self-reported time of onset of symptoms. We speculate that this may reflect a tendency of patients to dismiss or ignore minor initial symptoms that would otherwise have represented the true onset of disease. This is also consistent with an insidious onset of disease that would be characteristic of a long preclinical prodromal phase of the disease,25 and we note that long prodromal phases have been observed in SOD1 mice25 and human SOD1 case reports.26

Although our study is limited by relatively small numbers for the complexity of the models used, our findings shed new light on the evolution of ALSFRS-R subscores over time. We have shown than ALSFRS-R_bulb slope and ALSFRS-R_motor slope together provided a better fit in survival modelling than ALSFRS-R slope as a single variable, and should be considered separately in future analyses of clinical decline. Longitudinal analysis showed that ALSFRS-R_bulb and ALSFRS-R_motor show distinct trajectories for patients with bulbar and spinal onset. Convenience estimates of ALSFRS-R_bulb slope and ALSFRS-R_motor slope were important prognostic variables, but can also confirm site of onset with high sensitivity and specificity. Our results offer improved prognostication and may allow for more detailed analysis of progression in epidemiological studies and clinical trials, and provide strong evidence that ALSFRS-R should be not be reported as a single score, but rather analysed as domain specific subscores.


View Abstract


  • Contributors JR designed the study, analysed the data and wrote the manuscript. TB assisted with study design, collected data and edited the manuscript. AV collected data, managed the register database and edited the manuscript. MH collected data, managed the register database and edited the manuscript. OH set up and led the Irish ALS register, collected data, contributed to the study design and analysis and edited the manuscript.

  • Funding The research leading to these results has received funding from the Health Research Board Interdisciplinary Capacity Enhancement Programme, the European Community's Seventh Framework Programme (FP7/2007-2013) under the Health Cooperation programme and the project EUROMOTOR (number 259867), from the European Joint Progamme in Neurodegeneration (SOPHIA and ALS-CarE), from the Irish Institute of Clinical Neuroscience (IICN: 12549. 201616) and the Charities Research Motor Neuron and Irish Motor Neuron Disease Association. The funding sources played no role in the preparation of this manuscript.

  • Competing interests JR was funded by the Health Research Board Clinical Fellowship Programme (HPF-2014-527). TB was funded on a Clinical Research Fellowship from the Irish Institute of Clinical Neuroscience. OH is funded by the Health Research Board Clinician Scientist Programme. OH has received speaking honoraria from Novarits, Biogen Idec, Sanofi Aventis and Merck-Serono. She has been a member of advisory panels for Biogen Idec, Allergen, Ono Pharmaceuticals, Novartis, Cytokinetics and Sanofi Aventis. She serves as Editor-in-Chief of Amyotrophic Lateral Sclerosis and Frontotemporal Dementia.

  • Ethics approval The Irish ALS Register complies with Irish Data protection legislation (1988 and 2003), and has been approved by the Beaumont Hospital Ethics Committee (02/28 and 05/49).

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Data sharing requests are made in writing through Professor Hardiman ( and require a formal data sharing agreement with approval from the University Technology Transfer Department. Data sharing agreements must include details on how the data will be stored, who will have access to the data and intended use of the data, and agreements as to the allocation of intellectual property.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles