Evidence of Validity and Factorial Invariance of a Diet and Healthy Lifestyle Scale (DEVS) in University Students

Abstract

Background: University students continue to face health challenges related to a healthy diet and lifestyle. In this context, the measurement of diet and health status is important for institutions interested in health care and promotion. Objective: The objective of this study was to translate into Spanish, evaluate the internal structure, reliability, and factorial invariance of the Diet and Healthy Lifestyle Scale (DEVS) scale. Methods: The participants were 4482 university students aged 18 to 59 years (Mean [M] = 21.32, Standard deviation [SD] = 2.81). Data analysis included exploratory factor analysis (EFA) and confirmatory factor analysis (CFA), internal consistency, and through multigroup analysis, gender invariance was assessed. Results: The results showed the unidimensionality of the DEVS in Spanish and that it provides good reliability indices: Total sample (Ordinal Alpha [ordinal α] = 0.80, Omega [ω] = 0.83, Coefficient H [H] = 0.84), men (ordinal α = 0.79, ω = 0.83, H = 0.81), women (ordinal α = 0.84, ω = 0.85, H = 0.86). Configurational, scalar, and strict metric invariance was verified, indicating that the construct can be evaluated in both men and women. Conclusion: The DEVS is a valid, reliable, and invariable measure to measure the healthy lifestyle in university students.

1. Introduction

It is important to lead a healthy lifestyle from an early age [1]. The adoption of healthy eating habits, such as a balanced diet, rest habits, and physical exercise, decrease the risk of mortality [2,3]. On the contrary, unbalanced diet, lack of rest, sedentary lifestyle, smoking and alcohol consumption are associated with long-term non-communicable diseases (NCDs), such as diabetes mellitus, cardiovascular diseases, and different types of cancer [4,5]. These pathologies are increasingly becoming a challenge for the quality of life of people and health systems, as they represent one of the main causes of death in the world, decisively replacing infectious diseases and malnutrition [6,7,8]. According to a World Health Organization (WHO) report, NCDs cause approximately 41 million deaths each year, accounting for 71% of deaths worldwide [8].

The prevalence of unhealthy lifestyles is increasing among university students worldwide [9,10]. University students are a vulnerable group and constantly face a range of health problems [11]. Previous research on university students has shown low fruit and vegetable consumption, sedentary lifestyles, poor sleep quality, and unhealthy eating habits [10,12], and more than half have used tobacco at some point during their university years [13]. Among Peruvian university students, alcohol is the most commonly consumed drug [14]. A report estimated that 16% of the Peruvian population over 20 years of age suffers from coronary heart disease and more than 2000 die from some form of heart failure [15]. Cardiovascular risk factors, such as changes in lipid metabolism, obesity, and overweight, are more prevalent in Peruvian students [16]. Thus, male university students have a higher prevalence of overweight (49.65%) compared to females (16.51%) [17].

Healthy behaviors can promote better overall health, leading to better prevention and control of certain NCDs [18,19]. The benefits of dietary patterns based on the consumption of minimally processed plant foods are also evident in the prevention of NCDs. Several studies have highlighted the importance of plant-based diets in overall health [20] and, particularly, in the reduction of the prevalence of chronic conditions [21,22]. Vegetarian dietary patterns characterized by the consumption of whole grains, legumes, vegetables, fresh fruits, nuts, dried fruits, and seeds have been linked to a lower risk of all-cause mortality [23], lower prevalence of obesity [24], hypertension [20], diabetes [25], some types of cancer [26], and cardiovascular diseases [27]. The prevention of these diseases is conditioned not only by diet but also by lifestyle [10,12].

It is necessary to have practical instruments to measure or evaluate fundamental aspects of lifestyle and diet in the university population in order to establish programs or alternatives that contribute to individual or collective healthy behaviors [28]. In Peru, a 34-item instrument has been developed that measures physical/mental, ethical/moral, academic/family aspects [29] and another 25-item instrument in university students in a quarantine setting measuring dietary and unhealthy eating habits, physical activity, and use of media [30]. However, no brief questionnaires have been developed that measure dietary intake, physical activity, water consumption, and sun exposure, which are easy to administer. This is essentially relevant and of great importance for public health if we consider that research on lifestyles is relatively recent and is carried out, particularly in Western countries, which hinders the availability of validated instruments in middle-income countries, such as Peru.

Diet and lifestyles differ between men and women [31]. In general, men tend to have a high consumption of meats, foods rich in salt and fats, on the other hand, they rarely consume adequate amounts of fruits and vegetables [32]. In the Peruvian context, previous studies report similar dietary patterns according to sex [33,34]. Two studies using the Healthy Lifestyle Index revealed that [35,36] women had a higher score compared to men. Furthermore, men tend to prefer dietary patterns containing meat, such as non-vegetarian (pattern with no specific dietary restrictions on frequencies of meat, fish, and dairy consumption) and semi-vegetarian (consumption of red meat, poultry, or fish no more than once a week) patterns. Dietary patterns with an abundant consumption of minimally processed plant foods showed higher proportions of women [20,24]. Considering the importance of measuring lifestyle in all strata of the population, the evaluation of the factorial invariance between groups is considered important since, despite its importance, to date the invariance of diet and healthy lifestyles between groups of women and men has not been reported. Invariance will make it possible to demonstrate the comparison between groups since its absence may suggest that the comparison between groups is biased [37].

In Peru, although questionnaires have been developed and validated to measure lifestyle, to date, there is no instrument to estimate dietary intake and aspects of lifestyle. In view of this need, this study aims to translate into Spanish and adapt the Diet and Healthy Lifestyle Scale [35], evaluate the factorial structure (exploratory and confirmatory factor analysis) and reliability by means of the coefficients ordinal α, ω, and H, apart from evaluating the factorial invariance in men and women [38,39].

2. Materials and Methods

2.1. Design and Participants

This research has a cross-sectional and instrumental observational design [40]. The participants were selected through a non-probabilistic convenience sampling method. The participants study at a private university that has 3 campuses in different regions of Peru (coast, highlands, and jungle). Data were collected during the months of March and May 2021. The university students were recruited through an invitation sent via the university’s virtual platform in which the purposes of the research and their rights as potential participants were briefly described.

The effect size considering the number of observed and latent variables in the model, the anticipated effect size (λ = 0.10), the desired statistical significance (α = 0.05), and the statistical power level (1 − β = 0.90) considering a recommended minimum sample of 199 for structural equation studies (SEM) using Soper’s software [41]. To be more representative, the participants were 4482 Peruvian university students aged 18 to 59 years (M = 21.32, SD = 2.81) from the three regions of the country (coast, highlands, and jungle) selected by non-random sampling. The female subsample (47%) had a mean age of 21.66 (SD = 2.94), and the male subsample (53%) had a mean age of 21.01 (SD = 2.66). Inclusion criteria were considered: (1) Peruvian university students (2), over 18 years of age (3), accept the informed consent (4), and have access to the online survey. Those who did not give their consent to participate and foreign students were not considered.

2.2. Instrument

Diet and Healthy Lifestyle Scale (DEVS): We considered the unidimensional model based on the Vegetarian Lifestyle Index (VLI), which assesses dietary patterns between vegetarians and non-vegetarians and lifestyle adherence, as well as various theoretical proposals related to healthy eating and healthy living [35]. It is made up of 14 items, which assess the consumption of whole foods of plant origin, such as fruits, vegetables, legumes, nuts, and whole grains, foods of animal origin, such as milk and dairy products, eggs, reliable sources of vitamin B-12 and candies. In addition, they consider aspects of lifestyle, such as physical activity, adequate water intake, and moderate exposure to sunlight. The response options for each question were limited to 3, from 1 = (minimum perceived intake) up to 3 = (maximum perceived intake). The total scores are obtained by adding each item, where higher scores indicate greater adherence to diet and lifestyle. The VLI was translated into Spanish in accordance with the guidelines established for the translation and intercultural validation of psychometric instruments [42].

• Initially, two registered dietitians, experts in public health and vegetarian nutrition, and fluent in English and Spanish, conducted a direct and independent translation of the Vegetarian Lifestyle Index into Spanish (Peru).
• Second, the first Spanish version was independently back-translated into English by two translators whose native language was English and who were fluent in Spanish.
• Third, based on both versions, the research team, together with the translators mentioned above, evaluated the translated versions. This process was facilitated by the fact that the instrument had previously been translated into Spanish (Argentina) [36]. Therefore, a comparative analysis was carried out with this existing version, considering some linguistic and cultural similarities. The items were evaluated by registered dieticians and experts in the field who considered that the items were adequate and that the instrument was relevant to the public health of the Peruvian population. That allowed developing the initial version.
• Fourth, a pilot test was carried out, in which the initial version was applied to 10 students to check the readability and understanding of the items.
• Fifth, the research group evaluated the pilot test, and no modifications were suggested, which allowed having the final version of DEVS in Spanish (See Appendix A).

2.3. Procedure

The data were collected between March and May 2021 through a digital platform implemented by Universidad Peruana Unión. This method of data collection was due to the social isolation measure, as part of the sanitary restrictions in Peru by COVID-19, and followed the internet-mediated research (IMR) procedure, applying the ethical and methodological recommendations for this medium [43,44].

2.4. Ethical Aspects

On the home page of the survey, relevant information was provided on the purpose of data collection and the objective of the study. This page also contained a note of an invitation to participate in the study and written informed consent. Consent was obtained from the participants by clicking on the “I wish to participate” icon after having read and accepted the terms of informed consent. The project was approved by the Research Ethics Committee of the Universidad Peruana Unión and registered under the number 2021-CEUPeU-0009. In addition, it was performed in accordance with the ethical criteria established in the Declaration of Helsinki.

2.5. Statistical Analysis

Prior to the descriptive analysis of the data, the total sample (n = 4482) of the participants was divided into a sample of men (n = 2374) and women (n = 2108) in order to perform a CFA for both sexes [45]. Descriptive statistics, such as mean (M), standard deviation (SD), skewness (g₁ = +/−1.5), and kurtosis (g₂ = +/−1.5), were used according to Kline’s criteria [46] and the corrected item-test correlation analysis was considered for the elimination of items in case of being r(i-tc) ≤ 0.2 or multicollinearity (i-tc) ≤ 0.2 [46] and internal consistency was estimated using the ordinal α coefficient.

An exploratory factor analysis (EFA) was performed to determine the factor structure of the items. Parallel analysis was performed to determine the optimal number of factors and the Kaiser–Meyer–Olkin test and Bartlett’s sphericity test [47]. KMO values >0.8 and a p-value < 0.05 would indicate that the matrix is suitable for factor analysis [47]. The factorial estimation was carried out with the MINimum RESiduals (minres) method, which seeks to minimize the residuals of ordinary least squares and promax rotation [48]. The elimination of items with factor loadings lower than 0.3 was considered, and the correlations between factors were analyzed where values higher than 0.3 were considered adequate. Reliability analysis was performed with the ordinal α coefficient, McDonald’s ω [49], and H coefficient [50], which values >0.70 (acceptable), >0.80 (good), or 0.90 (excellent) to assess the construct [38,51,52].

Confirmatory Factor Analysis (CFA) of the DEVS was performed through structural equation modeling (SEM) using the weighted least squares weighted mean and variance adjusted robust least squares (WLSM) estimation method [53]. For the evaluation of the fit of the model, the proposals of Escobedo et al. [54] and Kline [46]. An acceptable model is considered when the comparative fit index (CFI) and the Tucker–Lewis index (TLI) have indices ranging between 0.90 and 0.95, and an adequate fit if the indices are greater than 0.95. Root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR) indices with values between 0.05 and 0.08 would indicate an acceptable fit, and values below 0.05 would indicate an adequate fit.

The factorial invariance was analyzed by means of restrictive progressive stage [55], considering the value of CFI (ΔCFI ≤ 0.01) and RMSEA (ΔRMSEA ≤ 0.015) [56], since χ² (Δ χ²) as a difference test is sensitive to sample size [57]. A reference model was established by evaluating the configural invariance (M1) that allows estimating factor loadings, intercepts, and residuals. We also evaluated the metric invariance model (M2) that constrained factor loading equally across groups and tested the scalar invariance model (M3) that allows factor loading and intercepts to be equal across groups. Finally, strict invariance (M4) was tested, which equals factor loading, intercepts, and residuals.

Statistical analyses were performed using R software, version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org, accessed on 5 March 2022). The “lavaan” package was used for the CFA [58]. For the factorial invariance, the “semTools” package was used [59].

3. Results

3.1. Descriptive Analysis

Table 1. Descriptive of the items.

Item	M	SD	g1	g2	r-itc	αordinal
1	1.5	0.61	0.81	−0.35	0.52	0.81
2	1.73	0.64	0.32	−0.71	0.52	0.81
3	1.64	0.64	0.48	−0.68	0.57	0.81
4	1.78	0.69	0.31	−0.89	0.56	0.81
5	1.56	0.68	0.83	−0.51	0.55	0.81
6	1.43	0.60	1.05	0.08	0.59	0.80
7	1.81	0.59	0.06	−0.34	0.53	0.81
8	1.95	0.58	0.00	−0.07	0.45	0.81
9	1.48	0.61	0.87	−0.24	0.45	0.81
10	1.76	0.66	0.31	−0.79	0.55	0.81
11	1.85	0.55	−0.06	0.03	0.34	0.82
12	1.94	0.66	0.07	−0.70	0.49	0.81
13	1.87	0.69	0.18	−0.88	0.48	0.81
14	1.82	0.70	0.26	−0.95	0.40	0.82

Note. M = Mean; SD = standard deviation; g₁ = Asymmetry; g₂ = Kurtosis; r-itc = Correlation of item-total-corrected; α_ordinal = Ordinal alpha (n = 4482).

shows the descriptions of the DEVS elements. The means range between 1 and 2, and the standard deviations range between 0.5 and 1, indicating that the variability of the items is adequate. Regarding the coefficients of asymmetry and kurtosis, all the items present values lower than 1.5 in absolute value, thus indicating that the items present univariate normality [60]. The presence of extreme scores and multicollinearity was not detected (Table 1). The reliability coefficients were optimal (ordinal α > 0.70), and the corrected item-total correlation coefficients were above 0.30, except for item 11 of the male sample (r-itc = 0.28), which is an acceptable value.

3.2. Exploratory Factor Analysis

Factor estimation was performed through exploratory factor analysis. The sample adequacy analyses using the Kaiser–Meyer–Olkin tests and the sphericity test (KMO > 0.8; p < 0.000) indicate that the sample matrix is adequate to perform the factorial analysis [52]. With the parallel analysis, which is a method of estimating the number of factors proposed by Horn [61], the presence of an underlying factor was estimated. With the minres method and the promax rotation method, the factor loads were estimated to be greater than 0.3 in all the items (

Table 2. Exploratory factor analysis.

Item	Factor	h2	u2
1	0.53	0.278	0.72
2	0.53	0.277	0.72
3	0.59	0.347	0.65
4	0.57	0.330	0.67
5	0.57	0.323	0.68
6	0.60	0.359	0.64
7	0.52	0.269	0.73
8	0.42	0.179	0.82
9	0.44	0.193	0.81
10	0.54	0.296	0.70
11	0.31	0.097	0.90
12	0.48	0.228	0.77
13	0.47	0.224	0.78
14	0.39	0.151	0.85

Note. h2 = Communalities; u2 = Unicities.

).

3.3. Confirmatory Factor Analysis and Reliability

With the confirmatory factor analysis, the model was evaluated in the total sample in men and women (

Table 3. Goodness-of-fit index.

Model	χ2	p	CFI	TLI	RMSEA [90% CI]	SRMR	ω	H
Total sample	2334.356	0.000	928	915	0.068 [0.069–0.075]	0.074	0.83	0.84
Men	1146.411	0.000	908	891	0.077 [0.072–0.081]	0.070	0.80	0.81
Women	1387.884	0.000	940	929	0.074 [0.074–0.071]	0.071	0.85	0.86

Note. Total sample n = 4485; women (n = 2108); men (n = 2374); χ² = Goodness-of-fit test; CFI, Comparative fit index; TLI = Tucker–Lewis Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square Residual.

). Considering the goodness-of-fit indices obtained and the criteria proposed by Kline [46], it was observed that the one-dimensional model had an adequate fit (χ² = 2334.356, p = 0.000; CFI = 0.928; TLI = 0.915; RMSEA = 0.068 [90% CI: 0.066–0.071]; SRMR = 0.068), also the reliability coefficients ω (0.83) and H (0.84) were adequate (>0.7). Furthermore, the one-dimensional model was tested separately in university students, obtaining adequate goodness-of-fit indices: Men (χ² = 1146.411; CFI = 0.908; TLI = 0.891; RMSEA = 0.077 [90% CI: 0.072–0.081]; SRMR = 0.070), with a good coefficient of ω (0.80) and H (0.81), and women (χ² = 1387.884; CFI = 0.940; TLI = 0.929; RMSEA = 0.071 [90% CI: 0.074–0.071]; SRMR = 0.071) and a good ω (0.85) and H (0.86).

3.4. Factor Invariance by Gender

The factorial invariance model showed that the unidimensional model is adequate for the male and female groups (

Table 4. Measurement Invariance.

	χ2 (df)	Δχ2 (Δdf)	RMSEA [90% IC]	TLI	p	CFI	(ΔCFI)	(ΔRMSEA)
M1	1760.008 (154)	-	0.056 [0.054, 0.059]	0.908	<0.001	0.922	-	-
M2	1673.231 (167)	86.777 (13)	0.056 [0.054, 0.059]	0.908	<0.001	0.915	0.007	0.000
M3	1816.516 (180)	143.285 (13)	0.057 [0.054, 0.059]	0.907	<0.001	0.908	0.007	0.001
M4	1918.05 (194)	101.534 (14)	0.056 [0.054, 0.059]	0.908	<0.001	0.901	0.007	−0.001

Note. M1 = configural; M2 = Metric, M3 = Scalar; M4 = Strict; χ²: chi-square; df = degrees of freedom; Δχ²: chi-square difference; TLI = Tucker–Lewis Index; CFI = Comparative Fit Index; ΔCFI = Comparative Fit Index difference; RMSEA = Root Mean Square Error of Approximation.

). The configural invariance model (M1) was evaluated as a basis for the evaluation of the others, indicating a good fit. Metric invariance (M2) and scalar invariance (M3) were a good fit with statistically significant values (p < 0.01). Finally, residuals were equalized between groups for strict invariance (M4). The ΔCFI in all models was within the expected range and showed that progressive restrictions do not modify the fit of the single factor. Therefore, factorial invariance is assumed for the male and female groups.

4. Discussion

University students continue to face health challenges, such as being overweight and obese [11]. Generally, students who are obese are at higher risk of various non-communicable diseases [62], including type 2 diabetes mellitus, cardiovascular disease, and even some types of cancer and an increased risk of mortality [63]. Prevention of these associated pathological problems should be done by adopting a healthy lifestyle with the purpose of maintaining or improving physical and emotional well-being [64]. Consequently, the evaluation of lifestyle parameters in university students is necessary. In addition, there is a need to deepen the understanding of these indicators in both men and women students to implement personalized health promotion and intervention programs aimed at preventing disease and improving their quality of life.

On the other hand, to date, a scale has not been validated to assess eating and lifestyle behaviors in a Latin American country, so the objective of the research is to present the first evidence of the validity of the DEVS. The study of healthy eating and aspects of health-related behaviors is in continuous development and, at present, demographic variables, such as sex, are important for a better understanding of their relationship [35].

Through parallel analysis and CFA, a unidimensional model was considered that provides an adequate fit with adequate reliability. The results corroborate the unidimensionality as well as the original proposal [35]. The coefficients ω and H considered to evaluate the limitations of Cronbach’s alpha coefficient were estimated and were good (>0.80). In addition, the item-totals (>0.20) show adequate homogeneity. No previous studies were found that examined the factorial invariance of the DEVS according to gender. Configural metric and scalar invariance was maintained for both genders, indicating that the construct can be measured in both males and females. Considering that most studies analyze men and women separately, it is important to provide statistical information that allows for an accurate comparison of gender roles. In fact, the lifestyles of male and female students are different, for example, in regard to food, males are more likely to consume animal foods, such as meat and poultry, while females are more likely to report eating fruits and vegetables [20]. In general, women tend to have healthier eating habits, a higher rate of moderate-vigorous physical activity, and a lower rate of smoking and obesity compared to men [31]. Therefore, gender differences in lifestyle should be considered a relevant topic that deserves special attention among researchers and health professionals.

The relevance and practical importance of the DEVS lies in the identification of university students with an inadequate lifestyle index score to implement intervention programs with a preventive approach to non-communicable diseases considering the gender factor. Although financial costs can be a problem for public health intervention programs, interventions aimed at improving the lifestyles of university students can be carried out at a low cost in accordance with the health policies of higher education institutions. In addition, students, regardless of gender, who present any chronic condition associated with an unhealthy lifestyle, should be a priority group within university intervention programs to reduce the consequences of these diseases through adequate management and control.

The present study is subject to several limitations. First, invariance has only been evaluated in relation to sex. Therefore, other studies should evaluate differences with other sociodemographic variables of interest. Second, although the study collected data on the lifestyles of students at a private university with campuses in the three regions of Peru (coast, highlands, and jungle), it is not possible to generalize to other Spanish-speaking cities. In addition, it is a non-probabilistic convenience sample. On the other hand, the applicability of the instrument is limited because university students represent a group with a solid academic background, and the other strata of society were not considered. Finally, the cross-sectional design represents an important disadvantage due to its susceptibility to errors or confounding factors. Longitudinal designs that allow causal evaluations and provide reliable inferences are suggested. Moreover, it is recommended that future studies could include representative samples from other regions or other segments of the population, including public universities or older adults. Future studies could also evaluate invariance according to body mass index (BMI). Despite the limitations, the study evaluates the psychometric properties of the DEVS and provides the first empirical evidence of this instrument in Spanish.

5. Conclusions

The psychometric evaluation supports the unidimensionality of the DEVS. Moreover, it is evident that the factorial validity and reliability of the index are acceptable for the university setting. Consequently, this DEVS can be used in the evaluation of dietary patterns and healthy behaviors, such as water consumption, sunlight exposure, and physical activity, in the Latin American university context. In addition to being valid and reliable, it is an instrument that stands out for its brevity and solidity.