Health-related quality of life associates with change in FEV1 in COPD: results from the COSYCONET cohort

Background Forced expiratory volume in one second (FEV1) characterizes the pathophysiology of COPD and different trajectories of FEV1 decline have been observed in patients with COPD (e.g. gradual or episodic). There is limited information about the development of patient-reported health-related quality of life (HRQL) over the full range of the natural history of COPD. We examined the longitudinal association between change in FEV1 and change in disease-specific and generic HRQL. Methods We analysed data of 1734 patients with COPD participating in the COSYCONET cohort with up to 3 years of follow-up. Patients completed the Saint George’s Respiratory Questionnaire (SGRQ) and the EQ-5D Visual Analog Scale (EQ VAS). Change score models were used to investigate the relationship between HRQL and FEV1 and to calculate mean changes in HRQL per FEV1 change categories [decrease (≤ − 100 ml), no change, increase (≥ 100 ml)] after 3 years. Applying hierarchical linear models (HLM), we estimated the cross-sectional between-subject difference and the longitudinal within-subject change of HRQL as related to a FEV1 difference or change. Results We observed a statistically significant deterioration in SGRQ (total score + 1.3 units) after 3 years, which was completely driven by the activity component (+ 4 units). No significant change was found for the generic EQ VAS. Over the same period, 58% of patients experienced a decrease in FEV1, 28% were recorded as no change in FEV1, and 13% experienced an increase. The relationship between HRQL and FEV1 was found to be approximately linear with decrease in FEV1 being statistically significantly associated with a deterioration in SGRQ (+ 3.20 units). Increase in FEV1 was associated with improvements in SGRQ (− 3.81 units). The associations between change in FEV1 and the EQ VAS were similar. Results of the HLMs were consistent and highly statistically significant, indicating cross-sectional and longitudinal associations. The largest estimates were found for the association between FEV1 and the SGRQ activity domain. Conclusions Difference and change in FEV1 over time correlate with difference and change in disease-specific and generic HRQL. We conclude, that deterioration of HRQL should induce timely re-examination of physical status and lung function and possibly reassessment of therapeutic regimes. Trial registration NCT01245933. Date of registration: 18 November 2010.


Background
Chronic obstructive pulmonary disease (COPD) is defined by the presence of post-bronchodilator airways obstruction, respiratory symptoms such as breathlessness, cough, and sputum production and a history of exposure to inhalational injury [1]. Patients with COPD experience an accelerated decline in FEV 1 compared to healthy never smoking individuals, where a decline of about 20 ml per years was shown [2]. However, the natural history of COPD is not always characterized by a gradual accelerate decline but can also present as episodic accelerated decline of FEV 1 . Here, episodes of deteriorated and improved lung function mark the overall downward trajectory of lung function over time [3]. Accordingly, patients with declining or rapidly declining FEV 1 but also patients with stable or even improved FEV 1 over time have been identified in large COPD cohorts [4][5][6].
While measures like FEV 1 and blood gases reflect the pathophysiology of COPD, measures of health-related quality of life (HRQL) reflect the patient's perspective of his/her disease. They are meaningful instruments to monitor the course of COPD as they cover the severity of symptoms, the impact of the disease on daily life and have also been found to predict mortality [7][8][9]. The longitudinal association between change in FEV 1 and change in HRQL is not fully understood. Estimates based on RCTs and only few observational studies range from only a weak correlation [10] to strong correlations [11][12][13] and often focus on one direction of FEV 1 change-i.e. decrease only [14] or increase only [11,13]. Furthermore, the transferability of findings from RCTs to routine care is limited, because of highly selected patient samples.
In summary, there is limited information about the development of HRQL over the full range of the natural history of COPD, which includes FEV 1 decrease in the context of exacerbations, FEV 1 increase as a consequence of treatment, as well as unchanged FEV 1 . We therefore analysed data from a large, real-world observational cohort of COPD patients followed for 3 years, with the aim to analyse and possibly quantify the association between longitudinal FEV 1 change and change in generic and disease-specific HRQL.

Study design and study population
Between September 2010 and December 2013, the prospective, multicentre COSYCONET ("German COPD and Systemic Consequences -Comorbidities Network") study recruited 2741 participants in 31 study centres across Germany and re-examinations took place after 18-and 36-months. Briefly, baseline inclusion criteria of COSYCONET were age ≥ 40 years and a physician's diagnosis of COPD. Detailed information about the inclusion and exclusion criteria and the recruitment process are available elsewhere [15].
For the present analysis, we excluded patients with (a) missing FEV 1 values at baseline, (b) FEV 1 /FVC ≥ 0.7 at baseline, and (c) no further study participation after baseline. Patients with alpha-1-antiythrypsin deficiency were not excluded, since their HRQL was found to be comparable to patients without the immune deficiency in a cross-sectional analysis [16]. An overview of the study population is given in Fig. 1.

COPD definition and HRQL assessment
Participants underwent standardized postbronchodilator spirometry at each visit. GOLD grades 1-4 were assigned at baseline based on FEV 1 predicted, whereby reference values were taken from the Global Lung Initiative [17]. For the stratified analysis, GOLD grades were further aggregated in two groups (GOLD 1/ 2 and 3/4) because of limited numbers of patients in GOLD grades 1 and 4 (each less than 10% of the total sample).
At each visit, HRQL was assessed using two selfadministered questionnaires: the generic 3-level version of Euro-Qol 5D (EQ-5D-3 L) and the disease-specific Saint George's Respiratory Questionnaire for COPD (SGRQ) [18,19]. The generic instrument EQ-5D is designed to assess HRQL regardless of a specific disease and consists of two parts, the descriptive section and the valuation section. For the present analysis, we used the descriptive section only, i.e. the Visual Analogue Scale (EQ VAS), since this descriptive section was found to better discriminate between COPD grades compared to the EQ-5D valuation section. Furthermore, the VAS was preferred as a simple measure of generic HRQL since the EQ-5D utility index requires a country-dependent tariff and is less sensitive due to its skewed distribution [20]. When using the EQ VAS, participants value their current health status on a scale between 0 (worst possible) and 100 (best imaginable) and a 6.9 units change has been proposed as the minimal clinically important difference [21]. To assess disease-specific HRQL, we used the SGRQ in its COPD specific version. This questionnaire consists of 40 questions related to three components of HRQL (symptoms, activity, and impacts). The total score ranges between 0 and 100 with higher values indicating worse HRQL. Its reliability, validity and responsiveness has been demonstrated in patients with COPD and a 4 units change is considered to indicate the minimal clinical important difference [22].

Assessment of covariates
Age, sex, education, and smoking status were assessed in standardized interviews complemented by selfadministered questionnaires. Body mass index (BMI) was calculated based on measured height and weight. Information on 33 comorbid conditions was obtained by asking "Has a physician ever diagnosed one of the following diseases?". This information was summarised into a single count indicating the number of comorbidities (range 0-33) at each visit. This approach has been previously proven to be a sufficient proxy for total comorbidity burden [23]. Again based on self-reports the history of exacerbations was defined according to GOLD guidelines as no exacerbation, mild, moderate, or severe exacerbation. For each patient, only the most severe exacerbation that occurred in the 12 months preceding the respective study visit was coded. In this way, we attempted to minimize a potential recall bias especially with regard to an underestimation of lighter events. In case of missing values, we imputed the most frequent category or the mean value for continuous data. Considering all three visits and > 4500 observations, a total of only 25 values were imputed for the covariates.

Statistical analysis
Since loss of lung function and HRQL are both dependent on disease severity, patient characteristics including FEV 1 and measures of HRQL at baseline and all analyses are reported stratified by GOLD grade (1/2 vs. 3/4) [4,24]. For 1182 patients with participation in the 3 year follow-up, change in FEV 1 and HRQL over 3 years was evaluated based on t-tests for paired data. To investigate the association between FEV 1 and HRQL over time, two statistical approaches were employed: change score analysis and hierarchical linear models. All models were adjusted for age, sex, BMI, education, smoking status, number of comorbidities, and exacerbation history.

Change score analyses
First, using ordinary least squares linear regression models, we regressed the change in HRQL between baseline and 36 months follow-up on three categories of FEV 1 change and covariates to calculate mean changes in HRQL. The within-subject change in FEV 1 after 36 months was defined as either decrease in absolute FEV 1 ≥ 100 ml, increase in absolute FEV 1 ≥ 100 ml, and no change (in between). The 100 ml cut-off in FEV 1 was chosen in accordance with the previously published minimal important difference for COPD [25]. As we considered the change in FEV 1 to be dependent on baseline lung function, an interaction term to account for the relation between the FEV 1 change category and baseline FEV 1 was incorporated.
Second, generalized additive models (GAM) were conducted, to investigate the relationship between HRQL and a continuous measure of FEV 1 . This nonparametric regression models the association between the dependent variable change in HRQL and the independent variable change in FEV 1 using a smoothing function while adjusting for covariates. Further details have been published elsewhere [26].

Hierarchical linear model
We applied hierarchical linear models (HLM), which enable the inclusion of time-variant and time-invariant covariates and can be applied on datasets with missing variables at different time points (i.e. patient dropped out after second follow-up). These models were designed to provide information regarding mean population trends and individual change over time. Considering time points as time nested in individuals, the model divides the original independent variable into the mean over time (between-subject differences) and the deviation from the mean over time (within-subject change) [27]. In our specific case, the model distinguished between the cross-sectional between-subject and the longitudinal within-subject association of FEV 1 (included as a continuous variable with the unit 100 ml difference or change) and HRQL.

Sensitivity analysis
To account for selective dropout bias, we performed a sensitivity analysis including Inverse Probability Weights (IPW) in the change score-and hierarchical linear models. We first modelled the probability of follow-up based on baseline characteristics (demographics, disease characteristics and quality of life). Weights were then assigned to all patients, who were included in the present analysis, by calculating the inverse of the estimated probability of follow-up. Using this approach, patients, who were found to be similar to those who dropped out, were given greater weights resulting in a weighted population simulating a population without dropout.
All analyses were carried out using the SAS software (SAS Institute Inc., Cary, NC, USA, Version 9.4) package.

Results
Of the 2741 patients recruited into the COSYCONET cohort, 450 had to be excluded because of missing or non-obstructive spirometry at baseline. Of those entering the cohort (n = 2291), 1724 were seen at the second, and 1182 at third follow-up visit. Another 10 participants skipped the first follow-up, but were re-examined in the second follow-up and thus included for the present analysis, resulting in a sample size of n = 1734 at baseline. Table 1 displays the baseline characteristics of the study sample, stratified by GOLD grade 1/2 versus 3/4. Patients with COPD GOLD 1/2 were found slightly older and reported a greater number of comorbidities.
The proportion of patients reporting at least one severe exacerbation in the 12 months before the baseline examination was greater for GOLD grade 3/4, as was the proportion of underweight patients. Similarly, mean baseline SGRQ total score and EQ VAS indicated worse HRQL for GOLD 3/4 compared to GOLD 1/2.

Change in FEV 1 and HRQL over three years
For 1182 COPD patients with participation at baseline and at the 36 months follow-up visit, mean change in FEV 1 and HRQL was calculated ( Table 2). Baseline characteristics of this subpopulation are available in Additional file 1. For the 3 years time period, a 150 ml FEV 1 decrease was observed for all patients, while this decrease was 180 ml for those with GOLD 1/2 and 90 ml for those with GOLD 3/4 at baseline. Over the same period, we also observed a statistically significant deterioration in disease-specific HRQL (SGRQ total score + 1.3 units) on the population level. This overall change in SGRQ was fully driven by a + 4 units change in the activity component, which was present for both baseline GOLD strata. On an individual level, 73% of patients experienced a clinically relevant change in SGRQ after 3 years (40% deterioration, 33% improvement) (Additional file 1 Table A2).
Analysing the change of the generic EQ VAS in the same way, no significant change was observed on the population level. However, 66% of patients experienced a clinically relevant change in EQ VAS (34% deterioration, 32% improvement).

Relationship between FEV 1 and HRQL over time Change score analysis
We then analysed all pairs of repeated FEV 1 and HRQL measurements stemming from 1173 patients who completed the follow-up after 36 months. Altogether, COPD patients with GOLD 1/2 at baseline contributed 695 pairs of observations, while those with GOLD 3/4 contributed 478 pairs. We observed a ≥ 100 ml FEV 1 decrease in 58% of the total sample, 28% were recorded as no change in FEV 1 , and the remaining 13% experienced a ≥ 100 ml FEV 1 increase over the 3 years period (Table 3). Figure 2 displays the adjusted mean change in SGRQ and EQ VAS as associated with FEV 1 change (decrease, no change, increase) for all participants and stratified by baseline GOLD grade of severity. Overall, a decrease in FEV 1 was associated with a deterioration in diseasespecific and generic HRQL ( . Regarding the category no change in FEV 1 , we found non-significant improvements in EQ VAS while the SGRQ remained unchanged. Both GOLD strata mirrored the results of the total sample. Our data indicated an approximately linear relationship between change in HRQL and FEV 1 (Fig. 3) . However, the graph was found to be shifted to the left side of the x-axis meaning that a zero change in FEV 1 did not correspond to a zero change in HRQL but was associated with slight improvements in HRQL. Consequently, a clinical relevant deterioration in SGRQ was associated with a decrease in FEV 1 of more than − 600 ml, while an increase of more than 200 ml FEV 1 was associated with a clinical relevant improvement in SGRQ.

Hierarchical linear model
The results of the HLM analysis detailing the crosssectional (between-subject) and longitudinal (within-subject) estimates for HRQL and FEV 1 are presented in Tables 4 and 5. Regarding the SGRQ total score (Table 4) and according to the cross-sectional estimate, higher FEV 1 was associated with better HRQL with 100 ml more (difference) FEV 1 corresponding to a mean improvement by − 1.42 units in SGRQ. Corresponding estimates for the single GOLD strata were − 1.00 for GOLD 1/2, and − 1.57 for GOLD 3/4. Conversely, the longitudinal within-subject estimate indicated the effect of a 100 ml FEV 1 decrease within a patient. Overall, a 100 ml decrease in FEV 1 resulted in a deterioration in disease-specific HRQL, indicated as a 0.83 units change in SGRQ (p < 0.0001).
Regarding the three component scores of the SGRQ, we observe statistically significant longitudinal associations between a 100 ml decrease in FEV 1 and all domains (activity, symptoms, and impacts). The biggest impact of FEV 1 decrease was found on the activity domain, followed by the symptoms-and impacts components (data not shown).  Regarding the generic EQ VAS (Table 5), we observed estimates of the same direction, but overall estimates were smaller regarding both the between-and withinsubjects analysis in relation to a 100 ml FEV 1 difference or change, respectively.

Sensitivity analysis
The results of the sensitivity analysis are displayed in Additional file 2 (Change Score model) and Additional file 3 (HLM). Overall, the inclusion of IPW confirmed our results since all estimates and p-values were nearly identical. However, it also indicated a slight underestimation of the effect of change in FEV 1 on HRQL particularly in patients with GOLD 3/4 at baseline when excluding dropouts. For example, in patients with baseline GOLD 3/4, the deterioration in SGRQ associated with decrease in FEV 1 was more pronounced when considering participants who dropped out through IPW (SGRQ mean change + 4.11 [1.37 to 6.84] including IPW vs. + 3.59 [0.79 to 6.38] without IPW (see Additional file 2).

Discussion
We analysed the change in HRQL over 3 years associated with change in FEV 1 and investigated both the cross-sectional and the longitudinal association of FEV 1 Fig. 2 Absolute adjusted mean change in SGRQ (a) and EQ VAS (b) after 36 months. Ordinary least square regression models were adjusted for age, sex, BMI, education, smoking status, number of comorbidities, exacerbation history, and FEV 1 change*baseline FEV 1 . Error bars indicate 95% confidence intervals. Change categories in FEV 1 were defined as decrease in absolute FEV 1 ≥ 100 ml, increase in absolute FEV 1 ≥ 100 ml, and no change (in between) after 36 months and HRQL. On the population level, the overall decline in SGRQ total score was small but statistically significant and was completely driven by a significant deterioration of + 4 units in the activity domain. On the individual patient level, more than one-third of patients experienced a clinically relevant deterioration in SGRQ. We found a linear relationship between change in FEV 1 and change in HRQL meaning that decrease in FEV 1 was associated with a deterioration in HRQL whereas an increase in FEV 1 was similarly found associated with improved Fig. 3 Relationship between change in FEV 1 and SGRQ (a), EQ VAS (b). Generalized additive models were adjusted for age, sex, BMI, education, smoking status, number of comorbidities, and exacerbation history. The solid curves represent the estimated smooth functions of the association between FEV 1 and HRQL. The shaded areas indicate 95% confidence intervals Remarkably, a no change in FEV 1 was also associated with slight improvements in HRQL. We found a highly significant relation between a 100 ml withinsubject FEV 1 decrease and generic and disease-specific HRQL, with the largest estimate for the activity domain of the SGRQ. The overall decline in the disease-specific HRQL in COPD is in line with the literature [12,28,29]. Noteworthy enough, the decline was not steady over time, as we found small reductions in the SGRQ total score (i.e. improvement in HRQL) and symptoms component as well as in the EQ VAS for the first 18-months of followup, similar to what has been previously observed by Yoo and co-workers [30].
The deterioration in SGRQ was completely driven by the activity component. This aligns with Waschki et al. who reported a substantial decrease in physical activity over 3 years in a COPD cohort [31]. The finding, that the symptoms and impact component remained unchanged or even improved, would be compatible with the assumption that these factors can be managed through adequate medical or non-medical therapy [14]. Our data underline that maintenance of physical activity should play a much greater role in the treatment of COPD.
The mean changes in HRQL as related to the three FEV 1 change categories did not exceed the MCID. However, a mean deterioration in HRQL on the population level, which is significantly different from zero, indicates an important development, given that a relevant proportion of patients experienced a clinical relevant change in HRQL after 3 years. Furthermore, our results are in line with a systematic review by Westwood and co-workers, who summarized the information of 22 randomized controlled trials on the effects of long-acting bronchodilator therapy and analysed the relationship between increase in FEV 1 and patient-reported outcomes, including HRQL as measured using the SGRQ [13]. According to this analysis, a mean 2.5 units decrease in SGRQ total score (i.e. improvement) was estimated for a 100 ml increase in FEV 1 .
Our results partly concur with Westwood et al., finding that even no change in FEV 1 is associated with improved HRQL. While the GAM indicated slight improvements in HRQL for a zero change in FEV 1 , the stratified analysis confirmed this only for patients with baseline GOLD grades 1/2, but indicated a trend for deteriorations in SGRQ for the more severe grades GOLD 3/4. Westwood et al. discuss a potential Hawthorne effecta phenomenon whereby patients modify their behaviour because of their active participation in science and their awareness of being observed [32]. However, in our study, this effect might be small because the intensity of supervision is rather low with more than a year between study visits. Adaptation processes or changes in treatment after recruitment into the cohort might rather play a role and additional research is needed to further explore this.
The observational Japanese COPD cohort HOK-KAIDO evaluated the relationship between FEV 1 decline and change in SGRQ and its component scores. Based on the degree of the annual decline in FEV 1 , the cohort was split into three categories: rapid decliner (− 63 ± 2 ml/year), slow decliner (− 31 ± 1 ml/year) and sustainers (including improvements in FEV 1 (− 2 ± 1 ml/year)). The authors report deterioration in HRQL for the rapid decliners indicated by a change of 5 units of the SGRQ total score after 5 years, zero change for slow decliners and an improvement in HRQL (− 4 units SGRQ) for the sustainers [14]. Calculation of the change in SGRQ per 100 ml FEV 1 decrease based on the data given for the rapid decliner, results in a mean deterioration in HRQL by a 1.59 units change in SGRQ total score. The withinsubject estimate of our HLM indicated a deterioration in HRQL of half the size (+ 0.83 units SGRQ per 100 ml FEV 1 decrease), which is not surprising, considering that our population was not stratified by categories of FEV 1 decline.
Both HRQL measures differentiated between GOLD strata at baseline and the longitudinal within-subject association between FEV 1 and HRQL showed a similar relationship. However, the overall change in EQ VAS after 3 years (− 0.5 units, n.s.) might have been too small to detect significant mean changes in EQ VAS as related to the FEV 1 change category decrease. Methodological aspects could explain part of the observed differences. Whereas the SGRQ covers history and current health status, the EQ VAS refers to the patients' current shortterm health status, which might show more variation than a sort of averaging as implemented in the SGRQ. Moreover, the EQ VAS as a generic measure of HRQL includes aspects of the patients' life that are not related to his/her COPD all. We conclude that disease-specific instruments are more suitable for the longitudinal assessment of HRQL in patients with COPD. Selective dropout of patients is an issue in long-term observational cohort studies. Regarding our data set of 2291 eligible patients recruited at baseline of the COSYCONET cohort study, 557 and 552 patients were not re-examined at the 18-month and 36-month followup visit, respectively. Of those 1109 patients, 153 (14%) died and 296 (27%) terminated their participation due to worsening of their health status. However, we do not think that dropout severely affected our findings and the sensitivity analysis including the IPW confirmed this hypothesis. One reason might be that our aim was to analyse the association of change in FEV 1 and HRQL and not to predict HRQL development. The latter would indeed be influenced by dropout as one would expect those with deteriorating COPD to also experience worse HRQL. Second, the hierarchical linear model also included patients who were available for only two examinations, therefore minimizing the number of patients not considered.
With regard to the observational and longitudinal design of our study, some limitations need to be addressed. First, regression to the mean might have occurred in the repeated measurement of lung function and HRQL values [33]. This bias seems, however, unlikely since we were interested in the association between the change in FEV 1 and the change in HRQL, which was independent from FEV 1 group assignment. Furthermore, longitudinal results were also confirmed by the HLMs, which are thought to be robust against a bias from regression to the mean. Second, our analyses do not allow drawing conclusions regarding treatment effects on lung function. All patients were under their usual therapy, but medication-specific variables were not considered in the models. This aspect might, however, be less important, as in general the treatment in the COSYCONET cohort is very intense and broad [34].

Conclusions
To conclude, our study provides estimates for both the cross-sectional and longitudinal association between FEV 1 and HRQL and these were highly statistically significant regarding both outcomes: disease-specific and generic HRQL. Overall, change in HRQL followed change in FEV 1 , however, increases in FEV 1 were associated with greater HRQL gains than equal decreases in FEV 1 with HRQL losses. To monitor the progression of COPD from the patient's perspective, the diseasespecific SGRQ was found superior to the generic EQ-VAS. As quality of life is an important aspect in patients' life, determining the course of the disease and therapeutic requirements, the findings suggest that optimal treatment of lung function and a minimization of its deterioration over time has an impact beyond the patients' functional status. Furthermore, deterioration of HRQL should induce timely re-examination of physical status and lung function and possibly reassessment of therapeutic regimes, particularly in patients with severe airflow obstruction.