Cohen 2004.

Methods	Randomised, double‐blind, placebo‐controlled, 3‐arm parallel study Study flow: Recruitment, 1751 women screened for study, 327 randomised, 103 to 12.5 mg paroxetine (70 completed, 95 analysed),113 to paroxetine 25 mgs (72 completed, 111 analysed), 110 to placebo (79 completed, 107 analysed)
Participants	Country: USA and Canada Site: Multi‐centre. Recruited from 43 outpatient centres Inclusion: PMDD. Women aged 18‐45 years with regular menstrual cycles (22 to 35 days) meeting diagnostic criteria for PMDD using DSM‐IV. Having symptoms of PMDD in at least 9 of 12 menstrual cycles over previous year. Onset of severe symptoms in the luteal phase subsided in the follicular phase on the four core symptoms of PMDD. Required to show a 200% worsening on one core symptom or a 100% worsening on two or more core symptoms during the luteal phase relative to their follicular phase score. A baseline Clinical Global Impression severity of illness (CGI‐S) score of ≥3. Exclusion: Presence of a primary psychiatric disorder (Axis 1 using DSM‐IV) except specific phobias in the previous 6 months; gynaecological or other clinically significant diseases, significant depressive symptoms during the follicular phase or current use of medication for PMDD symptoms.
Interventions	Screening: Two reference cycle for screening Placebo run‐in: Single blind placebo for one cycle. An additional cycle was available to patient who met all entry criteria before the first reference cycle but failed to achieve the predefined severity of core PMDD symptoms after a period of symptom tracking Interventions: Paroxetine CR 12.5 mg or 25mg, versus placebo for three treatment cycles administered daily Duration: Three treatment cycles Timing of administration: Study visits began within the first three days of menses Summary measures: Data was based on the study end point using the LOCF
Outcomes	Patient rated VAS ‐ Mood (irritability, tension, depressed mood and affective lability); Clinical Global Impression of Severity (CGI‐S); Clinical Global Impression of disease Improvement (CGI‐I); Patient rated Sheehan Disability Scale (SDS); Adverse events
Notes	Funding source not stated, but some of investigators employed by GlaxoSmithKline
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"computer‐generated randomisation code"
Allocation concealment (selection bias)	Unclear risk	Method not reported
Blinding (performance bias and detection bias) All outcomes	Low risk	"All the participants in the study, including those administering the study visits and those assessing the outcomes, were blinded to the study assignment" ..."similar appearing placebo"
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	As only 68% of women randomised completed the trial (221/327) a large number of data were imputed Text states "Fourteen randomized patients were excluded from all efficacy and safety analyses" ‐ no further explanation. Text also states "For all efficacy measures, the primary conclusions were based on the TC3 study end point using the last observation carried forward (LOCF) approach to handle missing data", but results for response rates include only 91% of "ITT population" and only 87% of those randomised
Selective reporting (reporting bias)	Unclear risk	Adverse events not a clearly prestated outcome and does not state whether adverse event data collected prospectively
Other bias	Unclear risk	Apparent exclusion of placebo responders