A comparative analysis of structural equation modeling and traditional regression analysis in investigating factors affecting health-related quality of life in general Thai population

Abstract

Background

Previous studies identified individual characteristics as factors affecting health-related quality of life (HRQoL) scores. Nevertheless, most studies employed univariate or multiple regression analyses, which have limitations in handling confounding variables and measurement errors across multiple dependent variables. This study therefore aimed to assess psychometric properties, investigate factors affecting HRQoL using the World Health Organization Quality of Life — Brief Version (WHOQOL-BREF) and the EuroQol five-dimension, five-level questionnaire (EQ-5D-5L) among general Thai population using structural equation modelling (SEM) and to compare the results with the traditional regression analyses.

Methods

The study utilized secondary data from the 2023 Thai population norms survey. Face-to-face interviews were conducted with 2,000 adults. However, SEM analysis was performed on data from 1,927 participants, after excluding 73 individuals to meet the assumption of normality. The study adapted Ferran’s model to fit the available data and evaluated measurement properties (i.e., internal consistency and convergent and discriminant validity). It also employed partial least squares SEM (PLS-SEM) to examine structural relationships and model properties through predictive relevance (Q²), effect size (f²), and goodness of fitness (GOF).

Results

Age significantly impacted WHOQOL-BREF scores (β = 0.264, f² = 0.037). For EQ-5D-5L, age (β = 0.304, f² = 0.071), occupation (β = 0.422, f² = 0.036), and drug (β = 0.288, f² = 0.052) were key factors. However, traditional regression yielded different results for drug and occupation factors on WHOQOL-BREF and MCS scores. General health perception was the strongest predictor of HRQoL for both models exhibiting acceptable reliability and validity. For most dependent variables explained by the model, predictive power was medium to large, except for the mental component summary, which displayed a small predictive value (Q² = 0.115). Both models demonstrated a high fit (GOF: 0.539 and 0.521 for WHOQOL-BREF and EQ-5D-5L, respectively).

Conclusions

Different statistical approaches yield varying results as SEM indicates that age, occupation, and drug significantly influence HRQoL scores, whereas traditional regression finds no significant effects for drug and occupation. Nevertheless, these findings can guide policymakers in allocating resources to targeting population groups with low levels of HRQoL.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12955-025-02450-3.

Introduction

Health-related quality of life (HRQoL) is a patient-reported outcome that reflects overall health without clinical interpretation from health professionals [1, 2]. It has become a key measure in population health surveys and in monitoring chronic conditions such as end-stage renal disease [3] and diabetes [4]. Furthermore, HRQoL can be expressed in terms of utility values, which are used to calculate quality-adjusted life years, a key outcome in cost–utility analysis, as recommended by health technology assessment guidelines worldwide [5].

HRQoL emphasizes the impact of health conditions on overall quality of life (QoL) [6]. Wilson and Cleary generated a conceptual framework of HRQoL and linked several factors to HRQoL, including biological function, symptoms, functional status, general health perception (GHP), non-medical factors, and characteristics of individuals and environments [7]. However, Ferrans et al. refined this model by omitting non-medical factors and reclassifying them under personal or environmental characteristics. This revised model enabled the generation of broader hypotheses beyond clinical manifestations, which renders it suitable for real-world settings. Hence, a systematic review recommended Ferran’s model for identifying causal relationships and guiding targeted health care interventions [8, 9].

To assess HRQoL, multidimensional instruments are typically applied across general and clinical populations [10], thus capturing physical, psychological, and social health domains. Generic instruments are commonly adopted to enable comparison across populations, although they may lack sensitivity to specific clinical changes. These instruments report health profiles across health dimensions or as a single index (utility) score [11, 12]. The utility score can be used to calculate quality-adjusted life years for economic analyses [13]. Notably, the utility score is specifically derived from health preference-based instrument, which is one of the generic instruments for measuring HRQoL. Therefore, generic instruments are frequently used in conducting large-scale health surveys [10].

HRQoL is increasingly used in health monitoring for the general population and patient groups in Thailand. Previous studies typically applied generic instruments such as the EuroQol five-dimension, five-level questionnaire (EQ-5D-5L) for the general population [14] and patients with chronic hepatitis B [15], the World Health Organization QoL — Brief Version (WHOQOL-BREF) for patients taking warfarin [16], and the Short Form Health Survey-version 2 (SF-12v2) for patients with HIV [17]. These studies also identified various sociodemographic factors that influence HRQoL levels, including sex, age, body mass index (BMI), number of comorbidities (NOC), level of education, and occupation, among others. Univariate and multiple regression analyses were employed to explore these associations. However, both approaches pose certain limitations. Univariate analysis does not control for confounding variables, while multiple regression analysis assumes simple linear relationships between one observed dependent and multiple independent variables. In addition, multiple regression analysis may increase measurement error if a separate regression analysis on multiple observed dependent variables is conducted, and it can overlook underlying latent variables [18].

Structural equation modeling (SEM) offers a robust statistical approach in various fields of study [19–21]. It enables a simultaneous analysis of complex relationships among multiple observed and latent variables, thus accounting for measurement error. Moreover, it is used for conducting confirmatory factor analysis and examining the construct validity and reliability of instruments. Therefore, SEM is beneficial for testing theoretical models and overcoming the limitations of multiple regression approaches [18].

SEM has been used across various disciplines such as organization performance [22], transportation [23], patient care [24], and suicidal behaviors [25]. SEM has also been recognized as a valuable analytical approach for studying HRQoL due to its ability to manage the complexity of interrelated factors and latent constructs inherent in HRQoL measures. Several studies have employed SEM to examine various factors affecting HRQoL and to capture its multidimensional structure of HRQoL more comprehensively among patients with rheumatoid arthritis [26], Parkinson’s disease [27], coronary heart disease [28], and in special populations such as youths with substance use disorders [29] and both younger and older adults [30]. Given this growing body of evidence, SEM allows researchers to simultaneously examine complex relationships among dependent and independent variables, yielding more accurate estimations of the theoretical concept interest [18]. For instance, previous studies on rheumatoid arthritis in China, individuals with diabetes in South-Korea, colorectal cancer and chronic obstructive pulmonary disease (COPD) patients in Thailand demonstrated that SEM could simultaneously examine and adequately capture both direct and indirect effects of factors affecting HRQoL [24, 26, 31, 32]. Furthermore, SEM provides concrete evidence on whether empirical data fit the hypothesized model through multiple model fit indices, which were not offered by traditional regression analysis [33–36]. Therefore, SEM is considered a preferred method for the multidimensional structure of HRQoL in health outcomes research [37]. Our use of SEM in the present study aligns with this methodological trend and aims to provide deeper insights into the structural relationships among factors influencing HRQoL, which has not been done in Thailand.

According to Ferran’s model, HRQoL consists of five core elements: biological function, symptoms, functional status, GHP, and the characteristics of individuals and their environments. Previous studies have demonstrated causal relationships between individual characteristics, functional status, and HRQoL levels. Multimorbidity, classified as a biological function in Ferran’s model, has been directly associated with increased functional limitations and decreased HRQoL levels among middle-aged and older adults in Europe [38] and China [39], as well as among general community-dwelling populations [40]. Some previous Thai studies have found that female gender, older age, obesity, and poor health perception are associated with decreased HRQoL in the general Thai population [14, 41]. Taken together, these patterns of direct association provide a theoretically grounded rationale for applying Ferran’s model to explore the relationships among these factors in the general Thai population. While scholars recommend the use of Ferran’s model on patient populations [6], it is also applicable to other populations such as children, community-dwelling elderly, and healthy women [9]. However, no study has applied SEM to examine relationships among sociodemographic factors and their latent variables using Ferran’s model within the general Thai population. Thus, the current study aimed to (a) explore pathways that link biological factors, functional status, and GHP to HRQoL using Ferran’s model, (b) assess the construct validity and reliability of HRQoL instruments in the current context, and (c) identify individual characteristics associated with HRQoL using SEM, and compare the results with traditional regression analysis.

Methods

Study design and participants

This study used data from a project entitled “EQ-5D-3L and EQ-5D-5L population norms for Thailand [14]” in which a cross-sectional survey was conducted via face-to-face interviews with 2,000 Thai adults. The study used a four-stage stratified random sampling to select provinces, districts, subdistricts, and villages for data collection. Details of the selected provinces were explained elsewhere [14]. For each stratum, participants were selected based on age bands, sex, and area of residence. The inclusion criteria were: age 18 years, fluency in Thai, and the ability to complete the data collection process. Exclusion criteria included acute or life-threatening conditions or signs of cognitive impairment based on their self-reported medical history and the researcher’s (KK) or well-trained interviewer’s observation whether the participant could comprehend and respond to questions appropriately during the interviews. A quota sampling method was used to select participants in proportion to age, sex, and area of residence (metropolitan and rural), ensuring representation of the general Thai population. All eligible individuals from the convenient sample were invited to participate in face-to-face interviews with well-trained interviewers, which were conducted at participants’ homes by trained interviewers between May and June 2023. All well-trained interviewers completed at least a bachelor’s degree in public health or management sciences with more than 5 years of professional experience in conducting field-based survey research. Prior to data collection, the researcher (KK) provided specific training to all interviewers on how to perform face-to-face interviews using a paper and pencil approach to record participants’ responses in a consistent manner to minimize interviewer bias.

Conceptual model

This study employed and adapted Ferran’s model to structure the relationships among variables and to investigate factors influencing HRQoL. As shown in Fig. 1a, the Ferran’s model links biological factors, symptoms, functional status, and GHP to overall HRQoL, while also incorporating characteristics of the individual and the environment that may influence each component altogether. In this study, we adapted Ferran’s model to fit with the collected data as follows: (1) symptoms of the Ferran’s model have not been used (2), the functional status was partially modified by treating the physical component summary (PCS) and mental component summary (MCS) as simultaneous measures of functional status in order to examine their mediating effects on GHP (3), the characteristics of the individual influencing GHP were not investigated, and (4) the characteristics of the environment have not been used (see Fig. 1b). The associations were also examined between each element of Ferran’s model and participant characteristics including age, BMI, number of daily medications (drugs), sex, and occupation (see Fig. 1b). Each component of the adapted Ferran’s model with the equivalent variables was measured using standardized instruments in the Thai language in which they have been validated among the general Thai population, SF-12v2 [42], WHOQOL-BREF [43], and EQ-5D-5L [44]. Notably, two conceptual models were employed to investigate factors associated with HRQoL as measured by the WHOQOL-BREF and EQ-5D-5L questionnaires, respectively, because they can be used to report different types of HRQoL scores (see Fig. 2). The WHOQOL-BREF provides domain-specific health profile scores across its four domains, whereas the EQ-5D-5L generates a single index-based health utility score.

Fig. 1.

Conceptual model for health-related quality of life (a) Ferran’s model, (b) adapted Ferran’s model with variables used

Fig. 2.

Conceptual model for health-related quality of life scores for WHOQOL-BREF and EQ-5D-5L

Instruments

As shown in Fig. 1b, each component of the model with the equivalent variables was measured using validated instruments in the Thai population and Thai language. These components were considered as endogenous variables in the SEM, including NOC, physical component summary (PCS), mental component summary (MCS), general health perception (GHP), WHOQOL, and EQ-5D-5L.

Biological function

In the Ferran’s model, biological function was assessed using laboratory tests, physical assessment, and medical diagnoses [6]. However, in the present study, the NOC was used to represent the biological function component because the study was conducted outside of clinical settings. Therefore, data were self-reported rather than obtained from medical records.

Functional status

The SF-12v2 was used to assess functional status and well-being [45]. The tool comprises 12 items on eight domains, namely, physical functioning (PF: 2 items), role limitations due to physical problems (RP: 2 items), bodily pain (BP: 1 item), general health perception (GH: 1 item), vitality (VT: 1 item), social functioning (SF: 1 item), role limitations due to emotional problems (RE: 2 items), and mental health (MH: 2 items). The participants rated their HRQoL during the past four weeks (recall period) using a three- or five-point Likert-type scale, depending on the subscale. Furthermore, the 12 items were further categorized under two summary components, namely, PCS (derived from PF, RP, BP, and GH) and MCS (derived from VT, SF, RE, and MH). As the SF-12v2 is a validated instrument for assessing HRQoL and functional status, PCS and MCS scores were utilized as indicators of physical and mental aspects of functional status. This approach aligns with Ferran’s model, which distinguishes functional status from GHP and overall QoL. It also allowed for analysis of how functional status mediates the relationship between biological factors and GHP.

GHP

We extracted relevant items from the WHOQOL-BREF reflecting subjective perceptions of health satisfaction (item 1) and overall QoL (item 26). This approach aligns with Ferran’s model, in which GHP represents an individual’s subjective perception of their general health status. Each item was rated using a five-point Likert-type scale, with a two-week recall period.

HRQoL

In this study, the HRQoL was measured using two generic instruments including WHOQOL-BREF and EQ-5D-5L. Details of each generic instrument are explained below.

WHOQOL-BREF

The Thai version of the WHOQOL-BREF has 26 items out of which 24 are categorized under four domains, namely, physical health (7 items), psychological health (6 items), social relationships (3 items), and environment (8 items). For each item, the participants rated their health status during the two-week recall period using a five-point Likert-type scale (1 = not at all, 5 = completely). WHOQOL-BREF domain scores were calculated based on the items corresponding to each respective domain. Subsequently, ordinal-to-interval conversion tables were applied to transform the ordinal-scale scores into interval-scale scores for each WHOQOL-BREF domain [43]. Each dimension score is calculated by summing the item scores within the respective domain, with higher scores indicating better HRQoL.

EQ-5D-5L

The EQ-5D-5L is the health preference-based instrument employed to elicit the utility score. It is composed of five items; each item examines one of the following dimensions: mobility (MO), self-care (SC), usual activities (UA), pain/discomfort (PD), and anxiety/depression (AD). The respondents rated their current health status through these five dimensions using a five-point scale (1 = no problem, 5 = extreme problem/unable to perform). A Thai-specific algorithm was used to elicit utility scores from the responses to the EQ-5D-5L questionnaire, in which the Thai utility score ranged from 0.4212 to 1.000 [46]. Higher utility scores also indicate better HRQoL, where a score of 1.00 represents perfect health and 0.00 represents the worst possible health state or death. Notably, negative utility scores indicate health states considered worse than death.

Participant characteristics

A range of demographic variables was collected from each participant, including sex (male/female), age, occupation (employed/unemployed), weight and height (used to calculate BMI), and number of daily medications (drug). These factors have been identified in previous studies conducted in Thailand as influential determinants of HRQoL.

Data collection

Data collection was conducted between May and June 2023. Eligible participants were given an information sheet explaining the title, objectives, data collection process, and benefits of the study, including the confidentiality of participants’ data. Furthermore, they were informed of their rights to refuse to participate or withdraw from study at any time for any reason. Afterward, they provided written informed consent and were interviewed by well-trained interviewers without interpretive assistance. The Institutional Review Board at Burapha University (IRB1-031/2566) provided ethical approval following the Declaration of Helsinki.

Data analyses

The study used descriptive statistics to summarize the participant demographics. Data distribution was assessed using kurtosis and skewness coefficients (acceptable range from 2.00 to 2.00) [47]. Outliers were omitted using the Mahalanobis distance method to ensure normal distribution across variables [48]. Chi-square or independent sample t-test was used to compare between full and SEM samples. Prior to SEM, one variable (drug) exhibited kurtosis and skewness values that exceeded ± 2; therefore, 73 participants were excluded from the dataset to achieve normal distribution for the drug variable. SEM was subsequently performed on 1,927 participants (SEM participants).

According to a priori sample size estimation for SEM, the required sample size depends on the number of observed and latent variables in the model, as well as the anticipated effect size, desired statistical power, and significance level [49]. In this study, the model included 4 latent variables (PCS, MCS, GHP, WHOQOL-BREF) and 25 observed variables (12 items from the SF-12v2 for PCS and MCS, Item 1 and Item 26 from the WHOQOL-BREF for GHP, four domains of the WHOQOL-BREF, NOC, Age, BMI, Drug, Sex, Occupation and utility score from EQ-5D-5L) as shown in Tables 2 and 3. We set the expected effect size at 0.3, statistical power at 0.8, and significance level at 0.05. Based on these parameters, the minimum required sample size was 241 participants. Our final sample consisted of 1,927 participants, which exceeds this requirement and supports the adequacy of the sample size for the analyses conducted, including path analysis involving multiple observed variables.

Table 2.

Reliability and convergent validity analysis of first-order constructs for WHOQOL-BREF as the dependent variable

Latent variables	Items	Factor loading	t-value	AVE	CR	CA
	(Observed variables)	> 0.7		> 0.5	> 0.7	> 0.7
Physical component summary (PCS)	PF02	0.845	103.351	0.682	0.928	0.906
	PF04	0.839	105.472
	RP02	0.866	105.143
	RP03	0.872	109.22
	BP02	0.805	74.694
	GH01	0.718	70.345
Mental component summary (MCS)	RE02	0.839	79.539	0.535	0.871	0.817
	RE03	0.830	74.415
	MH03	0.568	24.518
	MH04	0.805	69.200
	VT02	0.572	28.354
	SF02	0.722	50.023
GHP	WHO1	0.843	104.712	0.684	0.812	0.538
	WHO26	0.811	85.383
Health-related quality of life (WHOQOL-BREF)	Physical	0.861	141.914	0.714	0.909	0.866
	Psychological	0.885	171.491
	Social	0.782	68.798
	Environment	0.849	127.509

All loadings are significant, AVE = Average Variance Extracted, CR = Composite Reliability, CA = Cronbach’s alpha, GHP = General health perception

Table 3.

Reliability and convergent validity analysis of first-order constructs for EQ-5D-5L as the dependent variable

Latent variables	Items	Factor loading	t-value	AVE	CR	CA
	(Observed variables)	> 0.7		> 0.5	> 0.7	> 0.7
Physical component summary (PCS)	PF02	0.846	104.114	0.682	0.928	0.906
	PF04	0.840	106.303
	RP02	0.865	104.225
	RP03	0.871	108.225
	BP02	0.805	74.041
	GH01	0.718	70.818
Mental component summary (MCS)	RE02	0.842	81.073	0.536	0.871	0.817
	RE03	0.833	75.848
	MH03	0.561	23.810
	MH04	0.808	70.019
	VT02	0.566	27.409
	SF02	0.724	50.287
GHP	WHO1	0.901	128.084	0.676	0.805	0.538
	WHO26	0.735	43.046

All loadings are significant, AVE = Average Variance Extracted, CR = Composite Reliability, CA = Cronbach’s alpha, GHP = General health perception

Partial least squares SEM (PLS-SEM) was used due to the complex model structure and the limitations of the secondary dataset, which lacked some information of certain parts of Ferran’s model, particularly the biological function, as we did not collect the data related to objective assessments. The NOC was the only one aspect considered to represent the biological function in this study [50] (see Fig. 1a and b). Confirmatory factor analysis was performed to verify if any items reached factor loadings of 0.70. If this is the case, then they were retained in the model, and we consider other indices, such as Cronbach’s alpha (CA), composite reliability (CR), and average variance extracted (AVE), if they met the acceptable threshold. Internal consistency was assessed using CA and CR (acceptable value: ≥0.70). Convergent validity was evaluated using AVE (acceptable value: ≥0.50), and discriminant validity was confirmed if intervariable correlations reached less than the square root of AVE [51]. If satisfactory discriminant validity was observed, then the study inferred that no multicollinearity existed among the independent variables [52]. Variance inflation factor (VIF) was also examined to confirm the absence of multicollinearity among the independent variables. A VIF value of less than 5 was considered acceptable [53, 54].

The structural model was evaluated using several criteria: the coefficient of determination for the endogenous variable (R²), effect size (f²), predictive relevance (Q²), and goodness of fitness (GOF). R² values greater than 0.25, 0.50, and 0.75 indicate small, medium, and large effects, respectively [55]. Similarly, f² and Q² values above 0.02, 0.15, and 0.35 are considered small, medium, and large effects, respectively [56–58]. GOF was calculated using the square root of the mean AVE multiplied by the mean R². Values more than 0.10, 0.25, and 0.36 signify low, medium, and high model fit, respectively [59]. The path coefficients, structural model, significant values, and all indices were calculated via PLS algorithm and 5,000 bootstrapping samples [54]. The study hypothesized that female, older, and unemployed participants and those with higher BMI, number of daily medications, or comorbidities would report low levels of HRQoL (WHOQOL-BREF and utility scores) compared with their counterparts. To test the robustness of the adapted Ferran’s model, we also explored the correlations between the individual characteristics and GHP to see if there were any significantly different findings with the model without the correlations between those parameters.

Multiple linear regression was performed to investigate the associations between sociodemographic factors and HRQoL scores from WHOQOL-BREF and EQ-5D-5L, and the PCS and MCS scores from the SF-12v2 in order to compare the results with those from SEM. It was also performed to investigate the mediation effect of PCS and MCS for the relationship between NOC and GHP.

SEM was conducted using SmartPLS 4 Professional-Free Trial (v.4.1.1.2, SmartPLS GmbH, Bönningstedt, Germany) [60], while descriptive statistics were performed via IBM SPSS version 23 (IBM Corporation, Armonk, NY, USA).

Results

Participant characteristics

Table 1 depicts the characteristics of the total participants (n = 2000) and those for SEM (n = 1927). For SEM participants, the majority were women (52.3%), employed (79.9%), without reported comorbidities (65.4%), and without daily medications (66.2%). Mean age and BMI were 45.5 (SD = 16.57) and 22.75 (SD = 3.50), respectively. Participant characteristics also demonstrate that the same trend was found for the total samples, and no significant difference was observed between the two groups of samples.

Table 1.

Demographic characteristics of the participants

Characteristics	SEM (n = 1927)	Total (n = 2000)	P-value
Sex, n (%)
Male	920 (47.7)	940 (47.0)	0.641
Female	1007 (52.3)	1060 (53.0)
Age
Mean ± SD	45.5 (16.57)	46.16 (16.86)	0.208
Occupation, n(%)
Employed	1540 (79.9)	1571 (78.6)	0.291
Unemployed	387 (20.1)	429 (21.5)
Number of daily taken medicines
No	1275 (66.2)	1283 (64.2)	0.072
1–3	474 (24.60)	488 (24.4)
4	178 (9.24)	229 (11.5)
Number of Comorbidities
No comorbidities	1260 (65.4)	1267 (63.4)	0.061
1–2 comorbidities	580 (30.10)	610 (30.5)
3–5 comorbidities	87 (4.51)	123 (6.2)
Body mass index (BMI)
Mean ± SD	22.75 (3.50)	22.82 (3.60)	0.557

Evaluation of the measurement model

Table 2 presents the reliability and convergent validity of the first construct for the WHOQOL-BREF model. CA and CR indicated that the reliability of each construct ranged from 0.538 to 0.906 and from 0.812 to 0.928, respectively. AVE, which determines convergent validity for each construct, surpassed the threshold of 0.50 (0.5350.714), while desired external factor loadings exceeded 0.70 [54], except for MH03 and VT02 (0.568 and 0.572, respectively).

Table 3 also displays the reliability of the first construct for the EQ-5D-5L model. Each construct obtained CA and CR values that surpassed the threshold (0.7), except for GHP (CA = 0.538). AVE was greater than the threshold of 0.50 (0.5360.682). Similar to the WHOQOL-BREF model, the majority of items reached desired external factor loadings greater than 0.7 [54], except for MH03 and VT02 (0.561 and 0.566, respectively). The WHOQOL-BREF and EQ-5D-5L models demonstrated that all items were statistically significant (all p-values < 0.05).

Table 4 presents the results for discriminant validity using the Fornell–Larcker criterion for the WHOQOL-BREF and EQ-5D-5L models. The results indicated that the correlation coefficients between two latent variables (constructs) were less than the square root of the AVE for each latent variable and the VIF values were less than 5, which indicates satisfactory discriminant validity and the absence of multicollinearity among the independent variables for both models.

Table 4.

Discriminant validity: Fornell-Larcker criterion for WHOQOL-BREF and EQ-5D-5L as the dependent variables

Construct	Age	BMI	Drug	GHP	MCS	NOC	Occupation	PCS	Sex	WHOQOL	Utility
Age	1.000
BMI	0.232	1.000
Drug	0.522	0.188	1.000
GHP	-0.381	-0.180	-0.328	0.827
MCS	-0.277	-0.111	-0.316	0.505	0.731
NOC	0.583	0.218	0.832	-0.389	-0.380	1.000
Occupation	0.099	-0.129	0.206	-0.104	-0.120	0.215	1.000
PCS	-0.573	-0.137	-0.536	0.500	0.623	-0.592	-0.225	0.826
Sex	0.053	-0.053	0.055	-0.054	-0.038	0.072	0.095	-0.079	1.000
WHOQOL	-0.422	-0.181	-0.315	0.746	0.580	-0.388	-0.098	0.527	-0.056	0.845
Utility	-0.494	-0.108	-0.476	0.497	0.545	-0.556	-0.257	0.710	-0.065	N/A	1.000

N/A: Non-assessment, GHP = General health perception, NOC = Number of comorbidities, bolded values = the square root of average variance extracted (AVE)

Evaluation of the structural model

Structural models were evaluated using adjusted R², f², Q², and model fit (GOF). Table 5 illustrates that the adjusted R² values for the endogenous latent variables ranged from 0.149 to 0.726 and from 0.150 to 0.726 for the WHOQOL-BREF and EQ-5D-5L models, respectively.

Table 5.

Goodness of fit (GOF) results for both WHOQOL-BREF and EQ-5D-5L as the dependent variables

Construct	WHOQOL-BREF		EQ-5D-5 L
	AVE	Adjusted R2	AVE	Adjusted R2
Age
BMI
Drug
GHP	0.684	0.322	0.676	0.336
MCS	0.535	0.149	0.536	0.150
NOC		0.726		0.726
Occupation
PCS	0.682	0.443	0.682	0.443
Sex
WHOQOL	0.714	0.578	N/A	N/A
Utility	N/A	N/A	1.000	0.414
Average value	0.654	0.444	0.631	0.414
AVE x R 2	0.290		0.271
GOF	0.539		0.521

N/A: Non-assessment, AVE: Average variance extracted, GOF: Goodness-of-fit, GHP = General health perception, NOC = Number of comorbidities

The effect size (f²) of the endogenous construct for both models displayed similar results. For the WHOQOL-BREF model, the highest value was that of drug on NOC (1.276) followed by that of GHP on WHOQOL (0.903) and age on PCS (0.139). However, no effect size (f² = 0) was detected for the following pairs of causal relationships: drug/WHOQOL, occupation/WHOQOL, sex/WHOQOL, BMI/PCS, drug/MCS, and sex/MCS (Supplementary material 1). For the EQ-5D-5L model, the highest value was that of drug on NOC (1.276) followed by age on PCS (0.139) and GHP on utility (0.135), respectively. No effect size (f² = 0) was also observed in multiple pairs of causal relationships: sex/utility, BMI/PCS, drug/MCS, and sex/MCS (Supplementary material 2).

As shown in Supplementary material 3, predictive relevance (Q²-values) obtained the highest and lowest values for NOC (0.722) and MCS (0.115), respectively, for both models.

Table 5 also displays the model fit (GOF) for both models. The GOF values for the WHOQOL-BREF and EQ-5D-5L models were 0.539 and 0.521, respectively, indicating high model fit for both models.

Path analysis and hypothesis testing

Table 6 presents the total effects and hypothesis testing for the WHOQOL-BREF model. The majority of the hypotheses were found to be statistically significant, except for those concerning the correlations between sex and all endogenous constructs, as well as between BMI and PCS, all of which had p-values greater than 0.05.

Table 6.

Total effects and hypothesis testing for the WHOQOL-BREF as the dependent variable

Path	Original sample (O)	T statistics (|O/STDEV|)	P values
Age ->GHP	-0.163	10.972	0.000
Age ->MCS	-0.145	5.539	0.000
Age ->NOC	0.195	12.187	0.000
Age ->PCS	-0.404	19.252	0.000
Age ->WHOQOL	-0.264	12.714	0.000
BMI ->GHP	-0.022	1.978	0.048
BMI ->MCS	-0.046	1.973	0.049
BMI ->NOC	0.047	3.024	0.003
BMI ->PCS	-0.004	0.196	0.845
BMI ->WHOQOL	-0.039	2.187	0.029
Drug ->GHP	-0.232	13.631	0.000
Drug ->MCS	-0.217	8.665	0.000
Drug ->NOC	0.709	39.925	0.000
Drug ->PCS	-0.298	12.559	0.000
Drug ->WHOQOL	-0.162	7.874	0.000
GHP ->WHOQOL	0.680	52.568	0.000
MCS ->GHP	0.313	12.574	0.000
MCS ->WHOQOL	0.213	11.486	0.000
NOC ->GHP	-0.306	11.907	0.000
NOC ->MCS	-0.329	7.618	0.000
NOC ->PCS	-0.294	7.924	0.000
NOC ->WHOQOL	-0.208	11.468	0.000
Occupation ->GHP	-0.137	4.781	0.000
Occupation ->MCS	-0.163	2.724	0.006
Occupation ->NOC	0.133	3.731	0.000
Occupation ->PCS	-0.302	6.458	0.000
Occupation ->WHOQOL	-0.125	2.569	0.010
PCS ->GHP	0.224	7.865	0.000
PCS ->WHOQOL	0.152	7.781	0.000
Sex ->GHP	-0.028	1.415	0.157
Sex ->MCS	-0.030	0.688	0.491
Sex ->NOC	0.041	1.732	0.083
Sex ->PCS	-0.060	1.733	0.083
Sex ->WHOQOL	-0.042	1.140	0.254

GHP = General health perception, NOC = Number of comorbidities

Table 7 displays the total effects and hypothesis testing for the EQ-5D-5L model. The majority of the hypotheses were supported, because the significant level was less than 0.05. However, the hypotheses that represent correlations among BMI, sex, and all endogenous variables were not supported, except for BMI and NOC (p = 0.003).

Table 7.

Total effects and hypothesis testing for the EQ-5D-5L as the dependent variable

Path	Original sample (O)	T statistics (|O/STDEV|)	P values
Age ->GHP	-0.171	11.566	0.000
Age ->MCS	-0.145	5.543	0.000
Age ->NOC	0.195	12.187	0.000
Age ->PCS	-0.404	19.269	0.000
Age ->Utility	-0.304	13.016	0.000
BMI ->GHP	-0.021	1.939	0.053
BMI ->MCS	-0.045	1.927	0.054
BMI ->NOC	0.047	3.024	0.003
BMI ->PCS	-0.004	0.200	0.842
BMI ->Utility	0.020	0.866	0.387
Drug ->GHP	-0.245	14.413	0.000
Drug ->MCS	-0.218	8.673	0.000
Drug ->NOC	0.709	39.925	0.000
Drug ->PCS	-0.298	12.564	0.000
Drug ->Utility	-0.288	10.363	0.000
GHP ->Utility	0.312	14.789	0.000
MCS ->GHP	0.294	11.509	0.000
MCS ->Utility	0.092	8.892	0.000
NOC ->GHP	-0.321	12.382	0.000
NOC ->MCS	-0.330	7.623	0.000
NOC ->PCS	-0.294	7.922	0.000
NOC ->Utility	-0.100	9.209	0.000
Occupation ->GHP	-0.142	4.973	0.000
Occupation ->MCS	-0.163	2.720	0.007
Occupation ->NOC	0.133	3.731	0.000
Occupation ->PCS	-0.302	6.465	0.000
Occupation ->Utility	-0.422	7.835	0.000
PCS ->GHP	0.245	8.545	0.000
PCS ->Utility	0.077	6.637	0.000
Sex ->GHP	-0.030	1.494	0.135
Sex ->MCS	-0.031	0.704	0.481
Sex ->NOC	0.041	1.732	0.083
Sex ->PCS	-0.060	1.740	0.082
Sex ->Utility	-0.023	0.601	0.548

GHP = General health perception, NOC = Number of comorbidities

In addition, path analysis indicated that NOC impacted HRQoL (WHOQOL-BREF and EQ-5D-5L) through the parallel mediation of PCS and MCS, which was identified because the total indirect effect between NOC and GHP was significant ( = 0.169, p < 0.001) for the WHQOOL and EQ-5D-5L models. Notably, supplement material 4, 5, 6 and 7 showed direct and total indirect effects for the WHOQOL-BREF and EQ-5D-5L as dependent variables.

Notably, the adapted Ferran’s model with the correlations between individual characteristics and GHP (supplement material 10) did not fully converge for both models.

Comparison of individual characteristics influencing WHOQOL-BREF, EQ-5D-5L, PCS, and MCS scores between SEM and multiple linear regression analysis

As shown in supplement material 8, there were some dissimilar results regarding factors affecting WHOQOL-BREF, EQ-5D-5L, and MCS scores when comparing SEM and multiple linear regression approaches. Specifically, the variable “drug” showed dissimilar associations with WHOQOL-BREF, EQ-5D-5L, and MCS scores, while “occupation” was found to be a dissimilar factor affecting WHOQOL-BREF and MCS scores between the two analytical approaches. However, multiple linear regression confirmed the presence of parallel mediation, as all pathways from NOC to GHP through the parallel mediators PCS and MCS were statistically significant (p < 0.001) as shown in supplement material 9.

Discussion

This is the first study aiming to explore the pathways linking biological factors, functional status, and GHP to HRQoL based on Ferran’s model, evaluate the psychometric properties of HRQoL instruments, and identify and compare individual characteristics associated with HRQoL between SEM and traditional regression analysis among the general Thai population. The findings partially supported Ferran’s model in explaining the relationships of various factors with HRQoL and demonstrated acceptable psychometric properties within the general Thai population. The discussion has been structured to align with the study’s objectives.

Construct validity and reliability of the HRQoL instruments

GHP failed to reach a satisfactory level for CA but reached the acceptable level for CR. This discrepancy can be attributed to the fact that CA assumes equal item contribution, while the CR considers varying factor loadings and is used in conjunction with SEM [61, 62]. Hence, in this context, CR is more appropriate, because the two GHP items were not equally contributed to the construct.

MH03 and VT02, under the MCS construct, exhibited factor loadings below 0.70; however, these two factor loadings remained statistically significant. A study conducted on multiethnic populations in Singapore presented similar results [63]. This result may be due to both studies recruiting participants from the general population, which is mainly related to chronic physical conditions and, thus, may not be sensitive to mental health items.

Ferran’s model linking biological factors, functional status, and GHP to HRQoL

Within Ferran’s framework, this study classifies NOC as a biological function, which is consistent with previous findings among Thai men with COPD [24], in which the indirect effect of biological dysfunction (exacerbation and FEV-1 scores) was correlated with poor functional status measured using the SF-12.

Functional status (PCS and MCS) was also positively associated with GHP, which supports Ferran’s conceptual model. Previous research reported better functional status associated with more positive GHP in patients with HIV/AIDs [64] and COPD [24], while increased functional limitation was associated with diminished GHP among patients with heart failure [65]. Furthermore, the current study found a direct positive association between GHP and HRQoL, thus reinforcing the role of GHP as a key factor for overall QoL. Studies on various patient populations, including those with HIV/AIDs [64], COPD [24], and ESRD [66], supported this phenomenon. Therefore, this study confirmed the causal relationships between these factors and HRQoL among the general Thai sample using Ferran’s model, although it was originally developed to explain linkages among several factors related to HRQoL among patient groups.

This study also revealed that NOC indirectly affected GHP, but PCS and MCS exerted parallel mediation effects on this relationship, because the total indirect effects of NOC and GHP were significantly different for both models. These findings support Ferran’s model of functional status as a mediating variable in the relationship between biological function and GHP [6].

Individual characteristics associated with HRQoL using SEM and result comparision with traditional regression analysis

There were slight differences in the factors affecting WHOQOL-BREF and EQ-5D-5L scores. For example, BMI was significantly associated with WHOQOL-BREF scores, whereas no such association was observed with the EQ-5D-5L. This discrepancy may be due to the different conceptual frameworks and scoring methods of the two instruments. While WHOQOL-BREF provides domain-specific scores of HRQoL (physical, psychological, social, and environmental domains), EQ-5D-5L gives a utility score. Therefore, the differences in results may stem from distinct concepts and sensitivity to health changes of these two instruments [67]. In addition, some inconsistencies were found in the factors associated with WHOQOL-BREF, EQ-5D-5L, and MCS scores when comparing the results from SEM and traditional multiple linear regression analyses. These differences highlight a key limitation of multiple linear regression, which treats each dependent variable separately and may be more prone to measurement error—especially when analyzing a complex structure with several components linked to the construct such as HRQoL. In contrast, SEM allows for simultaneous modeling of relationships among multiple observed and latent variables while accounting for measurement errors. Therefore, SEM may provide more accurate and consistent estimates of the factors influencing HRQoL. While these limitations of using multiple linear regression for a complex structure of construct like HRQoL are recognized, many scholars are more likely to use SEM for investigating factors affecting HRQoL because it can identify direct, indirect, and total effects of the variables among various groups of populations. Moreover, several GOF indices can be employed to determine the structural model fitness [68, 69].

Compared with earlier Thai studies [14, 41, 70], the current findings were partially consistent, because they identified that female, older, and unemployed participants and those with higher BMI, higher NOC, and poor GHP reported lower utility scores. Similar to the studies from South Korea [71] and Singapore [72], they also reported that older participants with comorbidities were associated with diminished physical functioning, lower GHP, and reduced HRQoL measured by both SF-36 and EQ-5D-5 L. Our findings reinforce the need for public health strategies to better allocate health care resources to older adults and individuals with multi-morbidities in order to improve physical, mental functioning and overall HRQoL at the population level.

Conversely, the study did not report sex and BMI as significant predictors for both models and EQ-5D-5L, respectively. This result may be due to methodological differences, given that previous studies employed univariate and multiple regression analyses, while the current study applied SEM, which could account for complex relationships. Additionally, a number of previous studies were conducted prior to the outbreak of the COVID-19 pandemic, while the current study applied post-pandemic data where the acute phase of infection has passed. In addition, other previous studies reported that the effect of the COVID-19 pandemic on physical and psychological health may have caused certain disparities in factors that influence HRQoL scores [73, 74].

Notably, the NOC and MCS had the highest and lowest adjusted R² among the latent variables. These findings implied that participant characteristics as the exogenous variables could adequately explain the NOC construct, while those characteristics might not be able to explain well enough for the MCS construct. We explained that all of the collected participant characteristics might not be relevant to the mental health of the general Thai samples and those with impaired health status was mainly related to physical health. Therefore, future research should further include participants with a wide range of health statuses covering both physical and mental health impairments.

When the researchers aim to investigate the relationships among variables, they should consider the characteristics of causal relationships within the conceptual model. SEM could be used to simultaneously investigate complex relationships consisting of a large number of endogenous and exogenous variables, including mediating and latent variables, to simultaneously estimate the relationship of those variables [18]. It is considered an approach to obtain a more precise measurement of theoretical model [18]. Conversely, the traditional regression analysis could account for only simple causal relationships between dependent and independent variables [18]. SEM could be used to tackle a large number of endogenous and exogenous variables to simultaneously estimate the relationship of variables within the conceptual model [75]. Furthermore, SEM can precisely increase estimation when larger numbers of samples are recruited due to complex model estimations [76]. Notably, SEM has two subtypes: covariance-based SEM (CB-SEM) and PLS-SEM. CB-SEM generally requires a larger sample size than PLS-SEM [77]; however, a previous study suggested that both PLS-SEM and CB-SEM can yield similar results when the sample size exceeds 250 [78].

Research implication for ferran’s model

The findings from this study should be interpreted within the context of unique cultural and environmental aspects of Thailand. It differs from many western countries where most HRQoL research and instruments were originally conducted and developed. These contextual differences could impact the perception of HRQoL perceived by the Thai population. In Thailand, the health care system mainly operates under a universal health coverage (UHC) in which a majoriry of the Thai population can access all needed medication and health care services. It is considered an important factor affecting the perception of HRQoL dimensions, especially for physical and psychological health because the UHC could reduce some risky behaviors and promote some healthy behaviors, resulting in increased life expectancy and reduced mortality from communicable diseases [79]. Unlike in many high-income countries, where access to healthcare is more uniform or privatized, the Thai system may shape expectations and coping strategies differently. Furthermore, cultural and religious values can play an important role in how individuals perceive HRQoL constructs and dimensions differently from the western countries. In Thailand, Buddhism is the religion believed by the majority of Thai people, and Buddhism plays an important role in Thai lifestyle and culture. Buddhism’s core belief is karma, which refers to the correlation between intentional deed and its subsequent outcome in the future, so that karma is a driving factor that affects the concept of health and disease believed by Thai people. Generally, good health is perceived as a result of good karma in past lives, whereas sickness is the result of bad karma or misdeeds in the previous life [80, 81]. Making merits is generally perceived as the action that can accumulate the merit to avoid misfortune and bring good health to the Thais [82]. It may shape the subjective evaluation of QoL, particularly in the domains related to mental health [80]. Therefore, Buddhism can make Thai people uniquely conceptualize the concept of HRQoL and dimension differently from Western people.

These findings provide valuable insights into the application of Ferran’s model to illustrate interrelationships among various factors that influence HRQoL in the general Thai population. This model incorporates the use of the SF-12v2 and WHOQOL-BREF to represent its components, thus demonstrating that multiple generic instruments can be jointly employed to offer a comprehensive assessment of HRQoL. The results also implied that certain mental health items (specifically MH03 and VT02) may inadequately capture the mental health status of the general Thai population. To enhance the sensitivity of these items, future studies should conduct a new population health survey with a more diverse Thai sample that includes individuals with a broad spectrum of physical and mental health conditions. Additionally, the findings indicate that various sociodemographic factors may influence item responses and contribute to differences in HRQoL across population subgroups. These insights could support policymakers in allocating healthcare resources to those with low levels of HRQoL more effectively.

Although there is the Engel’s biopsychosocial (BPS) model developed to provide a broad framework for understanding health through the correlations among three factors, including biological, psychological, and social factors, its general nature was not originally developed for measuring a complex construct like HRQoL, as it was primarily designed for mental disorders and it is applicable to explain the correlations among those three factors within the patient groups [83–85]. Conversely, the Ferrans’ model was employed and adapted to illustrate the interrelationships among multiple factors associated with HRQoL. It offers a comprehensive framework that integrates biological, physical, psychological, and social functioning, along with individual perceptions and characteristics that influence overall HRQoL [6]. A key advantage of using the Ferrans’ model is its ability to demonstrate how individuals with varying health statuses may assign different levels of importance to specific life domains and how these domains, in turn, impact their HRQoL. Furthermore, the Ferran’s model can support cross-population comparisons when global measures of functional status and general health perceptions—such as the SF-12v2 or WHOQOL-BREF—are employed in both general and patient populations [8]. Therefore, the model can facilitate a more nuanced understanding of how personal, environmental, and socioeconomic factors collectively shape QoL and it is appropriate for our study’s objectives and data, thereby supporting more person-centered and equitable approaches in public health policy for the general population [8].

Model recommendations and future research directions

While the adapted Ferran’s model could provide the understanding of relationships between the factors and the HRQoL in the general Thai samples, some concerns might have arisen in terms of the limited use, robustness, and generalizability of the findings.

Although this adapted Ferran’s model is useful for investigating factors affecting HRQoL, it is limited used within some groups of the general Thai population because this study did not include individuals with specific chronic diseases or ethnic minorities. Therefore, this adapted model and its finding might not fully represent the diversity of the Thai population.

Regarding the robustness of the findings, we mainly obtained the data from the self-reported measures for each element of the adapted Ferran’s model. The data from the laboratory testing was not collected because it was conducted with the general Thai samples residing at their residences. These might have caused some response bias because we lacked the objective data from the laboratory testing to verify the self-reported data. Further studies should collect the objective data from the laboratory testing to verify the relationship with the HRQoL construct.

The generalizability was limited in terms of the findings, which might not fully reflect the heterogeneity of the general Thai population. In addition, some elements that were not included in the adapted Ferran’s model could bring some limitations of generalizability to other populations. Future research should therefore aim to investigate the model with full elements across different populations to strengthen the external validity.

Notably, the elements of “characteristic of environment” and “symptoms” were excluded from the adapted Ferran’s model due to unavailability of the collected data in this study. The omission of these factors may introduce variable bias because these factors might have influenced functional status and the overall HRQoL, and it might have distorted the interpretation of the causal relationships with the HRQoL.

While the biological function is also considered an important element, its operationalization slightly differed from the original Ferran’s model. In the Ferran’s model, the biological function is drawn from objective clinical data including laboratory tests, physical assessments, and medical diagnoses typically collected from medical records. In this study, we did not collect those variables; instead, the biological function was investigated using self-reported NOC as the proxy indicator that might not fully capture clinical symptoms of individuals because this study was collected from the general samples.

Given these limitations, we recommend that researchers should collect clinical data from the medical records and characteristics of the environment and symptom profile. This may allow the researchers to comprehensively investigate causal relationships between these factors and the HRQoL construct through the Ferran’s model to maintain the rigor of theoretical concept while adapting to the clinical context in the future research. Furthermore, future research should compare between the original and adapted Ferran’s model on the same dataset to provide the magnitude of bias due to the omission of some variables in the adapted Ferran’s model.

A comparison of the adapted ferran’s model with and without the inclusion of individual characteristics and GHP

Although the adapted Ferran’s model revealed causal relationships between participant characteristics and several elements, the directional associations with GHP were not examined in this study because the GHP is conceptualized as a mediator of the HRQoL construct in Ferran’s model. Specifically, this study mainly focused on how individual characteristics influence each component of the Ferran’s model without overcomplicating the structural model. Incorporating the path linking between participant characteristics and GHP would have increased the model complexity, which would more likely lead to overfitting and biased estimates [86, 87]. Unlike a previous study conducted in Korea, which utilized a standardized and validated instrument specifically designed to assess the GHP [66], the present study employed items from the WHOQOL-BREF. Although the WHOQOL-BREF includes general health-related items and is a validated instrument for HRQoL measurement in the Thai population, it may not be considered a specific tool for GHP assessment. This was further reaffirmed by the non-convergence of the adapted Ferran’s model when individual characteristics and GHP were tested, possibly due to the increased model complexity resulting from the correlations between these variables [88]. Non-convergence in SEM also indicated that the data did not support the parameter estimations and it could lead to biased estimates of the model [75]. Given these technical constraints, the adapted Ferran’s model without the inclusion of correlations between individual characteristics and GHP should be prioritized to avoid increasing complexity of the model and biased estimates of the model parameters and aligned with the theoretical framework.

To enable the successful implementations, future research should explore this potential causal pathway with a more diverse Thai population and more specific tool to assess the GHP to strengthen the robustness of the model and enhance the external validity of the findings.

Strengths and limitations

This study has several strengths. First, it compared the results obtained from traditional regression analysis and SEM, demonstrating that SEM is a more appropriate statistical approach for investigating the factors affecting the HRQoL construct. Second, the study confirmed that the Thai HRQoL instruments used exhibit good psychometric properties for the general Thai population, as determined through SEM analysis. Third, identifying the factors that affect HRQoL can help guide the effective allocation of limited resources to targeted population groups with diminished HRQoL levels in Thailand.

This study also has some limitations. First, it used secondary data, which may have omitted important variables such as specific symptoms or biological disease markers. Future research could include these factors to validate Ferran’s model in the target population. Second, participants with impaired health conditions were related to physical illness, which could limit sensitivity to mental health items. Future studies that target populations with mental health issues are required. Third, this study did not capture some elements highly relevant to the Thai context such as access to medical facilities and religious and cultural differences that affect the perception of HRQoL construct perceived by the Thais. Future research could benefit from incorporating such context and sociocultural variables in the structural model to contribute to a more holistic understanding of the determinants of HRQoL in Thailand. Fourth, we did not collect data on the types of medications used by each participant. Therefore, future research should include this information to further investigate which types of medications most significantly affect HRQoL scores. Fifth, the findings provided partial support for Ferran’s model, as certain data relevant to its elements—such as objective assessments of biological function—were not collected. This discrepancy may have contributed to the absence of certain participant characteristics and latent variables. Therefore, future research should re-examine the findings using comprehensive data that fully encompass all components of Ferran’s model. Sixth, this study did not employ a specific instrument for assessing GHP and did not examine the relationships between participant characteristics and GHP. Therefore, future research should utilize a validated GHP-specific tool and explore this potential causal pathway more comprehensively.

Conclusions

Although Ferran’s model was originally developed for patient groups, the current study confirms that it can be used to effectively explain relationships between key variables and HRQoL within the general Thai population. GHP emerges as the strongest predictor of HRQoL for both models. The differences in findings between WHOQOL-BREF and EQ-5D-5L may be due to their respective health dimensions and recall periods. Moreover, these findings can guide policymakers in allocating healthcare resources more effectively toward individuals with diminished HRQoL. To validate these findings, future research should be conducted on representative samples of the Thai population, addressing both physical and mental health conditions, and employing Ferran’s model with a standardized instrument for GHP assessment along with comprehensive data across all its components.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AD

Anxiety/depression

AVE

Average variance extracted

BMI

Body mass index

CA

Cronbach’s alpha

CR

Composite reliability

EQ-5D-5L

EuroQol five-dimension, five-level questionnaire

GHP

General health perception

GOF

Goodness-of-fit

HRQoL

Health-related quality of life

MCS

Mental component summary

MH

Mental health

NOC

Number of comorbidities

PCS

Physical component summary

PD

Pain/discomfort

QoL

Quality of life

SC

Self-care

SF-12v2

The Short Form Health Survey-version 2

SEM

Structural equation modeling

UA

Usual activities

WHOQOL-BREF

World Health Organization QoL — Brief Version

Author contributions

Krittaphas Kangwanrattanakul was involved in study conception and design, data collection, data analyses, interpretation, and wrote the first draft of the manuscript. Apinya Ingard rechecked the data interpretations. All authors edited, read, and approved the final manuscript and are all in agreement with the manuscript. The content has not been published elsewhere.

Funding

This study was funded by the Research Grant of Burapha University through National Research Council of Thailand under Grant No. Rx6/2566.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The protocol and data collection process was approved by the ethical committees: Burapha University Institutional Review Board (IRB1-031/2566). All participants gave their informed consent for study participation before the study commenced. Furthermore, it adhered to the principles outlined in the Declaration of Helsinki.

Consent for publication

This work does not contain any individual person’s data in any form (including any individual details, images, or videos.

Competing interests

The authors declare no competing interests.