This longitudinal study of patients with CIN showed that the EQ-5D-5L was responsive to change in health after surgery, and the effect size was small to moderate. The index value and EQ VAS after treatment were average 0.039 and 5.35, which can be considered an improvement in health from patients’ perspective. However, the MCID estimated in this study can only represent truly meaningful change of HRQoL score at group levels, not individual levels.
Among all dimensions, the anxiety/depression dimension was the most improved post-surgery, and the only one with a significant score change. This is similar to the results of a longitudinal HRQoL assessment of patients with CIN by Xie and colleagues, who assessed patients one month after treatment, and found that the average improvement in mental component summary scores (MCS) measured by the SF-36 questionnaire was higher than that of physical component summary scores (PCS; △MCS:7.05 vs. △PCS:1.47) [43]. A possible explanation is that, in general, CIN does not produce symptoms or signs that affect patients’ ability to perform, whereas a CIN diagnosis has a negative psychological impact [2, 3]. Howbeit, the psychological support of doctors, good prognosis examples of patients, and increased awareness of disease may ameliorate psychological.
In all patients, the positive changes in the index value and EQ VAS also coincided with other studies. A prospective study of Chinese patients with CIN conducted by Zhao and colleagues found that EQ-5D scores 1 month after treatment were significantly better than at baseline [6]. Therefore, we considered that post-surgical changes to patients’ health can be qualitatively judged by the change in EQ-5D-5L score. Interestingly, the index value and EQ VAS of patients whose response to the GRCQ was “improvement” increased significantly, while the different result was discovered in “about the same.” Bilbao and colleagues revealed similar results among patients who underwent surgery for hip or knee osteoarthritis, that is the mean change of the EQ-5D-5L score was positive in “improved” group [28]. Patients’ perceived health changes, as measured with the GRCQ, were consistent with EQ-5D-5L score changes; even though the GRCQ has only one question and the EQ-5D-5L is a multi-dimensional, multi-attribute questionnaire. Thus, the GRCQ is a simple and credible choice for determining whether health changes occurred when multiple-items questionnaires cannot be used.
Two of the most commonly used indicators of responsiveness—ES and SRM—were used to estimate the degree of change in patients’ health [44–46]. The effect size of the EQ-5D-5L across the entire sample was only between small and moderate, which mirrored previous studies. Chen and colleagues assessed the responsiveness of the EQ-5D-5L with 65 Taiwanese patients who were receiving rehabilitation after a stroke, and the effect sizes ranged from 0.40 to 0.63 for index value and 0.30 to 0.34 for EQ VAS—suggesting small to moderate responsiveness [47]. Furthermore, the effect size of index value was only 0.20 in patients after cataract surgery [48]. Another study of obese patients showed that the index value and EQ VAS had only small responsiveness after bariatric surgery [49]. These findings suggest that the EQ-5D-5L is responsive to various conditions, which clarifies that health changes were clinically relevant rather than random errors; nonetheless, the small responsiveness is noteworthy. The reason may be that the study population had chronic diseases, who experienced slow health deteriorating and had weak perception of health change as compared to patients with an acute disease who may recovery rapidly.
Some researchers believe that responsiveness may depend on the direction of changes in health state and individuals’ health state at baseline [36], and this theory was supported by the current results. We found a moderate responsiveness of index value and EQ VAS in patients with improved health states, while small or no responsiveness was found in patients with no change. In addition, the baseline scores of index value and EQ VAS in “improvement” surgical patients were lower than those that were “about the same,” while score change was higher than latter. Statistically, the responsiveness of patients with improved health states must be better than that of “about the same” patients.
Responsiveness of the EQ-5D-5L in patients with improved health states was also studied in other populations; however, the results were inconsistent. In patients with acute asthma, who underwent one month of treatment and self-reported improved health states, the index value had moderate to large responsiveness with the effect size ranged 0.63 to 0.95 [50]. Golicki and colleagues revealed that the EQ-5D-5L was consistently responsive in patients who had a stroke, who displayed improved health four months after treatment: the index value showed a moderate ES (0.51–0.71) and a moderate to large SRM (0.69–0.86), while the ES of EQ VAS ranged from 0.51 to 0.65 and the SRM ranged from 0.59 to 0.69 [51]. Another study of patients with osteoarthritis six months after surgery showed that patients with improved health states had ES and SRM of 1.48 for index value, and ES of 0.82 and SRM of 0.90 for EQ VAS [28]. Through the above, we found that although the responsiveness of “improvement” patients was at least moderate, the effect size of each study was quite different. The source of the difference may be participants’ unique characteristics or different time intervals for the two measurements [47]. Because longer time intervals allow for sufficient time to respond to one’s physical condition, this is reflected in larger score changes, resulting in a larger effect size to reflect the degree of change in health upon full recovery, and vice versa [37].
MCID is a vital component of questionnaire application. Previous studies have utilized the mean change of MCID scores in the anchor-based method [52, 53]; however, this does not consider the possible impact of HRQoL scores over time in patients who reported no health changes during follow-up [39]. However, in this study, the absolute value of score change in participants that scored “a little better” minus the score change in participants that scored “about the same” was used as the MCID; thus, we eliminated the potential impact of time on the MCID estimation.
Besides the distribution-based and anchor-based methods, the instrument-defined method can also be used to triangulate the MCID. Luo and colleagues used the instrument-defined method to estimate the MCID for the EQ-5D-3L, and the result was parallel to the published estimate; therefore, the instrument-defined method was regarded as an effective method for MCID estimation [34]. Owing to our results, we deem that the instrument-defined method can be used for the MCID estimation of the EQ-5D-5L in patients with CIN.
Concerning the relationship between MCID and MDC, the results demonstrated that the MCID estimated for index value and EQ VAS by the three methods can, at the group level, explain that the score change was owing to health changes rather than measurement error. However, MCID of index value and EQ VAS both cannot account for individual health changes at the 95% confidence level, possibly because of the inclusion of patients with different histopathological histories. In this study, the proportion of patients with carcinoma in situ was 22.00%. Although this belongs to CIN [1], compared with other pathological grades, it involves a higher risk of progressing to invasive cancer [54], and patients had lower psychological expectations of health changes; therefore, the result may be owing to different criteria that patients use to judge their health changes. Another possible explanation may be that, although we only included first-diagnosed patients, the HRQoL scores at baseline of some patients with a longer disease duration may be more improved compared to those more recently diagnosed, resulting in the baseline score of the entire sample being raised. Therefore, the possibility of underestimating MCID leads to it being less than MDC95%(ind). The current results should be further validated in patients with the same pathological grade and the same disease duration.
This study had several advantages. First, we used a combination of qualitative and quantitative approaches to assess responsiveness, which increases the credibility of the results. Second, in addition to the distribution-based and anchor-based methods, using the instrument-defined method for MCID estimation highlights the value of our results. Third, we analyzed whether the MCID estimated by each method can reflect true health changes at individual and group levels, which allowed us to determine the reliability of MCID and avoid incorrect application or interpretation of MCID. Although judging whether MCID differs from measurement error is a logical next step after MCID estimation [42], only a few studies have done this [55, 56]. Finally, there was no investigator-based measurement bias because both time-point surveys for each patient were performed by the same investigator.
This study also had several limitations. Apart from the GRCQ, a disease-specific questionnaire was a commonly used anchor in previous studies [25, 38]; however, we did not use a disease-specific questionnaire for CIN, such as the Functional Assessment of Chronic Illness Therapy–Cervical Dysplasia, since there is no Chinese version [57]. Although the GRCQ has only one question, it is the accepted anchor for MCID estimation at this stage [17]. Studies have shown that, if health state changes in different directions, the MCID may also be different [58]. Because no patients reported a worsened health change in this study, MCID could not be estimated for this group of patients. Future studies could develop MCID for these patients to determine whether it differs from improved patients. It is well known that MCID changes are associated with demographic characteristics, interventions, and so on [33, 59]; therefore, the current results cannot be generalized to other clinical settings. Another limitation is that different interview methods used during baseline and follow-up surveys may lead to information bias. Furthermore, the small sample size may affect MCID accuracy; although, this study met the basic requirements for MCID estimation [60].