Clustering of Risk Behaviors and their Social Determinants among Primary School Learners in Beijing, China: A Cross-sectional Study.

BACKGROUND: Studies in developed countries reveal that poor lifestyle choices triggering diseases typically cluster among children. However, there is insufficient evidence on the clustering of risk behaviors among children in developing countries. This study aimed to determine the clustering of risk behaviors and their social determinants among 4 th -and 5 th -grade learners in Beijing, China.
METHODS: The sample comprised of 967 learners from six primary schools enrolled migrant and resident learners by two-stage stratified cluster sampling. Prevalence denoted the risk behaviors and their clustering. A log-linear model was used to explore the clustering patterns. Ordinal logistic regression determined the influence of demographic characteristics, school environment, and family context on behavioral clustering.
RESULTS: The prevalence of none, one, two, and three or more risk factors was 61.2%, 20.0%, 10.8%, and 8.1% for infectious diseases and 46.0%, 30.6%, 15.4%, and 8.0% for chronic diseases, respectively. Some behaviors appeared dependent and were more likely to be observed together. The three most influential factors for infectious diseases were school type (odds ratio [OR] =4.47, 95% confidence interval [CI] 3.00-6.66), school located in an inner suburb (OR = 0.27, 95% CI 0.18-0.38), and gender (OR = 0.56, 95% CI 0.42-0.74). Regarding risk behaviors for chronic diseases, clustering was not associated with household registration status and number of appliances, but was significantly associated with school type (OR = 5.36, 95% CI 3.72-7.73), school located in an inner suburb (OR = 0.59, 95% CI 0.43-0.81), and gender (OR = 0.61, 95% CI 0.47-0.78). School environment variables were the most significant contributor to the number of risk behaviors.
CONCLUSIONS: The characteristics of schools enrolling migrants and residents influenced the number of risk behaviors. Therefore, improved school conditions and integrated behavioral interventions are particularly recommended for health promotion.

INTRODUCTION

Unhealthy behaviors contribute significantly toward the development of diseases.[1234] Moreover, people typically maintain childhood behavioral patterns.[567] Studies in developed countries reveal that poor lifestyle choices triggering noncommunicable diseases typically co-occur or “cluster” among children and adolescents.[89101112] These provide valuable information for interventions against risk behaviors. However, there is insufficient evidence on the clustering of risk behaviors among children in developing countries. Meanwhile, developing countries contend with the burden of infectious and chronic diseases[1314] less so than risk behaviors for infectious diseases (RBID) among children and adolescents.

Social variables, such as family socioeconomic status (SES) and family structure, are known to contribute toward risk behaviors;[151617] therefore, they contribute to risk behavior clustering. A low SES, which incorporates a low income and mothers’ educational levels, is associated with the clustering of risk behaviors.[81819] Therefore, interventions for risk behaviors must focus on disadvantaged children.

Rapid development in China is continuing to increase the number of rural-urban migrants. China's rural-urban migrant population (henceforth, “migrant population”) is defined as individuals leaving their rural residences for urban cities for a certain period (more than 6 months) without changing household registration. In 2012, the migrant population was approximately 236 million.[20] According to 2010 data, 20.8% of the migrant population were children under 14 years old[21] with living conditions characterized by instability, overcrowded housing, poor sanitation, children attending disadvantaged schools, and social and cultural isolation.[22] These constitute risk factors for health and the cultivation of healthy behaviors. However, the clustering of risk behaviors and their determinants among children in China is under-researched.

This study examined the clustering of risk behaviors for infectious and chronic diseases, and the accompanying influence of social determinants among learners in the high grades in primary schools enrolling migrants and residents in Beijing.

METHODS

Sample

We employed two-stage stratified cluster sampling. First, we stratified schools into urban districts, inner suburbs, and outer suburbs. Beijing comprises 16 districts, including 6 urban districts, 6 inner suburbs, and 4 outer suburbs. We randomly selected one district from each of these. Second, we randomly selected two medium-scale (800–1200) primary schools from each district, which enrolled migrant and resident learners. We selected two grade 4 and 5 classes from each school randomly, with all learners in those classes invited to participate. Among the 1021 learners recruited, 41 declined to participate, and 13 were excluded because their survey forms were unusable. The final sample size was 967, yielding a response rate of 94.7%. The data were collected from February to May 2011 by trained fieldworkers.

In this study, migrants were those whose household registration records (hukou) were not in Beijing. Given their different hukou statuses, learners were divided into two groups: Migrants and residents.

Informed consent was obtained from the participating learners and their parents by letters. This study was approved by the Institutional Review Board at Peking University (IRB00001052-11025).

Measures

The questionnaires were self-completed, and the research assistants were available to clarify arising questions.

Self-reported risk RBID included six items, namely: “Do you wash hands before eating?” “Do you wash hands after defecation?” “Do you wash fruit before eating?” “Do you spit in public?” “Do you drink tap water that is not boiled?” and “Do you share towels with others?” Response choices were “always,” “sometimes,” and “never.” “Sometimes” or “never” were risk behaviors for the first three questions, whereas “always” was a risk behavior for the latter three. Related risk behavior interventions for infectious diseases are recommended in health education courses in Chinese primary schools, particularly public schools.

Risk behaviors for chronic diseases (RBCD) and growth included: Smoking; passive smoking (PS) (≥4 days/week); breakfast (≤3 days/week); and consumption of vegetables (≤3 days/week), fruits (≤3 days/week), and milk (≤3 days/week). We defined “smoking” as having ever smoked or tried[23] and PS as exposure to other people's tobacco smoke for more than 15 min/day. The frequency method was used to measure risk behaviors, and arbitrary cut-offs were adopted to identify whether behaviors posed health risks.

Clustering of risk behaviors was demonstrated by number of risk behaviors and clustering patterns of individual risk behaviors as previous studies.[2425]

Data analysis

Statistical analysis was conducted through SPSS (version 13.0, SPSS Inc., USA). Descriptive analysis summarized the outcome of demographic characteristics and prevalence of risk behaviors with the percentage for categorical variables and mean for continuous variables. Chi-square tests were used to compare prevalence and clustering for the number of risk behaviors among the different groups, with a significance threshold of 0.05. A log-linear model was used to explore the clustering patterns. The main effects model would be a priority, assuming that the risk behaviors were mutually independent of one another. If it did not adequately fit, interactions between variables would be considered to establish an appropriate model. Ordinal logistic regression analysis demonstrated the social determinants of the number of risk behaviors. Four models were explored to distinguish the role of school and family context. Model 1 included only demographic characteristics (i.e., age and gender). Variables relating to the school environment were added to Model 2. Model 3 introduced family context variables to Model 1. All variables were included in Model 4. The test of parallel lines was used to judge whether the data fit the ordinal logistic regression model (P > 0.05). Goodness of fit was tested using − 2 log likelihood (−2 LL) (P < 0.05) and Pearson's Chi-square test (P > 0.05). Cox and Snell's pseudo R2 were calculated to assess the models’ interpretation power.

RESULTS

Demographic characteristics

As can be seen in Table 1, boys outnumbered girls in all the districts. Two-thirds of the learners (age range: 9–14 years; mean = 11.25 years) were in public schools; 70.2% of the subjects were migrants. The education level of guardians in the outer suburbs was the lowest; only 13.5% had attended high school. The percentage of ownership of <3 household appliances increased with distance from an urban area, with 21.6% in an urban district, 54.9% in an inner suburb, and 62.2% in an outer suburb.

Table 1

Demographic characteristics by district

Variables	Items	Urban district n (%)	Inner suburb n (%)	Outer suburb n (%)	Total n (%)
Gender	Boys	122 (50.8)	227 (55.2)	160 (53.9)	509 (53.7)
	Girls	118 (49.2)	184 (44.8)	137 (46.1)	439 (46.3)
Age (years)	Mean	11.26 ± 0.9	11.10 ± 1.1	11.46 ± 1.2	11.25 ± 1.1
School type	Public	245 (100.0)	192 (45.8)	218 (71.9)	655 (67.7)
	Private	0 (0.0)	227 (54.2)	85 (28.1)	312 (32.3)
Household registration	Resident	92 (37.6)	134 (32.0)	62 (20.5)	288 (29.8)
	Migrant	153 (62.4)	285 (68.0)	241 (79.5)	679 (70.2)
Number of appliances	≤3	53 (21.6)	230 (54.9)	189 (62.4)	472 (48.8)
	≥4	192 (78.4)	189 (45.1)	114 (37.6)	495 (51.2)
Guardians’ education level	Secondary school and below	168 (68.6)	343 (81.9)	262 (86.5)	773 (79.9)
	High school and above	77 (31.4)	76 (18.1)	41 (13.5)	194 (20.1)

Multiple risk behaviors and clustering patterns

Table 2 shows the RBID and RBCD. Risk behavioral frequencies were higher in suburban than urban districts. The most prevalent RBID and RBCD were “not washing hands after defecation (WHAD)” and “consuming milk ≤3 days/week,” respectively. The prevalence of none, one, two, and three or more RBID was 61.2%, 20.0%, 10.8%, and 8.1%, respectively, and for RBCD it was 46.0%, 30.6%, 15.4%, and 8.0%, respectively.

Table 2

Multiple risk behaviors and their clustering by district

Items	Urban n (%)	Inner suburb n (%)	Outer suburb n (%)	χ2	P
RBID
WHBE	31 (12.7)	53 (12.6)	63 (20.8)	10.70	0.005
WHAD	34 (13.9)	58 (13.8)	81 (26.7)	23.49	0.000
WFBE	14 (5.7)	36 (8.6)	42 (13.9)	11.18	0.004
SIP	5 (2.0)	15 (3.6)	16 (5.3)	4.01	0.135
Drinking tap water that is not boiled	8 (3.3)	25 (6.0)	44 (14.5)	27.44	0.000
Sharing towels with others	20 (8.2)	59 (14.1)	66 (21.8)	20.20	0.000
Number of risk behaviors
0	176 (71.8)	280 (66.8)	136 (44.9)	54.39	0.000
1	41 (16.7)	68 (16.2)	84 (27.7)
2	16 (6.5)	44 (10.5)	44 (14.5)
≥3	12 (4.9)	27 (6.4)	39 (12.9)
RBCD
Ever smoking	10 (4.1)	29 (6.9)	37 (12.2)	13.26	0.001
Passive smoking: ≥4 days/weeks	43 (17.6)	101 (24.1)	70 (23.1)	4.10	0.129
Breakfast: ≤3 days/weeks	11 (4.5)	49 (11.7)	41 (13.5)	13.07	0.001
Vegetable consumption: ≤3 days/weeks	7 (2.9)	21 (5.0)	11 (3.6)	2.04	0.361
Fruit consumption: ≤3 days/weeks	18 (7.3)	78 (18.6)	48 (15.8)	15.81	0.000
Milk consumption: ≤3 days/weeks	41 (16.7)	140 (33.4)	102 (33.7)	24.90	0.000
Number of risk behaviors
0	151 (61.6)	186 (44.4)	108 (35.6)	49.22	0.000
1	63 (25.7)	118 (28.2)	115 (38.0)
2	26 (10.6)	69 (16.5)	54 (17.8)
≥3	5 (2.0)	46 (11.0)	26 (8.6)

RBID: Risk behaviors for infectious diseases; RBCD: Risk behaviors for chronic diseases; WHBE: Not washing hands before eating; WHAD: Not washing hands after defecation; WFBE: Not washing fruit before eating; SIP: Spitting in public.

The main effects in the log-linear model showed inadequate fit for the data and interactions between two of the variables demonstrated an appropriate model. As shown in Table 3, this was deemed to have a better fit to the data (χ2 = 45.42, P = 0.538; χ2 = 61.64, P = 0.125). In RBID, not washing hands before eating with WHAD (odds ratio [OR] =6.06, 95% CI 4.00–9.17), spitting in public (SIP) with drinking unboiled tap water (DUTW) (OR = 4.57, 95% CI 2.10–9.98), and WHAD with not washing fruit before eating (OR = 4.09, 95% CI 2.48–6.75) were behaviors more likely to be observed together. In RBCD, FC with MC (OR = 5.49, 95% CI 3.75–8.03), HB with VC (OR = 2.64, 95% CI 1.23–5.69), and PS with FC (OR = 2.58, 95% CI 1.77–3.77) were more likely to occur dependently.

Table 3

Clustering patterns for risk behaviors

RBID	OR	95% CI	P	RBCD	OR	95% CI	P
WHBE × WHAD	6.06	4.00–9.17	0.000	ES × FC	2.57	1.51–4.37	0.001
WHBE × WFBE	2.60	1.54–4.39	0.000	PS × FC	2.58	1.77–3.77	0.000
WHBE × STWO	2.26	1.44–3.55	0.000	HB × VC	2.64	1.23–5.69	0.013
WHAD × WFBE	4.09	2.48–6.75	0.000	HB × FC	2.09	1.26–3.45	0.004
WHAD × SIP	2.67	1.29–5.53	0.008	HB × MC	1.98	1.26–3.10	0.003
WHAD × STWO	2.96	1.92–4.55	0.000	VC × MC	2.36	1.22–4.55	0.010
WFBE × DUTW	2.67	1.48–4.83	0.001	FC × MC	5.49	3.75–8.03	0.000
SIP × DUTW	4.57	2.10–9.98	0.000
SIP × STWO	2.37	1.10–5.12	0.028
DUTW × STWO	3.41	2.03–5.73	0.000

RBID: Likelihood ratio goodness-of-fit tests χ2 = 45.42, P = 0.538; RBCD: Likelihood ratio goodness-of-fit tests χ2 = 61.64, P = 0.125. WHBE: Not washing hands before eating; WHAD: Not washing hands after defecation; WFBE: Not washing fruit before eating; SIP: Spitting in public; DUTW: Drinking unboiled tap water; STWO: Sharing towel with others; ES: Ever smoking; PS: Passive smoking; HB: Having breakfast; VC: Vegetables consumption; FC: Fruits consumption; MC: Milk consumption; OR: Odds ratio; CI: Confidence interval; RBID: Risk behaviors for infectious diseases; RBCD: Risk behaviors for chronic diseases.

Social determinants of risk behavior clustering

Ordinal logistic analysis results showed that no matter the model for RBID or RBCD, the data fit the ordinal logistic regression model well, with P > 0.05 in the test of parallel lines; the goodness of fit was also fine with P < 0.05 in − 2 LL and P > 0.05 in Pearson's Chi-square test [Tables 4 and 5]. Results of R2 showed the explanatory power was lowest in Model 1 and highest in Model 4. R2 in Model 2 was greater than it was in Model 3, whether in RBID or RBCD, and disclosed the important role of the school. Once all variables were included in Model 4, more RBID was exhibited by boys, migrants, those living in the outer suburbs, private school learners, learners with less educated guardians, and those from families with fewer household appliances. The three most influential factors were school type (OR = 4.47, 95% CI 3.00–6.66), school located in an inner suburb (OR = 0.27, 95% CI 0.18–0.38), and gender (OR = 0.56, 95% CI 0.42–0.74). Regarding RBCD, behavioral clustering was not associated with household registration status and number of appliances, but was significantly associated with school type (OR = 5.36, 95% CI 3.72–7.73), school located in an inner suburb (OR = 0.59, 95% CI 0.43–0.81), and gender (OR = 0.61, 95% CI 0.47–0.78).

Table 4

Ordinal logistic regression by number of RBID (reference: No risk behaviors)

Variables	Items	Model 1			Model 2			Model 3			Model 4
		OR	95% CI	P	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
Age	Continuous	1.43	1.27, 1.62	0.000	1.13	1.00, 1.28	0.059	1.22	1.07, 1.38	0.002	1.08	0.95, 1.23	0.248
Gender	Girls (reference: Boys)	0.59	0.45, 0.77	0.000	0.53	0.41, 0.71	0.000	0.62	0.47, 0.81	0.000	0.56	0.42, 0.74	0.000
School location	Urban				0.55	0.38, 0.79	0.002				0.64	0.44, 0.95	0.028
	Inner suburb (reference: Outer suburb)				0.23	0.16, 0.32	0.000				0.27	0.18, 0.38	0.000
School type	Private (reference: Public)				6.48	4.49, 9.36	0.000				4.47	3.00, 6.66	0.000
Household registration	Migrants (reference: Residents)							3.00	2.07, 4.36	0.000	1.86	1.25, 2.77	0.002
Guardians’ education level	High school and above (reference: Middle school and below)							0.58	0.40, 0.86	0.007	0.61	0.41, 0.90	0.014
Number of household	More than 4 (reference: No more than 3)							0.53	0.40, 0.71	0.000	0.73	0.53, 1.00	0.049
appliances
Test of parallel lines (P)			0.620			0.650			0.406			0.471
Model fitting (−2 LL, P)			152.2, 0.000			460.1, 0.000			448.1, 0.000			848.9, 0.000
Goodness of fit			34.8, 0.573			196.5, 0.053			208.9, 0.641			555.3, 0.968
(Pearson’s χ2, P)
Pseudo R2 Cox and Snell			0.060			0.198			0.152			0.221

OR: Odds ratio; CI: Confidence interval; RBID: Risk behaviors for infectious diseases.

Table 5

Ordinal logistic regression by number of RBCD and growth (reference: No risk behaviors)

Variables	Items	Model 1			Model 2			Model 3			Model 4
		OR	95% CI	P	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
Age	Continuous	1.49	1.33, 1.66	0.000	1.21	1.08, 1.37	0.002	1.30	1.16, 1.46	0.000	1.18	1.05, 1.34	0.006
Gender	Girls (reference: Boys)	0.65	0.51, 0.83	0.001	0.60	0.47, 0.77	0.000	0.64	0.50, 0.83	0.001	0.61	0.47, 0.78	0.000
School location	Urban				0.61	0.43, 0.86	0.005				0.69	0.48, 0.99	0.042
	Inner suburb (reference: Outer suburb)				0.54	0.39, 0.73	0.000				0.59	0.43, 0.81	0.001
School type	Private (reference: Public)				6.48	4.66, 9.01	0.000				5.36	3.72, 7.73	0.000
Household registration	Migrants (reference: Residents)							1.98	1.46, 2.69	0.000	1.19	0.85, 1.66	0.305
Guardians’ education level	High school and above (reference: Middle school and below)							0.68	0.49, 0.94	0.021	0.71	0.51, 0.99	0.046
No. of household appliances	More than 4 (reference: No more than 3)							0.51	0.39, 0.66	0.000	0.78	0.59, 1.05	0.098
Test of parallel lines (P)			0.438			0.141			0.183			0.156
Model fitting (−2 LL, P)			169.6, 0.000			465.6, 0.000			527.2, 0.000			958.4, 0.000
Goodness of fit (Pearson’s χ2, P)			45.0, 0.173			174.0, 0.320			222.7, 0.380			578.4, 0.877
Pseudo R2 Cox and Snell			0.069			0.224			0.145			0.232

OR: Odds ratio; CI: Confidence interval; RBCD: Risk behaviors for chronic diseases; −2 LL: −2 log likelihood.

DISCUSSION

This study assessed the clustering of multiple risk behaviors for infectious and chronic diseases and their social determinants among learners in schools enrolling migrants and residents in Beijing, China. The prevalence of none, one, two, and three or more risk factors was 61.2%, 20.0%, 10.8%, and 8.1%, respectively, for infectious diseases, and 46.0%, 30.6%, 15.4%, and 8.0%, respectively, for chronic diseases.

Regarding some RBID, the participants engaged in more healthy behaviors than those in a nationwide, representative Chinese survey conducted in 2005, wherein elementary school learners washed their hands 72.5% (before eating) and 75.4% (after defecation) of the time.[26] The survey showed that, regarding some RBCD, daily breakfast and vegetable consumption was at 84.0% and 67.0%, respectively, among the learners; this was 79.4% and 81.2% in our study. As most Beijing schools provide lunch in recent years, vegetable intake was boosted in students.

Children who begin smoking earlier are more likely to develop severe nicotine addiction.[27] In a longitudinal Canadian study, the prevalence among 10–11-year-old children of having ever smoked was 6% from 2000 to 2001,[15] compared to 7.9% in our study. Data from the United Kingdom showed that 12–16-year-old children's breakfast intake ≥5 days a week was at 76.7%.[28] However, these studies are not comparable with the current one because of differences in the survey periods, regions, and population.

Although few studies have focused on behaviors for infectious diseases, there are a number of studies that have examined the clustering of RBCD.[11122429] Our results from the log-linear models showed clear evidence that risk behaviors for infectious and chronic diseases were not independent, but were instead clustered together, such as washing hands before eating and after defecation, SIP and DUTW, fruit consumption and milk intake, and breakfast and vegetable consumption. Although the analytic techniques used to define clustering differed across studies,[30] it was clear that some risk behaviors clustered.[31] These findings would have substantial implications for health promotion. Multiple behavior interventions (i.e., targeting more than one health behavior simultaneously) would have the potential for a much greater public health impact compared to single behavior interventions. The limited clustering of smoking with the other factors was very different from the other studies. A possible reason for this was that the risk behaviors we investigated were more about diet and nutrition, and less about physical activity and alcohol consumption, which were observed clustering with smoking by other researchers. In addition, the population we studied was primary school students, among whom the smoking rate is comparatively low.[32] However, our results also showed that interventions for smoking and fruit consumption together would be good for primary school students.[33]

The literature shows that many factors – including gender, ethnicity, grade level, family, schooling, and community contexts – influence risk behaviors.[343536] In our study, we considered demographic characteristics (i.e., gender and age), school environment (i.e., school type: Private or public; location: Urban district or suburb), and family context as determinants of the clustering of risk behaviors. The family context in our study referred to SES, including place of household registration (Beijing or other), number of household appliances, and guardians’ education levels. Here, household appliances (i.e., television, washing machine, air conditioner, refrigerator, and personal computer) were considered a measure of wealth because collecting data on income and savings is a sensitive issue in China and learners may not give accurate income-related information. Therefore, obtaining nonmonetary indicators of wealth is a simpler and considerate proxy.[3738] As expected, demographic characteristics, school environment, and family context variables contributed toward the number of risk behaviors, with the school environment variables being the most significant contributor.

As learners spend most of their time in school, school conditions, interventions, and teachers’ and peers’ support reportedly facilitate the formation of healthy behaviors.[3940] In China, urban schools’ physical conditions and health education capacity are slightly better than those of suburban schools. However, while the conditions in urban schools are generally better, the gaps between urban and suburban schools are not qualitative but quantitative, and more differences exist between public and private schools enrolling migrant children. In public schools, health education courses are regularly offered by qualified teachers, and basic hygiene amenities (sufficient taps, washrooms, and classroom lights), desks and chairs adjusted according to child height, and a balanced diet can be provided. In most public schools, learners annually undergo professional health checks, participate in vaccination programs, and receive counseling; none of these is provided in most private schools recruiting migrant learners. Private schools have fewer studying facilities and lower teaching capacity.[2141] Similarly, in our study, schools’ characteristics mattered more than the family context as R2 showed.

The family context, reflected by the SES, including household income, parents’ educational level, and occupation, was another important determinant of risk behaviors.[154243] In our study, number of appliances and guardians’ educational levels (constituting the family context) significantly influenced the number of RBID, possibly because of misinformation regarding these and related prevention measures among learners of a lower SES.[44] Regarding RBCD, the guardians’ education level significantly influenced behavioral clustering but proved less significant than the school environment. According to many studies, teenagers from low SES backgrounds are likely to report inadequate fruit and vegetable consumption.[45464748] However, comparisons are limited because differences in estimates may be confounded by differences in the risk factors investigated, data-gathering instruments, and analysis. The schools’ characteristics in our study could be particularly influential because urban public primary schools in Beijing provide nutritional lunch – comprising fruit, vegetables, and milk – to learners.

Our study primarily suggests that integrated interventions may be optimal in facilitating healthy lifestyles. However, demographic and social determinants can directly influence the number of unhealthy behaviors. Since clustered risk behaviors may be particularly harmful to children's health,[49] public health interventions should target three or more risk behaviors according to socioeconomic conditions.

This study had several limitations. It was comprised of a cross-sectional survey, making it impossible to make causal inferences about risk behaviors. Moreover, the survey comprised self-reported data by the learners, increasing the possibility of misreporting. Further, we did not measure daily vegetable and fruit servings, but assessed the weekly frequency, with arbitrary cut-offs. However, we did maximize the likelihood of honest reporting by ensuring the learners’ anonymity.

Thus, despite the limitations, this study is one of the few to focus on social determinants of multiple health behaviors among learners in developing countries, particularly migrant learners. Our results highlight the extent of health promotion among learners.

In conclusion, characteristics of schools enrolling migrants and residents influenced the number of risk behaviors. Therefore, improved school conditions and integrated behavioral interventions are particularly important for health promotion.