Performance-enhancing drug use in sports: a Bayesian variable selection and profile regression approach to identifying risk factors and estimating prevalence.

BACKGROUND: This study has two objectives: (1) to identify the attitudinal and demographic factors most closely associated with performance-enhancing drugs (PED) use in sports; (2) to propose a new method for estimating PED use prevalence that mitigates the limitations of self-report surveys. METHODS: We analyse survey data from a sample of 2113 athletes attending ten UK universities, using a two-step approach. First, Bayesian Variable Selection (BVS) is used to identify the factors most closely associated with PED use. Second, these factors are incorporated into a Bayesian Profile Regression (BPR) model to identify clusters of athletes with similar attitudes toward PEDs. The prevalence of PED use is then estimated by linking these clusters with reported PED use via a logit model. RESULTS: Four key factors are significantly associated with PED use: being male, believing PED use is necessary to excel in sports, having a perceived estimate of PED use prevalence among elite athletes higher than the sample average, and having a perceived estimate of PED use prevalence among sportspeople in general higher than the sample average. BPR identifies four distinct clusters of athletes: (1) non-users; (2) use-admitters; (3) use-admitters and non-admitters; (4) most likely non-users. Using this new methodology, PED use prevalence is estimated at 13.7%, considerably higher than the 3.4% obtained through direct questioning. CONCLUSION: The proposed method identifies attitudinal and demographic factors associated with PED use and provides a more accurate estimate of its prevalence, which can inform the development of more effective anti-doping programmes.