- Email: [email protected]
Characterizing and Quantifying the Symptomatic Features of COPD Exacerbations
CHEST
Original Research COPD
Characterizing and Quantifying the Symptomatic Features of COPD Exacerbations Paul W. Jones, PhD; Wen-Hung Chen, PhD; Teresa K. Wilcox, PhD; Sanjay Sethi, MD; and Nancy Kline Leidy, PhD; for the EXACT-PRO Study Group*
Background: There is a need for a standardized, valid, and reliable instrument for quantifying exacerbations of COPD. The objective of this study was to identify symptom items that characterize COPD exacerbations to form a new patient diary for evaluating exacerbation frequency, severity, and duration. Methods: Twenty-three symptom items identified from patient interviews were administered to 410 patients with COPD aged (mean ⫾ SD) 65 ⫾ 10 years with stable FEV1 of 51% predicted ⫾ 20% predicted and 1.8 ⫾ 1.8 exacerbations in the preceding 12 months. A total of 222 patients had a physician-diagnosed exacerbation; 188 were stable. Item-level analyses (floor and ceiling effects, criterion keying, item-total correlation) were used in the first stage of item reduction. Further reduction was conducted using Rasch model and descriptive item analyses. Exploratory factor analysis was performed on the items that survived the exclusion process. Results: No item behaved differently between stable and exacerbation conditions. One item was removed after item-level analysis, and eight were removed following Rasch analysis. Together, the surviving 14 items met the criteria for a unidimensional measure of exacerbation severity. Internal consistency (person separation index) was excellent at 0.92. Post hoc exploratory factor analysis revealed one dominant factor, with three domains (breathlessness, cough and sputum, and chest symptoms) that accounted for 68% of the variance. Conclusions: An exacerbation appears to be a quantitative rather than qualitative change from the stable state. This analysis identified a range of symptoms that form a unidimensional construct of overall exacerbation severity. The 14 items identified form the Exacerbations of Chronic Pulmonary Disease Tool (EXACT), a daily diary for detecting and quantifying exacerbation severity in COPD. CHEST 2011; 139(6):1388–1394 Abbreviations: EFA 5 exploratory factor analysis; EXACT 5 Exacerbations of Chronic Pulmonary Disease Tool; SGRQ-C 5 St. George Respiratory Questionnaire-COPD
are an important feature of COPD, Exacerbations leading to significant morbidity and mortality, and 1
they often are used as outcomes in clinical studies.2,3 Currently, there is no agreed-upon method for quantifying exacerbations of COPD4,5; the most widely used method is based on an increase in symptoms that requires treatment in a clinic, ED, or hospital. Exacerbations identified using this type of eventbased definition do not always correspond with those identified using symptom-based definitions,6 so methods that detect and quantify exacerbations using explicit clinical criteria are needed. COPD exacerbations are characterized by an acute worsening of symptoms7,8 and have been defined as
“a sustained worsening of the patient’s condition, from the stable state and beyond normal day-today variations, that is acute in onset and necessitates a change in regular medication.”9 This definition does not specify the components that worsen. Sputum color has attracted attention because it may be predictive of a high bacterial load,10 but less attention has been paid to other symptoms that together may serve as a measure of severity. Several groups have used exacerbation diary cards1; however, there has never been a formal analysis of exacerbation symptoms to identify those that can provide reliable detection and quantification of changes.
1388
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
This study describes part of the development of a validated diary card for detecting and quantifying exacerbations: the Exacerbations of Chronic Pulmonary Disease Tool (EXACT) Patient-Reported Outcome (known as EXACT-PRO). It examined items identified through extensive patient interviews about the symptomatic features of an exacerbation.11 We tested two hypotheses: (1) that an exacerbation is a quantifiable change in the intensity of preexisting symptoms and (2) that the transition from stable state to exacerbation may be characterized by the emergence or shift in relative importance of a new symptom or symptom cluster. Materials and Methods Study Design and Participant Recruitment A total of 23 potential items were developed from a comprehensive series of interviews and qualitative analyses in patients with COPD.11 These items assessed patients’ breathlessness, cough, sputum, chest discomfort, difficulty breathing, activity limitations, feelings of tiredness or weakness, sleep disturbance, and scared or worried feelings about their condition. Patients completed the diary each evening before bedtime, reflecting back on “today” and rating the severity of each attribute or item on a 3-, 5-, or 6-point ordinal scale (eg, from not at all to extremely). See e-Appendix 2 for additional information. The candidate items and final instrument were tested in a prospective, two-group observational study. Methods and results for tests of reliability, validity, and responsiveness of the final instrument are presented elsewhere.12 Briefly, the groups were stable patients (stable) and those with a clinician-confirmed exacerbation (acute). Participants were recruited from 38 sites in the United States. Inclusion criteria for both groups were previous diagnosis of COPD or chronic bronchitis, age . 40 years, smoking history . 10 pack-years, telephone landline for uploading electronic diary data, and ability to read and speak English. Exclusion criteria included concurrent diagnosis of asthma, clinically relevant bronchiectasis, visual or cognitive impairment that would interfere with completing questionnaires, and a concurrent medical or psychiatric condition that would affect participation. Stable participants were required to have a history of one or more acute exacerbations requiring systemic corticosteroids or Manuscript received May 14, 2010; revision accepted October 2, 2010. Affiliations: From St. George’s (Dr Jones), University of London, London, England; United BioSource Corporation (Drs Chen, Wilcox, and Leidy), Bethesda, MD; and University at Buffalo (Dr Sethi), SUNY, Buffalo, NY. *A list of study group members is available in e-Appendix 1. Funding/Support: The following companies have provided unrestricted funds to support the EXACT-PRO Initiative: AstraZeneca, GlaxoSmithKline, Pfizer Inc, Boehringer Ingelheim, Merck & Company, Sepracor Inc, Forest Laboratories Inc, Novartis, Schering-Plough, Adams Respiratory Therapeutics, Bayer, Atlanta AG (Nycomed), and Ortho-McNeil. Correspondence to: Nancy Kline Leidy, PhD, United BioSource Corporation, 7101 Wisconsin Ave, Ste 600, Bethesda, MD 20814; e-mail: [email protected] © 2011 American College of Chest Physicians. Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (http://www.chestpubs.org/ site/misc/reprints.xhtml). DOI: 10.1378/chest.10-1240 www.chestpubs.org
antibiotics within the previous 24 months and no clinic, ED, or hospital visit for an exacerbation within the previous 60 days. Acute participants were enrolled at a clinic visit triggered by worsening symptoms. They were recruited after full clinical evaluation, including physical examination and tests thought appropriate by the physician. A consensus exacerbation definition was used for determining recruitment.9 Sputum purulence was not a requirement. Procedure The study protocol was approved by the Sterling Institutional Review Board (Atlanta, Georgia) (IRB ID #2405), and all participants provided written consent prior to initiation of data collection. All participants completed a 27-item diary, including the pool of 23 items under test, each evening on a personal digital assistant (Palm Tungsten E; Hewlett-Packard Company; Palo Alto, California). Stable patients completed the diary for 7 days. Acute patients completed the diary on days 1 to 29 (⫾ 2) of the acute event (day 1 being the day of first presentation with the exacerbation), returning to the clinic on day 60 (⫾ 7 days) for 7 days of diary assessment (days 60-67) during a stable state. All participants completed the St. George Respiratory QuestionnaireCOPD (SGRQ-C).13 Physicians and patients also completed global assessments of health state, exacerbation severity, and global assessments of change. Stable-state FEV1 was recorded. All recruitment and study procedures met Health Insurance Portability and Accountability Act requirements. Additional information on study methods is provided in e-Appendix 2. Analyses A statistical analysis plan was developed prior to initiating the analyses, prespecifying the stepwise item reduction process to be used. The data analysis for this article was generated using SAS/BASE and SAS/STAT software version 9.1.3 of the SAS System for PC (SAS Institute; Cary, North Carolina). Item Analyses: The following item characteristics were examined using day-1 data of the pooled sample and by group (acute, stable): mean, frequencies of item responses, percentages of minimum and maximum responses, percent missing, inter-item correlation, and item-total correlation. An item was flagged for potential exclusion if it showed a minimum response (floor effect) . 50%; maximum response (ceiling effect) . 50%; interitem correlation . 0.80, suggesting that it may be redundant; or interitem correlation , 0.20, suggesting that it did not fit with the others. Each item also was examined by criterion keying14 using clinician global rating of patient health on day 1 as the external criterion. Similar tests were used to identify items with sex or age bias. Items were flagged for possible exclusion if they had ⱕ 4% shared variance with clinician rating of patient health state or ⱖ 6% shared variance with sex or age. Factor Analysis: To assess the presence of specific domains within the item set, exploratory factor analysis (EFA) was performed on the initial 23 items using a random split-half data sample from day 1. This analysis was carried out using pooled data and separately by groups. Methodology details are provided in e-Appendix 2. Rasch Model Analysis: With the Rasch model, the more severe the exacerbation (as judged by the patients’ response to all of the items combined), the higher the probability of a positive response to any given constituent item.15 The model assumes that all items have uniform discrimination power to distinguish between high and low severity. Statistical tests were used to test the fit of each individual item to the mathematically determined relationship with overall severity that is seen with the Rasch model, if the item CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1389
was a perfect measure of exacerbation severity. More details are available in e-Appendix 2 and in the article by Meguro et al13 and its online supplement. Rasch analyses were conducted using day-1 data from the entire pooled sample. In addition, each item was examined for differential item functioning to test whether it behaved consistently between stable state and exacerbation. This examination was a formal test of the hypotheses that an exacerbation is either a quantitative or a qualitative change from the stable state. Item severity was measured in terms of logits—the units in a Rasch model. The logit is the log odds for an item of given severity being affirmed by 50% of the study population. The patient’s severity can be measured in the same units. With Rasch analysis, item reduction is an iterative process of individual item deletion, beginning with the worst-fitting items, with consideration also given to item descriptive statistics, factor analyses, qualitative data, and clinical experience. Each time an item is deleted, the data are reanalyzed to evaluate fit of the remaining items to the model. The process continues until a core set of items is identified. SAS, version 8.2 statistical software was used to characterize the sample and item-level descriptive statistics. Mplus (Muthén & Muthén; Los Angeles, California) was used for EFA,16 and RUMM2020 (RUMM Laboratory Pty Ltd; Duncraig, Western Australia, Australia) was used for Rasch modeling.17
48% were men. Stable-state FEV1 was 51% predicted ⫾ 20% predicted. Using independent sample t tests for continuous variables and x2 tests for categorical variables, there were no significant differences between the groups on any demographic or clinical characteristics with two exceptions. The acute group reported more frequent events the previous year (2.2 ⫾ 0.5) than the stable group (1.3 ⫾ 1.2) (t 5 5.69 using the Satterthwaite method for unequal variances; P , .001). On day 1, SGRQ-C scores of the acute group (62.1 ⫾ 17.7) were significantly higher than those of the stable group (50.8 ⫾ 19.9) (t 5 5.97 pooled; P , .0001).
Results
Scree plots were created using all of the candidate items. One dominant factor was seen in the pooled sample; its eigenvalue was 9.7, and it accounted for 42% of the total variance (Fig 1). A dominant singlefactor pattern also was seen in both the acute group
Sample There were 222 acute patients and 188 stable patients. Their mean ⫾ SD age was 65 ⫾ 10 years, and
Item-Level Analyses During item-level analyses, items 5, 7, 11, 12, 13, 14, 16, 18, and 21 were flagged for possible exclusion. Item-level descriptive statistics and rationale for flagging are shown in Table 1. Factor Analysis
Table 1—Item Descriptive Statistics for the Pooled Sample Day 1 (N 5 410) Item Number and Description 1. Chest congested 2. Cough 3. Mucus when coughing 4. Difficulty with mucus 5. Mucus color 6. Chest discomfort 7. Chest hurt 8. Chest tight 9. Breathless 10. How breathless 11. SOB while sittingb 12. Difficulty breathing while sittingb 13. How active 14. Usual personal care 15. SOB with personal care 16. Usual indoor 17. SOB with indoor 18. Usual outdoor 19. SOB with outdoor 20. Weak/tired 21. Sleep disturbed 22. How much sleep 23. Scared/worried
Flag
Mean ⫾ SD
Range
Median
Floor, No. (%)
Ceiling, No. (%)
Item-Total Correlationa
… … … … *§ … * … … … X*§ *§
2.5 ⫾ 1.05 3.1 ⫾ 1.02 2.5 ⫾ 1.05 2.6 ⫾ 1.29 1.8 ⫾ 1.13 2.1 ⫾ 0.96 1.9 ⫾ 0.93 2.1 ⫾ 0.97 2.7 ⫾ 0.96 2.8 ⫾ 0.99 1.7 ⫾ 0.84 1.7 ⫾ 0.80
1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5
2 3 3 3 1 2 2 2 3 3 2 2
79 (19.3) 34 (8.3) 95 (23.2) 114 (27.8) 241 (58.8) 127 (31.0) 182 (44.4) 118 (28.8) 37 (9.0) 36 (8.8) 190 (46.3) 190 (46.3)
14 (3.4) 23 (5.6) 7 (1.7) 31 (7.6) 10 (2.4) 4 (1.0) 3 (0.7) 6 (1.5) 14 (3.4) 28 (6.8) 2 (0.5) 1 (0.2)
0.73 0.58 0.48 0.59 0.35 0.67 0.68 0.71 0.71 0.59 0.74 0.72
X*O *§ … X … X§ … … … XO …
3.3 ⫾ 0.72 1.5 ⫾ 0.56 2.3 ⫾ 1.03 2.2 ⫾ 0.84 2.6 ⫾ 1.42 2.3 ⫾ 0.77 2.7 ⫾ 1.61 2.5 ⫾ 0.97 2.3 ⫾ 1.02 3.4 ⫾ 0.79 2.3 ⫾ 1.13
1-5 1-3 1-6 1-3 1-6 1-3 1-6 1-5 1-5 1-5 1-5
3 1 2 2 2 3 2 2 2 3 2
1 (0.2) 232 (56.6) 102 (24.9) 116 (28.3) 94 (22.9) 76 (18.5) 111 (27.1) 67 (16.3) 95 (23.2) 3 (0.7) 121 (29.5)
21 (5.1) 13 (3.2) 3 (0.7) 179 (43.7) 35 (8.5) 216 (52.7) 53 (12.9) 10 (2.4) 9 (2.2) 34 (8.3) 22 (5.4)
0.20 0.54 0.66 0.38 0.67 0.32 0.63 0.68 0.57 0.24 0.59
There were no missing item-level data by design. * 5 based on pooled sample analyses; § 5 based on acute sample analyses; O 5 based on exploratory factor analysis loading , 0.3; SOB 5 shortness of breath; X 5 based on stable sample analyses. aPearson product-moment correlation (r). bInter-item correlation (r) 5 0.88. 1390
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
Figure 1. Scree plots from the exploratory factor analysis (EFA) for the pooled, acute, and stable samples with data from day 1. The x axis shows the number of identified factors. The y axis shows the eigenvalue associated with each factor; this is related to the amount of variance associated with each factor. One factor had the highest eigenvalue showing that this was the dominant item grouping.
(eigenvalue 7.6; 33% of variance) and the stable group (eigenvalue 8.3; 36% of variance). As a further test, the analysis was repeated with the item responses used as categorical variables. The findings were the same (e-Table 1), which lent support to the starting hypothesis for the Rasch analysis that there is an underlying unidimensional structure to the data. Rasch Analysis The Rasch analyses started with 22 items because item 13 was excluded due to poor performance across multiple criteria. None of the other eight flagged items was judged to have sufficiently poor performance to warrant exclusion from the Rasch analysis. By the end of 13 rounds of Rasch analyses, however, the eight items originally flagged for possible exclusion all had been removed because of a lessthan-good performance within a Rasch model. The 14 items that were retained conformed to Rasch
model requirements for a unidimensional model. Further item removal did not improve the tests for overall item fit to the model. The item parameters and model fit (x2 and residual) for the final instrument are shown in e-Table 2. The final summary statistics for the Rasch analyses showed an overall x2 of 149.3; degrees of freedom, 84; P , 0.01. The person separation index (analogous to Cronbach a) was 0.92, suggesting very good discriminant power among patients of different levels of severity. The item map (severity scale) of the 14 items by state (acute or stable) is shown in Figure 2. The severity of the items covered the severity range of the patients. The mean logit for the acute state was 0.16, which is almost at the center of the scaling range, whereas the mean for the stable state was 1.4 logits lower (ie, milder). This finding shows that the severity of items was very well targeted to the severity of patients’ COPD during an exacerbation. None of the surviving 14 items showed evidence of differential item functioning between stable and exacerbation states (analysis of variance P . 0.05 in each case). Post Hoc EFA A post hoc EFA was performed to explore the presence of domains within the 14 retained items. The goal was to test whether subcharacteristics of exacerbation could be identified empirically through the groupings of final EXACT items. Three factors were identified: breathlessness, cough and sputum, and chest symptoms. They accounted for 68% of the total variance (Table 2). Items that did not load onto a specific domain largely reflected systemic effects. The correlations between breathlessness and cough and sputum and between breathlessness and chest symptoms were 0.59 and 0.65, respectively, whereas the correlation between cough and sputum and chest symptoms was 0.66. As a sensitivity analysis, the data were analyzed both as continuous and as categorical
Figure 2. Rasch score distribution for patients and items by state (stable, acute). Severity on the x axis is a continuous scale in which a negative logit is milder and positive logit more severe. The top histogram shows the frequency distribution of patients with stable and exacerbating conditions. The bottom histogram shows the location (ie, severity) of the item categories. Note that one or more items may be closely adjacent, so more than one item may appear at one location. www.chestpubs.org
CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1391
Table 2—Three-Factor Solution for 14-Item EXACT (Postitem Reduction)a Item Number and Description 9. Breathless today 10. How breathless today 15. SOB with personal care 17. SOB with indoor 19. SOB with outdoor 2. Cough today 3. Mucus when coughing 1. Chest congested 6. Chest discomfort 8. Chest tight 4. Difficulty with mucus 20. Weak/tired 21. Sleep disturbed 23. Scared/worried
Factor 3 Breathlessness
Factor 2 Cough and Sputum
Factor 1 Chest Symptoms
0.614 0.604 0.800 0.877 0.863 20.120 20.024 20.002 20.085 0.170 0.147 0.455 0.250 0.336
0.009 0.199 0.118 20.101 20.120 0.745 0.863 0.199 20.016 20.002 0.408 0.062 0.401 0.071
0.204 20.095 20.101 0.015 20.023 0.150 20.116 0.707 0.929 0.706 0.233 0.285 0.169 0.373
Items are listed in the order of factor loading. Bold numbers indicate dominant factor loading. EXACT 5 Exacerbations of Chronic Pulmonary Disease Tool. See Table 1 legend for expansion of other abbreviation. aMaximum likelihood estimation with promax rotation.
variables; the item groupings using these two methods were the same (e-Table 1). The mean level of severity addressed by each domain was as follows: breathlessness, 20.36 logits; cough and sputum, 0.56 logits; and chest symptoms, 0.57 logits. The mean severity of the remaining items was 0.04 logits. These differences were small, and the distribution of the item response thresholds shows that all domains covered the scaling range symmetrically, although breathlessness captured the widest range of severity (Fig 3). Scoring The 14 items identified in this analysis were used to form the EXACT. The units of measurement within the Rasch model are logits, but for ease of interpretation of the EXACT, scores are transformed to a linear 0 to 100 interval scale in which higher scores indicate a worse health state. The three subdomains are scored similarly.
deliberately did not include sputum purulence because the intention was to produce an instrument for exacerbations of any etiology, not just those of bacterial origin. The final 14 items in the Rasch model had stable measurement properties between the two states, providing evidence that an exacerbation is characterized by a quantitative change in the severity of symptoms that are present in the stable state and worsen with exacerbation. In view of the extensive qualitative interviews undertaken to provide the items for this study11 and the range and depth of the current analysis, we suggest that this attempt is the most comprehensive to date to characterize the key symptomatic features of a COPD exacerbation. Previous symptomatic definitions have been limited to the cardinal symptoms of
Discussion This study identified a set of symptoms (14 items) that characterize an exacerbation of COPD, including breathlessness, chest tightness, chest congestion, cough, sputum production, chest discomfort, feeling weak or tired, sleep disturbance, and feeling scared or worried. No items performed differently between the stable and exacerbation conditions, and none appeared to be uniquely characteristic of an exacerbation. Sputum purulence or color—the item most expected to differ between states10—was removed because it demonstrated floor effects and was only weakly related to other items. This low positive response may reflect the inclusion criteria, which
Figure 3. Rasch item map showing the relative severity of the items that form domains within the 14 items. Each point indicates the level of severity at the boundary point between adjacent response categories for all 14 items. By convention, the mean severity indicated by all the items aggregated together is centered on zero. The unit is logits; mild severity is indicated by negative logits and more-severe severity by positive logits.
1392
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
cough, sputum, and breathlessness, but this comprehensive assessment demonstrates that exacerbation symptoms show a much broader spectrum than previously recognized. Rasch modeling, the principal methodology used in this analysis, tests each item for its fit to a unidimensional construct. The performance of the retained items and the overall quality of the fit of these items to the overall model suggest that the 14 items identified here fit together as an instrument that can be used to quantify COPD exacerbation severity. Three subgroups of items were identified empirically within this symptom complex: breathlessness, cough and sputum, and chest symptoms. It should be appreciated that the items forming these domains were retained because they associated well with all the other items as a measure of overall exacerbation severity. Future work will determine whether there are any material differences in the way in which these domains change with an exacerbation or whether they are simply definable symptom clusters within the overall exacerbation symptom group. The item reduction process used a combination of classic and modern item-response test theory to identify the set of items that best captured and quantified the key attributes of exacerbation. Several items flagged for exclusion were related to limitation of usual activity. Before they were eliminated, discussions took place with COPD experts to review the original classification of activity limitation as an attribute of exacerbations rather than as a consequence of worse symptoms. Qualitative patient descriptions of activity disturbance during exacerbation also were reexamined. Although patients clearly described a reduction in activity, this was tied closely to respiratory symptoms and tiredness, which were all captured through other items. The interval-scaling properties of an instrument developed using a Rasch model and scored in logits mean that a given change in score reflects the same change in severity, regardless of baseline severity. For example, an increase in EXACT score of 12 points from the patient’s baseline stable state could signal the onset of exacerbation whether occurring from a baseline score of 10, 20, or 30 units. By contrast, the result of the change (ie, the consequent severity of symptoms during the exacerbation) will be determined by baseline severity. The minimum changes that represent onset of and recovery from an exacerbation are under investigation, but the invariant properties of the instrument make it probable that these thresholds will remain the same across the scaling range. In summary, the 14 items that comprise the EXACT were derived from the descriptions of exacerbations provided by patients in extensive qualitawww.chestpubs.org
tive interviews and rigorously tested for their measurement properties. Measured daily over time, the EXACT should capture exacerbation onset (based on change from a stable state), severity, and duration. Acknowledgments Author contributions: Dr Jones: contributed to the study design, data analysis, and writing of the manuscript. Dr Chen: contributed to the data analysis and writing of the manuscript. Dr Wilcox: contributed to the analyses, report preparation, and the development of content for the manuscript. Dr Sethi: contributed to the study design, data analysis, and writing of the manuscript. Dr Leidy: contributed to the study design, execution, data analyses, report preparation, and writing of the manuscript. Financial/nonfinancial disclosures: The authors have reported to CHEST the following conflicts of interest: Dr Jones has received consultation or lecture fees from GlaxoSmithKline, AstraZeneca, Boehringer Ingelheim/Pfizer, Roche, Spiration, and Almirall but not for the work described in this article. Drs Chen, Wilcox, and Leidy are employed by United BioSource Corporation, which provides consulting and other research services to pharmaceutical, device, government, and nongovernment organizations. In their salaried positions, they work with a variety of companies and organizations and are precluded from receiving payment or honoraria directly from these organizations for services rendered. Dr Sethi has received consultation or lecture fees from Bayer, Boehringer Ingelheim, GlaxoSmithKline, AstraZeneca, Pfizer, Mpex, Schering-Plough, Novartis, and Nycomed but not for the work described in this article. Role of sponsors: Funds provided by sponsoring companies were unrestricted. Sponsors were kept informed throughout the EXACT development and validation process and invited to comment on protocols and study reports; representatives attended expert panel meetings as observers. All scientific decisions were made independently and without restrictions of any kind. Other contributions: We thank Carrie Lancos, BA, for formatting the manuscript to journal specifications. In addition, we acknowledge the participation and dedication of the following members of the United BioSource Corporation EXACT-PRO Initiative Team who participated in various stages of data collection, SAS programming, statistical analyses, and report preparation: Laurie Roberts, MPH; Randall Winnette, BS; Lindsey Murray, BA; Ren Yu, MA; Kellee Howard, MA; Jennifer Petrillo, BS; Charlotte Cates, MA; and Chris Thompson, BS. Members of the EXACTPRO Study Group contributed to the design of the work presented here and in the interpretation of study results but did not participate in the analysis of data or the writing of this manuscript. Additional information: The e-Appendices and e-Tables can be found in the Online Supplement at http://chestjournal.chestpubs. org/content/139/6/1388/suppl/DC1.
References 1. Seemungal TA, Donaldson GC, Bhowmik A, Jeffries DJ, Wedzicha JA. Time course and recovery of exacerbations in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2000;161(5):1608-1613. 2. Bach PB, Brown C, Gelfand SE, McCrory DC; American College of Physicians-American Society of Internal Medicine; American College of Chest Physicians. Management of acute exacerbations of chronic obstructive pulmonary disease: a summary and appraisal of published evidence. Ann Intern Med. 2001;134(7):600-620. 3. Metlay JP, Singer DE. Outcomes in lower respiratory tract infections and the impact of antimicrobial drug resistance. Clin Microbiol Infect. 2002;8(suppl 2):1-11. CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1393
4. Cazzola M, MacNee W, Martinez FJ, et al; American Thoracic Society; European Respiratory Society Task Force on outcomes of COPD. Outcomes for COPD pharmacological trials: from lung function to biomarkers. Eur Respir J. 2008;31(2):416-469. 5. Rennard SI, Leidy NK. Definition and severity of COPD exacerbations. In: Wedzicha W, Martinez F, eds. Exacerbations of Chronic Obstructive Pulmonary Disease (COPD). New York, NY: Informa Healthcare; 2009:1-14. 6. Calverley P, Pauwels Dagger R, Löfdahl CG, et al. Relationship between respiratory symptoms and medical treatment in exacerbations of COPD. Eur Respir J. 2005;26(3): 406-413. 7. Anthonisen NR, Manfreda J, Warren CP, Hershfield ES, Harding GK, Nelson NA. Antibiotic therapy in exacerbations of chronic obstructive pulmonary disease. Ann Intern Med. 1987;106(2):196-204. 8. Balter MS, La Forge J, Low DE, et al. Canadian guidelines for the management of acute exacerbations of chronic bronchitis. Can Respir J. 2003;10(suppl B):3B-32B 9. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest. 2000;117(5 suppl 2):398S-401S. 10. Stockley RA, O’Brien C, Pye A, Hill SL. Relationship of sputum color to nature and outpatient management of acute exacerbations of COPD. Chest. 2000;117(6):1638-1645.
11. Leidy NK, Wilcox T, Jones PW, et al. Development of the EXAcerbations of Chronic Obstructive Pulmonary Disease Tool (EXACT): a patient-reported outcome measure. Value Health. 2010;13(8):965-975. 12. Leidy NK, Wilcox T, Jones PW, et al. Standardizing measurement of COPD exacerbations: reliability and validity of a patient-reported diary [published online ahead of print September 2, 2010]. Am J Respir Crit Care Med. doi:10.1164/ rccm.201005-0762OC. 13. Meguro M, Barley EA, Spencer S, Jones PW. Development and validation of an improved, COPD-specific version of the St. George Respiratory Questionnaire. Chest. 2007; 132(2):456-463. 14. O’Leary CJ, Jones PW. The influence of decisions made by developers on health status questionnaire content . Qual Life Res. 1998;7(6):545-550. 15. Reeve B, Fayers P. Applying item response theory modeling for evaluating questionnaire item and scale properties. In: Fayers P, ed. Assessing Quality of Life in Clinical Trials: Methods and Practice. South Melbourne, Victoria, Australia: Oxford University Press; 2005:55-73. 16. Muthén LK, Muthén B. Mplus User’s Guide. 3rd ed. Los Angeles, CA: Muthén & Muthén, 1998-2004. 17. Andrich D. Controversy and the Rasch model: a characteristic of incompatible paradigms? Med Care. 2004;42(suppl 1):I7-I16.
1394
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
Original Research COPD
Characterizing and Quantifying the Symptomatic Features of COPD Exacerbations Paul W. Jones, PhD; Wen-Hung Chen, PhD; Teresa K. Wilcox, PhD; Sanjay Sethi, MD; and Nancy Kline Leidy, PhD; for the EXACT-PRO Study Group*
Background: There is a need for a standardized, valid, and reliable instrument for quantifying exacerbations of COPD. The objective of this study was to identify symptom items that characterize COPD exacerbations to form a new patient diary for evaluating exacerbation frequency, severity, and duration. Methods: Twenty-three symptom items identified from patient interviews were administered to 410 patients with COPD aged (mean ⫾ SD) 65 ⫾ 10 years with stable FEV1 of 51% predicted ⫾ 20% predicted and 1.8 ⫾ 1.8 exacerbations in the preceding 12 months. A total of 222 patients had a physician-diagnosed exacerbation; 188 were stable. Item-level analyses (floor and ceiling effects, criterion keying, item-total correlation) were used in the first stage of item reduction. Further reduction was conducted using Rasch model and descriptive item analyses. Exploratory factor analysis was performed on the items that survived the exclusion process. Results: No item behaved differently between stable and exacerbation conditions. One item was removed after item-level analysis, and eight were removed following Rasch analysis. Together, the surviving 14 items met the criteria for a unidimensional measure of exacerbation severity. Internal consistency (person separation index) was excellent at 0.92. Post hoc exploratory factor analysis revealed one dominant factor, with three domains (breathlessness, cough and sputum, and chest symptoms) that accounted for 68% of the variance. Conclusions: An exacerbation appears to be a quantitative rather than qualitative change from the stable state. This analysis identified a range of symptoms that form a unidimensional construct of overall exacerbation severity. The 14 items identified form the Exacerbations of Chronic Pulmonary Disease Tool (EXACT), a daily diary for detecting and quantifying exacerbation severity in COPD. CHEST 2011; 139(6):1388–1394 Abbreviations: EFA 5 exploratory factor analysis; EXACT 5 Exacerbations of Chronic Pulmonary Disease Tool; SGRQ-C 5 St. George Respiratory Questionnaire-COPD
are an important feature of COPD, Exacerbations leading to significant morbidity and mortality, and 1
they often are used as outcomes in clinical studies.2,3 Currently, there is no agreed-upon method for quantifying exacerbations of COPD4,5; the most widely used method is based on an increase in symptoms that requires treatment in a clinic, ED, or hospital. Exacerbations identified using this type of eventbased definition do not always correspond with those identified using symptom-based definitions,6 so methods that detect and quantify exacerbations using explicit clinical criteria are needed. COPD exacerbations are characterized by an acute worsening of symptoms7,8 and have been defined as
“a sustained worsening of the patient’s condition, from the stable state and beyond normal day-today variations, that is acute in onset and necessitates a change in regular medication.”9 This definition does not specify the components that worsen. Sputum color has attracted attention because it may be predictive of a high bacterial load,10 but less attention has been paid to other symptoms that together may serve as a measure of severity. Several groups have used exacerbation diary cards1; however, there has never been a formal analysis of exacerbation symptoms to identify those that can provide reliable detection and quantification of changes.
1388
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
This study describes part of the development of a validated diary card for detecting and quantifying exacerbations: the Exacerbations of Chronic Pulmonary Disease Tool (EXACT) Patient-Reported Outcome (known as EXACT-PRO). It examined items identified through extensive patient interviews about the symptomatic features of an exacerbation.11 We tested two hypotheses: (1) that an exacerbation is a quantifiable change in the intensity of preexisting symptoms and (2) that the transition from stable state to exacerbation may be characterized by the emergence or shift in relative importance of a new symptom or symptom cluster. Materials and Methods Study Design and Participant Recruitment A total of 23 potential items were developed from a comprehensive series of interviews and qualitative analyses in patients with COPD.11 These items assessed patients’ breathlessness, cough, sputum, chest discomfort, difficulty breathing, activity limitations, feelings of tiredness or weakness, sleep disturbance, and scared or worried feelings about their condition. Patients completed the diary each evening before bedtime, reflecting back on “today” and rating the severity of each attribute or item on a 3-, 5-, or 6-point ordinal scale (eg, from not at all to extremely). See e-Appendix 2 for additional information. The candidate items and final instrument were tested in a prospective, two-group observational study. Methods and results for tests of reliability, validity, and responsiveness of the final instrument are presented elsewhere.12 Briefly, the groups were stable patients (stable) and those with a clinician-confirmed exacerbation (acute). Participants were recruited from 38 sites in the United States. Inclusion criteria for both groups were previous diagnosis of COPD or chronic bronchitis, age . 40 years, smoking history . 10 pack-years, telephone landline for uploading electronic diary data, and ability to read and speak English. Exclusion criteria included concurrent diagnosis of asthma, clinically relevant bronchiectasis, visual or cognitive impairment that would interfere with completing questionnaires, and a concurrent medical or psychiatric condition that would affect participation. Stable participants were required to have a history of one or more acute exacerbations requiring systemic corticosteroids or Manuscript received May 14, 2010; revision accepted October 2, 2010. Affiliations: From St. George’s (Dr Jones), University of London, London, England; United BioSource Corporation (Drs Chen, Wilcox, and Leidy), Bethesda, MD; and University at Buffalo (Dr Sethi), SUNY, Buffalo, NY. *A list of study group members is available in e-Appendix 1. Funding/Support: The following companies have provided unrestricted funds to support the EXACT-PRO Initiative: AstraZeneca, GlaxoSmithKline, Pfizer Inc, Boehringer Ingelheim, Merck & Company, Sepracor Inc, Forest Laboratories Inc, Novartis, Schering-Plough, Adams Respiratory Therapeutics, Bayer, Atlanta AG (Nycomed), and Ortho-McNeil. Correspondence to: Nancy Kline Leidy, PhD, United BioSource Corporation, 7101 Wisconsin Ave, Ste 600, Bethesda, MD 20814; e-mail: [email protected] © 2011 American College of Chest Physicians. Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (http://www.chestpubs.org/ site/misc/reprints.xhtml). DOI: 10.1378/chest.10-1240 www.chestpubs.org
antibiotics within the previous 24 months and no clinic, ED, or hospital visit for an exacerbation within the previous 60 days. Acute participants were enrolled at a clinic visit triggered by worsening symptoms. They were recruited after full clinical evaluation, including physical examination and tests thought appropriate by the physician. A consensus exacerbation definition was used for determining recruitment.9 Sputum purulence was not a requirement. Procedure The study protocol was approved by the Sterling Institutional Review Board (Atlanta, Georgia) (IRB ID #2405), and all participants provided written consent prior to initiation of data collection. All participants completed a 27-item diary, including the pool of 23 items under test, each evening on a personal digital assistant (Palm Tungsten E; Hewlett-Packard Company; Palo Alto, California). Stable patients completed the diary for 7 days. Acute patients completed the diary on days 1 to 29 (⫾ 2) of the acute event (day 1 being the day of first presentation with the exacerbation), returning to the clinic on day 60 (⫾ 7 days) for 7 days of diary assessment (days 60-67) during a stable state. All participants completed the St. George Respiratory QuestionnaireCOPD (SGRQ-C).13 Physicians and patients also completed global assessments of health state, exacerbation severity, and global assessments of change. Stable-state FEV1 was recorded. All recruitment and study procedures met Health Insurance Portability and Accountability Act requirements. Additional information on study methods is provided in e-Appendix 2. Analyses A statistical analysis plan was developed prior to initiating the analyses, prespecifying the stepwise item reduction process to be used. The data analysis for this article was generated using SAS/BASE and SAS/STAT software version 9.1.3 of the SAS System for PC (SAS Institute; Cary, North Carolina). Item Analyses: The following item characteristics were examined using day-1 data of the pooled sample and by group (acute, stable): mean, frequencies of item responses, percentages of minimum and maximum responses, percent missing, inter-item correlation, and item-total correlation. An item was flagged for potential exclusion if it showed a minimum response (floor effect) . 50%; maximum response (ceiling effect) . 50%; interitem correlation . 0.80, suggesting that it may be redundant; or interitem correlation , 0.20, suggesting that it did not fit with the others. Each item also was examined by criterion keying14 using clinician global rating of patient health on day 1 as the external criterion. Similar tests were used to identify items with sex or age bias. Items were flagged for possible exclusion if they had ⱕ 4% shared variance with clinician rating of patient health state or ⱖ 6% shared variance with sex or age. Factor Analysis: To assess the presence of specific domains within the item set, exploratory factor analysis (EFA) was performed on the initial 23 items using a random split-half data sample from day 1. This analysis was carried out using pooled data and separately by groups. Methodology details are provided in e-Appendix 2. Rasch Model Analysis: With the Rasch model, the more severe the exacerbation (as judged by the patients’ response to all of the items combined), the higher the probability of a positive response to any given constituent item.15 The model assumes that all items have uniform discrimination power to distinguish between high and low severity. Statistical tests were used to test the fit of each individual item to the mathematically determined relationship with overall severity that is seen with the Rasch model, if the item CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1389
was a perfect measure of exacerbation severity. More details are available in e-Appendix 2 and in the article by Meguro et al13 and its online supplement. Rasch analyses were conducted using day-1 data from the entire pooled sample. In addition, each item was examined for differential item functioning to test whether it behaved consistently between stable state and exacerbation. This examination was a formal test of the hypotheses that an exacerbation is either a quantitative or a qualitative change from the stable state. Item severity was measured in terms of logits—the units in a Rasch model. The logit is the log odds for an item of given severity being affirmed by 50% of the study population. The patient’s severity can be measured in the same units. With Rasch analysis, item reduction is an iterative process of individual item deletion, beginning with the worst-fitting items, with consideration also given to item descriptive statistics, factor analyses, qualitative data, and clinical experience. Each time an item is deleted, the data are reanalyzed to evaluate fit of the remaining items to the model. The process continues until a core set of items is identified. SAS, version 8.2 statistical software was used to characterize the sample and item-level descriptive statistics. Mplus (Muthén & Muthén; Los Angeles, California) was used for EFA,16 and RUMM2020 (RUMM Laboratory Pty Ltd; Duncraig, Western Australia, Australia) was used for Rasch modeling.17
48% were men. Stable-state FEV1 was 51% predicted ⫾ 20% predicted. Using independent sample t tests for continuous variables and x2 tests for categorical variables, there were no significant differences between the groups on any demographic or clinical characteristics with two exceptions. The acute group reported more frequent events the previous year (2.2 ⫾ 0.5) than the stable group (1.3 ⫾ 1.2) (t 5 5.69 using the Satterthwaite method for unequal variances; P , .001). On day 1, SGRQ-C scores of the acute group (62.1 ⫾ 17.7) were significantly higher than those of the stable group (50.8 ⫾ 19.9) (t 5 5.97 pooled; P , .0001).
Results
Scree plots were created using all of the candidate items. One dominant factor was seen in the pooled sample; its eigenvalue was 9.7, and it accounted for 42% of the total variance (Fig 1). A dominant singlefactor pattern also was seen in both the acute group
Sample There were 222 acute patients and 188 stable patients. Their mean ⫾ SD age was 65 ⫾ 10 years, and
Item-Level Analyses During item-level analyses, items 5, 7, 11, 12, 13, 14, 16, 18, and 21 were flagged for possible exclusion. Item-level descriptive statistics and rationale for flagging are shown in Table 1. Factor Analysis
Table 1—Item Descriptive Statistics for the Pooled Sample Day 1 (N 5 410) Item Number and Description 1. Chest congested 2. Cough 3. Mucus when coughing 4. Difficulty with mucus 5. Mucus color 6. Chest discomfort 7. Chest hurt 8. Chest tight 9. Breathless 10. How breathless 11. SOB while sittingb 12. Difficulty breathing while sittingb 13. How active 14. Usual personal care 15. SOB with personal care 16. Usual indoor 17. SOB with indoor 18. Usual outdoor 19. SOB with outdoor 20. Weak/tired 21. Sleep disturbed 22. How much sleep 23. Scared/worried
Flag
Mean ⫾ SD
Range
Median
Floor, No. (%)
Ceiling, No. (%)
Item-Total Correlationa
… … … … *§ … * … … … X*§ *§
2.5 ⫾ 1.05 3.1 ⫾ 1.02 2.5 ⫾ 1.05 2.6 ⫾ 1.29 1.8 ⫾ 1.13 2.1 ⫾ 0.96 1.9 ⫾ 0.93 2.1 ⫾ 0.97 2.7 ⫾ 0.96 2.8 ⫾ 0.99 1.7 ⫾ 0.84 1.7 ⫾ 0.80
1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5 1-5
2 3 3 3 1 2 2 2 3 3 2 2
79 (19.3) 34 (8.3) 95 (23.2) 114 (27.8) 241 (58.8) 127 (31.0) 182 (44.4) 118 (28.8) 37 (9.0) 36 (8.8) 190 (46.3) 190 (46.3)
14 (3.4) 23 (5.6) 7 (1.7) 31 (7.6) 10 (2.4) 4 (1.0) 3 (0.7) 6 (1.5) 14 (3.4) 28 (6.8) 2 (0.5) 1 (0.2)
0.73 0.58 0.48 0.59 0.35 0.67 0.68 0.71 0.71 0.59 0.74 0.72
X*O *§ … X … X§ … … … XO …
3.3 ⫾ 0.72 1.5 ⫾ 0.56 2.3 ⫾ 1.03 2.2 ⫾ 0.84 2.6 ⫾ 1.42 2.3 ⫾ 0.77 2.7 ⫾ 1.61 2.5 ⫾ 0.97 2.3 ⫾ 1.02 3.4 ⫾ 0.79 2.3 ⫾ 1.13
1-5 1-3 1-6 1-3 1-6 1-3 1-6 1-5 1-5 1-5 1-5
3 1 2 2 2 3 2 2 2 3 2
1 (0.2) 232 (56.6) 102 (24.9) 116 (28.3) 94 (22.9) 76 (18.5) 111 (27.1) 67 (16.3) 95 (23.2) 3 (0.7) 121 (29.5)
21 (5.1) 13 (3.2) 3 (0.7) 179 (43.7) 35 (8.5) 216 (52.7) 53 (12.9) 10 (2.4) 9 (2.2) 34 (8.3) 22 (5.4)
0.20 0.54 0.66 0.38 0.67 0.32 0.63 0.68 0.57 0.24 0.59
There were no missing item-level data by design. * 5 based on pooled sample analyses; § 5 based on acute sample analyses; O 5 based on exploratory factor analysis loading , 0.3; SOB 5 shortness of breath; X 5 based on stable sample analyses. aPearson product-moment correlation (r). bInter-item correlation (r) 5 0.88. 1390
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
Figure 1. Scree plots from the exploratory factor analysis (EFA) for the pooled, acute, and stable samples with data from day 1. The x axis shows the number of identified factors. The y axis shows the eigenvalue associated with each factor; this is related to the amount of variance associated with each factor. One factor had the highest eigenvalue showing that this was the dominant item grouping.
(eigenvalue 7.6; 33% of variance) and the stable group (eigenvalue 8.3; 36% of variance). As a further test, the analysis was repeated with the item responses used as categorical variables. The findings were the same (e-Table 1), which lent support to the starting hypothesis for the Rasch analysis that there is an underlying unidimensional structure to the data. Rasch Analysis The Rasch analyses started with 22 items because item 13 was excluded due to poor performance across multiple criteria. None of the other eight flagged items was judged to have sufficiently poor performance to warrant exclusion from the Rasch analysis. By the end of 13 rounds of Rasch analyses, however, the eight items originally flagged for possible exclusion all had been removed because of a lessthan-good performance within a Rasch model. The 14 items that were retained conformed to Rasch
model requirements for a unidimensional model. Further item removal did not improve the tests for overall item fit to the model. The item parameters and model fit (x2 and residual) for the final instrument are shown in e-Table 2. The final summary statistics for the Rasch analyses showed an overall x2 of 149.3; degrees of freedom, 84; P , 0.01. The person separation index (analogous to Cronbach a) was 0.92, suggesting very good discriminant power among patients of different levels of severity. The item map (severity scale) of the 14 items by state (acute or stable) is shown in Figure 2. The severity of the items covered the severity range of the patients. The mean logit for the acute state was 0.16, which is almost at the center of the scaling range, whereas the mean for the stable state was 1.4 logits lower (ie, milder). This finding shows that the severity of items was very well targeted to the severity of patients’ COPD during an exacerbation. None of the surviving 14 items showed evidence of differential item functioning between stable and exacerbation states (analysis of variance P . 0.05 in each case). Post Hoc EFA A post hoc EFA was performed to explore the presence of domains within the 14 retained items. The goal was to test whether subcharacteristics of exacerbation could be identified empirically through the groupings of final EXACT items. Three factors were identified: breathlessness, cough and sputum, and chest symptoms. They accounted for 68% of the total variance (Table 2). Items that did not load onto a specific domain largely reflected systemic effects. The correlations between breathlessness and cough and sputum and between breathlessness and chest symptoms were 0.59 and 0.65, respectively, whereas the correlation between cough and sputum and chest symptoms was 0.66. As a sensitivity analysis, the data were analyzed both as continuous and as categorical
Figure 2. Rasch score distribution for patients and items by state (stable, acute). Severity on the x axis is a continuous scale in which a negative logit is milder and positive logit more severe. The top histogram shows the frequency distribution of patients with stable and exacerbating conditions. The bottom histogram shows the location (ie, severity) of the item categories. Note that one or more items may be closely adjacent, so more than one item may appear at one location. www.chestpubs.org
CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1391
Table 2—Three-Factor Solution for 14-Item EXACT (Postitem Reduction)a Item Number and Description 9. Breathless today 10. How breathless today 15. SOB with personal care 17. SOB with indoor 19. SOB with outdoor 2. Cough today 3. Mucus when coughing 1. Chest congested 6. Chest discomfort 8. Chest tight 4. Difficulty with mucus 20. Weak/tired 21. Sleep disturbed 23. Scared/worried
Factor 3 Breathlessness
Factor 2 Cough and Sputum
Factor 1 Chest Symptoms
0.614 0.604 0.800 0.877 0.863 20.120 20.024 20.002 20.085 0.170 0.147 0.455 0.250 0.336
0.009 0.199 0.118 20.101 20.120 0.745 0.863 0.199 20.016 20.002 0.408 0.062 0.401 0.071
0.204 20.095 20.101 0.015 20.023 0.150 20.116 0.707 0.929 0.706 0.233 0.285 0.169 0.373
Items are listed in the order of factor loading. Bold numbers indicate dominant factor loading. EXACT 5 Exacerbations of Chronic Pulmonary Disease Tool. See Table 1 legend for expansion of other abbreviation. aMaximum likelihood estimation with promax rotation.
variables; the item groupings using these two methods were the same (e-Table 1). The mean level of severity addressed by each domain was as follows: breathlessness, 20.36 logits; cough and sputum, 0.56 logits; and chest symptoms, 0.57 logits. The mean severity of the remaining items was 0.04 logits. These differences were small, and the distribution of the item response thresholds shows that all domains covered the scaling range symmetrically, although breathlessness captured the widest range of severity (Fig 3). Scoring The 14 items identified in this analysis were used to form the EXACT. The units of measurement within the Rasch model are logits, but for ease of interpretation of the EXACT, scores are transformed to a linear 0 to 100 interval scale in which higher scores indicate a worse health state. The three subdomains are scored similarly.
deliberately did not include sputum purulence because the intention was to produce an instrument for exacerbations of any etiology, not just those of bacterial origin. The final 14 items in the Rasch model had stable measurement properties between the two states, providing evidence that an exacerbation is characterized by a quantitative change in the severity of symptoms that are present in the stable state and worsen with exacerbation. In view of the extensive qualitative interviews undertaken to provide the items for this study11 and the range and depth of the current analysis, we suggest that this attempt is the most comprehensive to date to characterize the key symptomatic features of a COPD exacerbation. Previous symptomatic definitions have been limited to the cardinal symptoms of
Discussion This study identified a set of symptoms (14 items) that characterize an exacerbation of COPD, including breathlessness, chest tightness, chest congestion, cough, sputum production, chest discomfort, feeling weak or tired, sleep disturbance, and feeling scared or worried. No items performed differently between the stable and exacerbation conditions, and none appeared to be uniquely characteristic of an exacerbation. Sputum purulence or color—the item most expected to differ between states10—was removed because it demonstrated floor effects and was only weakly related to other items. This low positive response may reflect the inclusion criteria, which
Figure 3. Rasch item map showing the relative severity of the items that form domains within the 14 items. Each point indicates the level of severity at the boundary point between adjacent response categories for all 14 items. By convention, the mean severity indicated by all the items aggregated together is centered on zero. The unit is logits; mild severity is indicated by negative logits and more-severe severity by positive logits.
1392
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
cough, sputum, and breathlessness, but this comprehensive assessment demonstrates that exacerbation symptoms show a much broader spectrum than previously recognized. Rasch modeling, the principal methodology used in this analysis, tests each item for its fit to a unidimensional construct. The performance of the retained items and the overall quality of the fit of these items to the overall model suggest that the 14 items identified here fit together as an instrument that can be used to quantify COPD exacerbation severity. Three subgroups of items were identified empirically within this symptom complex: breathlessness, cough and sputum, and chest symptoms. It should be appreciated that the items forming these domains were retained because they associated well with all the other items as a measure of overall exacerbation severity. Future work will determine whether there are any material differences in the way in which these domains change with an exacerbation or whether they are simply definable symptom clusters within the overall exacerbation symptom group. The item reduction process used a combination of classic and modern item-response test theory to identify the set of items that best captured and quantified the key attributes of exacerbation. Several items flagged for exclusion were related to limitation of usual activity. Before they were eliminated, discussions took place with COPD experts to review the original classification of activity limitation as an attribute of exacerbations rather than as a consequence of worse symptoms. Qualitative patient descriptions of activity disturbance during exacerbation also were reexamined. Although patients clearly described a reduction in activity, this was tied closely to respiratory symptoms and tiredness, which were all captured through other items. The interval-scaling properties of an instrument developed using a Rasch model and scored in logits mean that a given change in score reflects the same change in severity, regardless of baseline severity. For example, an increase in EXACT score of 12 points from the patient’s baseline stable state could signal the onset of exacerbation whether occurring from a baseline score of 10, 20, or 30 units. By contrast, the result of the change (ie, the consequent severity of symptoms during the exacerbation) will be determined by baseline severity. The minimum changes that represent onset of and recovery from an exacerbation are under investigation, but the invariant properties of the instrument make it probable that these thresholds will remain the same across the scaling range. In summary, the 14 items that comprise the EXACT were derived from the descriptions of exacerbations provided by patients in extensive qualitawww.chestpubs.org
tive interviews and rigorously tested for their measurement properties. Measured daily over time, the EXACT should capture exacerbation onset (based on change from a stable state), severity, and duration. Acknowledgments Author contributions: Dr Jones: contributed to the study design, data analysis, and writing of the manuscript. Dr Chen: contributed to the data analysis and writing of the manuscript. Dr Wilcox: contributed to the analyses, report preparation, and the development of content for the manuscript. Dr Sethi: contributed to the study design, data analysis, and writing of the manuscript. Dr Leidy: contributed to the study design, execution, data analyses, report preparation, and writing of the manuscript. Financial/nonfinancial disclosures: The authors have reported to CHEST the following conflicts of interest: Dr Jones has received consultation or lecture fees from GlaxoSmithKline, AstraZeneca, Boehringer Ingelheim/Pfizer, Roche, Spiration, and Almirall but not for the work described in this article. Drs Chen, Wilcox, and Leidy are employed by United BioSource Corporation, which provides consulting and other research services to pharmaceutical, device, government, and nongovernment organizations. In their salaried positions, they work with a variety of companies and organizations and are precluded from receiving payment or honoraria directly from these organizations for services rendered. Dr Sethi has received consultation or lecture fees from Bayer, Boehringer Ingelheim, GlaxoSmithKline, AstraZeneca, Pfizer, Mpex, Schering-Plough, Novartis, and Nycomed but not for the work described in this article. Role of sponsors: Funds provided by sponsoring companies were unrestricted. Sponsors were kept informed throughout the EXACT development and validation process and invited to comment on protocols and study reports; representatives attended expert panel meetings as observers. All scientific decisions were made independently and without restrictions of any kind. Other contributions: We thank Carrie Lancos, BA, for formatting the manuscript to journal specifications. In addition, we acknowledge the participation and dedication of the following members of the United BioSource Corporation EXACT-PRO Initiative Team who participated in various stages of data collection, SAS programming, statistical analyses, and report preparation: Laurie Roberts, MPH; Randall Winnette, BS; Lindsey Murray, BA; Ren Yu, MA; Kellee Howard, MA; Jennifer Petrillo, BS; Charlotte Cates, MA; and Chris Thompson, BS. Members of the EXACTPRO Study Group contributed to the design of the work presented here and in the interpretation of study results but did not participate in the analysis of data or the writing of this manuscript. Additional information: The e-Appendices and e-Tables can be found in the Online Supplement at http://chestjournal.chestpubs. org/content/139/6/1388/suppl/DC1.
References 1. Seemungal TA, Donaldson GC, Bhowmik A, Jeffries DJ, Wedzicha JA. Time course and recovery of exacerbations in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2000;161(5):1608-1613. 2. Bach PB, Brown C, Gelfand SE, McCrory DC; American College of Physicians-American Society of Internal Medicine; American College of Chest Physicians. Management of acute exacerbations of chronic obstructive pulmonary disease: a summary and appraisal of published evidence. Ann Intern Med. 2001;134(7):600-620. 3. Metlay JP, Singer DE. Outcomes in lower respiratory tract infections and the impact of antimicrobial drug resistance. Clin Microbiol Infect. 2002;8(suppl 2):1-11. CHEST / 139 / 6 / JUNE, 2011
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians
1393
4. Cazzola M, MacNee W, Martinez FJ, et al; American Thoracic Society; European Respiratory Society Task Force on outcomes of COPD. Outcomes for COPD pharmacological trials: from lung function to biomarkers. Eur Respir J. 2008;31(2):416-469. 5. Rennard SI, Leidy NK. Definition and severity of COPD exacerbations. In: Wedzicha W, Martinez F, eds. Exacerbations of Chronic Obstructive Pulmonary Disease (COPD). New York, NY: Informa Healthcare; 2009:1-14. 6. Calverley P, Pauwels Dagger R, Löfdahl CG, et al. Relationship between respiratory symptoms and medical treatment in exacerbations of COPD. Eur Respir J. 2005;26(3): 406-413. 7. Anthonisen NR, Manfreda J, Warren CP, Hershfield ES, Harding GK, Nelson NA. Antibiotic therapy in exacerbations of chronic obstructive pulmonary disease. Ann Intern Med. 1987;106(2):196-204. 8. Balter MS, La Forge J, Low DE, et al. Canadian guidelines for the management of acute exacerbations of chronic bronchitis. Can Respir J. 2003;10(suppl B):3B-32B 9. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest. 2000;117(5 suppl 2):398S-401S. 10. Stockley RA, O’Brien C, Pye A, Hill SL. Relationship of sputum color to nature and outpatient management of acute exacerbations of COPD. Chest. 2000;117(6):1638-1645.
11. Leidy NK, Wilcox T, Jones PW, et al. Development of the EXAcerbations of Chronic Obstructive Pulmonary Disease Tool (EXACT): a patient-reported outcome measure. Value Health. 2010;13(8):965-975. 12. Leidy NK, Wilcox T, Jones PW, et al. Standardizing measurement of COPD exacerbations: reliability and validity of a patient-reported diary [published online ahead of print September 2, 2010]. Am J Respir Crit Care Med. doi:10.1164/ rccm.201005-0762OC. 13. Meguro M, Barley EA, Spencer S, Jones PW. Development and validation of an improved, COPD-specific version of the St. George Respiratory Questionnaire. Chest. 2007; 132(2):456-463. 14. O’Leary CJ, Jones PW. The influence of decisions made by developers on health status questionnaire content . Qual Life Res. 1998;7(6):545-550. 15. Reeve B, Fayers P. Applying item response theory modeling for evaluating questionnaire item and scale properties. In: Fayers P, ed. Assessing Quality of Life in Clinical Trials: Methods and Practice. South Melbourne, Victoria, Australia: Oxford University Press; 2005:55-73. 16. Muthén LK, Muthén B. Mplus User’s Guide. 3rd ed. Los Angeles, CA: Muthén & Muthén, 1998-2004. 17. Andrich D. Controversy and the Rasch model: a characteristic of incompatible paradigms? Med Care. 2004;42(suppl 1):I7-I16.
1394
Original Research
Downloaded from chestjournal.chestpubs.org by Kimberly Henricks on June 8, 2011 © 2011 American College of Chest Physicians