For the purpose of the present study, data from the 3 RCTs were pooled and analyzed irrespective of treatment assignments because none of the active therapies differed from placebo group in the primary and secondary outcomes. In the pooled dataset, 17% of observations had missing MRSS during course of the study. After the data were pooled, patients were divided into 5 groups according to their disease duration at baseline: group 1 (< 6 months), group 2 (6-<12 months), group 3 (12-<24 months), group 4 (24-<48 months) and group 5 (≥ 48 months). Disease duration was calculated from the date of onset of first symptom characteristic of dcSSc (other than Raynaud’s phenomenon). Box-and-whisker plots were used to illustrate the trend in MRSS decline in different disease duration categories by plotting the median MRSS at different scheduled visits. MRSS was obtained during 4 follow-up visits in the relaxin and collagen studies, while the D-Pen study included a fifth visit. The visits were conducted at the following intervals during the course of each study: D-Pen: baseline, 6, 12, 18 and 24 months; in the relaxin study, the visits were conducted at baseline, 1, 3 and 6 months; and in the collagen study the visits were conducted at baseline, 4, 8 and 12 months.
MRSS was modeled as a continuous variable with baseline disease duration and time in study (expressed in months after baseline) as covariates. Interaction terms between disease duration category and time in study were introduced in order to allow the rate of MRSS change to vary between disease duration groups. To account for the within-subject correlation of skin score over multiple visits, a mixed effects model with random intercept for subject was used. Mixed effects models allow all data points to be included in the analysis and are appropriate to use when data are missing at random and even remain relatively robust to data that are not missing at random.
We also analyzed the absolute change in MRSS in all 3 RCTs using one-way ANOVAs. For the D-Pen study, we assessed the MRSS change from baseline to 12 months and from 12 to 24 months; the change in MRSS was assessed from baseline to 6 months and from baseline to 12 months in the relaxin and collagen studies, respectively. Proportions of patients with an overall MRSS improvement and worsening were calculated in each study and in the pooled data. Based on a previous analysis of the D-Pen study(21 (link)), improvement and worsening in the MRSS was defined as a change of ≥ 5.3 points (minimum clinically important differences or MCID) in MRSS and analyzed in patients who completed the 3 RCTs .
We conducted 3 additional analyses: 1) disease duration was reclassified by including Raynaud’s phenomenon as part of disease duration in the D-Pen and relaxin studies (the collagen study did not capture the onset of Raynaud’s phenomenon), 2) course of MRSS was assessed in the placebo groups in the relaxin and collagen studies; D-Pen study compared the effects of low dose vs. high dose D-Pen and therefore was not included in the placebo group analysis, and 3) Since MRSS data was missing in 17% of observations during course of the trials in all 3 studies, we performed multiple imputation on the data. Multiple imputation uses a regression-type approach to estimate each missing datum. Imputed values are generated taking into account responses from the same participant on other correlated variables and responses to the same domain from participants who responded similarly. The three multiply-imputed datasets were then pooled and the pooled multiply-imputed dataset was analyzed using the mixed-effects regression model presented earlier.
All computations were achieved using the statistical SAS System Release 8.2 (SAS Institute Inc., Cary, NC, USA) and STATA 9.2 (College Station, TX, USA) software package.
MRSS was modeled as a continuous variable with baseline disease duration and time in study (expressed in months after baseline) as covariates. Interaction terms between disease duration category and time in study were introduced in order to allow the rate of MRSS change to vary between disease duration groups. To account for the within-subject correlation of skin score over multiple visits, a mixed effects model with random intercept for subject was used. Mixed effects models allow all data points to be included in the analysis and are appropriate to use when data are missing at random and even remain relatively robust to data that are not missing at random.
We also analyzed the absolute change in MRSS in all 3 RCTs using one-way ANOVAs. For the D-Pen study, we assessed the MRSS change from baseline to 12 months and from 12 to 24 months; the change in MRSS was assessed from baseline to 6 months and from baseline to 12 months in the relaxin and collagen studies, respectively. Proportions of patients with an overall MRSS improvement and worsening were calculated in each study and in the pooled data. Based on a previous analysis of the D-Pen study(21 (link)), improvement and worsening in the MRSS was defined as a change of ≥ 5.3 points (minimum clinically important differences or MCID) in MRSS and analyzed in patients who completed the 3 RCTs .
We conducted 3 additional analyses: 1) disease duration was reclassified by including Raynaud’s phenomenon as part of disease duration in the D-Pen and relaxin studies (the collagen study did not capture the onset of Raynaud’s phenomenon), 2) course of MRSS was assessed in the placebo groups in the relaxin and collagen studies; D-Pen study compared the effects of low dose vs. high dose D-Pen and therefore was not included in the placebo group analysis, and 3) Since MRSS data was missing in 17% of observations during course of the trials in all 3 studies, we performed multiple imputation on the data. Multiple imputation uses a regression-type approach to estimate each missing datum. Imputed values are generated taking into account responses from the same participant on other correlated variables and responses to the same domain from participants who responded similarly. The three multiply-imputed datasets were then pooled and the pooled multiply-imputed dataset was analyzed using the mixed-effects regression model presented earlier.
All computations were achieved using the statistical SAS System Release 8.2 (SAS Institute Inc., Cary, NC, USA) and STATA 9.2 (College Station, TX, USA) software package.
