This cross-sectional study was conducted in Hamadan from early January 2018 to late March 2018. Hamadan is a city in the west of Iran with a population of 573,449 based on the Health Center statistics from 2018.

The sample included 693 elementary-school students (7-12 years old) and their parents. Multistage stratified cluster sampling was conducted. Hamadan has 2 educational areas (strata), according to the divisions of the Hamadan Department of Education. In the first phase of sampling, a list was provided by the Department of Education and then 13 schools (clusters) were selected randomly from the above areas based on socioeconomic conditions. In the second phase, all 6 grades of each selected elementary-school (1 class from each grade) were considered, and finally students were selected randomly from each class.

The questionnaires developed by Dianat et al. [7] and Hatami et al. [19] were used to gather information about demographic variables, leisure time (including age, sex, PA, cell phone use, computer use, computer games, watching TV), family history of LBP, school-related factors (information about convenience and comfort of the school furniture based on a modification of the Chair Feature Checklist), classroom design and amount of homework, and some information about the type of school bag used, how students carried their school bags, and the length of time spent carrying the school bag. These questionnaires have been shown to be adequately reliable and valid. Minor revisions were made according to experts’ opinions to adapt the items of the questionnaires for the population of the present study. The questionnaires were tested in a pilot study of 60 students. According to their comments, partial revisions were made to a number of items in the questionnaire to improve transparency and understandability. Additionally, the test-retest approach was used to assess the reliability of the questionnaire items using the kappa coefficient (ranging from 0.73 to 0.95) and the intraclass correlation coefficient (ranging from 0.85 to 0.98).

The Rapid Upper Limb Assessment (RULA) checklist, which is an observational method for measuring the risk of upper limb musculoskeletal disorders, was used to observe students’ postures. On this checklist, the observations were recorded as numerical scores. These scores were converted into final scores using the RULA scoring matrix, according to which the priority level of corrective steps was determined (level 1: a score of 1 or 2, indicating that a posture is acceptable if not maintained for a long period of time; level 2: a score of 3 or 4, indicating that further investigation is needed and changes may be required; level 3, a score of 5 or 6, indicating that an investigation is needed and changes are required soon; and level 4, a score of 7 or higher, indicating that further investigation and immediate changes are required) [20]. The questionnaire’s reliability and validity were confirmed by Dianat & Salimi [21]. Each student’s posture was recorded by an observer using the RULA checklist to control for inter-observer variability. Inter-rater reliability was assessed at 2 different times, using a pilot study among 60 students in the current study. The Cronbach alpha coefficients for the upper arm, lower arm, wrist, neck, trunk, and leg were 0.79, 0.82, 0.78, 0.83, 0.84, and 0.86, respectively, indicating an acceptable level of reliability.

The standard Strengths and Difficulties Questionnaire (SDQ) was used to measure psychosocial variables. This questionnaire uses a 5-factor structure to measure behavioral-emotional problems and prosocial behaviors in 4-17-year-old children. It also involves 4 subscales related to difficulties (emotional problems, conduct problems, hyperactivity, and peer problems) along with a strengths measure (prosocial behaviors). This 25-item questionnaire was filled out by students’ parents. Each question was scored on a 3-point Likert scale as 0 (not true), 1 (somewhat true), and 2 (certainly true). Each subscale had 5 questions and the theoretical range of the strengths and difficulties scores was 0-10 and 0-40, respectively. The total difficulty score was obtained from the 4 subscales related to difficulties. According to the obtained scores, each subscale is divided into 3 categories (normal, borderline, and abnormal) [22]. The reliability and validity of the questionnaire have been reported to be acceptable [23].

A pre-shaded manikin picture to show areas of pain was used to measure the prevalence of LBP, with an accompanying question reading “Have you experienced pain or discomfort in your back area for a day or more during the last month? (responses: yes/no)” [7,24]. Pain intensity was measured using a visual analogue scale graded from 0 (no pain) to 10 (worst pain) [25]. All the questionnaires were filled out during interviews conducted by the principal researcher and the students’ posture was recorded by an ergonomic specialist.

Finally, the weights of students and their school bags were measured using a digital electronic scale (Beurer, Ulm, Germany) with a sensitivity of 100 g. Each participant’s standing height was measured using a portable stature meter (Yongkang Putai Hardware Factory, Yongkang, China). Each participant’s body mass index (BMI, kg/m²) was calculated and classified according to the World Health Organization (WHO) growth reference charts (2007) into 3 categories: healthy weight (healthy weight (5th-85th percentile), underweight (< 5th percentile), and overweight (≥ 85th percentile) [26].

LBP was considered as the outcome of interest (a binary response variable). The independent/exposure variables that were included in the analytical models were the students’ demographic and PA/leisure activity characteristics (age, sex, BMI, hours per week playing sports, hours per day using a mobile/tablet, hours per day using a computer, hours per day playing games, hours per day watching TV); classroom furniture/layout design (seat height, seat backrest height, seat backrest inclination, seat backrest curvature, seat depth, seat width, desk height, seat pan inclination, seatto-blackboard distance, classroom teacher placement, viewing the blackboard); amount of homework, position doing homework at home, history of accident or injury related to LBP, family history of LBP, school bag–related variables (type and weight of school bag, duration and method of carrying the school bag, method of travel to/from school); psychosocial factors (prosocial behavior, emotional symptoms, conduct problems, hyperactivity, peer problems, total difficulties), and RULA score.

Penalized logistic regression was utilized to select important correlates of LBP. We used the group smoothly-clipped absolute deviation (SCAD) penalty in the logistic regression model to select correlates and to measure the associations between LBP and demographic characteristics and PA/leisure activity, school-related factors, psychosocial factors, and RULA scores. Briefly, this model is a shrinkage regression model that imposes an L1 penalty on the regression coefficients. The chi-square test also was used to analyze the data in this study.

Stepwise logistic regression is the usual approach to select variables associated with a binary response; nevertheless, it suffers from some disadvantages, including instability of the selected variables and vulnerability to overfitting. Recently, penalized regression models, in which a penalty term is attached to the likelihood function and estimation and variable selection is done simultaneously, have shown promising results. Among the various penalties, the use of SCAD has been well established as a way to obtain reliable results [27].

The SCAD penalty is defined as follows:

In the above equation λ is the tuning parameter and its optimum value should be obtained through cross-validation. In this study, we used a 10-fold cross-validation strategy. The value of λ with the smallest Bayesian information criterion was chosen as the optimum value. This method was repeated 1,000 times and the estimated coefficients were averaged over all repetitions. To estimate the standard errors of the coefficients, a bootstrap strategy was used with 1,000 replications. Therefore, 1,000 samples (with replacement) were selected from the original data and then the standard errors of the coefficients were computed to calculate 2-sided p-values. A significance level of 0.05 was adopted for all statistical analyses.

The adequacy of the final model was checked using the Hosmer & Lemeshow test, in which a p-value greater than 0.05 indicates no evidence of poor fit [29].

The area under the receiver operating characteristic curve (AUC) was also used to check the predictive accuracy of the model. To do this, the data were divided randomly into 2 subsets (training and testing sets). The model was fitted to the training data and the AUC was obtained for the testing set. AUC values lie between 0.5 and 1, where 0.5 denotes a bad classifier and 1 denotes an excellent classifier.

It should be noted that penalized logistic regression simultaneously considers all independent variables, including potential confounding variables, when selecting variables. Therefore, this model can control for the effect of potential confounding variables. In the current study, demographic characteristics, physical leisure activities, psychosocial factors, and family history of LBP among school students, which could serve as potential confounding variables, were entered into the model as independent variables.

This study was approved by the Ethics Committee of Hamadan University of Medical Sciences (approval code: IR.UMSHA.REC. 1396.641). Informed consent was obtained from the students and their parents.

In total, 26.6% of the students reported LBP during the last month. The intensity of LBP differed significantly (p=0.001) between males (mean, 1.08; standard deviation [SD], 2.22) and females (mean, 1.79; SD, 3.06). However, there were no significant differences (p=0.106) by sex with regard to absence from school due to LBP.

The mean±SD age of the students’ was 9.70±1.61 years, and their BMI was 17.68±3.57 kg/m² (Table 1).

Using the chi-square test (Table 2), the prevalence of LBP was found to be significantly higher in older students (31.8%, p=0.001) and among female students (29.9%, p<0.05).

According to the penalized logistic regression analysis (Table 3), the following factors had an significant relationship with LBP: older age (odds ratio [OR], 3.08; 95% confidence interval [CI], 1.80 to 5.26; p<0.001), watching TV for more than 3 hours a day (OR, 2.62; 95% CI, 1.46 to 4.68; p<0.001), very short seat backrests (OR, 3.08; 95% CI, 1.61 to 5.90; p<0.001), excessively curved seat backrests (OR, 4.36; 95% CI, 2.08 to 9.13; p<0.001), very short desks (OR, 3.44; 95% CI, 1.61 to 7.35; p<0.001), a family history of LBP (OR, 2.49; 95% CI, 1.58 to 3.91; p<0.001), carrying a school bag on 1 shoulder (OR, 1.91; 95% CI, 1.03 to 3.54; p<0.05), and RULA scores of 3 (OR, 2.26; 95% CI, 1.13 to 4.50; p<0.05) or 4 (OR, 2.85; 95% CI, 1.37 to 5.91; p<0.01).

Model adequacy was confirmed by the Hosmer & Lemeshow test (χ²= 5.165; degree of freedom [df]= 8; p=0.740). Moreover, the predictive accuracy of the model using the selected variables shown in Table 3 was acceptable (AUC, 0.739) [29].