Introduction

COPD is a common serious disease with an overall global prevalence of 9–10%,1 and is predicted to increase from sixth to third place in the international ranking of diseases causing mortality between 1990 and 2020.2,3 In studies of patients discharged from hospital, 1-year mortality rates of 22%4 and 23%5, and 2-year mortality rates of 29.3%6 and 35.6%4, have been reported.

The Dyspnoea, Obstruction, Smoking, Exacerbation (DOSE) index combines the modified 5-point version of the Medical Research Council (MRC) dyspnoea scale7,8 with obstruction (forced expiratory volume in 1 second as percentage of predicted value, FEV1%pred),9 smoking status, and exacerbation rate during the previous year.10 By combining measures relevant to important clinical issues in COPD as well as providing a convenient measure of disease severity for use in routine clinical settings, the DOSE index appears to be a useful complement to other composite measures such as the body mass index, airflow obstruction, dyspnoea, and exercise capacity (BODE) index and the age, dyspnoea and airflow obstruction (ADO) index.11,12 The DOSE index predicts hospital admission, respiratory failure and exacerbation risk.10 To our knowledge, only one study has evaluated the DOSE index as a prognostic instrument in a population of male hospital outpatients.13

This study examined associations of the DOSE index with mortality in male and female COPD patients receiving primary and secondary care in a multicentre Swedish population.

Methods

Procedure

The participants were from a larger cohort of COPD patients in seven Swedish counties.14,15 A total of 1,548 patients aged 34–75 years with a COPD diagnosis (ICD code J44) recorded in medical records between 2000 and 2003 were randomly selected, 1,084 from primary care and 464 from hospital clinics.

Data collection

Data were collected in 2005 by self-completed questionnaire and record review for 2000–2003. The questionnaire response rate was 75%, and 98% (1,111 patients) consented to medical record review. The Swedish Board of Health and Welfare provided mortality data for 2005 to 2010, including underlying cause of death from death certificates. Two research nurses entered the data.

Measures

Age (<50 years, 51–60 years, 61–70 years and >70 years), sex, level of education, self-reported height and weight, dyspnoea (MRC range 0–4), smoking status (current or non-smoking), exacerbation rate, current treatment, and vaccination status were taken from patient questionnaires. The dichotomous education variable identified those who continued in full-time education for at least 2 years beyond the compulsory period of 9 years and those with shorter education. Obesity was defined as body mass index (BMI) ≥30, overweight as BMI <30 and ≥25, and underweight as BMI <20. An exacerbation was defined as an unscheduled or emergency visit or a course of oral steroids due to worsening of COPD. The questionnaire identified the number of exacerbations during the 6 months prior to questionnaire completion. Current treatment was categorised as: no maintenance therapy besides short-acting beta-agonists (SABA) or ipratropium; maintenance therapy with tiotropium or long-acting beta-agonists (LABA); inhaled corticosteroids (ICS) alone; and ICS combined with LABA or tiotropium therapy. Influenza vaccination in the last year and pneumococcal vaccination in the last 5 years were identified.

Information about co-morbid diagnoses and lung function was taken from patients' records. Heart disease was defined as ischaemic heart disease or heart failure, and types 1 or 2 diabetes mellitus were identified. Depression was defined as a diagnosis combined with antidepressant drug treatment. Lung function data were available for 594 (53%) patients. FEV1 was expressed as percentage of predicted using the European Community for Steel and Coal reference values.16 Mortality was categorised as death by respiratory disease, cardiovascular disease, cancer and other causes of death.

The DOSE index

The DOSE index was calculated using the MRC dyspnoea scale, FEV1%pred, smoking status, and the exacerbation rate in the previous year (Table 1).10 The DOSE index uses the sum of its components and the total score ranges from 0 to 8; the higher the score, the more severe the disease. Current smoking was defined as daily smoking. The exacerbation rate over the previous 6 months was used to estimate the annual rate (Table 1). Complete information was available for 319 patients from primary health care centres (PHCCs) and 243 from hospital clinics.

Table 1 DOSE index scoring system
Full size table

Statistical analysis

The analyses were performed using PASW Version 18.0 (SPSS Inc, Chicago, Illinois, USA). Kaplan-Meier curves and Cox regression assessed survival. Assessment using Kaplan-Meier curves indicated that the proportional hazards assumption was justified. The DOSE index was examined as a single score (categorised as scores of 6–8, 4–5 and <4) and other models examined its components separately. The analyses were adjusted for age, sex, and heart disease. Stepwise regression was used to identify the most predictive measure for mortality risk. The analyses were repeated with further adjustment for treatment and vaccination status. Cause-specific mortality associations with the DOSE index were adjusted for age and sex. The study period was from when patients received their questionnaires until death or September 2010. Stratification and interaction analyses investigated potential effect modification by sex, level of care, and age.

The smoking variable was modified to include categories for daily smokers, ex-smokers, never-smokers and occasional smokers, with adjustment for age, sex, and heart disease. Cox regression was used to analyse mortality risk by whether patients had (n=594) or did not have (n=517) spirometry data. A reduced ‘DSE index’ without lung function was created and its association with mortality analysed among excluded patients (n=549) and patients included in the main analysis (n=562).

Ethics

The study was approved by the Regional Ethical Review Board of Uppsala University (Dnr 2010/090). Written consent to use the information for future analysis was obtained for all participating patients.

Results

Patient characteristics

During the average study period of 5 years, 116 patients (20.6%) died. Information on cause of death was available for 113 patients (97.4%). No patient reached the maximum DOSE index score of 8. The number of deaths was higher in COPD stage 3 and 4,9 in secondary care, among men, older patients and those with co-morbid heart disease or those who were underweight. Influenza/pneumococcal vaccinations and maintenance therapy with the combination of ICS and LABA or tiotropium were more common among patients who died, and mortality was lower in the group treated with only SABA and ipratropium (Table 2).

Table 2 Patient characteristics
Full size table

In univariate analyses, male sex, older age, lower FEV1%pred, MRC dyspnoea scale values of 3 and 4, two or more exacerbations in the previous year, underweight, and heart disease were associated with raised mortality. Age, heart disease, smoking status, FEV1%pred, and MRC dyspnoea scale remained statistically significantly associated with mortality when all variables were included in the model (data not shown).

DOSE index

Mortality was higher in patients with a DOSE index score ≥4 (42.4%) than for lower scores (11.0%, p<0.0001; Table 3). The association between the DOSE index and mortality is shown by Kaplan-Meier curves and Cox regression analyses in Figures 1 and 2. DOSE index scores of 6–7 and 4–5 compared with 0-3 were statistically significantly and independently associated with higher mortality in Cox regression analyses before and after adjustment for age, sex, and heart disease (Table 4).

Table 3 DOSE index characteristics
Full size table
Figure 1
figure 1

DOSE index and mortality

Full size image
Figure 2
figure 2

DOSE index score and mortality

Full size image
Table 4 Results of Cox regression analysis of DOSE index
Full size table

Components of the DOSE index

When the components were included in the same model, all were statistically significantly associated with mortality, although the association with smoking was not statistically significant in univariate analysis (Table 5, Figure 3). When the DOSE index and its components were combined in a stepwise elimination regression model, the combined index remained, as it explained more variance than the individual components. Chi-square values (with adjustment for age, sex, and heart disease) were 146.4 for combined index, 116.0 for dyspnoea, 91.2 for obstruction, 54.0 for smoking, and 75.5 for exacerbations.

Table 5 Results of Cox regression analysis of DOSE index components
Full size table
Figure 3
figure 3

DOSE index components and mortality

Full size image

Sex, age and level of care

The relationship between a DOSE index score of 6–7 and mortality was of higher magnitude in men (hazard ratio (HR) 13.9, 95% CI 6.84 to 28.2) than in women (HR 4.00, 95% CI 1.62 to 9.88), with a p for interaction of 0.018. There was no effect modification by age or level of care (data not shown).

Treatment

Additional adjustment for treatment and vaccination did not alter the main results significantly (data not shown).

DOSE index and cause-specific mortality

Cox regression analysis of cause-specific mortality showed that a higher DOSE index score was statistically significantly and independently associated with an increased risk of respiratory disease-specific mortality. A DOSE index score of 4–5 was also independently and statistically significantly associated with an increased risk of cardiovascular mortality. No associations were found for DOSE index with death due to cancer or other diseases (Figure 4).

Figure 4
figure 4

DOSE index and cause-specific mortality

Full size image

Smoking definition and lung function

No statistically significant differences in mortality were found between ex-smokers and current smokers (data not shown). Patients with spirometry data (n=594) were compared with those without (n=517). Among patients with missing spirometry data (n=517), 25.0% died and, among patients with available spirometry data (n=594), 21.2% died (p=0.139). In the entire group no statistical significance was found for mortality associated with missing spirometry data (HR 1.10, 95% CI 0.86 to 1.41). The HR for mortality associated with missing spirometry data was statistically significantly raised when the analysis was restricted to primary care patients (HR 1.68, 95% CI 1.16 to 2.41), but not among secondary care patients (HR 1.20, 95% CI 0.79 to 1.81). In the group included in the main analysis (n=562), 20.6% died and, among patients excluded because of any missing data (n=549), 25.3% died (p=0.064). Excluding FEV1%pred from the DOSE index among the participants included in the main analysis produced a HR for the 4–5 group compared with the 0–3 group of 3.43 (95% CI 2.37 to 4.95) after adjustment for age, sex, and heart disease (no patients achieved a score of 6). Among the patients excluded from the main analysis due to missing data, the adjusted HR for mortality associated with a DOSE index score was 2.08 (95% CI 1.38 to 3.13) for a score of 4–5, and was 6.19 (95% CI 2.65 to 14.5) for a score of 6.

Discussion

Main findings

The primary finding of this multicentre study of COPD patients from primary and secondary care is that a higher DOSE index score is associated with increased mortality risk independent of sex, age, and level of care.

Interpretation of findings in relation to previously published work

A previous study has shown that the DOSE index can predict hospital admission, respiratory failure, and future exacerbations10, and another study has found an association with mortality in male hospital outpatients.13 To our knowledge, this is the first study where an association with mortality is shown for both sexes and in a population including patients from both primary and secondary care. Thus, this index could be a convenient clinical tool for estimating mortality risk among COPD patients in primary care.

The total DOSE index score was robustly associated with mortality. The majority of the separate components that constitute the index were also associated with mortality. The most established predictor of mortality in COPD is FEV1%pred,9,17 although results from one study suggested that dyspnoea measured by the MRC dyspnoea scale is a more reliable predictor.18 Our study demonstrates that the DOSE index — which includes both of these measures — is an even better predictor of mortality. Our analysis also found that the composite DOSE index was a more effective predictor of mortality than its separate components.

The univariate association between smoking status and mortality was not statistically significant. However, in analyses adjusted for the other DOSE index components, smoking was statistically significantly associated with mortality, thus possibly adding a useful contribution to the combined index. A dichotomous measure of current smoking status is potentially limited in precision since previous smoking is not taken into account. An alternative measure of smoking using pack-years has been shown to predict mortality in stable COPD,19 but a study of COPD patients with severe obstruction failed to show an independent association with mortality for number of pack-years.20 Other authors have found current smoking in COPD patients to be associated with increased mortality compared with never or ex-smokers, independent of pack-years.21–23

Although the DOSE index was associated with mortality in both men and women, the interaction analysis showed that the associations were not identical. An HR of higher magnitude was found for a specific DOSE index score in men compared with women. In contrast, a study of the BODE index24 found an HR of higher magnitude in women than in men. In another study,25 BMI was shown to be a more significant component of the total BODE index score in women than in men. Additional adjustment for BMI and depression26,27 in our study indicated that these factors did not explain sex differences in the associations with mortality.

The analyses of cause-specific mortality showed that the DOSE index is more predictive of death by respiratory disease than other causes. The association of the DOSE index with cardiovascular disease mortality was only statistically significant for a DOSE index score of 4–5, possibly due to a small number of individuals in the 6–7 score group. Determination of cause of death from death certificates can be problematic as certificates may have been incorrectly completed. Several studies have reported that COPD is predominantly under-reported as the underlying cause of death,28,29 especially when the primary cause is not pulmonary.30 This suggests that the association of the DOSE index with cardiovascular mortality might be overestimated due to over-recording of cardiovascular disease instead of COPD. However, the important association between the DOSE index and mortality by respiratory disease should not have been overestimated.

The BODE index combines BMI, obstruction (FEV1%pred), dyspnoea (MRC scale), and exercise capacity measured by the 6-minute walking distance (6MWD), and has also been demonstrated to predict mortality.11,12,31,32 Exercise capacity cannot be examined easily, which makes it unsuitable for most clinical settings. Another composite prognostic instrument is the ADO index, including age, dyspnoea (MRC scale) and obstruction (FEV1%pred).12 Both the ADO and the DOSE index are convenient to administer in clinical settings. An additional advantage of the DOSE index is that it includes measures of health and behaviour such as exacerbations and smoking that can be modified by intervention. Exacerbations have been associated with lung function decline,33 lower quality of life,34 and mortality5 and can be modified by treatment.35,36

Strengths and limitations

The strengths of this study are that it is longitudinal and that the study population is sampled from multiple centres representing both primary and secondary care. The data from medical records were initially recorded prospectively and should be reliable and not subject to recall bias. A possible limitation is that a proportion of the original study population was not included in the main analysis, mainly due to lack of spirometry data.

The fact that many COPD diagnoses are based on clinical findings rather than spirometry has been previously demonstrated.14 The exclusion of patients without this information may result in selection bias, which could potentially influence the results as there was higher mortality among patients with missing data in both primary and secondary care. Possible explanations might be that the patients with most severe disease have been unable or unwilling to perform a lung function test or that they had a diagnosis confirmed by spirometry before study entry. However, when we analysed the entire population after removal of the lung function component from the index, the association with mortality remained in patients both with and without spirometry data. Thus, an index excluding lung function data could be an alternative measure that can be used to predict mortality when spirometry is not available.

Implications for future research, policy and practice

In our opinion, the DOSE index appears to be very suitable for use in primary care since it combines information relevant to both clinical management and assessment of prognosis. Thus, it could be a useful instrument for clinicians in both primary and secondary care to identify mortality risk in COPD patients, enabling optimised management and treatment.

Conclusions

We conclude that the DOSE index can be used as a prognostic instrument for mortality in COPD. The combined DOSE index was a better predictor of mortality than any of its components alone. Our data also indicate that a composite index with dyspnoea, smoking and exacerbations without spirometry data can be clinically useful for assessing prognosis.