The minimal important difference for the St George's Respiratory Questionnaire in patients with severe COPD

Discussion

To the best of our knowledge, we established for the first time a SGRQ MID specifically for patients with severe COPD. Our findings differ substantially from the SGRQ MID of −4 units that has been described more generally in COPD [3]. Our results showed that the MID for patients with severe COPD was −8.3 units at 1 month follow-up and −7.1 units at 6 months follow-up. The relative MID was −13.4% at 1 month follow-up and −11.1% at 6 months follow-up.

Our calculated SGRQ MID for patients with severe COPD significantly differed compared with the earlier established MID [3]. One explanation could be the chosen time points. In Jones et al. [3], it is not clear for which time point the SGRQ MID was determined. Whilst inconclusive, most of their methods to establish the MID were applied at the 12-month follow-up interval. In contrast, our MID estimations are established at the short-term, after 1 and 6 months follow-up. At the 6-month follow-up interval, the established MID was lower compared with the 1-month follow-up interval. It can be hypothesised that the diminished perceived effect of treatment results in lower decrease of SGRQ scores after 6 months follow-up. In the long-term, a MID could be further decreased due to the progressive nature of COPD. For example, although treatments may produce an initial improvement in SGRQ scores, scores may subsequently return to the baseline level or worse due to disease progression [19, 20]. More research is needed to investigate whether the MID in severe COPD patients at 12 months follow-up differ from our findings at 1 and 6 months follow-up.

Our study population consisted of patients who were treated with BLVR and the question arises whether our newly established MID is only applicable to this treatment or to other treatment modalities for patients with severe COPD as well. There is no consensus on whether a MID is applicable to a specific treatment or a broad range of treatment modalities. Jones et al. [21] state that MIDs were originally not developed in the context of specific treatments. They stated that the reference point (at least with anchor-based methods) is patient- or clinician-perceived benefit [21]. Troosters et al. [22], however, question whether MIDs based on anchor techniques can be used across different interventions as several processes may underlie the observed effects of interventions. For example, the expectations of patients may be different, yielding different effect sizes with different interventions [22]. Therefore, more research is needed regarding the specificity of MIDs for different treatments.

Several methods for estimating MIDs have been described in the literature, the most important being anchor-based and distribution-based methods [14]. The combination of multiple methods is generally recommended to increase reliability of outcome [14]. Therefore, we relied on both anchor-based and distribution-based methods. The anchor-based method requires that a reasonably strong linear relationship exists between the anchor and the variable of interest [23]. However, there is no consensus on how strong this relationship should be exactly. One review recommends statistically significant (p<0.05) correlation coefficients of ≥0.3 as appreciable [14] and two studies in COPD patients performed the analyses when correlation coefficients were 0.3 or 0.5 [11, 24]. In line with our previous study, we chose a correlation coefficient threshold ≥0.4 [12]. Distribution-based methods are commonly considered inferior to anchor-based methods because they rely solely on statistical criteria and depend heavily on the characteristics of a particular study [23]. We agree with the general consensus that the distribution-based method should only be used to support estimates derived from anchor-based methods [2, 14]. Therefore, we have placed less weight on the distribution-based method in calculating our combined MID.

The quality of the anchors can influence the outcomes when performing an anchor-based analysis. A good anchor, in our opinion, is present if it is derived from several studies with comparable patient populations, in our case severe COPD patients. The anchor should be determined using several methods. Finally, there should be an appreciable association between the outcome variable and anchor [23]. Here, we will evaluate the strengths and weaknesses of the three anchors we used for our analysis. The 6MWD MID estimate was determined in a study investigating the effect of lung volume reduction surgery in severe COPD patients [11]. Therefore, the patient population was quite comparable to ours, but a different intervention was performed. Another study, which investigated the MID for 6MWD, found a similar outcome for treatment with pulmonary rehabilitation [24]. This 6MWD anchor was moderately correlated with SGRQ scores in our study (r=0.5). The MID for RV was established in almost the same population as we used (93 patients overlap) and therefore is not the best anchor for establishing our MID [12]. The MID was determined in one single study and we feel it should be re-tested in more studies to increase the reliability. The MID for RV was moderately associated with SGRQ score (r=0.4). A positive feature of the MID for FEV₁ is that it was determined in several studies comprising general COPD populations [10]. A downside was that determining the MID of FEV₁ was never the primary objective of these studies [10]. In our study, FEV₁ was moderately associated with SGRQ scores (r=0.4). In contrast to the anchor-based MIDs, the distribution-based MIDs were found to be higher at 6 months follow-up than at 1 month follow-up. This can be explained by the fact that, inherent to their methodology, distribution-based methods depend highly on variance in the population. On 6 months follow-up, the variance of measurements in the population is larger, leading to a higher distribution-based MID at 6 months, making it a more time-dependant method than the anchor-based method.

A limitation of our study is that most of our included studies were noncontrolled. Another limitation of our study is that all our patients participated in trials investigating a BLVR treatment, which is known to be a highly effective intervention resulting in a large decrease in SGRQ scores after treatment [25]. This could affect the size of the MID estimates. Future studies, compromising multiple treatment modalities other than BLVR, are needed to confirm our findings.

A strength of our study is that we were able to maintain a high level of standardisation as all measurements were performed in one specialised research hospital in The Netherlands, applying the same settings/measurement sequence, using the same equipment. Furthermore, compared with most other MID determination studies, we had a relatively large sample size [12, 13, 24].

In most studies determining a MID, results are solely expressed as absolute numbers. In our study, we have also determined the relative MIDs. Relative MIDs are able to evaluate changes adjusted for baseline SGRQ scores. We believe that when calculating a MID, besides the absolute MID, the relative MID should also be estimated.

In conclusion, this study is the first to establish a MID SGRQ specifically for patients with severe COPD. Using a combination of anchor- and distribution-based methods we established a MID 1 month after treatment of −8.3 units and a MID 6 months after treatment of −7.1. The relative MID 1 month after treatment was −13.4% and 6 months after treatment was −11.1%. Our new MID estimates could be applied for both interpreting SGRQ outcomes as well as sample size determination in future clinical trials investigating severe COPD patients.