Interstitial Lung Disease: Seasonality of Hospitalizations and In-Hospital Mortality 2005–2015

Background: The overall incidence of interstitial lung disease and disease-associated mortality have been found on the rise. Hospitalizations for interstitial lung disease are typically caused by airway infection or the acute exacerbation of the underlying disease. Seasonal variance in ambient air pollution has recently been linked to exacerbation and mortality. We sought to examine the seasonal pattern of hospitalizations in Germany, use of mechanical ventilation, and in-hospital mortality on a year-by-year basis to identify their overall trend and to characterize seasonal patterns. Methods: The national in-patient database of the federal statistical office of Germany was searched for cases of interstitial lung disease. Results: A total of 130,366 hospitalizations for ILD occurred from 2005 to 2015. Time series data were examined for seasonality using X-11 statistics. The incidence of hospitalizations, mechanical ventilation, and in-hospital mortality show clear seasonal peaks in the cold season. The observed seasonality cannot be attributed to the variance of selected comorbidities. Also, there is a significant overall upward trend regarding hospitalization counts, especially in the use of non-invasive ventilation. Conclusion: Time series analysis of in-hospital data shows an ILD-related rise of hospitalizations, in-hospital mortality, and non-invasive ventilation. This emphasizes a growing importance of interstitial lung diseases for health-care systems. Strong seasonality is seen in these variables. Data therefore support previous studies of ILD exacerbation. More research on infectious causes and environmental factors is warranted.


Introduction
Interstitial lung disease (ILD) is a term comprising numerous chronic lung diseases which primarily involve the lung parenchyma [1]. Among these are non-infectious inflammatory (e.g., sarcoidosis and connective tissue disease) and non-inflammatory entities (e.g. idiopathic pulmonary fibrosis) [2]. All ILDs bear, to a different extent, the risk of permanent tissue scarring with the end-result of pulmonary fibrosis [2,3]. Shortness of breath, poor physiologic performance, and premature mortality are the con-This article is licensed under the Creative Commons Attribution 4.0 International License (CC BY) (http://www.karger.com/Services/ OpenAccessLicense). Usage, derivative works and distribution are permitted provided that proper credit is given to the author and the original publisher.

Case Selection
The G-DRG classification of diseases parallels the WHO-ICD 10 definitions. The G-DRG database was searched for ICD code J84.1 comprising cases of interstitial lung disease and pulmonary fibrosis. Importantly, the definition includes IPF, but it is not exclusive (i.e., Hamman-Rich syndrome). Pulmonary fibrosis due to inhalation of chemicals, gases, fumes, or vapours or following radiation are excluded by definition. We further tried to estimate the specificity of the diagnosis by evaluating the prevalence of other conditions associated with pulmonary fibrosis. These include pneumoconiosis (J60-J65), hypersensitivity pneumonitis (HP) (J66-J67), rheumatoid arthritis (M06), and systemic connective tissue disorders (M30-36). In 11,002 cases (8.4%), either of these conditions occurred as a co-diagnosis. For the purpose of comparison and control, cases of lung cancer (ICD code C34.x) were searched and analysed accordingly. Patient characteristics regarding age and gender distribution, use of non-invasive and mechanical ventilation, and mortality were extracted. To evaluate different severity levels of the primary outcome of hospitalizations, we defined firstly prolonged duration (>5 days) and/or fatal cases, secondly cases with mechanical ventilation, and thirdly fatal cases.

Statistical Analysis
Serial quarterly hospitalization counts from 2005 to 2015 were analysed for seasonal variance using the X11 procedure as provided by SAS 9.4 [30]. PROC X11 is an adaptation of the US Bureau of the Census X-11 Seasonal Adjustment program and seasonally adjusts monthly or quarterly time series [30]. The output data sets contain the adjusted time series and statistical measures of seasonality [30]. Seasonal data (i.e., winter = January -March; spring = April -June; summer = July -September; and autumn = October -December) were analysed for the presence of seasonality. Upon comparison of high-and low-incidence seasons, the variances in quarter lengths were considered, and normalization to 90-day quarters (Q90) was performed. Generalized linear models (PROC GENMOD, using a Poisson regression) were used to confirm the significance of peak versus minimum seasons. Temporal trends were analysed using generalized linear models as provided by PROC GLM. Descriptive data regarding the use of non-invasive and/or mechanical ventilation, age and gender composition of the cohort are given as absolute and relative numbers and as mean and median ± standard deviation (SD), where applicable. Unless stated otherwise, the t test was used to compare continuous patient variables. Odds ratios including confidence intervals and p values were calculated, as previously described [31,32].  Numbers are given as absolute counts (N) and as rate per 100,000 by age group, gender, and in total. ORs of subgroups are in comparison to the total cohort of hospitalized and fatal cases, respectively. OR, odds ratio. In-Hospital Mortality ILD-related in-hospital mortality is high (7.83%) and can be even higher in advanced age (i.e., up to 13.07% in patients >80 year ( Table 1). The mortality of younger patients (<50 year) is lower but still considerable (2.48%) ( Table 1). As compared to female patients, male patients have a higher mortality rate (OR 1.89, p < 0.0001) (Table 1).

Seasonal Variation of Hospitalizations
Hospitalizations for ILD exhibit an obvious seasonal pattern, with highest adjusted counts in winter (3,227 in winter vs. 2,787 in summer). Likewise, hospitalizations for lung cancer follow a seasonal pattern, with highest average adjusted hospitalization counts in winter (21,882 in winter vs. 19,502 in autumn). In both cases, X-11 statistics can confirm the presence of seasonality (Fig. 1). Yet, important differences are noted. In the case of ILD, the season with the lowest average counts is summer, while autumn shows the lowest hospitalization count for lung cancer hospitalizations. Also, the magnitude of seasonality is greatly higher in ILD patients than in lung cancer patients (ILD, winter vs. summer: 15.8%, p < 0.0001; lung cancer, winter vs. summer: −2.6%, p < 0.0001) (Fig. 1). We hypothesized that cases with a more severe course (prolonged duration, cases with mechanical ventilation and fatal cases) are related to an increasing degree of seasonality. Likewise, the increase was more pronounced in cases with prolonged hospitalization (>5 days and/or fatal) in comparison to all cases (22.78% winter vs. summer p < 0.0001) (Fig. 2). The relation of winter versus summer in cases with the use of mechanical ventilation was slightly higher (23.62% winter vs. summer p < 0.001) (Fig. 2) and highest in fatal cases (32.77% winter vs. summer p < 0.0001) (Fig. 3). In contrast, lung cancer mortality was not related to season (1.06% winter vs. summer p = 0.0927) (Fig. 3).

Mechanical and Non-Invasive Ventilation
Descriptive statistics show that the use of mechanical and non-invasive ventilation is higher in male patients than in female patients (ORs 1.204 and 1.182, both p < 0.0001) ( Table 2). Differences in the age of patients needing mechanical ventilation are seen ( Table 2). Mechanical ventilation is more likely in younger individuals than in older patients. Very old patients (>80 years) are least likely to receive mechanical ventilation (Table 2). Non-invasive ventilation is preferably used in all age groups, except young patients (<50 years) and old patients (>80 years) ( Table 2). As in mechanically ventilated cases in total, invasive and NIV show a comparable seasonal pattern (21.2% and 16.3%, respectively, winter vs. summer p < 0.001, data not shown). Table 3 gives a summary of the absolute and relative frequencies of important comorbidities and comorbid conditions to be found in ILD patients. High-prevalence co-diagnoses are coronary artery disease (33.9%), chronic obstructive lung disease (12.6%) and pneumonia (10.2%). However, these do not explain the overall sea-sonality of hospitalizations as the odds ratio for winter hospitalizations are 1.02, 0.99, and 1.10. As expected, influenza as a secondary diagnosis, shows a clear winter peak. Yet the prevalence of this ICD code as a co-diagnosis is not high in patients with ILD (maximum 0.22% in winter). Likewise, pulmonary embolism shows a winter peak, but also has a low prevalence (maximum 1.21%, OR 1.17). Thus, influenza and pulmonary embolism do not explain the aforementioned seasonality in ILD. Other codiagnoses (HP, pneumoconiosis, systemic disease, rheumatoid disease) do not reveal high prevalence and/or winter peaks.

Discussion
We hereby demonstrate for the first time seasonality of hospital admissions, use of ventilation, and in-hospital mortality in the context of ILD on a population-based level. This phenomenon has been described for overall mortality in 2009 by Olson et al. as they showed seasonal variation of mortality from ILD in the USA [23]. Their  [23]. They were able to provide evidence that mortality is greatest during the colder seasons (winter > spring > autumn > summer) [23]. The investigators attributed the higher winter death toll at least in part to influenza epidemics, which recur in a yearly manner (December through March), and ensuing superimposed bacterial infections [23,27]. Yet, when excluding patient records with pneumonia, seasonal variation was consistently shown [23]. Clinically recognized airway infections may thus not be the only underlying cause leading to the death of PF patients [33]. This is in line with our own data, demonstrating that hospital admissions for interstitial lung disease and pulmonary fibrosis, use of mechanical ventilation, and in-hospital mortality follow a remarkably similar distinct pattern. This observation suggests a seasonal of variation not only mortality but also morbidity. Olson et al. [23] and Nakaji et al. [34] only found trivial increases (2.7 and 4.0%) of cancer mortality in winter as compared with summer. This is in accordance with our study as we found variation of mortality at the same magnitude.
It is meanwhile widely recognized that there are infectious and (partly unknown) non-infectious triggers leading to the acute exacerbation of interstitial lung disease [3,5,6,14,22]. Ambient air pollutants (O 3 and PM) have recently been described to be closely correlated to exacerbations of IPF [24][25][26]. While O 3 concentrations are usually higher in summer, higher particulate matter load is generally associated with winter weather [24][25][26]. Individual biological and behavioural aspects vary substantially between winter and summer [33]. While temperature as such has been linked to excess mortality, lack of daylight alters vitamin D levels and immunological host constitution via, for example, hormonal changes [33,35]. Cold temperatures are a motive for people to gather indoors, which readily facilitates pathogen spread from person to person [36]. These facts may contribute to the seasonal variation of ILD hospitalizations [36]. A trend towards more ILD-related hospitalizations and more fatal cases can be derived from our data set. GLM analyses of the above 11-year period conclusively show that an upward trend is superimposed on the seasonal pattern. As for US mortality data, a strong trend has been identified from 1992 to 2003, with an average age-and sex-adjusted increase of 28.3% in men and 41.3% in women [10,23]. ILDs are still considered rare orphan diseases, many of which are difficult to consider for randomized trials of pharmacological agents [1]. These findings are sign of a growing burden of and greater awareness for interstitial lung diseases and related mortality, which exceeds that of several malignancies [10,11]. A rise of hospitalization counts would directly impact health-care expenditure. A number of comorbidities in patients with certain interstitial lung diseases have been identified in the past, and their impact on morbidity and mortality has been characterized [37]. In ILD, comorbidities have the potential to impair quality of life and to reduce life expectancy [37]. It is reasonable to assume that co-morbidities and co-diagnoses may be driving factors for hospital admissions and thereby be one important underlying cause for the observed seasonality. Yet, the data, as compiled in Table 3, do not point in this direction. The prevalence of diagnoses with high seasonality and/or winter predominance is considerably low in patients with ILD. On the other hand, high-prevalence diagnoses do not show a relevant variance over the course of the seasons. Overall, these factors do not conclusively explain the observed seasonality of fibrotic ILD. There are certain limitations to this study. One important limitation arises from the fact that retrospective data are extracted from a database which has the primary purpose to facilitate health-care insurance processing [27][28][29]. We cannot directly ensure coding quality as such. Yet, coding accuracy is monitored by the professional medical service to the statutory health insurance, which reviews about 10% of all insurance claims [28,29]. Another limitation comes from the DRG system itself, which does not allow for a detailed resolution of ILDs regarding their aetiology since the ICD code J84.1 comprises diverse entities most prominently but not exclusively IPF. This phenomenon is consistently found in studies of ILD hospitalizations on a population-based level [11,38]. We therefore tried to classify cases with other specific causes of pulmonary fibrosis such as pneumoconiosis, HP, rheumatoid arthritis, and systemic connective tissue disorders. These were identified in a minority (8.3%) of all hospitalizations. This study does also not provide a direct insight into the underlying mechanisms that lead to hospitalization and mortality, although, considering the recent research in this field, it is very reasonable to assume that acute exacerbations, regardless of the trigger, account for much of the seasonal variation seen in ILD hospitalizations [6,18,[22][23][24][25]. However, pneumonia is only moderately more frequent in winter than during the rest of the year, and influenza with a strong seasonality is only present in a minuscule portion of overall cases (0.22% in winter). Influenza exhibits a strong seasonal distribution and marked winter peaks in our hemisphere [27,39]. Other authors have suggested influenza infection as one underlying cause of seasonality in DOI: 10.1159/000519214 ILD and other chronic lung disease [16,23]. While a direct explanation for this cannot be derived from our dataset, it is reasonable to assume that influenza vaccination is applied to ILD patients on a regular basis since these patients are seen by health-care providers on a frequent basis, and flu vaccination of patients with lung disease is officially recommended by the German Standing Committee on Vaccination (STIKO) [40].

Conclusion
Taken together, our data indicate that ILD-related hospitalizations show a strong seasonal variation, which is more pronounced in cases with prolonged hospitalization, mechanical ventilation, and fatal outcome. This is not common with other less seasonal lung diseases, such as lung cancer. Frequent comorbidities do not contribute to the observed seasonality. There is reason to believe that these findings are related to the activity of the disease and complications such as acute exacerbation of ILD.