We performed a post hoc data analysis on 2 studies performed involving 50 subjects (Clinical trial registrations ISRCTN75758249, retrospectively registered and ISRCTN58369615, retrospectively registered [10, 12]). The first was a randomized open-label parallel study with metformin and rimonabant in 20 patients with PCOS with a body mass index (BMI) ≥30 kg/m2. The 20 patients were randomized to either metformin 500 mg t.d.s. or rimonabant 20 mg daily. The second study was a randomized open-label parallel study with metformin, pioglitazone and orlistat in 30 patients with PCOS with a body mass index (BMI) ≥30 kg/m2. The 30 patients were randomized comparing treatment with metformin (500 mg three times a day), orlistat (120 mg three times a day) and pioglitazone (45 mg once daily) in hyperandrogenic, anovulatory Caucasian women with PCOS. It can be seen that the four groups were weight matched with a BMI range of 34.2 to 41.0. In both cases the diagnosis of PCOS was based on all three diagnostic criteria of the Rotterdam consensus, namely clinical and biochemical evidence of hyperandrogenaemia (Ferriman–Gallwey score > 8 and FAI > 4, respectively), oligomenorrhoea or amenorrhoea, and polycystic ovaries on transvaginal ultrasound [10, 12]. As detailed all subjects had a normal liver ultrasound with no evidence of a fatty liver indicative of NAFLD. Subjects had no concurrent illness, were not on any medication for the preceding 6 months and were not planning to conceive. None of the patients had had a successful pregnancy or a miscarriage at least 5 years prior to the study entry. Subjects were advised not to change their lifestyle, including physical activity and dietary habits, during the study period. Non-classical 21-hydroxylase deficiency, hyperprolactinaemia, Cushing’s disease and androgen-secreting tumours were excluded by appropriate tests. Adherence was monitored by counting the returned medication. All patients gave written informed consent. Additional IRB approval was not needed for this post hoc analysis as informed consent from the patients for future analyses on anonymous data had been given at the time of the initial consent procedure.
Randomization was performed using a random number generator. Both studies were approved by the Hull and East Riding Local Research Ethics Committee and both studies have been conducted in accord with the Consolidated Standards for Reporting Trials (CONSORT) 2010 statement .
Clinical and biochemical assessments were performed at randomization and at the end of the 3-month period for the first study [10, 12]. For the second biological variability study, clinical and biochemical assessments were performed at each visit on 10 consecutive occasions at 4-day intervals before and 12-weeks after treatment .
Methods have been detailed previously [10, 12] but briefly, study bloods and measurements were undertaken after an overnight fast. Fasting venous blood was collected into serum gel and fluoride oxalate tubes. Samples were separated by centrifugation at 2000 g for 15 min at 4oC, and the aliquots stored at –20oC. Serum testosterone was measured on an Architect analyser (Abbott Laboratories, Maidenhead, UK) and SHBG was measured by immunometric assay with fluorescence detection on a DPC Immulite 2000 analyser (Euro/DPC, Llanberis, UK) using the manufacturer’s recommended protocol. The FAI was obtained as the total testosterone × 100/SHBG. Total cholesterol, triglycerides and high-density lipoprotein cholesterol (HDL-C) levels were measured enzymatically using a Synchron LX20 analyser (Beckman-Coulter, High Wycombe, UK). Low-density lipoprotein cholesterol (LDL-C) was calculated using the Friedewald equation. Serum insulin was assayed using a competitive chemiluminescent immunoassay performed on the manufacturer’s DPC Immulite 2000 analyser (Euro/DPC). The analytical sensitivity of the insulin assay was 2μU/ml, the coefficient of variation was 6%, and there was no stated cross-reactivity with proinsulin. Plasma glucose was measured using the Synchron LX20 analyser (Beckman-Coulter), using the manufacturer’s recommended protocol. The coefficient of variation for the assay was 1·2% at a mean glucose value of 94·6 mg/dl (5·3 mmol/l) during the study period. The insulin resistance (IR) was calculated using the homeostasis model assessment (HOMA) method [HOMA-IR = (insulin x glucose)/22·5].
The Bio-Plex 200 system with HTF (Bio-Rad, Hercules, CA) was used to evaluate the sera and the levels of IL-1β, IL-6, IL-7, IL-10, IL12, TNF-α, MCP-1 and INF-γ and compared to a set of standards that were run simultaneously in the assay .
ALT was used specifically in this study to calculate the HAIR score (hypertension, ALT and insulin resistance) that is a predictive model for NASH (area under the receiver operator curve of 0.85) [16, 17]. Given that the mean ALT may not differ, but the biological variability may decrease specifically with weight loss, ALT biological variation was determined by ALT measurement performed at each of 10 consecutive visits at 4-day intervals, before and after 12 weeks treatment. All samples were taken between 8 and 9 am. Clinical assessments were also taken at these visits. Before analysis the samples were thawed thoroughly and mixed. Alanine aminotransferase was measured on the Unicel® DxC 80 analyser (Beckamn-Coulter, High Wycome, UK). Data are reported as mean ± SD. AST and gammaGT were not measured but have not been shown to be any better than ALT in NAFLD.
Statistical analysis was performed using SPSS for Windows version 16.0 (SPSS inc). The power of the studies was based on a significant reduction in total testosterone concentration after treatment with metformin [10, 12]. In data where distribution between individuals violated the assumptions of normality when tested using the Kolmogorov-Smirnov test, median values were used and differences calculated by paired Wilcoxon method. Given that these were randomised clinical trials baseline values were not compared or adjusted for . Biovariability data was analysed by calculating analytical, within subject, and between subject variances (SDA
2, respectively) as detailed previously (5,6). Using this technique, analytical variance (SDA
2) was calculated from the difference between duplicate results for each specimen (SDA
2 = Σd2/2N, where d is the difference between duplicates, and N is the number of paired results). The variance of the first set of duplicate results for each subject on the ten assessment days was used to calculate the average biological intraindividual variance (SDI
2) by subtraction of SDA
2 from the observed dispersion (equal to SDI
2 + SDA
2). Subtracting SDI
2 + SDA
2 from the overall variance of the set of first results determined the interindividual variance (SDG
2). The intraindividual (SDI) and interindividual (SDG) variations were estimated as square roots of the respective variance component estimates. For all analysis, a two-tailed P < 0.05 was considered to indicate statistical significance.