Study population
The study population included participants of the KNHANES IV, 2008-2009, which comprised a health interview survey, a health examination survey, and a nutrition survey. The health interview survey included items pertaining to sociodemographic status and health-related behaviors, and was used for obtaining data on morbidity, including self-reported CLBP status. The health examination survey comprised items on waist circumference and BMI as well as DXA measurements, from which we quantified regional fat mass.
Although the KHNAHES IV was initiated in 2007, the use of DXA examination was started only in the second half of 2008; accordingly, we included observations from the period after 2008. The KNHANES V (2010-2012) also conducted DXA, but the health interview survey lacked any questions on prevalent CLBP. Therefore, our study population was limited to the individuals included in the KNHANES IV, 2008–2009.
In this population, we included people with data from both the health interview survey and health examination survey. We excluded those without DXA measurements or information on CLBP status. Also excluded were people aged 18 years or younger, since the minimum age of those with CLBP data was 19 years, although the KNHANES IV targeted citizens aged 10 years or older.
DXA measurements
DXA uses X-ray of two energy levels with different absorptivity to tissue components; in this manner, it quantifies fat mass, lean mass, and bone mass [18]. The KNHANES IV gathered data on the fat, bone, or lean mass of each body part (upper limbs, lower limbs, trunk, and head), as measured by DXA (Discovery-W fan beam densitometer, Hologic, Inc., USA). We combined the measurements of the left and right into one value and did not use the measurement of the head itself, except for the total mass.
With the DXA measurements, we examined two aspects of obesity: body fat proportion against total body weight and regional fat distribution against total body fat; the former for the measurement of relative fat mass and the latter for the assessment of the manner in which the total fat mass is distributed along the whole body.
Outcomes
Of the aforementioned data, those pertaining waist circumference, body fat proportion, and regional fat distribution were used for the assessment of obesity status in three different ways. The waist circumference represents the apparent fat distribution at the trunk and was derived from the health examination survey (Aim 1). Data on body fat proportion and regional fat distribution were obtained using DXA. Body fat proportion signifies the total fat mass relative to the total weight (Aim 2), while regional fat distribution pertains to the fat mass distribution in each body part, which we defined as the regional fat mass divided by the total fat mass (head, trunk, and lower/upper limbs; Aim 3)
Exposure and covariates
Participants were assumed to have CLBP if they answered ‘yes’ to the following question in the health interview survey: “Have you experienced LBP persisting for three months or more in the most recent year?”
For Aim 1, we adjusted for sociodemographic factors (sex, age, education, occupation, household income, and region), health behaviors (smoking, drinking, and physical activity), and comorbidities (osteoporosis and depression). Region was included as a binary variable (Seoul vs. the other areas), although it originally had 16 levels in the data. Participants were considered smokers if they had a smoking habit at the time, or had smoked five or more packs of cigarettes in their life. The variable for drinking identified people who had consumed one or more glasses of alcohol per month during the most recent year. For physical activity, we used three variables. One pertained to whether a participant partook in at least 20 minutes of vigorous physical activity three or more days a week, while another was related to participation in at least 30 minutes of moderate physical activity five or more days a week. The third variable was associated with walking for at least 30 minutes five or more days a week. Comorbidities were determined by the response to a question in the health interview survey about whether participants had experienced a condition for three or more months in the most recent year. Participants with a BMI higher or equal to 25 kg/m2 were assigned to the obesity group according to the guideline for obesity diagnosis in Korea [19].
For Aims 2 and 3, we included the total lean body mass in addition to the aforementioned covariates. This allowed to obtain estimates conditional on the same total lean body mass. In other words, we tried to make estimates within people with similar body sizes except for fat mass.
Statistical analysis
We employed linear models to estimate average waist circumferences, body fat proportions, or regional fat proportions by CLBP status. For each aim, we assumed three models. First, we only included covariates as mentioned above for the estimation of averages by CLBP status conditional on the values of the covariates. The second model included an interaction term in the form of BMI category and CLBP status for the obtainment of estimates stratified by BMI category. For the third model, we added the interaction terms of BMI category, sex, and CLBP status for stratified estimates by sex and BMI category.
Following this, we calculated the marginal averages in each stratum. We first set sex, BMI category, or CLBP status to a value of each stratum and obtained predictions based on each model. Then, we averaged the predicted values, which is the marginal average of each stratum. For example, with the model including the interaction term of CLBP status and BMI category, we calculated the marginal average of either waist circumference, body fat proportion, or regional fat distribution in every combination of CLBP status (yes vs. no) and BMI category (lower than 25 kg/m2 vs. 25 kg/m2 or higher).
We conducted a hypothesis test to compare the marginal averages in the CLBP and non-CLBP groups. Another hypothesis test was conducted to test for sex- or BMI category-related differences in the marginal averages between the two groups. A p-value lower than 0.05 was assumed to be statistically significant.
To handle missing data, we employed the multiple imputation approach. Specifically, we imputed missing values using an R package, Amelia (version 1.7.6) [20] and created 10 imputed datasets. Using the imputed datasets, we performed the analyses, as described above, and combined the results, considering the variance among the estimates from the imputed datasets.
To account for the survey design of the KNHANES IV, we used the ‘survey’ package (version 4.0) with R 4.0.2. [21, 22].
Ethics approval and consent to participate
The Institutional Review Board (IRB) of the Korea Centers for Disease Control and Prevention (KCDC) approved the KNHANES survey protocol (No:2007-02CON-04-P, 2008-04EXP-01-C, 2009-01CON-03-2C), and all participants of the KNHANES signed a written informed consent form. This study was conducted in accordance with the ethical guidelines of the Declaration of Helsinki for medical research involving human research participants.