The investigators wanted to elucidate the effects of metformin in Polycystic Ovary Syndrome (PCOS) by performing a randomized, double-blinded, placebo-controlled cross-over study.

Eligible women who gave written informed consent were assigned to 6 months of treatment with either 850 mg of metformin or placebo twice daily, followed by a wash-out period of 3 months before cross-over to the alternate treatment for another 6 months. Randomization defining treatment sequence was done at inclusion by random number tables. The appearance of the tablets was identical, and patients and investigators were blinded to treatment allocation. The randomization code was stored in a closed envelope until all participants had finished the treatment. Participants were seen by one of the investigators before inclusion and every second month during treatment periods, always in the morning after an overnight fast of at least 8 h. They were weighed wearing light clothing. Waist circumference was measured at the umbilical level and hip circumference at the trochanter region. Systolic and diastolic blood pressure was measured with a semiautomatic blood pressure monitor and a blood sample was drawn for immediate analysis without respect to bleeding periods. All participants registered their bleeding periods in a calendar during both study periods and the 3 months wash-out period. Calculation of sample size was based on the assumption that at least 50% of the women would experience at least 30% more menstrual periods on metformin than on placebo. Based on a power of 90 (b = 0.10) to detect a significant difference [two-sided P-value of 0.05], the minimum sampling size was calculated to 44 subjects.

We also assumed a 10% drop-out rate, and thus aimed at including 50 women. As drop-out rate quickly rose higher than expected, we decided to include 60 women. In the intention-to-treat analysis, the values of each participant after 6 months of metformin or placebo were compared with the baseline values. Linear regression analysis with the changes in testosterone and homeostasis model assessment (HOMA) index as dependent variables was performed to examine potential relations between the changes. The per protocol analysis included data from participants completing both study periods, i.e. the difference between the values of each participant after placebo and metformin, respectively, was calculated, and a significance test performed on the differences.

Inclusion Criteria:

• testosterone value above the upper normal limit and
• oligo- or amenorrhea,

Exclusion Criteria:

• periclimacteric gonadotrophin values,
• hyperprolactinaemia,
• diabetes mellitus,
• impaired thyroid,
• renal or hepatic function,
• hormonal treatment,
• pregnancy,
• lactation or a wish for fertility treatment.