This study demonstrates longitudinal HRQL development in a prospective cohort of patients with ILDs. Our data illustrates that HRQL deteriorated by a clinically relevant amount in a third of the patients after this period. Key predictors for HRQL decline were lower lung function values at baseline across all K-BILD domains. Additionally, older age was a significant determinant in the deterioration of ‘psychological impact’.
In a comparison of the complete and incomplete cases, the incomplete cases (dropouts) had lower lung function values at baseline although they had had a shorter disease duration. This could be indicative of progressive fibrosing ILD. Kolb et al described that, progressive fibrosing ILDs are characterized by a rapid decline in lung function, which may have been the cause for the drop out [24]. Moreover, another study describes the patterns of progressive fibrosing ILD and a short disease duration is attributable to this type of ILD [25].
While comparing the patients that experienced HRQL decline and those who did not, we found that VAS scores were not significantly different between these two groups. We assume that this difference might be explained to some extent by the construction of both questionnaires. VAS asks the patient to rate how they feel on a scale from 1 to 100 today, while K-BILD covers a period of two weeks. First, the longer period presumably provides a more realistic picture of average HRQL, which is less affected by episodic highs and lows. We think, thus, VAS is biased upwards to some extent, which is however not the case for K-BILD. Second, the significant K-BILD differences can largely be explained by regression-to-the-mean. Since VAS also contains more general (non-ILD-specific) HRQL restrictions, it is probably only moderately correlated with the K-BILD and is therefore less affected by the regression effect.
The ILD categories (IPF, sarcoidosis and other) differed in terms of disease duration, number of comorbidities and type of medication. To explain these differences, IPF is a deadly disease, characterized by a short disease duration and high number of comorbidities than any other ILD subtype [26]. Regarding medication, apart from patients with IPF-associated cough or acute exacerbations and those who underwent steroid tapering, the majority of IPF patients received anti-fibrotic therapy, while most patients with other ILD subtypes received immunosuppressant medication.
We observed minimal changes in mean K-BILD scores on the population level after a year. These results are similar to results of the INBUILD trial [27], although the trial only recruited patients with progressive fibrosing ILD. Nevertheless, the baseline patient characteristics in this trial were congruent to those in our study, in terms of age, baseline DLCO % predicted and mean K-BILD scores. Baseline K-BILD scores in this study were 52.5 ± 11.0 and 52.3 ± 9.8 in the nintedanib and placebo arms respectively, (52.8 ± 12.1 in our study). After 52 weeks, the measured score differences were under one unit, similar to our study. The authors, however, did not report on the intra-individual changes in HRQL scores according to pre-specified MCIDs, and we could therefore not make a comparison on this level. Nonetheless, we must consider these small changes. Because a third of the patients worsened in HRQL but the mean K-BILD scores remained the same, it is possible that ILD was characterized by alternating good and bad episodes throughout the 12-month period. At the same time, this could be due to a threshold effect. We used the MCID as a threshold for defining deteriorations. Therefore, HRQL decline lower than MCID was defined as stable. As it is the issue with all threshold-based analyses, individuals closely below the threshold resemble those who just passed the threshold, be it from a clinical or a wellbeing-related point of view. To partially mitigate this issue of classifying individuals with similar pattern to different subgroups we performed a sensitivity analysis targeting linear HRQL change, which also captures those individuals which have HRQL decline slightly below MCID.
The present study found a relationship between HRQL decline and lung function decline. This is similar to previous findings. For example, although the study focused on IPF, Kreuter et al found that an increase in FVC % predicted (beta 0.04, p = 0.006) is a predictor for better HRQL scores [11]. Furthermore, a later analysis in the same population also found higher DLCO % predicted to be associated with higher VAS scores [12]. The only study that included different subtypes of ILD did not find any significant impact of lung function with HRQL, but the study was cross-sectional and thus could not detect these long-term associations [28].
Our study revealed alternating HRQL associations with either baseline FVC % predicted or DLCO % predicted in different K-BILD domains, with DLCO % predicted as an important predictor in respiratory domains and FVC % predicted in the psychological domain. Shortness of breath, dyspnea, is generally associated with HRQL decline [10] and with DLCO % predicted in ILD patients [29]. This could explain why DLCO % predicted was associated with ‘breathlessness and activities’ and ‘chest symptoms’ in our study. Similarly, depression is a common comorbidity in ILD [30] and a study found that depressive symptoms correlate with decline of FVC % predicted in ILD patients [31]. We, therefore, demonstrate lung function parameters predict different aspects of HRQL decline.
Having IPF was linked to less HRQL decline as compared to having sarcoidosis and other ILD subtypes in our study, and there are possible explanations for this. It is known that IPF is associated with substantial morbidity [26] and the prevalence of comorbidities is linked to mortality [23] and HRQL deterioration [11]. In our study, IPF patients had the highest number of comorbidities. We expect that comorbidities, especially in IPF patients, were detected early and managed rapidly, due to the severity of the disease. This in turn could have led to an improvement in overall HRQL. Beyond this, most IPF patients were treated with either nintedanib or pirfenidone. These medications have been shown to slow down lung function decline [6,32] and we assume IPF patients prescribed either nintedanib or pirfenidone possibly experienced less HRQL deterioration than patients with other ILD subtypes. Another possibility could be survival bias. According to literature, untreated IPF patients have a mean survival time of 3 years [33]. The patients in this cohort had already survived a mean time of 2.4 years at the time of recruitment. Therefore, these patients had a better prognosis and HRQL than typical IPF patients. Secondly, patients treated in ILD-expert centers may be assigned an ILD nurse. Studies have shown that ILD nurses have a beneficial role in IPF management [34] and this could have a positive influence on the HRQL of these patients.
We also found higher age to be a determinant of HRQL impairment in ‘psychological impact’ and VAS. It is widely known that functional impairment increases as one ages. This result is consistent with other studies [12].
Patients with secondary education had a significant association with HRQL decline compared to patients with higher education. It is possible that patients with a higher level of education may be able to better assess and understand their prognosis than patients with lower levels of education. Outside of ILD, a study found that lower education levels are associated with unhappiness, poor social relationships and poor self-assessed health [35].
Unexpectedly, the use of immunosuppressant medications was not associated with less HRQL decline. However, this association was present with the linear outcome for K-BILD change score in the ‘psychological’ domain. Immunosuppressive agents are used to treat various forms of ILD. It is believed that they improve the lung pathology [36]. We expected patients on immunosuppressive therapy to therefore have better HRQL scores after 12 months.
Another unanticipated finding was comorbidity burden was not a predictor of HRQL decline. Comorbidities may impact treatment decisions and thus impair HRQL [37,38]. The association between a higher number of comorbidities and HRQL decline was shown in IPF cohorts [12]. Perhaps, comorbidity burden is more significant in IPF than other ILD subtypes. Also, we cannot rule out the effect on HRQL of the medication used to treat comorbidities. On the contrary, however, examining individual comorbidities may have revealed important associations with HRQL decline. Unfortunately, due to the low number of cases, it was not feasible to explore these associations.
We reported a strong association between HRQL decline in K-BILD and study center, which is presumably linked to the center-specific distribution of ILD subtypes in both centers. Patients presenting at the two tertiary care centers for ILD were consecutively recruited; owing to different areas within Germany, the distribution of subtypes differed as described previously [14]. One center included particularly IPF patients, who tend to be older and have a higher comorbidity burden. Despite this center-specific heterogeneity, we decided on a center-adjusted pooled analysis, as analyses resulted in unstable, in part non-converging models for the smaller center. Furthermore, we strongly believe that pooling the data provides a more realistic picture on ILDs in general, as center-specific distributions are averaged to some extent.
Our first sensitivity analysis demonstrated the robustness of our results. Comparable to the main analysis, the main predictors for general HRQL change, DLCO % predicted and baseline HRQL scores predicted similar associations. The second sensitivity analysis excluding IPW showed the complete case analysis underestimated certain predictors of HRQL decline due to drop out bias. The associations that diminished can be explained by the characteristics of the dropouts. They had lower FVC % predicted, and the majority had other ILD subtypes. Moreover, the fact that patients who deteriorated were much older explains the difference of these results.
There are some crucial limitations in our study. This study was observational and therefore, we could only report on associations and not on causations. Furthermore, it is to note that K-BILD does not assess some health issues relevant to ILD such as cough. Cough has been shown to negatively influence HRQL in ILD [39]. However, it is possible that cough was indirectly assessed by its impact on “chest symptoms” and “breathlessness and activities”. In addition, our study did not collect data on weight or body mass index (BMI) and physical activity, and these covariates could provide more insight into the patient’s well-being. A review by Dowman et al. linked improved physical activity in the context of pulmonary rehabilitation with improved HRQL outcomes in ILD patients [40]. Weight loss is common among ILD patients and a study by Pugashetti et al. found an association between weight loss and mortality risk in IPF patients [41]. On the contrary, another study linked higher baseline BMI values to increased frequency of acute exacerbations [42]. It would be interesting to analyze the direction of the effect of weight on HRQL.
Another limitation is the substantial dropout rate of 27.3%, which is however expected in longitudinal studies on ILDs [43,44]. This drop out is presumably informative, as the dropout had lower baseline lung function values than the complete cases which indicates a more severe state of ILD. In consequence, unweighted analyses might underestimate the effect size of lung function. We best possibly addressed the issue of informative dropout by an IPW weighted analysis, and re-affirmed this model by an unweighted analysis (sensitivity analysis 1). Given the high consistency of both approaches, we think that loss-to-follow up might not crucially affect the identified associations.
One strength is that we are the first to apply K-BILD in a longitudinal analysis in a German cohort. We further demonstrate the validity of K-BILD as it captured similar results to the generic VAS in a longitudinal setting. Lastly, although the group is heterogeneous, our study includes otherwise not commonly studied subtypes of ILD, which is important because they have differing prognoses and patterns.