Epidemiology of interstitial lung diseases and their progressive-fibrosing behaviour in six European countries

Discussion

PERSEIDS addressed the paucity of detailed and updated incidence/prevalence data for ILDs and its PF forms. The study has various strengths: the parallel implementation in six countries and the participation of both rheumatology and pulmonology units provides a broad, contemporaneous picture of the burden of ILDs across countries and grants a study population large enough to obtain sufficiently precise initial estimates. Also, the period of 5 years considered was wide enough to observe temporal trends and between-country differences.

The few epidemiological studies on ILDs conducted so far in countries participating in PERSEIDS show quite homogeneous estimates. Incidence data were available for Belgium (1.0/10⁵ persons/year in 1998) [15], Greece (4.63/10⁵ persons/year in 2004) [16], Denmark (19.36–34.34/10⁵ person-years between 1995 and 2005, 3.8–6.6 per 10⁵/year between 2003 and 2009) [17, 18], and the Nordics including Finland and Norway (1.4–20.0/10⁵ persons/year in 2009) [19]. The Belgian and Greek studies above also provided prevalence estimates for ILDs (6.27/10⁵ and 17.3/10⁵, respectively) [15, 16]. In contrast, PERSEIDS estimates were higher and more variable among countries. IPF incidence estimates available so far for Belgium (0.22/10⁵/year), Greece (0.93/10⁵/year), Denmark (7.27/10⁵ person-years between 1995 and 2005 and 1.3/10⁵/year between 2003 and 2009) and the Nordics (0.4–10/10⁵/year) were lower than in PERSEIDS, as was IPF prevalence reported previously for Belgium (1.25/10⁵) and Greece (3.38/10⁵) [15–19]. A Norwegian nationwide cohort study estimated the incidence of SSc-ILD at 1.0–1.3/10⁵ persons between 2005 and 2012 and the prevalence at 13/10⁵ persons in 2013 [20], both in line with PERSEIDS estimates for Norway. Nevertheless, comparing the results of prior studies with PERSEIDS is difficult due to several differences in data collection periods, data sources, population composition/size, disease definitions, diagnostic criteria, availability of HRCT scans, specialists involved or even the ILDs considered. Of note, despite the comparatively higher estimates, in PERSEIDS the overall prevalence of individual ILD subtypes (IPF and non-IPF F-ILDs) remained below the European Union threshold for orphan diseases (50/10⁵ persons [21]).

Highest maximum crude estimates for ILDs, F-ILDs and non-IPF F-ILDs were consistently found in Belgium, and lowest in Portugal. Multiple factors may explain these findings. In Belgium, Liege Hospital was the only centre providing data for all ILDs in Phase 1. As rheumatological ILDs are referred to this centre from all over the country, we adjusted the reference population in these ILDs only. However, non-rheumatological referrals cannot be ruled out, so there may have been overestimation of these conditions and of the overarching ILD categories. Other contributing factors may be the concentration of heavy industries in the province of Limburg [15] and the high percentage of smokers among the Belgian population [22]. As for Portugal, the largest city (Lisbon) was not represented, and most patients with ILD are routinely managed in several hospitals in this country, possibly hindering the ability of the participating centres to identify incident/prevalent cases more accurately. The prevalence of other F-ILDs was high in Denmark, probably because the nationwide, comprehensive search conducted had a greater potential to detect these subtypes. For IPF, interestingly, the highest and lowest incidence estimates were consistently observed in the same country (Denmark) throughout the study period. In this country, maximum incidence was obtained based on incident cases identified by the systematic search (probably overestimated due to coding issues) and minimum incidence based on down-adjusted incident cases (see supplementary material Section B). The extent of overestimation and the strictness of the adjustment applied may explain the highest/lowest estimates compared to other countries.

Incidence of ILDs was relatively high compared to prevalence. This may point to a high mortality rate; but there may be other factors contributing, such as the aged population structure in Europe [17] and a better ascertainment of ILDs [23], in part due to an increasing use of HRCT [24]. Throughout the study period, incidence/prevalence were rather constant in most countries. The changes observed in Finland and Belgium seemed mainly driven by non-IPF F-ILDs, with exposure-related ILDs showing a downward trend in Finland, and other F-ILD and sarcoidosis increasing in Belgium. This suggests changes in occupational exposure and in ILDs classification. In Greece, the increasing prevalence (but not incidence) of F-ILD seemed to be driven by both IPF and non-IPF F-ILDs, perhaps pointing to improved survival. The relevance of these trends is not clear, as the study was not designed to assess the statistical significance of changes throughout the study period.

The relative distribution of non-IPF F-ILDs subtypes observed may help clinicians discern which patients should be screened for ILDs, to achieve prompt diagnoses and improved outcomes; and for the most prevalent subtypes, should trigger the development of proper recommendations for management, as they are currently lacking except for SSc-ILD [25]. In most countries, sarcoidosis was the most frequent subtype, as reported previously [10, 15, 16]. The exceptions were Belgium (RA-ILD) and Norway (other F-ILD). In Belgium, the only centre providing data for all subtypes in Phase 1 (Liege) was of reference for rheumatological ILDs (including RA-ILD), while in Norway, “other F-ILD” may have included other ILD subtypes due to the coding issues described in the supplementary material. The expected bias was reduced by using wider reference populations for calculating incidence/prevalence, applying a different PPV to rheumatological ILDs (Liège) and estimating F-ILDs incidence/prevalence based on data from surrounding countries (Norway), but some bias may have remained. The high percentage of exposure-related ILDs in Finland may due to its prominent ship-building industry [26], with its extensive use of asbestos throughout the 20th century [27]. The exposure–diagnosis lag-time may explain the still high proportion of exposure-related ILDs, while the effect of asbestos bans may be behind the progressive incidence/prevalence reduction observed in PERSEIDS. In Portugal, the high percentage of HP may be partly explained by the importance of the cork industry and the widespread practice of pigeon breeding [28, 29], both associated with HP [30–32]. The incidence of some non-IPF F-ILD subtypes exceeded that of IPF (sarcoidosis in most countries, but also other F-ILD in Belgium and Denmark, and iNSIP and uIIP in Belgium). The refinement of IPF definition and the improvement of diagnostic procedures in later years [33] may have reduced misdiagnoses of non-IPF F-ILDs as IPF, and is expected to further increase non-IPF F-ILDs estimates in the future.

Sarcoidosis was the most frequent non-IPF F-ILD subtype in Phase 1. Though rarely fibrotic in early stages, sarcoidosis was included among non-IPF F-ILDs in PERSEIDS because fibrosis ultimately affects 20% of patients with stage IV disease [14]. The codes/keywords used for sarcoidosis in Phase 1 were specific for lung involvement but could not discern between fibrotic and non-fibrotic disease, probably resulting in overestimated crude incidence/prevalence. This was later amended by the HRCT review in Phase 2, where false fibrotic cases were detected and accounted for (through the PPV) to adjust incidence/prevalence estimates for ILDs and PF-ILDS. The high proportion of PF behaviour in Phase 2 may be partly explained by differences in case-mix, which in Phase 2 favoured more progressive subtypes. PF behaviour definition in PERSEIDS was similar to the one used in the INBUILD trial [34], but being a retrospective study it was difficult to assess worsening of symptoms in patients with a moderate FVC decline (5 to <10% predicted). In these patients, surrogates such as hospitalisations and increasing fibrosis were used instead. PF behaviour was present in 30% of non-IPF F-ILD cases, as observed in the recent French PROGRESS study among patients not receiving antifibrotic treatment [35]. The percentage of PF behaviour among individual subtypes was generally consistent with that reported previously [10, 35, 36] except for SSc-ILD, though differences in PF behaviour definition must be considered [37]. Overall, PERSEIDS results point at subtypes more prone to progression (HP, other F-ILDs, uIIP and iNSIP) which could benefit most from a thorough follow-up and antifibrotic treatment. The proportion of UIP-like pattern in PERSEIDS was highest among SSc-ILD patients (70%), though generally <10% of SSc-ILDs show a definite UIP pattern [38, 39]. Also, UIP-like features were present in approximately a third of non-IPF F-ILD cases and in almost half of cases with PF behaviour. It must be considered that the UIP definition used in PERSEIDS included UIP-like HRCT patterns, and true UIP was not confirmed with histology. In the centre where most SSc-ILD cases with UIP-like pattern were reported, the percentage approached that previously reported for nonspecific idiopathic interstitial pneumonia (NSIP), a pattern far more common in SSc-ILD [40]. There is some overlap between NSIP (especially when fibrotic) and UIP features, which can be difficult to distinguish between [41]. The broader, UIP-like pattern used in PERSEIDS might have facilitated misclassification and thus the important overestimation of UIP in this centre.

Differences in the relative distribution of non-IPF F-ILD subtypes among countries and in their percentage of progression explain the variability in incidence/prevalence estimates of PF-ILDs, and reinforce the primary methodology chosen (weighted mean percentage of progression). Also, primary and sensitivity analyses gave similar results in most countries, further supporting the strength of results. Nevertheless, the primary approach had some limitations, as the weighted percentage was calculated by pooling progression data from all countries, and the I² statistic revealed a considerable heterogeneity among them.

Despite the efforts to reduce bias, the study has other limitations. It was not always possible to use a homogeneous methodology due to centre/country particularities, which may limit the comparability of results. Coding particularities/issues may explain some of the variability found between countries, as it may have influenced the ability to capture incident/prevalent cases and may have led to more misclassification in specific countries. Some centres (e.g. Portugal, Norway) did not provide incident/prevalent cases for the whole study period, precluding comparisons among countries for missing years and possibly affecting overall estimates. Between-country differences in ILD management, previously highlighted [18], also made comparisons difficult. Finally, the ratio of participating to total reference centres in each country and some participating centres being of reference for specific ILDs may have affected the representativeness of results at the country level. In this sense, future epidemiological studies should engage a higher percentage of centres managing ILDs within each country or use national registries where available.

Other limitations were not country-specific. ILDs are often rare and diagnosis can be difficult, leading to over/underestimation and misdiagnosis. Also, the use of ICD codes for ILDs has known limitations [42], but the PERSEIDS approach was reasonable as currently there is no standard methodology for ILD identification based on diagnostic codes. Some code/keywords for ILD-associated conditions were excluded from the search, but being extremely infrequent and/or unspecific, a relevant effect on results is not expected. Although the search algorithm was not previously validated, it was designed ad hoc following expert advice to maximise specificity and sensibility, and the high PPVs obtained in most countries are reassuring. The low PPV found in Finland may be partly explained by the ∼40% of sarcoidoses found in Phase 1, many of which were found to be non-fibrosing after the review of Phase 2. When calculating incidence/prevalence, populations to be used as denominators were not always clear, especially in centres receiving referrals, which can lead to under/overestimation. Furthermore, referrals were not always systematic, or varied during the study period. These uncertainties were addressed through sensitivity analyses wherever possible. Adjusting estimates by a PPV calculated based on non-IPF F-ILDs implied assuming the same rate of false positives across all ILDs, which may not be the case. Finally, non-IPF F-ILD cases reviewed during Phase 2 may represent a small proportion of total prevalent cases identified in the systematic search (especially in centres searching national or regional registries). However, as ≥100 consecutive cases were reviewed per centre, the sample is expected to be representative.