There is evidence in the literature that lifestyle factors associated with energy balance influence weight status in adolescents(Reference Narciso, Silva and Rodrigues1). Diet is a key factor in energy balance regulation. High intake of energy-dense, nutrient-poor foods is associated with overweight and obesity(Reference Liberali, Kupek and Assis2). In contrast, high-quality diets(Reference Liberali, Kupek and Assis2), low levels of sedentary behaviour(Reference de Rezende, Rodrigues Lopes and Rey-López3), regular physical activity(Reference Jiménez-Pavón, Kelly and Reilly4) and adequate sleep(Reference Fatima, Doi and Mamun5) appear to be protective factors. No single factor can be identified as a universal causal factor in overweight/obesity, given that several behaviours and determinants at different levels contribute to this issue(Reference Narciso, Silva and Rodrigues1). Many of these behaviours are interrelated within individuals and may have synergistic and cumulative effects on overweight/obesity(Reference Leech, McNaughton and Timperio6).
Clustering of multiple lifestyle behaviours, also known as the study of lifestyle patterns, has been successfully applied to understand the co-occurrence of different behaviours(Reference Spring, Moller and Coons7).
Lifestyle patterns can be derived using exploratory data-driven methods(Reference Carvalho, Fonsêca and Nobre8). These approaches aim to aggregate individuals who have similar behaviours or group behaviours that are highly correlated. Consequently, these techniques allow investigating the cumulative effect of combined behaviours on a given outcome(Reference Leech, McNaughton and Timperio6).
Leech etal.(Reference Leech, McNaughton and Timperio6) conducted a narrative review examining the clustering of diet, physical activity, and sedentary behaviour in children and adolescents. According to the authors, the association between cluster patterns and overweight/obesity was inconclusive. Studies examining lifestyle patterns and overweight/obesity often do not assess sleep-related factors. However, sleep, diet, physical activity, and sedentary behaviour all interact and influence each other to impact health(Reference Alberga, Sigal and Goldfield9). A recent systematic review examined the associations between lifestyle patterns including diet, physical activity, sedentary behaviour, and sleep and adiposity in children. The authors concluded that unhealthy lifestyle patterns were more frequently associated with adiposity risk(Reference D’Souza, Kuswara and Zheng10).
These previous reviews investigated studies conducted with children and adolescents (5–18 years)(Reference Leech, McNaughton and Timperio6) or children (5–12 years) only(Reference D’Souza, Kuswara and Zheng10), covering two distinct stages of life. Different from childhood, adolescence is a high-risk phase for weight gain, characterised by critical changes in body composition and lifestyle-related behaviours(Reference Alberga, Sigal and Goldfield9). In adolescence, the participation in physical activity can reduce, particularly among girls(Reference Alberga, Sigal and Goldfield9). Furthermore, dietary habits are altered with increasing autonomy(Reference Demory-Luce, Morales and Nicklas11).
Considering that: (i) there is a lack of consistent evidence about the relationship between lifestyle patterns and overweight/obesity in adolescents(Reference Leech, McNaughton and Timperio6); (ii) this phase is critical for weight gain, mainly in girls(Reference Alberga, Sigal and Goldfield9); and (iii) adolescents with overweight/obesity may continue to be overweight/obesity during adulthood(Reference Simmonds, Llewellyn and Owen12), it is pertinent to explore the direction of associations between lifestyle patterns and overweight/obesity in adolescents. We conducted a systematic analysis aimed at demonstrating the associations between lifestyle patterns and overweight/obesity in adolescents overall and by sex.
Methods
This systematic review followed the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Table S1, online Supplementary Material)(Reference Moher, Liberati and Tetzlaff13). The protocol (CRD42020151085) was registered with the International Prospective Register of Systematic Reviews (PROSPERO).
Eligibility criteria
Studies that met predefined criteria based on PECOS (Participants, Exposure, Comparison, Outcome and Study design) elements were considered eligible for inclusion in this systematic review (Table S2, online Supplementary Material). Inclusion criteria were as follows: (a) adolescents aged ≥10 to ≤19 years (or mean age within this range) according to the definition of the WHO(14); (b) application of exploratory data-driven methods to identify lifestyle patterns (such as cluster analysis, principal components analysis, treelet transform, reduced rank regression and latent class analysis) and the assessment of the diet domain in conjunction with at least one of the following behavioural domains: physical activity, sedentary behaviour and sleep; and (c) weight status (overweight/obesity) as outcome, determined from age- and sex-specific BMI percentiles, BMI Z-scores, BMI standard deviation scores and BMI cut-off points proposed by the International Obesity Task Force (IOTF)(Reference Cole and Lobstein15), the WHO(Reference de Onis, Onyango and Borghi16), the US Centers for Disease Control and Prevention (CDC)(Reference Kuczmarski, Ogden and Guo17) or national references. Only observational (cross-sectional and prospective) studies were included. There were no language or time restrictions.
Exclusion criteria were as follows: (i) studies in children under 10 years of age, adults or seniors; or with well-characterised samples of non-healthy adolescents (e.g. individuals with type 2 diabetes, hypertension or eating disorders); (ii) studies that did not use exploratory data-driven methods to determine lifestyle patterns and that did not include diet domain; (iv) studies that did not include overweight/obesity as the outcome; and (v) conference proceedings, case reports and letters to the editor.
Information sources
Specific search strategies were created for the following databases (Table S3, online Supplementary Material): Latin American and Caribbean Health Sciences Literature (LILACS), Scopus, PubMed Central, and Web of Science. We had support from a librarian at the Federal University of Santa Catarina in the search process(Reference Lefebvre, Glanville, Briscoe, Higgins, Thomas and Chandler18). Descriptors came from Health Sciences Descriptors, Medical Subject Headings and words related to the subject. A search restriction for terms in the ‘title or abstract’ was made to increase the specificity of the systematic search, given the scope of the descriptors. An additional search of grey literature documents was performed using ProQuest Dissertations & Theses Global and Google Scholar; in Google Scholar, the search was restricted to the first 100 studies. The systematic search was conducted on 6 November 2019 and updated on 21 July 2020. The reference lists of full-text articles were visually screened to identify other relevant articles. When articles were not available online or in full text, we contacted the authors by email or through Research Gate.
Study selection
Search results were transferred to EndNote Web version X9, and duplicate hits were removed. Study selection was performed in three stages. First, two reviewers (LJP and LHM) independently screened the titles and abstracts of all identified records to identify potentially relevant articles. Then, the reviewers read in full all selected articles to determine which papers met the eligibility criteria. Articles that did not meet the eligibility criteria were excluded. In the third stage, the reviewers screened the reference lists of selected articles for other potentially relevant papers. Any discrepancy between the two reviewers was resolved by consensus with a third reviewer (PFH).
Data collection
Two reviewers (LJP and LHM) independently extracted the data. This process was guided by the use of a form previously prepared by the authors and subjected to a pilot test to ensure consistency across reviewers. Extracted data were subsequently compared for agreements and disagreements. Divergences were resolved by consensus with a third reviewer (PFH).
Data items
The following information was retrieved from selected studies: authors, year of publication, country, study design, survey year, age range or school grade, sample size, diet variables, diet assessment method, physical activity variables, physical activity assessment method, sedentary behaviour variables, sedentary behaviour assessment method, sleep variables, sleep assessment method, lifestyle patterns, lifestyle pattern assessment method, outcome indicator, outcome measurement method, cut-off reference, method of analysis, and associations identified between overweight/obesity and lifestyle patterns. Research funding data and conflicts of interest were also extracted from the articles.
Risk of bias assessment
The Joanna Briggs Institute critical appraisal tools were used to assess the quality of selected studies(19). The instrument consists of eight items: (1) eligibility criteria; (2) study subjects and setting; (3) validity and reproducibility of exposure measures; (4) criteria for patient diagnosis; (5) confounding factors; (6) strategies for dealing with confounding factors; (7) validity and reproducibility of the outcome measure; and (8) statistical analysis. We developed specific criteria for item scoring to facilitate the analysis (Table S4, online Supplementary Material). In Item 1, the authors from the selected studies should describe in detail whether adolescents with physical or mental disabilities, diseases, pregnancy, lactation or restrictive diet were excluded from the sample (Item 1). The authors also should provide a clear description of the sample studied, including sex, age or school grade, socio-economic status, year of the research, location, sampling, and sample size estimation (Item 2). The studies should clearly describe whether the instruments used to measure all exposure and outcome variables were subjected to validity and reproducibility tests with the same population of interest, presenting the respective reference. If the method used was considered a gold standard (i.e. objective measurement of weight and height, accelerometer), this assessment was not necessary (Items 3 and 7). We assessed whether the authors reported typical confounders such baseline characteristics (age, sex and socio-economic status) (Item 5). Finally, we considered appropriate studies those that used multivariate analysis adjusted (multivariate ANOVA and regression analysis) for typical confounders as a statistical method to evaluate associations (Items 6 and 8). Items are scored as yes, no, unclear or not applicable. Item 4 was excluded from analysis because it was not applicable to the nature of the selected studies. Thus, the risk of bias was determined using the other seven items of the instrument. Two reviewers (LJP and LHM) independently assessed each study and resolved disagreements with a third reviewer (PFH). For classification of the risk of bias, we calculated the proportion of ‘yes’ responses. The risk of bias was determined as ‘high’ when the study reached a ‘yes’ score up to 49 %, ‘moderate’ between 50 % and 69 %, and ‘low’ when it was above 70 %(Reference Hinnig, Monteiro and De Assis20). The results of risk of bias assessment are presented in Table S5, Supplementary Material.
Summary measures
Lifestyle patterns (principal independent variable) identified by exploratory data-driven methods and their associations with overweight/obesity (outcome) were described as OR or β1 coefficients and 95 % CI. Data were also subjected to univariate ANOVA and Pearson’s χ 2 tests.
Synthesis of results
Given the heterogeneity of methods used to assess associations between lifestyle patterns and overweight/obesity in the selected studies, it was not possible to perform a meta-analysis. Therefore, the results are described according to the Synthesis Without Meta-analysis (SWiM) guideline. When the characteristics of the studies are very varied to produce a meaningful summary estimate of the effect, alternative methods of summarising the results may be adopted, such as counting votes based on the direction of the effect. As such, SWiM provides guidance for reporting these methods and results(Reference Campbell, McKenzie and Sowden21).
A wide variety of lifestyle patterns were identified; we chose to categorise them according to the healthiness or unhealthiness of related behaviours (Table S6, online Supplementary Material). Healthy behaviours included presence/high levels of physical activity, healthy diet, and adequate sleep habits as well as low levels/absence of sedentary behaviour and low consumption of unhealthy foods. Unhealthy behaviours were defined as presence/high levels of sedentary behaviour, unhealthy diet and inadequate sleep habits as well as low levels/absence of physical activity and low consumption of healthy foods. Moderate behaviours were defined as intermediate levels of diet quality, sleep quality, physical activity and sedentary behaviour. Lifestyle patterns that included only healthy behaviours were classified as completely healthy and those that included only unhealthy behaviours as completely unhealthy. Lifestyle patterns characterised by at least two healthy behaviours and one unhealthy or moderately unhealthy behaviour were classified as predominantly healthy, whereas lifestyle patterns including at least two unhealthy behaviours and one healthy or moderately healthy behaviour were classified as predominantly unhealthy. Finally, lifestyle patterns characterised by an equal proportion of healthy and unhealthy behaviours were classified as mixed. Only studies that used multivariate analysis adjusted for confounders (multivariate ANOVA and regression analysis) to assess associations between lifestyle patterns and overweight/obesity were included in the synthesis of results. The direction of association was described as positive, inverse or none. Positive and inverse associations were only considered valid for studies reporting statistically significant associations (i.e. P < 0·05, zero not included in the 95 % CI for β1 or OR ≠ 1). To examine differences in the direction of associations, we recorded the number of positive, inverse or null associations reported in the studies. Subsequently, the number of associations was analysed by sex (based on data from studies that used sex stratification) and by risk of bias.
Results
Study selection
A total of 6017 articles were identified in the database search. Additionally, six articles were identified through other sources. After removal of duplicates, 3662 articles remained and were screened by title and abstract, revealing forty-three potentially relevant for eligibility assessment. Of these, twenty-two articles were excluded (Table S7, online Supplementary Material): seven did not focus on adolescents(Reference Kontogianni, Farmaki and Vidra22–Reference Leech, McNaughton and Timperio28), one article did not use exploratory data-driven methods to identify lifestyle patterns(Reference Werneck, Agostinete and Cayres29), nine used outcomes that did meet our inclusion criteria(Reference Maia30–Reference Busch, Van Stel and Schrijvers38) and five were not accessible(Reference Turner, Dwyer and Edwards39–Reference Boone, Gordon-Larsen and Adair43). Twenty-one articles were retained for systematic review (Fig. 1).
Fig. 1. Flowchart of literature search and selection criteria. Adapted from PRISMA.
Study characteristics
A detailed description of the main characteristics of selected studies is provided in Table 1. Of the twenty-one articles selected, one reported the results of two studies with different samples (from Europe and Brazil), so their characteristics are presented separately in this review(Reference Moreira, da Veiga and Santaliestra-Pasías44). Two articles refer to a single study but used different designs; therefore, sample characteristics are described once but associations with the outcome are presented separately(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46) . Twenty studies were cross-sectional(Reference Spengler, Mess and Schmocker27,Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Laxer, Brownson and Dubin45,Reference Boone-Heinonen, Gordon-Larsen and Adair47–Reference Marttila-Tornio, Ruotsalainen and Miettunen63) . Most studies were performed in European countries(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Landsberg, Plachta-Danielzik and Lange48–Reference Ottevaere, Huybrechts and Benser52,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54–Reference Nuutinen, Lehto and Ray56,Reference Wadolowska, Hamulka and Kowalkowska58–Reference dos Santos, Picoito and Loureiro60,Reference Veloso, Matos and Carvalho62–Reference Spengler, Mess and Mewes64) , three in the USA(Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Iannotti and Wang53,Reference Berlin, Kamody and Thurston61) , two in Brazil(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Dantas, dos Santos and Lopes57) and one in Canada(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46) . Sample sizes ranged from 173(Reference Sevil-Serrano, Aibar-Solana and Abos59) to 18 587(Reference Laxer, Brownson and Dubin45) subjects. In one study, data collection was conducted before 2000(Reference Boone-Heinonen, Gordon-Larsen and Adair47), fifteen between 2001 and 2010(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Landsberg, Plachta-Danielzik and Lange48–Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Nuutinen, Lehto and Ray56,Reference dos Santos, Picoito and Loureiro60–Reference Spengler, Mess and Mewes64) , and five between 2011 and 2017(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Dantas, dos Santos and Lopes57–Reference Sevil-Serrano, Aibar-Solana and Abos59) (Table 1). The statement of the funding and conflict of interest of the included studies can be found in Table S8, Supplementary Material. No notable concern about conflict of interest was observed from the studies.
Table 1. Characteristics of studies included in systematic review
PA, physical activity; SB, sedentary behavior; LP, lifestyle pattern; B, boys; G, girls; C, cross-sectional; Q, questionnaire; TV, television; NA, not applicable; FPAQ, Flemish physical activity questionnaire; T, time; LTPA, leisure-time physical activity; YRBS, youth risk behaviour survey; HELENA, healthy lifestyle in Europe by nutrition in adolescence; DIAT, dietary assessment tool; MVPA, moderate vigorous physical activity; IPAQ, international physical activity questionnaire; SF, short form; SF-FFQ4PolishChildren, multicomponent dietary questionnaire to assess food frequency consumption, nutrition knowledge and lifestyle in Polish schoolchildren; MoMo-PAQ, Motorik-Modul physical activity questionnaire; KIGGS, German health interview and examination survey for children and adolescents; HBSC, health behaviour in school-aged children; YLSBQ, Spanish version of youth leisure-time sedentary behaviour questionnaire; P, prospective; CA, cluster analysis; LCA, latent class analysis; PCA, principal component analysis; LPA, latent profile analysis; BIC, bayesian information criterion; AIC, akaike information criteria; a-BIC, adjusted bayesian information criterion; CAIC, consistent akaike information criterion; LMR, Lo–Mendell–Rubin test.
* Values for reliability reported or can be found through the reference(s) provided.
† Values for validity reported or can be found through the reference(s) provided.
‡ Also examined other behaviours (e.g. smoking, drugs and alcohol use, psychological factors, dieting behaviours, parental involvement and scholar aspects).
Four studies investigated lifestyle patterns related to diet, physical activity, sedentary behaviour and sleep(Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54–Reference Nuutinen, Lehto and Ray56,Reference Sevil-Serrano, Aibar-Solana and Abos59) , sixteen analysed diet, physical activity, and sedentary behaviour(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Landsberg, Plachta-Danielzik and Lange48,Reference van der Sluis, Lien and Twisk50–Reference Iannotti and Wang53,Reference Dantas, dos Santos and Lopes57,Reference Wadolowska, Hamulka and Kowalkowska58,Reference dos Santos, Picoito and Loureiro60–Reference Spengler, Mess and Mewes64) , and one assessed diet and physical activity only(Reference Sabbe, De Bourdeaudhuij and Legiest49). Five studies included other health-related behaviours, such substance use (marijuana use, smoking and binge drinking)(Reference Laxer, Brownson and Dubin45–Reference Landsberg, Plachta-Danielzik and Lange48,Reference dos Santos, Picoito and Loureiro60,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) , dieting behaviours(Reference Boone-Heinonen, Gordon-Larsen and Adair47) and parental involvement(Reference Boone-Heinonen, Gordon-Larsen and Adair47) (Table 1). Lifestyle patterns composed of dieting behaviours and parental involvement in isolation were not considered in this systematic review.
The methods used to measure behaviours varied across studies. One study used the 24-h recall with a 3-d record (two consecutive weekdays and one weekend day) to collect dietary data(Reference Perez-Rodrigo, Gil and Gonzalez-Gross55), two used 2-d non-consecutive 24-h recall(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Ottevaere, Huybrechts and Benser52) and nineteen used FFQ(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Seghers and Rutten51,Reference Iannotti and Wang53,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Nuutinen, Lehto and Ray56–Reference Spengler, Mess and Mewes64) . Only one study used accelerometer to measure physical activity and sedentary behaviour(Reference Sevil-Serrano, Aibar-Solana and Abos59). One study used face-to-face interview to assess physical activity, sedentary behaviour and sleep habits(Reference Perez-Rodrigo, Gil and Gonzalez-Gross55), and all others used self-report questionnaires. Across the studies, dietary variables ranged from specific food groups such as soft drinks and fruit juices(Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54), fruits, vegetables, and junk foods(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference van der Sluis, Lien and Twisk50,Reference Iannotti and Wang53,Reference Nuutinen, Lehto and Ray56,Reference Dantas, dos Santos and Lopes57,Reference dos Santos, Picoito and Loureiro60,Reference Veloso, Matos and Carvalho62,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) to dietary indices(Reference Landsberg, Plachta-Danielzik and Lange48,Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Seghers and Rutten51,Reference Ottevaere, Huybrechts and Benser52,Reference Sevil-Serrano, Aibar-Solana and Abos59,Reference Spengler, Mess and Mewes64) or patterns based on the whole diet(Reference Perez-Rodrigo, Gil and Gonzalez-Gross55). Physical activity variables assessed were total,(Reference Iannotti and Wang53,Reference Dantas, dos Santos and Lopes57,Reference dos Santos, Picoito and Loureiro60) moderate and vigorous-intensity activities(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Laxer, Cooke and Dubin46,Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Ottevaere, Huybrechts and Benser52,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Sevil-Serrano, Aibar-Solana and Abos59,Reference Berlin, Kamody and Thurston61–Reference Marttila-Tornio, Ruotsalainen and Miettunen63) , physical activity at school(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46,Reference Wadolowska, Hamulka and Kowalkowska58,Reference Berlin, Kamody and Thurston61) , outside of school or on leisure time(Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Landsberg, Plachta-Danielzik and Lange48,Reference van der Sluis, Lien and Twisk50,Reference Seghers and Rutten51,Reference Nuutinen, Lehto and Ray56,Reference Wadolowska, Hamulka and Kowalkowska58,Reference Veloso, Matos and Carvalho62,Reference Spengler, Mess and Mewes64) , organised sports/competitions(Reference Laxer, Brownson and Dubin45–Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Berlin, Kamody and Thurston61,Reference Spengler, Mess and Mewes64) , active commuting(Reference Landsberg, Plachta-Danielzik and Lange48,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55) , and housework(Reference Boone-Heinonen, Gordon-Larsen and Adair47). Most sedentary behaviour variables were time of watching television(Reference Moreira, da Veiga and Santaliestra-Pasías44) and computer(Reference Landsberg, Plachta-Danielzik and Lange48,Reference van der Sluis, Lien and Twisk50,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Wadolowska, Hamulka and Kowalkowska58) , screen time (e.g. television, video or DVD, computer, video console games, or mobile phone)(Reference Laxer, Brownson and Dubin45–Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Seghers and Rutten51–Reference Iannotti and Wang53,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55–Reference Dantas, dos Santos and Lopes57,Reference Sevil-Serrano, Aibar-Solana and Abos59–Reference Spengler, Mess and Mewes64) , or non-screen activities (e.g. homework)(Reference Seghers and Rutten51,Reference Ottevaere, Huybrechts and Benser52,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55) . The measure of sleep habits had the least variation, as most studies included a question about the number of hours that adolescents usually sleep at night. A single study sought to describe sleep habits through duration, discrepancy (sleep duration on weekend night – sleep duration on schools nights) and quality of sleep(Reference Nuutinen, Lehto and Ray56). Four studies reported both the reliability and validity for all assessed measures(Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Iannotti and Wang53,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Nuutinen, Lehto and Ray56) .
Cluster analysis was the most frequently used method to identify lifestyle patterns (n 18/21). The techniques used were a combination of hierarchical (Ward’s method) and non-hierarchical (k-means) clustering(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Ottevaere, Huybrechts and Benser52,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Dantas, dos Santos and Lopes57,Reference Sevil-Serrano, Aibar-Solana and Abos59,Reference Spengler, Mess and Mewes64) , k-means(Reference Sabbe, De Bourdeaudhuij and Legiest49–Reference Seghers and Rutten51,Reference Nuutinen, Lehto and Ray56,Reference Wadolowska, Hamulka and Kowalkowska58,Reference Veloso, Matos and Carvalho62,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) and two-step(Reference Landsberg, Plachta-Danielzik and Lange48,Reference dos Santos, Picoito and Loureiro60) . In addition, two studies applied latent class analysis(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46,Reference Iannotti and Wang53) , and one latent profile analysis(Reference Berlin, Kamody and Thurston61) (Table 1).
Risk of bias assessment
Of the articles included in the synthesis of the results(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference van der Sluis, Lien and Twisk50,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55–Reference dos Santos, Picoito and Loureiro60) , only six out of eleven provided clear eligibility criteria. Most articles clearly specified their study population (n 9/11). No study used reliable and valid measures for all exposure variables, and five assessed their outcome measures objectively. Although all studies have established strategies to deal with confounders, these factors were adequately accounted for in only nine studies. Nine studies used appropriate statistical analysis models. Of these studies, six had a low risk of bias (54·5 %)(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Dantas, dos Santos and Lopes57,Reference Wadolowska, Hamulka and Kowalkowska58,Reference dos Santos, Picoito and Loureiro60) , four moderate risk (36·4 %)(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46,Reference van der Sluis, Lien and Twisk50,Reference Nuutinen, Lehto and Ray56) and one high risk (9·1)(Reference Sevil-Serrano, Aibar-Solana and Abos59). Risk of bias assessment of the studies is presented in Table S5, Supplementary Material.
Lifestyle patterns
Table 2 describes the lifestyle patterns identified in the twenty-one studies evaluated. We found sixteen completely healthy lifestyle patterns(Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference van der Sluis, Lien and Twisk50,Reference Ottevaere, Huybrechts and Benser52–Reference Nuutinen, Lehto and Ray56,Reference Wadolowska, Hamulka and Kowalkowska58–Reference Marttila-Tornio, Ruotsalainen and Miettunen63) and fourteen completely unhealthy patterns.(Reference Boone, Gordon-Larsen and Adair43,Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference van der Sluis, Lien and Twisk50,Reference Seghers and Rutten51,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Nuutinen, Lehto and Ray56,Reference Wadolowska, Hamulka and Kowalkowska58,Reference Sevil-Serrano, Aibar-Solana and Abos59,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) Among completely healthy lifestyle patterns, six included the diet, physical activity, sedentary behaviour and sleep domains(Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54–Reference Nuutinen, Lehto and Ray56,Reference Sevil-Serrano, Aibar-Solana and Abos59) , although different methods were used to measure indicators. Three completely healthy lifestyle patterns included low substance use(Reference dos Santos, Picoito and Loureiro60,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) . Five of the fourteen completely unhealthy lifestyle patterns were determined on the basis of the four domains(Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Nuutinen, Lehto and Ray56,Reference Sevil-Serrano, Aibar-Solana and Abos59) . Two completely unhealthy lifestyle patterns included high substance use(Reference Marttila-Tornio, Ruotsalainen and Miettunen63). Predominantly healthy (n 24) and predominantly unhealthy (n 31) lifestyle patterns prevailed among adolescents. High levels of both physical activity and sedentary behaviour co-occurred in some predominantly unhealthy patterns(Reference Ottevaere, Huybrechts and Benser52,Reference Dantas, dos Santos and Lopes57,Reference Veloso, Matos and Carvalho62) . However, most of the predominantly unhealthy lifestyle patterns included behaviours such as low physical activity levels, high sedentary behaviour levels, low healthy food consumption and low/moderate unhealthy food consumption(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Ottevaere, Huybrechts and Benser52–Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Dantas, dos Santos and Lopes57,Reference Spengler, Mess and Mewes64) . Some lifestyle patterns were considered mixed (n 22), with a balance of healthy and unhealthy behaviours(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Laxer, Brownson and Dubin45,Reference Landsberg, Plachta-Danielzik and Lange48,Reference Seghers and Rutten51,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Wadolowska, Hamulka and Kowalkowska58–Reference dos Santos, Picoito and Loureiro60,Reference Veloso, Matos and Carvalho62,Reference Spengler, Mess and Mewes64) (Table 2).
Table 2. Summary of lifestyle patterns identified
n, absolute frequency; HD, healthy diet; UD, unhealthy diet; PA, physical activity; SB, sedentary behaviour; S, sleep; SU, substance use; ↑, high; ↓, low; ↔, moderate.
Adapted from D’Souza et al. (2020)(Reference Simmonds, Llewellyn and Owen12).
Associations with overweight and/or obesity
Table 3 provides details of the identified lifestyle patterns and their associations with overweight/obesity. In eleven studies, weight and height measurements were used to determine BMI(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Landsberg, Plachta-Danielzik and Lange48,Reference Ottevaere, Huybrechts and Benser52,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Dantas, dos Santos and Lopes57,Reference Wadolowska, Hamulka and Kowalkowska58,Reference Berlin, Kamody and Thurston61,Reference Spengler, Mess and Mewes64) . Nine studies used self-reported weight and height(Reference Laxer, Brownson and Dubin45,Reference van der Sluis, Lien and Twisk50,Reference Seghers and Rutten51,Reference Iannotti and Wang53,Reference Nuutinen, Lehto and Ray56,Reference Sevil-Serrano, Aibar-Solana and Abos59,Reference dos Santos, Picoito and Loureiro60,Reference Veloso, Matos and Carvalho62,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) , and one study used parent-reported weight and height(Reference Sabbe, De Bourdeaudhuij and Legiest49). Nine studies were based on IOTF cut-offs for overweight/obesity(Reference Landsberg, Plachta-Danielzik and Lange48,Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Seghers and Rutten51,Reference Ottevaere, Huybrechts and Benser52,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54–Reference Dantas, dos Santos and Lopes57,Reference Veloso, Matos and Carvalho62) , one on IOTF and Polish standards(Reference Wadolowska, Hamulka and Kowalkowska58), five on WHO(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Laxer, Brownson and Dubin45,Reference dos Santos, Picoito and Loureiro60,Reference Marttila-Tornio, Ruotsalainen and Miettunen63) , two on CDC(Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Iannotti and Wang53,Reference Berlin, Kamody and Thurston61) , and one on national reference values(Reference Spengler, Mess and Mewes64).
Table 3. Associations between lifestyle patterns and overweight and obesity in adolescents
LP, lifestyle pattern; CDC, Centers for Disease Control and Prevention; IOTF, International Obesity Task Force; HELENA, healthy lifestyle in Europe by nutrition in adolescence; MVPA, moderate vigorous physical activity; SSB, sugar-sweetened beverage; B, boys; G, girls; β1, coefficient beta.
* CI not described in the text, only in the figure.
Three studies used BMI as a continuous measure(Reference Laxer, Cooke and Dubin46,Reference van der Sluis, Lien and Twisk50,Reference Sevil-Serrano, Aibar-Solana and Abos59) , and one was based on BMI Z-scores(Reference Berlin, Kamody and Thurston61). Eight studies used Pearson’s χ 2 tests(Reference Landsberg, Plachta-Danielzik and Lange48,Reference Sabbe, De Bourdeaudhuij and Legiest49,Reference Seghers and Rutten51–Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Marttila-Tornio, Ruotsalainen and Miettunen63,Reference Spengler, Mess and Mewes64) , one used the Bose-Chaudhuri-Hocquenghem (BCH) method(Reference Berlin, Kamody and Thurston61), one used one-way ANOVA(Reference Veloso, Matos and Carvalho62) and another used multivariate ANOVA(Reference Sevil-Serrano, Aibar-Solana and Abos59) to describe lifestyle patterns according to weight status. Eleven studies used regression analysis to identify associations(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference van der Sluis, Lien and Twisk50,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55–Reference Wadolowska, Hamulka and Kowalkowska58,Reference dos Santos, Picoito and Loureiro60) . Regarding the degree of adjustment for confounding factors, seven studies adjusted for the three typical confounders (age, sex and socio-economic status or some proxy for this variable)(Reference Boone, Gordon-Larsen and Adair43,Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55,Reference Dantas, dos Santos and Lopes57,Reference Wadolowska, Hamulka and Kowalkowska58,Reference dos Santos, Picoito and Loureiro60) . Three studies did not adjust for age(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference van der Sluis, Lien and Twisk50) , and one did not adjust for socio-economic status(Reference Nuutinen, Lehto and Ray56). The twelve studies that adopted adjusted analyses (multivariate ANOVA and regression analysis) were used here for synthesis of results of the total sample(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference van der Sluis, Lien and Twisk50,Reference Perez-Rodrigo, Gil and Gonzalez-Gross55–Reference dos Santos, Picoito and Loureiro60) . Five of these studies stratified data by sex(Reference Moreira, da Veiga and Santaliestra-Pasías44,Reference Boone-Heinonen, Gordon-Larsen and Adair47,Reference Nuutinen, Lehto and Ray56,Reference Dantas, dos Santos and Lopes57) .
Table 4 provides a synthesis of the results. Prospective analysis revealed positive associations between all predominantly unhealthy and mixed lifestyle patterns and overweight/obesity. In the cross-sectional studies, most lifestyle patterns were not associated with overweight/obesity. An inverse association between completely healthy lifestyle patterns and overweight/obesity was observed only once out of three times. For completely unhealthy lifestyle patterns, almost all associations (eight out of nine) were null. Two predominantly healthy lifestyle patterns were positively associated with overweight/obesity and nine were not associated. The association between predominantly unhealthy lifestyle patterns and overweight/obesity was positive in seven out of sixteen times. Only one mixed lifestyle pattern (out of twelve) was associated with overweight/obesity (Table 4).
Table 4. Direction of associations between lifestyle patterns and overweight and obesity in adolescents
* Included 1 lifestyle pattern with risk behaviours.
† Included 2 lifestyle patterns with risk behaviours.
When stratifying by sex, we observed a positive association between predominantly healthy lifestyle patterns and overweight/obesity in girls two out of five times. The association between predominantly unhealthy lifestyle patterns and overweight/obesity was tested eight times for girls and five times for boys, with three and two positive associations, respectively. No association was found between the other lifestyle pattern classifications and overweight/obesity in boys or girls (Table 5).
Table 5. Direction of associations between lifestyle patterns and overweight and obesity in adolescents by sex
Table S9 presents the synthesis of the results according to the risk of bias. For cross-sectional studies with moderate risk of bias, a positive association between predominantly unhealthy lifestyle patterns and overweight/obesity was found in three out of four times. A positive association with overweight/obesity was also found for the single mixed lifestyle pattern. The positive associations found in the prospective analysis included in this systematic review come from studies with moderate risk of bias (Table S9, online Supplementary Material).
Discussion
This review sought to identify the association between lifestyle patterns identified by data-driven exploratory analysis and overweight/obesity in adolescents. It was possible to note a co-occurrence of healthy and unhealthy behaviours in lifestyle patterns. Predominantly unhealthy lifestyle patterns were more frequently observed among adolescents. Overall, synthesis of the results of cross-sectional studies indicated that there was no association between lifestyle patterns and overweight/obesity, even when stratified by sex. However, analysis stratified by risk of bias showed a positive association between predominantly unhealthy and mixed lifestyle patterns and overweight/obesity in studies with moderate risk. Prospective analysis of a single study with moderate risk of bias suggested an increase in BMI over time with predominantly unhealthy and mixed lifestyle patterns.
We identified a variety of lifestyle patterns and weight status indicators in studies, hindering their comparison. The methods used to measure diet, physical activity, sedentary behaviour and sleep were heterogeneous across studies, as were the methods used to derive lifestyle patterns. Most studies used FFQ to collect food consumption data, while one study used 1-d 24-h recall combined with 3-d food record, and two studies used 2-d non-consecutive 24-h recall. FFQ refers to the respondent’s usual intake of food over a specific period, and the estimation tasks are complex. Regarding to recall/record methods, at least 2 d are recommended to estimate usual intake of the population and their relationship with other indicators(Reference Hinnig, Monteiro and De Assis20). For synthesis of results, it was necessary to categorise lifestyle patterns according to their behaviours. Patterns were classified into completely or predominantly healthy or unhealthy and mixed (i.e. balance between healthy and unhealthy behaviours).
We emphasise that previous reviews were conducted with children and adolescents (5–18 years)(Reference Leech, McNaughton and Timperio6) and only children (5–12 years)(Reference D’Souza, Kuswara and Zheng10). The first comprised a narrative review of studies that evaluated the clustering of diet, physical activity and sedentary behaviour(Reference Leech, McNaughton and Timperio6). The other systematically reviewed evidence on the clustering of diet, physical activity, sedentary behaviour and sleep(Reference D’Souza, Kuswara and Zheng10). Our systematic review is the first to analyse the association between lifestyle patterns and overweight/obesity in adolescents, which is a critical phase for weight gain. The above-mentioned reviews did not restrict the eligibility criteria regarding the behaviours included in the lifestyle patterns, so some studies reviewed contained only physical activity and sedentary behaviours in the patterns. In contrast to this, we considered the domain of diet as a fundamental behaviour in lifestyle patterns, as diet plays a major role in energy balance regulation(Reference Narciso, Silva and Rodrigues1). Consumption of ultra-processed foods has been identified as a risk factor for increasing obesity(Reference Costa, Del-Ponte and Assunção65). Additionally, dietary patterns with a lower percentage of obesogenic foods appears to be effective in reducing the risk of developing obesity(Reference Liberali, Kupek and Assis2).
As reported in previous reviews(Reference Leech, McNaughton and Timperio6,Reference D’Souza, Kuswara and Zheng10) , we identified the co-occurrence of both healthy and unhealthy behaviours in lifestyle patterns which indicates that protective and risk behaviours for overweight/obesity coexist in lifestyle patterns. This implies that we may not assume that healthy levels of a particular behaviour are indicative of an overall healthy lifestyle. The review studies(Reference Leech, McNaughton and Timperio6,Reference D’Souza, Kuswara and Zheng10) found a large number of clusters with high levels of sedentary behaviour, similar to our results. In the present review, almost all studies have assessed sedentary behaviour through screen time only. In addition, most lifestyle patterns found were composed of low levels of physical activity and high levels of sedentary behaviour(Reference Moreira, da Veiga and Santaliestra-Pasías44–Reference Laxer, Cooke and Dubin46,Reference Ottevaere, Huybrechts and Benser52,Reference Fernandez-Alvira, De Bourdeaudhuij and Singh54,Reference Nuutinen, Lehto and Ray56,Reference Dantas, dos Santos and Lopes57) . This suggests that sedentary behaviour contributes strongly to adolescent lifestyle, which is somewhat worrying, given the positive association between sedentary behaviour and unfavourable health indicators(Reference Carson, Hunter and Kuzik66). Moreover, we identified lifestyle patterns composed of high levels of both physical activity and sedentary behaviour. These findings demonstrate that one behaviour is not necessarily a barrier to the other and that adolescents find time for physical and sedentary activities throughout the day(Reference Seghers and Rutten51,Reference Ottevaere, Huybrechts and Benser52,Reference Dantas, dos Santos and Lopes57) .
Previous review from D’Souza etal. (Reference D’Souza, Kuswara and Zheng10), who analysed studies on children only, concluded that unhealthy lifestyle patterns were more often associated with adiposity risk than healthy and mixed patterns. In contrast, Leech etal.(Reference Leech, McNaughton and Timperio6) reported that the relationship of cluster patterns with excess weight was inconsistent. This was partially observed in the main findings analysed in the current review. Although we found some evidence of a positive association between lifestyle patterns and overweight/obesity in studies with moderate risk of bias, we considered these findings questionable due to the limited methodological quality of the studies. The associations found in cross-sectional studies are not, by themselves, evidence of causality. Prospective findings from a single study are insufficient. Additionally, all measured variables from these studies were based on self-reports, including weight and height variables, which compromises the quality of the findings(Reference Laxer, Brownson and Dubin45,Reference Laxer, Brownson and Dubin45,Reference Nuutinen, Lehto and Ray56) . There is a tendency to overestimate height and underestimate weight, leading to incorrect reports and inaccurate estimates of overweight/obesity rates(Reference Kuskowska-Wolk, Karlsson and Stolt67). Adolescents with overweight/obesity tend to underestimate their weight more often than normal-weight adolescents, and girls tend to underestimate their weight more often than boys(Reference Carson, Hunter and Kuzik66). Furthermore, children and adolescents who are overweight or obese were more likely to under-report energy intake when compared with their non-obese peers(Reference Walker, Ardouin and Burrows68).
Unexpectedly, we found that some predominantly healthy lifestyle patterns were positively associated with overweight/obesity, particularly in girls. The following hypotheses may explain these controversial results. It is not possible to infer whether healthy behaviours reflected the development of overweight and obesity or whether the presence of the latter led to the adoption of healthier habits as a strategy for weight loss. Studies have shown that adolescents with overweight/obesity often try to lose weight(Reference Brown, Skelton and Perrin69), and that these behaviours are more likely in girls than in boys(Reference Chung, Backholer and Wong70). Furthermore, although predominantly healthy, these lifestyle patterns comprised inadequate behaviours in the diet domain, such as low consumption of healthy foods and high consumption of unhealthy foods, which could explain, at least in part, this result(Reference Moreira, da Veiga and Santaliestra-Pasías44).
As in the recent review by D’Souza etal. (Reference D’Souza, Kuswara and Zheng10), we considered four domains of energy balance-related behaviours associated with overweight/obesity. Of the studies used for synthesis of results, those conducted by Perez-Rodrigo etal.(Reference Perez-Rodrigo, Gil and Gonzalez-Gross55), Nuutinen etal.(Reference Nuutinen, Lehto and Ray56) and Sevil-Serrano etal.(Reference Sevil-Serrano, Aibar-Solana and Abos59) addressed these four behaviours. The other studies included diet, physical activity and sedentary behaviour domains only, which represents a limitation, as sleep habits are related to overweight/obesity(Reference Fatima, Doi and Mamun5,Reference Fatima, Doi and Mamun71) . In the study by Nuutinen etal.(Reference Nuutinen, Lehto and Ray56) the indicators of sleep habits were related to duration, discrepancy and quality. The authors found a greater risk for overweight/obesity among girls in the high screen time, unhealthy lifestyle pattern, whose scores for sleep duration were moderate but for discrepancy and quality were low. This finding suggests that investigating only the effect of sleep duration on lifestyle patterns may not be enough. Future studies should consider this behaviour both in terms of duration and quality to understand the cumulative effects of sleep on weight status in adolescents.
Most studies evaluating the association between lifestyle patterns and overweight/obesity were carried out in European countries, that is, countries with high socio-economic levels. It is important to highlight that, whereas the prevalence of overweight/obesity among young people has stabilised in high-income countries, the prevalence is still on the rise in medium- to low-income countries(Reference Abarca-Gómez, Abdeen and Hamid72). Thus, more studies need to be carried out in medium- and low-income countries for cultural, economic, and demographic variability and representativeness of data.
About 85·6 % of the reviewed studies applied the cluster analysis method to identify the lifestyle patterns. Although there is evidence to indicate that latent class analysis substantially outperforms the cluster analysis, our results demonstrate that the latter technique has been more applied by studies. Both methods are centred on individuals; however, the cluster analysis use the distance in order to separate observations into different groups while latent class analysis is a model-based approach. An advantage of using a model-based approach is less arbitrary and more rigorous statistical techniques. More precisely, in latent class analysis is assumed that a mixture of underlying probability distributions generates the data. Furthermore, there are more formal criteria to make decisions about the best model fit or number of classes(Reference Leech, McNaughton and Timperio6,Reference Magidson and Vermunt73) .
The lack of evidence from longitudinal data precludes determination of whether, over time, lifestyle patterns contribute negatively to the weight status of adolescents. Only two studies with this population were found(Reference Laxer, Cooke and Dubin46,Reference Landsberg, Plachta-Danielzik and Lange48) , and only one was considered for the synthesis of results, which had a moderate risk of bias(Reference Laxer, Cooke and Dubin46). However, overall, the findings of the referred studies demonstrate that, in the long term, lifestyle patterns may have some effect on the weight status of adolescents. More longitudinal studies with well-designed methodology in other countries are needed to measure these effects over time. It could be also interesting for future prospective studies to examine the stability of the lifestyle patterns over time.
In the present review, five studies found that risk behaviours, such as substance use, tend to co-occur with energy balance-related behaviours(Reference Laxer, Brownson and Dubin45–Reference Landsberg, Plachta-Danielzik and Lange48,Reference dos Santos, Picoito and Loureiro60,Reference Berlin, Kamody and Thurston61) . Risk behaviours are common among adolescents and tend to increase with age(Reference Laxer, Brownson and Dubin45). Marijuana use, smoking and binge drinking are generally not the focus of studies evaluating the determinants of overweight/obesity. Excessive alcohol consumption can contribute to weight gain by increasing energy intake or stimulating consumption of other unhealthy foods(Reference Battista and Leatherdale74). However, little is known about the relationship between overweight/obesity and other types of substances. Laxer etal.(Reference Laxer, Brownson and Dubin45,Reference Laxer, Cooke and Dubin46) found positive cross-sectional and prospective associations in the lifestyle patterns of moderately active adolescents with low healthy food consumption, high unhealthy food consumption and high substance use.
This systematic review was rigorous. We designed a search strategy including a range of terms relevant to the topic that was applied to a variety of databases. An appropriate tool was used to assess the risk of bias, and criteria based on the tool items were determined to facilitate the analysis(19). However, the review becomes limited by the quality of evidence from the included studies. Most studies were cross-sectional and used self-reported questionnaires to assess the behaviour of adolescents, increasing susceptibility to memory and social desirability bias. Not all studies adopted valid or reproducible methods for these measures, which may have implications for the accuracy and reliability of the findings. Nevertheless, the results may provide important insights into how obesogenic behaviours cluster together, which can help in the design and improvement of public health policies aimed at combating obesity.
This review included only studies that used data-driven methods as cluster analysis and latent class analysis to determine lifestyle patterns. The strength of our study is that the approach allowed separating individuals into mutual groups that share similar characteristics. However, some statistical techniques require that arbitrary decisions and subjective interpretations be made by the researcher. Importantly, the methods used to determine lifestyle patterns in each study were data-driven, and the lifestyle patterns found may only be specific to the populations and cultures studied, which limit the generalisability of the findings. Additionally, ‘high’ or ‘low’ levels of a particular behaviour may refer to the highest or lowest scores in a specific pattern or to the highest or lowest probabilities to belong to a pattern and may not even meet the guidelines for the behaviour. ‘High’ or ‘low’ behaviour in one study population may not be classified as such in another population. This may have implications for understanding the influence of different lifestyle patterns in relation to overweight/obesity.
Data-driven methods can produce different lifestyle patterns even when applied to the same dataset(Reference D’Souza, Downing and Abbott75). Thus, comparison of results across studies employing different methods should be done with caution. Researchers should consider the choice of method based on their study objectives and subsequent analyses. It is also essential that authors justify the decisions made and the final model chosen. More studies including the four behavioural domains (diet, physical activity, sedentary behaviour and sleep) are needed. Studies should consider the use of objective methods, such as accelerometers or pedometers, to capture movement behaviours in order to obtain more accurate results. There is no consensus on which dietary assessment method is more accurate for adolescents(Reference Pérez-Rodrigo, Artiach Escauriaza and Artiach Escauriaza76). However, it is important to improve dietary assessment methods by including validated and reproducible measures to better capture dietary information. The choice of instrument should consider the objectives and the logistic of the study, which will influence the suitability and feasibility of different approaches
Finally, it is worth mentioning that we synthesised the directions of the association between lifestyle patterns and overweight/obesity, not the strengths of associations. It was not possible to carry out a meta-analysis because it was not clear whether the data were sufficiently comparable for quantitative analysis. In this case, vote counting based on the direction of association was the alternative adopted as an ‘acceptable’ method for the presentation and synthesis of the results. Vote counting can be used to synthesise results when there is inconsistency in the effect measures or data reported across studies(Reference McKenzie, Brennan, Higgins, Thomas and Chandler77). This method has some limitations. Vote counting does not consider the magnitude of effects and the differences in the relative sizes of the studies and could difficult the assessment of the certainty of the evidence(Reference Campbell, McKenzie and Sowden21,Reference McKenzie, Brennan, Higgins, Thomas and Chandler77) . However, this method may be more advantageous compared with a narrative review that only describes results study by study, which comes with the risk that some studies are privileged above others without appropriate justification(Reference McKenzie, Brennan, Higgins, Thomas and Chandler77).
Conclusion
Adolescents tend to simultaneously exhibit healthy and unhealthy lifestyle behaviours with predominantly unhealthy lifestyle patterns more frequently observed. The presence of unhealthy behaviours together with healthy behaviours suggests the need to be alert, even with those adolescents who appears to be doing well in a domain. The large number of lifestyle patterns with high levels of sedentary behaviour suggests that this behaviour has become increasingly important in adolescents’ lives. Overall, cross-sectional studies indicate that there is no association between lifestyle patterns and overweight and obesity in adolescents, even after sex stratification. However, when analysing the results stratified by risk of bias, a positive association between predominantly unhealthy and mixed lifestyle patterns with overweight/obesity was identified in cross-sectional studies with moderate risk of bias. The only prospective analysis of the topic found an increase in BMI over time associated with predominantly unhealthy and mixed lifestyle patterns. Because of the heterogeneity and quality of the studies, we consider the current evidence weak and inconsistent. Further research is needed, preferably longitudinal studies using objective methods or validated and reproducible tools to measure lifestyle behaviours in the adolescent population. Despite these limitations, the findings from this systematic review have considerable implications for public health policy and school-based health promotion initiatives, with an emphasis on integrated approaches. We highlight the importance of targeting multiple behaviours simultaneously to achieve more health benefits than when these behaviours are targeted separately.
Acknowledgements
None
This work was supported by Brazilian Coordination for the Improvement of Higher Education Personnel (CAPES) (L.J.P., grant number: 88882.438755/2019-01; and L.H.M., grant number: 88887.498373/2020-00).
L. J. P., F. G. K. V. and P. D. F. H. designed the study; L. J. P. and L. H. M. conducted the searches, studies selection, data collection process and quality assessment; L. J. P., L. H. M., F. G. K. V. and P. D. F. H. analysed the data and interpreted the results; L. J. P., L. H. M., F. G. K. V., P. D. F. H., P. F. P. and M. A. A. D. A. wrote the manuscript and made substantial contributions to the final version.
The authors declare no conflicts of interest.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114522000228
