Preference-Based Assessments| Volume 21, ISSUE 11, P1338-1345, November 2018

Ok

Health State Values Derived from People with Multiple Sclerosis for a Condition-Specific Preference-Based Measure: Multiple Sclerosis Impact Scale–Eight Dimensions–Patient Version (MSIS-8D-P)

Open ArchivePublished:June 08, 2018

Abstract

Objective

In economic evaluation, health outcomes are commonly quantified using quality-adjusted life-years (QALYs) derived from the preferences of a sample of the general population. It can be argued that this approach ignores the preferences of people with experience of the condition, and that patient preferences have a place in the valuation of health outcomes. Here we report the estimation of a preference-based index for an existing condition-specific preference-based measure for multiple sclerosis (MS), the MSIS-8D, based on the preferences of people with MS.

Study design

Internet time trade-off (TTO) survey, eliciting preferences from people with MS.

Methods

We elicited preferences from a sample of people with MS (N = 1635) across 169 MSIS-8D health states, using the TTO technique. We fitted ordinary least squares and random effects models to the survey data to estimate values for all health states described by the MSIS-8D.

Results

The new patient-derived index (the MSIS-8D-P) provides values ranging from 0.893 for the best possible health state to 0.138 for the worst state. The MSIS-8D-P exhibits good discriminative validity, identifying expected significant differences between groups based on presence/absence of MS, type of MS, and duration since diagnosis.

Conclusions

The MSIS-8D-P index of values for MS-specific health states provides an opportunity to estimate QALYs based on patient preferences, for use in economic evaluations of treatments for MS. More broadly, it adds to the methods and data available to consider the health-related quality of life of people with MS to inform resource allocation and individual-level decisions regarding treatments for MS.

Introduction

When considering the cost-effectiveness of health care interventions, the effects of treatment are frequently assessed using quality-adjusted life-years (QALYs). QALYs are calculated by weighting each year of life according to its quality, on a scale from 1 (equivalent to full health) to 0 (equivalent to being dead). This combines the impact of treatment on length and quality of life into one measure. QALY weights are generally estimated by eliciting preferences between health states from a sample of the general population, or from a sample of people with the condition that the intervention is designed to address (hereafter referred to as “patients”), using a preference elicitation technique [
• Weinstein M.C.
• Torrance G.
• McGuire A.
QALYs: the basics.
]. These techniques enable preferences to be quantified, thereby producing the quality weights, or health state values (HSVs), required for the calculation of QALYs [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
]. QALYs are commonly based on preference-based measures (PBMs) of health-related quality of life (HRQL), which use a standardized classification system for describing health states and a tariff of quality weights for all health states described by the classification system. The most commonly used PBMs, including the EuroQol EQ-5D, are designed to be generic, that is, suitable for any health condition, although a growing number of condition-specific PBMs are becoming available.
Most commonly, preferences are elicited from members of the general population, and this approach is specifically recommended in most policy settings, for example, by the National Institute for Health and Care Excellence [

National Institute for Health and Care Excellence (NICE). Guide to the methods of technology appraisal 2013. National Institute for Health and Care Excellence (NICE), 2013. Available from: http://www.nice.org.uk/article/pmg9/chapter/foreword. [Accessed October 23, 2017].

] and the Panel on Cost-Effectiveness in Health and Medicine [
• Gold M.R.
• Siegel J.E.
• Russell L.B.
• Weinstein M.C.
Cost-Effectiveness in Health and Medicine.
,
• Sanders G.D.
• Neumann P.J.
• Basu A.
• et al.
Recommendations for conduct, methodological practices, and reporting of cost-effectiveness analyses: Second Panel on Cost-Effectiveness in Health and Medicine.
]. However, this approach ignores the preferences of people with experience of the condition, and it has been argued that patient preferences have a place in the valuation of health [
• Nord E.
• Pinto J.L.
• Richardson J.
• et al.
Incorporating societal concerns for fairness in numerical valuations of health programs.
,
• Versteegh M.M.
• Brouwer W.B.F.
Patient and general public preferences for health states: a call to reconsider current guidelines.
]. The availability of patient preferences is all the more salient given the evidence of significant differences between public- and patient-derived HSVs [
• Stiggelbout A.M.
• de Vogel-Voort E.
Health state utilities: a framework for studying the gap between the imagined and the real.
]. Within a UK health policy context that is increasingly patient centered, with initiatives such as “No decision about me, without me” [
Department of Health
Liberating the NHS: no decision about me, without me—Government response to the consultation.
], we believe it is relevant and timely to review the role of patient preferences in economic evaluation. Whereas some discussions represent a dichotomy between public or patient values, others suggest using values from both perspectives [
• Gold M.R.
• Siegel J.E.
• Russell L.B.
• Weinstein M.C.
Cost-Effectiveness in Health and Medicine.
,
• Nord E.
• Pinto J.L.
• Richardson J.
• et al.
Incorporating societal concerns for fairness in numerical valuations of health programs.
,
• Versteegh M.M.
• Brouwer W.B.F.
Patient and general public preferences for health states: a call to reconsider current guidelines.
,
• Menzel P.
• Dolan P.
• Richardson J.
• Olsen J.A.
The role of adaptation to disability and disease in health state valuation: a preliminary normative analysis.
]. The common practice of reporting sensitivity analyses alongside base-case cost-effectiveness results provides an opportunity for achieving this, in the context of reimbursement considerations, resource allocation decisions, and individual-level treatment decisions [
• Stiggelbout A.M.
• de Vogel-Voort E.
Health state utilities: a framework for studying the gap between the imagined and the real.
].
In previous research, we developed a multiple sclerosis (MS)-specific PBM, the eight-dimension Multiple Sclerosis Impact Scale (MSIS-8D), with QALY weights based on the preferences of the UK general population [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
]. Here, we aim to provide an alternative tariff of QALY weights for the MSIS-8D, based on the preferences of people with MS, for potential use across this spectrum of decision making.

Methods

MS is a chronic inflammatory condition affecting the central nervous system. In the majority of cases, people with MS experience exacerbations of symptoms (relapses) interspersed with periods of total or partial remission, before developing a progressive disease course. In approximately 10% to 15% of cases, the disease is progressive from onset [
• Zajicek J.
• Freeman J.
• Porter B.
Mulitple Sclerosis Care: A Practical Manual.
]. Symptoms vary widely and can include physical, psychological, and cognitive effects [
• Hemmett L.
• Holmes J.
• Barnes M.
• Russell N.
What drives quality of life in multiple sclerosis?.
]. Here we describe the methods employed in conducting a valuation survey with a sample of people with MS and in undertaking analysis to estimate HSVs.

Health State Descriptions (the MSIS-8D)

Health states for MS were described using the MSIS-8D, which was developed in response to concerns about the content validity [
• Hemmett L.
• Holmes J.
• Barnes M.
• Russell N.
What drives quality of life in multiple sclerosis?.
,
• Fisk J.D.
• Brown M.G.
• Sketris I.S.
• et al.
A comparison of health utility measures for the evaluation of multiple sclerosis treatments.
,
• Kuspinar A.
• Mayo N.E.
Do generic utility measures capture what is important to the quality of life of people with multiple sclerosis?.
,
• Orme M.
• Kerrigan J.
• Tyas D.
• et al.
The effect of disease, functional status, and relapses on the utility of people with multiple sclerosis in the UK.
] and sensitivity [
• Hemmett L.
• Holmes J.
• Barnes M.
• Russell N.
What drives quality of life in multiple sclerosis?.
,
• Benito-León J.
• Morales J.M.
• Rivera-Navarro J.
• Mitchell A.
A review about the impact of multiple sclerosis on health-related quality of life.
,
• Gruenewald D.A.
• Higginson I.J.
• Vivat B.
• et al.
Quality of life measures for the palliative care of people severely affected by multiple sclerosis: a systematic review.
,
• Opara J.A.
• Jaracz K.
• Brola W.
Quality of life in multiple sclerosis.
] of generic PBMs in the context of MS. The MSIS-8D descriptive system was derived from the Multiple Sclerosis Impact Scale (MSIS-29), a well validated and frequently used patient-reported outcome measure for MS; this is described in detail elsewhere [
• Goodwin E.
• Green C.
A QALY measure for multiple sclerosis: developing a patient-reported health state classification system for an MS-specific preference-based measure.
]. In summary, it represents eight dimensions of importance to the HRQL of people with MS: physical functioning, mobility, social activities, daily activities, fatigue, cognitive function, emotional well-being, and depression. The original items of the MSIS-29 were based primarily on qualitative work with people with MS, alongside expert opinion and a literature review [
• Hobart J.C.
• Riazi A.
• Lamping D.L.
• et al.
Improving the evaluation of therapeutic interventions in multiple sclerosis: development of a patient-based measure of outcome.
]. The eight dimensions covered by the MSIS-8D descriptive system were informed by a number of previous studies that used qualitative techniques to explore HRQL among people with MS [
• Goodwin E.
• Green C.
A QALY measure for multiple sclerosis: developing a patient-reported health state classification system for an MS-specific preference-based measure.
]. Each dimension is represented by one MSIS-29 item with four response levels: not at all, a little, moderately, and extremely. This constitutes a descriptive system (Fig. 1) that describes 65,536 unique MS health states.
A tariff of HSVs for the MSIS-8D has been estimated previously [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
]), based on preferences elicited from a representative sample of the UK general population for a sample of 169 MSIS-8D health states, which was selected using the Rasch vignette approach [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
] to reflect states that are likely to be experienced by people with MS at different levels of severity. The same health states are used for the current study.

Valuation Survey

The valuation survey followed the protocol used to obtain MSIS-8D values from the general population, which was based on the Measurement and Valuation of Health (MVH) version of the time trade-off (TTO) technique. The MVH protocol was developed to generate the UK tariff of preference weights for the EQ-5D-3L [
• Gudex C.
Time trade-off user manual: props and self-completion methods.
,
• Dolan P.
Modeling valuations for EuroQol health states.
]. Respondents are presented with a choice between two hypothetical scenarios: living in a suboptimal health state for a given number of years or living in perfect health for a shorter period of time. The length of time spent in perfect health is varied until the respondent is indifferent between the two scenarios. In this way, HSVs are determined by asking respondents to trade between quality and length of life [
• Gudex C.
Time trade-off user manual: props and self-completion methods.
]. The survey was administered via the internet. To ensure its suitability for the target population, people with MS were involved in developing the survey protocol. We used pre-pilot testing and an online pilot (N = 55) with people with MS to finalize the valuation methods.
Before undertaking the TTO tasks, participants completed the MSIS-8D descriptive system for their own health to familiarize themselves with the descriptive system. As warm-up exercises, participants were asked to rank three MSIS-8D health states in order of preference and to complete a practice TTO exercise with detailed instructions. Participants were then asked to value six MSIS-8D health states. Each set of health states was stratified to include five health states covering a range of severity plus the worst possible health state (the “pits” state). Each participant was randomly assigned a set of health states. Previous MSIS-8D surveys had demonstrated that this represented an acceptable workload for participants [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
,
• Green C.
• Goodwin E.
• Hawton A.
“Naming and framing”: the impact of labeling on health state values for multiple sclerosis.
]; this was confirmed during pilot testing with people with MS.
Ethical approval was obtained from the University of Exeter Medical School Ethics Committee.

Sample of People with MS

Respondents were sourced from the UK Multiple Sclerosis Register (the MS Register). The MS Register was launched in May 2011 and had 13,920 members as of July 18, 2016 [

MS Register website. Available from: https://www.ukmsregister.org/Portal/Home#about. [Accessed July 18, 2016].

]. Members are requested to complete a range of patient-reported outcome measures, including the MSIS-29, via an internet portal every 3 months. Other available data include sociodemographic and clinical characteristics. Initial analysis indicates that members are broadly representative of people with MS in the United Kingdom in terms of key characteristics including gender, age at onset, and MS type [
• Ford D.V.
• Jones K.H.
• Middleton R.M.
• et al.
The feasibility of collecting information from people with multiple sclerosis or the UK MS Register via a web portal: characterising a cohort of people with MS.
,
• Mackenzie I.S.
• Morant S.V.
• Bloomfield G.A.
• et al.
Incidence and prevalence of multiple sclerosis in the UK 1990–2010: a descriptive study in the General Practice Research Database.
].
Invitation emails were circulated to all current members of the MS Register during March 2016. Our target sample size was 1500 respondents, to generate approximately 40 observations per health state, based on a review of the literature describing the valuation of health states for condition-specific PBMs [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
] with an adjustment to allow for the increased variance that may result from internet administration [
• Norman R.
• King M.T.
• Clarke D.
• et al.
]. In previous valuation surveys, this was sufficient for the estimation of a regression model to predict HSVs for the MSIS-8D [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
].

Data Cleaning and Descriptive Analysis

When estimating a tariff for a PBM, it is common practice to exclude data from respondents who provide responses that are internally inconsistent or illogical. This study adopted the exclusion criteria developed for previous MSIS-8D surveys, which were based on the condition-specific PBM development literature [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
]. During the practice TTO exercise, respondents were screened out of the survey if they considered 10 years in full health to be worse than or equivalent to 10 years with health problems, or considered 10 years in full health to be worse than or equivalent to being dead. After data collection, respondents were excluded from the analysis if they:
• gave the same value to all health states (unless they valued all health states as equivalent to full health);
• gave all states a value less than or equal to 0;
• valued the pits state at least as highly as all other states;
• gave the least severe state a lower value than all other states; or
• provided three or more inconsistent responses with a difference in HSV of at least 0.1; that is, they valued a dominated health state as better than a logically better alternative by the equivalent of 1 year in the TTO exercise.
Negative HSVs (i.e., health states considered to be worse than being dead) were transformed onto a scale from 0 to –1 following the method used to estimate the UK tariff for the EQ-5D-3L [
• Dolan P.
Modeling valuations for EuroQol health states.
].

Modeling to Obtain Health State Values

HSVs for all MSIS-8D health states were estimated using the standard regression model [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
]:
$hij=f(β′Xλ∂)+εij$

where hij represents the TTO value; i represents individual health states; j represents individual respondents; f represents the functional form; X represents a vector of dummy explanatory variables for each level λ of dimension ∂ of the classification system, where level λ = 1 acts as a baseline; and εij represents the error term.
We estimated individual-level and mean-level ordinary least squares (OLS) models, fixed or random effects (RE) models to account for the clustering of data by respondent, and RE Tobit models to allow for censoring of HSV data between –0.975 and 1 [
• Brazier J.
• Ratcliffe J.
• Salomon J.A.
• Tsuchiya A.
Measuring and Valuing Health for Economic Evaluation.
]. Any inconsistent coefficients, where a less severe item level resulted in a greater HRQL decrement than a more severe item level, were merged and the analysis was rerun to produce a consistent model. Additional versions of these models were generated by merging adjacent item-levels represented by coefficients that were nonsignificant at the 95% level [
• Versteegh M.M.
• Leunis A.
• Uyl-de Groot C.A.
• Stolk E.A.
Condition-specific preference-based measures: benefit or burden?.
]. We did not assume that the best MSIS-8D health state represents perfect health; therefore the constant was not constrained to unity [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
].
Models were compared in terms of the proportion of coefficients that were significant, mean absolute error (MAE) of predicted HSVs, and the number of health states with absolute errors greater than 5% and 10% (equivalent to 6 months and 1 year in the TTO exercise respectively) [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
]. This enabled selection of a preferred model to generate an index of HSVs for the MSIS-8D based on the preferences of people with MS. Here we do not compare the preferences of people with MS and those of the general public. A detailed examination of this comparison will be reported in a companion article.

Discriminative Validity

The sensitivity of the MSIS-8D-P index was assessed by exploring its discriminative validity [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
]. Empirical data suggest that progressive types of MS have a greater impact on HRQL than relapsing–remitting MS and that the HRQL of people with MS decreases over time; therefore we would expect these differences to be reflected in HSVs [
• Hawton A.
• Green C.
Health utilities for multiple sclerosis.
,
• Benito-León J.
• Morales J.M.
• Rivera-Navarro J.
Health-related quality of life and its relationship to cognitive and emotional functioning in multiple sclerosis patients.
]. Independent t tests were used to determine the ability of the MSIS-8D-P index to distinguish between subgroups of respondents to the valuation survey, based on type of MS and duration since diagnosis. Additional analysis was undertaken using MSIS-8D responses from previous surveys of the general population [
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
,
• Green C.
• Goodwin E.
• Hawton A.
“Naming and framing”: the impact of labeling on health state values for multiple sclerosis.
] to assess the ability of the MSIS-8D-P index to distinguish between people with and without MS.
Analyses were undertaken using Microsoft Excel and Stata.

Results

Valuation Survey

In total, 3565 members of the MS Register entered the website, of whom 1635 (46%) completed the survey. Of these, 39 (2.39%) provided inconsistent or illogical responses and were excluded from the analysis. No differences were apparent in the characteristics of excluded and included respondents.
Table 1 presents the sociodemographic and clinical characteristics of the 1596 respondents who were included in the analysis compared to those of all MS Register members (for whom data were available) at the time the survey was closed. The characteristics of the survey respondents reflected those of the MS Register members overall in terms of age, sex, employment status, type and duration of MS, and self-reported health status, although a higher proportion of respondents had a university education. Table 1 also reports respondents’ views of task comprehension and difficulty. The majority (84%) reported that they found the TTO questions easy or very easy to understand. Nearly half of respondents (47%) found it difficult to choose between the scenarios, but only 7% said they found this very difficult.
Table 1Characteristics of respondents to the preference elicitation survey
MSIS-8D survey sampleMS Register members (%)
CharacteristicNumberPercentage
Sex
Female11457372
Male4242728
Age group (years)
25 and younger701
26 to 358557
36 to 453041920
46 to 555043230
56 to 654633027
Older than 652051315
Employment status
Economically active6334139
Economically inactive9125961
Highest level of education
University6584333
Occupational4643034
Compulsory2981926
Other12587
Type of MS
Relapsing–remitting MS7454951
Secondary progressive MS3942625
Primary progressive MS2411614
Benign7955
Unknown5545
Time since diagnosis
Less than 10 years6024140
10 to 19 years5303636
20 years or longer3302324
Respondents’ self-reported raw scores on the MSIS-8D
MSIS-8D total score
Mild (score 8–16)44128
Moderate (score 17–24)68643
Severe (score 25–32)46829
MSIS-8D physical score
Mild (score 4–8)44328
Moderate (score 9–12)51432
Severe (score 13–16)63840
MSIS-8D psychological score
Mild (score 4–8)65241
Moderate (score 9–12)61639
Severe (score 13–16)32721
What were the questions like to understand?
Very easy39124.50
Easy94659.27
Difficult23914.97
Very difficult201.25
How easy or difficult was it to make choices between the options you were asked to think about?
Very easy1358.46
Easy58836.84
Difficult75547.31
Very difficult1187.39
The mean observed HSVs for the 169 health states included in the survey, presented in the Appendix, ranged from 0.15 for the pits state to 0.94 for the best state. The mean number of observations per health state was 48. The distribution of individual observed HSVs over the full possible range of values reflected the left skew and clustering at 0 and 1 that are typical of TTO data [
• Brazier J.
• Ratcliffe J.
• Salomon J.A.
• Tsuchiya A.
Measuring and Valuing Health for Economic Evaluation.
].
The 169 health states were assigned to 24 severity groups, based on the sum of response levels across all dimensions [
• Kind P.
A revised protocol for the valuation of health states defined by the EQ-5D-3L classification system: learning the lessons from the MVH study.
]. The average HSV for each severity group is shown in Table 2. The pattern is consistent with the expected direction of preferences (i.e., mean observed HSVs decrease as severity increases), with the exception of three discrepancies (group 2, group 4, and group 12).
Table 2Mean health state values by severity group
Severity groupTotal scoreMeanSDMinMaxObsNumber of health states
080.9430.15001541
190.8820.203014878
2100.8430.2140.02511183
3110.8530.216–0.50012034
4120.7940.262–0.97512475
5130.8190.233–0.212806
6140.8040.246–0.27513337
7150.8010.237–0.97513918
8160.7300.290–0.97514639
9170.7130.285–0.725142110
10180.6750.342–0.900143410
11190.6480.335–0.825144810
12200.6180.362–0.925152011
13210.6270.368–0.975149110
14220.5870.363–0.97514349
15230.5450.395–0.82513918
16240.4900.420–0.97513458
17250.4510.419–0.82513778
18260.4150.443–0.97513227
19270.4050.430–0.97512356
20280.3480.461–0.97512535
21290.3390.478–0.90011964
22300.2870.488–0.97511313
23310.1570.486–0.97514068
24320.1460.480–0.975115961
max, maximum observed value; min, minimum observed value; obs, observations.

Modeling Health State Values

The Hausman test yielded a nonsignificant result (χ2 (24) = 14.05; P = 0.95), indicating that a fixed effects specification would produce a similar result with reduced efficiency; therefore a random effects specification was used [
• Greene W.H.
]. The individual-level OLS and RE models each had one coefficient (corresponding to level 2 of the Fatigue dimension) with an inconsistent sign, and the aggregate-level OLS model produced three coefficients with an inconsistent sign (for Mobility levels 2 and 3 and for Fatigue level 2). Consistent versions of these models were created by merging the affected levels.
Table 3 summarizes the consistent individual OLS, aggregate-level OLS and RE models and the original Tobit model. All models had coefficients that were consistent with expected preferences; that is, for each dimension of the MSIS-8D, coefficient values decreased as the level of severity increased. The consistent individual-level OLS performed the least well, with the highest MAE (0.0469), number of errors greater than 0.1 (17), and number of errors greater than 0.05 (63). Just over one third (35%) of coefficients were significant.
Table 3Regression models for the estimation of health state values
Consistent individual OLSConsistent mean OLSConsistent RE modelTobit modelPreferred model: RE version 2Tobit version 2
CoeffPCoeffPCoeffPCoeffPCoeffPCoeffP
Physical
A little–0.0340.006–0.0370.045–0.0400.008–0.0720.000–0.0470.000–0.0800.000
Moderately–0.0360.082–0.0400.098–0.0420.052–0.0880.001–0.0650.000–0.0920.000
Extremely–0.1470.000–0.1510.000–0.1510.000–0.2250.000–0.1750.000–0.2300.000
Social
A little–0.0220.173–0.0220.228–0.0250.108–0.0320.089–0.0370.030
Moderately–0.0470.077–0.0440.069–0.0510.027–0.0700.009–0.0320.019–0.0830.000
Extremely–0.0710.039–0.0770.009–0.0860.001–0.1150.000–0.0670.001–0.1280.000
Mobility
A little–0.0010.935–0.0030.820–0.0060.752–0.0030.856–0.0060.716
Moderately–0.0010.960–0.0170.449–0.0190.462
Extremely–0.0840.018–0.0790.000–0.0920.001–0.0970.001–0.0770.000–0.0840.000
Daily activities
A little–0.0120.385–0.0100.5860.0000.996–0.0090.629
Moderately–0.0350.172–0.0320.190–0.0130.568–0.0290.267–0.0200.132–0.0270.077
Extremely–0.0640.063–0.0650.029–0.0390.135–0.0530.084–0.0480.015–0.0510.022
Fatigue
A little–0.0030.859–0.0030.842
Moderately–0.0340.068–0.0370.040–0.0200.206–0.0330.184–0.0210.137
Extremely–0.0770.005–0.0860.001–0.0620.004–0.0890.003–0.0630.002–0.0600.010
Emotion
A little–0.0170.173–0.0160.260–0.0160.187–0.0330.048–0.0150.203–0.0340.041
Moderately–0.0310.174–0.0300.197–0.0420.034–0.0600.014–0.0420.035–0.0770.001
Extremely–0.0490.165–0.0520.090–0.0700.008–0.0890.005–0.0690.009–0.1060.000
Cognition
A little–0.0270.054–0.0290.088–0.0270.058–0.0280.106–0.0270.030–0.0280.104
Moderately–0.0550.022–0.0530.033–0.0520.013–0.0580.018–0.0520.008–0.0570.019
Extremely–0.1070.002–0.1020.001–0.1150.000–0.1210.000–0.1160.000–0.1200.000
Depression
A little–0.0060.706–0.0010.9680.0000.974–0.0160.341–0.0300.047
Moderately–0.0440.102–0.0400.106–0.0410.050–0.0650.006–0.0400.008–0.0790.000
Extremely–0.1660.000–0.1700.000–0.1410.000–0.1560.000–0.1400.000–0.1680.000
Constant0.9020.0000.9020.0000.8940.0001.0890.0000.8930.0001.0920.000
Model performance
Coefficients232223241921
Sig coefficients8 (34.78%)10 (47.62%)11 (47.83%)14 (58.33%)15 (78.95%)17 (80.95%)
Mean absolute error0.04690.03490.03610.03910.03640.0399
No. of errors > 0.11722738
No. of errors > 0.05634650495251
Obs (respondents)9576 (1596)169 (1596)9576 (1596)9576 (1596)9576 (1596)9576 (1596)
Wald χ2NANA8897.14 (23)9305.22 (24)8893.62 (19)9300.19 (21)
Prob > χ2NANA<0.001<0.001<0.001<0.001<0.001
Overall R2NANA0.3014NA0.3013NA
Log likelihoodNANANA–3846.31NANA
F185.91 (24, 9551)133.94 (21, 147)NANANANA
Prob > F<0.001<0.001NANANANA
R20.30170.9503NANANANA
Root-mean-square error0.37670.0469NANANANA
Coeff, coefficient; OLS, ordinary least squares; RE, random effects.
The consistent mean-level OLS and RE models performed similarly well. The Tobit model performed better in terms of significant coefficients (58.33%); however, it had a slightly higher MAE (0.0391) and number of errors greater than 0.1 (7).
To avoid the loss of information that occurs when data are aggregated, the mean-level OLS model was not considered further. The remaining RE and Tobit models, however, had a relatively low proportion of significant coefficients. Therefore, various options for merging affected item levels were explored. The best-performing of these parsimonious models (RE version 2 and Tobit version 2) are presented in Table 3. Results for all other models are available from the authors on request.

Selection of Preferred Model

Overall, RE version 2 had superior predictive ability, with fewer errors greater than 0.1 and a lower MAE than Tobit version 2. On this basis, the parsimonious RE version 2 is the recommended model for estimation of a single index for the MSIS-8D based on the preferences of people with MS. The index ranges from 0.893 for the best MSIS-8D health state to 0.138 for the pits state. This indicates that respondents considered the best state to be less than perfect health; that is, they assumed decrements in HRQL beyond the dimensions included in the classification system. This is not unusual for a condition-specific PBM [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
].
This preferred model, as presented in Table 3, enables a HSV to be calculated for any MSIS-8D health state, by summing the constant and the coefficient for each item depending on its level. For example, the predicted value for the MSIS-8D health state (3,3,2,3,4,2,2,1) is calculated as:
Constant + Physical (3) + Social (3) + Mobility (2) + Daily activities (3) + Fatigue (4) + Emotion (2) + Cognition (2) + Depression (1)
= 0.893 – 0.065 – 0.032 – 0.003 – 0.020 – 0.063 – 0.015 – 0.027 + 0 = 0.668.

Discriminative Validity

Table 4 presents data describing the discriminative ability of the MSIS-8D-P. The results of the t tests provided strong evidence that the index is capable of discriminating between subgroups of people with MS that would be expected to differ in terms of their HRQL (P < 0.0001). Significantly lower HSVs were observed for those with progressive rather than relapsing–remitting MS, and for those with a disease duration of 10 or more years since diagnosis. A large, significant difference was also observed between survey respondents with and without MS.
Table 4Discriminative validity of the MSIS-8D-P
MeanSDFrequencyt statisticP value
Disease statusNo MS0.7660.172349028.931<0.0001
MS0.6130.1861635
Duration of MSLess than 10 years0.6450.1876124.943<0.0001
10 years or longer0.5970.182882
MS typeRelapsing0.6660.177760–12.651<0.0001
Progressive0.5470.175652

Discussion

We have elicited preferences from people with MS to derive an alternative tariff of HSVs for an existing health state classification system. We refer to this as the MSIS-8D-P (Multiple Sclerosis Impact Scale–8 Dimensions–People with MS). This provides an additional source of information about the impact of MS and its treatment on HRQL, alongside the original tariff of MSIS-8D values, which is based on the preferences of the general population. Both tariffs are suitable for use across all types of MS, to assess HRQL and to estimate QALYs, and can be derived directly from patient-reported responses to the MSIS-29, a well validated and frequently used patient-reported outcome measure for MS. The methods employed in this study are well accepted for the generation of HSVs for condition-specific descriptive systems [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
]. In terms of the mean absolute error and the proportion of health states with a prediction error greater than 0.05 or 0.1, the preferred model for the estimation of MSIS-8D-P values compares favourably with models that have been developed to estimate tariffs for other condition-specific PBMs [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
]. The MSIS-8D-P exhibits good discriminative validity, suggesting that it is sensitive to differences and changes in the HRQL of people with MS.
There are two potential uses for the MSIS-8D-P: providing QALY weights from the perspective of people with MS to inform economic evaluations and providing a source of data on the HRQL of people with MS to inform condition-specific resource allocation and individual-level treatment decisions. In addition, the availability of tariffs for the MSIS-8D classification system from people with MS and from the general population enables a full comparison to be drawn between public and patient values for MS health states. This analysis will be reported in a companion article.

Strengths and Limitations of the Study

The results of the questions regarding self-reported task comprehension, and the nature of the preference data gathered, indicate that it is possible to administer a complex technique such as the TTO via the internet to people with a chronic condition such as MS, which can, in some cases, affect cognitive functioning. The direct involvement of people with MS in reviewing the survey protocol, the careful construction of instructions and warm-up tasks, and the pre-pilot and pilot tests were instrumental in ensuring that the survey was appropriate for the target population.
Although the approach taken in this study was informed by best practice guidance on the development of condition-specific PBMs [
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
], it has some limitations. In keeping with national guidelines [

National Institute for Health and Care Excellence (NICE). Guide to the methods of technology appraisal 2013. National Institute for Health and Care Excellence (NICE), 2013. Available from: http://www.nice.org.uk/article/pmg9/chapter/foreword. [Accessed October 23, 2017].

], HSVs were elicited using the MVH version of the TTO. This asks respondents to imagine remaining in a specified health state for 10 years, with no changes in that health state during that time. MS is, however, usually characterized by alternating periods of relapse and remission, or by ongoing progression [
• Zajicek J.
• Freeman J.
• Porter B.
Mulitple Sclerosis Care: A Practical Manual.
]. This may have caused confusion for respondents with MS and may have affected the values they attributed to health states.
The proportion of coefficients that were significant in the initial RE model was relatively small compared to other models that have been estimated to predict HSVs for condition-specific PBMs. As a result, we merged four pairs of adjacent dimension levels to produce the preferred model (along with an additional pair that was merged to address one coefficient with an unexpected sign, which is not unusual for models of this type) [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
]. This increased both the number and the proportion of coefficients that were significant. The levels that were merged and the coefficients that remained nonsignificant in the preferred model indicate that the preferences of people with MS were not sensitive to shifts from level 1 (not at all) to level 2 (a little) for the Social, Mobility, Daily Activities, Emotion, and Depression dimensions, or to shifts from level 2 (a little) to level 3 (moderately) for the Mobility, Daily Activities, and Fatigue dimensions.

Patient or Public Values?

Two main arguments have been put forward to support the use of patient preferences in economic evaluation. The first rests on the theory of welfare economics, which posits that the well-being of a society equals the sum of the utilities of its individual members. This implies that it is more appropriate to base decisions regarding public programs on the preferences of those set to gain or lose directly from the decision, rather than a wider sample, many of whom will be unaffected. The second is more prosaic: patients are likely to have more experience of poor health and are hence better placed to value how this affects quality of life. Conversely, it is argued that societal preferences should guide resource allocation so as to reflect the views of those who are funding the service [
• Brazier J.
• Akehurst R.
• Brennan A.
• et al.
Should patients have a greater role in valuing health states?.
], while some have expressed concerns that strategic bias may be introduced into HSVs if patients attempt to maximize the possibility of treatments being considered cost effective [
• Dolan P.
Whose preferences count?.
]. Furthermore, there is debate over whether the differences between public and patient values represent a better understanding of the impact of health states by those who have experienced them, or less desirable effects of ill health on how people assess their situation such as distortions in HSVs or negative forms of adaptation (e.g., people failing to recognize how poor their current health is, what full health feels like and what it would allow them to do, or lowering their expectations) [
• Menzel P.
• Dolan P.
• Richardson J.
• Olsen J.A.
The role of adaptation to disability and disease in health state valuation: a preliminary normative analysis.
].
There are arguments for and against the use of patient values to inform resource allocation decisions; however, neither approach can claim superior theoretical or empirical validity, and the choice between the two is likely to affect which types of intervention are considered cost effective [
• Versteegh M.M.
• Brouwer W.B.F.
Patient and general public preferences for health states: a call to reconsider current guidelines.
]. It has therefore been suggested that cost-effectiveness results based on patient preferences should be used in conjunction with results based on the preferences of the general population [
• Gold M.R.
• Siegel J.E.
• Russell L.B.
• Weinstein M.C.
Cost-Effectiveness in Health and Medicine.
,
• Nord E.
• Pinto J.L.
• Richardson J.
• et al.
Incorporating societal concerns for fairness in numerical valuations of health programs.
,
• Versteegh M.M.
• Brouwer W.B.F.
Patient and general public preferences for health states: a call to reconsider current guidelines.
,
• Menzel P.
• Dolan P.
• Richardson J.
• Olsen J.A.
The role of adaptation to disability and disease in health state valuation: a preliminary normative analysis.
]. The MSIS-8D-P provides the information required to apply this approach in the context of MS.
The use of condition-specific PBMs is not limited to informing the allocation of resources across whole health care systems. It has been suggested that, although public preferences are better suited to system-wide decision making, patient values are more appropriate for informing condition-specific resource allocation and individual-level treatment decisions [
• Gold M.R.
• Siegel J.E.
• Russell L.B.
• Weinstein M.C.
Cost-Effectiveness in Health and Medicine.
]. A recent systematic review of the literature describing the development of condition-specific PBMs [
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
] identified 21 instruments with tariffs based on patient preferences. Of these, 18 were specifically designed to inform individual or clinical-level decision making, rather than system-wide resource allocation. Such measures provide useful information about the factors that influence patients’ experiences of living with disease and that inform their decisions between treatment alternatives, and the relative importance of these factors.

Conclusion

This new tariff of HSVs for the MSIS-8D, based on the preferences of people with MS, provides an additional source of information to assess the impact of MS and treatments for people with MS, providing a broader context for resource allocation decision making. We recommend that such information should be used to determine the impact of using QALYs based on public preferences versus QALYs derived from preferences of those with experience of the condition, and to consider more broadly the common policy guidance on use of public preferences when undertaking economic evaluation. The MSIS-8D-P may also help to inform resource allocation within ring-fenced budgets for MS and individual-level treatment decisions.

Supplementary material: The MSIS-8D-P valuation survey is available from the corresponding author upon request.

Acknowledgments

Financial support for this study was provided in part by the Multiple Sclerosis Society of Great Britain and Northern Ireland. Partial funding was received from the UK National Institute for Health Research Collaboration for Leadership in Applied Health Research and Care of the South West Peninsula (PenCLAHRC) to CG and AH.

References

• Weinstein M.C.
• Torrance G.
• McGuire A.
QALYs: the basics.
Value Health. 2009; 12: S5-S9
• Brazier J.E.
• Rowen D.
• Mavranezouli I.
• et al.
Developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
Health Technol Assess. 2012; 16: 1-114
1. National Institute for Health and Care Excellence (NICE). Guide to the methods of technology appraisal 2013. National Institute for Health and Care Excellence (NICE), 2013. Available from: http://www.nice.org.uk/article/pmg9/chapter/foreword. [Accessed October 23, 2017].

• Gold M.R.
• Siegel J.E.
• Russell L.B.
• Weinstein M.C.
Cost-Effectiveness in Health and Medicine.
Oxford University Press, New York1996
• Sanders G.D.
• Neumann P.J.
• Basu A.
• et al.
Recommendations for conduct, methodological practices, and reporting of cost-effectiveness analyses: Second Panel on Cost-Effectiveness in Health and Medicine.
JAMA. 2016; 316: 1093-1103
• Nord E.
• Pinto J.L.
• Richardson J.
• et al.
Incorporating societal concerns for fairness in numerical valuations of health programs.
Health Econ. 1999; 8: 25-39
• Versteegh M.M.
• Brouwer W.B.F.
Patient and general public preferences for health states: a call to reconsider current guidelines.
Soc Sci Med. 2016; 165: 66-74
• Stiggelbout A.M.
• de Vogel-Voort E.
Health state utilities: a framework for studying the gap between the imagined and the real.
Value Health. 2008; 11: 76-87
• Department of Health
Liberating the NHS: no decision about me, without me—Government response to the consultation.
• Menzel P.
• Dolan P.
• Richardson J.
• Olsen J.A.
The role of adaptation to disability and disease in health state valuation: a preliminary normative analysis.
Soc Sci Med. 2002; 55: 2149-2158
• Goodwin E.
• Green C.
• Spencer A.
Estimating a preference-based index for an eight dimensional health state classification system derived from the Multiple Sclerosis Impact Scale (MSIS-29).
Value Health. 2015; 18: 1025-1036
• Zajicek J.
• Freeman J.
• Porter B.
Mulitple Sclerosis Care: A Practical Manual.
Oxford University Press, Oxford2007
• Hemmett L.
• Holmes J.
• Barnes M.
• Russell N.
What drives quality of life in multiple sclerosis?.
QJM. 2004; 97: 671-676
• Fisk J.D.
• Brown M.G.
• Sketris I.S.
• et al.
A comparison of health utility measures for the evaluation of multiple sclerosis treatments.
J Neurol Neurosurg Psychiatry. 2005; 76: 58-63
• Kuspinar A.
• Mayo N.E.
Do generic utility measures capture what is important to the quality of life of people with multiple sclerosis?.
Health Qual Life Outcomes. 2013; 11: 1-10
• Orme M.
• Kerrigan J.
• Tyas D.
• et al.
The effect of disease, functional status, and relapses on the utility of people with multiple sclerosis in the UK.
Value Health. 2007; 10: 54-60
• Benito-León J.
• Morales J.M.
• Rivera-Navarro J.
• Mitchell A.
A review about the impact of multiple sclerosis on health-related quality of life.
Disabil Rehabil. 2003; 25: 1291-1303
• Gruenewald D.A.
• Higginson I.J.
• Vivat B.
• et al.
Quality of life measures for the palliative care of people severely affected by multiple sclerosis: a systematic review.
Mult Scler. 2004; 10: 690-704
• Opara J.A.
• Jaracz K.
• Brola W.
Quality of life in multiple sclerosis.
J Med Life. 2010; 3: 352-358
• Goodwin E.
• Green C.
A QALY measure for multiple sclerosis: developing a patient-reported health state classification system for an MS-specific preference-based measure.
Value Health. 2015; 18: 1016-1024
• Hobart J.C.
• Riazi A.
• Lamping D.L.
• et al.
Improving the evaluation of therapeutic interventions in multiple sclerosis: development of a patient-based measure of outcome.
Health Technol Assess. 2004; 8: 1-60
• Gudex C.
Time trade-off user manual: props and self-completion methods.
The MVH Group, Centre for Health Economics, University of York, York1994 (Available from: http://www.york.ac.uk/che/pdf/op20.pdf. [Accessed October 23, 2017])
• Dolan P.
Modeling valuations for EuroQol health states.
Med Care. 1997; 35: 1095-1108
• Green C.
• Goodwin E.
• Hawton A.
“Naming and framing”: the impact of labeling on health state values for multiple sclerosis.
Med Decis Making. 2017; 37: 703-714
2. MS Register website. Available from: https://www.ukmsregister.org/Portal/Home#about. [Accessed July 18, 2016].

• Ford D.V.
• Jones K.H.
• Middleton R.M.
• et al.
The feasibility of collecting information from people with multiple sclerosis or the UK MS Register via a web portal: characterising a cohort of people with MS.
BMC Med Inform Decis Mak. 2012; 12: 1-8
• Mackenzie I.S.
• Morant S.V.
• Bloomfield G.A.
• et al.
Incidence and prevalence of multiple sclerosis in the UK 1990–2010: a descriptive study in the General Practice Research Database.
J Neurol Neurosurg Psychiatry. 2014; 85: 76-84
• Goodwin E.
• Green C.
A systematic review of the literature on the development of condition-specific preference-based measures of health.
Appl Health Econ Health Policy. 2016; 14: 161-183
• Norman R.
• King M.T.
• Clarke D.
• et al.
Qual Life Res. 2010; 19: 499-508
• Brazier J.
• Ratcliffe J.
• Salomon J.A.
• Tsuchiya A.
Measuring and Valuing Health for Economic Evaluation.
Oxford University Press, Oxford2007
• Versteegh M.M.
• Leunis A.
• Uyl-de Groot C.A.
• Stolk E.A.
Condition-specific preference-based measures: benefit or burden?.
Value Health. 2012; 15: 504-513
• Hawton A.
• Green C.
Health utilities for multiple sclerosis.
Value Health. 2016; 19: 460-468
• Benito-León J.
• Morales J.M.
• Rivera-Navarro J.
Health-related quality of life and its relationship to cognitive and emotional functioning in multiple sclerosis patients.
Eur J Neurol. 2002; 9: 497-502
• Kind P.
A revised protocol for the valuation of health states defined by the EQ-5D-3L classification system: learning the lessons from the MVH study.
Centre for Health Economics, University of York, York2009
• Greene W.H.
Econometric Analysis. 7th ed. Prentice Hall, Boston, MA2012
• Brazier J.
• Akehurst R.
• Brennan A.
• et al.
Should patients have a greater role in valuing health states?.
Appl Health Econ Health Policy. 2005; 4: 201-208
• Dolan P.
Whose preferences count?.
Med Decis Making. 1999; 19: 482-486