Data sources
This study used data from the UCLA Low Back Pain Study conducted from 1995 to 2000. In this study, 681 participants with LBP were enrolled. Comprehensive data were collected regarding the current episode of LBP, as well as the LBP history of each participant [20]. Supplementary information was collected regarding the demographics, occupation history, disability, health status and mental health status of each patient. Outcomes were collected on each patient at 2 weeks, 4 weeks, 6 weeks and 6 months after the initial enrollment. This study used the data gathered at the 6-month follow-up period. There was a 4.3 % loss to follow-up, leaving 652 participants.
Study design
Patients presenting to the Friendly Hills Healthcare Clinic with a complaint of LBP were offered information regarding the study. Those interested and agreeing to participate were subsequently enrolled. Individuals presenting with fractures, tumors, infections, rheumatic disease or other severe coexisting conditions were excluded from participation in the study. The majority of clinic patients belonged to a HMO and did not have to pay for services. Patients utilizing MediCal or Medicare insurance, had a Workman’s Compensation injury, or who were fee for service or had services paid by a third-party payer were eliminated from consideration. Also eliminated were patients who were unable to effectively communicate in English and those who had been under care for LBP within the previous 1 month. Of the 1203 potential study participants, 273 (22.7 %) were excluded due to ineligibility and 249 (20.7 %) were eligible but declined participation.
Patients who were enrolled were randomized into one of four treatment groups: medical care only, medical care with physical therapy, chiropractic care only, or chiropractic care with the use of physical modalities. Participants were then treated according to the treatment plan prescribed by their assigned doctor.
Outcome variables
Participants completed the Roland-Morris disability questionnaire at baseline and at the 6-month follow-up period. The difference between these two scores was measured, and an improvement of three or more was considered a positive outcome (clinically meaningful improvement). Participants also completed an 11-point numerical rating scale (NRS) in which 0 is no pain and 10 is unbearable pain, indicating the average and most severe level of LBP they had experienced in the previous week. Again, the 6-month value was subtracted from the baseline value. Clinically meaningful improvement was considered to be a positive change on the NRS greater than or equal to two.
Exposure variable
BMI was used as the primary exposure variable. It was modeled two ways. The first was according to National Health Guidelines, in which BMI less than 19.0 is considered underweight, greater than or equal to 19.0 and less than 25.0 is considered normal weight, greater than 25.0 and less than 30.0 is considered overweight, and greater than 30.0 is considered obese [21, 22]. BMI was also considered in a dichotomous format, obese (>30) versus not obese (≤30). BMI data was available for 618 study patients for which 6 month follow up occurred. Of these, 7 were categorized as underweight (1.13 %), 178 as normal weight (28.80 %), 250 as overweight (40.45 %), and 183 obese (29.61 %).
Potential confounders
Potential confounders were sociodemographic variables (gender, race, age), lifestyle variables (smoking status, alcohol consumption, coffee consumption, weight change over 6 months), LBP variables (duration of current episode, history of previous episode, baseline Roland-Morris and NRS scores), assigned treatment group and patient confidence in the treatment plan. Race was categorized as white and non-white. Age was categorized by decade. Due to the low number of participants under the age of 20 and over the age of 79, participants under 20 were combined with 20–29 year olds to create a single category. Participants 80–95 were similarly combined into a single category. Other than weight change, lifestyle variables were all set as dichotomous (yes/no), as was the history of LBP variable. Duration of LBP was dichotomized to less than 3 months vs. over 3 months. Weight change was categorized as lost weight, gained weight, or had little or no weight change (less than five pounds in either direction).
Confidence was categorized as high (8–10 on a 0–10 scale), moderate (4–7) and low (0–3). Confidence was established by questioning patients on their belief that the treatment received would actually help their back condition to improve. Finally, the baseline scores for the three outcome variables were considered for the purpose of stratification. It was hypothesized that those with lower initial scores (less disability and/or less subjective pain) may have less opportunity to and therefore be less likely to improve. The Roland-Morris scores were grouped as 0–6, 7–10, 11–14, and 15–24 (approximating quartiles). The pain variables were grouped as low (0–3), moderate (4–7) and high (8–10).
Statistical methods
Binary logistic regression was used to estimate the effect of and test the association between BMI and the outcome variables, controlling for confounders. Odds ratios derived from the model results were used to measure effects. Comparisons were made between obese participants and non-obese participants for each of the three outcome variables.
Initially, binary logistic regression models included gender, race, age, treatment group, smoking status, alcohol consumption, coffee consumption, LBP history, duration of the current episode of LBP, baseline scores of the outcome variable, and patient confidence in the treatment. Gender, race, and smoking status were included because of a priori knowledge regarding their effects on LBP. Treatment group was also included to account for varying treatment protocols and their effects on treatment outcome.
Weight change, a history of LBP, and the duration of the current episode of LBP were considered as possibly having confounding influence on the BMI effect on treatment outcome. Likewise, patient confidence in treatment and the baseline scores of the outcome variable were also considered as possibly confounding the BMI effect. Models were then run with and without these variables to evaluate the influence each variable had on the effect of BMI on outcomes. Variables influencing the odds ratios of the remaining variables by more than 10 % were kept in the model. As a result, LBP history and weight change were eliminated from the disability and average pain models. LBP history was also removed from the severe pain model, but the weight change variable remained. Statistical analyses were performed using SAS (SAS Institute Inc., Cary, NC).