Derivation and Validation of a Composite Index of Severity in Chronic Obstructive Pulmonary Disease The DOSE Index Rupert C. Jones1, Gavin C. Donaldson2, Niels H. Chavannes3, Kozui Kida4, Maria Dickson-Spillmann1, Samantha Harding1, Jadwiga A. Wedzicha2, David Price5, and Michael E. Hyland6 1
Respiratory Research Unit, Peninsula Medical School, University of Plymouth, Devon, and 2Department of Medicine, Academic Unit of Respiratory Medicine, University College London, London, United Kingdom; 3Department of Public Health and Primary Care, Leiden University Medical Center, Leiden, The Netherlands; 4Department of Pulmonary Medicine, Infection, and Oncology, Respiratory Care Clinic, Nippon Medical School, Tokyo, Japan; 5Department of General Practice and Primary Care, Foresterhill Health Centre, University of Aberdeen, and 6School of Psychology, University of Plymouth, Devon, United Kingdom
Rationale: Chronic obstructive pulmonary disease (COPD) is increasingly recognized as a multicomponent disease with systemic consequences and effects on quality of life. Single measures such as lung function provide a limited reflection of how the disease affects patients. Composite measures have the potential to account for many of the facets of COPD. Objectives: To derive and validate a multicomponent assessment tool of COPD severity that is applicable to all patients and health care settings. Methods: The index was derived using data from 375 patients with COPD in primary care. Regression analysis led to a model explaining 48% of the variance in health status as measured by the Clinical COPD Questionnaire with four components: dyspnea (D), airflow obstruction (O), smoking status (S), and exacerbation frequency (E). The DOSE Index was validated in cross-sectional and longitudinal samples in various health care settings in Holland, Japan, and the United Kingdom. Measurements and Main Results: The DOSE Index correlated with health status in all data sets. A high DOSE Index score (>4) was associated with a greater risk of hospital admission (odds ratio, 8.3 [4.1–17]) or respiratory failure (odds ratio, 7.8 [3.4–18.3]). The index predicted exacerbations in the subsequent year (P < 0.014). Conclusions: The DOSE Index is a simple, valid tool for assessing the severity of COPD. The index is related to a range of clinically important outcomes such as health care consumption and predicts future events. Keywords: health status; outcome assessment; composite outcomes; disease severity
Chronic obstructive pulmonary disease (COPD) is defined as airflow limitation that is not fully reversible (1). The disease is increasingly recognized as a multicomponent disease with important systemic effects such as skeletal muscle dysfunction and cardiovascular deconditioning, which impair the functional status of patients (2, 3). The disorder is also associated with impaired health status (4, 5). These different aspects interact to cause a spiral of decline; breathlessness reduces physical activity, inactivity causes physical deconditioning and demoti-
(Received in original form February 19, 2009; accepted in final form September 23, 2009) Supported by a Researcher Development Award of the Department of Health, UK (R.J.); no other funding was obtained. Correspondence and requests for reprints should be addressed to Rupert C. Jones, M.D., Respiratory Research Unit, Peninsula Medical School,1 Davy Road, Plymouth PL6 8BX, UK. E-mail:
[email protected] This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org Am J Respir Crit Care Med Vol 180. pp 1189–1195, 2009 Originally Published in Press as DOI: 10.1164/rccm.200902-0271OC on September 24, 2009 Internet address: www.atsjournals.org
AT A GLANCE COMMENTARY Scientific Knowledge on the Subject
Chronic obstructive pulmonary disease (COPD) is a complex disease, but current guideline recommendations for assessment of severity are based on airflow obstruction. Multicomponent indices have been developed but are not widely used in clinical practice. What This Study Adds to the Field
The DOSE Index (MRC Dyspnea Scale, airflow obstruction, smoking status, and exacerbation frequency) was derived and validated in international data sets. Unlike other severity indices, this index is intended for use in routine clinical settings, not just as a measure of disease severity but also as a quick guide to management.
vation, leading to further breathlessness, inactivity, and reduced quality of life (3). Traditionally, the FEV1 has been the main measure of COPD severity for clinicians and still has a prominent place in guidelines. Although patients are concerned mainly with symptoms, exacerbations, and functional capacity (6), airflow obstruction is important to clinicians to measure lung damage and to determine treatment. A composite measure could account for various dimensions of the disease, and take into account both the patient’s and the physician’s perspectives. One highly regarded composite measure is the BODE [body mass index (B), degree of airflow obstruction (O), dyspnea (D), and exercise capacity (E)] Index (7), which was originally designed to predict mortality in COPD. However, the BODE Index involves a 6-minute walk test (6MWT), which limits its use in routine clinical settings as it takes time, supervision, and space. Another validated prognostic index, the COPD Prognostic Index (8), is also cumbersome to use in routine clinical settings as it includes seven items, one of which is a health status questionnaire. The aim of this study was to derive a multicomponent assessment index of current COPD severity. The index was intended to include items that are clinically important, applicable to all grades of disease severity and all health care settings, and simple and clear to use. Once derived, we aimed to validate the index against other markers of current disease severity such as health care consumption. We also aimed to test the index in relation to predicting future events, such as hospital admissions, exacerbations, and courses of treatment.
1190
AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 180
METHODS The DOSE Index was derived from the Devon Primary Care COPD audit data set. This consisted of 375 patients in primary care with confirmed COPD, of whom 197 (53%) had GOLD (Global Initiative for Chronic Obstructive Lung Disease) stage II disease, 144 (38%) had GOLD stage III disease, and 34 (9%) had GOLD stage IV disease. Patients underwent a comprehensive, guideline-based assessment of their disease, performed by a respiratory specialist. The process of the audit has been described elsewhere (9). Exacerbations in this cohort were recorded by patient self-report with confirmation from primary care medical records; the definition of an exacerbation was that stated in the National Institute for Health and Clinical Excellence (NICE, London, UK) guidelines (10). Unscheduled visits refer to emergency consultations at the patient’s home, or in an emergency treatment center outside normal office hours, for example, at night or at the weekend. On the basis of theoretical clinical considerations a number of possible predictors of health status were selected (Table 1). The number of exacerbations per year is known to predict outcomes (11), is needed for determining therapy, and is recognized as a clinically important marker in current guidelines (10). Smoking status and packyears are known to affect prognosis, as are measures of airflow obstruction, for example, FEV1% predicted and exercise capacity (1). Health status was measured using the Clinical COPD Questionnaire (CCQ) (12). Evaluation of the CCQ, including treatment of missing data, was in line with instructions given by the original authors. For 12% of patients (n 5 43) no CCQ total score could be calculated. In a first step, the relationships of the possible predictor variables with health status were quantified through bivariate correlations. Subsequently, those variables showing strongly significant relationships (i.e., P < 0.01) with health status were entered as predictors into a multiple regression analysis with health status as the dependent variable. The variance in health status explained by the model was assessed. Finally, a scoring system was designed to weigh the components according to their clinical and statistical strength. Further details of the weighting process are provided in the online supplement. Details of validation of the DOSE Index in other data sets are also given in the online supplement.
Cross-Sectional Validation Procedures To assess external validity of the DOSE Index as a marker of current health status, Spearman’s rank correlations between the DOSE Index and the St. George’s Respiratory Questionnaire (SGRQ) (13) in the Holland, London, and Tokyo data sets were calculated. Further, the DOSE Index was tested against other markers of disease severity. The relationship between the DOSE Index score and health care consumption such as bed days in hospital was analyzed using scatterplots to establish a cutoff point into two categories: low and high DOSE Index score. Using the DOSE Index categories as the predictor, linear regression analyses were performed on markers of disease severity, such as bed days or current resting hypoxia in the
TABLE 1. SPEARMAN’S CORRELATION (r) BETWEEN PROPOSED MARKERS OF HEALTH STATUS AND CLINICAL COPD QUESTIONNAIRE TOTAL SCORE IN DEVON DERIVATION DATA SET Variable MRC Dyspnea Scale score FEV1% predicted Obstruction grade Steroid courses Out-of-hours visits A&E attendances Exacerbations Current smoking status Pack-years BMI
2009
Devon validation data set. Respiratory failure was defined as resting oxygen saturation (SaO2) of less than 92% (14). In the Tokyo data set, the relationships between the DOSE Index score and the BODE Index, the body mass index (BMI), and the 6MWT were examined using Spearman’s rank correlations. Details of methods used in assessing longitudinal changes in the DOSE Index score are given in the online supplement.
RESULTS Sample Characteristics
The index was derived in one sample of people with COPD and validated in four different samples; the five samples investigated in this study are characterized in Table 2. Derivation of the DOSE Index
The correlations of all variables initially considered to be potentially related to health status are shown in Table 1. Four correlations were significant: The Medical Research Council (MRC, London, UK) Dyspnea Scale score, the obstruction grade based on FEV1% predicted, the number of exacerbations, and the current smoking status. Colinearity between these predictors of health status was not significant as correlations among the predictors ranged from r 5 20.37 to r 5 0.22; colinearity is judged as significant if it exceeds r 5 60.80. Multiple regression analysis showed that these four variables explained 48% of the variance in health status (Table 3). Thus, the DOSE Index consists of the MRC Dyspnea Scale score (D), the airflow obstruction grade (O), the current smoking status (S), and the number of exacerbations (E). To design the index scoring system, a number ranging between 0 and 3 was assigned to each factor. Weighting was given according to both clinical and statistical considerations. Further details are provided in the online supplement. For each component of the scoring system, cutoff values were chosen. FEV1% predicted cutoffs were selected using the obstruction grading based on the GOLD and NICE guidelines (1, 10). For exacerbations the cutoff values related to the frequency of exacerbations, which predicts outcome: values above 2.9 exacerbations per year were associated with faster decline in lung function (15). Two or more exacerbations per year are considered to be an indication for certain drug treatments (1, 10). The DOSE Index score is built from the sum of its components and the total score ranges from 0 to 8; the higher the score, the more severe the disease status (Table 4). In the Devon validation cohort, the distribution of the DOSE Index was analyzed (Figure 1). The median (interquartile range) was 2 (1–3). The distribution was skewed toward lower DOSE Index scores, which corresponded to a high proportion of mild to moderate COPD cases in this cohort. Validity of DOSE Index as a Marker of Current Health Status
n
r
P Value
329 329 329 320 328 328 329 329 321 329
0.67 0.31 0.31 0.31 0.21 0.19 0.26 0.18 0.06 0.05
,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.01 ,0.001 ,0.001 0.32 0.37
Definition of abbreviations: A&E 5 accident and emergency; BMI 5 body mass index; MRC 5 Medical Research Council.
The associations of the DOSE Index with health status were found to correlate significantly with health status in the London, Japan, and Holland samples (Table 5). Validity of DOSE Index as an Indicator of Current Health Care Consumption
Visualization of the DOSE Index against the number of bed days in the previous year indicated that above a DOSE Index score of 4, bed days increased rapidly (Figure 2). Linear regression analysis confirmed the validity of a DOSE Index score cutoff of 4 to predict increased health care consumption (Table 6). On average, patients with a DOSE Index score above 4 were eight times more likely to be admitted and had spent 5.1 more bed days in hospital than patients with
Jones, Donaldson, Chavannes, et al.: DOSE: An Index of COPD Severity
1191
TABLE 2. DESCRIPTIVE DATA OF DERIVATION AND VALIDATION COHORTS Devon Derivation
Devon Validation
Tokyo Validation
Holland Validation
London Validation
375 224 (60%) 69.2 (8.6) 50 (14) 0/52/39/9 120 (32%) 1.3 (1.6) 26.7 (5.6) 1.66 (1.0) — 2.0 (1.1)
460 265 (58%) 69.5 (8.7) 49 (15) 0/51/37/12 144 (31%) 1.5 (1.9) 26.7 (5.7) 1.79 (1.1) — —†
81 77 (95%) 73.2 (6.9) 49.4 (14.8) 0/49/40/11 2 (3%) 1.01 (1.7) 22.2 (3.1) 1.21 (0.8) 33.2 (13.2)* —
152 96 (63%) 63.1 (11.0) 66.9 (18.4) 10/61/26/3 62 (41%) 0.47 (0.5) 26.7 (5.0) 2.26 (1.0) 30.8 (20.3) 1.4 (0.9)
133 94 (71%) 67 (8.0) 41.8 (15.8) 1/29/39/31 49 (37%) 1.04 25.7 3.14 (1.0) 53.6 (17.0) —
Number Sex, males Age, years Mean FEV1% predicted GOLD I, II, III, IV, % Current smokers Exacerbations per year BMI MRC Dyspnea Scale SGRQ, total CCQ, total
Definition of abbreviations: CCQ 5 Clinical COPD Questionnaire; GOLD 5 Global Initiative for Chronic Obstructive Lung Disease; SGRQ 5 St. George’s Respiratory Questionnaire. Age, number of exacerbations, BMI, and all questionnaire total and subscale scores are presented as means (SD). * Japanese version of the SGRQ. † CCQ scores were not assessed in the oxygen clinic subsample of the Devon validation sample.
a DOSE Index score of 4 or less. There were few emergency department attendances in the Devon data set, but the DOSE Index cutoff of 4 predicted emergency department attendances at a near-significant level (b 5 0.10, P 5 0.06, R2 5 0.01).
Hospitalization
Validity of DOSE Index as an Indicator of Current Respiratory Failure
Respiratory failure was defined as an oxygen saturation level of less than 93% on air at rest, using pulse oximetry. Using the cutoff of 4, the DOSE Index score was a highly significant predictor of desaturation. In those with a lower DOSE Index score, 9% were desaturated compared with 45% in the higher DOSE Index score group (odds ratio, 7.8 [3.4–18.3], x2 P , 0.0001). Validity of DOSE Index as an Indicator of Current Exercise Capacity, BMI, and the BODE Index
Exercise capacity (as measured by the distance walked in the 6MWT) showed a strong negative correlation (r 5 –0.54, P < 0.01) with the DOSE Index in the Tokyo data set. The BMI was also negatively associated with DOSE Index (r 5 –0.30, P < 0.01). The BODE Index was highly correlated with the DOSE Index (r 5 0.78, P < 0.01). Changes in the DOSE Index over Time
In the London exacerbation cohort, 338 DOSE Index scores were available on 175 patients. Over a 9-year period the DOSE Index score increased at a rate of 0.18 unit/year (z 5 5.26; P , 0.001; 95% confidence interval [CI], 0.11 to 0.25). In Figure 3, raw mean DOSE Index score (with standard error bars) was plotted against year of follow-up. No allowance has been made
TABLE 3. MULTIPLE LINEAR REGRESSION ANALYSIS RESULTS FOR MAIN INDICATORS OF HEALTH STATUS (CLINICAL COPD QUESTIONNAIRE)
MRC Dyspnea Scale score Number of exacerbations (12 mo) Current smoking status FEV1% predicted (Constant)
for patients withdrawing from the study or missing DOSE Index scores in any year. Year 0 refers to the first year of the study.
b
SE Coefficient
b
0.62 0.11 0.35 20.01 0.42
0.05 0.03 0.10 0.01 0.25
0.58* 0.15* 0.15* 20.09† 0.09
R2 5 0.48, F(4, 324) 5 74.54, SE 5 standard error of the b coefficient. * P , 0.001. † P , 0.05.
In the London study, there were a total of 338 annual DOSE Index scores on 175 patients, during which there were 50 hospital admissions for an acute exacerbation. The DOSE Index was related by random-effects Poisson regression to the number of hospitalizations in the current year (exponentiated coefficient [which has the interpretation of incidence rate ratios], 1.44; P , 0.001; 95% CI, 1.20–1.72). Fewer data were available for the subsequent year, as patients withdrew or died. For the data available (217 annual DOSE Index scores on 107 patients) there was no relationship between DOSE Index score and hospital admission (coefficient, 1.03; P 5 0.77; 95% CI, 0.82– 1.30). Using Pearson correlations, DOSE Index score correlations with number of bed days were stronger than FEV1% predicted and the MRC Dyspnea Scale in the Devon validation cohort. DOSE Index score correlations were r 5 0.334 and P 5 0.001; for FEV1% predicted they were r 5 0.027 and P 5 0.698; and for the MRC Dyspnea Scale they were r 5 0.288 and P 5 0.001. In the London Cohort, based on whether patients were admitted or not, DOSE Index score was a stronger predictor of future admissions than the FEV1% predicted. The receiver operator characteristics analysis gave an area under the curve (AUC) of 0.755 for the DOSE Index, against FEV1% predicted (AUC 5 0.254) the difference being significant at x2 5 12.16 (P 5 0.0005), and slightly but not significantly better than the MRC Dyspnea Scale (AUC 5 0.6278, x2 5 3.54, P 5 0.060). TABLE 4. DOSE INDEX SCORING SYSTEM DOSE Index Points
MRC Dyspnea Scale score Obstruction FEV1% predicted Smoking status Exacerbations per year
0
1
2
3
0–1 .50 Nonsmoker 0–1
2 30–49 Smoker 2–3
3 ,30
4
.3
To build the DOSE Index score, the DOSE Index points associated with every category of all four variables are added up. Example: Patient has an MRC Dyspnea Scale score of 2 (1 DOSE point), an FEV1% predicted of 61 (0 DOSE points), is a smoker (1 DOSE point), and has had two exacerbations over the previous year (1 DOSE point). The patient’s DOSE Index score is 1 1 0 1 1 1 1 5 3.
1192
AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 180
2009
TABLE 5. SPEARMAN’S RANK CORRELATIONS BETWEEN MEAN SCORE ON ST. GEORGE’S RESPIRATORY QUESTIONNAIRE AND DOSE INDEX IN LONDON, JAPAN, AND HOLLAND COHORTS
SGRQ SGRQ SGRQ SGRQ
total symptoms activity impact
London
Japan
Holland
0.44* 0.22† 0.47* 0.41*
0.49* 0.27† 0.51* 0.38*
0.56* 0.28† 0.60* 0.47*
* P , 0.01. † P , 0.05.
Figure 1. Distribution of DOSE Index scores in the Devon validation cohort, presented for various airflow obstruction grades. Solid 5 severe disease; gray 5 moderate disease; dashed 5 mild disease.
Exacerbations, Antibiotic and Steroid Courses
There were 149 patients and 284 annual DOSE Index scores for years when treatment with antibiotics and oral steroids were recorded in the London cohort. The DOSE Index score was related to the annual number of exacerbations in that year (random-effects Poisson regression coefficient 5 1.30; P < 0.001; 95% CI, 1.23–1.39). The DOSE Index score was also related to the number of exacerbations in the subsequent year
although there were fewer patients (n 5 128) and observations (n 5 245) (coefficient, 1.07; P 5 0.01; 95% CI, 1.02–1.14). A similar relationship was seen between the DOSE Index score and the number of exacerbations treated with antibiotics in the subsequent year (coefficient, 1.08; P 5 0.009; 95% CI, 1.02–1.15), and with exacerbations treated with oral corticosteroids in the subsequent year (coefficient, 1.10; P 5 0.007; 95% CI, 1.03–1.18). Associations between the DOSE Index scores and various markers of disease severity indicate that the DOSE Index score reflects these markers better than its components (Table 7). One exception is the MRC Dyspnea Scale; the DOSE Index has similar predictive strength compared with the MRC alone in relation to many variables examined in cross-sectional and longitudinal data.
DISCUSSION In this article we describe the derivation and validation of a simple index that could be conveniently performed in a routine clinical setting and that includes items that are clinically important in their own right. In a large data set from primary care we have shown that the MRC Dyspnea Scale, airflow obstruction (FEV1% predicted), current smoking status, and number of exacerbations per year were predictors of health status as measured by the CCQ total score. A weighted index was derived and validated in a range of patient populations with varying nationality, disease status, and clinical settings.
Figure 2. Mean number of days spent in hospital as a function of the DOSE Index score. Note: DOSE Index scores higher than 5 are not displayed because of the small number of cases in these categories (8 and 2, respectively).
Jones, Donaldson, Chavannes, et al.: DOSE: An Index of COPD Severity
1193
TABLE 6. HEALTH CARE CONSUMPTION FOR PATIENTS WITH DOSE INDEX SCORE OF 4 OR LESS COMPARED WITH THOSE WITH DOSE INDEX SCORE GREATER THAN 4 DOSE Index Score
