REF-1592 · Sagoe et al., 2026 · Sports Medicine - Open · Paper · PubMed · Free full text

Pooling anonymous randomised-response-type surveys from 34 studies (46,052 respondents, mostly in Europe), admitted doping was 12.7-22.6% for lifetime and 8.7-15.5% for the past year across competitive and recreational, single- and multi-sport samples

About one in six single-sport recreational sportspeople in the past year — with wide confidence intervals, no significant differences between the groups, and an average of about 29% of respondents not following the survey instructions

Main caveat: Self-reported use under randomised-response designs; about 29% average noncompliance and modelling assumptions (for example, the cheater-detection midpoint)…
moderate evidenceMeasured on recreational to elite athletesRead in full

Sagoe and colleagues pooled surveys that used indirect estimation models — question designs such as the unrelated question or forced response, in which a randomiser hides whether any one person's "yes" is an admission, so that an honest answer cannot be traced to the person.

They reviewed 46 records and pooled 55 estimates from 34 independent studies in a multilevel model, split by lifetime or past-year use, competitive or recreational samples, and single- or multi-sport samples.

Lifetime admitted doping was highest in multi-sport competitive samples (22.6%) and lowest in single-sport competitive samples (12.7%); past-year use was highest in single-sport recreational samples (15.5%) and lowest in multi-sport recreational samples (8.7%). Confidence intervals overlapped and the group differences were not significant.

The authors flag noncompliance with survey instructions, which averaged 28.81%, and re-analyses of the same data in several outputs, as challenges for these estimates.

Participants

34 pooled studies, 46,052 participants (16,046 men and 3,118 women where sex was given); samples of 249 to 4,629 (mean 1,354).
Mostly multi-sport (22 studies); competition levels from school and local to international; mostly Germany (10), the Netherlands (4) and the UK (3).

Protocol

Models in the pooled studies: unrelated question 8, forced response with cheater detection 6 and without 3, crosswise 4, Kuk's design 3, single sample count 2, combinations 3.
Pooling: estimates converted to the probit scale, weighted by inverse variance, multilevel (estimates within studies) with prevalence type, sample type and sport type as predictors; cheater-detection estimates entered as the midpoint between lower and upper bounds.
Data collection in the studies: 2003 to 2022, online, paper or tablet surveys. The results text says most data (28) were collected at sport events, while the abstract and discussion say most were collected outside events.
Funding and interests: no funding beyond open-access fees (University of Bergen); four authors (AP, DS, MC, OH) members of WADA's Prevalence Working Group since 2021, with expenses and meeting honoraria.

Key Results

SampleLifetime, % (95% CI)Past year, % (95% CI)
Competitive, multi-sport22.6 (14.3-33.1)14.1 (8.9-21.1)
Competitive, single-sport12.7 (1.9-41.7)12.7 (5.6-24.5)
Recreational, multi-sport15.0 (7.1-27.2)8.7 (6.0-12.2)
Recreational, single-sport15.4 (1.5-55.4)15.5 (5.5-33.3)
No significant lifetime or past-year difference by sample type or sport type in the multilevel model (authors).
Instruction noncompliance averaged 28.81% (SD 17.44, range 0.0-64.9); 13 studies estimated it, one corrected for self-protective noncompliance, 15 did not consider it.
A quarter of studies did not give the exact wording of the doping question; wording varied in precision and neutrality.
For contrast, the discussion cites adverse analytical findings at about 1% of tests a year and sanctioned violations at about 0.6-0.7% of tested athletes a year.

Practical Application

**Does everyone dope, amateurs included?In anonymous surveys built so that no single answer gives the person away (34 studies, mostly European), 8.7-15.5% of sportspeople admitted doping in the past year and 12.7-22.6% ever, and recreational samples did not differ significantly from competitive ones, so amateurs are not exempt.** The point estimates leave most respondents in every group not admitting use (our arithmetic: at least 77% in each group), these are self-reports with wide uncertainty, and 16 studies defined doping other than by the WADA code (steroids, SARMs, illicit, prescription or controlled drugs), so not every estimate counts the same thing.
Admitted use in these surveys runs far above the about 1% of tests that come back positive (our comparison, from figures in the discussion). Inference, not tested here: a clean test record is not a measure of how common use is.
"Everyone does it" is not what these numbers show: the highest past-year estimate was about one in six (15.5%, single-sport recreational samples; authors' framing), though its upper confidence limit was 33.3%.
For what hidden drugs in supplements mean for a tested athlete, see SYNTH-67 and REF-358.

Limitations

Self-reported use under randomised-response designs; about 29% average noncompliance and modelling assumptions (for example, the cheater-detection midpoint) can move the estimates either way.
Definitions of doping and time frames differ between studies; a quarter did not report question wording.
Mostly European and German samples; elite and non-elite competitors could not be separated.
No formal risk-of-bias tool; one author extracted the data; several outputs re-analyse the same data.
The paper is internally inconsistent on whether most data came from sport events.
Four authors are members of WADA's Prevalence Working Group.
How this entry was checked and changed
Why this level

a registered review and meta-analysis read in full, but no formal risk-of-bias tool was applied to the surveys (instruction noncompliance and question wording were described instead, and the first author extracted the data); the rubric's A- needs the risk of bias read, and the pooled body is self-reported prevalence