Low levels of health-related physical fitness (HRF) in youth can influence mortality and morbidities in adulthood for several disorders, including cardiovascular disease, obesity, and metabolic syndrome [1
]. It has been reported that fitness is a better predictor of health outcomes in adults than physical activity (PA) levels [6
]. In children, data from cross-sectional and prospective studies have been used to suggest that increasing PA is insufficient since future cardiovascular risk is more dependent on fitness than on the amount of PA performed [7
]. In particular, inadequate levels of aerobic fitness in children and adolescents can influence overweightness, metabolic disorders, and cognitive diseases that predict morbidities in adulthood [9
]. Thus, it is important to establish healthy habits early in life to attain a desirable level of fitness during childhood and adolescence [13
] by acting on the factors that could influence this behaviour [14
]. As noted in other reports, adolescents’ low fitness levels could be due to several factors involving genetic, biological, familial, environmental, and behavioural aspects [13
]. At present, female sex, low income, low consumption of dairy products and bread/cereals, increased consumption of sweetened beverages, insufficient PA level, excessive screen time, and excess body fat have been found to be associated with low aerobic fitness levels in youth [17
]. Additionally, the contemporary construct of paediatric dynapenia, an acquired condition characterized by low levels of muscular strength and power, has been used to identify youth with consequent functional limitations not caused by neurologic or muscular disease [21
The latent class analysis (LCA) is a promising statistical tool that has been frequently used in recent years to identify small homogenous groups based on behavioural patterns related to health, obesity, diet, PA, sedentariness, substance use, and smoking [22
]. Moreover, LCA allows the identification of latent classes that can be used to target proper interventions [28
]. Of interest, previous investigations have used cluster analysis to identify potential patterns of health-related behaviours in adolescents [31
], and other colleagues have already used LCA to determine patterns of behaviours related to obesity or to PA and sedentariness on different population targets [26
]. However, to our knowledge, no studies have identified distinctive classes of patterns with respect to low fitness risk factors. The aims of this paper were to describe profiles of adolescents’ fitness level, identify latent classes of fitness-related risk behaviours, describe sociodemographic and environmental predictors of class membership, and evaluate the association of the identified patterns with overall fitness level and single health-related fitness components.
2. Materials and Methods
2.1. Study Design and Sample
The ASSO (Adolescents Surveillance System for the Obesity prevention) Project funded by the Italian Ministry of Health was aimed at developing a surveillance system in schools to collect data on adolescents’ lifestyles and food consumption, and anthropometric and fitness measurements [42
A total of 883 students were recruited, through a multistage sampling, during the years 2012 and 2013 from the classes 1 to 4 of seven public and private high schools within Palermo city (Italy). All participants were provided with information sheets and had to supply the informed consent signed by their parents before starting the study.
A standardized methodology with standard operating procedures (SOPs) has been developed for the data collection, and training sessions were organized for all the ASSO team members and teachers of the participating schools.
The principles of the Italian data protection (196/2003) were guaranteed and the Ethical Committee of the Azienda Ospedaliera Universitaria Policlinico “Paolo Giaccone” in Palermo approved the project and the study protocol (approval code n.9/2011).
Fitness measures were collected through the ASSO-Fitness Tests Battery (ASSO-FTB) [45
], composed by five accurately selected tests for the assessment of five physical fitness components: (1) the hand-grip strength test (HG) to assess upper body maximal strength; (2) the standing broad jump test (SBJ) to assess lower body strength and power; (3) the sit-up test to exhaustion (SUe) to assess local muscular endurance; (4) the 4 × 10 m shuttle run test (4 × 10 m SR) to assess speed and agility; and (5) the 20 m shuttle run test (20m SR) to assess endurance/aerobic capacity. Classes of fitness levels were derived, and the detailed description of the applied methodology can be found in Bianco et al. [45
]. Fitness measures were complete for all five components for a total of 544 students aged 13–19, with mean age 16.3 ± 1.4 years; M = 369 (67.8%); F = 175 (32.2%). Anthropometric measurements of weight, height, and waist circumference were collected by the teachers through the use of a calibrated scale, a stadiometer, and a nonelastic meter, respectively, all available within the schools. Personal information and lifestyle aspects were collected through the web-based questionnaires of the ASSO-NutFit software administered within the schools: ASSO-PIQ (Personal Information Questionnaire), ASSO-PASAQ (Physical Activity, Smoke, and Alcohol Questionnaire) and ASSO-FHQ (Food Habits Questionnaire). The ASSO-PIQ included questions regarding participant and family information and neonatal and clinical assessment. The ASSO-PASAQ consisted of three sections: physical activity, smoking, and alcoholic drinks and other beverages. Finally, the ASSO-FHQ consisted of six items, regarding: breakfast, school break, lunch, afternoon break, dinner, and various habits such as eating out, eating ready meals, organic food, fresh food, or food from vending machines [42
Fitness levels and the five fitness components considered (upper body maximal strength, lower body maximal strength, muscular endurance, speed and agility, and endurance/aerobic capacity) were categorized into three classes (0 = high, 1 = medium, and 2 = low), and subsequently for the purpose of the logistic regression analysis into two classes (0 = high/medium and 1 = low).
Variables eventually associated with fitness level and single components addressed in the survey included initially 87 items. Since including too many variables in the model for LCA could negatively affect the analysis, a total of 18 variables (included in Table 1
) out of the 87 items contained in the ASSO questionnaires were selected and gathered into the following three dimensions: (1) biological and genetic; (2) sociocultural and environmental; (3) life habits. The inclusion of items in the different dimensions was based on previous literature suggesting risks within the context of a larger conceptual framework for health characteristics and for sedentary behaviours [50
] and subsequently was adapted to fitness.
Moreover, among these factors, a total of eleven latent class indicators were chosen to represent multiple dimensions of fitness risks, i.e., biological (health risk/status) and lifestyle (physical activity/sedentariness, alcohol/smoking, and meal patterns and habits) dimensions; the other seven variables were investigated in a multivariate analysis for their role as possible predictors of class membership (gender, age, having at least one parent overweight/obese, parents’ education, family affluence scale (FAS), town of residence’s size, and school type) [50
Binary latent class indicators were created for convenience of use, with a recoding scheme that dichotomized fitness-related behaviours, with ‘0’ representing a healthier behaviour and ‘1’ representing a less healthy behaviour.
Among the biological and genetic determinants, weight status, health risk, malaise frequency and diagnosed diseases were selected for the latent classes’ assessment, while sex, age, and having at least one parent overweight/obese were selected as predictors.
2.3. LCA and Logistic Regression Analyses
An LCA was performed to identify latent classes of fitness-related risk behaviours. The number of classes that best fit the data was chosen by evaluating an increasing number of classes, through the log likelihood, Akaike information criterion (AIC), Bayesian information criterion (BIC) [23
], sample-size-adjusted Bayesian information criterion (adjusted BIC), and a consistent version of AIC (CAIC) [54
]. Latent classes were then defined by the probabilities that individuals in each class met the criteria for the considered eleven variables.
The chi-squared test was applied to assess differences between the classes, and multiple logistic regression models were used to examine the relationship between latent classes and predictors and between latent classes and overall fitness level and single fitness components. All regressions were controlled for potential confounders, and adjusted odds ratios (Adj ORs) were obtained.
Alpha level was set at 0.05, and 95% CIs were calculated for the ORs derived.
In this study, Stata/MP 12.0 software was used, and LCA was performed using the LCA Stata Plugin (PennState).
The ethical approval was given by the ethical committee of the “Azienda Ospedaliera Universitaria Policlinico Paolo Giaccone” (approval code n.9/2011). All the participant students provided an informed consent form signed by their parents.
The present data describe underlying patterns of HRF factors and their sociodemographic and environmental predictors, and report on the association of latent classes with fitness level and single fitness components.
Cluster analysis has been previously adopted to identify potential patterns of health-related behaviours in adolescents [31
], and LCA has also been frequently used to determine patterns of behaviours on different population targets [26
]. To our knowledge, there are no studies investigating fitness patterns in adolescents. Therefore, although comparison of our results with findings from other studies is limited, this study adds important information on possible predictors of low fitness levels in adolescents.
Five latent classes of fitness-related lifestyle patterns and health risk/status have been identified in our model. A first class was composed mainly of virtuous subjects, and this is consistent with one study on patterns of PA, sedentariness, and diet that identified a “healthful” class characterized by participants meeting recommendations for PA, consuming healthy foods, and showing a high overall health status and life satisfaction or low depression [22
]. Moreover, other studies found a positive association of high PA levels, low sedentariness, and healthful diet with other physical and psychological factors in children, adolescents, and adults [30
]. With regard to latent class 2, the major risk for unhealthy behaviours in this class was in the dimension of PA/sedentariness. The suboptimal food intake and overweight/obese status could be presumably related to their low PA and sport practice. Of note, subjects from this class were mostly males, older (more than 16 years), had low educated parents, and attended professional/technical vocational schools. This is consistent with findings from other studies that found that younger participants were generally more active than older youth [56
] or that adolescents with low socioeconomic status (SES) were less active [58
]. These observations suggest that interventions to enhance PA and sport involvement, which in turn may improve food intake and weight status, could be addressed to help fighting the global pandemic of physical inactivity among children and adolescents [60
Moreover, it is likely that the lack of daily PA in this class could contribute to lower fitness levels and lower fitness abilities (excepted for upper body maximal strength) assessed through the ASSO-FTB reported in this class. This is also consistent with latent class 5, where almost all adolescents comply with the PA recommendations and a small number had low fitness levels. These results are in line with several studies. Morrow et al. in 2013 showed that adolescents failing to meet national aerobic and muscle-strengthening PA guidelines have higher odds of not achieving healthy physical fıtness levels of aerobic capacity [2
]. Other longitudinal studies showed that PA interventions and PA combined with nutrition interventions were effective at increasing fitness levels in school-age youth [61
]. Of interest, Silva et al. recently reported an association between hand-grip strength, body mass, body height, and PA levels in youth. The authors found that performance on these measures was positively related with greater body mass (probably muscle mass) and greater height (probably reflecting greater leverage) and only partially related with PA levels [64
]. Collectively, these findings can indirectly explain the lack of association found in our study between latent class 2 and low upper body maximal strength (Table 4
In the third latent class, unhealthy lifestyles characterize a high proportion of adolescents, who are mostly males and older compared to latent class 1, and this could explain the higher frequency of alcohol consumption that is generally more common in males and older adolescents [65
] and poor food habits that are more common in males [66
]. Despite these unhealthy behaviours, there is a high proportion of adolescents belonging to this class that have a high fitness level. This pattern was also found in one study [26
], where adolescents from an “unhealthy” class were found to consume huge quantities of energy-dense foods, while not taking advantage of the health benefits associated with their more active lifestyle [39
]. Moreover, since no association was found for fitness level and for all single fitness components in this class compared to the virtuous class, it is possible that these behaviours are not mere determinants of fitness levels. Currently, there is not a consensus on the relationship between fitness/PA and alcohol consumption in adolescents [65
]. One study showed that adolescents with low upper-body musculoskeletal strength had a lower risk of alcohol consumption [71
Thus, this alcohol-related behaviour should be further analysed before suggesting strategies and interventions aimed at decreasing the risk of low fitness. When, for example, latent class 4 is considered, where almost all subjects are at health risk, a strong association was found with fitness level; health risk was assessed through two components, one of this being the alcoholic risk; it could be suggested that in itself, drinking more than 12 g of ethanol per day could be a risk factor for low overall fitness level, while drinking alcoholic beverages in amounts <12 g of ethanol per day is not a risk factor in people more than 15 years old [72
]. Moreover, adolescents from latent class 4 are more likely to be overweight/obese. The findings of the strong positive association evidenced between this class and low fitness level are in line with other studies confirming that fatness is inversely related to fitness [18
]. A positive association in latent class 4 was also found with low muscular endurance. In accordance with Chen et al., our study confirms that overweight/obese adolescents have poorer performance in muscle endurance tests [75
Subjects from latent class 5 are characterized by a health status that can be interrelated to the high probabilities of being overweight/obese, being at health risk, and practising sport less than 3 h/week that could be found in this class, as well as drinking alcohol. For example, high rates of adolescent sedentary behaviours have been associated with patterns of physical and psychological health in one study by Ussher et al. [76
]. Different variables were found as predictors of this class, and this is in line with several studies that found children and adolescents with highly-educated parents as being more likely to display positive psychological health and fewer health complaints than youth with less educated parents, which highlights the need for programs helping people access university studies [77
]. Although adolescents from class 5 show the highest percentage of high fitness levels, they are significantly more at risk of lower body maximal strength, muscular endurance, and speed and agility compared with class 1. This is partially in line with findings from other authors showing that children and adolescents with both upper- and lower- body muscular fitness had higher ORs of reporting fair (vs. excellent) perceived health status [78
Importantly, it should be noted that the relationship between malaise and poor fitness level could be reciprocal. Generally, it could be hypothesised that the relationship is indirect, with those with a high malaise level, frequent health complaints, and diseases diagnosed by a medical doctor, and at the same time drinking alcohol and tending to be overweight, would likely have a less active lifestyle, which could lead to a low overall fitness level. Of interest, Farooq et al. recently reported that moderate-to-vigorous physical activity (MVPA) begins to decline at age 7 years [80
], so it may be too late to start interventions during adolescence. This is an important public health message as interventions need to start early in life before risk factors such as low PA/sport (class 2) or incorrect alcohol/food habits (class 3) become present. Moreover, our research findings support the modern-day concept of exercise deficit disorder, which is aimed at identifying children with low levels of MVPA before they become more resistant to exercise interventions during adolescence [14
It can be suggested that subgroups of adolescents with high malaise frequency and diseases, coming from low educated families, and attending technical/professional schools could be targets of interventions aimed at improving fitness levels and, in particular, muscular fitness. Secular trends indicate levels of muscular fitness in contemporary youth are decreasing [82
], and therefore targeted interventions are needed to address growing concerns related to paediatric dynapenia in youth [21
]. In our study, there was not a particular class characterized by a high level of sedentary behaviours, and this is in contrast with other studies comparing PA, sedentary behaviour, and diet contributing to obesity. One study analysing PA and sedentary behaviours [38
] found three distinctive classes for boys and girls of active, sedentary, and low/moderate PA behaviour; another study applied LCA to PA and sedentary behaviours, showing a model with five patterns for average intensity, sedentary behaviour, light activity, MVPA, and vigorous activity [37
]. However, since time spent watching TV, PC, and videogames are equally distributed in almost all classes of our study (from classes 2 to 5), interventions to increase fitness could convey the message of reducing screen time to all the subgroups identified.
One of the main strengths of the present study is that it used the LCA to identify latent patterns underlying possible fitness-related behaviours, which has been demonstrated to be a valid approach to clustering subjects with similar characteristics. Risk factor clustering is an important tool in terms of public health and prevention to plan targeted interventions early in life. The choice of the LCA as the method of analysis was guided by the numerous categorical variables originally collected within the project through the ASSO toolkit. Compared to other clustering methods, this model-based clustering approach allows deriving clusters using a probabilistic model that describes distribution of data through a top-down approach. The classes’ identification can effectively improve our understanding of specific joint behaviours that should be modified to improve the health of school-age youth [27
Moreover, the logistic regression analysis for the association of the latent patterns with overall fitness level and single fitness components was carried out for the purpose of checking the validity of the identified latent classes, and this is a valid approach for addressing proper strategies and interventions. Another strength of this paper is that the different fitness components were assessed through the ASSO-FTB, a validated tool that assessed various health-related fitness components [46
]; these components were subsequently used to evaluate the overall fitness level through a principal components analysis (PCA) described in detail in another previous published paper [45
]. One of the limitations of the present study is that for the assessment of PA and sedentariness, a validated web-based questionnaire was used with self-reported information, but no motor skills, muscular strength, or MVPA levels have been directly assessed. Moreover, statistical testing power of the regression analyses could have been decreased for some predictors, such as non-sedentary activities, FAS, or town size, because one of the groups of the binary variables was too small. It has to be also considered that dichotomizing variables, even if this is a common approach applied in LCA, could have reduced sensitivity and lost some key information. Another limitation is that the study sample was composed of adolescents from an area in Southern Italy, thus the sample was not representative of the entire national population. This did not allow the comparison of behaviours with adolescents from the northern and central parts, and can help with suggesting strategies that are valid at the local but not at the national level.