Establishing Minimal Clinically Important Differences for the Quality of Life Instrument in Patients with Esophageal Cancer QLICP-ES (V2.0) Based on Anchor-Based and Distribution-Based Methods

Background: The minimal clinically important difference (MCID) is an important phrase with big appeal in a eld strug gling to interpret quality of life (QOL) and other patient-reported outcomes (PRO), is also a bridge between statistics and clinical medicine. This paper is aimed to determine the MCID of esophageal cancer scale among Quality of Life Instruments system for Cancer Patients, QLICP-ES (V2.0). Methods: According to the scoring rule of QLICP-ES (V2.0), the scores of each domain and the overall of the scale were calculated. The MCID values of this scale were established by anchor-based and distribution-based methods. Two criteria A (improves one level after treatments) and B (at least improves one levels after treatments) were dened treatments effects in anchor-based methods, while methods of ES, SEM and RCI were used in distribution-based methods. Results: Using the anchor-based method, according to standard A, the MCID values of physical domain, psychological domain, social domain, common symptom and side-effects domain, the specic domain and the overall were 15.1, 4.4, 3.1, 6.7, 8.5 and 6.0 respectively. According to standard B, the MCID values of above domains and the overall were 19.3, 4.2, 4.8, 7.7, 9.5 and 7.5 respectively. Under the distribution-based methods, the MCID values above calculated by each method (ES, SEM and RCI) are in different ranges from 1.1 to 13.3. Conclusion: All methods have its own advantages and disadvantages to develop the MCID values, so it is necessary develop the MCID values with a variety methods considering the actual


Background
Esophageal cancer (ES) is the sixth most common cancer in the world and the fourth most common cause of cancer deaths in China. Nearly half a million new ES cases are diagnosed each year [1,2]. In China, the incidence of esophageal cancer ranks the sixth among malignant tumors, and the incidence of male cancer is higher than female cancer, and the incidence of rural cancer is higher than urban cancer [3]. The clinical symptoms of early esophageal cancer are not obvious, and dysphagia is often the main symptom, which may be combined with gastrointestinal and other system-related symptoms [4], resulting in somatic or functional damage, which seriously affects the physical and mental health of patients and reduces their quality of life (QOL) and other patient-reported outcomes (PRO). With the medical model entering the biopsycho-social medical model (modern medical model), people pay more attention to the improvement of QOL while paying more attention to the prolonging of life.
Therefore, the assessments of QOL have been applied as signi cant outcome indicators for patients with ES [5.6], given the time course of the disease, the di culty in curing and the burden of treatments. Consequently, several speci c instruments for patients with ES have been developed and are used in cancer clinical researches, including European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Group questionnaire, the QLQ-OES23 [7.8], the Functional Assessment of Cancer Therapy-Esophageal cancer questionnaire (FACT-E) [9]. In China, the QOL team led by professor Chonghua Wan developed a scale system called QLICPs (Quality of Life Instruments for Cancer Patients), in which the QLICP-ES (the rst version and the second version) is an instrument for quality of life measurement of esophageal cancer [10,11]. The QLICPs is a Chinese QOL instruments system developed by module approach with a general module (QLICP-GM) being used with all types of cancer, and some speci c modules for different cancers [12,13].
The QLICP-ES (V2.0) includes a generic module QLICP-GM (V2.0) and an esophageal cancer speci c module, with the QLICP-GM (V2.0) including physical function (8 items), psychological (9 items), social functions (8 items), common symptoms and side effects (7 items), a total of 10 side and 32 items in 4 domains. The esophageal cancer speci c module includes four domains and 16 items. The whole scale consists of 14 facets and 48 items in 5 domains (dimensions), which can be used for patients with various types of esophageal cancer, and can be used for the measurement of patients' quality of life in the period of onset, treatment and rehabilitation.
For reasonable explanation of the actual clinical signi cance of questionnaire survey and the scale score, the minimal clinically important differences (MCID) as an important change assessment tools or scale responsiveness of appropriate benchmarks have been proposed for the rst time in 1987. Canadian scholar Jaeschke and others de ned MCID as the minimum variation of questionnaire dimension score acceptable to patients without considering side effects and costs [14,15].
The MCID is an important phrase with big appeal in a eld strug gling to interpret QOL and other patientreported outcomes (PRO), is also a bridge between statistics and clinical medicine. How to develop reasonable and reliable MCID values has become a hot topic for scholars. There are mainly two traditional methods, including anchor-based method and distribution-based method. At present, although QLICP-ES (V2.0) has been developed, MCID has not been developed. For this reason, in this paper, the MCID of QLICP-ES (V2.0) was developed by using the commonly used anchor-based method and distribution-based method, and the advantages and disadvantages of the two methods were compared to lay a foundation for clinical application.

Data sources
In this study, 232 inpatients with esophageal cancer diagnosed by pathological examination and diagnosed by thoracic surgery in a provincial cancer hospital were selected as the research objects. The inclusion and exclusion criteria are as follows: Inclusion criteria: (1)Patients diagnosed with esophageal cancer by X-ray and beroptic esophagoscopy or gastroscopy biopsy; (2) Inpatients receiving treatments for esophageal cancer; (3) Understanding of the questionnaire due to the fact that the patient himself have a primary school education or above; (4)Patients who voluntarily participated in the test on this quality of life scale [16].
Exclusion criteria: (1) Patients who are illiterate or lack the ability to read and write;(2) Patients who are unable to express their true feelings clearly due to vague consciousness during hospitalization;(3) Patients who are unable to participate the test due to suffering from other serious diseases.
The QLICP-ES (V2.0) scale and EORTC QLQ-C30 were given to 232 patients with esophageal cancer before and after treatments, and was lled in once before and after treatments.

Anchor-based methods
The anchor-based method is to clarify the meaning of the scale score change by examining the relationship between the scale and another independent measurement tool score or other indicators. There are crosssectional anchors and longitudinal anchors. In this study, longitudinal anchors were selected to compare the curative effect before and after treatments. First, the Raw Score (RS) of domains were computed according to the number of questions and the answers of patients in each domain. Then, linear transformation is carried out with range method to convert the original score into a standardized score within 0-100.The score calculation method of each domain is as follows [17]: Second, the Q29 item in the EORTC QLQ-C30 scale, "how would you evaluate your overall health in the past week", was selected as an anchor after considering the correlation coe cient between Q29 and QLICP-ES (V2.0) . Then, patients with a difference of one grade (standard A) and at least one grade (standard B) in Q29 before and after treatments were selected, and the score differences in various domains before and after treatments were calculated respectively, and the mean value of the difference was denoted as MCID.

Distribution-based methods
The distribution method is to determine the MCID from a statistical point of view by using the distribution(variation) of the sample data of the evaluation tool. The commonly used indexes to calculate variation include Effect-Size (ES), Standard Error of the Measurement (SEM), and Reliable Change Index Effect-Size (ES) is obtained by dividing the difference in mean scores from baseline 0 to post-intervention 1 by the standard-deviation of the baseline score (SD baseline ) [18], the calculation formula and the corresponding MCID are as follows: ES is often used to compare two or more groups to measure the size of the difference between groups. In health-related quality of life assessments, ES is currently the most recognized parameter in determining the importance of group or individual changes. Cohen [19] empirically de ned an effect size of 0.2 as small,0.5 as moderate, and 0.8 as large.
Standard Error of the Measurement (SEM), de ned as the baseline SD multiplied by the square root of one minus sample test-retest reliability coe cient, were also calculated for comparison purpose [20]. The reliability is usually estimated using a test-retest reliability estimate, but some authors also use an internal consistency estimate, for example Cronbach's alpha [21]. The calculation formula and the corresponding MCID are as follows: SEM is assumed to be fairly sample-independent [22], which is its best advantage: a growing standard deviation is balanced by a higher reliability. Some authors like Wyrwich et al. consider one SEM as an approximation of the MCID [23,24]. X can be assigned to 1(small effect), 1.96 (medium effect), 2.77(large effect).
Reliable Change Index (RCI) is the change value of the questionnaire score divided by the square root of SEM. If RCI is greater than 1.96, then the change value has a 95% chance of being a meaningful change [25]. The calculation formula and the corresponding MCID are as follows:

Results
The demographic characteristics of the sample Table 1 showed the general demographic characteristics of the sample. A total of 232 patients with esophageal cancer were investigated in this study. There were 204 males and 28 females. The age distribution was 35-82 years old, with the mean of 59.3(SD8.9). Male elderly patients are more common. The occupations were mostly workers and farmers, with 20 workers (20.0%) and 35 farmers (35.0%) respectively.
The educational level varies, among which Middle school or High school has the highest proportion(55.6%). The MCID values using Anchor-based methods Table 2   The MCID values using Distribution-based methods    [26,27]. Table 4 and Table 5 showed the MCID values calculated using SEM and RCI methods in detail.

Discussion
The main function of MCID is to assist researchers and clinical staff to explain the signi cance of changes or differences in the score of evaluation tools. MCID values can be used to determine whether changes in the score of the scale brought by clinical interventions have "clinical signi cance" [28]. For clinicians and researchers, the key for using measurement instruments is the effectiveness and stability of the MCID score. The correlation coe cient r between the selected anchor Q29 and the score of the total scale was 0.72, but the correlation coe cient between the anchor Q29 and some domains, such as psychological domain and social domain, was 0.17 and 0.32, which were not too high. Therefore, the distribution-based methods were further adopted and the MCID values under ES, SEM and RCI were calculated respectively. Studies have shown that when calculating MCID values by ES method, 0.5 medium effect was the most appropriate, while 0.2 and 0.8 were easy to be too large or too small in calculation [29]. And the results are closer to the MCID values calculated by the anchor-based method. SEM is not completely dependent on the sample information and is slightly affected by the sample size. Therefore, scholars recommend 1SEM or 1.96 SEM as MCID values.
Anchor-based methods and distribution-based method have their own advantages and disadvantages, so this study used a variety of methods to calculate the MCID values. The advantage of the anchor-based methods is that the change in the outcome measure score is associated with a meaningful external anchor that re ects the patient's point of view and provides a professional explanation for the minimum change in clinical signi cance determined. However, the disadvantage is that the measurement error is not taken into account, different criteria will produce different changes in the minimum clinical signi cance, and it is often di cult to nd a suitable anchor. Moreover, in this paper, the sample size is not too large and only subjective anchors are used, subsequent studies will expand the sample size and use objective anchors for further studies. The distribution-based methods take measurement error into account to determine the MCID values, which is relatively easy to implement, but the results obtained from different samples may be different, so the professional explanation cannot be given for the determined minimum measurable change value, and there is a lack of recognized judgment criteria for the obtained minimum measurable change values. In developing MCID, the anchor-based method is generally preferred. When there are no good criteria or a small sample size, the distribution-based methods will be considered comprehensively. In this paper, a lot of different MCID values were presented so that it can be easy and convenient to select by users.
Although MCID as a minimum threshold plays an important role in determining the scale score changes, its stability and variability may be affected by many factors, such as defects in the measurement method, demographic characteristics, patient comprehension and study cycles and so on [30].

Conclusion
In conclusion, appropriate methods should be adopted to evaluate the MCID of each scale. When MCID is applied to evaluate the clinical e cacy of esophageal cancer, it is necessary to carefully review and fully understand the concepts and in uencing factors of each method, so as to evaluate the e cacy of esophageal cancer patients with scales more objectively. In this paper, a lot of different MCID values were presented so that it can be easy and convenient to select by users.