Hefler 2006.

Methods	True randomisation with allocation by computer‐generated blocks. Parallel design without cross‐over. Patient and physicians were blinded. No mention of pathologist. Analysis by ITT. No mention of toxicity assessment. Enrollment May to October 2005. Study was closed after 87 days due to withdrawal of rofecoxib from the market. Randomisation code broken at this point
Participants	16 women with histological evidence of CIN 2 or 3, a fully visible transformation zone and lesion margin, compliant patient, and safe contraception. Exclusions included presence of (micro‐)invasive cancer, endocervical lesion, upper margin of lesion not visible on colposcopy, non‐compliant patient, prior history of an adverse gastrointestinal event (ulcer, haemorrhage), age greater than 60 years; concurrent use of glucocorticoids, or NSAID allergy. Age: range 24 to 36 years, mean 27.1 years for treatment arm and 31.2 years for placebo arm
Interventions	Rofecoxib 25 mg or placebo, by mouth, daily for three months. After the screening visit a physical examination was performed, a questionnaire was answered and study medication was distributed. At initial visit all patients meeting inclusion criteria were evaluated for HPV status via Hybrid Capture (Digene Corp., Gaithersburg, MD, USA). Follow‐up examinations performed at three and six months and included gynaecological examination with ecto‐ and endocervical cytology smears, colposcopy and biopsy, HPV test and pregnancy test. These modalities were used for the evaluation of regression, persistence, or progression of cervical dysplasia
Outcomes	Primary outcomes assessed were regression and remission of CIN over a 3‐month treatment period, and secondary were safety and side effects. 16 women participated: eight in the treatment arm (four with CIN 2; four with CIN 3) and eight in the placebo arm (five with CIN 2; three with CIN 3). Regression occurred in two of eight (25%) in the treatment arm and one of eight (12.5%) in the placebo arm (P value 0.9), after a mean of 87 days of treatment
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Participants were randomised using a random number sequence with permuted block size of 12
Allocation concealment (selection bias)	Unclear risk	Not stated
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Randomisation code broken when rofecoxib withdrawn from the market. Statistician and pathologist blinded at all times
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Randomisation code broken when rofecoxib withdrawn from the market. Statistician and pathologist blinded at all times
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Not clear
Selective reporting (reporting bias)	Unclear risk	Insufficient information to judge
Other bias	Unclear risk	Early closure of study. Insufficient information to assess whether an important risk of bias exists