The causal relationships between leisure-time physical activity and body mass index in adulthood: a triangulation study

Background Previous studies have presented conflicting findings regarding the potential causal relationships between leisure-time physical activity (LTPA) and body mass index (BMI). Here, we use individual-level data and apply a triangulation framework that incorporates three complementary methods to investigate the bidirectional causal associations between LTPA and BMI. Methods We used data from a longitudinal Finnish twin cohort with assessments in 1975, 1981, 1990 and 2011. The data included 22,696 twin individuals aged 18‒50 years at baseline (52.4% women); 8,527 had genetic data available. We applied three analytical approaches suggested to strengthen causal inference in observational studies: Random intercept cross-lagged panel model (RI-CLPM) for longitudinal data, one-sample Mendelian Randomisation (MR) and Direction of Causation (DoC and MR-DoC) twin models for cross-sectional data at each measurement point. Results At the within-person level, higher BMI in 1981 was associated with lower LTPA in 1990 (RI-CLPM: standardised regression coefficient, B=-0.071, 95% CI [-0.106, -0.036]), and higher BMI in 1990 was associated with a lower LTPA in 2011 (B=-0.149 [-0.190, -0.108]). Estimates from the MR analyses of the effect of BMI on LTPA were generally negative in both 1990 and 2011 (e.g., MR-Egger: -0.096 [-0.177, -0.016] and - 0.238 [-0.319, -0.156], respectively). Twin models yielded negative estimates at each measurement point (e.g., MR-DoC: B=-0.087 [-0.161, -0.013], -0.250 [-0.358, -0.142], -0.205 [-0.291, -0.119] and - 0.257 [-0.377, -0.137], respectively). In contrast, the RI-CLPM did not indicate a within-person association of higher LTPA and subsequent lower BMI. Results from the MR analyses were inconsistent, partly due to weak instruments. The twin models yielded consistently negative estimates for the effect of LTPA on BMI (e.g., MR-DoC: B=-0.029 [-0.054, -0.004], -0.085 [-0.126, -0.044], -0.079 [-0.112, -0.046], -0.132 [-0.191, -0.073], respectively). Conclusions Consistent results across the three methodological approaches indicated that higher BMI was associated with lower LTPA at later follow-up, supporting a potential causal effect. In the opposite direction, findings were inconsistent. Potentially greater measurement error in self-reported LTPA than in BMI, together with uncertainty in questionnaire-based MET scoring and weak genetic instruments for LTPA may partly explain the inconsistent findings.