A multicenter longitudinal study on development and validation of the quality of life scale QLICD-DM (V2.0) for diabetes mellitus based on classical test theory and generalizability theory

Abstract

Purpose

Quality of life (QOL) research in diabetes has long been a concern with many specific instruments being developed, but no scale for diabetes has been developed based on the modular approach. This study aimed to develop and validate the Diabetes Mellitus Scale of the System of Quality of Life Instruments for Chronic Diseases QLICD-DM (V2.0) by a modular approach and mixed methods.

Methods

The Scale was developed based on procedural decision-making methods and by combining the general module and the specific module. The instrument was used to assess the quality of life of 242 diabetic patients both before and after treatments. Under Classical Test Theory (CTT), the psychometric properties of the scale were assessed with regard to validity, reliability and responsiveness by correlation analyses, structural equation modeling, as well as t-tests. In addition, G-study and D-study in Generalizability Theory (GT) were used to validate the scale further.

Results

The QLICD-DM (V2.0) was developed with a 14-item specific module and a 28-item general module in the final scale. For all domains, Cronbach’s α values were greater than 0.70 with the exception of physical function (0.68), the test-retest reliability correlations r and ICCs were greater than 0.80 with the exception of physical and social function (0.65,0.64). The theoretical construct was supported by correlation analyses and confirmatory factor analysis using structural equation modeling, which demonstrated good construct validity. There were significant differences (P < 0.05) in the domains of physical function, specific module and the total scale before and after treatments. The standardized response means (SRMs) of the physical, psychological, social, and the specific module were 0.36, 0.14, 0.11 and 0.28 respectively. All G-coefficients were all greater than 0.70 with the exception of the physical domain (0.683), further confirming the reliability of the scale further. The overall error was found to be small in the G-study and in the D-study, indicating a high level of accuracy.

Conclusion

The QLICD-DM (V2.0) not only integrated characteristics of generic and disease-specific instruments but also demonstrated good reliability, validity, and moderate responsiveness, and it could serve as a quality of life assessment tool for diabetes patients.

Introduction

Diabetes mellitus encompasses a group of metabolic disorders resulting from the interaction of genetic and environmental factors. The prevalence of diabetes mellitus and its associated complications has increased substantially worldwide, becoming a major global public health concern and posing a serious threat to individuals’ health and well-being [1]. Currently, diabetes affects approximately 8.30% of the global population, these two statistics are projected to rise to 592 million individuals and 10.10% respectively. The danger of diabetes is not only its high prevalence, but also the serious complications that occur as the disease progresses, such as ketoacidosis and peripheral neuropathy, in addition to being including ketoacidosis, peripheral neuropathy, nephropathy, blindness, and lower-limb amputation [2]. In addition to impacting work performance, increasing disability and mortality, diabetes also places a significant economic and emotional strain on the individual, substantially reducing their overall quality of life (QOL). Furthermore, diabetes does not have a definitive remedy, necessitating ongoing therapy that also impacts patients’ behaviours and ways of life, consequently influencing their overall quality of life.

Previously, the management and therapeutic effectiveness of diabetes were primarily evaluated using biochemical indicators. However, in the contemporary healthcare system, diabetes management involves not only glycemic control but also the improvement of patients’ overall QOL [3]. Quality of life research in diabetes has attracted considerable attention, and the evaluation in diabetic patients has been conducted since the 1960s. Quality of life scales can effectively assess patients’ quality of life and provide valuable information for clinical decision-making and treatment planning [4]. The large number of available instruments and the lack of clear guidance for selecting appropriate scales seriously hinder the promotion and application of scales. Some of the scales were not developed using standardized procedures or have not been rigorously validated based on psychometric theories resulting in a waste of resources. When selecting instruments to evaluate quality of life, the general scales are often used such as the SF-36,WHOQOL-100,WHOQOL-BREF [5–7]. However, these generic instruments often fail to adequately capture diabetes-specific symptoms, treatment-related burdens, and disease-related side effects.

Consequently, some specific instruments have been developed and increasingly applied in many studies. Current diabetes-specific scales include DCCT, DQLCTQ, ADDQOL, Diabetes-39(D-39), DTSQ, ITR-QOL, Diabetes Distress Scale(DDS) [8–14]. The general scales are widely used by the population, but disease-specific issues are not well represented and have poor responsiveness, whereas disease-specific instruments are more sensitive to the unique experiences of diabetic patients. However, to the best of our knowledge, no scale for diabetes was developed using a modular approach that integrates a generic module with disease-specific modules (a general module plus specific modules).

In China, due to sociocultural differences, quality of life needs to be examined in the context of Chinese society [15]. However, in the Chinese cultural context, there are only a limited number of scales specifically designed for diabetic patients, and the majority of these scales are translated or culturally adapted versions of foreign instruments. The content and items vary widely, and the framework is uneven, creating difficulties for both researchers and clinicians.

To address these limitations and to meet the demand for culturally appropriate instruments for research and clinical studies, we have developed a system of Quality of Life Instruments for Chronic Diseases (QLICD) by modular approach [15–16], This system combines a generic QOL module with disease-specific modules for each of the disease considered. The general module, called QLICD-GM, can be used with all types of chronic disease patients, while the specific module captures unique aspects of QOL associated with particular diseases, thereby compensating for the limited specificity of generic instruments in the general module by capturing the unique aspects of QOL pertaining to the specific disease. For example, the diabetes instrument QLICD-DM is constructed by combining QLICD-GM with the specific module for diabetes [17]. The latest version of the system QLICD(V2.0) contained 34 chronic disease-specific scales including QLICD-CG for Chronic Gastritis [18], QLICD-HY for hypertension [19], QLICD-SLE for systemic lupus erythematous [20], QLICD-RA for rheumatoid arthritis [21], QLICD-OS for osteoporosis [22], and QLICD-CHD for coronary heart disease [23].

This study aimed to report the development and validation process and results of QLICD-DM (V2.0), with a focus on its development procedures, psychometric evaluation, and major findings.

Methods

Development of the QLICD-DM(V2.0)

The QLICD-DM(V2.0) was developed by combining the generic module for chronic diseases, QLICD-GM [15, 16], and the newly developed diabetes-specific module.

The development of the QLICD-GM (V2.0) strictly followed the internationally recognized programmatic decision-making approach, including multiple procedures such as establishment of a theoretical framework, proposing a pool of alternative items, item screening, scale evaluation, etc. A focus group comprising statisticians, chronic illness doctors, diabetes patients and their families, researchers, and a nominal group made up of these individuals designed and evaluated the instrument. Semi-structured interviews were conducted with doctors and nurses, as well as with patients suffering from various chronic diseases. Following the interviews, some items considered less relevant or less important were revised after discussion. Finally, after the analysis and screening of testing data and several rounds of focus group discussion, the final version of QLICD-GM (V2.0) was developed and demonstrated good psychometric properties [15, 16]. QLICD-GM (V2.0) includes 3 domains of physical function (9 items, coded GPH1-GPH9), psychological function (11 items, coded GPS1- GPS11) and social function (8 items, coded GSO1-GSO8), with a total of 28 items and 9 facets (See Fig. 1 in detail).

Fig. 1.

Steps towards development and validation procedure of QLICD-DM(V2.0)

Similar to the development of QLICD-GM [15, 16] and other specific modules for chronic gastritis, hypertension, systemic lupus erythematous, rheumatoid arthritis, osteoporosis and coronary heart disease [18–23], the disease-specific module for diabetes mellitus was developed. Based on the conceptual framework, the specific module consisted of four facets, complication symptoms, treatment-related psychology, and treatment related mentality. The existing domestic and international QOL specific scales for diabetes, clinical manifestations of diabetic patients, and their particular psychological and social characteristics were reviewed and synthesized to generate candidate items for the specific module for diabetic patients. Based on a comprehensive literature review and experts’ experience, the members of the research group independently proposed 18 non-overlapping items which formed the initial item pool of the specific module. To determine the specific module, two rounds of expert discussions were conducted, and 14 non-overlapping items were selected with 4 facets of Specific symptoms(SPS), complication symptoms (COS), disease-related psychology (DRP), treatment-related psychology (TRP) .

The entire development and evaluation process was summarized in Fig. 1.

Validation of the QLICD-DM(V2.0)

The two modules were subsequently combined into the QLICD-DM(V2.0). It includes 4 domains, 13 facets and 42 items, of which the physiological domain (PHD, 9 items) is coded GPH1-GPH9, the psychological domain (PSD, 11 items) is GPS1- GPS11, social function (SOD, 8 items) is GSO1-GSO8, the specific module (SPD, 14 items) is DM1-DM14.The instrument was administered to patients with diabetes mellitus to validate the QLICD-DM(V2.0).

Subjects

In this research, 242 patients with diabetes mellitus who were recruited from the affiliated hospitals of Guangdong Medical University, Shilong Boai Hospital and Dalang Hospital in Dongguan City were selected for the self-rated scale survey.

Inclusion criteria: ①patients diagnosed with diabetes mellitus and the ability to read, write and communicate; ② no other major mental illness or mental disorder; ③ informed consent and voluntary cooperation.

All included patients gave informed consent and voluntarily participated in this study.

Survey methods

In accordance with the Helsinki Declaration, each respondent participated in the survey with informed consent. This study was approved by the Ethics Committee of Guangdong Medical University with the approval number PJ2015050KT. A brief explanation of the study objectives and procedures was provided by the investigators. The investigators administered the QLICD-DM (V2.0) after obtaining their informed consent. On the first day of hospitalization to the hospital, each participant finished the QLICD-DM(V2.0) and Chinese version of SF-36 questionnaires [24]. The test-retest reliability was evaluated using retest surveys on the next day following admission using the same questionnaires, and a responsiveness assessment was made before discharge in the third survey. In order to ensure the accuracy of the answers, investigators reviewed the questionnaires immediately after completion.

Scoring methods

The QLICD-DM(V2.0) uses a five-point Likert scale for each item. These levels include “not at all,” “a little bit,” “somewhat,” “quite a bit,” and “very much.” Scores ranging from 1 to 5 were assigned to each level, reverse-scored items were scored in the opposite direction. By summing the scores of domain/facet items, the raw scores of facets and domains were obtained. The raw score is the sum of the scores of each item in that domain or facet. Domain scores were then summed to determine the total score.

All domain scores were linearly transformed into standardized scores (SS) between 0 and 100 to facilitate comparisons. Specifically, SS = (RS − Min) × 100 / R, where RS, Min, and R represented the original score, the lowest score, and score range.

Validation of the QLICD-DM(V2.0) based on CTT

Under CTT, Cronbach’s α coefficient was commonly used to assess the internal consistency reliability in the scale development, with Cronbach’s ɑ coefficient > 0.70 indicating good internal consistency [25–27]. Test-retest reliability for the QLICD-DM was assessed using Pearson correlation r with values greater than 0.80 being considered acceptable.

To evaluate construct validity of the study, Pearson correlation coefficients were utilized to compare the relationship between the items and their respective domains, with a correlation coefficient greater than 0.40 being considered acceptable. Also structural equation modelling (SEM) was employed to evaluate construct validity, in which CFI(comparative fit index), TLI(Tucker-Lewis index), RMSEA(root-mean-square error of approximation) and SRMR(standardized root mean square residual) are indices recommended as sensitive to model misspecification, with the CFI and TLI with values greater than 0.90 and RMSEA, SRMR less than 0.08 reflecting a good fit of the model to the data. Correlation coefficients between QLICD-DM domain scores and SF-36 domain scores can be used to determine criterion-related validity.

For the assessment of responsiveness, the mean difference between pre- and post-treatment was compared, and the standardized response mean (SRM) was calculated to represent the degree of responsiveness, and 0.20, 0.50 and 0.80 representing small, medium, and large responsiveness respectively [28, 29].

Validation of the QLICD-DM(V2.0) based on GT

Generalizability theory is based on variance analysis to estimate multiple sources of measurement error from different sources by interacting them. It provides a more refined analysis of reliability than CTT. To control the measurement errors, GT introduces potential sources of measurement error that may affect test scores into the measurement model, such as differences among the research objects, item difficulty, scoring criteria, and interaction of these factors. The effects of these variables or factors on test scores were subsequently evaluated by variance analysis, in which the variance components are used as indexes [23, 30–33].

GT is mainly divided into G-studies (Generalizability studies) and D-studies (decision studies). This study used a one-facet multivariate crossed p* i design of multivariate Generalizability theory; in the model, the operating premise is that the subjects and the measurement items are completely random with a crossover relationship; where the subjects (p) are the diabetic patients, and the measurement items (i) are the individual items [23, 32, 33].

Statistical software

In the selection of software for this study, IBM SPSS Statistics 27.0 was used for the CTT component, MPLUS 8.0 for confirmatory factor analysis, and mGENOVA for Generalizability Theory analyses. All statistical tests were two-sided, with a significance level of 0.05.

Results

Characteristics of the patients

In this study, 242 patients with diabetes mellitus were recruited for the self-administered questionnaire survey, all of whom had type 2 diabetes mellitus. The majority of the patients in this survey were male (53.7%), most of the patients were married (86.40%), 27.30% of the patients were farmers by occupation, approximately half of the patients reported a moderate family economic status, 59.10% of the patients had social health insurance, and the mean age of the patients was 52.64 ± 32.36 years. In terms of educational level, 103 patients had a primary school education, Junior high school 81, high school or technical secondary school 35, Undergraduate and above 23 cases. All patients gave informed consent to this study and voluntarily participated in the study.

Results from analysis based on CTT

Reliability

The Cronbach’s α coefficients for the four domains (physical, psychological, social and the specific module) and also the general module and the total scale of QLICD-DM(V2.0) were 0.68,0.84,0.74,0.82, 0.88 and 0.91,respectively, exceeded 0.70, except for the physical domain (0.68), indicating acceptable to good internal consistency reliability. Additionally, the split-half reliability coefficients for the general module and the overall scale were 0.81 and 0.74, respectively.

The results of the first and second assessments were used to determine the test-retest reliability for each domain. The results showed that There were no statistically significant differences across domains between the first and second evaluations (p > 0.05). Therefore, the Pearson correlation r can be used to evaluate test-retest reliability with that of 4 domains and also the general module and the total scale of QLICD-DM(V2.0) being 0.65,0.80,0.64,0.83,0.81 and 0.86,respectively. The correlation coefficients of physical and social domains were less than 0.80 (0.65 and 0.64, respectively). These findings indicate that the QLICD-DM demonstrated acceptable to good test–retest reliability.

Validity

The research project on the system of Quality of Life Instruments for Chronic Diseases (QLICD) served as the foundation for the development of this scale. Following evaluation and discussion by a panel of experts, the items’ content encompassed the WHO’s suggested definitions of health and quality of life and included issues and concerns specific to patients with diabetes mellitus, indicating good content validity.

The results of the correlation analyses (Table 1) revealed that the correlations between items and their corresponding domains were stronger compared to the correlations of the items with other domains (most correlation coefficients were > 0.50). Also, an acceptable model fit was observed by examining the individual components of the modules, which were divided into four domains for the general module and four facets for the specific module.

Table 1.

Correlation coefficients r among items and domains of QLICD-DM V2.0(n = 242)

Items	Items brief description	Domains
		Physical	Psychological	Social	Specific
GPH1	Appetite	0.46 **	0.20**	0.27**	0.07
GPH2	Sleep	0.38 **	0.11	0.07	0.18**
GPH3	Sexual function	0.37 **	0.22**	0.15*	0.22**
GPH4	Excrement	0.45 **	0.24**	0.23**	0.23**
GPH5	Pain	0.57 **	0.40**	0.21**	0.35**
GPH6	Daily activities	0.67 **	0.34**	0.44**	0.20**
GPH7	Work	0.68 **	0.36**	0.38**	0.23**
GPH8	Walk	0.66 **	0.32**	0.38**	0.17**
GPH9	Fatigue	0.52 **	0.29**	0.17**	0.40**
GPS1	Attention	0.57**	0.53 **	0.42**	0.27**
GPS2	Memory deterioration	0.36**	0.59 **	0.24**	0.38**
GPS3	Joy of life	0.17**	0.29 **	0.38**	0.04
GPS4	Restless	0.24**	0.62 **	0.26**	0.30**
GPS5	Family burden	0.36**	0.65 **	0.45**	0.42**
GPS6	State of health	0.20**	0.58 **	0.30**	0.48**
GPS7	Depression	0.35**	0.79 **	0.40**	0.48**
GPS8	Disappointment	0.39**	0.79 **	0.46**	0.46**
GPS9	Fear	0.33**	0.73 **	0.41**	0.49**
GPS10	Positive attitude	0.37**	0.45 **	0.48**	0.08
GPS11	Bad Temper	0.31**	0.74**	0.32**	0.42**
GS01	Social contact	0.44**	0.37**	0.61*	0.14*
GS02	Family relationship	0.13*	0.27**	0.58 **	0.01
GS03	Friend relationship	0.20**	0.28**	0.62 **	0.13*
GS04	Family support	0.37**	0.32**	0.67 **	0.10
GS05	Other people’s care	0.25**	0.35**	0.74 **	0.16*
GS06	Economic hardship	0.15*	0.40**	0.51 **	0.37**
GS07	Labor status	0.33**	0.48**	0.53 **	0.45**
GS08	Family role	0.43**	0.35**	0.68 **	0.12
DM1	Thirsty and dry mouth	0.15*	0.16*	0.07	0.45 **
DM2	Easily hungry	0.17**	0.20**	0.15*	0.49 **
DM3	Frequent urination	0.25**	0.16**	0.22**	0.49 **
DM4	Numbness or tingling in the hands and feet	0.39**	0.46**	0.25**	0.55 **
DM5	Shaky hands. Heartburn. Sweating	0.40**	0.35**	0.21**	0.60 **
DM6	Diminished visual acuity/Blurred vision	0.38**	0.40**	0.20**	0.56 **
DM7	Skin itching	0.30**	0.32**	0.15*	0.50 **
DM8	Skin infections or wounds	0.30**	0.22**	0.13*	0.47 **
DM9	Long-term treatment troubles	0.22**	0.41**	0.30**	0.59 **
DM10	Eyelid or bilateral lower extremity edema	0.29**	0.26**	0.23**	0.47 **
DM11	Fear of more serious complications	0.13*	0.34**	0.15*	0.64 **
DM12	Inheritance to offspring	0.14*	0.29**	0.14*	0.60 **
DM13	Trouble with diet control	0.25**	0.38**	0.24**	0.68 **
DM14	Troubled by frequent blood glucose testing	0.16**	0.39**	0.24**	0.64 **

Correlations between each item and its designated domain are in bold type

** There was a significant at the level of 0.01. * There was a significant at the level of 0.05

Additionally, the results of the confirmatory factor analysis using SEM of the specific module demonstrated a good fit to the proposed model, χ2/df = 2.12(χ2 = 144.29, df = 68), RMSEA = 0.06, CFI = 0.92, TLI = 0.90, 90%CI ranging from 0.05∼0.08, SRMR = 0.06 (Fig. 2; Table 2). The findings of the aforementioned analyses supported the theoretical construct and demonstrated strong construct validity.

Fig. 2.

The structure of the specific module of QLICD-DM(V2.0) by structural equation modelling

Table 2.

Structure of the specific module of the QLICD-DM(V2.0) confirmed by SEM

Facets	Indicator	Standardized estimates	SE	Est./S.E.	P
Specific symptoms(SPS)	DM1	0.61	0.09	6.45	< 0.001
	DM2	0.79	0.10	7.83	< 0.001
complication symptoms (COS)	DM4	0.77	0.07	10.65	< 0.001
	DM5	0.79	0.07	11.65	< 0.001
	DM6	0.73	0.08	9.16	< 0.001
	DM7	0.57	0.07	7.88	< 0.001
	DM8	0.51	0.07	6.80	< 0.001
disease-related psychology (DRP)	DM10	0.04	0.08	0.46	< 0.001
	DM11	1.18	0.08	14.28	< 0.001
	DM12	1.10	0.09	11.93	< 0.001
treatment-related psychology (TRP)	DM3	0.34	0.07	4.80	< 0.001
	DM9	0.76	0.08	9.07	< 0.001
	DM13	0.84	0.08	10.13	< 0.001
	DM14	0.77	0.09	8.96	< 0.001

SF-36 was selected as the criterion instrument because it is now a very well-established instrument. As shown in Table 3, the correlation between corresponding or conceptually similar domains was generally higher than the correlation between different and dissimilar domains. Since SF-36 does not contain a specific module, the correlation coefficients of corresponding domain are larger than those with other domains, and the specific module domain is showed relatively low correlations with the SF-36 domains.

Table 3.

Correlation coefficients between domains of the QLICD-DM(V2.0) and those of the SF-36

SF-36	QLICD-DM
	PHD	PSD	SOD	SPD
Physical function (PF)	0.62**	0.37**	0.37**	0.19**
Role function (RF)	0.37**	0.38**	0.35**	0.18**
Bodily pain (BP)	0.49**	0.46**	0.25**	0.43**
General health (GH)	0.32**	0.37**	0.33**	0.33**
Vitality (VT)	0.38**	0.40**	0.36**	0.30**
Social function (SF)	0.51**	0.48**	0.47**	0.40**
Role emotional (RE)	0.37**	0.38**	0.35**	0.18**
Mental health (MH)	0.39**	0.51**	0.45**	0.36**

Note: **P < 0.01

Responsiveness

Table 4 shows that there were significant changes in several domains of the scale (p < 0.05) when paired t-tests were conducted before and after treatment, except for the psychological and social domains, which did not show statistically significant changes. The physical function domain showed the highest SRM value (0.357), while the other domains had lower SRM values.

Table 4.

Responsiveness of the quality-of-life instrument QLICD-DM V2.0 (n = 242)

Domain/Facets	Before treatment		After treatment		t	p	SRM
	Mean	Standard deviation	Mean	Standard deviation
Physical function (PHD)	65.86	14.12	70.90	13.68	-5.63	0.000	0.35
Basic physical function (BPF)	56.92	14.74	62.46	14.51	-5.07	0.000	0.37
Independence (IND)	84.71	23.68	87.23	21.74	-1.60	0.111	0.10
Energy and discomfort (EAD)	55.47	23.79	63.29	21.58	-4.63	0.000	0.32
Psychological function (PSD)	65.25	17.88	67.78	17.75	-1.69	0.091	0.14
Cognition (COG)	68.54	22.86	70.42	21.03	-1.16	0.247	0.08
Emotion (EMO)	63.16	19.55	65.99	19.49	-1.45	0.148	0.14
Will and personality(WIP)	69.26	21.81	71.40	21.18	-1.12	0.263	0.09
Social function (SOD)	74.45	15.35	76.11	15.66	-0.95	0.339	0.10
Interpersonal communication (INC)	79.27	16.99	79.83	17.22	0.15	0.880	0.03
Social support (SSS)	72.72	18.73	75.12	18.29	-1.55	0.121	0.12
Social role (SOR)	69.83	22.00	72.01	20.28	-0.75	0.454	0.09
General domain (CGD)	68.07	13.41	71.16	13.54	-3.37	0.001	0.23
Specific domain (SPD)	60.27	16.68	64.89	15.81	-4.66	0.000	0.27
Specific symptoms(SPS)	58.16	23.26	65.79	21.68	-5.00	0.000	0.32
complication symptoms (COS)	72.33	19.58	75.17	17.80	-2.36	0.019	0.14
disease-related psychology (DRP)	50.82	24.42	57.76	23.09	-4.14	0.000	0.28
treatment-related psychology (TRP)	53.35	22.13	56.95	21.69	-2.21	0.028	0.16
Total score	66.14	12.64	69.07	12.73	-4.63	0.000	0.23

Results from the generalizability theory analysis

The results of the G study are shown in Table 5. The bold values along the main diagonal are the variance component values of the domain, and the rest are the correlation coefficients between the domains. The distribution of the variance component values of patients (p) in each domain ranges from 0.218 to 0.431, and the distribution of the variance component of the items (i) ranges from 0.060 to 0.365, all of which were below 0.50, indicating relatively small differences of the various domains and the items are relatively small. The variance components for the patient × item interaction (p*i) has a variance component distribution between 0.766 and 1.064, indicating that the primary source of measurement error originated from the patient–item interaction, and whereas the item-related error variance was relatively small. In addition, the generalizability coefficient (Eρ²) of physical domain, psychological domain, social domain, specific domain under the current item configuration are 0.683,0.842,0.746 and 0.830, respectively. Therefore, the QLICD-DM (V2.0) demonstrated good discriminative ability and measurement accuracy.

Table 5.

Estimation of variance and covariance components for each domain in the p*i designed G study of the QLICD-DM(V2.0)

Source of variation/ Coefficients	PHD	PSD	SOD	SPD
Patients(p)	0.218	0.704	0.674	0.597
	0.216	0.431	0.765	0.676
	0.167	0.267	0.281	0.444
	0.169	0.270	0.143	0.369
Items(i)	0.365
		0.060
			0.244
				0.257
Patients×Items(p*i)	0.912
		0.890
			0.766
				1.064
Generalizability coefficient(Eρ2)	0.683	0.842	0.746	0.830
index of dependability(Ф)	0.606	0.833	0.690	0.797

The elements on the main diagonal are the estimates of the variance components of each effect in the corresponding domains (shown in bold), the elements below the main diagonal are the estimates of the covariance components of the effects in different domains, and the elements above the main diagonal are the correlation coefficients between each domain. PHD, physical domain; PSD, psychological domain; SOD, social domain; SPD, specific domain

The results obtained from the D study are shown in Table 6. It can be seen that at current items the Generalizability coefficient of the total scale is 0.917 and the reliability index is 0.899, the Generalizability coefficients of the domains are between 0.683 and 0.842, which were all greater than 0.70 except for the physical domain(0.683), indicating that the scale is reliable. It can also be observed that, regardless of the domain, as the number of items increases, the generalizability coefficient also increases. If a generalizability coefficient greater than 0.70 is used as the standard, the physical function domain would require one additional item (thus reaching 10 items).

Table 6.

p×i - designed D-study results of the various domains of QLICP-DM(V2.0)

Domain	Number of Items	σ2(P)	σ2(I)	σ2(PI)	σ2(δ)	σ2(Δ)	σ2(XPI)	Eρ2	Φ
Physical domain	7	0.218	0.053	0.130	0.130	0.182	0.053	0.626	0.545
	8	0.218	0.047	0.114	0.114	0.160	0.047	0.657	0.577
	9	0.218	0.042	0.101	0.101	0.141	0.042	0.683	0.606
	10	0.218	0.038	0.091	0.091	0.128	0.038	0.705	0.631
	11	0.218	0.034	0.083	0.083	0.116	0.034	0.724	0.653
Psychological domain	6	0.431	0.012	0.148	0.148	0158	0012	0.744	0.731
	7	0.431	0.011	0.127	0.127	0.136	0.011	0.772	0.761
	8	0.431	0.010	0.111	0.111	0.119	0.010	0.795	0.784
	9	0.431	0.009	0.099	0.099	0.106	0.008	0.814	0.804
	10	0.431	0.008	0.089	0.089	0.095	0.008	0.829	0.820
	11	0.431	0.008	0.081	0. 081	0.086	0.008	0.842	0.833
	12	0.431	0.007	0.074	0.074	0.079	0.007	0.853	0.845
	13	0.431	0.007	0.068	0.068	0.073	0.007	0.863	0.855
Social domain	6	0.281	0.042	0.128	0.128	0.168	0.042	0.688	0.626
	7	0.281	0.036	0.109	0.109	0.144	0.036	0.720	0.661
	8	0.281	0.032	0.096	0.096	0.126	0.032	0.746	0.690
	9	0.281	0.029	0.085	0.085	0.112	0.029	0.768	0.715
	10	0.281	0.026	0.077	0.077	0.101	0.026	0.786	0.736
Specific domain	9	0.369	0.031	0.118	0.118	0.147	0.031	0.758	0.716
	10	0.369	0.028	0.106	0.106	0.132	0.028	0.776	0.737
	11	0.369	0.025	0.097	0.097	0.120	0.025	0.793	0.755
	12	0.369	0.023	0.089	0.089	0.110	0.023	0.806	0.770
	13	0.369	0.022	0.082	0.082	0.102	0.022	0.819	0.784
	14	0.369	0.020	0.076	0.076	0.094	0.020	0.830	0.797
	15	0.369	0.019	0.071	0.071	0.088	0.019	0.839	0.807
	16	0.369	0.018	0.066	0.066	0.083	0.018	0.847	0.817

Bold values represent results for the current number of items in the domain

σ²(δ), the variance components of relative error; σ²(Δ), the variance components of absolute error; σ²(X_PI), the variance components of error when estimating the universe score by using sample mean; Eρ², the Generalizability coefficient, Φ, the index of dependability

Discussion

This study is a component of the Quality of Life Instruments system for Chronic Diseases (QLICD), which was developed using a combination of a generic module (QLICD-GM), applicable to various chronic diseases and widely accepted, together with disease-specific modules. This modular approach integrates all disease-specific QLICD instruments under a unified generic module QLICD-GM. On the basis of QLICD-GM(V2.0), the QLICD-DM(V2.0) is specially designed for DM patients and can capture disease-specific symptoms, complications, and psychological characteristics, complications and psychological characteristics of DM patients in a targeted manner. Therefore, contrast to other QOL instruments for DM such as DCCT, DQLCTQ, ADDQOL, Diabetes-39(D-39), DTSQ, ITR-QOL, Diabetes Distress Scale(DDS) [8–14], the QLICD-DM (V2.0) possesses the characteristics of both generic and disease-specific instruments. It can utilize the general module to compare the quality of life across different diseases through QLICD-GM, and it can also use the specific module to more accurately assess disease-specific symptoms, treatment side effects, and related psychological and social issues.

The use of objective clinical indicators to evaluate disease severity and treatment effectiveness has become standard clinical practice. However, patients’ subjective experiences and the impact of disease on daily life are often overlooked. Therefore, QLICD-DM(V2.0) combines four domains: physical, psychological, social, and disease-specific aspects. These domains reflect patients’ subjective perceptions of the disease and provide a more comprehensive assessment of quality of life.

In order to evaluate the reliability of QLICD-DM(V2.0), this study evaluated internal consistency and test–retest reliability. A Cronbach’s α coefficient above 0.70 is generally considered indicative of acceptable internal consistency [25–27]. The physical domain’s Cronbach coefficient is 0.683, slightly below the recommended threshold of 0.70, but Cronbach coefficient and split-half reliability of the total scale are greater than 0.70 (0.90, 0.74 respectively). The fundamental assumption of test-retest reliability is that the construct being measured remains stable over time [34]. If the patient’s health status has not significantly changed over time, the scores for diabetics will not be statistically significant, as observed in the present study (all p > 0.05). Therefore, the correlation coefficients for each domain were calculated for the data collected at two time points, with a test-retest reliability coefficient greater than 0.80 (the exception of the physical and social domains 0.65 and 0.64, respectively), indicating acceptable test–retest reliability [35].

However, it must also be noted that the internal consistency and test-retest reliability of the physical function domain, as well as the test-retest reliability of social function, did not meet the ideal standards (0.7, 0.8, respectively). This is very rare in our QOL scale system research, and the underlying reasons require further in-depth investigation. It is possible that the medical condition of diabetic inpatients underwent significant changes after admission (which strictly speaking violates the assumption of test-retest), or there may be differential item functioning due to varying interpretations of the meaning of physical function items among diabetic patients. Subsequent studies will conduct in-depth analysis to explore potential causes, and it may even be necessary to optimize the items.

Validity refers to the extent to which an instrument accurately measures the constructs it is intended to assess, as well as the degree of measurement, i.e., to what extent the indicators or observations are accurately reflected objectively and truthfully. It also reflects the extent of measurement bias and systematic error [36]. Good construct, content, and criterion-related validity were demonstrated by the QLICD-DM(V2.0). The four domains provide a more comprehensive representation of quality of life, and the items are concise and easy for patients to understand. Item-domain correlation analysis and SEM were used to examine the construct validity of the scale [37]. The findings were generally consistent with the theoretical framework, and the four latent factors effectively reflected the multidimensional problems experienced by patients with diabetes as well as the mental and physical problems they produce.

However, it should be noted that individual path coefficients in the structural equation model exceed 1 (e.g., the coefficients for DRP with DM11 and DM12). Possible reasons may include high correlation between DM11 and DM12 (0.83), indicating potential multicollinearity to some extent, as well as relatively small sample size leading to less stable estimated coefficients. Subsequent research will further investigate the underlying causes and implement appropriate revisions, including potential item optimization or merging.

The Chinese version of the SF-36, which has strong reliability and validity and is thought to be a potential QOL assessment tool for a variety of applications, was used as the criterion instrument for criterion-related validity [5, 38].

Responsiveness refers to the ability of a scale to detect changes over time, such as those brought about by a course of therapy. According to the assessment’s findings, patients’ conditions may have remained relatively stable over a brief period of time and did not change substantially in several categories, such as their psychological and social functioning, where scores were not statistically different. Previous studies have suggested, the responsiveness is low at around 0.2, moderate at about 0.5, and large responsiveness above 0.8 [29, 38]. The SRM values for all of the domains in this study were lower than the 0.5, which may be associated with the relatively short hospitalization period despite the chronic nature of diabetes, and the relatively low SRM values for the facets, which may have also contributed to the low SRM values for each domain. It is also possible that certain domains or facets do not undergo changes during the hospitalization period themselves (e.g., in terms of social support and social security). In addition, there may be issues with the item formulation that require optimization. Subsequent research will further explore potential underlying reasons to address this issue.

In addition to classical test theory analysis, this study also applies generalizability theory. G-studies can decompose multiple sources of measurement error according to the number of items for the current scale, whereas D-studies can evaluate the effects of changing the number of items that meet the reliability requirements [23, 38, 39]. This study evaluated both the G-coefficient and Φ coefficient but also introduced their changes under different item configurations. For psychological domain, social domain and the specific module the currently designed G coefficients are 0.842, 0.746 and 0.830, respectively. It can be considered that it meets the 0.70 standards. For the physical domain, the current design G coefficient is 0.683, which is lower than but close to the acceptable 0.70. Therefore, the items of this domain can be improved if needed. For an alternative design with 10 items, the G coefficient is estimated to be 0.705, which will satisfy acceptable reliability. Conversely, psychological function, social function, and specific modules may potentially reduce the number of items if needed, reducing them to 6, 7, and 9 respectively, all of which can achieve high reliability. This finding may provide a basis for future scale simplification.

This study also has several limitations that should be taken into account. The subjects in this study were selected from the inpatient population at hospitals, which has severe conditions and symptoms than the outpatients. The exclusive use of inpatients limits external validity. Further studies are needed to assess the generalizability of the instrument to other settings and populations such as outpatients at a local clinic.

In addition, the sample size was relatively small, which may also affect the results related to SEM. Other problems include the fact that the sample size of the survey depends on a number of variables and that quality of life is a multidimensional outcome, which makes it possible that the results are limited representativeness. In the future, the sample size should be increased to include patients with other forms of the disease and different settings.

In summary, the QLICD-DM (V2.0) can be used not only to evaluate patients with diabetes mellitus, but also by doctors to evaluate and select appropriate treatment strategies. The scale has been tested to have strong reliability, validity, and acceptable responsiveness. It provides a reliable tool for the evaluation of the quality of life of diabetes patients as a tool to gather reliable and valid data, which can enhance the quality of relevant research and evaluation.

Author contributions

YL, CW, RC designed the study. CZ, QZ, XZ, RG performed the data collection. YL, CZ, CW performed data analyses and drafted the manuscript. CW and RC revised the manuscript deeply. All authors contributed to interpreting the data, and have read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (71373058, 30860248).

Data availability

The data sets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol and the informed consent form were approved by the IRB (institutional review board) of the affiliated hospital of Guangdong Medical University (PJ2013037). The respondents were voluntary and provided written consent for participation. The Declaration of Helsinki’s ethical guidelines were followed in the study. All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.