Integrating Patient Preferences Into Health Outcomes Assessment

Integrating Patient Preferences Into Health Outcomes Assessment

Integrating Patient Preferences Into Health Outcomes Assessment* The Multiattribute Asthma Symptom Utility Index Dennis A. Revicki, PhD; Nancy Kline L...

5MB Sizes 0 Downloads 7 Views

Integrating Patient Preferences Into Health Outcomes Assessment* The Multiattribute Asthma Symptom Utility Index Dennis A. Revicki, PhD; Nancy Kline Leidy, PhD; Fiona Brennan-Diemer, BA; Sonja Sorensen, MPH; and Alkis Togias, MD

Study objective: To develop and evaluate a brief, easy-to-administer symptom assessment scale for use as a preference-based outcome measure in clinical trials and cost-effectiveness studies in asthma. Design: Cross-sectional survey with 2-week reproducibility assessment. Setting: Ambulatory care: university asthma and allergy center. Participants: One hundred sixty-one adults with asthma, 59% female, mean age 35 ± 11 years. Mean FEV 1 percent predicted was 86 ± 17%. Interventions: The 11-item Asthma Symptom Utility Index (ASUI). Measurements and results: Mean ASUI score for this sample was 0. 71 ± 0.23, with a range from 0.02 to 1.0. The ASUI was reproducible (intraclass correlation coefficient = 0. 74) and able to distinguish patients known to differ on disease severity according to clinician ratings (p < 0.001) and by an asthma disease severity scale score (p < 0.001). The instrument was also significantly correlated with FEV 1 percent predicted (r = 0.27, p < 0.001), the Asthma Quality of Life Questionnaire (r 0.77, p < 0.001), and the Health Utilities Index Mark 2 (r 0.36, p < 0.001). Conclusion: The results of this study support the reliability and validity of the ASUI, suggesting it will be a useful, complementary patient outcome measure for clinical trials and costeffectiveness studies comparing treatment alternatives for persons with asthma. (CHEST 1998; 114:998-1007)



Key words: asthma-related symptoms; health preferences; quality of life; patient outcomes

Abbreviations: ADSS == Asthma Disease Severity Scale; ANCOVA == analysis of covariance; AQLQ == Asthma Quality of Life Questionnaire; ASUI == Asthma Symptom Utility Index; HRQL == health-related quality of life; HUI == Health Utilities Index; PSRS == Physician Severity Rating Scale; QALY == quality-adjusted life year; SALY == symptom-adjusted life year; SG == standard gamble; VAS == visual analog scale


oncern about limited resources and the high costs associated with new as well as existing medical treatments has placed greater emphasis on the evaluation of the cost-effectiveness of treatment alternatives. To be relevant to providers of care, "effectiveness" in economic evaluation must be clinically meaningful, that is, reflect important characteristics of the disease, be responsive to treatment effects, and represent aspects of health important to *From the MEDTAP International, Inc (Drs. Revick:i and Leidy, and Ms. Sorensen), Bethesda, MD, and Johns Hopkins Asthma and Allergy Center (Ms. Brennan-Diemer and Dr. Togias), Baltimore, MD. This study was funded by SmithKiine Beecham and in part by NIH grant HlA9545. Manuscript received November 3, 1997; revision accepted May 7, 1998. Correspondence to: Dennis Revicki, PhD, Center for Health Outcomes Research, MEDTAP International, Inc, 7101 Wisconsin Ave, Suite 600, Bethesda, MD 20814, e-mail: [email protected] 998

the patients themselves. To be suitable for modeling, the indicator must accurately and directly represent the appropriate health outcome, have the necessary temporal characteristics, and be interpretable within and across treatment groups. The purpose of this article is to propose a new approach for evaluating outcomes in clinical trials and cost -effectiveness studies of asthma. BACKGROUND

There is general consensus that the evaluation of treatment outcomes in clinical trials of asthma should include patient perceptions of the frequency and severity of key symptoms, the adverse effects of treatment, and the impact of disease and treatment on daily life, that is, the patient's health-related quality of life (HRQL). Instruments designed to assess HRQL have been used successfully in trials of Clinical Investigations

pharmacologic and behavioral therapies. 1- 16 While these instruments are helpful for understanding the impact of disease and treatment on HRQL, they are less useful for cost-effectiveness analyses, primarily because change scores are difficult to interpret relative to therapy costsJ7 Further complications arise when an HRQL measure generates a profile of scores, necessitating multiple cost-effectiveness ratios that can yield conflicting results. Because of these difficulties, the episode- or symptom-free day has been proposed as a measure of effectiveness. Models using this indicator compare treatment costs per episode or symptom-free day.1 8 While this appears to be the best alternative currently available, it is a relatively crude approach for evaluating outcomes, particularly in patients with severe disease. For example, successful treatment leading to clinically significant improvements in symptom frequency and severity but minimal to no impact on symptomjree days would not be captured with this method. It is likely that certain symptoms and side effects are more troubling to patients than others, and that patients prefer treatments that are more effective in reducing the severity or frequency of those symptom, with minimal undesirable side effects. The symptom-free day does not reflect these preferences. Utility or health preference scales are a better alternative for evaluating outcomes in cost-effectiveness analysis. Generally, these measures evaluate patient preferences for multidimensional health states on a 0- to 1-point scale, where 0 represents death and 1 represents complete or perfect health. 19•20 Methods for gathering utilities include direct measures (eg, standard gamble [SG], time trade-off, or visual analog scales [VAS]), or established questionnaires such as the Health Utilities Index (HUI), EuroQol, and the Quality of WellBeing Scale with empirically based weighting schemes to reflect patient preferences. 21 •22 Time spent in a specific health state can be weighted by the utility score for the health state to yield estimates of quality-adjusted life years (QALYs). Costs associated with treatment can thus be expressed as cost per QALY. Although the Quality of Well-Being Scale has been used to estimate utilities in patients with COPD, relatively few studies have gathered these types of data from patients with asthma. Results to date indicate these patients assign relatively high utility values to their health state and respond within a fairly narrow range. Rotten-van Molken et al, 12 for example, reported baseline SG scores of 0.87 (±0.13) and rating scale scores of 0.70 (±0.14) in their study of patients with moderate asthma (FEV 1 percent predicted= 59%). Kerridge and col-

leagues 23 found that patients with asthma had an average quality of life value (Rosser index) of 0.93. Given that many asthma patients have mild-tomoderate disease, utility scores, whether directly measured or derived from multiattribute instruments, may not be sufficiently sensitive to the burden of asthma. 12 Cost-effectiveness studies in asthma may require a condition-specific utility index with optimal sensitivity to the symptomatic experiences of the disease and the efficacy and side effects of treatment. For example, the evaluation of alternative treatments for mild-to-moderate asthma requires effectiveness measures that are less blunt than symptom-free days or utility scores . The ideal outcome would be an index that assesses asthma symptoms, symptom characteristics, and side effects that patients find most distressing and practitioners target for intervention. The index will yield an easily interpreted, intervallevel scale reflecting the true range of symptomatic experience, from worst state imaginable (O) to the best state imaginable (1). Unlike the symptom-free day, however, the index would reflect symptom gradations (ie, variations in type, frequency, and severity of symptoms) as well as side effects, weighted according to patient preferences. Effectiveness could then be discussed in terms of symptom-adjusted life years (SALYs), providing similar metric advantages of a QALY, but offering greater sensitivity to the symptom variation associated with asthma. The purpose of this study was to develop and test such an instrument, the Asthma-Symptom Utility Index (ASUI), to complement generic and asthmaspecific HRQL measures in traditional trial settings and provide a useful metric for evaluating the costeffectiveness of treatment alternatives for patients with asthma. MATERIALS AND METHODS

Sample One hundred sixty-one adults, 95 women and 66 men, with asthma receiving treatment at the Johns Hopkins University Asthma and Allergy Center participated in the study. Patients were recruited for this study from the population of patients in receiving asthma-related services in the clinical center. Consecutive patients were approached as they visited the clinic for regular appointments and clinic patients with a history of asthma were contacted by telephone about the study. The intent was to identifY patients with a broad range in asthma disease sevetity. Most patients approached agreed to participate in the study (76%). Mean age of the sample was 34.7 years (SO = 10.7). Most patients (79%) were white, with 17% African-American and 3% Asian. The sample was well-educated, 34% completed some college education, and 46% completed an undergraduate or CHEST I 114 I 4 I OCTOBER, 1998


graduate college education. Seventy percent of patients were working full time. Mean duration of illness for the sample was 17.3 y ears (SD = 11.22) with an average age of onset at 17.7 years (SD = 13.3). Sixty-nine percent had a family history of asthm a. M ean FEV 1 percent predicted was 85.6% (SD = 17.1). A significant number were only using bronchodilators (45%) or a bronchodilator/inhaled steroid combination (45%) . T en patients ( 6%) indicated they were using no drug therapy at the time of the study. A random subsample of 25 patients were inte rviewed 2 weeks afte r the initial interview to evaluate test-retest r eliability of the ASUI.

Asthma-Symptom Utility Index The four symptoms (cough, wheeze, shortness of breath, and awakening at night) and two dimensions (frequency and severity) comprising the ASUI were identified through clinical practice, a review of the literature, patient interviews, and discussion with physicians regarding symptoms of primary concern in practice and the evaluation of treatment effectiveness. Qualitative data concerning the patient's perspective of symptoms and side effects were gathered through interviews with 10 patients with varying severity of asthma. The interviews consisted of open-ended questions concentrating on the identification of asthm a-related symptoms and problems that were troublesome and distressing to patients. Patients were asked to rank order the relative importance, in terms of their functioning and well-being, of the symptoms illicited. We continued conducting patient interviews until no new information was generated. A oc ntent analysis was done to identify symptoms that were most important and distressing from the patient perspective. The ASUI is represented b y 11 items. Frequency and severity of each symptom are measured on 4-point Likert scales: not at all, 1 to 3 days, 4 to 7 days, and 8 to 14 days during a 14-day period for frequency, and not applicable, mild, moderate, and severe for severity. Two ite ms address the frequency and severity of medication side effects. One open-ended item is a lsoincluded in the measure, asking patients to list adverse effects of asthma treatment. Responses to this item s erve as qualitative anchors for the two items addressing frequency and severity of side effects but do not contribute to the scoring of the ASUI. The entire

questionnaire is reproduced in Appendix l. Procedures for deriving th e preference weights and ASUI scores aresummarized below.

Derivation of ASUI Scoring Algorithm Patient Pref erence Measures: The preference-weighting scheme used to compute ASUI scores was b asedon multiattribute utility assessment methods. 24 - 2 " Utilities are measures of the strength of an individual's preference for a particular h ealth outcome or state under conditions of uncertainty 1 9 - 2 1 .27,2R Preferences represent the individual's subjective rating when uncertainty is not a condition of measurement. Several methods are available to derive utilities and preferences and the most commonly used are the VAS , SG, and time trade-off.l 9 ·21 •27 -2B For this study, a ocmbination of VAS and SG techniques was used. Health states w ereconstructed that r eflected single or multiple asthma symptoms of different frequency and severity. Five multiple symptom states were developed in which each symptom was individually described at the worst possible level of severity and fi·equency and the other four remaining symptoms were described a s absent. Five multiple symptom states describing selected combinations of severity and frequency of the different symptoms were constructed to represent the range of symptom combinations often experienced b y persons with asthma. The corner and multiple symptom states are presented in Table l. Visual Analog Scale: The VAS evaluates preferences under certainty, providing data on the value associated with v raious health states. Briefly, subj ects were asked to review a s eries of health states and identifY a value for each along a fixed line in an order consistent with their preference ranking, ie, relative to one another and spaced in agreement with their relative preference among states, such that the distances are proportional to preference differences 2 1 A visual prop, called the Feeling Thermometer, was used t o facilitate the ranking task. 28 This is a vertical thermometer-shaped scale, 55 em long, and numerically scaled in units from 0 to 100. The numeric value adjacent to the placement of the health state is the assigned preference score. For the development of the ASUI, health states were limited to symptom atttibutes individually or in combination. The VAS was used to determine patient preferences for the 5 individual

Table 1-Mean VAS Preferences, SG Utilities, and ASUI-Derived Utilities for Multi-symptom States State Corner states* Severe cough Severe w heeze Severe dyspnea Severe awaken at night Severe medication side effects Multisymptom states t Moderate cough and dyspnea (1-3 d) Moderate cough and wheeze (4-7 d) Severe cough; moderate wheeze and dyspnea (1-3 d) Severe cough; moderate w heeze, dyspnea, and awaken at night (1- 3 d ) Severe cough, dyspnea, and awaken at night; moderate wheeze and side e fects (1-3 d)

VAS t Mean (SD)

SG Utility t Mean (SD)

ASUI § Mean

0.26 (025) 0.24 (025) 0.16 (0.21 ) 0.25 (025) 0.25 (0.26)

0.69 (0.21) 0.66 (0.21) 0.60 (0.25) 0.67 (0.25) 0.66 (023)

0.70 0.67 0.61 0.68 0.67

0.31 (025) 0.22 (0.20) 0.20 (0.20) 0.17 (0.18)

0.67 (0.23) 0.62 (0.24) 0.60 (0.21) 0.59 (0.24)

0.73 0.56 0.54 0.50

0.08 (0.12)

0.46 (0.27)


*One symptom is described assevere for 8 to 14 days and remaining symptoms are not present. fFor multisymptom states if a symptom is not mentioned, it is described a s mild in the health state. tThe VAS and SG utility scores are on a 0 to 1 scale with higher scores indicating better h ealth. §ASUI-derived utility score based on multiattribute utili ty fun ctions (see text for details). 1000

Clinical Investigations

symptoms, by frequency and severity, and 10 multiattribute symptom states. For example, for cough, patients were asked to assign preferences for the different frequency-severity combinations (ie, moderate cough 4 to 7 days, severe cough 1 to 3 days) on the VAS where 100 was anchored as no cough and 0 was anchored as severe cough 8 to 14 days. The corner states and the multiple symptom states were assigned preferences on the VAS where 100 was anchored as no cough, wheeze, dyspnea, sleep problems, or medication side effects (ie, best possible state) and 0 was anchored as severe cough, wheeze, dyspnea, sleep problems, and medication side effects for 8 to 14 days (ie, worst possible state). Standard Gamble: The SG method for deriving utilities is grounded in von Neumann-Morgenstern's model for describing rational decision-making under conditions of uncertainty and it involves the evaluation of health states under conditions of risk. 19 ·21 Briefly, subjects were presented with two alternatives: A and B. A was an uncertain choice with two alternatives, in this case the worst symptom state or the best symptom state; B was a certain choice, the symptom state under consideration with a designated time frame of 2 weeks. Generally, the less desirable the state under consideration, the greater the willingness to take a chance to escape the state, the lower the indifference probability, and the lower the utility score for that state. 19 A probability wheel was used to help subjects choose between alternatives, with color-coded sectors representing the health state under consideration. The size of the sectors varies according to changes in the gamble probabilities. Chance p was varied systematically in increments of 10% until the respondent was indifferent between options A and B. The indifference point represented the state's utility for the patient. Patients were asked to rate each of the 10 multiple symptom states using the SG method. Therefore, there were 10 multiple symptom states with both VAS and SG scores. Multi-Attribute Utility Function: Multiattribute utility theory is a method for determining a mathematical formula that enables the estimation of utility or preference scores for a number of health states defined in a multiattribute classification system, based on measured preference scores for a subset of these states. 26 In this case, the classification system is made up of five dimensions or attributes: cough, wheeze, dyspnea, awakening at night, and side effects of asthma medications. The measured levels for each attribute range from the most severe and frequent symptoms to the absence of these symptoms. Mean VAS preference and SG utility scores for the five corner states and the five multiple symptom states are summarized in Table l. For the

corner states, severe dyspnea was rated as least desirable followed by wheeze, awakened at night, and cough. For the multisymptom states, states describing more severe symptoms, such as severe cough, dyspnea, and awakening at night, were rated as less desirable. We used multiattribute utility theory to construct a model for calculating ASUI utilities from individual patient responses. The analysis of the VAS preferences and SG utilities and the development of the ASUI multiattribute utility functions followed the techniques outlined by Torrance. 24 ·26 The multiattribute utility function was based on a multiplicative model. A multiattribute utility function was constructed for combining ratings within and between symptoms. Derivation of the utility function is detailed in Appendix 2. The multiattribute utility function is ASUI = 1.200 X (s 1 X s2 X s3 X s4 X s5 ) - 0.200 where ASUI is the utility of the symptom state on a scale where the best state (no symptoms) has a score of 1 and worst possible symptoms (severe symptoms for 8 to 14 days) has a score of 0; and s1 is the score for the level on symptom l. Table 2 contains the coefficients for calculating ASUI scores. For example, the ASUI score for a person classified as level3 on cough (S 1 ), level4 on wheeze (S 2 ), level 2 on dyspnea (S 3 ), level 3 on awaken at night (S 4 ), and level 2 on medication side effects (S 5 ) is 0.701 (1.200(0.963 X 0.913 X 0.946 X 0.931 X 0.970]) - 0.200). Clinical and Health-Related Validation Measures The concurrent and construct validity of the ASUI was evaluated by examining the relationship between the ASUI and indicators of disease severity, generic health utility, and asthmaspecific HRQL. Disease Se1/erity: Pulmonary function data (FEV 1 percent predicted and FEV /FVC) were gathered by trained personnel using conventional spirometric methods. All patients were asked to refrain from inhaled and oral bronchodilation prior to the visit. Patients also responded to an asthma-specific medical history instrument developed at the Johns Hopkins University Asthma and Allergy Center that captures information on asthma severity, seasonality and temporal pattern of symptoms, environmental exposures, known triggers, allergic history, family history, and current medications.3o Two multi-dimensional rating schemes were used to categmize patients by disease severity: the Physician Severity Rating Scale (PSRS) and the Asthma Disease Severity Scale (ADSS). The

Table 2-Multiattribute Utility Function on Worst Possible Symptom State to No Symptoms Scale for ASUI* Symptom (Attribute) Cough



Level, d


1 None 2 Mild, 1-3 3 Mild, 4-7 4 Mild, 8-14 5 Moderate, 1-3 6 Moderate, 4-7 7 Moderate, 8-14 8 Severe, 1-3 9 Severe, 4--7 10 Severe, 8-14



0.985 0.963 0.935 0.955 0.920 0.875 0.863 0.813 0.751

0.962 0.940 0.913 0.913 0.886 0.851 0.810 0.772 0.729


Awaken at Night S4

1.0 0.946 0.920 0.885 0.892 0.860 0.818 0.771 0.729 0.681

0.955 0.931 0.899 0.909 0.880 0.845 0.821 0.781 0.734



Medication Side Effects S5 1.0 0.970 0.954 0.930 0.924 0.900 0.862 0.824 0.789 0.730

*Calculating ASUI scores is as follows: ASUI = 1.200 X (S 1 X S2 X S3 x S4 X S5 ) - 0.200. For example, if a person is classified as level 3 on cough (S 1 ), level 4 on wheeze (S 2 ), level 2 on dyspnea (S 3 ), level 3 on awaken at night (S 4 ), and level 2 on medication side effects (S.5 ), his or her ASUI score equals (1.200 [0.963 X 0.913 X 0.946 X 0.931 X 0.970] - 0.200) or 0.701. CHEST I 114 I 4 I OCTOBER, 1998


PSRS is a physician global assessment of the severity of asthma on a scale of 1 (mild ) to 6 (severe) based on the patient's pulmona1y function tests and medical and symptom history information from th e asthma medical history instrument. Previous research has demonstrated the reliability and validity of PSRS. 30 Patients were classified into one of four groups based on the PSRS . The sample distribution was as follows: mild (n = 41); mild/moderate (n = 50); moderate ( n = 43); and moderate/severe ( n = 24). The ADSS is based on the Asthma Control Scale developed by Juniper et aJ.3 The ADSS is a composite of resource use, spirometry, and symptoms, including emergency dep artment visits during the past 12 months (2:: 1); hospitalizations dming the last 12 months (2:: 1); FEV 1 percent predicted :570%; chronic cough or chronic phlegm; chronic wheeze; chronic breathlessness; and chroni c nighttim e symptoms. Each positive item is assigned a score of one. ADSS scores could range from 0 to 7. Patients were assigned to one of four groups based upon ADSS score. The sample was distributed as follows: mild (0) ( n = 49); mild/moderate ( 1) ( n= 39); moderate ( 2,3) (n = 48); and moderate/seve re (2::4) (n = 24). As expected, there was a significant but imperfect r elationship between these two indicators of disease severity (x 2 = 66.67, p < 0.01 ). Because th ere is a varian ce in th e mann er in whi ch di sease severity is determin ed in asthma, we believed it important to test th e ASUI's sensitivity unde r more than one rating scheme. Generic Health Utility: The HUI Mark 2 (HUI2) was used as a generic utility measure to validate th e ASUI.24 ·29 The HUI2 is a multi -attribute HUI that contains dim ensions of sensory ability (ie, vision, hearing, speech), mobility, emotional function , selfcare, cognitive function , and pain and discomfort. A multiattribute utility function is used to combine the levels on each dim ension into a summmy utility score. 24 The mean HUI2 score in this sample was 0.84 (SD = 0.17). Disease-Specific Quality of Life: The Asthma Quality of Life Questionn aire (AQLQ) was selected to measure HRQL related to asthmaa· 7 The AQLQ was designed as a disease-specific measure of health status in persons with asthma and evaluates activity limitations, symptoms, emotional function , and environmental exposure. Response options are on a seven-point scale, with 1 = maximal impairment and 7 = no impairment. Scores are e;.pressed as the mean score per ite m for each subscale, with an overall quality of life score (ie, mean of all AQLQ items). Highe r scores indicate better quality of life. The AQLQ has good evidence of tes t-retest r eliability, construct validity, and responsiveness to change. 3 ·6 Internal consistency reliabili ty (Cronbach 's alpha) for th e four subscales ranged from 0.80 to 0.93 and internal consistency reliability for th e total score in this sample was 0.97. Intraclass correlations were high, ranging from 0.81 to 0.93. Mean AQLQ total score in our sample was 5.24 (S D = 1.18).

Data Analysis Intraclass correlation coefficients were used to estimate the reproducibility (test-retest reliability) of the ASUI, supplemented by paired t tests. Analysis of vmiance procedures were used to test th e ability of the ASUI score to discriminate among levels of disease severity, measured by the PSRS and ADSS. A statistically significant F-statistic was followed by post hoc comparisons (Scheffe) to identify the source of group differences in mean ASUI scores. Pearson correlation coefllcients were also used to describe th e relationship between th e ASUI and FEV 1 percent predicted, HUI2 scores, and AQLQ subscales and total scale scores. Finally, mean ASUI scores were examined for sociodemographic effects (ie, age, gender, education) using t tests for independent groups. Analysis of covariance (ANCOVA), adjusting for disease sevelity, compared mean ASUI scores by education level.


The distribution of ASUI scores is shown in Figure l. For comparative purposes, the distribution of

HUI2 scores is also displayed. Scores on the ASUI ranged from 0.04 to 1.0. Mean ASUI score was 0.71 (SD = 0.23) ; median score was 0.76; and the mode was 0.89 (n = 7). The distribution was relatively flat with a kurtosis of -0.37. In contrast, mean HUI2 score was 0.84, ranging from 0.17 to 1.0. The median score was 0.90 with a mode of 1 (n = 31). The distribution was peaked with a kurtosis of 3.28. The intraclass correlation coefficient, describing the 2-week reproducibility of the ASUI, was 0.74,


x= .71 SO= .23 Md= .76 Mode= .89









ASUI Score

Data Collection Procedures All data were collected b y trained inte rviewers during face-toface interviews. The intervi ews were divided into three parts: part 1 included the ASUI questions, the HUI, the AQLQ, and questions on demographic characteristics. Part 2 consisted of th e VAS and SG preference methods. Pmt 3 consisted of the pu lmonary function tes ts and detailed m edical and symptom histmy. The entire interview took 1 to 1lf2 h to complete . All patients completed the three-part inte 1views. Patients were paid $25 per inte1view to compensate th e m for their tim e and parking costs. To examin e the reproducibility of th e ASUI, 25 subjects return ed for a econd s inte1view, completing part 1 only. The study was approved by th e John s Hopkins University Bayview Medical Center Institutional Review Board and all p ati ents gave informed consent prior to th e onset of the inte1v iew. 1002


n • 74


x=.s4 SD= . I7 Md= .90 Mode= 1.0










HUI Score

FIGURE 1. Distributional characteristics of the ASUI and th e HUI. Clinical Investigations

and the Pearson correlation coefficient was 0.75 (p < 0.001 ). Mean (M)scores between the two observations (n = 25) were not significantly different (Mn = 0.71 :±:: 0.26; MT2 = 0.66 :±:: 0.26; Mclur = 0.05 (se = 0.04), t = 1.40, p = 0.17). The direct SG utility scores and ASUI-derived utility scores were compared for the multi-symptom states (Table 1) . The differences between the ASUIderived and SG utilities for the corner states were small, ranging from 0.02 to 0.03. For the other multisymptom states, the differences were somewhat larger, ranging from 0.06 to 0.15 points. The ASUI was significantly correlated with FEV 1 percent predicted (r = 0.27, p < 0.01 ), as well as with FEV/FVC (r = 0.27, p < 0.001 ). In contrast, correlations between pulmonary function and the AQLQ total were 0.15 and 0.08, respectively; correlations with the AQLQ symptom scores were 0.21 (p < 0.01 ) and 0.13, respectively. The HUI2 was unrelated to pulmonary function (r = 0.12 and r = 0.05, respectively) . Figure 2 displays mean ASUI scores by asthma seve1ity groups using the two rating schemes. The

ASUI discriminated between asthma severity groups under both methods: the PSRS (F 3 154 = 27.64, p < 0.001 , R2 = 0.35) and ADSS (F 3 : 1s 6 = 34.98, p < 0.001, R2 = 0.40 ). Post hoc Scheffe tests indicated statistically signiflcant differences between all pairs (p < 0.05), \.vith the exception of the comparison between the mild and mild/moderate groups. For comparative purposes, mean HUI2 scores by disease severity are shown in Figure 3. This generic utility measure also discriminated between asthma severity groups under both rating schemes: the PSRS (F 3 , 151 = 5.91 , p < 0.001, R2 = 0.11) and ADSS (F 3 , 153 = 6.28, p < 0.001, R2 = 0.11). Post hoc analyses indicated the measure could not differentiate mild from either mild/moderate or moderate or mild/moderate from moderate. The Pearson correlation coefficient between the ASUI and the HUI2 was 0.36 (p < 0.001). Correlations between the ASUI and the four subscales and total scale scores of th e AQLQ were as follows: activity limitations 0.59; symptoms 0.85; emotional function 0.63; environmen tal exposure 0.60; and total score 0.77, all statistically significant (p < 0.001) . For comparative purposes, correlations between the HUI2 and the four subscales and total

ASUI Score HUl Score

Mild n =41

Mild/Modcmte n - 50


n =43

Modemte/Sevcre n •24

Disease Severity: Physician Severity Rating Scale (PSRS) F=27.64, p<.OO!, R 2=.35

Disease Severity: Physician Severity Rating Scale (PSRS) F=S.91, p<.OOI, R 2=. 11

ASUl Score

HUI Score


n =39

- n =48

n =24

Disease Severity: Asthma Disease Severity Scale (ADSS) F=3 4.98, p<. OO!, R 2=.40

Disease Severity: Asthma Disease Severity Scale (ADSS) F=6.28, p<.OOI, R 2=. 11 I "" perfect heailh O= death

I "" best possible state 0 =worst possible state


2. Mean ASUI score by disease severity.


3. Mean HUI score by disease severity. CHEST / 114 I 41 OCTOBER, 1998


scale scores of the AQLQ were as follows: activity limitations 0.60; symptoms 0.51 ; emotional function 0.40; environmental exposure 0.48; and total score 0.57, all statistically significant (p < 0.001 ). No relationship was found between ASUI score and age (r = 0.18), and there was no difference in mean score between men (Mm = 0.72 :±: 0.24) and women (Mw= 0.70 :±: 0.22) (t = 0.71 , p = 0.48 ). Education effects were significant, however (F 2 ,158 = 5.92, p < 0.01 ), with significant differences between those \vith a high school education or less and those with a college or postgraduate education. Mean ASUI scores were lower for patients with a high school education or less (0.63 :±: 0.27) than those with some college education (0.71 :±: 0.22) or college graduates (0.76 :±: 0.19). Further analyses revealed a significant relationship between educational l evel and disease severity (PSRS , x2 = 16.9, p < 0.05; ADSS, l = 23.3, p < 0.01 ). Patients with less education were more likely to have worse asthma severity. The findings from an ANCOVA, controlling for ADSS , indicated no statistically significant impact of education level on mean ASUI scores (F 2 , 154 = 0.63, p = 0.535). The ANCOVA, controlling for PSRS scores, also showed no statistically significant difference for education on mean ASUI scores (F 2 , 154 = 1.54, p = 0.218).


The purpose of this study was to evaluate the qualities of the ASUI. The 11-item index was designed to summarize the frequency and severity of selected asthma-related symptoms and side effects reported during a 2-week p eriod , weighted according to patient preferences. The combination of the five symptoms and ratings of frequency and severity yields a wide variety of multidimensional symptom states, presenting an opportunity for evaluating multiple symptom combinations, concentrating on those most applicable to the treatment under investigation. The ASUI displayed evidence of reliability and validity a s anoutcome measure for clinical trials and cost-effectiveness studies of asthma. As designed, the patient scores were well distributed across the 0 to 1.0 scale. In contrast, the range of scores on the HUI2 was relatively narrow, with a substantial number of the patients scoring a p refect 1.0. Mean HUI2-derived utility score of 0.84(:±:0.17) in this sample was remarkably similar to the SG score of 0.87 (:±:0.13) reported by Rutten-van Molken et al.l 2 AS UI scores were relatively stable and reproducible over a 2-week time period. The ASUI was significantly correlated with FEV 1 percent predicted and FEV / FVC. In addition, the 1004

measure was capable of differentiating patient groups of known disease severity. Similar results were demonstrated using physician-generated ratings of severity of asthma and an asthma disease severity rating system b ased on emergency medical service use, spirometiy, and symptoms. ASUI scores produced a stair-step pattern of decreasing scores with increasing disease severity. Patients categorized as having mild disease according to the physician rating scheme scored the highest on the ASUI at 0.88 and were distinguishable from the groups with moderate and moderate-to-severe disease by 0.24 and 0.41 points, respectively. Patients with mild! moderate, moderate, and moderate/severe disease were also distinguishable with a value of 0.30 separating the mild/moderate from the moderate/severe . This suggests the instrument is sensitive to subtle differences in disease severity, a characteristic necessary for responsiveness to change with treatment and meaningful cost-effectiveness studies. Further research is needed to quantify the responsiveness of the ASUI to clinically meaningful change. As expected, the HUI2 generic utility measure was less sensitive to differences in disease severity. The instrument was unable to differentiate among patients with mild-to-moderate disease and mean HUI2 scores of patients with mild/moderate and moderate disease were identical at 0.86. A value of 0.13 separated these patients from those with moderate-to-severe asthma; 0.16 separated the mild and the moderate-to-severe groups. From a discriminant validity p erspective, the ASUI appears to be superior to the HUI2 in its potential for sensitivity to treatment effects . However, it should be noted that when comparing HUI2 and ASUI scores, HUI2 is relative to death and the ASUI is relative to the worst asthma-related health state. The validity of the ASUI was also supported by strong, statistically significant correlations with the AQLQ, a condition-specific indicator of health-related quality of life. These correlations ranged from 0.59 to 0.85, indicating that the ASUI shares 35 to 72% of variance with the AQLQ scales. As expected, the largest correlation was b etween tl1e ASUI and the AQLQ symptom scale ( r = 0.85), suggesting that the two scores are measuring similar concepts. As intended, the two measures appear to be complementary. The advantage of the ASUI for cost-effectiveness studies lies in its preference-based weighting scheme and interval-level m etric. The ASUIderived utilities w ere c om parable to direct SG utilities for the multisymptom states, providing additional evidence supporting the validity of the ASUI scores . Two caveats should be considered in interpreting the results of this study. First, one must keep in mind Clinical Investigations

that the measure is unidimensional and therefore valid in situations in which reduction in symptom frequency and intensity is the primary clinical outcome. The instrument is not intended as an indicator of quality of life or as a tool for the calculation of QALYs . Trials evaluating the effectiveness of interventions in terms of health-related quality of life require a comprehensive index that includes assessment of the physical functioning, social functioning, and psychological well-being domains that are characteristic of HRQL. The ASUI offers a domainspecific outcome measure with the advantages of an interval level metric. Second, a limitation of any multiattribute utility system is the generalizability of the weights to other cultures. The sample in this study was drawn from an asthma and allergy center and the subjects were relatively well educated. Further research is needed to determine the stability of the weights across socioeconomic, ethnic, and cultural groups. The purpose of this study was to test the feasibility of a multiattribute symptom utility index and open methodologic dialogue concerning the usefulness of such as measure for evaluating outcomes in asthma. The results of this study suggest that ASUI scores are unrelated to age, gender, and education level. ASUI score and age were independent and there were no differences between mean scores of men and women. The relationship observed between ASUI score and education can be attributed to differences in disease severity. As suggested above, the sample consisted of relatively well-educated and working adults with asthma and may not be representative of all patients with asthma. We need to evaluate the stability of the weights and the reliability and validity of the ASUI with patients from lower socioeconomic status populations and other ethnic and cultural backgrounds. The appropriateness of the ASUI for assessing symptoms in children with asthma also needs to be evaluated. This new outcome measure represents patient preferences for combinations of asthma-related symptoms and side effects on a scale from worst possible state to best possible state. The ASUI is not a health state utility score on the death to complete health scale, as is the HUI2. 24 •26 Therefore, SALYs based on ASUI scores are not directly comparable to QALYs derived from SG or generic multiattribute utility scores. As suggested previously, placing ratings of milder symptoms that are relatively infrequent on a death to complete health scale would constrain ASUI scores to the upper range of the scale and compromise discriminative ability and clinical responsiveness. The ASUI is intended as an outcome indicator for clinical trials and decision analytic studies comparing different symptom-targeted inter-

ventions for asthma, with cost-effectiveness outcomes expressed in ASUI score and symptomadjusted life years or SALYs. The clinical and economic evaluation of alternatives for the treatment of asthma requires the careful assessment of clinical outcomes, symptoms, HRQL, and medical resource use and costs. The results of this study suggest the ASUI is a complementary outcome for use in clinical trials and a viable alternative for cost-effectiveness studies comparing medical treatments for asthma. ACKNOWLEDGMENTS: The assistance of Linda deFranco with interview training and coordination, and Christine Thompson with data cleaning, management, and SAS programming is greatly appreciated. We also appreciate the review and helpful comments of Michael Halpern, MD, PhD, Randel Richner, MPH, and Jennifer Ehreth, PhD, on earlier drafts of this manuscript. APPENDIX


Asthma Symptom Utility Index I would like to ask you some questions about different symptoms of asthma and how often you were bothered by these symptoms in the past 2 weeks. l. How many days were you bothered by coughing during the past 2 weeks? 0 Not at all (skip to question 3) 1 1-3 days 2 4-7 days 3 8-14 days 2. On average, how severe was your coughing during the past 2 weeks? 1 Mild 2 Moderate 3 Severe 3. How many days were you bothered hy wheezing during the past 2 weeks? 0 Not at all (skip to question 5) 1 1-3 days 2 4-7 days 3 8-14 days 4. On average, how severe was your wheezing during the past 2 weeks? 1 Mild 2 Moderate 3 Severe 5. How many days were you bothered by shortness of breath during the past 2 weeks? 0 Not at all (skip to question 7) 1 1-3 days 2 4-7 days 3 8-14 days 6. On average, how severe was your shortness of breath during the past 2 weeks? 1 Mild 2 Moderate 3 Severe 7. How many days were you awakened at night during th e past 2 weeks? 0 Not at all (skip to question 9) l l-3 days 2 4-7 days 3 8-14 days CHEST f 114 f 4 f OCTOBER, 1998


8. On average, how much of a problem was being awakened at night during the past 2 weeks? 1 Mild 2 Moderate 3 Severe 9. How many days were you bothered by side effects of your asthma medication during the past 2 weeks? 0 Not at all 1 1-3 days 2 4-7 days 3 8-14 days 10. If 1 day or more, what side effects did you have? 11. On average, how severe were the side effects of your asthma medication during the past 2 weeks? 1 Mild 2 Moderate 3 Severe APPENDIX 2

Development of the Multiattribute Utility Function for the ASUI Multiattribute utility theory can be used with either preference values or SG utilities as the underlying metric.26 The construction of a multiatttibute utility function requires the assumption of first-order utility independence, mutual utility independence, or additive utility independence for the preferences for levels of attributes within the classification system. (See Torrance et al26 for more complete discussion of independence assumptions and multiattribute utility theory.) Basically, these assumptions require that the relative scaling within a symptom, say wheeze, remains constant regardless of the relative levels of the other symptoms. A multiattribute utility function was constructed based on the VAS preference and SG utility data collected from th e patient sample. The person-mean approach was used to construct the multiattribute utility function, that is estimating one function from the mean responses of the sample. The derived function combines the ratings within symptoms and the ratings between symptoms to calculate a single index score. The steps taken for developing the multiattribute utility function are as follows: l. Means and SDs were calculated for the VAS ratings within symptoms. The VAS scores were converted to utilities through a power curve estimate of the risk aversion function using data from the five corner states and the five multiple symptom states (see Torrance et al 24 •26 ). A regression model predicting SG utilities from VAS scores for the set of multiple symptom states resulted in an unstandardized regression coefficient of 3.83 (model R2 = 0.96). Therefore, the function was u = 1 - (1 - vf 83 where u equals utilities and v equals VAS values. Using this power function, we converted VAS values to estimated utilities. 2. The general multiplicative multiattribute utility function for five attributes was as follows: Equation 1: (1-p)=(l/c)[n 5 (l +c cp-pJ))-1] j=1 Equation 2: 1+c= n


(1+c c)

j=1 \Vhere pis the preference score of the multisymptom state on the worst possible to best possible symptom scale, that is, where 1006

worst state is severe cough, wheeze, dyspnea, awakened at night, and side effects for 8 to 14 days and best state is no cough, wheeze, dyspnea, awakened at night, or side effect symptoms. Each Pi is the single symptom preference score for symptom j, on a scale where the worst level of symptom j has a preference score of 0 and the best level has a preference score of l. The ci and c values are scaling parameters for the multiattribute utility function that are estimated from the data. Following the approach of Torrance et a!, 24 ·26 parameter c is solved for iteratively using equation 2 and then entered in equation l. The value for c is -0.80037. The result is an estimated utility on the worst symptom state (O) to best symptom state (1) scale. The resulting multiplicative multiattribute utility function is ASUI = 1.200* (s 1 X s2 X s3 X s4 X s5 ) - 0.200 where: ASUI is the utility of the symptom state on a scale where the best state (no symptoms) has a score of 1 and worst possible symptoms (severe symptoms for 8 to 14 days) has a score of 0; and s 1 is the score for the level on symptom l. Table 2 contains the coefficients needed for calculating ASUI scores for the states defined by the ASUI classification system.

REFERENCES 1 Juniper EF. Quality-of-life considerations in the treatment of asthma. Pharmacoeconomics 1995; 8:123-138 2 Rothman ML, Revicki DA. Issues in the measurement of health status in asthma research. Med Care 1993; 3l:MS82MS96 3 Junipe r EF, Guyatt GH, Ferrie PJ, eta!. Measuring quality of life in asthma. Am Rev Respir Dis 1993; 147:832-838 4 Marks GB, Dunn SM, Woolcock AJ. An evaluation of an asthma quality oflife questionnaire as a measure of change in adults with asthma. J Clin Epidemiol1993; 46:1103-1111 5 Rowe BH, Oxman AD. Performance of an asthma quality of life questionnaire in an outpatient setting. Am Rev Respir Dis 1993; 148:675-681 6 Juniper EF, Johnston PR, Borkhoff CM, et al. Quality of life in asthma clinical trials: comparison of salmeterol and salbutamol. Am J Respir Crit Care Med 1995; 151:66-70 7 Juniper EF, Guyatt GH, Epstein RS, et al. Evaluation of impairment of health-related quality of life in asthma: development of a questionnaire for use in clinical trials. Thorax 1992; 47:76-83 8 Hyland ME, Finnis S, Irvine SH . A scale for assessing quality of life in adult asthma sufferers. J Psychosom Res 1991; 35:99-110 9 Marks GB , Dunn SM , Woolcock AJ. A scale for the measurement of quality of life in adults >vith asthma. J Clin Epidemiol 1992; 45:461-472 10 Creer TL, Wigal JK, Kotses H, et a!. A life activities questionnaire for adult asthma. J Asthma 1992; 29:393-399 11 Bousquet J, Knani J, Dhivert H, et a!. Quality of life in asthma: I. Internal consistency and validity of the SF-36 questionnaire. Am J Respir Crit Care Med 1994; 149:371-375 12 Rutten-van Molken MPMH, Clusters F, van Doorslaer EKA, et al. Comparison of performance of four different instruments in evaluating the effects of salmeterol on asthma quality of life. Eur Respir J 1995; 8:888-898 13 Israel E, Cohn J, Dube L, et al. Effect of treatment with zileuton, a 5-lipoxygenase inhibitor, in patients with asthma. JAMA 1996; 275:931-936 14 Marin JM , Carrizo SJ, Garcia R, et al. Effects of nedocromil sodium in steroid-resistant asthma: a randomized controlled trial. J Allergy Clin Immunol1996; 97:602-610 15 Peters DH, Faulds D. Salmeterol: an appraisal of its qualityof-life benefits and potential pharmacoeconomic positioning Clinical Investigations

in asthma. Pharmacoeconomics 1995; 7:562-574 16 Hyland ME, Finnis S, Irvine SH. A scale for assessing quality of life in adult asthma suffers. J Psychosom Res 1991; 35:99-110 17 Revicki DA. Relationship of pharmacoeconomics and healthrelated quality of life. In: Spilker B, ed. Quality of life and pharmacoeconomics in clinical trials. 2nd ed. Philadelphia, PA: Lippincott-Raven, 1996; 1077-1084 18 Sculpher MJ, Buxton MJ. The episode-free day as a composite measure of effectiveness: an illustrative economic evaluation of formoterol versus salbutamol in asthma therapy. Pharmacoeconomics 1993; 4:345-352 19 Torrance GW. Measurement of health state utilities for economic appraisal. J Health Econ 1986; 5:1-30 20 Feeny D, Torrance GW. Incorporating utility-based qualityof-life assessment measures in clinical trials: two examples. Med Care 1989; 27:S190-S204 21 Bennett KJ, Torrance GW. Measuring health state preferences and utilities: rating scale, time trade-off, and standard gamble techniques. In: Spilker B, ed. Quality of life and pharmacoeconomics in clinical trials. 2nd ed. Philadelphia, PA: Lippincott-Raven, 1996; 253-266 22 Berzon RA, Mauskopf JA, Simeon GP. Choosing a health profile (descriptive) and/or patient-preference (utility) measure for a clinical trial. In: Spilker B, ed. Quality of life and pharmacoeconomics in clinical trials. 2nd ed. Philadelphia, PA: Lippincott-Raven, 1996; 375-418 23 Kerridge RK, Glasziou PP, Hillman KM. The use of'quality-








adjusted life years' (QALYs) to evaluate treatment in intensive care. Anaesth Intensive Care 1995; 23:322-331 Torrance GW, Feeny DH, Furlong WJ, eta!. Multi-attribute preference functions for a comprehensive health status classification system: health utilities index mark 2. Med Care 1996; 34:702--722 Torrance GW, Boyle MH, Hmwood SP. Application of multi-attribute utility theory to measure social preferences for health states. Operations Res 1982; 30:1043-1069 Torrance GW, Furlong W, Feeny D, et a!. Multi-attribute preference functions: health utilities index. Pharmacoeconomics 1995; 7:503--520 Revicki DA. Health care technology assessment and healthrelated quality of life. In: Banta D, Luce B, eds. Health care technology and its assessment: an international perspective. Oxford: Oxford University Press, 1993; 114-131 Furlong W, Feeny D, Torrance GW, et a!. Guide to design and development of health-state utility instrumentation. Hamilton, Ontario: Centre for Health Economics and Policy Analysis, McMaster University, 1990 Feeny DH, Torrance GW, Furlong W. Health utilities index (HUI). In: Spilker B, ed. Quality of life and pharmacoeconomics in clinical trials. 2nd ed. Philadelphia, PA: LippincottRaven, 1996; 239-252 Diemer FB, Horowitz E, Horton M, et a!. Validation of a questionnaire to assess seve1ity of asthma in adults. J Allergy Clin Immunol 1997; 99:560

CHEST I 114 I 4 I OCTOBER, 1998