On the relevance of surrogate parameter deduction in biomedical research: mediated regression analysis for variance explanation of cervical range of motion.

Abstract

Research on cervical range of motion (ROM) often includes age and body mass index (BMI) as potential variables to explain inter-individual variances. The BMI may not be a predictor of ROM but an age-affected surrogate parameter: the described effect of BMI on ROM is, thus, suspected being partially or completely mediated by age. Healthy and adult volunteers (n = 139, 65 female, age 19-75 years, BMI 24.2 ± 3.8 kg m^-2) performed five repetitive maximal cervical movements in the sagittal plane to assess maximal ROM (primary outcome). After the examination of underlying assumptions, data were analysed by mediation regression analyses using a SPSS-macro provided by Hayes. ROM represented the outcome variable, independent variable was BMI and mediator variable was age. Total as well as direct and indirect effects were calculated: (1) for all subjects included and (2) for subject with a BMI <35 kg m^-2. Analysis including all subjects revealed both a direct (-1.1, s _e .46, p < .05, 95 %CI -2; -1.7) and an existing indirect effect (mediated by age, -2.4, s _e .33, p < .05, 95 %CI -3.1; -1.8) of BMI on ROM. Analysis without obese 2 subjects showed no direct effect of BMI (effect -1, s _e .54, p > .05, 95 %CI -2.1; +.1) but a systematic indirect effect, mediated by age, on ROM (effect -2.4, s _e .33, p < .001, 95 %CI -3.1; -1.8). After the withdrawal of the surrogate parameter BMI, age explains 53 % of maximal ROM. No impact of BMI on ROM was detected after excluding highly obese participants. Our results illustrate the relevance of including each supposable predictor in causal mediation models development.