Which outcomes PCOS trials report
A count of 576 randomised PCOS treatment trials published 2016 to 2026, and which outcomes each one reports.
Quality of life or satisfaction
10.6%of 576 trials report quality of life or treatment satisfaction (95% interval 8.3 to 13.4). experiments/outcome_audit/LINKAGE_RESULTS.md:52-53
This is the cleanest number on the page. Quality of life and satisfaction are two of the outcomes the 2020 PCOS core outcome set recommends every trial report, so the definition comes from outside this project. experiments/outcome_audit/CROSS_CONDITION_RESULTS.md:169,179-180 "Reported" means the trial's paper gives a result for it or clearly states it was assessed. A trial can collect an outcome and never report it; we can only count what is in print.
Outcomes on our patient-centred list
Our patient-centred list is: quality of life, treatment satisfaction, depression, anxiety, eating disorders, diabetes as a diagnosis, live birth, and patient-reported symptom scores. experiments/outcome_audit/RESULTS.md:23-25 How many trials report "any of them" depends on what goes on the list, so we give the range rather than one number.
| Definition | Trials reporting at least one |
|---|---|
| List without live birth and diabetes | 15.7% |
| List as registered | 22.2% (19.0 to 25.8) |
| List plus periods, hair and acne | 45.0% |
The range is the honest finding: the answer depends on the definition. Periods, hair and acne are symptoms women live with, but they were not on the registered list.
What trials report most, and least
From the first reading, 412 published trial reports in the audit sample. experiments/outcome_audit/RESULTS.md:40-62
| Outcome | Share reporting it |
|---|---|
| Hormone levels | 54% |
| Glucose and insulin | 47% |
| BMI | 45% |
| Androgens (blood) | 41% |
| Periods | 22% |
| Hair growth | 16% |
| Acne | 8% |
| Quality of life | 6.6% |
| Depression | 6.6% |
| Anxiety | 6.1% |
| Treatment satisfaction | 2.7% |
| Diabetes as a diagnosis | 0.5% (2 trials) |
- Fertility trials (95): 76% report pregnancy; 17% report live birth. experiments/outcome_audit/RESULTS.md:60
- 9.5% of trials report only lab values and weight measures. experiments/outcome_audit/RESULTS.md:62
Most trials last about 12 weeks, which is too short for diabetes to be diagnosed. The low diabetes figure says as much about trial length as about reporting. experiments/outcome_audit/RESULTS.md:15-16
Over time
This comparison was chosen after the first results were seen.
| Publication years | Quality of life or satisfaction |
|---|---|
| 2016 to 2019 | 6.8% (4.1 to 11.1) |
| 2020 to 2022 | 10.2% (6.5 to 15.7) |
| 2023 to 2026 | 14.7% (10.5 to 20.2) |
- Adjusted odds ratio for the latest period against the first: 2.2 (1.06 to 4.4). experiments/outcome_audit/PUBLISH_FIRST_PLAN.md:50 docs/REVIEW_PMOS_AUDIT_2026-10-26.md:174
- By the registered rule, no change is called: the first and last intervals overlap. experiments/outcome_audit/LINKAGE_RESULTS.md:77-80
- A registered test of the trend, on records that did not produce it, is drafted and has not been run. experiments/outcome_audit/PUBLISH_FIRST_PLAN.md:50-51
Why these are lower bounds
- 325 of 576 trials print no registry ID, so their follow-up papers could not be linked. experiments/outcome_audit/LINKAGE_RESULTS.md:25
- Linking follow-up papers added a patient-centred outcome to 3 trials, and a second primary paper added one to 3 more. experiments/outcome_audit/LINKAGE_RESULTS.md:12-13
- The search covered MEDLINE, and is English-heavy.
Retractions and notices
A later check of the 602 eligible papers against PubMed found 17 with a linked retraction, 22 with a linked expression of concern and 14 with a linked erratum. experiments/source_check/outcome-audit/SOURCE_CHECK.md:3,13-15
The figures above still include these papers. A rule for handling them will be written down before the numbers are recomputed.
How it was read
Each trial report was coded against a fixed codebook by an AI reader. A second reading of a sample used the same model again experiments/outcome_audit/PUBLISH_FIRST_PLAN.md:53, so agreement between those two readings measures consistency, not accuracy. The independent check is the re-reading by models from two other vendors, described in the box above. The plan was committed before any reading began. experiments/outcome_audit/RESULTS.md:18