Background
Although the stroke mortality rate has been declining [
1], the estimated prevalence rate of stroke-related disability is about 331 per 100,000 [
2]. Stroke disability and morbidity cause reduced quality of life (QOL) among stroke survivors [
3]. The greater the disability, the lower the QOL is [
4]. With ongoing rehabilitation, however, improvements in functional status are possible [
5] and contribute to increase QOL for stroke survivors. Therefore, the assessment of stroke rehabilitation should include disability and QOL domains, which are influenced by the disease [
6‐
9].
Generic QOL instruments such as the Medical Outcomes Study Short-Form 36-item survey (SF-36) may underestimate the effect of stroke [
10]; therefore, disease-specific tools are considered more helpful in providing information about the difficulties that patients with stroke may experience [
7,
11]. Because the information from the patients' perspective on the consequences of disease and the therapeutic benefits is considered critical in the evaluation of health care, patient-reported outcome measures have been used to supplement clinical decisions made from physician-based outcome measures [
12]. Of the stroke-specific scales, the Stroke-Specific Quality of Life Scale (SS-QOL) [
13], in addition to the Stroke Impact Scale version 3.0 (SIS 3.0) [
14], is the most comprehensive [
15] and frequently used patient-reported outcome measure [
16‐
19].
The SS-QOL is a self-report questionnaire consisting of 49 items in the 12 domains of energy, family roles, language, mobility, mood, personality, self-care, social roles, thinking, upper extremity (UE) function, vision, and work/productivity. The domains are scored separately, and a total score is also provided. The psychometric properties of the SS-QOL have been validated in patients with ischemic stroke and intracerebral hemorrhage [
10,
18,
20]. In patients with subarachnoid hemorrhage, the 12 SS-QOL domains and the total score demonstrated good internal consistency [
21]. The SS-QOL items also have acceptable agreement with the categories of the International Classification of Functioning, Disability, and Health, which indicates that the SS-QOL covers multidimensional components meaningful for patients with stroke [
22]. The clinical utility of the SS-QOL remains understudied, however, and several clinimetric properties, such as the minimal detectable change (MDC) and the clinically important difference (CID) of the SS-QOL, have not yet been investigated. This information helps inform clinical decision making on the discontinuation or alteration of a treatment program that aims to improve patients' physical function.
The MDC is the smallest change that can be detected by the instrument beyond measurement error. The CID is a related concept that shows how much change can be deemed as clinically important [
23]. That is, CID is the threshold score that a group of patients perceive as noticeable. The MDC and CID facilitate the interpretation of treatment outcomes. For example, the study by Lin et al [
24] reported that a true change in the SIS mobility subscale that occurs after rehabilitation needs to show an increase of at least 15.1 points or the change is likely due to an error in the measurement.
In some instances, CID scores do not exceed the MDC scores but still convey information about whether a patient group experienced a clinically important change. In the study of Plummer et al [
25], for example, the improvement of 0.11 m/s in gait speed was lower than the measurement error of 0.17 m/s reported by Evans et al [
26], indicating that the improvement of gait speed might not be real and beyond measurement error. However, the change of 0.11 m/s gait speed in the Plummer et al [
25] study indicated that this patient group improved from the category of physiologic ambulatory to that of full-time home ambulatory, according to the walking categories developed by Perry et al [
27]. Without these important benchmarks against which the clinical interpretation is based, clinicians may make erroneous conclusions about the effect of a treatment.
Therefore, this study sought to establish the MDC and CID score estimates of the SS-QOL subscales and assess the proportion of patients' change scores on the SS-QOL subscales that exceeded the MDC and CID in a cohort of patients with stroke who received rehabilitation therapies.
Discussion
To the best of our knowledge, this is the first study to determine the MDC and CID scores of the SS-QOL subscales that can be used to differentiate patients treated with stroke rehabilitation who experience real improvement and clinically meaningful change from those who do not. Our findings suggest that a patient's change score has to reach 5.9, 4.0, and 5.3 on the mobility, self-care, and UE function subscales to indicate a true change. That is, when the change scores between the patient's 2 measurements (e.g., baseline and follow-up) reach 24.6%, 20.0%, and 26.5% of the scale width on the mobility, self-care, and UE function subscales, the clinicians may interpret the changes in that patient as true and reliable, given the 95% confidence level.
There is no universally accepted standard for determining the CID [
48‐
52]. An integrated system for defining CID is recommended that combines anchor-based and distribution-based methods [
48]. The value and limitations of anchor-based and distribution-based methods in estimating CID have been recognized. The anchor-based approach emphasizes the primacy of a patient's perspective, but anchor-based CID scores may vary with demographic characteristics such as age [
49]. Although the distribution-based CID scores are easy to generate, these SD-based scores are associated with some bias due to sample heterogeneity [
38]. As a result, a number of recent clinical reports have advocated an approach that combines the anchor-based and distribution-based methods to refine the range of CID [
24,
50,
51].
Using a 1 SEM distribution-based approach, we found that the CIDs for the mobility, self-care, and UE function subscales are 1.7 (7.1% scale width), 1.2 (6.0% scale width), and 1.3 (6.5% scale width), respectively. The SEM incorporates a sample's variability and the reliability of the instrument. Several previous studies have shown that 1 SEM is close to the estimate of CID [
53‐
56]. Despite being theoretically constant [
56], the SEM may become larger with a low reliability [
57].
Furthermore, the CID scores using 1 SEM would be always less than the MDC values mathematically. Therefore, values of 0.5 SD were calculated as supportive information for determining the CID. On the basis of the 0.5 SD approach, we found that the CID scores for the subscales were 2.4 (10% scale width) for mobility, 1.9 (9.5% scale width) for self-care, and 1.8 (9% scale width) for UE function.
The CID values produced by the anchor-based method were 1.5 (6.3% scale width) for mobility, 1.3 (6.5% scale width) for self-care, and 1.2 (6.0% scale width) for UE function. These estimates were comparable with those obtained from the distribution-based approaches. Because a cutoff threshold of the group-level CID may potentially undermine the clinical interpretation of trial data [
58], we reported ranges rather than a single value. We found the CID ranges were 1.5 to 2.4 for mobility, 1.2 to 1.9 for self-care, and 1.2 to 1.8 for UE function. That is, patients with stroke who achieve mean scores in the ranges of 6.3% to 10.0%, 6.0% to 9.5%, and 6.0% to 9.0% of the scale width on the mobility, self-care, and UE function subscales are likely to have clinically meaningful change in these domains.
Of note, there is a concern about the differences between group and individual clinical importance [
59]. Average effects across a group may not be meaningful to the individual patient. Group-derived CID values are suitable to interpret the results of clinical trials or group studies, but they are often directly applied to interpret the individual's change [
59]. For individual-level use, it may be reasonable to expect that the MDC would be less than or equal to the minimal CID. However, some researchers have suggested that this is not always the case [
24,
60], which is also consistent with our current findings. When the MDC exceeds the minimal CID, the change score reaching a CID does not mean that patients have exceeded the measurement error, and both values are suggested to be considered in clinical decision making [
61].
Taking our cohort sample of stroke rehabilitation as an example, the mean change scores on the mobility, self-care, and UE function subscales were 3.5, 2.8, and 4.1 points, which exceeded the minimal CID ranges. This indicated that the improvements achieved after rehabilitative therapies in this cohort were meaningful to the patients. A mean change score of 1.2 on the self-care subscale in a previous study of the Chronic Disease Self-Management course [
17] was reported to achieve statistical significance. This improvement at the group level failed to achieve the lower bound of the minimal CID range established by our current study, which may weaken the validity of the study conclusion about the effect of the self-management education on the quality of self-care after stroke.
Although the validity of a self-rated global assessment scale has been criticized for its "retrospective bias" [
50,
62,
63], we recognized that clinical interpretation of the MDC and CID scores would be enhanced if a patient-driven anchor were included in the study design. Therefore, the reliable-change approach, as proposed by Davidson and Keating [
64], was adopted to expand the clinical application of the MDC
95 and CID established by the current study. The reliable-change approach addresses the question about the proportion of patients exceeding the threshold of MDC and CID. The concept is similar to the event rate, which represents the number of people in whom an event is observed [
65]. For example, the event rate is 40% if 40 of 100 patients experience an adverse event such as side effect. On the basis of our results, 9.5%, 6.8%, and 12.2% of patients achieved functional improvement beyond measurement error on the mobility, self-care, and UE function subscales. The greatest proportion of patients that exceeded the lower bound of the minimal CID was observed for the UE function subscale (33.8%), followed by the self-care (28.4%) and mobility (28.4%) subscales. According to Schmitt and Fabio [
66], the better the responsiveness of a scale is, the greater the numbers of patients who will exceed the minimal change criteria. Thus, the UE function subscale appears the most responsive subscale among those in the physical category of the SS-QOL for the patients of this study. Because the focus of the rehabilitation used in the current study was on the functional recovery of the paretic arm, it is also possible that the intervention effect was responsible for the relatively greater proportion of patients who exceeded the MDC and CID of the UE function subscale. Further research using larger samples is needed to validate the findings.
It is important to note that the participants included in this study were assigned to receive different treatment programs; thus, the variance in the change scores might have varied among the different treatment groups. Additional analyses of the CIDs for each intervention group showed that the differences in CID values represented by 1 SEM between the participants of each intervention group and the overall participants were less than 0.6 points in the mobility subscale (each intervention participants: 1.3-2.3; vs. overall participants: 1.7) and 0.4 point in the self-care (1.1-1.6 vs. 1.2) and UE function subscales (1.1-1.7 vs. 1.3); and the differences in CID values represented by 0.5 SD between each intervention group participants and the overall participants were less than 0.7 points (mobility: 1.8-3.1 vs. 2.4, self-care: 1.6-2.4 vs. 1.9, and UE function: 1.6-2.5 vs. 1.8). Generally speaking, the CID values in each intervention group are arguably close enough to allow collapse of data from all intervention groups into one group for analysis in each subscale. Given the above information and the fact that the same amount of treatment duration and intensity were used across the different treatment programs, we felt the method of collapsing the data from various intervention groups would be justifiable. For example, some recent studies [
67,
68] have combined the data from different intervention groups for clinimetric analyses.
The current investigation has some limitations that warrant consideration when interpreting and generalizing the study findings. First, the generalizability of the current findings might be limited. Because we only included patients from departments of rehabilitation with the demonstration of Brunnstrom stage III or higher for the affected UE, the current findings may not be suitable for stroke patient at a Brunnstrom stage of less than III. In addition, some patients were excluded from the current investigation due to cognitive difficulties. To increase the external validity of the results of this study, it is warranted to recruit a wider sample of patients with stroke with differing levels of motor impairment and cognitive difficulty.
Second, because of the relevance of proxy reports for QOL outcome evaluations, particularly in patients with stroke with language impairments [
69], there is a need for extended research on the clinimetric properties of the proxy version of the SS-QOL to establish the minimal significant change perceived by the proxies.
Third, although patients who have received different treatment programs with the same treatment duration are often pooled together for clinimetric analysis of the outcome measures [
67,
68], further research is needed that may investigate the MDC and CID of the SS-QOL for specific interventions based on larger samples to provide further insights into the clinimetric properties of the SS-QOL in specific contexts.
Finally, there are potential clinimetric differences in patient-reported QOL outcomes due to the modes of administration [
70]; thus, further research may study clinimetric attributes of the SS-QOL administered in different modes, such as paper-and-pencil administration vs. telephone interviews vs. Web-based electronic data collection.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
KCL conceived the study, participated in its design and coordination, and helped to draft the manuscript. TF participated in the design of the study, performed the statistical analysis, and participated in the writing of the manuscript. CYW contributed to secure the research funding, designed and conducted the study, and participated in the data interpretation. CJH contributed to the revision of the manuscript. All authors read and approved the final manuscript.