ILD-GAP Combined with the Charlson Comorbidity Index Score (ILD-GAPC) as a Prognostic Prediction Model in Patients with Interstitial Lung Disease

Background The ILD-GAP scoring system has been widely used to predict the prognosis of patients with interstitial lung disease (ILD). The ability of the ILD-GAP scoring system combined with the Charlson Comorbidity Index score (CCIS) (ILD-GAPC) to predict ILD prognosis was investigated. Methods In ILD patients, including idiopathic pulmonary fibrosis (IPF), idiopathic nonspecific interstitial pneumonia (iNSIP), collagen vascular disease-related interstitial pneumonia (CVD-IP), chronic hypersensitivity pneumonitis (CHP), and unclassifiable ILD (UC-ILD), treated between April 2013 and April 2017, the relationships between baseline clinical parameters, including age, sex, CCIS, ILD diagnosis, pulmonary function test results, and disease outcomes, were retrospectively assessed, and the ability to predict prognosis was compared between the ILD-GAP and ILD-GAPC models, respectively. Results A total of 185 patients (mean age, 71.9 years), all of whom underwent pulmonary function testing, including percentage predicted diffusion capacity for carbon monoxide, were assessed. ILD diagnosis consisted of IPF in 57 cases, iNSIP and CVD-IP in 117 cases, CHP in 6 cases, and UC-ILD in 5 cases. The ILD-GAPC provided a greater area under the receiver operating characteristic curve (0.758) for predicting 3-year ILD-related events than the ILD-GAP (0.721). In addition, log-rank tests showed that the Kaplan−Meier curves differed significantly among low, middle, and high ILD-GAPC scores (P < 0.001), unlike ILD-GAP scores (P = 0.083). Conclusions The ILD-GAPC model could provide more accurate information for predicting prognosis in patients with ILD than the ILD-GAP model.


Background
Interstitial lung disease (ILD) is characterized by alveolar infammation leading to progressive fbrosis. Te clinical course and rate of progression of ILD are extremely variable among patients due to various radiological and pathologicalmorphological patterns, such as usual interstitial pneumonia (UIP), nonspecifc interstitial pneumonia (NSIP), organizing pneumonia, respiratory bronchiolitis, desquamative interstitial pneumonia, difuse alveolar damage, and their combinations [1]. An ofcial statement of the American Toracic Society, the European Respiratory Society, the Japanese Respiratory Society, and the Latin American Toracic Association (ATS/ERS/JRS/ALAT) proposed various clinical parameters associated with an increased risk of mortality, such as clinical symptoms, pulmonary function, and the extent of UIP on high-resolution computed tomography (HRCT); however, clinical parameters for accurately predicting the prognosis of ILD have not been established [2].
To provide more accurate prognostic information in patients with ILD, various composite approaches have been reported using peripheral blood biomarkers and physiological and radiographic measurements [3][4][5][6][7]. Ley et al. proposed the GAP index as a mortality prediction model for idiopathic pulmonary fbrosis (IPF) patients, consisting of four parameters including gender (G), age (A), percent predicted forced vital capacity (%FVC), and difusion capacity of carbon monoxide (%D Lco ) (P) [3]. In addition, to predict mortality in major chronic ILD subtypes including IPF, idiopathic NSIP (iNSIP), collagen vascular diseaserelated interstitial pneumonia (CVD-IP), chronic hypersensitivity pneumonia (CHP), and unclassifable ILD (UC-ILD), ILD-GAP has been reported to be useful [4]. Both GAP and ILD-GAP have been widely used in the clinical setting, but these mortality prediction models do not take into account the presence or severity of comorbidities, despite previous research showing that comorbidities, such as cardiovascular disease, arteriosclerosis, and cancer, afect the long-term prognosis of ILD [8,9].
Te present study retrospectively investigated the accuracy of predicting ILD prognosis using the ILD-GAP scoring system combined with the Charlson Comorbidity Index score (CCIS) (ILD-GAPC), which has been widely used as a prognostic indicator for patients with colorectal cancer, advanced nonsmall cell lung carcinoma, acute myocardial infarction, and so on [10][11][12][13].

Study Location and Enrolled
Patients. Tis retrospective, observational study was performed using data from patients treated at Yokohama City University Hospital between April 2013 and April 2017. Te medical records of all patients with ILD who met the following inclusion criteria were reviewed: patients with IPF, iNSIP, CVD-IP, CHP, and UC-ILD in a stable condition who were able to perform pulmonary function tests, including D Lco . ILD patients in a stable condition were defned as patients who had not experienced acute respiratory worsening such as an acute exacerbation (AE), infection, pulmonary embolism, pneumothorax, or pulmonary edema until a pulmonary function test [14]. As shown in Figure 1, pulmonary sarcoidosis, lung cancer with ILD at the time of enrollment, cryptogenic organizing pneumonia, drug or radiation-induced lung injuries, chronic obstructive pulmonary disease, bronchial asthma, and infectious pulmonary disease were excluded.

Data Collection.
Te relationships between baseline clinical parameters include age, sex, CCIS, ILD diagnosis, blood biomarker results, pulmonary function test results, and the disease outcome. CCIS is a summed score of 19 comorbidities weighted according to severity, which was developed to assess the risk of death from comorbidities and has been widely used as a prognostic indicator for patients with colorectal cancer, advanced nonsmall cell lung carcinoma, and acute myocardial infarction [10][11][12][13]. In recent years, large-scale cohort studies and clinical trials in fbrotic ILD have also been recognized as important factors afecting long-term prognosis, including mortality and acute exacerbation [14,15]. Te disease outcome included 3-year ILDrelated events and 3-year all-cause mortality. Tree-year   ILD-related events mean ILD-related mortality such as respiratory failure and frst AE after the pulmonary function  test within 3 years. Tree-year all-cause mortality includes  respiratory events such as chronic respiratory failure due to  ILD and nonrespiratory events such as extrapulmonary  malignancy after the pulmonary function test within 3 years. For patients who did not die in our hospital, the disease outcomes were confrmed by telephone. In addition, only one patient (0.5%), who was transferred to another hospital for best supportive care due to severe deterioration of respiratory status, was lost to follow-up; therefore, the transfer date of that patient was selected as the decision date of the disease outcome.

Diagnosis of ILD.
A diagnosis of idiopathic interstitial pneumonias (IIPs) was confrmed by physical fndings, serological testing, fndings from HRCT, and lung biopsy specimens, based on the ofcial statement for IIP [1,2]. Patients from whom a lung biopsy could not be obtained were diagnosed based on the radiological classifcation [1,2]. Te diagnosis of CVD-IP was confrmed by physical fndings, serological testing, and HRCT fndings consistent with ILD. CHP was diagnosed based on previously established criteria [16]. An AE of ILD was defned as: unexplained worsening of dyspnoea; hypoxemia or severely impaired gas exchange; new alveolar infltrates on radiography; and absence of an alternative explanation such as infection, pulmonary embolism, pneumothorax, or pulmonary edema [17][18][19].

Te Details of ILD-GAP and ILD-GAPC Classifcation.
Te ILD-GAP model was developed for application across all ILD subtypes, including iNSIP, CVD-IP, CHP, and UC-ILD to provide cause-specifc survival estimates using a single risk prediction model compared to the original GAP model for IPF patients that accounted for better adjusted survival in these patients [4]. As shown in Table 1, the predictor variables considered in this model include gender, age, lung physiology variables (%FVC and %D Lco ), and these ILD subtypes. Te ILD-GAP score is calculated by combining the points assigned to these variables that is then divided into stages I (≤1 point), II (2, 3 points), III (4, 5 points), and IV (>5 points) or low score (≤1 point), moderate score (2, 3 points), and high score (≥4 points) that predict mortality risks at 1, 2, and 3 years. CCIS was scored as follows: (0-1: 0 points, 2-3: 1 point, ≥4: 2 points).Te ILD-GAPC score is calculated by combining CCIS and the original ILD-GAP scores and then divided into low score (≤1 point), moderate score (2-3 points), and high score (≥4 points). Te rationale for creating the ILD-GAPC model will be presented in the result section.

Statistical
Analysis. Data were statistically analysed using JMP12 (SAS Institute, Cary, NC) and R software, version 3.5.1 (Te R Foundation for Statistical Computing, Vienna, Austria), and are expressed as means ± standard deviation. Groups were compared using the chi-square test and Wilcoxon rank-sum tests. To determine the primary predictors of 3-year ILD-related events, including causespecifc mortality and the frst AE, univariate analyses were performed. Te predictive performance of the scoring systems was investigated using the areas under the timedependent receiver operating characteristic curve (ROC) analysis (AUC), the concordance index (C-index), and Akaike's information criterion (AIC). When comparing 3year ILD-related events and 3-year all-cause mortality among groups depending on the scoring system, Kaplan−Meier curves were used. Log-rank testing was also performed with strata based on the identifed predictors. Values of P < 0.05 were considered signifcant. Table 2 shows the clinical characteristics of the 185 patients evaluated, including IPF in 57 cases, iNSIP and CVD-IP in 117 cases, CHP in 6 cases, and UC-ILD in 5 cases. CVD-IP included rheumatoid arthritis in 11 cases, antineutrophil cytoplasmic antibodyassociated vasculitis in 5 cases, polymyositis/dermatomyositis in 7 cases, and Sjögren's syndrome in 8 cases. Especially in the IPF group, the incidence of males was the highest, and %D Lco was the lowest. Te ILD-GAP score between IPF and UC-ILD was similar and higher than the other ILDs. Te antifbrotic agents were used in 10 patients, including 9 with Consecutive patients undergoing pulmonary function test including difusion capacity of lung for carbon monoxide due to respiratory diseases between April 2013 and April 2017 (n = 315) Excluded Chronic obstructive pulmonary disease n = 60 Pulmonary sarcoidosis n = 28 Lung cancer without interstitial lung disease (ILD) n = 13 Drug or radiation induced lung injuries n = 12 Cryptogenic organizing pneumonia n = 9 Bronchial asthma n = 6 Infectious pulmonary disease n = 2
IPF and 1 with iNSIP. Antiinfammatory agents, including corticosteroids or immunosuppressants, were used mainly in patients with CVD-IP or iNSIP. Te enrolled ILD patients were divided into 4 stages according to the ILD-GAP model (stage I, 117 cases; stage II, 57 cases; stage III, 10 cases; stage IV, 1 case). As shown in Figure 2, the Kaplan−Meier curves for predicting 3-year ILD-related events (P � 0.169) or 3year all-cause mortality (P � 0.153) proved to be not signifcant between the 4 stages.

Univariate Analysis of Primary Predictors of 3-Year ILD-Related Events.
To determine the primary predictors of 3year ILD-related events, univariate analysis was performed with the following parameters: age, sex, CCIS, diagnosis of ILD (IPF vs. non-IPF), ILD-GAP score, %FVC, and %D Lco (Table 3). Tis showed that CCIS, the ILD-GAP score, and the FVC were signifcant predictors of 3-year ILD-related events.

Accuracy of Composite Scoring Models in Predicting 3-
Year ILD-Related Events. It was hypothesized that the ILD-GAP model combined with CCIS (the ILD-GAPC model) is more accurate for predicting 3-year ILD-related events than the ILD-GAP model. Table 1 shows the details of ILD-GAPC scoring. Based on the previous research using nomogram analysis, CCIS was classifed into three categories and scored as follows: (0-1: 0 points, 2-3: 1 point, ≥4: 2 points) [7]. Ten, in the ILD-GAPC model, the score of CCIS was added to the ILD-GAP score. To investigate the accuracy of the ILD-GAP and ILD-GAPC models for 3-year ILD-related events, AUCs, C-index values, and AIC values for these models were calculated. All of the AUCs, C-index values, and AIC values were higher with the ILD-GAPC model than with the ILD-GAP model (Table 4). Figure 2, using the classifcation from the original ILD-GAP model, the number of stage III and IV cases is very small. Considering the equality of cases in each group, we attempted to change the staging of the ILD-GAP and ILD-GAPC models. Te Kaplan−Meier curves for 3-year ILD-related events were compared according to the ILD-GAP score (low score ≤1 point vs. moderate score 2, 3 points vs. high score ≥4 points), and the log-rank test showed that these groups did not difer signifcantly (Figure 3(a) (P � 0.083)). On the other hand, these curves were compared according to the ILD-GAPC score (low score ≤1 point vs. moderate score 2-3 points vs. high score ≥4 points), and the log-rank test showed that the Kaplan−Meier survival curves of these groups difered signifcantly (Figure 3(b) (P < 0.001)). Furthermore, logrank tests showed that the Kaplan−Meier curves for 3year all-cause mortality difered signifcantly among low,  Canadian Respiratory Journal middle, and high ILD-GAPC scores (P < 0.001) (Figure 3(d)), unlike ILD-GAP scores (P � 0.074) (Figure 3(c)).

Discussion
Although the clinical course and rate of progression of ILD are extremely variable among patients, clinical parameters for accurately predicting the prognosis of ILD have not been established [1,2]. From the viewpoint of clinical simplicity and versatility, various composite approaches such as GAP or ILD-GAP including age, sex, ILD diagnosis, and physiological measurements have been widely used to provide more accurate prognostic information in clinical settings [3,4]. However, these mortality prediction models do not take into account the presence or severity of comorbidities.
In the present study, the ILD-GAPC model was found to better predict 3-year ILD-related events and 3-year all-cause mortality than the ILD-GAP model. FVC is widely used as a biomarker in patients with ILD for predicting prognosis or evaluating treatment efcacy [3,4,[20][21][22][23][24][25][26]. Longitudinal variation of FVC is reported to be more reliable than baseline FVC, since baseline FVC may oversimplify the staging process because disease activity in patients with ILD does not always progress in a linear pattern [2,26]. Actually, we demonstrated that the most infuential prognostic factor was CCIS, not the baseline FVC. Te CCIS, as a summed score of 19 comorbidities weighted according to severity, was developed to assess the risk of death from comorbidities and has been widely used as   AIC, akaike's information criterion; AUROC, areas under the receiver operating characteristic curve; GAP, gender/age/physiology; GAPC, gender/age/ physiology/Charlson comorbidity indec score; ILD, interstitial lung disease. a prognostic indicator for patients with various diseases [11][12][13]. Also, in patients with ILD in both stable and AE conditions, the CCIS has been recently reported to be a prognostic indicator [6,7,14,15,27,28]. Interestingly, in the present study, although the number of events was small, the ILD-GAPC model was shown to more sensitively predict ILD-related events, including frst AE and mortality, rather than nonrespiratory mortality, for which only CVD-IP/ iNSIP patients showed nonrespiratory death, and the calculation of the ILD-GAPC score is expected to be a prognostic biomarker specifc to ILD (Supplement Figure 1). Tese suggest that the comorbidity itself has a direct impact on the progression of ILD, rather than simply coexisting with it. In order to prove this, it is necessary to analyze whether the defnitive treatment of comorbidities improves the prognosis of ILD, however, it can be said that this is a future task.
ILD can be associated with a large number of comorbidities, such as lung cancer, diabetes mellitus, coronary artery disease, heart failure, pulmonary hypertension (PH), gastroesophageal refux disease (GERD), and so on [29][30][31][32][33][34]. As shown above, the progression of comorbidities may be pathophysiologically linked to the progression of ILD itself; however, their prognostic impact and mechanism are not fully understood. Previous studies have revealed a high incidence of lung cancer in IPF (7% to 20%), though the true cumulative incidence of lung cancer after the diagnosis of IPF and its predictive factors at the initial diagnosis of IPF remain unknown. Various mechanisms such as endoplasmic reticulum stress, alterations of growth factors expression, oxidative stress, and large genetic and epigenetic variations, myofbroblast/mesenchymal transition, myofbroblast activation and proliferation can contribute to predispose the patient to develop IPF and lung cancer [29]. Diabetes mellitus is a systemic disorder characterized by a chronic hyperglycemic state that is associated with infammation and oxidative stress, leading to interstitial fbrosis and alveolar capillary microangiopathy [30]. MicroRNAs (miRNAs) regulate gene expression at the posttranscriptional level, contributing to all major cellular processes, including oxidative stress and cell death. Several miRNAs have been reported to crosstalk with oxidative stress in both the cardiac and pulmonary systems [31]. Fibrogenic mediators such as transforming growth factor-β promote fbroblast migration, proliferation, and activation in the heart and lungs [32]. Mechanisms contributing to the development of PH in patients with IPF are complex, including hypoxia causing smooth muscle hypertrophy and collagen deposition in pulmonary arteries, the destruction and obstruction of pulmonary vasculature by the progression of pulmonary fbrosis, and vascular remodeling contributed by fbroblast growth factor and platelet-derived growth factor [33]. In patients with GERD and IPF, microaspiration of gastric material may play a fundamental role in the fbrotic transformation of pulmonary parenchyma, and IPF may favor GERD by increasing the negative intrathoracic pressure [34]. From these, the progression of ILD seems to crosstalk with other comorbidities, suggesting that comorbidities may contribute to ILD-related events even if they do not directly cause death. Tus, high CCIS not only indicates an increased risk of death from comorbidities but may also indicate a poorer prognosis for ILD itself.
Te ILD-GAP model has been reported to accurately predict mortality in major chronic ILD subtypes such as IPF, iNSIP, CVD-IP, and CHP [4]. In the present study, the ILD-GAPC model was a better predictor of 3-year ILDrelated events than the ILD-GAP model, though there was a signifcant correlation between these models (Supplement Table 1). Although all patients in the high ILD-GAP score group were included in the high ILD-GAPC score group, the patients in the moderate ILD-GAP score group were divided into the moderate and high ILD-GAPC score groups, and the patients in the low ILD-GAP score group were divided into the low and moderate ILD-GAPC score groups. Te previously reported ILD-GAP model is a model for ILD patients with higher severity than in the present study, in fact, the enrolled patients in the original research on the ILD-GAP model had much lower %FVC and %D Lco than those in the present study [4]. Based on the above, the ILD-GAP model is considered a prognosis prediction model for severe cases, while the ILD-GAPC model is considered a highly versatile model for patients with a wide range of severity from mild to severe.
Although the ILD-GAPC model might have been shown to be a useful scoring system to predict the incidence of AE or future mortality in patients with ILDs, there are several limitations in the present research. Te number of enrolled patients was still small from a single institution. Especially, the clinical diagnoses of the patients enrolled with CHP or UC-ILD were much smaller than the others. Also, we used the CCIS as an assessment of the severity of ILD comorbidities, but we have not been able to compare it with other scoring model, such as the COPD specifc comorbidity test (COTE) index, and so on [35,36]. Te reproducibility of the fndings of this study needs to be confrmed through validation cohorts that increase the number of patients in the future. Te majority of patients enrolled were not so severely ill that pulmonary function tests, including D Lco could not be tolerated, which suggests a possible source of bias in the present research. Actually, the number of patients with a high ILD-GAP score is very small. Te ILD-GAPC model is useful for examining the long-term prognosis of relatively mild cases, and future validation including more severe cases is also necessary, though only in the %FVC >75% (%FVC score: 0 point) populations, we found that ILD-GAPC better predicted the 3-year ILD-related events than ILD-GAP (Supplement Figure S1). A treatable traits approach has been proposed as a new paradigm for the management of chronic lung diseases such as chronic airway disease, bronchiectasis, and ILD [37][38][39]. Especially in ILD, from the recent reports of the clinical efcacy of anti-fbrotic agents, the detection and severity evaluation of lung fbrosis or infammation as the treatable trait has become more important in considering therapeutic intervention [24,25,40]. As in the previous research, in which the CCIS proved to be an important prognostic indicator in patients with ILD, comorbidities such as lung cancer, cardiovascular disease, GERD, and PH have been reported to have prognostic impacts [8,9]. Tus, not only lung involvements but also CCIS seemed to be important treatable traits for patients with ILD, though it is unclear whether treatment for these comorbidities will improve the prognosis of ILD patients (Supplement Figure 3).

Conclusions
We speculate that comorbidity itself has a direct impact on the progression of ILD, rather than simply coexisting with it. Also, a high CCIS not only indicates an increased risk of death from comorbidities but may also indicate a poorer prognosis for ILD itself. From the above, the ILD-GAPC model could provide more accurate information for predicting prognosis in patients with ILD than the ILD-GAP model.

Data Availability
Te data used to support the fndings of this study are available from the corresponding author upon request.

Ethical Approval
Tis study followed the guidelines of the Declaration of Helsinki and was approved by the Institutional Review Board at Yokohama City University Hospital (approval number B190300005).

Consent
Not applicable.

Conflicts of Interest
Te authors declare that they have no conficts of interest.

Authors' Contributions
FH and HY conceptualized and designed the study, provided administrative support, involved in provision of study materials or patients, collected, assembled, analyzed, and interpreted data and wrote and approved the manuscript. SY conceptualized and designed the study, provided administrative support, analyzed and interpreted data, and wrote and approved the manuscript. TY, MK, NR, AA, IA, and SK conceptualized and designed the study, involved in provision of study materials or patients, analyzed and interpreted data, and wrote and approved the manuscript. WK, HN, KN, and KT conceptualized and designed the study, provided administrative support, involved in provision of study materials or patients, and wrote and approved the manuscript. All authors have read and approved the fnal manuscript. DisclosureTis research was performed as part of the employment of the Department of Pulmonology, Yokohama City University Graduate School of Medicine. A preprint has previously been published [41].