Latent profile analysis of adolescents’ active health behaviors and the predictive factors

Abstract

This study aimed to explore the latent profiles of adolescents’ active health behaviors and analyze the predictive factors of different latent profiles of active health behaviors. In December 2024, a survey was conducted among 1,093 middle school students from two rural schools in Anhui Province, China, using convenience sampling. The Adolescent Active Health Behavior Scale (AAHES), Family Functioning Scale (FF), and modified eHealth Literacy Scale (m-eHEALS) were adopted. Latent Profile Analysis (LPA) was used to identify the latent profiles of active health behaviors, and multivariable logistic regression analysis was applied to explore the relevant factors of active health behaviors. The active health behaviors of adolescents could be divided into three latent profiles: Negative Coping Type (27.5%), Unstable Type (46.1%), and Positive Development Type (26.3%). The predictive factors of adolescents’ active health behaviors included family functioning, digital health literacy, class cadre status, father’s educational level, exercise habits, dietary habits, electronic product usage, frequency of electronic product use, and frequently focused online information types. It is recommended that families and educators develop and implement targeted interventions based on the relevant factors to enhance adolescents’ active health behaviors.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-48179-x.

Introduction

Adolescence is a period of adventure and thrill-seeking. The asynchrony among age, cognitive ability, and psychosocial development stages leads to variations in adolescents’ capacity to consider the potential risks and consequences associated with their behaviors¹. During this period, adolescents not only make lifestyle choices but also establish behavioral patterns that impact their current and future health. According to a report by the United Nations Children’s Fund (UNICEF), approximately 70% of premature deaths among adolescents are linked to health risk behaviors that emerge during adolescence, such as unhealthy dietary behaviors, low physical activity, substance abuse, and problematic social media use^2–4. Moreover, these behaviors currently exhibit a clustered and co-occurring distribution^5,6. Therefore, there is an urgent need to implement interventions focused on promoting active health behaviors to reduce the occurrence of health risk behaviors at the source. Active health behaviors⁷ refer to the active development of a health literacy system in various aspects of oneself and the cultivation of positive behavioral habits. Through these actions, individuals acquire sustained health capabilities, a pleasant physical and mental state, and achieve sound social adaptability. With active health knowledge as the guarantee, a positive mindset and social relationships as the support, and a sound social adaptation state as the target, individuals can enhance their health status and comprehensive literacy, as well as reduce the occurrence of health-risk behaviors. Therefore, it is of crucial importance to help adolescents establish awareness of active health, improve their level of health behaviors, and support their healthy growth.

The family is not only a crucial environment for adolescents’ growth and development but also serves as the foundation for promoting their positive development⁸. According to Family Systems Theory, the various subsystems and elements within a family interact with one another, forming a social-ecological system. These interactions reflect interdependent, mutually conditional, and mutually influential relationships among subsystems and constituent elements. Such dynamic interactions profoundly influence the growth and development of children and adolescents⁹. Studies have shown that in an environment with healthy family functioning, parents’ positive lifestyle habits directly impact their children, who in turn tend to exhibit better mental health, lower criminal tendencies, and higher overall health levels¹⁰. Conversely, impaired family functioning may lead to various adaptability problems in adolescents¹¹. Therefore, improving family functioning is critically important for promoting adolescents’ active health behaviors.

Currently, although social media activities have become an indispensable part of adolescents’ lives¹², their cognition and behaviors are still in the developmental stage. This results in variations in adolescents’ ability to consider the potential risks and consequences associated with their behaviors, when facing various life events, they are prone to encountering obstacles in developing active health behaviors⁹. Existing studies have shown that adolescents’ inappropriate use of the Internet can lead to unhealthy lifestyles¹³. Digital health literacy¹⁴ refers to an individual’s ability to search for, understand, evaluate, apply, share, and create health information by integrating personal and social factors as well as technical constraints within a dynamic environment. Individuals with this literacy can respond more scientifically to health opportunities and risks in the digital environment, thereby reducing health harm caused by misinformation or inappropriate behaviors. This literacy can effectively moderate the impact of Internet use on unhealthy behaviors. However, adolescents with low digital health literacy, when using social media, not only independently increase the occurrence of internalizing and externalizing problems but also significantly raise the risk of comorbidity between the two¹⁵. Therefore, improving adolescents’ digital health literacy is a key factor in promoting their adoption of active health behaviors.

The present study

Previous studies have primarily focused on urban adolescents, with insufficient attention given to rural adolescents. Moreover, these studies have predominantly adopted variable-centered approaches (e.g., multiple regression, structural equation modeling) to explore the relationships among active health behaviors, family functioning, and digital health literacy. However, such methods are limited in their ability to capture heterogeneity within groups. Therefore, investigating the diverse characteristics within the rural adolescent population can more accurately elucidate the patterns and influencing factors of their active health behaviors¹⁶. Latent Profile Analysis (LPA), a person-centered approach, can identify distinct subgroups within a population and reveal more nuanced patterns of relationships among variables¹⁷. LPA has been widely applied in social and behavioral science research¹⁸, including studies in mental health¹⁹, nursing²⁰, elderly populations²¹, and injury epidemiology²². Based on this, the present study employs LPA to classify adolescents’ active health behaviors and explore the factors influencing these behaviors across different adolescent subgroups, with the aim of providing insights to promote healthy adolescent development.

Materials and methods

Study design

This study was structured as a cross-sectional survey and complied with the STROBE guidelines for cross - sectional studies.

Participants

This study collected data from 1,100 middle school students at two rural schools in Anhui Province in December 2024. The inclusion criteria were as follows: (1) middle school students present at school on the day of the survey; and (2) those who provided informed consent and voluntarily agreed to participate in the study. The exclusion criteria were: (1) students on leave of absence or who had dropped out of school; and (2) students who had participated in related research projects.

Sample size

Based on Kendall’s sample size estimation method, the sample size for multivariable analysis should be 10 to 20 times the total number of independent variables²³. A multiplier of 20 was used for calculation in this study. Given that the questionnaire comprised 24 items, the estimated sample size was 480 cases. With a 20% allowance for invalid questionnaires, the minimum required sample size was 600 cases. Additionally, when the Bayesian Information Criterion (BIC) was used as the primary indicator for model selection, the sample size needed to meet the corresponding requirements. Moreover, the robustness of variable classification in latent profile analysis is influenced by sample size²⁴, which suggests that the sample size should be approximately 500 cases. Therefore, by comprehensively considering these two sample size calculation methods and the actual number of students at the survey sites, the minimum sample size was determined to be 600 participants. In summary, a total of 1,100 samples were obtained in this study, exceeding the minimum sample size requirement.

Data collection

In December 2024, the research team emailed the principals of two rural middle schools in Anhui Province, China, outlining the research objectives, content, and researcher information. After obtaining approval, the team distributed written informed consent forms to the parents of adolescents through the head teachers of each class, along with relevant questionnaires assessing whether the students had recently participated in research related to adolescent emotional or behavioral problems. The team also confirmed the specific date and time for their visit to conduct the questionnaire survey in the classrooms. Parents were required to electronically return the signed consent forms before the researchers’ second visit. Due to the academic pressure on ninth-grade students, the study participants were limited to adolescents in seventh and eighth grades.

On the designated date, the research team conducted a second visit to the schools and held a pre-survey training session for the investigators. This training aimed to ensure a consistent understanding of the questionnaire content and survey procedures, clarify ethical requirements such as privacy protection and informed consent, and prevent practices that could compromise data objectivity (e.g., suggestive questioning and subjective judgments). Subsequently, questionnaires were distributed to eligible adolescents in a classroom setting. Students were seated individually at separate desks to ensure independent completion and were informed that the survey could be completed anonymously. The entire process lasted approximately 15–20 min. If any student had questions, they could raise their hand, and an investigator would approach their desk to provide on-site clarifications using standardized response scripts. For the 14 adolescents absent on the survey day, make-up surveys were collectively arranged within one week in consultation with the head teachers of the respective classes. Based on findings from relevant preliminary surveys, 36 students who had participated in similar research projects were directed to engage in independent learning and were not provided with the questionnaires.

Upon completion, the questionnaires were checked on-site, and adolescents were asked to complete any missing items. After collection, the research team collaborated with another researcher to assess their validity. Questionnaires were deemed invalid if they contained incomplete responses, logical inconsistencies, incorrect or misunderstood answers, patterned responses, or identical selections across multiple questions. In cases of discrepancies, the team consulted a third researcher to reach a final decision. Of the 1,100 distributed questionnaires, no missing data were present in the final sample used for analysis, as incomplete or logically inconsistent responses were excluded during data cleaning, resulting in an effective recovery rate of 99.36%. Since no missing values were present in the analytic dataset, missing data handling procedures (e.g., FIML) were not required in the LPA. This study received approval from the Yanbian University Medical Ethics Committee (Ethics No.: 20241223). No adolescents or their parents declined to participate in the study.Informed consent was obtained from the parents or legal guardians of all minors under 16 years old, the minors themselves, and the school authorities.

Measures

Adolescent active health behavior scale (AAHES)

Active health behaviors were assessed using the Adolescent Active Health Behavior Scale (AAHES)⁷, which measures the level of active health behavior among general adolescent population. This scale comprises 5 dimensions with a total of 24 items (total score ranging from 24 to 96), including 5 items for “health responsibility”, 5 items for “physical activity”, 6 items for “nutritional diet”, 4 items for “mental health” and 4 items for “temperance and self-discipline”. A 4-point Likert scale was used for scoring, with higher scores indicating a higher level of active health behavior among adolescents. The Cronbach’s α coefficient of the original scale was 0.946, while the Cronbach’s α coefficient of the total scale in this study was 0.922.

General family functioning scale (FF)

Family functioning was assessed using the General Family Functioning Scale (FF)²⁵, which measures the family functioning scores of rural middle school students. This scale is unidimensional and consists of 6 items, employing a 5-point Likert scale, where higher scores indicate better overall family functioning. The Cronbach’s α coefficient of the original scale was 0.878, while the Cronbach’s α coefficient of the total scale in this study was 0.910.

The mobile-eHealth literacy scale (m-eHEALS)

Digital health literacy was assessed using the Mobile-eHealth Literacy Scale (m-eHEALS)²⁶, which measures the digital health literacy abilities of rural middle school students. This scale consists of 3 dimensions with a total of 12 items: 3 items for “self-perception”, 5 items for “information acquisition”, and 4 items for “interactive evaluation”. A 5-point Likert scale was employed, with higher total scores indicating a greater level of digital health literacy. The Cronbach’s α coefficient of the original scale was 0.91, while the Cronbach’s α coefficient of the total scale in this study was 0.862.

Covariates

It included three domains: (1) General demographic information (gender, age, grade, class cadre status, only child status, single-parent family background, parental educational level (Both paternal and maternal education levels were included in the analysis. In the absence of income measures, paternal education level was considered a reasonable representation of the family’s socioeconomic resources, which has been commonly used as a proxy indicator of socioeconomic status in adolescent health research and was therefore incorporated into the model as a key socioeconomic factor); (2) Health-related status (family history of chronic diseases [chronic diseases were classified according to the International Classification of Diseases, 10th Revision (ICD-10)]²⁷, exercise habits, dietary habits); (3) Internet usage (electronic product usage, frequency of electronic product use, frequently focused online information types). The binary variables mentioned above include gender, grade, class cadre status, only child status, single-parent family background, family history of chronic diseases, and electronic product usage; all other variables are categorical with multiple categories.

Statistical analysis

The 24 items of the AAHES scale used in this study were treated as continuous variables and analyzed using raw scores without standardization. Given that the AAHES items were measured on a 4-point Likert scale and demonstrated approximately normal distributions in preliminary analyses, they were treated as continuous indicators in the LPA. Here are the specified Mplus estimation settings for the Latent Profile Analysis (LPA) in this study: the default MLR (Maximum Likelihood with Robust Standard Errors) estimator was used, with robust standard errors applied for standard error estimation. Two hundred initial sets of random starting values were specified, and the top 10 optimal sets were selected for the final-stage iteration to avoid local optima. No parameter constraints were imposed, and all model parameters were freely estimated. Additionally, parallel processing was enabled using PROC = 4 to improve computational efficiency. Model fit test indicators included three information evaluation criteria: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-adjusted Bayesian Information Criterion (aBIC). Lower values of these three indicators indicate a better model fit. Additionally, the Entropy value was used as an indicator to measure classification accuracy, with a range of 0 to 1. An Entropy value > 0.8 means the classification accuracy exceeds 90%²⁸. SPSS 28.0 software was employed for statistical analysis of the data. Qualitative data were presented as frequencies and percentages. Following subgroup identification through LPA, chi-square tests examined demographic differences across latent classes. Significant covariates, family function, and digital health literacy were subsequently incorporated into multivariate logistic regression models, with class membership as the outcome variable. A P-value < 0.05 was considered statistically significant.

Results

The general demographic data among rural middle school students

The general demographic data indicated that there were 573 male students (52.4%) and 520 female students (47.6%); the age of the participants ranged from 11 to 15 years. Among them, 591 students (54.1%) were in the first grade of middle school, and 502 students (45.9%) were in the second grade. Detailed information is presented in Table 1.

Table 1.

Characteristics of valid sample distribution.

Variable	Numbers	Proportions (%)
Gender
Male	573	52.4
Female	520	47.6
Age(years)
<12	359	32.8
13	552	50.5
>14	182	16.7
Grade
The first grade of middle school	591	54.1
The second grade of middle school	502	45.9
Class cadre status
Yes	251	23
No	842	77
Only child status
Yes	87	8
No	1006	92
Single-parent family background
Yes	87	8
No	1006	92
Father’s educational level
Junior college degree or above	82	7.5
Senior high school or technical secondary school	354	32.4
Junior high school or below	657	60.1
Mother’s educational level
Junior college degree or above	80	7.3
Senior high school or technical secondary school	315	28.8
Junior high school or below	698	63.9
Living status
Living with both parents	231	21.1
Living with either father or mother	429	39.2
Living with grandparents (paternal or maternal)	373	34.1
Others	60	5.5
Family history of chronic diseases
Yes	32	2.9
No	1061	97.1
Exercise habits
Every day	584	53.4
3–5 times per week	270	24.7
1–2 times per week	200	18.3
Never	39	3.6
Dietary habits
Mainly light diet	75	6.9
Balanced mix of meat and vegetables	782	71.5
Prefer oily and spicy foods	186	17
Delivery food and others	50	4.6
Electronic device usage
Yes	899	82.3
No	194	17.7
Frequency of electronic product use
Every day	176	16.1
3–5 times per week	250	22.9
1–2 times per week	611	55.9
Never	56	5.1
Frequently focused online information types
Health information	542	49.6
Weight loss and fitness	241	22
Nutrition and health preservation	310	28.4

Latent profile analysis of adolescents’ active health behaviors

Table 2 shows that starting from the 1-class model, the number of profiles was gradually increased to explore latent profiles of 1 to 6 classes, and the model fit indicators for each model. As the number of classes increased, the information criteria (AIC, BIC, and aBIC) gradually decreased. The Bootstrap Likelihood Ratio Test (BLRT) indicated statistical significance for all models (all P-values < 0.01), while the Lo-Mendell-Rubin Test (LMR) showed statistical significance only for the 2-class, 3-class, and 5-class models (all P-values < 0.01), thus excluding the 4-class and 6-class models. The two-class solution was overly general and failed to capture meaningful heterogeneity in active health behaviors. In contrast, the five-class solution yielded at least one very small class, suggesting limited stability and interpretability. Considering both the interpretability and parsimony of the models comprehensively, the 3-class model (C1, C2, C3) was ultimately selected as the optimal fitting model. Based on the above analysis, the average probability of participants being assigned to each of the 3 latent classes was calculated. For each individual, the probability of being assigned to their corresponding class was above 0.95, and the probability of being assigned to the other classes was below 0.1. This indicates that the results of the 3-class model are reliable (Table 3).

Table 2.

Comparison of model metrics for different categories.

Model	AIC	BIC	aBIC	LMP (P)	BLRT (P)	Entropy	Category Probabilities (%)
1	74397.593	74637.434	74484.975	-	-	-	-
2	68844.086	68805.844	68573.979	<0.001	<0.001	0.916	55.7/44.3
3	66633.735	67123.410	66812.140	<0.001	<0.001	0.916	27.5/46.1/26.4
4	66257.046	66871.638	66480.962	0.5810	<0.001	0.852	21.1/27.8/33.1/18.0
5	65491.219	66230.728	65760.647	<0.001	<0.001	0.941	8.8/19.5/5.4/40.6/25.7
6	65142.216	66006.642	65457.155	0.1325	<0.001	0.898	8.8/16.2/5.4/30.1/23.7/15.8

Table 3.

Average posterior probabilities.

Model	Number of Individuals	Proportion of Each Category(%)	Membership Probability			x̄ ± s
			C1	C2	C3
Class1	301	27.5	96.6	3.4	0	0.28 ± 0.447
Class2	504	46.1	2.3	96.0	1.7	0.46 ± 0.499
Class3	288	26.3	0	3.2	96.8	0.26 ± 0.441

Naming of latent classes for adolescents’ active health behaviors

The score distribution of active health behavior latent classes across the 24 items is shown in Fig. 1. Latent classes were named based on the score characteristics of each class across all items. Class 1: Scores on all items were at a relatively low level, so it was named the “Negative Coping Type”. This class included 301 participants, accounting for 27.5% of the total sample. Class 2: Average scores on all items were at a moderate level, and there were greater fluctuations across dimensions compared to the other two classes. Thus, it was named the “Unstable Type”, with 504 participants (46.1% of the total sample). Class 3: Average scores on all items were at a relatively high level, indicating that individuals in this class held a more positive attitude toward active health behaviors. Therefore, it was named the “Positive Development Type”, consisting of 288 participants (26.3% of the total sample).

Fig. 1.

Characteristic distribution of the three latent classes of adolescents’ active health behaviors.

Differences in dimensions of adolescents’ active health behaviors among different latent types

A multivariate analysis of variance (MANOVA) was conducted on individuals from the three latent classes across various dimensions of active health behaviors to verify the heterogeneity of the classification results. As shown in Table 4, statistically significant differences were observed among the three groups across all five dimensions. The overall results were as follows: Wilks’ λ = 0.152, F (10,2174)=339.211, P<0.001, = 0.453. Specifically, significant differences were found in the dimensions of health responsibility [F(2,1090)＝1045.478, P < 0.001, = 0.657], physical activity [F(2,1090)＝926.847, P<0.001, = 0.453], nutritional diet [F(2,1090)＝673.745, P<0.001, = 0.553], mental health [F(2,1090)＝388.468, P<0.001, =0.416], and self - discipline [F(2,1090)＝382.294, P<0.001, = 0.412]. Subsequently, post - hoc multiple comparisons were performed using the least significant difference (LSD) method within a multivariate general linear model. The results revealed that there were statistically significant differences among the three groups across all five dimensions (all P < 0.05). This indicates that the latent classification of active health behaviors in this study is valid and can effectively distinguish levels of active health behaviors among individuals.

Table 4.

Test and comparison of scores in each dimension of active health behavior among adolescents with different types of active health behavior ().

Statistic	Health Responsibility	Physical Activity	Nutritional Diet	Mental Psychology	Temperance and Self-Discipline
C1	9.08 ± 2.739	8.70 ± 2.357	13.60 ± 3.109	8.57 ± 2.524	9.28 ± 2.395
C2	12.47 ± 2.257	12.57 ± 2.474	18.57 ± 2.635	11.33 ± 2.427	11.89 ± 2.221
C3	17.87 ± 2.071	17.16 ± 2.251	21.66 ± 2.348	13.97 ± 1.998	14.15 ± 1.664
F	1045.478***	926.847***	613.745***	388.468***	382.294***
	0.657	0.630	0.553	0.416	0.412
C1 vs. C2	−3.389	−3.863	−4.971	−2.753	−2.610
C1 vs. C3	−8.788	−8.455	−8.065	−5.397	−4.870-
C2 vs. C3	−5.400	−4.592	−3.094	−2.645	−2.260
Post-hoc comparison	C1< C2< C3	C1< C2< C3	C1< C2< C3	C1< C2< C3	C1< C2< C3

C1 Negative Coping Type, C2 Unstable Type, C3 Positive Development Type.

*P<0.05, ***P<0.001.

Univariate analysis of each latent class of adolescents’ active health behaviors

Univariate analysis results showed that among demographic indicators, statistically significant differences were observed across the three latent classes of adolescents’ active health behaviors in terms of class cadre status, father’s educational level, living status, exercise habits, dietary habits, electronic product usage, frequency of electronic product use, and frequently focused online information types (all P < 0.05), as shown in Table 5.

Table 5.

Univariate analysis of latent profile of adolescents’ active health behaviors.

Variable	C1 (n = 301)	C2 (n = 504)	C3 (n = 288)	χ2
Gender
Male	167(55.5)	250(49.6)	156(54.2)	3.087
Female	134(44.5)	254(50.4)	132(45.8)
Age(years)
<12	104(34.6)	159(31.5)	96(33.3)	5.051
13	144(47.8)	253(50.2)	155(53.8)
>14	53(17.6)	92(18.3)	37(12.8)
Grade
The first grade of middle school	165(54.8)	259(51.4)	167(58.0)	3.305
The second grade of middle school	136(45.2)	245(48.6)	121(42.0)
Class cadre status
Yes	44(14.6)	127(25.2)	80(27.8)	17.047 ***
No	257(85.4)	377(74.8)	208(72.2)
Only child status
Yes	22(7.3)	50(9.9)	15(5.2)	5.795
No	279(92.7)	454(90.1)	273(94.8)
Single-parent family background
Yes	30(10.0)	38(7.5)	19(6.6)	2.506
No	271(90.0)	466(92.5)	269(93.4)
Father’s educational level
Junior college degree or above	17(5.6)	31(6.2)	34(11.8)	18.532 **
Senior high school or technical secondary school	83(27.6)	167(33.1)	104(36.1)
Junior high school or below	201(66.8)	306(60.7)	150(52.1)
Mother’s educational level
Junior college degree or above	19(6.3)	38(7.5)	23(8.0)	4.217
Senior high school or technical secondary school	78(25.9)	144(28.6)	93(32.3)
Junior high school or below	204(67.8)	322(63.9)	172(59.7)
Living status
Living with both parents	52(17.3)	102(20.2)	77(26.7)	16.797 **
Living with either father or mother	114(37.9)	195(38.7)	120(41.7)
Living with grandparents (paternal or maternal)	114(37.9)	184(36.5)	75(26.0)
Others	21(7.0)	23(4.6)	16(5.6)
Family history of chronic diseases
Yes	11(3.7)	16(3.2)	5(1.7)	2.106
No	290(96.3)	488(96.8)	283(98.3)
Exercise habits
Every day	118(39.2)	277(55.0)	189(65.6)	83.690 ***
3–5 times per week	70(23.3)	129(25.6)	71(24.7)
1–2 times per week	87(28.9)	88(17.5)	25(8.7)
Never	26(8.6)	10(2.0)	3(1.0)
Dietary habits
Mainly light diet	29(9.6)	27(5.4)	19(6.6)	83.135 ***
Balanced mix of meat and vegetables	165(54.8)	373(74.0)	244(84.7)
Prefer oily and spicy foods	76(25.2)	89(17.7)	21(7.3)
Delivery food and others	31(10.3)	15(3.0)	4(1.4)
Electronic product usage
Yes	266(88.4)	423(83.9)	210(72.9)	25.885 ***
No	35(11.6)	81(16.1)	78(27.1)
Frequency of electronic product use
Every day	84(27.9)	76(15.1)	16(5.6)	83.788 ***
3–5 times per week	65(21.6)	139(27.6)	46(16.0)
1–2 times per week	144(47.8)	267(53.0)	200(69.4)
Never	8(2.7)	22(4.4)	26(9.0)
Frequently focused online information types
Health information	148(49.2)	228(45.2)	166(57.6)	16.697 **
Weight loss and fitness	79(26.2)	113(22.4)	49(17.0)
Nutrition and health preservation	74(24.6)	163(32.2)	73(25.3)

C1 The Negative Coping Type, C2 The Unstable Type; C3 Positive Development Type.

*P<0.05, **P<0.01, ***P<0.001.

Multivariable analysis of latent profile of adolescents’ active health behaviors

Multivariable analysis results showed that when different categories of active health behaviors were taken as dependent variables (with the Positive Development Type as the reference group), and variables with statistical significance in the univariate analysis, family function (raw score) and digital health literacy (raw score) were included as independent variables for multivariable logistic regression analysis. This study did not apply a formal correction for multiple testing. Instead, we prioritized effect sizes and the consistency of results to reduce the risk of Type I error. For model fit, the following indices were reported: χ² = 687.611, df = 40, P < 0.001; Cox & Snell R² = 0.467; Nagelkerke R² = 0.530; McFadden R² = 0.296. The Nagelkerke R² of 0.530 indicates a moderate explanatory power of the model. Overall, the model shows acceptable fit and performs well in distinguishing between outcome categories. The results indicated that the Negative Coping Type was primarily characterized by low family functioning, poor digital health literacy, non-holding of class cadre cadre, father’s educational level (junior high school or below), exercise habits (never), dietary habits (delivery food and others), frequency of electronic product use (every day), 3–5 times per week. The Unstable Type was mainly associated with low family functioning, poor digital health literacy, frequently focused online information types (nutrition and health preservation), frequency of electronic product use (every day, 3–5 times per week)(P < 0.05 for all above associations). For details, see Table 6.

Table 6.

Multivariable logistic regression analysis of factors related to latent profile of adolescents’ active health behaviors.

IV		B	SE	Waldχ2	OR(95%CI)	P
C1VC3	Constant	20.077	1.736	133.727		<0.001 *
	Family function(FF)	−0.260	0.026	101.257	0.771(0.733,0.811)	<0.001*
	Digital health literacy (m-eHEALS)	−0.262	0.020	164.684	0.769(0.739,0.801)	<0.001*
	Class cadre status
	Yes	−1.029	0.283	13.224	0.357(0.205,0.622)	<0.001*
	No	-	-	-	1.000(−)	-
	Father’s educational level
	Junior college degree or above	−0.878	0.470	3.492	0.415(0.165,1.044)	0.062
	Senior high school or technical secondary school	−0.632	0.251	6.364	0.532(0.325,0.869)	0.012*
	Junior high school or below	-	-	-	1.000(−)	-
	Exercise habits
	Every day	−2.273	0.863	6.940	0.103(0.019,0.559)	0.008*
	3–5 times per week	−2.049	0.878	5.443	0.129(0.023,0.721)	0.020*
	1–2 times per week	−0.558	0.893	0.391	0.572(0.099,3.295)	0.532
	Never	-	-	-	1.000(−)	-
	Dietary habits
	Mainly light diet	−2.308	0.820	7.926	0.099(0.020,0.496)	0.005*
	Balanced mix of meat and vegetables	−2.132	0.692	9.489	0.119(0.031,0.460)	0.002*
	Prefer oily and spicy foods	−1.086	0.747	2.115	0.338(0.078,1.459)	0.146
	Others	-	-	-	1.000(−)	-
	Electronic product usage
	Yes	0.727	0.354	4.219	2.069(1.034,4.141)	0.040*
	No	-	-	-	1.000(−)	-
	Frequency of electronic product use
	Every day	2.345	0.725	10.446	10.429(2.516,43.227)	0.001*
	3–5 times per week	1.590	0.682	5.439	4.906(1.289,18.676)	0.020*
	1–2 times per week	0.718	0.631	1.292	2.050(0.595,7.067)	0.256
	Never	-	-	-	1.000(−)	-
C2VC3	Constant	10.502	1.443	52.943		<0.001 *
	Family function(FF)	−0.136	0.022	39.232	0.873(0.837,0.911)	<0.001*
	Digital health literacy (m-eHEALS)	−0.131	0.015	76.295	0.878(0.852,0.904)	<0.001*
	Frequency of electronic product use
	Every day	1.440	0.503	8.208	4.220(1.576,11.299)	0.004*
	3–5 times per week	1.176	0.441	7.122	3.241(1.367,7.688)	0.008*
	1–2 times per week	0.376	0.389	0.934	1.456(0.679,3.122)	0.334
	Never	-	-	-	1.000(−)	-
	Frequently focused online information types
	Health information	−0.460	0.207	4.939	0.631(0.421,0.947)	0.026*
	Weight loss and fitness	−0.233	0.259	0.810	0.792(0.477,1.316)	0.368
	Nutrition and health preservation	-	-	-	1.000(−)	-

Reference group: Positive Development Type; The reference category for all categorical covariates was set as the last category. IV independent variable, B Unstandardized Coefficient, SE Standard Error, Waldχ² Wald Chi-Square, P Probability Value, OR(95%CI) Odds Ratio (95% Confidence Interval). ^*Significant at the 0.05 level.

Discussion

Adolescence is a critical period for developing healthy lifestyle habits. To improve the health status of rural middle school students, this study found that their active health behaviors can be categorized into three types: Negative Coping Type (27.5%), Unstable Type (46.1%), and Positive Development Type (26.3%). It also examined the relevant significant factors. Based on this, schools, families, and health departments in rural areas need to attach importance to the holistic development of adolescents, and implement targeted interventions that consider the characteristics of the three behavior types and their predictive factors, thereby comprehensively improve the health level of adolescents.

This study focused on rural middle school students as the individual unit to explore the heterogeneity of their active health behaviors. The results showed that there were 3 latent classes of active health behaviors among middle school students, which was consistent with similar studies on adolescents’ health behaviors at home and abroad^29,30, but there were differences in specific classifications. Yang Qinwen et al.²⁹ divided the health-risk behavior patterns of students aged 11–18 in Shanghai into three groups: a low-risk group across all behaviors, a psychological addiction group, and a substance addiction group. Webster et al.³⁰ classified the health behavior patterns of students from two British schools into four types (high, low, moderate, and poor sleep and diet) based on physical activity, screen time, sleep quality, dietary habits, and mental health. In contrast, the present study identified three latent classes: Negative Coping Type (27.5%), Unstable Type (46.1%), and Positive Development Type (26.3%). Differences in classification results may be attributed to variations in research regions, cultural backgrounds, or the variables included in the models.

In this study, only 26.3% of middle school students belonged to the Positive Development Type. Students in this type had the highest scores across all dimensions of active health behaviors, with significantly higher autonomy than those in other groups. They could internalize health needs into daily habits and maintain their health status by strengthening their advantages (e.g., motivating other students). The Unstable Type accounted for a high proportion of 46.1%. Students in this type showed poor stability in health behaviors, were easily affected by external factors, and had weak awareness of health. Therefore, the focus should be on addressing shortcomings and improving stability—for instance, designing fragmented exercise programs to tackle insufficient physical activity and enhancing health education for them. Additionally, 27.5% of middle school students were classified into the Negative Coping Type. These students had the lowest scores in all dimensions of active health behaviors, with a complete lack of initiative. Moreover, multi-dimensional risks were prone to accumulate, forming a vicious cycle of health problems. The priority for this type lies in building a foundation and implementing intensive interventions: through family and school interventions, basic health issues should be resolved first, followed by gradual cultivation of proactive awareness.

All three types in this study scored lowest in the mental health dimension. Therefore, greater emphasis should be placed on designing intervention strategies targeting the mental health, while considering the behavioral characteristics and psychological needs of each student type. For example, students classified as the Positive Development Type with stress management and enhancing psychological resilience; those identified as the Unstable Type should be guided to focus on the connection between psychological cognition and health behaviors; and students of the Negative Coping Type should be encouraged to rebuild their psychological foundation through accessible, low-threshold psychological interventions. Simultaneously, efforts must be made to elevate the priority of mental health initiatives in rural areas and to deeply integrate psychological support with health behavior guidance, thereby preventing psychological deficiencies from impeding overall health improvement.

Univariate and multivariable logistic regression analyses in this study identified class cadre status, father’s educational level, living status, exercise habits, dietary habits, electronic products usage, frequency of electronic product use, and frequently focused online information types as significant predictors.

1. Sociodemographic and school factors

The results of this study reveal that middle school students with fathers who have higher educational levels and those who serve as class cadres are more likely to be categorized as the Positive Development Type, consistent with the findings of previous studies^31,32. Relevant research³³ indicates that parents with lower educational levels tend to adopt outdated parenting styles when responding to their children’s psychological changes during adolescence and struggle to implement appropriate coping strategies, which constitutes a risk factor for their children’s low levels of health behaviors. Additionally, parents’ educational levels influence their parenting styles, and scientific, reasonable parenting approaches can promote the healthy and comprehensive development of children³⁴. In contrast, students who hold class cadre positions tend to have higher levels of self-efficacy, which can extend to the health domain—enabling them to be more confident in managing unhealthy habits such as staying up late and being picky eaters. These factors collectively create favorable conditions for the development of healthy behaviors, ultimately leading to relatively higher levels of health behaviors³⁵. Therefore, for parents with relatively low educational levels, communities or schools can be leveraged to disseminate parenting knowledge in an accessible manner. Collaborations with professional institutions (e.g., psychological counseling centers, maternal and child health hospitals) can be established to provide free consulting services, thereby helping parents promote the healthy and holistic development of their children through scientific and appropriate parenting practices. Meanwhile, students who serve as class cadres should be encouraged to support their peers’ development by setting positive examples, organizing relevant activities, and facilitating mutual supervision, which in turn helps maintain their own health status.

In the univariate analysis, students living with only one parent (either father or mother) or with grandparents (paternal or maternal) demonstrated a higher proportion of classification as the Unstable Type. A study conducted among urban adolescents yielded the same result³⁶. However, this association did not reach statistical significance in the logistic regression analysis, possibly due to the confounding effects of other variables. Studies have shown³⁷ that the family is one of the most important agents of socialization for children, with different family members playing unique roles in promoting child development. Parental absence may hinder the effective fulfillment of nurturing functions, while grandparents often face limitations in educational capacity and inadequate use of digital technologies. Consequently, middle school students in these living situations are more likely to be classified as Unstable.

• 2)Behavioral factors

The present study found that middle school students who engaged in more frequent weekly physical activity and maintained healthy dietary habits were more likely to be classified as the Positive Development Type, and they also exhibited higher levels of active health behaviors, consistent with findings from previous studies^38,39. Survey results indicated that only 23.3% of middle school students exercised more than three days per week⁴⁰. Regular physical activity not only plays a crucial role in preventing obesity, regulating blood pressure, and inhibiting the development of hypertension^41,42 but also positively impacts the metabolic health of middle school students⁴³. Meanwhile, balanced and healthy dietary habits can reduce anxiety levels in adolescents⁴⁴ and influence the physical development and overall health of children and junior high school students⁴⁵. Conversely, unhealthy dietary habits hinder the establishment of good long-term health patterns⁴⁶. Families and schools can collaborate to supervise and encourage middle school students to participate in daily physical activities, optimize food offerings in school canteens, and regularly distribute family dietary guidelines, thereby transforming healthy eating from a passive necessity into an active choice.

Compared to middle school students who used electronic devices more frequently per week, those who never used them or with lower usage frequency were more likely to be categorized as the Positive Development Type, which was consistent with the findings of Wang⁴⁷. According to the survey, urban adolescents tend to use the Internet more for learning and development, benefiting from more adequate supervision and guidance. In contrast, rural minors often experience relatively insufficient Internet management, with most of their online time spent playing games or browsing short videos⁴⁸. Studies have indicated that a higher degree of internet dependence not only elevates the risk of suicidal behaviors but also raises the likelihood of engaging in additional health-risk behaviors (e.g., smoking and drinking) when it progresses to Internet Addiction Disorder (IAD)⁴⁵. Therefore, it is advisable to strengthen educational guidance on electronic device usage in schools, enhance daily supervision within families, and improve internet café regulation in communities, thus enabling the internet serves as a tool to improve health literacy and prevent health-risk behaviors such as smoking and drinking associated with internet addiction.

Notably, middle school students of the Unstable Type were more inclined to focus primarily on nutrition and health preservation, whereas those identified as the Positive Development Type paid attention to a broader range of health information. This finding aligns with the results reported by Svestkova⁴⁹. Health information⁵⁰ encompasses multidimensional health elements related to public health, diseases, and health preservation, providing systematic and scientific guidance for individuals’ active health behaviors; in contrast, focusing exclusively on nutrition and health preservation may limit the comprehensiveness and scientific rigor of health behaviors due to cognitive bias. Research has demonstrated that individuals who actively seek and disseminate health information tend to exhibit higher levels of health literacy⁵¹. Therefore, it is recommended to implement comprehensive health information science outreach programs targeting unstable middle school students to expand their health cognition, encourage active learning and dissemination of health knowledge, overcome the narrow focus on nutrition and health maintenance, and facilitate their transition to the Positive Development Type.

• 3)Family and digital environment

The results of this study demonstrated that middle school students classified as the Negative Coping Type and Unstable Type were predominantly associated with poor family functioning and low levels of digital health literacy. In contrast, those with higher levels of family functioning and digital health literacy exhibited greater engagement in active health behaviors, consistent with prior research findings^8,52,53. The sound establishment and effective functioning of fundamental family roles contribute to reducing individuals’ internet addiction⁵⁴, alleviating academic anxiety⁵⁵, and promoting healthy lifestyles⁵². Conversely, when families fail to provide effective emotional support and a positive interaction environment, individuals are more likely to be influenced by the negative emotions, attitudes, and inappropriate behaviors of family members, leading them to unconsciously adopt and replicate these behavioral patterns. Over time, this may develop into overt practical problems such as aggression, disciplinary violations, and emotional outbursts⁵⁶. Furthermore, when confronted with a vast array of health resources of varying quality, individuals lacking the necessary skills to access, evaluate, and make informed decisions regarding electronic health information may face increased health risks⁵⁷. Research indicates that good digital health literacy helps alleviate personal psychological stress; higher digital health literacy is associated with a lower risk of depressive symptoms^58,59 and a more pronounced reduction in psychological symptoms^60,61, thereby contributing to better overall health. Therefore, digital health literacy is considered a fundamental factor for adolescents in maintaining a healthy lifestyle and good health status^62,63.

Rural students often face issues such as being left-behind children, guardians (mostly grandparents) with limited health knowledge, insufficient family health atmosphere, and vulnerability to misleading information like folk remedies, rumors, and vulgar health content. Therefore, efforts should first be made to build a supportive healthy family environment: establishing a “cloud-based parent-child health link,” popularizing simple health knowledge, and integrating healthy behaviors into daily life. Secondly, digital health education activities should be carried out to help rural students identify the authenticity of health information, and digital health tools focusing on common health problems in rural areas should be developed. At the same time, family and school interventions should cooperate with each other to form a closed loop of “students learn knowledge → families practice actions → schools promote reinforcement”, ensuring that family support and digital health literacy effectively translate into practice, thereby improving adolescent health outcomes.

Limitations and future research

First, this study adopted a cross-sectional survey design, whose inherent methodological limitations that cannot be overlooked. Future research should consider using a longitudinal design combined with trajectory analysis to more accurately assess causal relationships among variables. Second, the Active Health Behavior Scale measures subjective perceptions, and its results may not be entirely consistent with objective behavioral performance. This limitation should be fully taken into account when interpreting the findings of this study. Finally, the research was conducted only in two schools in Anhui Province with a convenience sampling method, which restricted the sample size and the diversity of participants. It is recommended that future studies adopt a multicenter design to expand the applicability and generalizability of the results.

Conclusion

The active health behaviors of rural middle school students can be classified into three types: the Negative Coping Type (27.5%), the Unstable Type (46.1%), and the Positive Development Type (26.3%). Factors predictive of adolescents’ active health behaviors include class cadre status, father’s educational level, living status, exercise habits, dietary habits, electronic product usage, frequency of electronic product use, frequently focused online information types, family functioning, and digital health literacy. Therefore, schools and families should pay close attention to students’ behavioral changes, provide support and guidance, and help students develop positive lifestyles and healthy habits based on the characteristics of different types and the significant factors identified. In addition, medical and health institutions should provide education and resources tailored to the specific attributes of schools, enhance students’ awareness of the risks associated with unhealthy behaviors at different academic stages, and contribute to the prevention and reduction of such behaviors.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AAHES

The Adolescent Active Health Behavior Scale

FF

General Family Functioning Scale

m-eHEALS

The Mobile-eHealth Literacy Scale

Author contributions

Zhenzhen Zhang, Yuhan Xu: Conceptualized and designed the study, data analysis, manuscript writing. Jinzhen Jin, Yinhe Xuan and Xiangdan Shen: Refinement of the study and critical revisions of the manuscript. All authors reviewed the manuscript.

Funding

The project did not receive any funding.

Data availability

Although the raw data cannot be shared publicly due to privacy and ethical restrictions, the Mplus syntax used for latent profile analysis and the analysis codebook are available as supplementary materials to enable reproducibility of the results.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All procedures performed in the study involving human participants were in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. All experimental protocols has obtained the review and approval of the Yanbian University Medical Ethics Committee (Ethics No.: 20241223). In this study, informed consent was obtained from all parents or legal guardians of minors under the age of 16 for their children’s participation in the research.