Statistical analyses

Statistical analyses were conducted using STATA version 13 (StataCorp, College Station, TX, USA). The weighted prevalence (using the “svy” estimators and “pweight” as the weighting variable) of foot pain obtained from Stage 2 and Stage 3 was determined and the frequency of each of the variables measured as part of the Stage 2 assessment, for those with and without foot pain on most days in the last month, was determined. These variables were then used as predictors for the presence of foot pain as determined at Stage 3 and thus, tests aimed at examining how well variables predicted the presence of foot pain were used. The sensitivity (probability that the associated factor is present when foot pain is present), specificity (probability that the associated factor is not present when foot pain is not present), positive predictive value of each of the predictor variables (the probability that foot pain is present given that the associated factor is present) and the area under the Receiver Operating Characteristic (ROC) curve (the summary of the performance of each variable in terms of the ability to predict foot pain) were all determined. An area under the ROC curve equal to one indicates that a variable perfectly predicts those at risk of the outcome whereas a value of 0.5 represents a predictor that is no better than chance at predicting the outcome of interest. The area under the ROC curve was calculated using predictors as binary variables as this is the most common occurrence in clinical practice, however is must be noted that generally continuous variables are more appropriate to use for ease of interpretability.

Generalised linear models using the binary outcome variable of presence of foot pain/no foot pain on most days in the past month were used with the “svy” estimators and weighted data to determine the relative risks (RR) of each of the predictors in association with the outcome variable. Multivariable models were created with all relevant predictor variables included. The participants reporting toe and nail pain were combined (due to the small number of participants reporting the presence of toe pain). In addition, the areas of joint pain were combined. Hand and shoulder pain were defined as upper limb pain and hip and knee pain were defined as lower limb pain. Back pain remained a separate variable. In total, four separate multivariable models were created with adjustment for all possible predictors.