Patient selection

Details of patient selection and recruitment have been previously published by Reveille and colleagues [25]. Patients who fulfilled the following inclusion criteria are enrolled: age ≥18 years; diagnosis according to the American College of Rheumatology (formerly the American Rheumatism Association) preliminary classification criteria for SSc [26]; disease onset within the previous 5 years (defined as the first non-Raynaud's phenomenon symptom); and defined ethnicity with all four grandparents from the same ethnic group. Patients who had SSc-like illnesses associated with environmental, ingested, or injected agents were excluded from the study.

All enrolled patients at the time of analysis were included in the current study. Patients were assigned to the following four ethnic groups: white, African American, Hispanic and other. A total of 10 patients were assigned to the other group consisting of one Native American and nine Asian patients. The study was approved by the institutional review board of all participating sites, and written informed consent was obtained from all study subjects.

Data collection

A standardized clinical manifestation form is completed at baseline and at follow-up visits [25]. This assessment includes calculation of the modified Rodnan Skin Score [24] and a previously validated severity scale, the Medsger Severity Index [27]. Electrocardiograms, chest radiographs, and laboratory tests including complete blood count, metabolic panel, creatine phosphokinase and urinalysis are also obtained at baseline and annually thereafter. Pulmonary fibrosis on chest radiograph is defined as increased interstitial markings not explained otherwise. Information on annual income and educational levels is also collected and categorized as described in Supplement 1 in Additional file 1. Furthermore, the following autoantibodies are determined in the laboratories of the University of Texas Health Science Center at Houston: anti-nuclear antibodies and ACA by indirect immunofluorescence utilizing HEp-2 cell substrates (Antibodies Incorporated, Davis, CA, USA); ATA, anti-Ro, anti-La, anti-Smith antibody, and anti-U1-ribonucleoprotein by passive immunodiffusion against calf thymus extract (INOVA Diagnostics, San Diego, CA, USA); and anti-RNA polymerase III antibodies by ELISA (MBL, Nagoya, Japan). Anti-fibrillarin antibodies (U3-ribonucleoprotein) are determined on patients with anti-nucleolar antibodies by immunoprecipitation of the recombinant protein at the University of Calgary, as previously described [28].

The disease duration was calculated from the time of first non-Raynaud's phenomenon symptom in our study. We also utilized a second method for calculation of the disease duration, however, in which the disease onset was defined as the time of first symptom attributable to SSc (first Raynaud's phenomenon symptom or non-Raynaud's phenomenon symptom).

Chest CT, echocardiogram and right heart catheterization are not included as a routine part of the GENISOS visit, but the results of these studies are captured if obtained as part of the usual clinical care. Only a small group of subjects had undergone a chest CT or right heart catheterization at enrollment, so the results of these studies were not included in our analysis. A larger portion of patients (23.68%), however, had undergone an echocardiogram at enrollment; the predictive significance of elevated systolic pulmonary arterial pressure, defined as a right ventricular systolic pressure >40 mmHg, was therefore investigated in this subgroup of patients.

We also determined the vital status of each patient as of November 2009 by review of the National Death Index and the Social Security Death Database.

Pulmonary function test

PFTs are performed at the three sites on the baseline visit and annually thereafter. Variables are expressed as the percentage predicted FVC, forced expiratory volume in 1 second, total lung capacity, and diffusing capacity (liters) of carbon monoxide (DLco). The predicted values are calculated according to the patient's age, height, weight, gender and ethnicity utilizing consistent reference values. The DLco percentage predicted values are also corrected for the patient's actual hemoglobin level [29]. Furthermore, all PFTs are reviewed by a pulmonologist (RME-Y-M) and studies are excluded that do not fulfill the American Thoracic Society/European Respiratory Society criteria for pulmonary function testing [30–32].

Statistical analysis

The generalized linear mixed models were used to evaluate the effects of the measured baseline demographic and clinical variables on the percentage predicted FVC. First, the association of the baseline variables with serial measurements of percentage predicted FVC was examined. Subsequently, predictor-follow-up time interactions were included in the models to evaluate the effects of the baseline and clinical variables on the change in percentage predicted FVC over time. Those interactions evaluated whether a predictor was an effect modifier (that is, whether the change in percentage predicted FVC over time depended on the baseline variable). In other words, a baseline variable was considered a predictor of decline rate in FVC if the interaction term between the variable and the follow-up time was significant with regard to serial measurements of percentage predicted FVC.

The identification of predictors was conducted for two different follow-up periods. First, we limited our analysis to FVC measurements obtained within 3 years after enrollment. The cut-off point of 3 years was selected because this approximated to the average disease duration of 5 years in our cohort. Next, we analyzed all FVC measurements throughout the entire follow-up time. This approach was chosen was chosen to diminish the effects of survival bias because patients with rapidly progressive pulmonary disease had higher mortality (see results) and were less likely to be present in the long-term follow-up group (up to 11.4 years).

We treated patients as a sample from a larger population, and modeled between-patient variability in percentage predicted FVC as a random intercept. We also modeled between-patient variability in the change in percentage predicted FVC over time by a random slope (that is, we estimated a separate slope for each patient). We accounted for the correlations among random-effect parameters by an independent covariance matrix. Exchangeable or unstructured covariance matrices did not improve model fit evaluated by the Aikaike Information Criteria. The relationship between follow-up time and FVC was appropriately modeled by a linear function (a second-order polynomial did not improve the model fit). Generally, mixed-effect models allow inclusion of all data points in the analysis and can be used when some data points are missing.

Box-and-whisker plots were used to illustrate the trend in progression of percentage predicted FVC in the overall population and subgroups of patients based on disease type [33] and on disease duration at enrollment. P < 0.05 was considered significant. Analyses were conducted with Stata 10.1 statistical software (StataCorp LP, College Station, TX, USA).

Sample characteristics

Between January 1998 and October 2009, 266 patients were enrolled in the GENISOS cohort. The mean (± standard deviation) follow-up time was 3.76 (± 3.35) years, ranging up to 11.39 years. A total of 926 FVC measurements belonging to 244 patients fulfilled the American Thoracic Society/European Respiratory Society criteria and were included in the analysis. Overall, 182 patients had at least two FVC measurements that fulfilled these criteria.

The mean (± standard deviation) disease duration at enrollment was 2.53 (± 1.63) years; 41.4% of patients had limited cutaneous disease. ACA, ATA and anti-RNA polymerase III antibodies were present in 12.1%, 18.4% and 22.9% of patients, respectively. The proportion of patients with a baseline FVC <80% was 41.9%. Furthermore, only 22 patients (8.3%) had received cyclophosphamide before enrollment or during the follow-up period. Clinical characteristics of the GENISOS cohort at enrollment are presented in Table 1 and in Supplement 2 in Additional file 1. Moreover, a detailed comparison of baseline demographic and clinical characteristics between the investigated ethnic groups has been published previously [18].

Factors associated with percentage predicted FVC

As shown in Table 2, the presence of anti-RNA polymerase III, anti-fibrillarin and anti-U1-ribonucleoprotein antibodies, the body mass index and other examined demographic and clinical variables were not significantly associated with percentage predicted FVC.

In addition, the smoking status did not show a significant relationship to FVC. Furthermore, none of the above-mentioned associations changed after correction for current smoking status. Similarly, the above-mentioned associations did not change after exclusion of patients ever treated with cyclophosphamide (data not shown).

Role of ethnicity in progression of restrictive lung disease in systemic sclerosis

We have previously reported that African American patients with SSc had a higher frequency of pulmonary fibrosis and a lower percentage predicted FVC at the baseline visit in the GENISOS cohort [18]. As demonstrated in Table 2, this group of patients also had lower percentage predicted FVC after inclusion of all FVC measurements, whereas this outcome variable among Hispanic patients did not differ from that for white patients. In a multivariable model, percentage predicted FVC among the African American patients remained significantly lower than white patients even after adjustment for disease duration, ATA and ACA positivity, disease type, and smoking status, in addition to income and education levels (P = 0.038; regression coefficient (RC) = -7.73, confidence interval (CI) = -15.02 to -0.45), whereas Hispanic patients did not differ from white patients (P = 0.733; RC = 1.21, CI = -5.33 to 7.58). These results indicate that African American patients had lower percentage predicted FVC independent of smoking status, socioeconomic factors, disease duration/type (limited versus diffuse cutaneous disease) and serology.

Predictors of rate of decline in percentage predicted FVC

As shown in Figure 1, our data did not indicate that the overall decline in percentage predicted FVC slowed down during the follow-up time. This was also demonstrated by the fact that the curvilinear model for progression of percentage predicted FVC did not show a better fit to the data than the linear model. The frequency of patients with ≥10% decline in percentage predicted FVC increased at a slower pace, however, from 16.7% in the first year of follow-up to 45.9% after ≥8 years of follow-up (Table 4).

We next investigated which baseline variables are predictors of accelerated decline in FVC. For this purpose, the interaction term between the respective variable and follow-up time was examined. As shown in Tables 2 and 3, presence of ATA was the only variable associated with sequentially obtained FVC levels that also predicted the rate of decline within the first 3 years of follow-up. Presence of ATA was associated with a higher decline rate in FVC in this time period (P = 0.022, RC = -2.49, CI = -4.62 to -0.36). Presence of ATA was not a significant predictor of FVC decline rate, however, when the follow-up time was not restricted (P = 0.121). The latter analysis included FVC measurements that were obtained up to 11.4 years after enrollment.

The interaction term for elevated systolic pulmonary arterial pressure on the echocardiogram was also significant when the analysis was restricted to 3-year follow-up. However, this variable was not associated with sequentially obtained FVC levels (insignificant main effect, P = 0.997). Furthermore, an echocardiogram was obtained only in 23.7% of patients at enrollment.

Other baseline variables associated with differential FVC levels did not predict progression of this variable over time in 3-year or long-term follow-up analyses. Specifically, ethnicity, autoantibody status, subjective dyspnea, baseline PFT values, Medsger Severity Index subscores and smoking status did not predict the rate of progression in FVC.

Figure 2a,b shows graphically that the progression of the outcome variable was not significantly different among patients with limited versus diffuse disease types; this was further supported by the fact that the interaction term between disease type and follow-up time was not significant (P = 0.442 for 3-year follow-up, P = 0.932 for long-term follow-up). Similarly, the interaction term for disease duration at enrollment and follow-up time also was not significant (P = 0.767 for 3-year follow-up, P = 0.94 for long-term follow-up), indicating that shorter disease duration did not predicted a significantly faster rate of decline in FVC. To demonstrate this observation graphically, SSc patients were divided into two groups based on disease duration of 2 years. One hundred and nineteen (44.7%) patients had disease duration <2 years at the initial visit. As indicated in Figure 2c,d, the slope of decline in the percentage predicted FVC was not significantly different between these two groups.

Similarly, the disease duration measured from the time of first symptom attributable to SSc (see Table 2, Disease duration (alternative method)) did not predict the rate of decline in the percentage predicted FVC (P = 0.582 for 3-year follow-up, P = 0.753 for long-term follow-up.

Rate of FVC decline among deceased patients

Seventy-three patients (27.4%) died during follow-up. As expected, the deceased patients had lower serial FVC values (P = 0.001, RC = -10.08, CI = -16.26 to -3.89). Furthermore, the examination of the interaction term between follow-up time and vital status revealed that the deceased patients had a significantly faster rate of decline in percentage predicted FVC compared with the remainder of the patients (P = 0.001, RC = -1.66, CI = -2.63 to -0.69) for long-term follow-up).