Abstract

Introduction

The Scleroderma Lung Study (SLS) [1] was a double-blind, randomized, placebo (PLA)-controlled trial of oral cyclophosphamide (CYC), administered for 1 year, on the course of forced vital capacity percentage (FVC%) predicted in patients with evidence of active SSc-related interstitial lung disease (SSc-ILD). The study showed that CYC produced an improvement in the FVC% predicted and dyspnoea [as measured by the Transition Dyspnoea Index (TDI)] [2, 3].

The Mahler Baseline Dyspnoea Index (BDI) and TDI have recently been shown to be feasible, reliable and valid in patients with SSc-ILD [3, 4]. In any given individual, the TDI score represents either an improvement (positive score) or a worsening (negative) score compared with the BDI in the same individual. As future studies in SSc-ILD are likely to include a dyspnoea index, our current objective was to assess the minimally important difference (MID) or minimal clinically important difference (MCID)—the smallest improvement in score that patients perceive as beneficial and that may lead to a change in the patient's management of dyspnoea [5].

Patients and methods

Analysis

MID estimate was assessed using an anchor-based approach. We utilized five patient-reported anchors from the SF-36 instrument to assess MID for TDI–SF-36 transition question and Questions 3D, E, H and I pertaining to walking on a flat surface or climbing stairs. These questions were judged a priori to be pertinent to assess dyspnoea related to SSc-ILD. Because previous studies have shown an inherent uncertainty around the MID estimates [7–9], we included several anchors. On the SF-36 transition question (compared with 1 year ago, how would you rate your health in general now), patients who rated as little better or little worse (12-month baseline visit) were defined as the minimally changed subgroup. For Question 3 (Tables 1 and ​and2),2), patients who reported improvement from ‘Limited a lot’ to ‘Limited a little’ and ‘Limited a little’ to ‘No limit’ and vice versa (12 month vs baseline visit) were defined as the minimally important changed subgroup for improvement and worsening, respectively. The changes in the mean TDI scores (time_{12 month visit} − time_{baseline visit}) for the group that reported a little better or a little worse and for groups that changed one step (from ‘Limited a lot’ to ‘Limited a little’ and ‘Limited a little’ to ‘No limit’ and vice versa) were determined in order to estimate the MID. By inspecting the quantile–quantile plot and histogram, the distribution of TDI score appears to be unimodal and symmetrical around 0. We report the MID estimates as mean and 95% CIs.

Table 1.

Pearson correlation coefficient between anchors and TDI

	n	Correlation coefficient	P-value
SF-36 health transition	130	0.62	<0.001
Question 3d (climbing several flights of stairs)	129	0.33	<0.001
Question 3e (climbing one flight of stairs)	129	0.24	0.007
Question 3h (walking several hundred yards)	129	0.23	0.009
Question 3i (walking 100 yards)	129	0.24	0.008

Open in a separate window

Table 2.

MID estimates for improvement and worsening of the TDI by different anchors

	n	Mean TDI	95% CI	P-value*
SF-36 health transition item
Much worse	6	−5.67	−10.39, −0.94	0.030
Somewhat worse	29	−2.55	−3.62, −1.49	<0.001
About the same	47	−0.13	−0.58, 0.32
Somewhat better	25	2.16	0.62, 3.70	0.007
Much better	23	3.3	1.79, 4.82	<0.001
Question 3d (climbing several flights of stairs)
Worse 2	1	−4		0.240
Worse 1	23	−1.87	−3.09, −0.65	0.010
Same	82	0.17	−0.60, 0.94
Improve 1	22	1.68	−0.05, 3.41	0.080
Improve 2	1	12		0.001
Question 3e (climbing one flight of stairs)
Worse 2	7	−3	−6.58, 0.58	0.010
Worse 1	21	−1.52	−3.15, 0.10	0.017
Same	74	0.59	−0.22, 1.41
Improve 1	25	1.32	−0.16, 2.80	0.380
Improve 2	2	−4	−29.40, 21.41	0.250
Question 3h (walking several hundred yards)
Worse 2	5	−4.2	−9.27, 0.87	0.007
Worse 1	31	−0.61	−2.03, 0.81	0.130
Same	69	0.59	−0.29, 1.48
Improve 1	21	1.05	−0.28, 2.38	0.610
Improve 2	3	−2.33	−10.32, 1.48	0.180
Question 3i (walking 100 yards)
Worse 2	5	−4.6	−9.46, 0.26	0.002
Worse 1	21	−1.05	−28.4, 0.74	0.090
Same	77	0.4	−0.36, 1.16
Improve 1	23	1.52	−0.16, 3.21	0.180
Improve 2	2	−2.5	−46.97, 41.97	0.230

Open in a separate window

* P-value in comparison to same group.

To assess the usefulness of an anchor, we assessed the association between the anchors and changed the score using Pearson product-moment correlation coefficient. Concordance between the MID estimate vs change in FVC% predicted was assessed based on 3% FVC change as the cut-off. We defined a >3% change in FVC as improvement, −3% to 3% as no change and a decrease of >3% as worsening. Similarly for SF-36 items, we summarized the SF-36 with the median of the change in the five items of SF-36 transition question and Questions 3D, E, H and I; a value of >1 is considered as an improvement, a value between −1 and 1 is no change and a value less than −1 as worsening. Cross-tabulations among FVC% change, SF-36 change and estimated MID were tabulated, total percentage concordance was estimated and Kendall's τ b-statistic and its significance were also calculated (to evaluate the strength of concordance). The data were analysed using STATA 9.2. P-value of <0.05 was deemed to be indicative of statistical significance.

Results

The main findings of the SLS, including patient-reported outcome analysis (including BDI and TDI), have been published elsewhere [1, 3]. Briefly, the mean ± s.d. of the average age of the study population was 48.5 ± 12.3 years and most of the participants were females (71%), had disease duration of 3.1 (2.1) years and moderate dyspnoea on BDI (5.68 ± 1.89). The correlation between the anchors and TDI ranged from 0.23 to 0.62, P < 0.01 (Table 1). The health transition item and climbing several flights of stairs (Q3d: climbing several flights of stairs) had higher correlation coefficients compared with the other three items.

The MID estimates for the TDI improvement and worsening ranged from 1.05 to 2.16 (mean score = 1.5) U and −0.61 to −2.55 (mean score = −1.5) U, respectively (Table 2). This change was larger than the no-change group (−0.13 to 0.59 U; mean score = 0.38 U), providing face validity to the estimates. The numbers of patients who improved or worsened greater than MID group were too small to make definitive conclusions about greater degrees of change.

Clinical association of MID estimates with treatment group and FVC% predicted

We have previously shown that CYC resulted in a statistically significant improvement in TDI (+1.4 CYC vs − 1.3 PLA; P < 0.001) at 12 months [3]. We assessed the proportion of patients who achieved MID for improvement and worsening in the CYC and PLA group in the SLS. A higher proportion of patients on CYC (39%) achieved a clinically important improvement in dyspnoea (defined as TDI 1.5 U) compared with the PLA group (7%). Conversely, a higher proportion of patients on PLA (34%) worsened in their dyspnoea (TDI 1.5 U) compared with the CYC arm (13%; P < 0.001 favouring CYC).

Patients who showed a clinically important improvement (the MID estimate or more) at the end of 12 months had a greater improvement in their FVC% predicted (+3.6%; 95% CI 1.3, 6) compared with those who did not achieve MID (−3.3%; 95% CI −1.5, −5.1; P < 0.001; Fig. 1). Similarly, patients who worsened greater than TDI −1.5 or more at 12 months showed a greater numerical decline in the FVC% predicted compared with no change group (−5.9; 95% CI −3.1, −8.8; P = 0.11).

Open in a separate window

Fig. 1.

Change in FVC% predicted (baseline to 12 months).

Next, we assessed the concordance between the MID estimate vs change in FVC% and SF-36 items. Concordance was 53% for MID estimate vs FVC%, and 59% for SF-36 items, and Kendall's τ b-statistic was estimated to be 0.40 and 0.38 (P < 0.001 for both), respectively.

Discussion

Dyspnoea is the most common symptom in patients with ILD. The baseline and transition dyspnoea indices have recently been shown to be feasible, reliable and valid in the SLS [3, 4]. In addition, the TDI was able to discriminate between CYC and PLA at the end of the 1-year study and complemented the changes seen in the physiological measures [1]. In this manuscript, we show that the MID estimates for improvement and worsening of TDI score are +1.5 U and −1.5 U, respectively.

MDIs were developed and validated in patients with chronic obstructive pulmonary disease (COPD) [2, 10]. Witek and Mahler [10] reported the MID estimate for the TDI improvement score as 1 U in patients with COPD, somewhat different from our estimate. This discrepancy may result from the patient population studied. In addition, Witek and Mahler used physician- rather than patient-reported anchors.

Our MID estimates showed a positive association with change in FVC% predicted. Patients who achieved the MID for improvement at 12 months demonstrated a greater increase in mean FVC% predicted compared with those who did not. Conversely, patients who worsened greater than the MID estimates had a greater decline in mean FVC predicted than those who did not, thereby suggesting clinical relevance and validating our estimates. Although one may question the clinical relevance of these findings, the statistically significant improvements in FVC% predicted in those subjects with a positive MID (+3.6%; 95% CI 1.3, 6) contrasted with reciprocal statistically significant declines in FVC% predicted of similar magnitude in those subjects with a negative MID (−3.3%; 95% CI −1.5, 5.1) for a mean difference between these two groups of subjects of ~7% predicted. This difference in FVC% predicted between the two groups is equivalent to a difference in absolute FVC of ~250 ml. Although the MID for FVC has not been estimated, estimates for the MID for FEV1 in COPD range between 100 and 140 ml [11]. In intervention trials in COPD, improvements in FVC tend to be nearly twice as large as those in FEV1, so that one might reasonably infer an MID for FVC in COPD and, by extension, in ILD, of ~250 ml, which is the difference observed between those who reached a positive vs a negative MID for the TDI in the present study. The clinical relevance of the relatively modest changes in FVC% predicted in the SLS between the two treatment groups (2.5%; 95% CI 0.28, 4.79) is also suggested by the associated treatment-related improvements in some HRQOL measures, as well as in the change scores for dyspnoea (TDI), which approximated three times the MID (for COPD) of 1 for the TDI [1]. In the present analysis, moreover, the MID estimates were also able to discriminate between the effects of CYC vs PLA on dyspnoea scores.

MID estimates can help clinicians understand whether TDI score differences between two treatment groups are meaningful and if changes within one group over time are clinically meaningful in clinical trials or clinical practice [12]. MID estimates are interpreted at a group level and not at an individual level. Therefore, a change of 1.5 points here is considered as MID, although TDI change (improvement and worsening) in an individual patient is a whole unit.

Our study has many strengths. We used prospective data from a large randomized study to assess MID estimates. In addition, we used multiple patient-reported anchors to reach a consensus, as recommended by the experts [13]. Physician report of global assessment has also been used to assess MID estimates, but was not obtained in SLS [8]. Although certain authors have argued that the anchor should be an overall or global change (e.g. SF-36 health transition instrument), there is an inherent uncertainty around the MID estimates [13]. We chose anchors that were relevant to the symptoms of dyspnoea, thus providing face and content validity. Third, we provide MID estimates for both improvement and worsening, since these can be different [14].

Our study has some limitations. First, the analysis was post hoc rather than a priori. Secondly, the dyspnoea indices were originally designed and validated in patients with COPD, although we have previously shown that these dyspnoea indices have construct validity in patients with SSc-ILD [3, 4]. The TDI was even proposed as an outcome measure by scleroderma experts in a recent Delphi panel [15]. Thirdly, majority of the patients did not show a change in their TDI scores during 1-year clinical trial, and the MID estimates are based on a small number of subjects who reported change. However, our MID estimates were larger than the no-change group (mean score = 0.38 U), and the MID estimates were clinically relevant as they were able to discriminate between the FVC% predicted for improvement and worsening. These estimates should be confirmed in future clinical trials and observational studies.

These limitations notwithstanding, we show that a change (improvement/worsening) of 1.5 U in the TDI score is the MID in patients with SSc-ILD. These estimates can aid in interpreting TDI scores in future clinical trials of SSc-ILD, as well as other forms of ILD, and may be relevant for sample size calculations in clinical studies.

Acknowledgement

References