Evaluation of the Norwegian version of the Dyspnoea-12 questionnaire in patients with COPD

Data collection

The study included 249 potentially eligible patients with COPD aged 18 years and over attending a 4-week inpatient pulmonary rehabilitation programme at the LHL Hospital Gardermoen in South-Eastern Norway over a 6-month period from June 2018. Patients were included if they had no cognitive impairments and sufficient understanding of the Norwegian language. The self-completed pen and paper questionnaire which included the D-12, were completed within the first week of attendance. Patients received the questionnaire in a group setting for completion at the end of the session in the same room or in their own room. There was a nurse available to answer any questions during completion and on collection from patients who completed it in their room.

Patient-reported outcomes

The D-12 has 12 items with four-point scales of none, mild, moderate and severe in relation to breathlessness ‘these days’.11 The first seven items sum to the physical domain with scores from 0 to 21. The remaining five items sum to the affective domain scores from 0 to 15. All twelve items sum to the D-12 total scores from 0 to 36. Higher scores represent greater severity.11 Translation of the original English version of the D-12 into Norwegian followed international recommendations for forward-backwards translation of patient-reported outcome measures (PROMs)17 18 including two independent forward and backwards translations. The original time frame of ‘these days’ was retained following translation. Minor discrepancies between translations were resolved by discussion within the research group which included the translators. The Norwegian translation (online supplemental figure S1), including instructions, questions and scaling, was found to be acceptable and easy to understand following interviews with 20 patients with COPD who did not participate in the main study.

In addition to the D-12, the questionnaire included the COPD Assessment Test (CAT) and Hospital Anxiety and Depression Scale (HADS). The CAT has eight items with six-point scales with endpoint-only descriptors.19 Four items relate to symptoms and four relate to broader aspects of health and quality of life. The items sum to give a score from 0 to 40, where higher scores represent a greater impact of COPD on health. The CAT has been widely translated and has evidence for reliability, validity and responsiveness.19 20 The HADS was developed to assess mood disorder in non-psychiatric hospital outpatients and has 14 items with four-point descriptive scales that refer to the last week.21 Seven items sum to give the anxiety scale and the remainder sum to give the depression scale. Scores range from 0 to 21 with higher scores representing more symptoms. The HADS has been widely used alongside the D-129 and has evidence for internal consistency and validity in Norwegian patients with COPD.22

Clinical measures

Dyspnoea was assessed by the mMRC Dyspnoea Scale, as registered by physicians in the electronic medical record. The single-item mMRC assesses disability attributable to shortness of breath and has been widely used in Norwegian healthy and patient populations.10 Spirometry was conducted according to American Thoracic Society/European Respiratory Society recommendations using appropriate reference values.23 A 6 min walk test followed standard recommendations reporting walked distance in metres.24 Dyspnoea before and after the walk test was assessed using the Borg CR10 scale with a range of 0–10.25 Further information retrieved from medical records included age, sex, smoking status, employment status, body mass index, number of COPD-related hospitalisation in the last year, whether there was a current acute exacerbation, use of long-term oxygen treatment and comorbidities.

Statistical analysis

Descriptive statistics are presented with the mean (SD), median (range) or number (%), as appropriate. Missing data and floor and ceiling effects were assessed at the item and domain level. CFA with robust-weighted least squares (WLSMV)26 was used to assess the structural validity of the D-12, or the extent to which the item scores adequately contribute to physical, affective and overall domains.16 Model fit was assessed by the root mean square error approximation (RMSEA, acceptable fit if <0.06), the Comparative Fit Index (CFI, acceptable fit if >0.95, poor fit if <0.90, otherwise marginal) and the Tucker Lewis Index (TLI, acceptable fit if >0.95, poor fit if <0.90, otherwise marginal).16 27

Unidimensionality of D-12 domains was tested using the Rasch partial credit model. This extension of the Rasch model for polytomous items has separable item and person parameters, sufficient statistics and conjoint additivity allowing item and person comparisons.28 Overall and item fit statistics were used to assess whether items within the domains fitted the one-dimensional model. Item fit was assessed with the χ² statistic, standardised residuals, which should be between ±2.5, and item characteristic curves. Local independence was assessed through examination of the residual correlation matrix, coefficients of ≥0.2 indicating item redundancy.29 30

Differential item functioning (DIF) occurs when different groups with the same levels of construct being measured, respond differently to an item. Consistent differences across the construct represent uniform DIF. Inconsistent differences that vary across the construct represent non-uniform DIF. Both forms of DIF were assessed for gender, pensionable age (67 years), receipt of disability payment and four comorbidities: anxiety/depression, asthma, hypertension and obesity. Differences of ≥0.5 logits in item difficulties were considered meaningful in interpretation of DIF.31 32

Internal consistency was assessed by the Person Separation Index (PSI)33 and Cronbach’s alpha.16 Estimates of 0.70 and 0.90 are necessary for group and individual comparisons respectively.16

Hypothesis testing was used to further assess the validity of D-12 scores through comparisons with those for the HADS and clinical measures.9–15 Criteria for expected levels of correlation followed those from a systematic review of PROMs.34 First, correlations ≥0.60 were expected for scores assessing the same construct: D-12 affective and HADS anxiety and depression. Second, correlations <0.60 and ≥0.30 were expected for scores assessing largely related but dissimilar constructs: D-12 domains/total and CAT. Third, correlations <0.50 and ≥0.20 were expected for moderately related but dissimilar constructs: D-12 physical/total and HADS, mMRC, Borg scores, FEV₁, number of COPD hospitalisations in the last year and 6 min walk distance. Fourth, correlations <0.30 were expected for weakly related or unrelated constructs: D-12 affective and mMRC, Borg scores, FEV₁, number of COPD hospitalisations in the last year and 6 min walk distance. At least 75% of the results should correspond with the hypotheses.16 34 Pearson’s correlation was used to aid comparisons with published studies, and findings are shown alongside those from a systematic review.9

Statistical analyses were undertaken using RUMM2020 V.4.1 (Rumm Laboratory, Perth, Western Australia), Lisrel version 7 (Muthe’n & Muthe’ n, Los Angeles, CA) and Stata V.15.0 (StataCorp).

Patient and public involvement

Patients and public were not involved in development of the research question, study design, study conduct or dissemination.

Data collection

Of the 249 patients assessed for eligibility, 12 did not meet inclusion criteria, 32 did not want to participate and 203 (86%) consenting participants completed the questionnaire. Table 1 shows their background and clinical characteristics. The mean age (SD) was 65.2 (9.0), 49% were female, and 54% were receiving disability benefits.

Distribution of scores

Levels of missing data were low for all D-12 items (table 2). Median item scores corresponded with the mild and moderate response categories for seven and five items, respectively. Item scores were approximately normally distributed, except for the affective domain item relating to agitation, which had a slightly lower number of responses for the mild compared with adjacent response categories. For physical domain items, floor effects ranged from 6% to 27% for “I feel short of breath” and “My breathing is uncomfortable”, respectively. For affective domain items, floor effects ranged from 13% to 29% for “My breathing is irritating” and “My breathing makes me agitated”, respectively. For physical domain items, ceiling effects ranged from 10% to 31% for “My breathing is uncomfortable” and “I feel short of breath”, respectively. For affective domain items, ceiling effects ranged from 11% to 21% for “My breathing makes me agitated” and “My breathing is irritating”, respectively.

Statistical analysis

CFA results showed good model fit for the D-12 physical and affective domains according to all fit indices (table 3). The D-12 unidimensional model had poor fit according to the RMSEA and TLI, and marginal fit according to the CFI. The χ² test also showed that the bidimensional model had better model fit. The two domain scores had a correlation of 0.57.

After adjusting for missing values based on recommendations,11 there was complete data for the physical domain and one missing response for the affective domain scores. Both mean domain scores were approximately in the middle of the possible score ranges with floor and ceiling effects of between 3% and 7% (table 2).

Table 4 shows that items comprising the D-12 physical and affective domains fit the Rasch model according to the p values for the χ² statistics. No items had disordered thresholds and there was no evidence for DIF. Item residual correlations did suggest a lack of local independence for first two items within the affective domain, “My breathing makes me depressed” and “My breathing makes me miserable”. These two items were summed to a single item and satisfactory fit to the Rasch model was found including item fit residuals between ±2.5. Levels of Cronbach’s alpha met the criterion for group and individual comparisons (table 4). PSI levels were borderline for the more stringent criterion relating to individual comparisons.

Correlations between D-12 scores and those for the other instruments and clinical variables were highly consistent with a priori hypotheses (table 5). The highest levels were found for D-12 affective domain scores and those for the HADS, followed by those for the CAT. Lower levels of correlation were found for mMRC, Borg scores and remaining clinical measures. Of the 27 correlations, 21 (78%) were within the hypothesised range. Two exceptions related to HADS scores. First, correlation between the D-12 affective domain scores and those for HADS depression fell just below the criterion of 0.6. Second, correlation between the D-12 total scores and HADS anxiety were above the criterion of 0.6. The remaining exceptions were the four correlations between D-12 physical and total scores and those for the Borg before the walking test and FEV₁ predicted. These were below the criterion of 0.2.

Strengths and limitations

The study was broadly similar in scope to eight international studies reporting on the translation and testing of the D-12.9 The sample size was adequate for the application of CFA and the Rasch partial credit model. Most studies reporting on the D-12 had sample sizes of well under 100,9 which limited more advanced testing for measurement properties, including structural validity. This may be why the D-12 continues to be scored as a unidimensional instrument. Again, facilitated by sample sizes, this was the first time the Rasch partial credit model was used to assess the two separate domains with evidence for fit to the model. The current study included additional PROMs and clinical variables which have been widely used in testing of the D-12.9 Testing followed international recommendations and was based on explicit hypotheses.16

The study did not include a test–retest design which would have provided information on the stability of scores together with the SE of measurement and minimal important differences.9 16 Two studies have reported test–retest reliability for domain scores with similarly high levels of reliability and internal consistency found with UK asthma patients,12 while somewhat lower levels were found for a mixed population in Sweden.35 The cross-sectional nature of the study also precluded testing for responsiveness to changes in health. Finally, this study was limited to patients with COPD, and further testing in other Norwegian populations is necessary before the D-12 can be more widely recommended.