Correlations between exposures were estimated using the Spearman rank correlation coefficient. Hazard ratios (HR) between exposures and incident CTS were estimated using Cox proportional hazards regression with robust confidence intervals. Guided by DAGs, the models were adjusted for potential confounding by personal factors related to both exposure and outcome that were not on the causal pathway. Using the forward stepwise procedure, variables were retained in the model if inclusion resulted in a change of the effect estimate of the primary exposure variable by 10% or more17 (link). Ultimately, age, gender, BMI and study site were included in all models. Models where specific biomechanical exposures were the primary exposure of interest were adjusted for dissimilar biomechanical exposures (ie., exposures of a different type)9 (link). For example, the relationship between Peak Hand Force and CTS was adjusted for Total Repetition Rate and wrist posture whereas the model assessing the relationship between Forceful Repetition Rate and CTS was only adjusted for wrist posture. Assessment of confounding of one class of exposures by another (e.g., biomechanical, work psychosocial) used the same process and criteria described above. All analyses were implemented with the Stata statistical package (Stata, College Station, TX).