Farley 2006.

Methods	True randomisation with allocation by computer‐generated blocks. Parallel design without cross‐over. Patient, provider, pathologist blinded. All pathologic specimens were reviewed by a blinded central pathologist. Enrollment June 2002 to October 2003. Analysis was by ITT
Participants	25 women 18 years or older with biopsy proven CIN 2 or CIN 3. Exclusions included: pregnancy or breast feeding, history of stomach bleeding, NSAID allergy, severe kidney disease (such as creatinine greater than 1.2 mg/dL) or liver problems (such as AST/ALT greater than 80 U/L), immunosuppression, NSAIDs use for other medical conditions, or NSAID‐associated asthma. Also women with positive dysplasia on endocervical curettage or cervical cytology diagnosis of atypical glandular cells, carcinoma in situ, adenocarcinoma in situ, or invasive carcinoma. Age range 19 to 27 years. Race distribution not stated. Study performed at Tripler Army Medical Center, Hawaii, USA.
Interventions	Celecoxib 200 mg or placebo by mouth twice daily with meals for six months or until progression of dysplasia. All patients underwent initial colposcopy, thin‐prep liquid‐based cervical cytology, reflex HPV testing, and colpography of lesion with specimens reviewed by a blinded pathologist. Patients were seen at eight‐week intervals for six months with thin‐prep liquid‐based cervical cytology, colposcopy, photography of the cervix and biopsy. If colposcopy was normal, representative biopsy was taken from the area of previous dysplastic abnormality. Participants were removed from the study and underwent loop electrosurgical excision procedure (LEEP) for any increase in severity of CIN on cytology or histology. After six months, persistent CIN 3 required removal from study and treatment with LEEP. Participants missing an eight‐week follow‐up were contacted and asked to return for evaluation as soon as possible. Anyone not attending follow‐up was removed from the study and treated with LEEP
Outcomes	Primary objectives were to determine response rate to treatment and toxicity of treatment. Toxicity was assessed at each eight‐week follow‐up visit by the treating pharmacist by direct questioning about the most common side effects of celecoxib. No toxicity was reported by participants. Response was defined as any decrease in severity of CIN on histology. A complete response was defined as a complete visual resolution of the lesion on colposcopy and normal thin‐prep liquid‐based cytology and tissue histology. Partial response was defined as a decrease in severity of CIN on histology with no cytologic change in severity of CIN. Stable disease defined as no change in histologic degree of CIN from entry cervical biopsy. Progression was defined as any increase in severity of CIN on cytology or histology. Of the 25 participants 13 were randomised to the placebo and 12 to the treatment arm. Five participants (two in the treatment and three in the placebo arm) discontinued the study early because they desired definitive treatment. Overall response was N = 4 (31%) placebo versus N = 9 (75%) treatment. Mean time to response was 72 (+/‐ 38) days in the placebo arm and 73 (+/‐ 26) days in the treatment arm (P value 0.39). Complete response N = 2 (15%) placebo versus N = 4 (33%) treatment. Partial response was N = 2 (15%) placebo versus N = 5 (42%) treatment. Progression to higher degree of CIN was N = 2 (15%) placebo and N = 1 (8%) treatment arm with a mean time to progression of 65 days. No patients progressed to invasive carcinoma. The treatment arm was 2.5 x more likely to have regression of their cervical dysplasia
Notes	100% of treatment arm women were positive for high‐risk types of HPV by Hybrid Capture II in comparison with 85% in the placebo arm. Eleven of 12 (92%) women in the treatment arm and 11 of 13 (85%) in the placebo arm had CIN 2 with 1 of 12 (8%) in the treatment arm and 2 of 13 (15%) in the placebo arm having CIN 3. Two women in the treatment arm regressed from CIN 2 to CIN 1 in eight weeks, but wished then to be treated with LEEP, having final diagnoses of moderate dysplasia
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using a computer generator Random Allocation Software, which randomised each participant to a single treatment, by using the method of randomly permuted blocks
Allocation concealment (selection bias)	Low risk	Allocation concealment was accomplished by the Department of Pharmacy. All physicians who participated in the implementation of the treatments, medication or placebo, were blinded to the randomisation process. Only the Pharmacy co‐ordinator (E.G.) was aware of the allocation of the medications. Participants obtained their medications from E.G. in a separate encounter to their clinical follow‐up
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Both the patients and the examining physicians were blinded to the treatment option to which the patients were randomised
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Both the patients and the examining physicians were blinded to the treatment option to which the patients were randomised
Incomplete outcome data (attrition bias) All outcomes	High risk	Five patients (20%) discontinued the study early but were included in the final statistical analysis
Selective reporting (reporting bias)	Unclear risk	Insufficient information
Other bias	Unclear risk	Insufficient information to assess whether an important risk of bias exists