Advertisement

Mapping ALSFRS-R and ALSUI to EQ-5D in Patients with Motor Neuron Disease

Open ArchivePublished:July 25, 2018DOI:https://doi.org/10.1016/j.jval.2018.05.005

      Abstract

      Background

      The Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised (ALSFRS-R) is the preferred measure of health outcome in clinical trials in motor neuron disease (MND). It, however, does not provide a preference-based health utility score required for estimating quality-adjusted life-years in economic evaluations for health technology assessments.

      Objectives

      To develop algorithms for mapping from measures used in MND clinical studies to allow for future prediction of the five-level EuroQol five-dimensional questionnaire (EQ-5D-5L) utility in populations of patients with MND when utility data have not been collected.

      Methods

      Direct mapping models were developed using ordinary least squares and Tobit regression analyses to estimate EQ-5D-5L utilities (based on English tariffs), with ALSFRS-R total, domain, and item scores used as explanatory variables, using patient-level data from a UK cohort study. Indirect mapping models were also used to map EQ-5D-5L domains, using the same variables, along with the Neuropathic Pain Scale and the Hospital and Anxiety Depression Scale for MND using multinomial logistic regression analysis. Goodness of fit was assessed along with predicted values for each mapping model.

      Results

      The best-performing model predicting EQ-5D-5L utilities used five items of the ALSFRS-R items as explanatory variables in a stepwise ordinary least squares regression. The mean squared error was 0.0228, and the mean absolute error was 0.1173. Prediction was good, with 55.4% of estimated values within 0.1 and 91.4% within 0.25 of the observed EQ-5D-5L utility value. Indirect mapping using the Neuropathic Pain Scale and the Hospital and Anxiety Depression Scale for MND provided less predictive power than direct mapping models.

      Conclusions

      This is the first study to present mapping algorithms to crosswalk between ALSFRS-R and EQ-5D-5L. This analysis demonstrates that the ALSFRS-R can be used to estimate EQ-5D-5L utilities when they have not been collected directly within a trial.

      Keywords

      Introduction

      Motor neuron disease (MND) (also known as amyotrophic lateral sclerosis [ALS]) is a progressively degenerative neurological condition that affects motor neurons in the brain and spinal cord. Life expectancy is between 3 and 5 years from symptom onset [
      • Van Es. M.A.
      • Hardiman O.
      • Chio A.
      • et al.
      Amyotrophic lateral sclerosis.
      ] and quality of life (QOL) is greatly impaired. Established treatments are symptom management, riluzole (which increases median survival by about 3 months) [
      • Miller R.G.
      • Mitchell J.D.
      • Moore D.H.
      Riluzole for amyotrophic lateral sclerosis (ALS)/motor neuron disease (MND).
      ], and palliative care [
      • Miller R.G.
      • Jackson C.E.
      • Kasarskis E.J.
      • et al.
      Practice parameter update: the care of the patient with amyotrophic lateral sclerosis: multidisciplinary care, symptom management, and cognitive/behavioural impairment (an evidence-based review).
      ].
      The recent approval of edaravone [
      Writing Group; Edaravone (MCI-186) ALS 19 Study Group
      Safety and efficacy of edaravone in well-defined patients with amyotrophic lateral sclerosis: a randomised, double-blind, placebo-controlled trial.
      ] by the US Food and Drug Administration and the potential for other new treatment options [

      Motor Neurone Disease Association. MND treatment trials. Available from: http://www.mndassociation.org/research/mnd-research-and-you/treatment-trials/. [Accessed October 2, 2016].

      ,
      • Goutman S.A.
      • Chen K.S.
      • Feldman E.L.
      Recent advances and the future of stem cell therapies in amyotrophic lateral sclerosis.
      ] will increase the need for evidence to support health technology assessment (HTA) and reimbursement decisions. At present, there is limited literature on preference-based health utilities in patients with MND [
      • Moore A.
      • Young C.A.
      • Hughes D.A.
      Economic studies in motor neurone disease: a systematic methodological review.
      ], which are required for the calculation of quality-adjusted life-years for cost-utility analyses.
      The EuroQol five-dimensional questionnaire (EQ-5D) is the preferred measure of the National Institute for Health and Care Excellence [
      • Longworth L.
      • Rowen D.
      Mapping to obtain EQ-5D utility values for use in NICE health technology assessments.
      ] for calculating quality-adjusted life-years and the most widely used generic preference-based health outcomes measure, facilitating comparisons of health technologies between different diseases [
      • Brazier J.
      • Ara R.
      • Rowen D.
      • et al.
      A review of generic preference-based measures for use in cost-effectiveness models.
      ]. Nevertheless, concerns have been expressed in applying this measure to patients with MND, because it does not account for a range of symptoms, including communication, fatigue, swallowing, and respiratory difficulty [
      • Van Es. M.A.
      • Hardiman O.
      • Chio A.
      • et al.
      Amyotrophic lateral sclerosis.
      ]. Experience of the three-level EQ-5D in patients with MND is that the measure can be used but with cautions of ceiling/floor effects, among other issues [
      • Jones A.
      • Jivraj N.
      • Balendra R.
      • et al.
      Health utility decreases with increasing clinical stage in amyotrophic lateral sclerosis.
      ,
      • Kiebert G.M.
      • Green C.
      • Murphy C.
      • et al.
      Patients’ health-related quality of life and utilities associated with different stages of amyotrophic lateral sclerosis.
      ].
      When EQ-5D data are not available, the National Institute for Health and Care Excellence allows for utilities to be estimated by mapping from other health-related QOL measures [

      National Institute for Health and Care Excellence. Process and methods: guide to the methods of technology appraisal. 2013. Available from: https://www.nice.org.uk/process/pmg9. [Accessed March 24, 2017].

      ]. A number of studies concerned with mapping disease-specific QOL instruments to the EQ-5D have been published [
      • Longworth L.
      • Rowen D.
      Mapping to obtain EQ-5D utility values for use in NICE health technology assessments.
      ,
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ] and guidelines have been produced for best practice [
      • Petrou S.
      • Rivero-Arias O.
      • Dakin H.
      • et al.
      The MAPS reporting statement for studies mapping onto generic preference-based outcome measures: explanation and elaboration.
      ,
      • Wailoo A.J.
      • Hernandez-Alava M.
      • Manca A.
      • et al.
      Mapping to estimate health-state utility from non–preference-based outcome measures: an ISPOR Good Practices for Outcomes Research Task Force Report.
      ]. Mapping from a non–preference-based measure to the EQ-5D can be performed by predicting either the EQ-5D health utility values (direct mapping) or each of the five domain responses (indirect mapping). There is, however, limited use of either approach in the context of neurological conditions [
      • Dams J.
      • Klotsche J.
      • Bornschein B.
      • et al.
      Mapping the EQ-5D index by UPDRS and PDQ-8 in patients with Parkinson’s disease.
      ,
      • Sidovar M.F.
      • Limone B.L.
      • Lee S.
      • et al.
      Mapping the 12-item multiple sclerosis walking scale to the EuroQol 5-dimension index measure in North American multiple sclerosis patients.
      ].
      The Amyotrophic Lateral Sclerosis Functioning Rating Scale-Revised (ALSFRS-R) [
      • Cedarbaum J.M.
      • Stambler N.
      • Malta E.
      • et al.
      The ALSFRS-R: a revised ALS functional rating scale that incorporates assessments of respiratory function. BDNF ALS Study Group (Phase III).
      ] is recommended for use in clinical trials of treatments for MND [
      • Leigh P.N.
      • Swash M.
      • Iwasaki Y.
      • et al.
      Amyotrophic lateral sclerosis: a consensus viewpoint on designing and implementing a clinical trial.
      ] to capture clinical changes in motor, bulbar, and respiratory function. Although this is not a preference-based measure, the ALS Utility Index (ALSUI), which is derived from five items of the ALSFRS-R and based on US general population tariff scores, does allow for utilities to be estimated [
      • Beusterien K.
      • Leigh N.
      • Jackson C.
      • et al.
      Integrating preferences into health status assessment for amyotrophic lateral sclerosis: the ALS Utility Index.
      ], but has not been used in patients with MND.
      The aim of our study was to develop algorithms for mapping, both directly and indirectly, from measures used in MND clinical studies to allow for future prediction of five-level EQ-5D (EQ-5D-5L) utility in populations of patients with MND when utility data have not been collected.

      Methods

      Data

      Data were sourced from the ongoing Trajectories of Outcomes in Neurological Conditions (TONiC) study [

      Trajectories of Outcomes in Neurological Conditions. Motor Neurone Disease. Available from: https://tonic.thewaltoncentre.nhs.uk/ [Accessed March 24, 2017].

      ]. This longitudinal study of QOL and economic outcomes includes a large cohort of patients with MND recruited throughout the United Kingdom. Participants complete a series of outcome measures and provide demographic and clinical information.
      For the analysis, we used baseline responses from a cross section of patients recruited by MND clinical and research teams up to January 2017, who were at different stages of the disease course. Cross-sectional rather than longitudinal data were used because only 106 from 636 patients had returned any follow-up questionnaires at the time of analysis for this article. All questionnaires used in the mapping analysis were returned in a single pack, which the participant was requested to complete on the same day if possible. Clinicians allocated MND to limb, bulbar, or respiratory onset types and performed disability assessment using the ALSFRS-R.
      Ethical approval was granted from National Research Ethics Service Committee North West—Greater Manchester West (reference no. 11/NW/0743).

      Missing Data

      Mapping was conducted only for those participants for whom complete data were available. A logistic regression was used to test whether participants who had returned incomplete questionnaires were comparable with those who had fully completed questionnaires in terms of their age, sex, MND onset type, independent completion of questionnaires, and recruiting center.

      Measures

      The EQ-5D-5L was included in the TONiC study to estimate health utilities. It covers the health domains of mobility, self-care, usual activities, pain/discomfort, and anxiety/depression, each with five levels of severity [
      • Devlin N.J.
      • Shah K.K.
      • Feng Y.
      • et al.
      Valuing health-related quality of life: an EQ-5D-5L value set for England.
      ]. A preference-based single index score can be generated with any combination of responses, anchored at 0 to represent death and 1 to represent full health, and, on the basis of an English tariff, includes the worst health state of −0.281. These health utility values have been developed using general public responses to a standard gamble survey.
      Three measures were selected from the TONiC data set for the purposes of mapping to the EQ-5D-5L: 1) the ALSFRS-R (from which the ALSUI was derived), 2) the Neuropathic Pain Scale (NPS), and 3) the Hospital and Anxiety Depression Scale for MND (MND-HADS).

      Amyotrophic Lateral Sclerosis Functioning Rating Scale-Revised

      The revised version of the ALSFRS incorporates respiratory items, increasing the sensitivity of the instrument to changes in the disease course of MND [
      • Cedarbaum J.M.
      • Stambler N.
      • Malta E.
      • et al.
      The ALSFRS-R: a revised ALS functional rating scale that incorporates assessments of respiratory function. BDNF ALS Study Group (Phase III).
      ]. The ALSFRS-R is a validated MND-specific 12-item questionnaire concerning bulbar, limb, and respiratory function. Responses range from a score of 0 (severe problems) to 4 (no change). Responses to the ALSFRS-R are often used to derive a single index value and this value is reported in many clinical studies, but recent evidence suggests that the ALSFRS-R should be examined on a domain level to generate either three or four domain scores to overcome concerns of unidimensionality [
      • Franchignoni F.
      • Mora G.
      • Giordano A.
      • et al.
      Evidence of multidimensionality in the ALSFRS-R Scale: a critical appraisal on its measurement properties using Rasch analysis.
      ] (Fig. 1).
      Fig. 1
      Fig. 1Structure of ALSFRS-R, showing breakdown by four and three domains and items. Bulbar items are related to speech and communication. The fine motor domain is concerned with actions such as hand and finger movements, whereas gross motor domain captures changes in areas such as walking and climbing. The respiratory domain captures issues around the ease of breathing. ALSFRS-R, Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised.
      The ALSUI is derived from the following ALSFRS-R domains: speech and swallowing, eating and self-care, leg function, and respiratory function [
      • Beusterien K.
      • Leigh N.
      • Jackson C.
      • et al.
      Integrating preferences into health status assessment for amyotrophic lateral sclerosis: the ALS Utility Index.
      ]. Preference weights were generated from members of the general public in the United States using the standard gamble method and can be used to calculate a single preference-based utility score for persons with MND.

      Neuropathic Pain Scale

      The NPS [
      • Bradley S.
      • Galer M.D.
      • Jensen M.P.
      Development and preliminary validation of a pain measure specific to neuropathic pain: the Neuropathic Pain Scale.
      ] measures the intensity, unpleasantness, and sharpness of neuropathic pain. The questionnaire consists of 10 scales with varying descriptions of pain, each with a possible response value between 0 (no pain) and 10 (worst pain imaginable). A further item concerns the length of time the patient has experienced pain, with a score of between 0 and 2. Responses to the scales and the time item are summed to provide an NPS index score.

      Hospital and Anxiety Depression Scale for MND

      The MND-HADS [
      • Gibbons C.
      • Mills R.J.
      • Thornton E.W.
      • et al.
      Rasch analysis of the Hospital Anxiety and Depression Scale (HADS) for use in motor neurone disease.
      ] is a modified version of the Hospital and Anxiety Depression Scale (HADS) [
      • Zigmond A.S.
      • Snaith R.P.
      The Hospital Anxiety and Depression Scale.
      ], developed for use in MND populations to address concerns that items in the original HADS may be confounded by physical disability. The modified HADS-Anxiety and HADS-Depression, which have acceptable psychometric properties, resulted from the removal of one item from both seven-item scales.

      Statistical Methods

      With our aim of developing a crosswalk between the selected measures available in the TONiC study and the EQ-5D-5L, we tested various model types and structures to arrive at a preferred model, and present alternative acceptable models that may suit different scenarios depending on data availability. Models based on direct mapping to EQ-5D-5L utilities (based on the English tariff [
      • Devlin N.J.
      • Shah K.K.
      • Feng Y.
      • et al.
      Valuing health-related quality of life: an EQ-5D-5L value set for England.
      ]) and indirect mapping to EQ-5D-5L domains were tested. We randomly divided our data set into estimation and validation samples in a 2:1 ratio, allowing algorithms generated in the estimation sample to predict values in the validation sample.
      For the direct mapping analysis, we considered the ALSFRS-R by individual items, three and four domain variables, and index score (Table 1; Fig. 1). Individual item responses to the ALSFRS-R provide the greatest granularity, domain variables of the ALSFRS-R offer more concise information on distinctive features of MND [
      • Franchignoni F.
      • Mora G.
      • Giordano A.
      • et al.
      Evidence of multidimensionality in the ALSFRS-R Scale: a critical appraisal on its measurement properties using Rasch analysis.
      ], and the index score was selected on the basis of it being reported in many clinical studies in MND. The ALSUI was analyzed by index score because this measure is preference-based and therefore the index value combined weighted domain responses.
      Table 1Mapping models used in statistical analysis
      Model numberExplanatory variablesStatistical methods
      Direct mapping
      1aALSFRS-R indexOLS and Tobit
      1bALSFRS-R index and demographicsOLS and Tobit
      2ALSFRS-R four domainsOLS and Tobit
      3ALSFRS-R three domainsOLS and Tobit
      4ALSUIOLS and Tobit
      5ALSFRS-R itemsOLS and Tobit
      6Stepwise ALSFRS-R itemsOLS and Tobit
      Indirect mapping
      7ALSFRS-R indexMultinomial logistic
      8ALSFRS-R four domainsMultinomial logistic
      9ALSFRS-R three domainsMultinomial logistic
      10ALSUIMultinomial logistic
      11ALSFRS-R itemsMultinomial logistic
      12Stepwise ALSFRS-R itemsMultinomial logistic
      13ALSFRS-R index score, NPS, and MND-HADSMultinomial logistic
      14ALSFRS-R four domains, NPS, and MND-HADSMultinomial logistic
      15ALSFRS-R three domains, NPS, and MND-HADSMultinomial logistic
      16ALSUI score, NPS, and MND-HADSMultinomial logistic
      17ALSFRS-R items, NPS, and MND-HADSMultinomial logistic
      18ALSFRS-R items stepwise selection, NPS, and MND-HADSMultinomial logistic
      ALS, amyotrophic lateral sclerosis; ALSFRS-R, Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised; ALSUI, ALS Utility Index; MND, motor neuron disease; MND-HADS, Hospital and Anxiety Depression Scale for MND; NPS, Neuropathic Pain Scale; OLS, ordinary least squares.
      Two model types were chosen for the direct mapping. First, we used ordinary least squares (OLS) regression, which has been used extensively in comparable studies with acceptable performance [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ]. Given that EQ-5D-5L utility data are skewed, however, violating the assumption of normality, and are censored at the upper limit of 1, we also used a Tobit regression model [
      • McDonald J.F.
      • Moffitt A.
      The uses of Tobit analysis.
      ] and compared the results with OLS regression models.
      For all indirect mapping analyses, we used multinomial logistic regression to account for the categorical nature of EQ-5D domains and the ordering of EQ-5D domain levels (Table 1). Initially, we used the same combinations of explanatory variables as in our direct mapping analysis. We then undertook a second indirect mapping analysis, which included the additional measures of the NPS and the MND-HADS. These were included to overcome the lack of pain and mental health domains within the ALSFRS-R, therefore aiding our indirect mapping analysis. All models, direct and indirect, were run with and without the demographic variables of age, sex, and MND onset type. All regression analyses were performed on the estimation sample, with generated results used to predict values using the validation sample. Furthermore, a stepwise selection was used to examine whether a reduced ALSFRS-R item model was more appropriate, with regard to removing variables whose coefficients were not rationally directed, and to test whether a more efficient model could be obtained.
      Data management was carried out using Microsoft Excel (Microsoft, Washington, DC), and R statistical software version 3.0 (R Foundation for Statistical Computing, Vienna, Austria) [
      • Core Team R.
      R: A Language and Environment for Statistical Computing.
      ] was used for statistical analysis.

      Assessing Model Performance

      Model performance was examined by the mean squared errors (MSEs) and mean absolute errors (MAEs), in line with mapping guidance [
      • Longworth L.
      • Rowen D.
      Mapping to obtain EQ-5D utility values for use in NICE health technology assessments.
      ,
      • Petrou S.
      • Rivero-Arias O.
      • Dakin H.
      • et al.
      The MAPS reporting statement for studies mapping onto generic preference-based outcome measures: explanation and elaboration.
      ], to identify the best predictive models. For optimal model selection, we used MSE results from our validation sample. The MAE was included to complement the MSE analysis and ensure that models selected on the basis of a lower MSE score also had a lower MAE score.
      Tests of systematic bias in selected models, chosen by lowest MSE score, were performed by examining the percentage of predicted values that deviated from observed values by more than 0.10 and 0.25. To identify whether the selected models performed better for particular ranges of utility values, we also present the errors for the following categories of EQ-5D-5L utility scores: less than 0, 0 to less than 0.2, 0.2 to less than 0.4, 0.4 to less than 0.6, 0.6 to less than 0.8, and 0.8 to 1. The plotting of histograms of the residuals of observed and predicted values of the selected model provided visual evidence of the nature of errors present in the models. Examination of mean differences in utility values between data sets was also undertaken. Finally, the Akaike information criterion (AIC) [
      • Vrieze S.
      Model selection and psychological theory: a discussion of the differences between the Akaike information criterion (AIC) and the Bayesian information criterion (BIC).
      ] was used to test the fit of models with the lowest MSE for each of the explanatory variable groups in the direct mapping and also for all indirect mapping models.
      The conduct and reporting followed guidance from the MApping onto Preference-based measures reporting Standards (MAPS) statement [
      • Petrou S.
      • Rivero-Arias O.
      • Dakin H.
      • et al.
      The MAPS reporting statement for studies mapping onto generic preference-based outcome measures: explanation and elaboration.
      ].

      Results

      Data Characteristics

      Questionnaires were posted to 958 patients. A response rate of 66.4% for our cross-sectional data set was achieved, resulting in 636 returned questionnaires. Forty-one questionnaires were incomplete for direct mapping, leaving a total of 595 completed patient questionnaires for inclusion in this analysis. Respondents who did not fully complete the questionnaires were not statistically different from those who returned completed questionnaires, with respect to the variables tested (see the Appendix in Supplemental Materials found at 10.1016/j.jval.2018.05.005). For direct mapping, 397 patients were randomly assigned to the estimation sample and 198 to the validation sample. For indirect mapping, 18 patients had not completed the required additional questionnaires, and therefore 385 patients were in the estimation sample and 192 in the validation sample. Estimation and validation samples were well balanced in terms of age, sex split, MND onset type (bulbar, limb, and respiratory), severity of EQ-5D domain responses, their EQ-5D-5L and ALSUI utility values, and ALSFRS-R, NPS, and MND-HADS scores (Table 2, Table 3). The mean age of respondents was 65.1 years, which is in line with reported average ages of patients with MND, and the sex split of 61% male is also reflected within the literature [
      • Chiò A.
      • Logroscino G.
      • Traynor B.J.
      • et al.
      Global epidemiology of amyotrophic lateral sclerosis: a systematic review of the published literature.
      ].
      Table 2Patient characteristics
      CharacteristicWhole sample (n = 595)Estimation sample (n = 397)Validation sample (n = 198)
      Male, n (%)363 (61.0)243 (61.2)120 (60.6)
      Age, mean ± SD65.07 ± 10.8965.25 ± 10.8964.70 ± 10.6
      MND onset, n (%)
       Limb404 (69.9)265 (66.8)139 (70.2)
       Bulbar159 (26.7)112 (28.2)48 (26.7)
       Respiratory11 (2.5)8 (2.0)3 (2.5)
      Measures, mean ± SD
       EQ-5D-5L index0.57 ± 0.260.57 ± 0.260.58 ± 0.27
       EQ-5D VAS0.60 ± 21.300.61 ± 22.010.60 ± 21.78
       ALSFRS-R score31.95 ± 8.3331.85 ± 8.1332.15 ± 8.73
       ALSUI0.40 ± 0.240.40 ± 0.240.41 ± 0.24
       NPS30.02 ± 16.4028.74 ± 16.9532.62 ± 15.01
       MND-HADS8.02 ± 5.457.90 ± 5.518.25 ± 5.32
      ALS, amyotrophic lateral sclerosis; ALSFRS-R, Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised; ALSUI, ALS Utility Index; EQ-5D-5L, five-level EuroQol five-dimensional questionnaire; MND, motor neuron disease; MND-HADS, Hospital and Anxiety Depression Scale for MND; NPS, Neuropathic Pain Scale; VAS, visual analogue scale.
      Table 3Distribution of responses by EQ-5D-5L domains
      EQ-5D domainWhole sample (n = 595)Estimation sample (n = 397)Validation sample (n = 198)
      Mobility, n (%)
       Level 199 (16.6)63 (15.9)36 (18.2)
       Level 281 (13.2)54 (13.6)27 (17.6)
       Level 3157 (26.4)106 (26.4)52 (26.3)
       Level 4152 (25.6)100 (25.2)52 (26.3)
       Level 5106 (17.8)75 (18.9)31 (15.7)
      Self-care, n (%)
       Level 1118 (19.8)85 (21.4)33 (16.7)
       Level 2152 (25.6)88 (22.2)64 (32.3)
       Level 3162 (27.2)110 (27.7)52 (26.3)
       Level 471 (11.9)52 (13.1)19 (9.6)
       Level 592 (15.5)62 (15.6)30 (15.2)
      Usual activities, n (%)
       Level 153 (8.9)35 (8.8)18 (9.1)
       Level 2117 (19.7)71 (17.9)46 (23.2)
       Level 3174 (29.2)118 (29.7)56 (28.3)
       Level 4118 (19.8)85 (21.4)33 (16.7)
       Level 5115 (22.4)88 (22.2)45 (27.7)
      Pain/discomfort, n (%)
       Level 1179 (30.1)116 (29.2)63 (31.8)
       Level 2213 (33.8)140 (35.3)73 (36.9)
       Level 3161 (27.1)114 (28.7)47 (23.7)
       Level 437 (3.6)22 (5.5)15 (7.6)
       Level 55 (0.8)5 (1.3)0 (0.0)
      Anxiety/depression, n (%)
       Level 1268 (45.1)181 (45.6)87 (43.9)
       Level 2203 (34.1)131 (33.0)72 (36.4)
       Level 398 (16.5)66 (16.6)32 (16.2)
       Level 420 (3.3)15 (3.8)5 (2.3)
       Level 56 (1.0)4 (1.0)2 (1.0)
      EQ-5D-5L, five-level EuroQol five-dimensional questionnaire.
      Figure 2 shows the distributions of the EQ-5D-5L utilities, ALSFRS-R index values, and ALSUI scores in both samples. The number of individuals reporting negative EQ-5D-5L utilities in our full data set was 13 (2.2%). EQ-5D-5L utility ranged from −0.21 to 1, whereas the ranges of other measures were as follows: ALSFRS-R, 1 to 48; ALSUI, 0 to 1; NPS, 0 to 85; and MND-HADS, 0 to 28.
      Fig. 2
      Fig. 2Distributions of EQ-5D-5L utilities, ALSFRS-R index scores, and ALSUI scores by sample. ALSFRS-R, Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised; ALSUI, ALS Utility Index; EQ-5D-5L, five-level EuroQol five-dimensional questionnaire.
      The distributions of responses varied across the EQ-5D domains (Table 3), with mobility and usual activities associated with greater proportions of severe problems, compared with other domains, reflecting the impact of MND on patients’ motor functioning. There were fewer responses in the more severe categories of pain/discomfort and anxiety/depression, with 5 (0.8%) and 6 (1.0%) individuals reporting severe problems, respectively.

      Model Performance

      The results of our mapping analysis by model type are presented in Table 4. Patient demographic characteristics were significant predictors of EQ-5D-5L utilities in only model OLS (1b); results for the other models with demographic variables are therefore not presented.
      Table 4Mapping results
      ModelEstimation sample (n = 397)Validation sample (n = 198)
      Mean ± SDMinimum, maximumMSEMAEMean ± SDMinimum, maximumMSEMAE
      Observed EQ-5D-5L utility0.57 ± 0.26−0.2, 1NANA0.58 ± 0.26−0.21, 1NANA
      Direct models
      OLS (1)0.57 ± 0.190.1, 0.860.04040.15940.57 ± 0.18−0.06, 0.90.0370.1552
      OLS (1b)0.57 ± 0.190.04, 10.03390.14480.57 ± 0.19−0.06, 10.03060.1407
      OLS (2)0.57 ± 0.210.08, 0.960.02390.12020.57 ± 0.150.1, 0.960.04610.1794
      OLS (3)0.57 ± 0.200.05, 0.940.04470.12450.57 ± 0.150.08, 0.940.02810.1306
      OLS (4)0.57 ± 0.160.03, 0.920.02190.12010.57 ± 0.160.3, 0.950.04410.1731
      OLS (5)0.57 ± 0.220.09, 0.980.02240.11350.57 ± 0.220.1, 0.980.02450.1218
      OLS (6)0.57 ± 0.210.09, 0.960.02210.11120.58 ± 0.210.1, 0.970.02280.1173
      Tobit (1)0.57 ± 0.170.09, 0.870.04050.15890.59 ± 0.18−0.06, 0.910.03710.1545
      Tobit (1b)0.57 ± 0.190.05, 0.990.03560.14530.57 ± 0.20−0.01, 0.920.03100.1423
      Tobit (2)0.57 ± 0.170.07, 0.850.04210.16250.51 ± 0.150.03, 0.810.04660.1801
      Tobit (3)0.57 ± 0.210.03, 0.970.02710.12830.55 ± 0.200.01, 0.920.02800.1329
      Tobit (4)0.57 ± 0.160.3, 0.930.04470.17110.58 ± 0.160.3, 0.970.04420.1730
      Tobit (5)0.57 ± 0.220.08, 10.02190.11320.57 ± 0.220.09, 0.990.02550.1288
      Tobit (6)0.57 ± 0.210.08, 0.90.02330.11490.57 ± 0.210.09, 0.980.02500.1241
      Indirect models
      Mlogit (7)0.65 ± 0.240.17, 0.950.56600.17940.66 ± 0.230.17, 10.05970.1812
      Mlogit (8)0.66 ± 0.220.17, 10.03900.12850.67 ± 0.580.17, 10.03200.1415
      Mlogit (9)0.64 ± 0.240.17, 10.03600.13790.60 ± 0.25−0.02, 10.03030.1342
      Mlogit (10)0.61 ± 0.230.17, 10.05010.18110.62 ± 0.220.17, 0.950.05100.1732
      Mlogit (11)0.62 ± 0.210.01, 0.950.02740.11650.62 ± 0.22−0.02, 10.03150.1526
      Mlogit (12)0.61 ± 0.220.01, 0.950.02520.11400.62 ± 0.21−0.02, 10.03100.1563
      Mlogit (13)0.57 ± 0.22−0.07, 10.01990.10340.58 ± 0.22−0.02, 10.03080.1310
      Mlogit (14)0.72 ± 0.230.34, 10.09890.24210.58 ± 0.210.17, 10.05340.2181
      Mlogit (14)0.74 ± 0.220.49, 0.930.09540.23390.60 ± 0.210.34, 0.940.06630.2316
      Mlogit (15)0.59 ± 0.230.09, 10.15810.15810.59 ± 0.22−0.02, 10.04970.1757
      Mlogit (16)0.49 ± 0.22−0.09, 10.18700.18700.51 ± 0.110.51, 10.06570.1956
      Mlogit (17)0.59 ± 0.22−0.1, 10.20100.20100.59 ± 0.210.17, 10.04410.2301
      EQ-5D-5L, five-level EuroQol five-dimensional questionnaire; MAE, mean absolute error; MSE, mean squared error; NA, not applicable; OLS, ordinary least squares.

      Direct mapping

      Direct mapping models were compared in terms of their fitted values deviating by more than 0.1 and 0.25 of the true utility. This ranged from 31.3% to 55.4% for within 0.10 of true value, and from 56.3% to 91.4% for within 0.25 of true value. Direct mapping models generally performed well in estimating mean utility in the estimation sample, with all models predicting the mean correctly to two decimal places. In the validation sample, however, only three mapping models predicted the mean to two decimal places, and only three predicted negative utility values. Model OLS (5) demonstrated the lowest MSE (0.0245), MAE (0.1218), and AIC values in the validation sample; it, however, contained nonsignificant coefficients, and negative (counterintuitive) coefficients on items 1 to 4 and 12. For these reasons, among direct mapping models, the use of the reduced ALSFRS-R item model with stepwise selection of explanatory variables (model OLS (6)) is preferred. Although MSE (0.0228), MAE (0.1173), and AIC values all indicated model OLS (6) to provide the best fit of the data, the predicted errors were not uniform across the range of EQ-5D-5L utility scores (see the Appendix in Supplemental Materials found at https://doi.org/10.1016/j.jval.2018.05.005). Larger errors were apparent for negative utilities and for utilities in the range of 0 to 0.2. Figure 3 presents the fitted versus observed values, and Figure 4 plots the residuals. The model was the strongest when predicting values from 0.2 to 0.8. In total, 91.7% of estimations were within 0.25 of the observed EQ-5D-5L values, with 55.4% within 0.10 of the true value. The algorithm generated from this regression is as follows:
      EQ5D5Lutility=0.086203+0.057486×item6+0.046674×item7+0.058688×item8+0.035927×item9+0.021126×item10.


      Fig. 3
      Fig. 3Selected model OLS (6) fitted values vs. observed values, validation sample. EQ-5D-5L, five-level EuroQol five-dimensional questionnaire; OLS, ordinary least squares.
      Fig. 4
      Fig. 4Residuals of selected model OLS (6) based on the validation sample. OLS, ordinary least squares.

      Indirect mapping

      All indirect mapping models using the ALSFRS-R or ALSUI were upwardly biased because they consistently predicted higher utility values. They reported higher MSEs and MAEs than the direct mapping models using the same clinical information, but although the use of the additional measures of the NPS and MND-HADS resulted in lower errors, these models did not outperform direct mapping models.
      To researchers who may benefit from our mapping analysis, and recognizing that data availability may differ from one study to another, we present the complete results of the best-performing models for various levels of required information in the Appendix in Supplemental Materials.

      Discussion

      Our study provides evidence that the ALSFRS-R, conceptually, could be a good candidate for mapping to the EQ-5D-5L in patients with MND because the domain themes that appear in the EQ-5D (pain/discomfort and anxiety/depression), but not in the ALSFRS-R, are reported in less severe terms in patients with MND. This may partially explain why our mapping results fell within the reported MSE ranges of other mapping studies [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ] and allowed us to assert that mapping from the ALSFRS-R to the EQ-5D-5L is viable.
      The various ALSFRS-R mapping models showed markedly better predictive results than the models using the ALSUI when estimating EQ-5D-5L utilities. This may be in part due to the use of US preference tariff in the ALSUI, contrasting with our use of the English EQ-5D-5L tariff given the population from which the data were derived, but also the different selection of ALSFRS-R domains in their construct. The ALSUI estimated utility from items 1, 6, 8, 10, and 12 of the ALSFRS-R, whereas our best-fitting model, OLS (6), used items 6 to 10. More research is needed to confirm the external validity of the ALSUI and the extent to which it can be used to complement generic preference-based measures. On the basis of our mapping analysis, we cannot recommend using this measure to crosswalk to the EQ-5D-5L in patients with MND.
      As with most of the previous mapping studies, our analysis found OLS regressions to have the strongest predictive power, slightly bettering the results from the Tobit regressions for direct mapping [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ]. Indirect mapping models with the same specifications as the direct models showed higher MSEs using a multinomial logistic regression and consistently estimated larger mean EQ-5D utilities compared with observed values. The addition of the NPS and HADS to the indirect models reduced reported MSEs, but not to the extent as estimated in the direct mapping models. Demographic information did not significantly improve the predictive power of the models, with the exception of model 1b; this result has been reflected in other MND research [
      • Chiò A.
      • Gauthier A.
      • Montuschi A.
      • et al.
      A cross sectional study on determinants of quality of life in ALS.
      ].
      Our preferred model OLS (6), using a selection of ALSFRS-R items as explanatory variables, had MSE and MAE values comparable with other neurological statistical mapping work [
      • Dams J.
      • Klotsche J.
      • Bornschein B.
      • et al.
      Mapping the EQ-5D index by UPDRS and PDQ-8 in patients with Parkinson’s disease.
      ,
      • Sidovar M.F.
      • Limone B.L.
      • Lee S.
      • et al.
      Mapping the 12-item multiple sclerosis walking scale to the EuroQol 5-dimension index measure in North American multiple sclerosis patients.
      ] and with errors reported in mapping studies in general [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ]. The fact that our most accurate model, in terms of lowest MSE, contained only 5 items from the 12-item ALSFRS-R highlights the limitations of the use of the EQ-5D-5L within MND populations. There are characteristics of the disease, as defined by the main disease-specific measure in MND, that do not influence the metric of EQ-5D-5L health utility. These are communication, salivation, swallowing, hand use, and respiratory function.
      This study is a useful addition to the literature in that it presents results for both direct and indirect mapping algorithms, using various model structures. Few previous mapping studies have carried out both approaches on the same data set [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ]. Ours is the first study, to our knowledge, to have carried out such an analysis within an MND population and provides useful evidence for the development of economic analyses in MND when EQ-5D data have not been collected directly. A strength of the analysis was the completeness of returned questionnaires with no evidence that data were not missing at random.
      Our analysis may have been more robust, however, if we had access to data for a greater number of patients. In being a longitudinal study, TONiC offered the opportunity for an analysis of repeated measures to increase the power of the study, but because only 106 patients (of 636 patients) had returned at least one follow-up questionnaire pack at our cutoff date, we considered this to be an insufficiently representative sample for such an analysis. TONiC nonetheless represents both the largest and one of the most detailed QOL studies for MND in the world. The strongest models within this study were unable to predict negative utility values for patients with MND and had a higher error rate for low utility scores. This is of concern because MND is associated with relatively low utility values reflecting very poor health-related QOL, although our data had only a few patients reporting negative utilities (2.2%). The mapping algorithms presented in this study were validated from a sample of data that stems from the same study. Although this is commonplace in the literature [
      • Dakin H.
      Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
      ,
      • Petrou S.
      • Rivero-Arias O.
      • Dakin H.
      • et al.
      The MAPS reporting statement for studies mapping onto generic preference-based outcome measures: explanation and elaboration.
      ,
      • Wailoo A.J.
      • Hernandez-Alava M.
      • Manca A.
      • et al.
      Mapping to estimate health-state utility from non–preference-based outcome measures: an ISPOR Good Practices for Outcomes Research Task Force Report.
      ], external validation would have been preferable in the context of assessing broader generalizability. Finally, it should be noted that directly collected data on EQ-5D-5L utilities always supersede predicted values based on mapping algorithms.

      Conclusions

      Many studies in MND have not used preference-based utility measures, which are required increasingly to support HTA and reimbursement decisions. The algorithms presented here provide an option for estimating EQ-5D-5L utility when this has not been collected directly from patients with MND. This study has shown that it is possible to predict, with reasonable accuracy (based on reported MSE ranges for other mapping studies), EQ-5D-5L utility values from the ALSFRS-R. It is also possible to map indirectly to EQ-5D-5L domains if the NPS and MND-HADS have been used alongside the ALSFRS-R. These findings should aid HTA of interventions for MND by providing evidence linking commonly used clinical outcome measures to a widely adopted generic preference-based measure, the EQ-5D-5L.

      Acknowledgments

      We thank the participants for their invaluable contribution and the clinical and research staff involved in the TONiC study.
      Source of financial support: The National Institute for Health Research Comprehensive Local Research Network provided research support and the Motor Neurone Disease Association UK and the Walton Neurological Disability Fund provided funding for this study.

      Supplemental Materials

      References

        • Van Es. M.A.
        • Hardiman O.
        • Chio A.
        • et al.
        Amyotrophic lateral sclerosis.
        Lancet. 2017; 390: 2084-2098
        • Miller R.G.
        • Mitchell J.D.
        • Moore D.H.
        Riluzole for amyotrophic lateral sclerosis (ALS)/motor neuron disease (MND).
        Cochrane Database Syst Rev. 2012; (CD001447)
        • Miller R.G.
        • Jackson C.E.
        • Kasarskis E.J.
        • et al.
        Practice parameter update: the care of the patient with amyotrophic lateral sclerosis: multidisciplinary care, symptom management, and cognitive/behavioural impairment (an evidence-based review).
        Neurology. 2009; 73: 1227-1233
        • Writing Group; Edaravone (MCI-186) ALS 19 Study Group
        Safety and efficacy of edaravone in well-defined patients with amyotrophic lateral sclerosis: a randomised, double-blind, placebo-controlled trial.
        Lancet Neurol. 2017; 16: 505-512
      1. Motor Neurone Disease Association. MND treatment trials. Available from: http://www.mndassociation.org/research/mnd-research-and-you/treatment-trials/. [Accessed October 2, 2016].

        • Goutman S.A.
        • Chen K.S.
        • Feldman E.L.
        Recent advances and the future of stem cell therapies in amyotrophic lateral sclerosis.
        Neurotherapeutics. 2015; 12: 428-448
        • Moore A.
        • Young C.A.
        • Hughes D.A.
        Economic studies in motor neurone disease: a systematic methodological review.
        Pharmacoeconomics. 2017; 35: 397-413
        • Longworth L.
        • Rowen D.
        Mapping to obtain EQ-5D utility values for use in NICE health technology assessments.
        Value Health. 2013; 16: 202-210
        • Brazier J.
        • Ara R.
        • Rowen D.
        • et al.
        A review of generic preference-based measures for use in cost-effectiveness models.
        Pharmacoeconomics. 2017; 35: 21-31
        • Jones A.
        • Jivraj N.
        • Balendra R.
        • et al.
        Health utility decreases with increasing clinical stage in amyotrophic lateral sclerosis.
        Amyotroph Lateral Scler Frontotemporal Degener. 2014; 15: 285-291
        • Kiebert G.M.
        • Green C.
        • Murphy C.
        • et al.
        Patients’ health-related quality of life and utilities associated with different stages of amyotrophic lateral sclerosis.
        J Neurol Sci. 2001; 191: 87-93
      2. National Institute for Health and Care Excellence. Process and methods: guide to the methods of technology appraisal. 2013. Available from: https://www.nice.org.uk/process/pmg9. [Accessed March 24, 2017].

        • Dakin H.
        Review of studies mapping from quality of life or clinical measures to EQ-5D: an online database.
        Health Qual Life Outcomes. 2013; 11: 151
        • Petrou S.
        • Rivero-Arias O.
        • Dakin H.
        • et al.
        The MAPS reporting statement for studies mapping onto generic preference-based outcome measures: explanation and elaboration.
        Pharmacoeconomics. 2015; 33: 993-1011
        • Wailoo A.J.
        • Hernandez-Alava M.
        • Manca A.
        • et al.
        Mapping to estimate health-state utility from non–preference-based outcome measures: an ISPOR Good Practices for Outcomes Research Task Force Report.
        Value Health. 2017; 20: 18-27
        • Dams J.
        • Klotsche J.
        • Bornschein B.
        • et al.
        Mapping the EQ-5D index by UPDRS and PDQ-8 in patients with Parkinson’s disease.
        Health Qual Life Outcomes. 2013; 11: 35
        • Sidovar M.F.
        • Limone B.L.
        • Lee S.
        • et al.
        Mapping the 12-item multiple sclerosis walking scale to the EuroQol 5-dimension index measure in North American multiple sclerosis patients.
        BMJ Open. 2013; 3: e002798
        • Cedarbaum J.M.
        • Stambler N.
        • Malta E.
        • et al.
        The ALSFRS-R: a revised ALS functional rating scale that incorporates assessments of respiratory function. BDNF ALS Study Group (Phase III).
        J Neurol Sci. 1999; 169: 13-21
        • Leigh P.N.
        • Swash M.
        • Iwasaki Y.
        • et al.
        Amyotrophic lateral sclerosis: a consensus viewpoint on designing and implementing a clinical trial.
        Amyotroph Lateral Scler Other Motor Neuron Disord. 2004; 5: 84-98
        • Beusterien K.
        • Leigh N.
        • Jackson C.
        • et al.
        Integrating preferences into health status assessment for amyotrophic lateral sclerosis: the ALS Utility Index.
        Amyotroph Lateral Scler Other Motor Neuron Disord. 2005; 6: 169-176
      3. Trajectories of Outcomes in Neurological Conditions. Motor Neurone Disease. Available from: https://tonic.thewaltoncentre.nhs.uk/ [Accessed March 24, 2017].

        • Devlin N.J.
        • Shah K.K.
        • Feng Y.
        • et al.
        Valuing health-related quality of life: an EQ-5D-5L value set for England.
        Health Econ. 2018; 27: 7-22
        • Franchignoni F.
        • Mora G.
        • Giordano A.
        • et al.
        Evidence of multidimensionality in the ALSFRS-R Scale: a critical appraisal on its measurement properties using Rasch analysis.
        J Neurol Neurosurg Psychiatry. 2013; 84: 1340-1345
        • Bradley S.
        • Galer M.D.
        • Jensen M.P.
        Development and preliminary validation of a pain measure specific to neuropathic pain: the Neuropathic Pain Scale.
        Neurology. 1997; 48: 332-338
        • Gibbons C.
        • Mills R.J.
        • Thornton E.W.
        • et al.
        Rasch analysis of the Hospital Anxiety and Depression Scale (HADS) for use in motor neurone disease.
        Health Qual Life Outcomes. 2011; 9: 82-90
        • Zigmond A.S.
        • Snaith R.P.
        The Hospital Anxiety and Depression Scale.
        Acta Psychiatr Scand. 1983; 67: 361-370
        • McDonald J.F.
        • Moffitt A.
        The uses of Tobit analysis.
        Rev Econ Stat. 1980; 62: 318-321
        • Core Team R.
        R: A Language and Environment for Statistical Computing.
        R Foundation for Statistical Computing, Vienna, Austria2013
        • Vrieze S.
        Model selection and psychological theory: a discussion of the differences between the Akaike information criterion (AIC) and the Bayesian information criterion (BIC).
        Psychol Methods. 2012; 17: 228-243
        • Chiò A.
        • Logroscino G.
        • Traynor B.J.
        • et al.
        Global epidemiology of amyotrophic lateral sclerosis: a systematic review of the published literature.
        Neuroepidemiology. 2013; 41: 118-130
        • Chiò A.
        • Gauthier A.
        • Montuschi A.
        • et al.
        A cross sectional study on determinants of quality of life in ALS.
        J Neurol Neurosurg Psychiatry. 2004; 75: 1597-1601