Rationale: In interstitial lung disease complicating systemic sclerosis (SSc-ILD), the optimal prognostic use of baseline pulmonary function tests (PFTs) and high-resolution computed tomography (HRCT) is uncertain.
Objectives: To construct a readily applicable prognostic algorithm in SSc-ILD, integrating PFTs and HRCT.
Methods: The prognostic value of baseline PFT and HRCT variables was quantified in patients with SSc-ILD (n = 215) against survival and serial PFT data.
Measurements and Main Results: Increasingly extensive disease on HRCT was a powerful predictor of mortality (P < 0.0005), with an optimal extent threshold of 20%. In patients with HRCT extent of 10–30% (termed indeterminate disease), an FVC threshold of 70% was an adequate prognostic substitute. On the basis of these observations, SSc-ILD was staged as limited disease (minimal disease on HRCT or, in indeterminate cases, FVC ⩾ 70%) or extensive disease (severe disease on HRCT or, in indeterminate cases, FVC < 70%). This system (hazards ratio [HR], 3.46; 95% confidence interval [CI], 2.19–5.46; P < 0.0005) was more discriminatory than an HRCT threshold of 20% (HR, 2.48; 95% CI, 1.57–3.92; P < 0.0005) or an FVC threshold of 70% (HR, 2.11; 95% CI, 1.34–3.32; P = 0.001). The system was evaluated by four trainees and four practitioners, with minimal and severe disease on HRCT defined as clearly < 20% or clearly > 20%, respectively, and the use of an FVC threshold of 70% in indeterminate cases. The staging system was predictive of mortality for all scorers, with prognostic separation higher for practitioners (HR, 3.39–3.82) than trainees (HR, 1.87–2.60).
Conclusions: An easily applicable limited/extensive staging system for SSc-ILD, based on combined evaluation with HRCT and PFTs, provides discriminatory prognostic information.
Optimal evaluation of prognosis in interstitial lung disease complicating systemic sclerosis, using a combination of high-resolution computed tomography and pulmonary function tests, is currently unavailable.
A combination of high-resolution computed tomography and pulmonary function test data, constructing a simple staging algorithm, is applicable to routine clinical practice and provides discriminatory prognostic information for interstitial lung disease due to systemic sclerosis.
However, the wide range of normal pulmonary function test (PFT) values (80–120% of predicted) is a major constraint. For example, a mildly reduced FVC level of 75% represents a clinically insignificant reduction from a premorbid FVC level of 80% but a striking decline from a premorbid value of 120%; thus, baseline FVC levels are most likely to be misleading when disease is either minimal or severe. In principal, the quantification of disease extent on high-resolution computed tomography (HRCT) would improve the accuracy of staging in these scenarios. However, because formal scoring of disease extent on HRCT is unrealistic in routine clinical practice, the optimal approach is likely to require a combination of semiquantitative HRCT evaluation and PFTs.
In this study, we identified the optimal HRCT extent threshold against mortality. For the purposes of analysis, HRCT subthresholds were set, such that disease extent was clearly below or above the threshold value and readily classifiable by less experienced observers as minimal or severe. A prognostic algorithm was created in which disease was classified as limited or extensive when disease extent on HRCT was overtly minimal or severe. In the remaining cases, with disease extent not readily classifiable as minimal or severe (i.e., “indeterminate” in extent), the distinction between limited and extensive disease was based on FVC threshold values. This staging system was compared with the use of HRCT and FVC in isolation. Some of this work has been presented in abstract form (8).
Consecutive patients (n = 330) referred to the Royal Brompton Hospital (London, UK) between January 1990 and December 1999 met criteria for SSc (9) with no overlap features: 277 (84%) had evidence of interstitial lung disease on HRCT. Investigations were performed as part of a routine clinical protocol. Exclusion criteria consisted of the following: (1) HRCT scans performed elsewhere, unavailable for scoring (n = 53); (2) baseline investigations separated by more than 90 days (n = 8); and (3) no follow-up (n = 1). The remaining 215 patients made up the study population. Excluded patients did not differ from the study cohort (see online supplement).
Forty-four of the 46 patients with complete data and no pulmonary fibrosis or pulmonary hypertension, as judged by HRCT and echocardiography, were compared with the study cohort.
Vital status at May 1, 2006, and serial PFT (2–12 monthly intervals) up to May 2006 were recorded. “Ever smokers” had smoked more than 1 cigarette/day for greater than 1 year. Treatment was defined as corticosteroid (prednisolone ⩾ 1mg/d) and/or immunosuppressant (cyclophosphamide, azathioprine, mycophenolate) therapy.
PFTs (% predicted), echocardiography, and HRCT were performed as previously reported (10, 11). Pulmonary hypertension was defined as a pulmonary arterial systolic pressure of 40 mm Hg or greater (12). HRCT scans were scored by two observers (A.U.W., S.R.D.) at five levels for total disease extent, extent of reticulation, proportion of ground glass, and coarseness of reticulation (see online supplement).
Mortality, disease progression (“time to decline” in either FVC levels of ⩾10% from baseline or DlCO levels of ⩾15% from baseline) and “progression-free survival” were quantified from the date of HRCT, as previously described (10, 13) (see online supplement).
Analyses were performed using STATA software (STATA Version 4; Computing Resource Centre, Santa Monica, CA). Data were expressed as means (SD) or medians (range), depending on distribution. Group comparisons were made using Student's t test, Wilcoxon rank sum, χ2 statistics and Fisher's exact test, as appropriate. A P value of less than 0.05 was considered significant. Outcome was examined using proportional hazards analysis (see online supplement).
Optimal HRCT extent and FVC thresholds, identified against mortality using proportional hazards analysis, were compared with alternative thresholds using a modified time-dependent receiver operating characteristic curve (ROC) analysis (14) (see online supplement). The prognostic value of a limited/extensive classification for disease extent, integrating HRCT observations and FVC values (as shown in Figure 1A), was compared with HRCT and FVC thresholds.
Observations were tested in two patient subgroups, matched for disease extent on HRCT (the patient with the median value was assigned to subgroup B). In subgroup A, the above analyses were repeated as a random subgroup test. In subgroup B, limited and extensive disease were defined as shown in Figure 1B: HRCT extent was graded as definitely greater than 20%, definitely less than 20%, or indeterminate by two radiologists, two physicians with more than 18 months of experience in interstitial lung disease and four trainees (two radiologists and two physicians, with <6 mo experience in interstitial lung disease). All sections from the arch of the aorta to the top of the hemidiaphragm were evaluated. Instruction was confined to three illustrative examples (with less than 10 min discussion).
As shown in Table 1, there was a female predominance, and a wide range of disease extent on HRCT (as shown in Figure E1 of the online supplement) and PFTs in the study cohort (n = 215). Age and male to female ratio did not differ significantly between the study cohort and the comparative group of 44 patients with no cardiopulmonary disease (age, 45.9 ± 11.5 yr; male to female ratio, 7:37). Nineteen of 215 patients (9%) had pulmonary hypertension. Sixty-one patients (28%) were receiving treatment at presentation and 38 patients (18%) had treatment introduced within the next 3 months. Seventy-five of 215 patients died (35%) (median follow-up, 89 mo; 10-yr survival, 59%). Decline in FVC was observed in 109 of 194 patients (56%; median time to decline, 61 mo). Decline in DlCO was also observed in 109 of 194 patients (56%; median time to decline, 64 mo). Progression (defined as death or deterioration in either FVC or DlCO) occurred in 150 of 194 patients (77%; median progression-free survival, 36 mo).
|Age, yr||49.1 ± 13.0|
|Sex, no. M:F||41:174|
|Smoking status||never = 123, ex = 74, current = 18|
|Baseline FEV1, % predicted||77.6 ± 18.6|
|Baseline FVC, % predicted||78.7 ± 21.4|
|Baseline DlCO, % predicted||55.1 ± 16.8|
|Extent of disease, %*||13.5 (1.0–84.0)|
|Extent of a reticular pattern, %*||6.5 (0–56.7)|
|Proportion of ground glass, %||49.0 ± 28.5|
|Coarseness (grades 0–3)*||5.5 (0–13.3)|
| Presence of honeycombing||53/215 (24.7%)|
Mortality was strongly linked to baseline DlCO levels, the extent of disease on HRCT, the extent of a reticular pattern on HRCT, and the presence of pulmonary hypertension on univariate analysis (Table 2). These severity variables were examined in separate multivariate models, with adjustment for age, sex, smoking status, the proportion of ground glass on HRCT, and the coarseness of reticulation on HRCT. HRCT disease extent was as predictive of mortality as DlCO levels and the presence of pulmonary hypertension (all P < 0.0005) and more predictive than the extent of a reticular pattern (P = 0.001) (Table E2).
|Variables||HR||95% CI||P Value|
|Extent of disease||1.03||1.01–1.04||<0.0005|
|Extent of a reticular pattern||1.05||1.03–1.07||<0.0005|
|Proportion of ground glass||0.99||0.98–0.99||0.001|
|Coarseness of reticulation||1.15||1.06–1.25||0.001|
|Presence of honeycombing||1.63||1.00–2.66||0.05|
|Presence/absence of PHT||4.78||2.68–8.54||<0.0005|
On univariate analysis, baseline PFTs, the extent of disease on HRCT, and the extent of a reticular pattern on HRCT were linked to decline in FVC, decline in DlCO and progression-free survival. The extent of a reticular pattern was the strongest determinant of time to decline in FVC and progression-free survival, whereas the presence of pulmonary hypertension was the strongest determinant of time to decline in DlCO. Full details are provided in the online supplement (Table E3).
The prognostic value of disease extent on HRCT was examined using threshold values, as shown in Table 3. An extent threshold of 20% on HRCT separated the study cohort into a larger subgroup (n = 151) with a 10-year survival of 67% and a smaller subgroup (n = 64) with a 10-year survival of 43%. In patients with extent of less than 20%, mortality was not related to the exact HRCT extent of disease (hazards ratio [HR], 1.02; 95% confidence interval [CI], 0.96–1.08; P = 0.52). Furthermore, compared with patients with no cardiopulmonary disease, mortality was higher in patients with HRCT extent of greater than 20% (HR, 3.03; 95% CI, 1.48–4.87; P = 0.001), but not in patients with HRCT extent of less than 20% (HR, 1.24; 95% CI, 0.64–2.34; P = 0.55) (Figure 2).
Extent of Disease (%)
Proportion of Patients (above/below cutoff)
On the basis of these observations, a cardinal HRCT extent threshold of 20% was selected. This threshold was further evaluated by comparing areas under the curve in a modified time-dependent ROC analysis for thresholds of 15, 20, and 25%, as shown in the online supplement (Figure E2). ROC values were significantly higher for a threshold value of 20% than with the use of threshold values of 15 or 25%.
Parallel analyses were performed in an attempt to identify an optimal threshold against mortality for the extent of a reticular pattern on HRCT. However, examining threshold values of 5, 10, 15, and 20% did not identify an optimal value (as shown in Table E4).
An FVC threshold of 70% corresponded to the optimal HRCT threshold of 20%, as judged by linear regression (Figure 3), and was selected for further evaluation. On proportional hazards analysis, this FVC threshold provided the greatest prognostic separation in the whole population, and in patients with HRCT disease extent between 10 and 30% (n = 92, 43%) (Table 4). FVC thresholds were further evaluated by comparing areas under the curve in a modified time-dependent ROC analysis for thresholds of 65, 70, and 75%, as shown in the online supplement (Figure E3). ROC values were significantly higher with the use of a threshold of 70%.
Patient Proportion (below/above threshold)
|Whole Cohort (n = 215)|
|10 to 30% Disease Extent on HRCT (n = 92)|
In keeping with the optimal HRCT extent threshold of 20%, subthreshold values of 10 and 30% were chosen for the purpose of further analysis, as indicative of easily classifiable minimal (limited; n = 87; see Figure E4) or severe (extensive; n = 36; see Figure E6) disease on HRCT. In the remaining 92 patients (43%) with an indeterminate extent of disease on HRCT (10–30%) (see Figure E5), FVC values were used to stage disease as limited or extensive as shown in Figure 1A. The following analyses were performed before the adaptation of the staging system for routine scoring.
The distinction between limited and extensive disease was more discriminatory than either HRCT (using an extent threshold of 20%) or FVC (using a threshold of 70%) in isolation (Figure 4).
The prognostic value of the staging system was reevaluated with adjustment for demographic data and the presence or absence of pulmonary hypertension. The staging system was the strongest determinant of mortality (HR, 3.66; 95% CI, 2.25–5.97; P < 0.0005) when age, sex, smoking status, and the presence of pulmonary hypertension were taken into account. The presence of pulmonary hypertension (HR, 2.26; 95% CI, 1.18–4.34; P = 0.01) and increasing age (HR, 1.03; 95% CI, 1.01–1.05; P < 0.005) were independently associated with increased mortality.
The limited/extensive staging system was strongly predictive of mortality, regardless of whether treatment was initiated or continued at presentation (n = 99; HR, 3.76; 95% CI, 1.91–7.40; P < 0.0005) or whether the initial management strategy was one of observation without immediate therapy (n = 115; HR, 3.02; 95% CI, 1.46–6.27; P < 0.005).
When retested as a random subgroup analysis (subgroup A, n = 107), the limited/extensive system (HR, 3.35; 95% CI, 1.64–6.84; P = 0.001) was more discriminatory than an HRCT disease extent threshold of 20% (HR, 2.51; 95% CI, 1.23–5.09; P = 0.01) or an FVC threshold of 70% (HR, 2.41; 95% CI, 1.19–4.88; P = 0.02).
For use in routine practice, formal HRCT scoring was substituted by an estimation of disease extent as definitely less than 20%, definitely more than 20%, or indeterminate (i.e., observer difficulty in extent subcategorization). Each observer staged disease severity as limited or extensive based on disease extent on HRCT and FVC levels, as shown in Figure 1B.
HRCT scans, evaluated by two radiologists, two physicians, and four trainees, required 1.49 ± 0.93 minutes/scan for evaluation. Interobserver agreement was good for practitioners (κ = 0.64), and fair to moderate for trainees (κ = 0.41). As shown in Table 5, the staging system was predictive of mortality for all scorers, with prognostic separation higher for practitioners (HR, 3.39–3.82) than trainees (HR, 1.87–2.60). The staging system was also predictive (n = 7) or marginally predictive (n = 1) of progression-free survival (Table 5; Figure 5).
|HR||95% CI||P Value||HR||95% CI||P Value|
We propose an easily applicable prognostic algorithm for SSc-ILD, validated against mortality and requiring (1) the rapid identification of overtly minimal or severe disease on HRCT, based on a disease extent threshold of 20%, and (2) the use of an FVC threshold of 70% in the remaining cases (with an indeterminate extent of disease on HRCT). Thus, HRCT is used to stage severity, provided that findings are clear-cut, with recourse to an FVC threshold of 70% in marginal cases requiring greater radiologic expertise for HRCT classification. The staging of interstitial lung disease as limited or extensive had a greater prognostic value than either HRCT or FVC thresholds in isolation, when HRCT extent was formally scored. The prognostic separation between limited and extensive disease was retained when tested by experienced practitioners, confirming that formal HRCT scoring, which is not practicable in routine practice, can be avoided. The HRCT staging of patients with obviously minimal or severe disease reduces misclassification of severity by an FVC threshold (due to the wide range in normal premorbid values). More importantly, the proposed system provides, for the first time in SSc-ILD, an easily applicable and rapid means of integrating HRCT extent and functional severity in routine prognostic evaluation.
Formal HRCT scoring was used to construct the staging system and to perform preliminary ancillary analyses. For this purpose, HRCT disease extent thresholds of 10 and 30%, equidistant to 20%, were used to identify patients readily classifiable as having minimal or severe disease, with FVC levels used for staging when disease extent lay between 10 and 30%. The prognostic distinction between limited and extensive disease was not weakened when the presence of pulmonary hypertension and demographic factors were taken into account. Furthermore, the prognostic separation was equally striking in patients considered to require immediate treatment and in the remaining patients in whom observation was initially believed to be appropriate.
However, the true test of the limited/extensive system was its application using rapid semiquantitative estimation of disease extent by four practitioners and four trainees. Strikingly, the two physicians with less than 2 years' experience in interstitial lung disease achieved prognostic separations, both in mortality and progression-free survival, comparable to those of two experienced radiologists, despite a training period of only 10 minutes. It is possible that the group κ value of 0.64, indicative of good interobserver agreement (15), would have risen significantly with lengthier training, but it was considered desirable to assess the system as it might be applied by observers in routine practice, with no formal training available. The lesser accuracy of trainees, as judged by outcome analyses and interobserver variation, is indicative of a significant training effect.
HRCT threshold values tested in analysis were confined to those that can practicably be identified on rapid semiquantitative HRCT evaluation. Thus, prognostic distinctions were examined using 5% increments in disease extent. The chosen threshold of 20% was selected because it signaled a substantial increase in mortality, compared with patients with no cardiopulmonary disease. The use of alternative thresholds of 15 or 25% failed to capture this key threshold effect, and this was confirmed using a recently described time-dependent ROC methodology (14). An HRCT extent threshold of 30% was equally discriminatory, but with this threshold a much smaller subgroup with extensive disease (n = 36) was selected, and in patients with less extensive disease, subgroups with very heterogeneous outcomes (<20%, 20–30%) were amalgamated.
The FVC threshold of 70% corresponded exactly to an HRCT disease extent of 20% on linear regression and could be justified on two other grounds. First, an examination of FVC thresholds at 5% increments (to parallel the HRCT thresholds) showed that an FVC threshold of 70% was optimal on proportional hazards analysis, and this observation was confirmed using time-dependent ROC methodology (14). Second, in the recent scleroderma lung studies (16, 17), an FVC level of 70% was the threshold below which therapy was effective in preventing disease progression, an effect that was lost with the cessation of treatment.
It should be stressed that the aim of this study was to construct a simple staging system and not a complex multivariate index in which all possible prognostic information is included. In idiopathic pulmonary fibrosis, the clinical-radiographic-physiologic indices have provided important insights into the disease (18, 19) but are not applied in routine practice because of their complexity. Furthermore, clinical-radiographic-physiologic scores and other multivariate systems are continuous and do not separate patients into discrete groups according to disease severity in a clinically useful fashion. The aim of our staging system was to enable clinicians to readily classify patients into two groups (“high risk” and “low risk”), based on obvious differences in outcome, using HRCT extent and FVC thresholds that can be realistically evaluated in clinical practice. Importantly, in preliminary proportional hazards analyses, disease extent on HRCT was shown to provide prognostic information that was comparable to other severity variables that are not specific to the interstitium (DlCO levels, the presence of pulmonary hypertension) or require expert radiologic assessment (the extent of reticulation on HRCT).
One limitation of our study is that the extent threshold at which treatment should be instituted cannot be distilled from our data. Patients with extensive disease had a worse outcome, despite the fact that the majority were treated from presentation, and this suggests that, ideally, treatment should be instituted earlier in the disease course, in the hope of preventing progression to the 20% HRCT extent/70% FVC thresholds. However, the therapeutic approach in limited disease was too variable to allow subanalysis, because it was also influenced by other prognostic factors, including observed disease progression.
Therapeutic status was categorized coarsely, according to whether treatment was believed to be warranted at presentation. The staging system provided equivalent prognostic information regardless of whether patients were treated or observed. This broad distinction was necessary because the later introduction of treatment could not be accounted for in analysis. The choice, timing, and duration of treatment varied widely during follow-up, with introduced therapy often modified or withdrawn due to side effects. Our approach has the advantage of categorizing therapeutic status at the time at which baseline information was available, thereby taking treatment into account using an “intention to prognosticate” approach.
A further limitation related to time to decline in PFTs. This variable is a relatively coarse outcome measure, compared with mortality, for which follow-up time is generally collated at monthly intervals. Patients with mild disease are often monitored infrequently. Thus, it is possible that, in some cases, functional decline might have been missed for many months. However, to put this limitation in perspective, it should be stressed that the maximum time interval for PFT follow-up and thus the maximum inaccuracy in time to decline analyses was 12 months, with the total follow-up averaging over 5 years. Furthermore, this imprecision in outcome applies equally to assessment of all baseline variables and should not, in principle, favor one variable over another in prognostic evaluation.
In conclusion, we propose a simple staging system for SSc-ILD as limited or extensive disease, based on simplified HRCT evaluation and FVC estimation, that provides more powerful prognostic information than either component in isolation. The system was readily applied by more experienced observers, who had been minimally trained in its use. In principle, this means of risk stratification is equally applicable to routine practice and the enrollment of patients in pharmaceutical studies.
|1.||Ferri C, Valentini G, Cozzi F, Sebastiani M, Michelassi C, La Montagna G, Bullo A, Cazzato M, Tirri E, Storino F, et al. Systemic sclerosis: demographic, clinical, and serologic features and survival in 1,012 Italian patients. Medicine (Baltimore) 2002;81:139–153.|
|2.||Scussel-Lonzetti L, Joyal F, Raynauld JP, Roussin A, Rich E, Goulet JR, Raymond Y, Senecal JL. Predicting mortality in systemic sclerosis: analysis of a cohort of 309 French Canadian patients with emphasis on features at diagnosis as predictive factors for survival. Medicine (Baltimore) 2002;81:154–167.|
|3.||Simeon CP, Armadans L, Fonollosa V, Solans R, Selva A, Villar M, Lima J, Vaque J, Vilardell M. Mortality and prognostic factors in Spanish patients with systemic sclerosis. Rheumatology (Oxford) 2003;42:71–75.|
|4.||Morgan C, Knight C, Lunt M, Black CM, Silman AJ. Predictors of end stage lung disease in a cohort of patients with scleroderma. Ann Rheum Dis 2003;62:146–150.|
|5.||Steen VD, Conte C, Owens GR, Medsger TA Jr. Severe restrictive lung disease in systemic sclerosis. Arthritis Rheum 1994;37:1283–1289.|
|6.||Steen VD, Medsger TA Jr. Severe organ involvement in systemic sclerosis with diffuse scleroderma. Arthritis Rheum 2000;43:2437–2444.|
|7.||Bouros D, Wells AU, Nicholson AG, Colby TV, Polychronopoulos V, Pantelidis P, Haslam PL, Vassilakis DA, Black CM, du Bois RM. Histopathologic subsets of fibrosing alveolitis in patients with systemic sclerosis and their relationship to outcome. Am J Respir Crit Care Med 2002;165:1581–1586.|
|8.||Goh NS, Desai SR, Veeraraghavan S, Hansell DM, Denton CP, Black CM, du Bois RM, Wells AU. The prognostic significance of HRCT in the interstitial lung disease of systemic sclerosis (SSc-ILD) [abstract]. Am J Respir Crit Care Med 2007;175:A297.|
|9.||Masi AT, Rodnan GP, Medsger TA Jr. Preliminary criteria for the classification of systemic sclerosis (scleroderma). Arthritis Rheum 1980;23:581–590.|
|10.||Goh NS, Veeraraghavan S, Desai SR, Cramer D, Hansell DM, Denton CP, Black CM, du Bois RM, Wells AU. Bronchoalveolar lavage cellular profiles in patients with systemic sclerosis-associated interstitial lung disease are not predictive of disease progression. Arthritis Rheum 2007;56:2005–2012.|
|11.||Desai SR, Veeraraghavan S, Hansell DM, Nikolakopolou A, Goh NS, Nicholson AG, Colby TV, Denton CP, Black CM, du Bois RM, et al. CT features of lung disease in patients with systemic sclerosis: comparison with idiopathic pulmonary fibrosis and nonspecific interstitial pneumonia. Radiology 2004;232:560–567.|
|12.||Plastiras SC, Karadimitrakis SP, Kampolis C, Moutsopoulos HM, Tzelepis GE. Determinants of pulmonary arterial hypertension in scleroderma. Semin Arthritis Rheum 2007;36:392–396.|
|13.||Raghu G, Brown KK, Bradford WZ, Starko K, Noble PW, Schwartz DA, King TE Jr. A placebo-controlled trial of interferon gamma-1b in patients with idiopathic pulmonary fibrosis. N Engl J Med 2004;350:125–133.|
|14.||Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005;61:92–105.|
|15.||Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam Med 2005;37:360–363.|
|16.||Tashkin DP, Elashoff R, Clements PJ, Goldin J, Roth MD, Furst DE, Arriola E, Silver R, Strange C, Bolster M, et al. Cyclophosphamide versus placebo in scleroderma lung disease. N Engl J Med 2006;354:2655–2666.|
|17.||Tashkin DP, Elashoff R, Clements PJ, Roth MD, Furst DE, Silver RM, Goldin J, Arriola E, Strange C, Bolster MB, et al. Effects of 1-year treatment with cyclophosphamide on outcomes at 2 years in scleroderma lung disease. Am J Respir Crit Care Med 2007;176:1026–1034.|
|18.||Watters LC, King TE, Schwarz MI, Waldron JA, Stanford RE, Cherniack RMA. Clinical, radiographic, and physiologic scoring system for the longitudinal assessment of patients with idiopathic pulmonary fibrosis. Am Rev Respir Dis 1986;133:97–103.|
|19.||King TE Jr, Tooze JA, Schwarz MI, Brown KR, Cherniack RM. Predicting survival in idiopathic pulmonary fibrosis: scoring system and survival model. Am J Respir Crit Care Med 2001;164:1171–1181.|