Introduction

Idiopathic pulmonary fibrosis (IPF) is a chronic, progressive fibrosing interstitial lung disease (ILD) of unknown cause[1,2,3].The disease is characterized by an aberrant accumulation of fibrotic tissue in the lung parenchyma, resulting in extensive alterations of lung structure and function and leads finally to respiratory failure and death [2, 4, 5]. Long-term observational studies in clinically diverse IPF populations from all over the world are increasing [1, 6,7,8,9,10,11,12,13,14,15] and provide us with important information on disease behaviour, management, and effectiveness of approved treatments.

Forced vital capacity (FVC), diffusion capacity for carbon monoxide (DLCO), composite physiological index (CPI) [7, 16] and GAP (gender, age and physiology) stage have been used to define the severity of IPF and to predict mortality[7, 17, 18]. In a recent study, a six-minute walking test (6MWT) was proved to be important predictors for survival [19]. GAP stage 1 has been commonly used as a mild physiological impairment criterion [7]. However, the impact of these physiological variables on disease progression and mortality in patients with mild or more advanced disease is largely unknown. Furthermore, we have previously indicated potential gender differences in patients with IPF [20]. Thus, an unsupervised cluster analysis may provide novel insights into the phenotypes of IPF with potential prognostic significance. Progress in the management of IPF has been made with the introduction of two antifibrotics, pirfenidone and nintedanib, which have been shown to reduce the rate of disease progression [21, 22]. However, strict inclusion and exclusion criteria in clinical trials may limit the generalizability of the results in real clinical settings. For instance, patients with comorbidities, lower lung function, and concomitant medications have been commonly excluded from participation in randomized clinical trials [23,24,25]. Therefore, many questions remain about the generalizability of these findings to a wider IPF population.

Given the lack of knowledge on disease course and mid- to long-term outcomes in IPF, our aims were to explore characteristics, disease severity, phenotype, and anti-fibrotic treatments in patients with IPF under real-life conditions and to assess associations to mortality. We also wanted to ascertain whether further characterization may help patients with IPF and aid the development of personalized management and/or therapy. Additionally, we compared our data with other registries to highlight clinical and geographical variability.

Patients and methods

Study population

The Swedish IPF Registry (SIPFR) is a nationwide registry collecting comprehensive longitudinal data of IPF patients and implemented in 22 respiratory medicine units across Sweden [13, 19, 20]. The SIPFR also includes patients diagnosed before the registry was launched in 2014. The registry relies on a web-based platform (Granitics Unify Med, Granitic Ltd, Espoo, Finland) which allows secure data collection at each respective center. Data entries are made by nurses and physicians at each site, and the quality of the data is evaluated and improved by source data verification performed by the registry coordinator (LC). To be eligible for inclusion in the registry, the patient has to have a confirmed diagnosis of IPF according to the national and international guidelines [13, 26, 27] by a specialist in respiratory medicine either at a university hospital or a local hospital. The registry applies no explicit exclusion criteria, thereby reducing selection bias. We included all patients enrolled in the registry from Sep 2014 until April 2020, and the patients followed ≥ 6 months were enrolled in survival analyses. The outcome of death was defined as patients dying or receiving a lung transplant during the observation period. Patients who were alive at the last visit date during the follow-up period of this study were censored and classified as survivors. The primary survival time was calculated from the enrolment date, with baseline data. Secondary survival time was calculated from the diagnosis date, without matched baseline data.

Variables

Data covering demographics, self-reported comorbidities, lung function, 6MWT, radiology, quality of life (assessed with the King's Brief Interstitial Lung Disease Questionnaire (K-BILD)) and anti-fibrotic therapy were included [13]. Charlson Comorbidity Index (CCI) was calculated with an ad hoc modified formula, i.e. coronary artery disease, other cardiovascular diseases, diabetes, arterial hypertension, chronic obstructive pulmonary disease, and acid reflux gave one point each and history of cancer gave two points. The composite physiologic index (CPI) was calculated using the formula: CPI = 91.0—(0.65 × % predicted DLCO)—(0.53 × % predicted FVC) + (0.34 × % predicted FEV1). The gender-age-physiology (GAP) index was extrapolated for each patient with available data in the registry using the variables of the scoring system combining gender, age, and lung physiology (FVC and DLCO) and classified as GAP stage I (0–3 points), GAP stage II (4–5 points), or GAP stage III (6–8 points). Patients with smoking history included ex-smokers and/or current smokers. Patients were considered as "incident” cases if diagnosed within 6 months from inclusion, while patients with a diagnosis of more than 6 months from inclusion were considered as "prevalent". Each patient's IPF diagnosis was evaluated by "clinic radiological", "thoracoscopic biopsy", "open lung biopsy", or " multidisciplinary conference". Exposure was ascertained by the answers to self-reported questions, such as microbes, particles from the atmosphere, irritants, pollutants, allergens, and pathogens [28]. Data collected at 6 months prior to or after the consent date in this study was considered as baseline data.

Anti-fibrotic therapy

Treatment status was classified into an anti-fibrotic treatment group (receiving anti-fibrotic therapy after diagnosis ≥ 6 months) and untreated group (no treatment or anti-fibrotic therapy after diagnosis < 6 months) [10]. The anti-fibrotic treatment group was further classified into three groups: (1) patients only treated with nintedanib ≥ 6 months; (2) patients only treated with pirfenidone ≥ 6 months; (3) patients switched between pirfenidone (≥ 6 months) and nintedanib (≥ 6 months).

Classification of disease severity using different mild definitions

Disease severity was evaluated as mild and moderate to severe physiological impairment using different criteria [7]. We compared GAP criteria for mild physiological impairment (GAP stage 1) against other proposed criteria: FVC ≥ 75% (exploratory analysis for ≥ 90%,  ≥ 80%, and ≥ 70%); DLCO ≥ 55% (exploratory analysis of  ≥ 60%,  ≥ 50%, and ≥ 45%), TLC ≥ 65% (exploratory analysis for  ≥ 75%,  ≥ 70%, and  ≥ 60%), and CPI ≤ 45 (exploratory analysis CPI ≤ 30, CPI ≤ 40, CPI ≤ 50) exploring the agreement in classifications and relationship with disease outcomes (data not shown).

Cluster analysis on baseline SIPFR data with disease severity

A two-step cluster analysis was used to differentiate the patients into distinct phenotypes. Input variables for the cluster analysis were based on the basis of factor analysis, including baseline characteristics (age, gender, BMI), comorbidities (the number of comorbidities, CCI, acid reflux and cardiovascular diseases), and severity (cut-off level: GAP stage 1, FVC ≥ 75%, DLCO ≥ 55%, TLC ≥ 65% and CPI ≤ 45). The Kaiser–Meyer–Olkin (KMO) value of the scale (> 0.6) and the Bartlett test value of sphericity (p < 0.05) were used to determine the sampling adequacy for factor analysis. Cluster analysis was carried out by using a two steps process [29]. First, the number of clusters were pre-evaluated by Ward hierarchical cluster analysis and factor analysis. Then, the K-means cluster analysis was carried out by using the pre-specified number of clusters. The stepwise discriminant analysis was performed to identify variables discriminating amongst the clusters. For validation, we carried out the leave-one-out method to ensure the stability and repeatability of the cluster model.

Other statistical analyses

Descriptive analysis was performed with medians with interquartile ranges (IQR), or mean ± standard deviation (SD) for continuous variables, and counts with percentages for categorical variables. Missing data, primarily due to data not being registered, was not estimated but was removed from the denominator in calculation. Comparisons between groups were performed using t-test, ANOVA, Mann–Whitney U test, Chi-squared test, or pairwise comparison as appropriate. Univariable and multivariable Cox regression models were performed to investigate the relationships between baseline variables. All models were examined for assumptions of normality of the residuals and homogeneity of variance by examination of residual plots. Kaplan–Meier estimates and a log rank test for mortality were performed to calculate mortality by selected variables. The log-rank test was used to test the differences in survival between the two groups of patients. Comparisons with other IPF cohorts are descriptive and based on published data [1, 6, 9, 10]. All statistical analyses were performed using IBM’s SPSS Statistics version 21 (SPSS, Chicago, IL, USA), Stata 13.1 software package (StataCorp LP, College Station, TX, USA), and GraphPad Prism version 6.0 (GraphPad Software, San Diego, CA, USA). We considered p < 0.05 as statistically significant.

Results

SIPFR cohort

Included patients (n = 662, median age 72.7 years, males 74.0%) were collected between Sep. 2014 and Apr. 2020 (Table 1). Almost two thirds of patients reported a history of smoking, with approximately 60% of patients being ex-smokers, and 24 patients (4%) current smokers (Table 1). The time from IPF diagnosis to enrolment was 2 months. GAP stage was available for 384 patients, and the distribution of GAP stage was I (51.0%), II (40.9%) and III (8.1%). The median value of CCI was 4. The most frequently reported group of comorbidities were cardiovascular diseases, with 54.5% of patients reporting at least one cardiovascular disease (these included hypertension 35.6% of all patients, other cardiovascular diseases 31.6%, and ischaemic heart disease 20.2%) (Fig. 1). Approximately over 70% patients reported at least one comorbidity, and more than 40% of patients had two or three comorbidities at baseline (Fig. 1). According to the primary survival timeline, 480 patients were followed ≥ 6 months from enrolment date (median (interquartile range) 28 (15–46.5) months), while the secondary survival timeline, included 540 patients who had been followed ≥ 6 months from diagnosis date (20 (12–32) months).

Table 1 Baseline characteristics of the SIPFR
Full size table
Fig. 1
figure 1

Prevalence of comorbidities in the SIPFR. The number and percent of a single comorbidity, b the combination of comorbidities. COPD chronic obstructive pulmonary disease

Full size image

During the follow-up time, 195 had died and 23 had undergone lung transplants. The increasing cumulative rate of death from the diagnosis date in one to five years was 7, 16, 30, 39, and 48%, respectively. The cumulative rate from the enrolment date in one to five years was 12, 32, 50, 62 and 78% (Fig. 2a). Figure 2 displays the different cumulative rates of death according to the GAP stage. A trend of shorter survival in male patients compared to women was observed (median: 35.0 vs. 44 months, log rank p = 0.067). In the univariate Cox analysis of baseline factors, decreased lung function, six-minute walking distance (6MWD), K-BILD, BMI, age, smoking history, CPI and GAP, were significant predictors of mortality (Table 2).

Fig. 2
figure 2

Kaplan–Meier analysis for survival in the cohort and in GAP stages. Kaplan–Meier analysis for mortality in the SIPFR cohort according to a time from the enrolment; b time from the diagnosis; c and d GAP stage GAP gender, age, physiology

Full size image
Table 2 Univariable Cox analysis for survival
Full size table

Comparison with other IPF-registries

Comparison of the SIPFR with the Australian IPF Registry [6] Finnish IPF Registry [10] (FinnishIPF, n = 453), the German INSIGHTS-IPF Registry [9] (INSIGHTS, n = 623) and European IPF registry [1] (EurIPFreg, n = 525) is outlined in Table 3. Age and gender distributions were similar in all registries, whereas patients in SIPFR had lower BMI. Baseline lung functions in SIPFR were more preserved than in INSIGHTS- and EurIPF -registries, but worse than in AIPFR and the FinnishIPF. The 6MWD was similar in SIPFR and AIPFR, whereas the distance was greater than the one reported in the INSIGHTS- and EurIPF-registry. Only two registries presented data on TLC% at baseline. Swedish IPF registry presented a lower TLC% compared to EurIPFreg. The cumulative rate of death data from reports were available for SIPFR, AIPFR and FinnishIPF, with one-year mortality of 7% 5%, and 5%, respectively.

Table 3 Comparison of baseline characteristics on SIPFR to other published IPF registries
Full size table

Anti-fibrotic therapy

Among the 540 patients with a follow up of ≥ 6 months from diagnosis, 347 (64.3%) received anti-fibrotic treatment for ≥ 6 months from diagnosis date, either with pirfenidone or nintedanib (33.9% and 26.3% respectively). A minor group of patients (4.1%) had switched treatment. Patients on anti-fibrotic therapy were younger compared to those who did not receive treatment (p = 0.018, Table 4). The median age at diagnosis of the “switched” group, “pirfenidone treated” group, and “nintedanib treated” group were 67.0 years, 72.0 years, and 72.0 years, respectively. However, the difference in age at diagnosis was not statistically significant (p = 0.056). Two thirds of patients (n = 218) had a smoking history (Table 4), with 3 current smokers receiving pirfenidone, 2 current smokers nintedanib, and 1 current smoker had switched treatment. The median age at diagnosis of the “switched” group, “pirfenidone treated” group, and “nintedanib treated” group were 67.0 years, 72.0 years, and 72.0 years, respectively. However, the difference in age at diagnosis was not statistically significant (p = 0.056). Two thirds of patients (n = 218) had a smoking history (Table 4), with 3 current smokers receiving pirfenidone, 2 current smokers nintedanib, and 1 current smoker had switched treatmentFVC % predicted and GAP stage did not differ between patients treated with anti-fibrotic and those who did not receive treatment (Table 4). GAP stage did not differ between nintedanib and pirfenidone treated patients (p = 0.807 and p = 0.116, respectively). Kaplan–Meier analysis showed improved survival in patients on anti-fibrotic therapy compared to untreated patients in all and in patients with GAP stage ≥ 2 ((log rank p = 0.037 and p = 0.034, Fig. 3a, b). When we separately analyzed the two anti-fibrotic drugs, we found that patients receiving nintedanib had better survival compared to untreated patients in all and in patients with GAP stage ≥ 2 (log rank p = 0.034 and p = 0.025, respectively, Fig. 3c, d). In addition, patients switching treatment also had a better survival compared to untreated patients (log rank p = 0.026, Fig. 3c). In the multivariate Cox regression analysis, patients with anti-fibrotic treatment still had a better prognosis than those without (p = 0.007, HR (95% CI): 1.797 (1.173–2.753)) after adjustment of age, gender, BMI, smoking status, FVC%, and DLCO%.

Table 4 Baseline characteristics in treatment
Full size table
Fig. 3
figure 3

Kaplan–Meier analysis for survival in treatment. Kaplan–Meier analysis for mortality in the SIPFR cohort according to a, b patients with and without anti-fibrotic treatment in patients in all and GAP stage over 1; c, d patients with anti-fibrotic treatment (nintedanib, pirfenidone, switched treatment) and untreated in patients in all and GAP stage over 1

Full size image

Classification of disease severity

Altogether 243 patients were followed ≥ 6 months, after exclusion of patients with missing data on FVC%, DLCO%, TLC%, CPI, and GAP stage. Mild physiological impairment defined by GAP stage 1 had a good agreement with CPI ≤ 45 (kappa value (k) = 0.62), and moderate agreement with DLCO ≥ 55% (k = 0.58), FVC ≥ 75% (k = 0.50), and TLC ≥ 65% (k = 0.47). Mild physiological impairment at baseline (DLCO ≥ 55%, TLC ≥ 65%, CPI ≤ 45, FVC ≥ 75% and GAP stage 1, respectively) was predictive of better survival compared to patients with moderate-severe disease in univariable analysis, as well as multivariable Cox analysis after adjustment of age, gender, BMI, smoking status and anti-fibrotic use (Table 5).

Table 5 Crude and adjusted hazard ratios (HR) of mortality for the disease severity in Cox regression model
Full size table

Cluster analysis

A two-step cluster analysis was performed with 15 variables selected on basis of baseline characteristics and severity (Table 6). Altogether, 164 patients were followed ≥ 6 months after exclusion of patients with missing data. Factor analysis showed the selected variables were suitable for further analysis, since the KMO measure of sampling adequacy was 0.612 and Bartlett's Test of sphericity demonstrated a significant difference (p < 0.001). Three clusters were identified in Fig. 4 A-D; patients in cluster 1 (n = 55) consisted mostly of heart diseases (96.4%), mainly male patients (87.3%) with moderate-severe disease at baseline; Cluster 2 (n = 70) was characterized by mild disease with more than 50% females and few comorbidities; Cluster 3 (n = 39) were younger, moderate-severe patients with few comorbidities. The discriminant analysis showed function 1 to mainly consist of the disease severity variables, while function 2 mainly contained comorbidity variables (Fig. 4e). Kaplan Meier analysis of clusters showed that patients in cluster 1 had a worst survival compared to cluster 2 and 3 (log rank p < 0.001 and p = 0.036), whereas patients in cluster 2 had the best survival compared to cluster 3 (log rank p = 0.017) (Fig. 4f). Multivariable Cox analysis showed that cluster 1 (HR: 3.154, 95%CI (1.855–5.364), p < 0.001) and cluster 2 (HR: 0.291, 95%CI (0.160–0.528, p < 0.001)) were predictors of survival, after adjustment of anti-fibrotic use.

Table 6 The characteristics of clusters in SIPFR
Full size table
Fig. 4
figure 4

Characteristics of clusters, distribution and survival. In a–d shown the basic characteristics of clusters; e The distribution in clusters, largest absolute correlation between each variable and any discriminant function in Function 1 (GAP stage 1, CPI ≤ 45% TLC ≥ 65%, DLCO ≥ 55%, males, LpSaO2, and 6MWD) and in Function 2 (CCI, the number of comorbidities, heart diseases, FVC ≥ 75%, age, smoking history, acid reflux and BMI); f Kaplan–Meier analysis for mortality in clusters

Full size image

For validation, we carried out discriminant analysis by the leave-one-out method to ensure stability and repeatability of the model. This method showed that 95.7% of the originally grouped cases were correctly classified, and 90.9% of the cross-validated grouped cases were correctly classified.

Discussion

Similar to other IPF-registries, we demonstrate a heterogeneous patient cohort with respect to age, disease severity, and co-morbidities. The cumulative 1, 2, 3, 4 and 5 year mortality was 7, 16, 30, 39 and 48%, respectively. We were able to confirm that lung function, 6MWD and BMI are significant predictors of mortality [17, 18, 24, 30]. Patients receiving anti-fibrotic therapy had better survival than untreated patients in all and in GAP stage above 1. We investigated the agreement of the GAP stage with single and composite measures of physiological impairment and found that patients with mild physiological impairment have better survival than patients with moderate-severe disease. Three clusters were identified of which one, consisting of males with heart diseases, multiple comorbidities, and high GAP stage, had the worst survival.

One of the important findings in this study is a stratification for a standardized approach to disease severity. Potential stratifications of disease severity have been a widely discussed topic in the community for a long time. Heterogeneity in IPF is multidimensional. Although it is difficult to define the "best" definition of disease stratification, classification requires consideration of these disparate domains. Some of these characteristics have been incorporated in indexes of different domains such as the GAP-index and the composite physiological index, CPI. Patient registries give us the opportunity to include a heterogeneous group of patients with wide ranges of baseline physiology and disease severity. Our results showed that CPI ≤ 45, DLCO ≥ 55%, FVC ≥ 75%, and TLC ≥ 65%, agreed well with GAP stage 1 for staging of mild physiological impairment. This was a first study to define the mild physiological impairment by TLC% in a large scale of IPF patients. Moreover, we also showed that the presence of mild impairment at baseline was predictive of better survival compared to patients with moderate-severe disease on univariable as well as multivariable Cox analysis adjusting for age, gender, BMI, smoking status, and anti-fibrotic use.

To the best of our knowledge, no cluster analysis has been done on the IPF registry cohorts using longitudinal data so far. We report an explorative analysis of potential phenotypes of IPF patients in SIPFR, including our newly defined "mild" IPF classification, comorbidities, and demographic data. More than 40% of the patients had two or three comorbidities. Although we did not find a significant association between comorbidities and outcome in the univariate analysis, comorbidities showed high predictor importance in the cluster analysis. This may reflect the real-world IPF patient since single or composite variables have some limitations as disease predictors. Three clusters were identified, with GAP, comorbidities, and gender deemed important factors between clusters. Heart diseases and severity factors had high predictor importance value in our cluster analysis. As shown in other studies, IPF and heart disease may share several risk factors, and IPF has been associated with atherosclerosis [31,32,33]. The cluster comprising moderately to severe diseased males with heart diseases had worst survival, and mild disease cluster with less comorbidities had best survival. Thus, phenotypes may offer a novel multidimensional approach for predicting outcomes of patients with IPF and suggest patients’ need for special management.

Registries provide the opportunity to study disease progression in patients with anti-fibrotic treatment [10, 30]. In Sweden, anti-fibrotic drugs are completely reimbursed, which results in a large number of patients being on treatment [34]. Thus, approximately 65% of the patients received anti-fibrotic treatment in our study, which is considerably more than in Germany (44%) [9], Finland (26%) [10], and Australia (23%) [6]. The present study shows that patients on anti-fibrotic therapy appear to survive longer than untreated patients, a result similar to what other registries have reported [6, 8, 20]. In order to avoid a potential bias in mortality analysis, we showed that there were no significant differences in baseline lung function between anti-fibrotic treated and untreated groups. Furthermore, we adjusted the potential confounders (age, gender, BMI, smoking history, and anti-fibrotic use) at baseline to identify the association between low lung function parameters and mortality. The curves of antifibrotic use and untreated could be clearly distinguished in the Cox model. Although the effect driven by lung function decline was not included in the current baseline project, it is a focus in an upcoming project.

Interestingly, twenty-two patients had been followed ≥ 6 months, who received the switched antifibrotics treatment (Table 4). Reasons for switching antifibrotics is not a dedicated variable in the registry, resulting in a risk of missing disease progression as a cause to the switch. This might be the case for some of these patients in our dataset. In our experience, side effects make up the main reason and are reported for some of the patients in this group. Disease progression is, in our experience, a minor reason for switching treatment, simply because there are no defined definition of stable or progressive disease when it comes to the individual patient. It is important to clarify that our registry, like all other registries, is not designed to compare treatment effects. Differences in characteristics of the compared groups, non-randomization, other undetected confounders, and missing registry data are important factors that require a cautious interpretation of these results. For the purpose of studying treatment effects, well-powered randomized controlled trials are the only gold standard. Thus, lack of improved survival, or survival benefits, does not imply the absence or presence of a true, underlying difference between the groups. The favourable effect on survival of the “switched group” may only be hypothesis-generating and interpreted with caution since pirfenidone and nintedanib have different mechanisms of action. The idea of sequential treatment strategies in IPF has been discussed before, with few retrospective studies on small cohorts, supporting such strategy [35,36,37].

A number of limitations are worth noting. Firstly, no estimates were made for missing categorical and continuous data and missing data was not involved in further analysis. Secondly, while other studies and registries have highlighted the poorer prognosis in patients with pulmonary hypertension and/or lung cancer, our registry does not collect that type of data, potentially missing other explanatory variables for prognosis. Thirdly, prevalent patients, consisting of 35%, may have a slower disease progression [10, 11], increasing the risk of bias in the survival analysis. Finally, we considered the two timelines from diagnosis and enrolment and adjusted the confounders, but residual confounding might be possible and may have affected the regression analysis [38]. In addition, the effect of smoking on IPF behaviour was not deeply analysed, due to the small numbers of current smokers (n = 24). Only 19 of 24 (4%) patients had been followed ≥ 6 months. Hence, only smoking history (ex-and current smokers) were evaluated in this study. The influence of current smoking on the disease will require a larger cohort. Potential preventive effects of antifibrotics on hospitalizations and exacerbations and thus also on mortality were not analysed in this paper. Currently, data related to exacerbations in the Swedish IPF-registry is limited and needs further distinguishment and collection (e.g. distinguishing hospitalizations related to comorbidities from IPF related exacerbations).

Conclusion

We conclude that both disease severity and phenotype are closely associated with outcome in IPF which may be important for disease behaviour and follow-up. Survival was significantly higher in IPF patients with anti-fibrotic therapy, especially in patients with moderate-severe disease. Mild physiological impairments could be defined by TLC ≥ 65% in SIPFR. IPF patients with mild physiological impairment have better survival than patients with moderate-severe disease. Phenotypes may contribute to predicting outcomes of patients with IPF and suggest the patients’ need for special management, whereas single or composite variables have some limitations as disease predictors. Our results provide an insight into the characteristics, management, and outcome of IPF-patients in real life.

Role of the sponsors

The study funders/sponsors had no role in this study, including the design, collection, management, analysis, writing, review, or approval of the manuscript; and decision to submit the manuscript for publication.