Abstract
Background
Interstitial lung disease (ILD) is an important cause of morbidity and mortality in patients with scleroderma (Scl). Risk prediction and prognostication in patients with Scl-ILD are challenging because of heterogeneity in the disease course.
Methods
We aimed to develop a clinical mortality risk prediction model for Scl-ILD. Patients with Scl-ILD were identified from two ongoing longitudinal cohorts: 135 patients at the University of California, San Francisco (derivation cohort) and 90 patients at the Mayo Clinic (validation cohort). Using these two separate cohorts, a mortality risk prediction model was developed and validated by testing every potential candidate Cox model, each including three or four variables of a possible 19 clinical predictors, for time to death. Model discrimination was assessed using the C-index.
Results
Three variables were included in the final risk prediction model (SADL): ever smoking history, age, and diffusing capacity of the lung for carbon monoxide (% predicted). This continuous model had similar performance in the derivation (C-index, 0.88) and validation (C-index, 0.84) cohorts. We created a point scoring system using the combined cohort (C-index, 0.82) and used it to identify a classification with low, moderate, and high mortality risk at 3 years.
Conclusions
The SADL model uses simple, readily accessible clinical variables to predict all-cause mortality in Scl-ILD.
Key Words: interstitial lung disease, prognosis, risk prediction, systemic sclerosis
Abbreviations: ANA, anti-nuclear antibody; Dlco, diffusing capacity of the lung for carbon monoxide; GERD, gastroesophageal reflux disease; HRCT, high-resolution computed tomography; ILD, interstitial lung disease; IPF, idiopathic pulmonary fibrosis; PH, pulmonary hypertension; RHC, right heart catheterization; SADL, ever smoking history, age, and diffusing capacity of the lung for carbon monoxide (% predicted); Scl, scleroderma; Scl-ILD, scleroderma-related interstitial lung disease; UIP, usual interstitial pneumonia
Pulmonary involvement, including pulmonary hypertension (PH) and interstitial lung disease (ILD), is common in scleroderma and is associated with significant morbidity and mortality.1, 2 The prevalence of ILD in scleroderma is high and has been reported to occur in up to 90% of patients.3 ILD is the most frequent cause of death among patients with scleroderma.1
Although many patients with scleroderma-related interstitial lung disease (Scl-ILD) will progress over time,4 the disease course is heterogeneous.5, 6, 7 This variability in disease evolution makes prognosis a challenging task for clinicians. Previous studies have attempted to identify factors that predict progression and mortality in Scl-ILD,8, 9, 10, 11, 12, 13, 14, 15, 16 but were limited by sample size, lack of validation, and model complexity.17 Goh et al11 proposed a staging system for patients with Scl-ILD that combines FVC (% predicted) and extent of ILD on high-resolution CT (HRCT) scanning to classify patients as having either limited or extensive disease. This staging system was found to be predictive of mortality in a single cohort of patients with Scl-ILD.
The goal of this study was to develop and validate a simple-to-use and practical all-cause mortality risk prediction model for patients with Scl-ILD using widely available clinical variables at the time of their initial visit, applying an approach we previously used in idiopathic pulmonary fibrosis (IPF).18 We demonstrate that 3-year mortality in Scl-ILD can be estimated using the SADL (ever smoking history, age, and diffusing capacity of the lung for carbon monoxide (% predicted)) model in two independent cohorts of patients with Scl-ILD.
Methods
Study Population
We retrospectively identified patients from ongoing longitudinal prospective cohorts at their respective centers. Patients were included in the study if they had received a rheumatologist-confirmed diagnosis of scleroderma19, 20 and had evidence of ILD on HRCT scanning and/or surgical lung biopsy. Patients without ILD on HRCT scanning and/or surgical lung biopsy were excluded (Fig 1). The derivation cohort included 135 patients from the University of California, San Francisco Interstitial Lung Disease Program. The validation cohort included 90 patients from the ILD clinic at the Mayo Clinic (Rochester, MN). Institutional review boards at both centers approved the parent databases (San Francisco Institutional Human Subject Review Committee [#10-01592] and Mayo Clinic Institutional Review Board [#12-009206]), and patients provided written informed consent.
Figure 1.
Cohort formation. Patients were excluded if they did not have interstitial lung disease on high-resolution CT and/or surgical lung biopsy.
Clinical Data
Demographic variables including age, sex, body mass index, ever smoking history, ethnicity, and long-term oxygen use were obtained from structured questionnaires at the time of their initial ILD clinic visit and medical chart review. Results of pulmonary function tests performed within 6 months of the ILD visit date were obtained. Scleroderma-specific variables, including comorbid conditions (eg, diffuse skin involvement, Raynaud’s disease, gastroesophageal reflux disease [GERD], renal disease), serologic profile (eg, anti-nuclear antibody [ANA], anti-topoisomerase 1 antibody, anti-centromere antibody), and immunosuppressive therapy, were obtained by chart review. Patients were identified as having “signs suggestive of pulmonary hypertension on echocardiogram” if the estimated pulmonary artery systolic pressure was higher than 36 mm Hg.21
An expert thoracic radiologist re-reviewed all available HRCT scans performed within 12 months of the initial ILD clinic visit date. The HRCT scan patterns were categorized as definite usual interstitial pneumonia (UIP), possible UIP, or inconsistent with UIP, according to published guidelines.22 In addition, the total extent of ILD on HRCT scanning was evaluated using previously described methods in the derivation cohort.11, 23 Briefly, the extent of disease was estimated (to the nearest 5%) at five different levels: (1) origin of great vessels, (2) carina, (3) pulmonary venous confluence, (4) halfway between levels 3 and 5, and (5) 1 cm above the right hemidiaphragm. The total extent of disease was calculated as the mean of the five scores. The presence or absence of emphysema was also recorded in the derivation cohort.
Outcome
The primary outcome was mortality. Medical chart review and the US Social Security Death Index were used to determine whether patients had died during the follow-up period. Patients who underwent lung transplantation were censored at the time of transplantation. Cause of death was available for the Mayo Clinic patients.
Statistical Analysis
The mortality risk prediction model was developed and validated using a previously described method.18 Using the derivation cohort, we estimated every potential candidate Cox model for time to death, each including three or four of a possible 19 clinical predictors, including age, sex, ethnicity, history of smoking, body mass index, Raynaud’s disease, skin involvement, GERD, renal disease, anti-topoisomerase 1 antibody, anti-centromere antibody, FVC (% predicted), Dlco (% predicted), definite UIP on HRCT scanning, extent of ILD on HRCT scanning, presence of emphysema on HRCT scanning, oxygen use, any immunosuppressive therapy, and signs suggestive of pulmonary hypertension on echocardiogram. The models were ranked by the C-index, a measure of discrimination, estimated using 20 repetitions of 10-fold cross-validation,24 with 95% bootstrap percentile confidence intervals, and the model with the highest cross-validated C-index was chosen. In preliminary analysis, we checked for nonlinearity effects of continuous predictors on mortality risk, as well as for interactions of all other factors with signs suggestive of pulmonary hypertension on echocardiogram. We then used the validation cohort to reassess the C-index of the selected model, also with a 95% bootstrap percentile confidence interval. We assessed calibration in each of the two cohorts, by comparing model-based and nonparametric estimates of mortality at 1, 2, and 3 years.
Next, we derived a point score system based on the original model, using data from the combined cohorts. Specifically, we refit the Cox model for time to death after categorization of the two continuous predictors (age and Dlco [% predicted]) at clinically meaningful cut points. Each of the resulting categories was assigned one to four points by rescaling and rounding the resulting regression coefficients. A classification system was developed by grouping the point score results into three clinically meaningful categories: low, moderate, and high risk. The discrimination and calibration of this final model were assessed by cross-validation in the combined cohort. Finally, we compared our model performance with that of the staging system proposed by Goh et al11 in the derivation cohort. Briefly, the grading system proposed by Goh et al11 includes grading HRCT scans according to the extent of lung disease (< 20%, limited; > 20%, extensive) and among those that were indeterminate, limited or extensive disease was defined by an FVC threshold of 70% predicted. STATA version 14 (Stata Corp) was used for all the statistical analyses.
Results
Patient Characteristics
Table 1 shows the baseline characteristics of all patients and of those in each cohort. As a group, the patients were predominantly white women with a mean age of 55.5 ± 12.3 years. A history of current or past smoking was reported in 42% of patients. Of those patients evaluated for Raynaud’s disease, most had Raynaud’s disease (95%) and diffuse skin involvement (82%), and of those for whom serologic data were available, all had a positive ANA result. Thirty-seven percent of patients had signs suggestive of pulmonary hypertension on echocardiogram. However, only 21 patients (9%) were receiving vasodilator therapies for pulmonary hypertension at their baseline visit (Table 2). The mean FVC (% predicted) was 71 and the mean Dlco (% predicted) was 50. Surgical lung biopsy was rarely performed in this population (9%).
Table 1.
Patient Characteristics
| Characteristic | All (N = 225) |
Derivation Cohort (n = 135) |
Validation Cohort (n = 90) |
P Value |
|---|---|---|---|---|
| Age, y (SD) | 55.5 (12.3) | 54.4 (12.8) | 57.1 (11.2) | .10 |
| Female, No./total (%) | 163/225 (72) | 101/135 (75) | 62/90 (69) | .33 |
| White, No./total (%) | 170/225 (76) | 87/135 (64) | 83/90 (92) | < .001 |
| Asian, No./total (%) | 21/225 (9) | 18/135 (13) | 3/90 (3) | .01 |
| Black or African American, No./total (%) | 14/225 (6) | 10/135 (7) | 4/90 (4) | .35 |
| Hispanic or Latino, No./total (%) | 12/225 (5) | 12/135 (9) | … | … |
| Native American, No./total (%) | 4/225 (2) | 4/135 (3) | … | … |
| Pacific Islander, No./total (%) | 4/225 (2) | 4/135 (3) | … | … |
| Ever smoker, No./total (%) | 94/225 (42) | 52/135 (39) | 42/90 (47) | .23 |
| Raynaud’s, No./total (%)a | 168/176 (95) | 85/93 (91) | 83/83 (100) | .01 |
| Diffuse skin involvement, No./total (%)a | 142/173 (82) | 64/83 (77) | 78/90 (87) | .09 |
| Renal disease, No./total (%)a | 9/185 (5) | 4/95 (4) | 5/90 (6) | .67 |
| GERD, No./total (%)a | 133/194 (69) | 64/104 (62) | 69/90 (77) | .02 |
| Signs suggestive of pulmonary hypertension on echocardiogram, No./total (%)a,b | 72/194 (37) | 40/104 (38) | 32/90 (36) | .68 |
| Anti-nuclear antibody, No./total (%)a | 167/167 (100) | 87/87 (100) | 80/80 (100) | … |
| Anti-topoisomerase I antibody, No./total (%)a | 59/145 (41) | 34/73 (47) | 25/75 (35) | .15 |
| Anti-centromere antibody, No./total (%)a | 20/102 (20) | 9/41 (22) | 11/61 (18) | .63 |
| FVC | .50 | |||
| Percent predicted (SD) | 71.0 (19.7) | 71.9 (22.6) | 69.9 (15.2) | |
| No./totala | 194/225 | 110/135 | 84/90 | |
| Dlco | .57 | |||
| Percent predicted (SD) | 50.0 (18.6) | 50.7 (20.4) | 49.2 (12.2) | |
| No./totala | 191/225 | 107/135 | 84/90 | |
| HRCT, No./total (%)c | 155/225 (69) | 89/135 (66) | 66/90 (73) | .24 |
| Definite UIP, No./total (%)a | 14/155 (9) | 4/89 (5) | 10/66 (15) | .02 |
| Possible UIP, No./total (%)a | 23/155 (15) | 10/89 (11) | 13/66 (20) | .14 |
| Inconsistent with UIP, No./total (%)a | 118/155 (76) | 75/89 (84) | 43/66 (65) | .01 |
| Extent of disease, median % (IQR)a | … | 14.9 (4-23) | … | … |
| Presence of emphysema, No./total (%)a | … | 6/89 (6.7) | … | … |
| Surgical lung biopsy, No./total (%) | 20/225 (9) | 11/135 (8) | 9/90 (10) | .60 |
| UIP pattern, No./total (%) | 5/20 (25) | 2/11 (18) | 3/9 (33) | .44 |
| NSIP pattern, No./total (%) | 6/20 (30) | 4/11 (36) | 2/9 (22) | .50 |
| Others, No./total (%) | 9/20 (55) | 5/11 (45) | 4/9 (44) | .96 |
| Staging system, No./total (%)d | ||||
| Limited diseasea | … | 59/89 (66) | … | … |
| Extensive diseasea | … | 30/89 (34) | … | … |
Dlco = diffusing capacity of the lung for carbon monoxide; GERD = gastroesophageal reflux disease; HRCT = high-resolution CT; NSIP = nonspecific interstitial pneumonia; UIP = usual interstitial pneumonia.
These variables have missing data as indicated by the denominator reporting the total number of observations.
Pulmonary hypertension was defined by an estimated pulmonary artery systolic pressure higher than 36 mm Hg on echocardiogram.21
According to the published guidelines.22
Using the staging system proposed by Goh et al.11
Table 2.
Treatment Data
| Characteristic | All (N = 225) |
Derivation Cohort (n = 135) |
Validation Cohort (n = 90) |
P Value |
|---|---|---|---|---|
| Long-term oxygen therapy, No./total (%)a | 26/194 (13) | 4/104 (4) | 22/90 (24) | < .001 |
| Immunosuppressive therapy, No./total (%)a | 106/196 (47) | 43/106 (41) | 63/90 (70) | < .001 |
| Prednisonea | 73/196 (37) | 32/106 (30) | 41/90 (46) | .03 |
| Cytoxana | 17/195 (9) | 5/105 (5) | 12/90 (13) | .03 |
| Mycophenolate mofetila | 54/195 (28) | 24/105 (23) | 30/90 (33) | .10 |
| Vasodilator therapies, No./total (%)a | 21/225 (9) | 8/105 (8) | 13/90 (14) | .13 |
These variables have missing data as indicated by the denominator reporting the total number of observations.
Patients in the derivation and validation cohorts exhibited some differences at baseline (Table 1). Patients in the validation cohort were more likely to be white and to have a history of Raynaud’s disease and GERD. A definite UIP pattern on HRCT scanning was more common in the validation cohort (15%) than in the derivation cohort (5%; P = .02). In addition, the proportion of patients receiving oxygen and immunosuppressive treatment at baseline differed between study cohorts (Table 2). Patients in the validation cohort were more likely to have been receiving immunosuppressive therapy (70%) at the time of their initial ILD clinic visit compared with the derivation cohort (41%; P < .001). Additional baseline characteristics are presented in e-Table 1 of the online supplement.
During the follow-up period, 56 patients died: 27 in the derivation cohort and 29 in the validation cohort. Five patients in the derivation cohort underwent lung transplantation; none occurred in the validation cohort. The median (range) follow-up time was 3.15 (0.07-15.16) and 4.64 (0.1-12.31) years in the derivation and validation cohorts, respectively. We were not able to identify the cause of death for patients in the derivation cohort. The majority of the deaths (70%) in the validation cohort were attributable to respiratory causes.
SADL Model Derivation and Validation
The model screening process identified a three-variable continuous model including history of ever smoking, age, and Dlco (% predicted) (SADL). The cross-validated C-index for the selected continuous model was 0.88 (95% CI, 0.78-0.95) in the derivation cohort and 0.84 (95% CI, 0.75-0.91) in the validation cohort. Table 3 shows the calibration of this continuous model in the derivation and validation cohorts for 1-, 2-, and 3-year mortality.
Table 3.
Continuous Model Performance in Derivation and Validation Cohorts
| Mortality | C-Index (95% CI) for Derivation Cohort: 0.88 (0.78-0.95) |
C-Index (95% CI) for Validation Cohort: 0.84 (0.75-0.91) |
||
|---|---|---|---|---|
| Predicted | Observed (95% CI) | Predicted | Observed (95% CI) | |
| 1 y, % | 2.9 | 3.0 (1-9.1) | 6.1 | 4.9 (1.9-12.6) |
| 2 y, % | 6.4 | 6.5 (3-14) | 11.8 | 16.6 (9.7-27.5) |
| 3 y, % | 14.0 | 13.3 (7.6-23) | 21.1 | 28.6 (19.1-41.3) |
Next, a point score system was developed by assigning values to subcategories within each of the three variables of the continuous SADL model, using both cohorts. On the basis of the SADL point score model, patients in the combined cohort with scores of 0 to 3, 4 or 5, and 6 or 7 points were classified as low, moderate, and high risk, respectively (Table 4). The distribution of patients within each risk category is presented in Table 5, and the results of the multivariable survival analysis using the point score SADL model are shown in Table 6. The C-index of the point score SADL model was 0.82 (95% CI, 0.77-0.87), and its calibration was satisfactory (Table 7). The C-index of the point score SADL model was similar in the individual cohorts: derivation cohort (C-index, 80.20; 95% CI, 71.76-88.29) and validation cohort (C-index, 82.92; 95% CI, 75.69-90.16). Figure 2 shows the cumulative mortality difference within the combined cohort when categorized by the three risk groups. We performed a sensitivity analysis excluding nonrespiratory deaths from the validation cohort. The performance of the SADL model was not affected by the exclusion of these patients (C-index, 0.84; 95% CI, 0.74-0.92) and provided satisfactory calibration. An additional sensitivity analysis was performed to evaluate the SADL model excluding patients with signs suggestive of PH on echocardiography. Again, the performance of the model was similar to the combined cohort (data not shown).
Table 4.
SADL Model and Staging System
| Predictor | Points | |
|---|---|---|
| S | Ever smoking history | |
| No | 0 | |
| Yes | 1 | |
| A | Age, y | |
| < 55 | 0 | |
| 55-70 | 1 | |
| > 70 | 2 | |
| DL | Dlco (% predicted) | |
| > 60% | 0 | |
| 40-60% | 3 | |
| < 40% | 4 |
| Total possible points: |
7 |
||
|---|---|---|---|
| Risk Category | |||
| Low | Moderate | High | |
| Points | 0-3 | 4-5 | 6-7 |
SADL = ever smoking history, age, and diffusing capacity of the lung for carbon monoxide (% predicted). See Table 1 legend for expansion of other abbreviation.
Table 5.
Distribution of Patients by SADL Score and Risk Category
| Risk Category Based on SADL Score |
|||
|---|---|---|---|
| Low Risk | Moderate Risk | High Risk | |
| Points | 0-3 | 4-5 | 6-7 |
| Distribution in combined cohort, % | 48 | 31 | 21 |
| Distribution in derivation cohort, % | 42 | 35 | 23 |
| Distribution in validation cohort, % | 56 | 24 | 20 |
See Table 4 legend for expansion of abbreviation.
Table 6.
Multivariate Survival Analysis Using SADL Point Score
| Point Score Variable | Hazard Ratio (95% CI) | P Value |
|---|---|---|
| Ever smoking history | 2.58 (1.32-5.02) | .005 |
| Age, y | ||
| 55-70 | 2.24 (1.00-5.03) | .05 |
| > 70 | 7.48 (3.03-18.5) | < .001 |
| Dlco (% predicted) | ||
| 40%-60% | 18.3 (2.24-150) | .007 |
| < 40% | 40.9 (4.97-337) | .001 |
Table 7.
Point Score Model Performance in Combined Cohort
| C-Index (95% CI) for Combined Cohort: 0.82 (0.77-0.87) |
||
|---|---|---|
| Predicted | Observed (95% CI) | |
| 1-y mortality, % | ||
| Low risk | 0.5 | 0 (0-5.6)a |
| Moderate risk | 3.8 | 4.7 (1.8-12.1) |
| High risk | 12.8 | 11.1 (3.7-30.6) |
| 2-y mortality, % | ||
| Low risk | 1.6 | 0 (0-6.1)a |
| Moderate risk | 11.3 | 10.5 (5.3-20) |
| High risk | 34.0 | 40.4 (24-62.3) |
| 3-y mortality, % | ||
| Low risk | 3.2 | 0 (0-7.0)a |
| Moderate risk | 22.2 | 22.8 (14.3-35.3) |
| High risk | 56.9 | 63.3 (44.3-82.1) |
When no events have occurred, Greenwood confidence intervals for the Kaplan-Meier incidence proportion are undefined. We substituted exact binomial confidence intervals for this proportion, with zero events, and the number at risk at the end of the year as the denominator.
Figure 2.
Cumulative mortality in the combined cohort by risk category. Red line, high-risk patients; blue line, moderate-risk patients; dashed gray line, low-risk patients.
Finally, we applied the staging system proposed by Goh et al11 to our derivation cohort to compare its performance with that of the SADL model in predicting mortality in patients with Scl-ILD. Although the calibration of this staging system was satisfactory in our cohort (e-Table 2 and e-Fig 1), the discrimination of the Goh et al11 staging system was significantly lower (C-index, 0.62; 95% CI, 0.50-0.74) than the SADL model.
Discussion
The SADL model is an easy-to-use mortality risk prediction classification system designed for patients with Scl-ILD. It is based on two clinical variables (age and ever smoking history) and one physiological variable (Dlco [% predicted]), which are commonly used and easy to obtain in clinical practice at the initial clinic visit. Model discrimination for all-cause mortality is good and calibration is acceptable, both in the derivation and validation cohorts, and as a continuous model and a point-scoring system. This risk classification system could be applied at the initial visit of a patient with Scl-ILD in an ILD clinic to help discuss prognosis and to guide management decisions.
Unlike risk prediction for IPF, FVC (% predicted) was not included in the top-scoring candidate models, all of which included Dlco (% predicted) instead. These results are not surprising as Dlco has been shown to be a significant predictor of mortality in several prior studies of Scl-ILD.9, 11, 13 For example, a recent substudy25 from the Scleroderma Lung Study 126 demonstrated that Dlco (% predicted) was the best predictor of both extent of lung fibrosis and total interstitial lung disease in patients with Scl-ILD. Baseline Dlco (% predicted) was also highly predictive of mortality and as predictive of HRCT disease extent in another model of mortality in Scl-ILD.11 In our model, replacing Dlco (% predicted) with FVC (% predicted) in the model dropped the C-statistic from 0.88 to 0.79. The Dlco, in addition to capturing the extent of ILD in scleroderma, is likely capturing the impact of pulmonary vascular disease in this patient population. Of note, echocardiographic signs suggestive of pulmonary hypertension were not included in the final model. This may be due to the poor sensitivity and specificity of echocardiographic screening for pulmonary hypertension when compared with right heart catheterization (RHC) in patients with ILD.27, 28 Additional studies should investigate the role of RHC-diagnosed pulmonary hypertension in this model.
The inclusion of smoking history in the model was somewhat unexpected. Smoking has been previously shown to be associated with worse vascular and gastrointestinal outcomes in patients with scleroderma,29 and the prevalence of smoking history among the patients with scleroderma is similar to what has been previously reported in the literature.29, 30 In patients with other forms of ILD, specifically IPF, smoking history has also been associated with worse survival time.31 One potential explanation for this finding could be the coexistence of emphysema (ie, combined pulmonary fibrosis and emphysema).32 Interestingly, the presence of emphysema in this cohort was low (6.7%), suggesting additional mechanisms for worse outcomes among patients with Scl-ILD who have a history of smoking. We did not find an association between the number of pack-years and mortality in these two cohorts, but it is worth further investigation, as this is the only modifiable factor in the model.
Mortality risk prediction models have previously been proposed for patients with Scl-ILD. A systematic review highlighted the current limitations of the prior studies, including methodologic quality and lack of validation.17 As previously mentioned, one Scl-ILD-tailored staging system (Goh et al11) uses FVC (% predicted) and the extent of disease on HRCT to categorize patients as having either limited or extensive disease. This staging system provided prognostic separation between groups in two different cohorts of patients with Scl-ILD,11, 33 but is limited by the need to interpret disease extent on HRCT imaging. The application of this staging system in our derivation cohort demonstrated significantly lower discrimination compared with the SADL model, although its calibration was satisfactory.
IPF risk prediction models have been applied to a cohort of patients with Scl-ILD.34 The modified du Bois index35 had the best combination of discrimination and calibration in predicting mortality at 1 year. This index is calculated on the basis of age, history of respiratory hospitalization in the last 24 weeks, FVC, change in FVC in 24 weeks, distance walked in a 6-min walk test, and change in distance walked in a 6-min walk test. Although the modified du Bois index appeared to predict mortality of patients with Scl-ILD at 1 year, it requires longitudinal assessments, has not been validated, and has not been shown to predict mortality beyond 1 year.
Well-thought-out mortality risk prediction models can be useful tools in clinical practice as well as in the research setting. Identification of those at differential risk for disease progression can help guide treatment and management decisions, as well as inform the counseling and education of patients and their caregivers. From a research standpoint, risk prediction models can be useful for selecting high-risk clinical trial populations. By using advanced statistical methodology to develop and validate a risk prediction model, with an unbiased approach to candidate variable selection, we have identified a group of three variables that can predict mortality in Scl-ILD. The SADL model uses simple variables readily available to clinicians caring for patients with Scl-ILD, thereby facilitating its use. This same statistical method has proven useful in the field of IPF.18, 36, 37
There are limitations to this study. First, both cohorts of patients originated from academic centers using a retrospective design. Patients with Scl-ILD seen in both of these clinics may differ from the general population of people with Scl-ILD and could limit the generalizability of this model. Second, although data were obtained primarily from ongoing prospective longitudinal ILD cohorts, missing data could bias this analysis. The SADL model should not be applied to patients who are unable to perform the Dlco maneuver, as it has not been tested in this population. Third, the majority of the enrolled patients were white. The impact of race/ethnicity on mortality in the Scl-ILD population needs further investigation. Furthermore, all the patients in our cohort had positive ANA results. A small proportion of patients with scleroderma have negative ANA results (5%-10%38, 39, 40), and the SADL model would need to be tested in this subset of patients before it could be applied to this population. Finally, the SADL model was derived to predict all-cause mortality in patients with Scl-ILD; it is not specific for ILD-related mortality. Since the patients in our cohort did not systematically undergo RHC, our estimation of PH is prone to bias and the observed mortality could be attributed to any cause.
In conclusion, the SADL model successfully predicts mortality using ever smoking history, age, and Dlco (% predicted). Future studies should evaluate the impact of incorporating biomarkers as well as longitudinal measures on risk prediction in Scl-ILD.
Acknowledgments
Author contributions: J. M. takes responsibility for the content of the manuscript, including the data and analysis. Involvement in conception, hypothesis, and design of the study (J. M., E. V., P. J. W., T. E. K., H. R. C., J. S. L.); acquisition of the data (J. M., X. H., B. M. E., S. L., J. H. R., K. D. J., J. A. G., A. H., J. S. L.); analysis and interpretation of the data (J. M., E. V., B. M. E., F. B., B. L., P. J. W., T. E. K., H. R. C., J. S. L.); substantial involvement in the writing and/or revision of the article (J. M., E. V., X. H., J. A. G., F. B., J. H. R., B. L., P. J. W., T. E. K., H. R. C., J. S. L.).
Financial/nonfinancial disclosures: J. M. reports personal fees from Boehringer Ingelheim and Hoffmann-La Roche, outside the submitted work. B. L. reports personal fees from Genentech, outside the submitted work. H. R. C. reports personal fees from Medimmune, Bayer, Biogen, Boehringer Ingelheim, Xfibra, Genoa, Gilead, GlaxoSmithKline, Mesoblast, Moerae Matrix, Pharmakea, Promedior, Prometic, Pulmatrix, the Pulmonary Fibrosis Foundation, Unity, Aeolus, aTry pharmaceuticals, Pfizer, UCB Celltech, GBT, Veracyte, Patara, Samumed, Alkermes, Five Prime and Takeda, outside of the submitted work. J. S. L. reports grants from the National Institutes of Health. None declared (E. V., B. M. E., X. H., S. L., J. H. R., K. D. J., A. H., J. A. G., F. B., P. J. W., T. E. K.).
Role of sponsors: The sponsors had no role in the design of the study, the collection and analysis of the data, or the preparation of the manuscript.
Other contributions: The authors thank the physician members and staff of the UCSF Interstitial Lung Disease Program and the Mayo Clinic for their assistance in recruiting patients for this study, Jane Berkeley, BA, for administering the UCSF clinical database, and the many patients who generously agreed to participate in our longitudinal cohort study.
Additional information: The e-Figure and e-Tables can be found in the Supplemental Materials section of the online article.
Footnotes
FUNDING/SUPPORT: This work was supported by the National Center for Advancing Translational Science, National Institutes of Health [grant UCSF-CTI KL2TR000143], the Scleroderma Research Foundation, and the Nina Ireland Program for Lung Health.
Supplementary Data
References
- 1.Steen V.D., Medsger T.A. Changes in causes of death in systemic sclerosis, 1972-2002. Ann Rheum Dis. 2007;66(7):940–944. doi: 10.1136/ard.2006.066068. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Solomon J.J., Olson A.L., Fischer A., Bull T., Brown K.K., Raghu G. Scleroderma lung disease. Eur Respir Rev. 2013;22(127):6–19. doi: 10.1183/09059180.00005512. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.White B. Interstitial lung disease in scleroderma. Rheum Dis Clin North Am. 2003;29(2):371–390. doi: 10.1016/s0889-857x(03)00025-5. [DOI] [PubMed] [Google Scholar]
- 4.Khanna D., Tseng C.H., Farmani N. Clinical course of lung physiology in patients with scleroderma and interstitial lung disease: analysis of the Scleroderma Lung Study Placebo Group. Arthritis Rheum. 2011;63(10):3078–3085. doi: 10.1002/art.30467. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Steen V.D., Conte C., Owens G.R., Medsger T.A., Jr. Severe restrictive lung disease in systemic sclerosis. Arthritis Rheum. 1994;37(9):1283–1289. doi: 10.1002/art.1780370903. [DOI] [PubMed] [Google Scholar]
- 6.Plastiras S.C., Karadimitrakis S.P., Ziakas P.D., Vlachoyiannopoulos P.G., Moutsopoulos H.M., Tzelepis G.E. Scleroderma lung: initial forced vital capacity as predictor of pulmonary function decline. Arthritis Rheum. 2006;55(4):598–602. doi: 10.1002/art.22099. [DOI] [PubMed] [Google Scholar]
- 7.Man A., Davidyock T., Ferguson L.T., Ieong M., Zhang Y., Simms R.W. Changes in forced vital capacity over time in systemic sclerosis: application of group-based trajectory modelling. Rheumatology. 2015;54(8):1464–1471. doi: 10.1093/rheumatology/kev016. [DOI] [PubMed] [Google Scholar]
- 8.Al-Dhaher F.F., Pope J.E., Ouimet J.M. Determinants of morbidity and mortality of systemic sclerosis in Canada. Semin Arthritis Rheum. 2010;39(4):269–277. doi: 10.1016/j.semarthrit.2008.06.002. [DOI] [PubMed] [Google Scholar]
- 9.Bouros D., Wells A.U., Nicholson A.G. Histopathologic subsets of fibrosing alveolitis in patients with systemic sclerosis and their relationship to outcome. Am J Respir Crit Care Med. 2002;165(12):1581–1586. doi: 10.1164/rccm.2106012. [DOI] [PubMed] [Google Scholar]
- 10.Fischer A., Swigris J.J., Groshong S.D. Clinically significant interstitial lung disease in limited scleroderma: histopathology, clinical features, and survival. Chest. 2008;134(3):601–605. doi: 10.1378/chest.08-0053. [DOI] [PubMed] [Google Scholar]
- 11.Goh N.S., Desai S.R., Veeraraghavan S. Interstitial lung disease in systemic sclerosis: a simple staging system. Am J Respir Crit Care Med. 2008;177(11):1248–1254. doi: 10.1164/rccm.200706-877OC. [DOI] [PubMed] [Google Scholar]
- 12.Launay D., Humbert M., Berezne A. Clinical characteristics and survival in systemic sclerosis-related pulmonary hypertension associated with interstitial lung disease. Chest. 2011;140(4):1016–1024. doi: 10.1378/chest.10-2473. [DOI] [PubMed] [Google Scholar]
- 13.Goh N.S., Veeraraghavan S., Desai S.R. Bronchoalveolar lavage cellular profiles in patients with systemic sclerosis-associated interstitial lung disease are not predictive of disease progression. Arthritis Rheum. 2007;56(6):2005–2012. doi: 10.1002/art.22696. [DOI] [PubMed] [Google Scholar]
- 14.Swigris J.J., Zhou X., Wamboldt F.S. Exercise peripheral oxygen saturation (SpO2) accurately reflects arterial oxygen saturation (SaO2) and predicts mortality in systemic sclerosis. Thorax. 2009;64(7):626–630. doi: 10.1136/thx.2008.111393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Trad S., Amoura Z., Beigelman C. Pulmonary arterial hypertension is a major mortality factor in diffuse systemic sclerosis, independent of interstitial lung disease. Arthritis Rheum. 2006;54(1):184–191. doi: 10.1002/art.21538. [DOI] [PubMed] [Google Scholar]
- 16.Fertig N., Domsic R.T., Rodriguez-Reyna T. Anti-U11/U12 RNP antibodies in systemic sclerosis: a new serologic marker associated with pulmonary fibrosis. Arthritis Rheum. 2009;61(7):958–965. doi: 10.1002/art.24586. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Winstone T.A., Assayag D., Wilcox P.G. Predictors of mortality and progression in scleroderma-associated interstitial lung disease: a systematic review. Chest. 2014;146(2):422–436. doi: 10.1378/chest.13-2626. [DOI] [PubMed] [Google Scholar]
- 18.Ley B., Ryerson C.J., Vittinghoff E. A multidimensional index and staging system for idiopathic pulmonary fibrosis. Ann Internal Med. 2012;156(10):684–691. doi: 10.7326/0003-4819-156-10-201205150-00004. [DOI] [PubMed] [Google Scholar]
- 19.Subcommittee for Scleroderma Criteria of the American Rheumatism Association Diagnostic and Therapeutic Criteria Committee Preliminary criteria for the classification of systemic sclerosis (scleroderma) Arthritis Rheum. 1980;23(5):581–590. doi: 10.1002/art.1780230510. [DOI] [PubMed] [Google Scholar]
- 20.van den Hoogen F., Khanna D., Fransen J. 2013 classification criteria for systemic sclerosis: an American College of Rheumatology/European League against Rheumatism collaborative initiative. Arthritis Rheum. 2013;65(11):2737–2747. doi: 10.1002/art.38098. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Galiè N., Hoeper M.M., Humbert M. Guidelines for the diagnosis and treatment of pulmonary hypertension: the Task Force for the Diagnosis and Treatment of Pulmonary Hypertension of the European Society of Cardiology (ESC) and the European Respiratory Society (ERS), endorsed by the International Society of Heart and Lung Transplantation (ISHLT) Eur Heart J. 2009;30(20):2493–2537. doi: 10.1093/eurheartj/ehp297. [DOI] [PubMed] [Google Scholar]
- 22.Raghu G., Collard H.R., Egan J.J. An official ATS/ERS/JRS/ALAT statement: idiopathic pulmonary fibrosis: evidence-based guidelines for diagnosis and management. Am J Respir Crit Care Med. 2011;183(6):788–824. doi: 10.1164/rccm.2009-040GL. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Desai S.R., Veeraraghavan S., Hansell D.M. CT features of lung disease in patients with systemic sclerosis: comparison with idiopathic pulmonary fibrosis and nonspecific interstitial pneumonia. Radiology. 2004;232(2):560–567. doi: 10.1148/radiol.2322031223. [DOI] [PubMed] [Google Scholar]
- 24.Harrell F.E., Jr., Lee K.L., Mark D.B. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15(4):361–387. doi: 10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4. [DOI] [PubMed] [Google Scholar]
- 25.Tashkin D.P., Volkmann E.R., Tseng C.H. Relationship between quantitative radiographic assessments of interstitial lung disease and physiological and clinical features of systemic sclerosis. Ann Rheum Dis. 2016;75(2):374–381. doi: 10.1136/annrheumdis-2014-206076. [DOI] [PubMed] [Google Scholar]
- 26.Tashkin D.P., Elashoff R., Clements P.J. Cyclophosphamide versus placebo in scleroderma lung disease. N Engl J Med. 2006;354(25):2655–2666. doi: 10.1056/NEJMoa055120. [DOI] [PubMed] [Google Scholar]
- 27.Thakkar V., Stevens W., Prior D. The inclusion of N-terminal pro-brain natriuretic peptide in a sensitive screening strategy for systemic sclerosis-related pulmonary arterial hypertension: a cohort study. Arthritis Res Ther. 2013;15(6):R193. doi: 10.1186/ar4383. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mukerjee D., St George D., Knight C. Echocardiography and pulmonary function as screening tests for pulmonary arterial hypertension in systemic sclerosis. Rheumatology. 2004;43(4):461–466. doi: 10.1093/rheumatology/keh067. [DOI] [PubMed] [Google Scholar]
- 29.Hudson M., Lo E., Lu Y., Hercz D., Baron M., Steele R. Cigarette smoking in patients with systemic sclerosis. Arthritis Rheum. 2011;63(1):230–238. doi: 10.1002/art.30071. [DOI] [PubMed] [Google Scholar]
- 30.Chaudhary P., Chen X., Assassi S. Cigarette smoking is not a risk factor for systemic sclerosis. Arthritis Rheum. 2011;63(10):3098–3102. doi: 10.1002/art.30492. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Antoniou K.M., Hansell D.M., Rubens M.B. Idiopathic pulmonary fibrosis: outcome in relation to smoking status. Am J Respir Crit Care Med. 2008;177(2):190–194. doi: 10.1164/rccm.200612-1759OC. [DOI] [PubMed] [Google Scholar]
- 32.Cottin V., Nunes H., Mouthon L. Combined pulmonary fibrosis and emphysema syndrome in connective tissue disease. Arthritis Rheum. 2011;63(1):295–304. doi: 10.1002/art.30077. [DOI] [PubMed] [Google Scholar]
- 33.Moore O.A., Goh N., Corte T. Extent of disease on high-resolution computed tomography lung is a predictor of decline and mortality in systemic sclerosis-related interstitial lung disease. Rheumatology. 2013;52(1):155–160. doi: 10.1093/rheumatology/kes289. [DOI] [PubMed] [Google Scholar]
- 34.Ryerson C.J., O’Connor D., Dunne J.V. Predicting mortality in systemic sclerosis-associated interstitial lung disease using risk prediction models derived from idiopathic pulmonary fibrosis. Chest. 2015;148(5):1268–1275. doi: 10.1378/chest.15-0003. [DOI] [PubMed] [Google Scholar]
- 35.du Bois R.M., Albera C., Bradford W.Z. 6-Minute walk distance is an independent predictor of mortality in patients with idiopathic pulmonary fibrosis. Eur Respir J. 2014;43(5):1421–1429. doi: 10.1183/09031936.00131813. [DOI] [PubMed] [Google Scholar]
- 36.Kim E.S., Choi S.M., Lee J. Validation of the GAP score in Korean patients with idiopathic pulmonary fibrosis. Chest. 2015;147(2):430–437. doi: 10.1378/chest.14-0453. [DOI] [PubMed] [Google Scholar]
- 37.Kishaba T., Shimaoka Y., Fukuyama H. Clinical characteristics of idiopathic pulmonary fibrosis patients with gender, age, and physiology staging at Okinawa Chubu Hospital. J Thorac Dis. 2015;7(5):843–849. doi: 10.3978/j.issn.2072-1439.2015.04.54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Salazar G.A., Assassi S., Wigley F. Antinuclear antibody-negative systemic sclerosis. Semin Arthritis Rheum. 2015;44(6):680–686. doi: 10.1016/j.semarthrit.2014.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Ho K.T., Reveille J.D. The clinical relevance of autoantibodies in scleroderma. Arthritis Res Ther. 2003;5(2):80–93. doi: 10.1186/ar628. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Hamaguchi Y. Autoantibody profiles in systemic sclerosis: predictive value for clinical evaluation and prognosis. J Dermatol. 2010;37(1):42–53. doi: 10.1111/j.1346-8138.2009.00762.x. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.