Serum androgen profiles in women with premature ovarian insufficiency: a systematic review and meta-analysis.

In this meta-analysis, all published studies in which serum androgen concentrations including total testosterone (T), androstenedione (A), dehydroepiandrosterone sulfate (DHEA-S), sex hormone-binding globulin (SHBG) were described for women with POI/POF and compared with healthy controls, were considered eligible. The predefined criteria for inclusion were as follows: POI/POF had to be defined as the disappearance of periods (amenorrhea) for the past 3 months; the disappearance of periods must have occurred even before the age of 40 is attained; in addition, the participants of the study must have had elevated follicle-stimulating hormone (FSH) levels and reduced estradiol concentrations; controls were required to be women without POI/POF. The predefined exclusion criteria were as follows: the studies were excluded if the participants had hyperandrogenemia, bilateral salpingo-oophorectomy (BSO), or gonadotoxic treatment performed after menopause; if the participants used hormone therapy, the study was excluded; studies including participants other than females were excluded; studies, which focused on chromosomally abnormal POI/POF women, were excluded; reviews, case reports, letters to the editor, and conference papers were not considered for inclusion. Only English language articles were examined. Study methodology was not considered as an exclusion criterion. The study selection processes were carried out in two phases. First and foremost, screening of titles and abstract was done to meet the inclusion criterion by two separate researchers. The consensus among themselves resolved any disagreements. The searched databases were PubMed, Cochrane library, and Medline. The time range was inception to February, 2018. Primarily the studies were evaluated by reading the titles and abstracts. The search terms were menopause, premature, ovarian, insufficiency, failure, POI, POF, testosterone, total testosterone, DHEA-S, androstenedione, and SHBG. The search syntax for PubMed central was:(“menopause, premature”[MeSH Terms] OR (“menopause”[All Fields] AND “premature”[All Fields]) OR “premature menopause”[All Fields] OR (“premature”[All Fields] AND “ovarian”[All Fields] AND “failure”[All Fields]) OR “premature ovarian failure”[All Fields]) AND (“androgens”[All Fields] OR “androgens”[MeSH Terms] OR “androgens”[All Fields] OR “androgen”[All Fields]). The search was modified for other databases used. Reference lists of relevant review articles were searched for potentially eligible studies. When required, authors were contacted for additional information. Studies were scrutinized by two reviewers independently according to the predefined inclusion-exclusion criteria. The final decisions about the included articles were made by examining the full articles. Two independent reviewers assessed the risk of bias with the Newcastle-Ottawa Scale for meta-analysis of observational studies. 29 Newcastle-Ottawa Scale for meta-analysis of observational studies grades studies based on three factors: selection, comparability, and exposure. The criteria for these assessments were as follows: is the case definition adequate? representativeness of the cases; selection of controls; definition of controls; comparability of cases and controls based on the design or analysis; ascertainment of exposure; the same method of ascertainment for cases and controls; and nonresponse rate. Two reviewers extracted data from all eligible articles independently. Data that were collected included the year of publication, study design, country, age, body mass index (BMI), number of participants, mean and standard deviation (SD) of testosterone concentrations (ng/dL), androstenedione (ng/dL), DHEA-S (μg/mL), and SHBG (nmol/L) for both cases and controls. Selection can have a maximum of 4 ∗ s, comparability a maximum of 2 ∗ s based on the two important confounding factors. In our study, we chose age and BMI as the two factors. Exposure can have a maximum of 3 ∗ s. So, altogether, 9 ∗ s are the highest that can be obtained. The high and low score ranges are arbitrarily given. Here the lowest score was 2 ∗ s and highest was 8 ∗ s. So we used a scale of 0 to 2 ∗ s being poor quality, 3 to 5 ∗ s as fair quality, and 6 to 9 ∗ s being higher quality. A study is considered good if it can score at least 1 ∗ in each category, and more ∗ s means better quality generally. In this meta-analysis, the SMD with 95% confidence intervals (CIs) were solved to determine T concentrations, A concentrations, DHEA-S, and SHBG between POI/POF cases and controls. Random-effect model (REM) 30 was used to calculate the pooled effect size with 95% CI. Analyses were conducted using Revman 5.3. 31 Heterogeneity among studies was assessed using chi-square and quantified with the I 2 index. If the I 2 index is 0% to 40%, it means low heterogeneity and might not be important; if the I 2 index is 30% to 60%, it means moderate heterogeneity; if the I 2 index is 50% to 90%, it means substantial heterogeneity; and if the I 2 index is 75% to 100%, it means considerable heterogeneity. 32 Funnel plot and Egger's test 33 were used to evaluate the existence of publication bias. A subgroup analysis was done to investigate the source of heterogeneity between studies, and sensitivity analysis with leave-one-out method was performed to assess robustness of the meta-analysis. The leave-one-out method was achieved by removing one study at a time and measuring the pooled estimate. Various system searches resulted in 2,464 studies discussing POI/POF. The duplicate of the studies was eliminated by use of reference manager; this yielded 1,621 studies. By use of selection criteria, abstract, and screening of the title generated in the differentiation of 192 studies, text papers were retrieved from this search and were analyzed by the selection criteria. During the full-text review, 178 articles were removed. Out of them, 98 studies were removed from the list of studies because of not fulfilling the inclusion criteria. An additional 80 studies were removed from the list of studies because POI/POF diagnosis failed to meet prescribed standards. Finally, 14 studies were included for this meta-analysis (Fig. 1 ). Out of these 14 studies, 14, 8, 10, and 9 studies were included for total T, DHEA-S, A, and SHBG, respectively. Study characteristics for each androgen are mentioned in Tables 1 - 4 . Search strategy, PRISMA flow chart for meta-analysis on total androgen concentrations. Study characteristics for meta-analysis on testosterone levels BMI, body mass index; NA, not available; POF, premature ovarian failure; T, testosterone concentration. Study characteristics for meta-analysis on DHEA-S levels BMI, body mass index; DHEA-S, dehydroepiandrosteronesulfate; NA, not available; POF, premature ovarian failure. Sudy characteristics for meta-analysis on androstenedione levels A, androstenedione; BMI, body mass index; NA, not available; POF, premature ovarian failure. Study characteristics for meta-analysis on SHBG levels BMI, body mass index; NA, not available; POF, premature ovarian failure; SHBG: sex hormone-binding globulin. The risk of bias assessment for studies selected for the comparison of androgen concentrations in POI/POF versus controls is reported in Table 5 [see Table, Supplemental Digital Content 1]. All the studies had clearly defined selection criteria for the cases, but for the controls, the selection is not described, although all the studies except one had defined the controls. All the studies except three fulfilled the comparability assessment. Most of the studies had fulfilled the exposure assessment. Altogether, seven studies had a score of more than 6 ∗ s, namely that of Kalantaridou et al, van der Stege et al, Gulhan et al, Doldi et al, Hartmann et al, Janse et al, and Ates et al. For sensitivity analysis of testosterone concentrations, the studies by Ates et al, Gulhan et al, and Doldi et al were not included because even though they scored 6 ∗ s on the scale, the selection methodology for cases and controls were not clearly mentioned. The studies by Kalantaridou et al, van der Stege et al, and Janse et al scored 7 ∗ s or more on the scale and were of the highest methodological quality. The study by Hartmann et al, while scoring only 6 ∗ s, had mentioned the selection methodology; it was just not clear if the participants were consecutively chosen, but because it mentioned how they were chosen as opposed to no description in the other three studies, it was included. Also the study by Doldi et al was included in the sensitivity analysis for androstenedione and DHEA-S because we wanted to compare a minimum of three studies to do the sensitivity analysis. The study by Doldi et al compared both androstenedione and DHEA-S, and scored more than 6 ∗ s on the Newcastle-Ottawa Scale. In the absence of other higher-quality studies for these two androgens (androstenedione, DHEA-S), we included this study in the sensitivity analysis. The study by Benetti-Pinto et al had no description of how cases and controls were selected. The study by Elias et al did not have age or BMI-matched cases and controls, nor did it have description of how cases and controls were selected. The study by Falsetti et al had no description of how cases and controls were selected, nor did it have description if both cases and controls used similar tests to check for exposure. Assessment of the risk of bias of involved studies using the Newcastle-Ottawa Scale for meta-analysis of observational studies The criteria for Quality Assessment using Newcastle-Ottawa Scale for meta-analysis. Selection: 1. Is the case definition adequate? (a) yes, with independent validation * ; (b) yes, for example, record linkage or based on self reports; (c) no description. 2. Representativeness of cases: (a) consecutive or obviously representative series of cases * ; (b) potential for selection biases or not stated. 3. Selection of controls: (a) community controls * ; (b) hospital controls; (c) no description. 4. Definition of controls: (a) no history of disease (endpoint) * ; (b) no description of source. Comparability: 1. Comparability of cases and controls on the basis of the design or analysis: (a) study controls for (select most important factor) * ; (b) study controls for any additional factor * (this criterion could be modified to indicate specific control for a second important factor). Exposure: 1. Ascertainment of exposure: (a) secure record (eg, surgical records) * ; (b) structured interview where blind to case/control status * ; (c) interview not blinded to case/control status; (d) written self-report or medical record only e) no description. 2. Same method of ascertainment for cases and controls: (a) yes * ; (b) no. 3. Nonresponse rate: (a) same rate for both groups * ; (b) nonrespondents described; (c) rate different and no designation. Note-1: The high and low score ranges are arbitrarily given. Note-2: 0 to 2 * s being poor quality, 3 to 5 * s as fair quality, 6 to 9 * s being higher quality. Note-3: Maximum score that can be obtained is 9 * s.

Women suffering from POI/POF have lower testosterone in comparison to the controls according to data retrieved from 14 studies (n = 1,656) (Table 1 ), SMD (IV, random, 95% CI) −0.73 [−0.99, −0.46] with P value < 0.05 (Fig. 2 ). There was significant heterogeneity observed in the meta-analysis (I 2  = 82%). With heterogeneity: Tau 2  = 0.19; χ 2  = 72.18, df = 13 ( P  < 0.00001); so, random-effects model was used. The funnel plot was evaluated to check for publication bias. A subgroup analysis was done to verify the source of heterogeneity. The sensitively analysis showed a consistency among the studies. Forest plot for meta-analysis of 14 comparative studies on total testosterone concentrations in women with POI/POF compared with fertile controls. Squares represent effect of each study, and the diamond represents the overall effect of the study at 95% confidence intervals. The values to the left of the solid line indicate SMD less than 1, decreased concentrations in POI/POF group. The values to the right of the solid line indicate SMD greater than 1, increased concentrations in POF/POI when compared to the control group. POF, primary ovarian failure; POI, premature ovarian insufficiency; SMD, standardized mean difference. The subgroups were divided according to the various assays used for measurement. For the analysis of total testosterone concentrations, there were four subgroups identified. The first subgroup consisted of four studies which directly applied radioimmunoassay (RIA) for the measurement of testosterone concentrations. The second subgroup consisted of five studies which used extraction or chromatography before applying RIA. The third subgroup consisted of four studies which utilized the chemiluminescent immunoassay. And the fourth subgroup consisted of only one study in which it was not clear which assay was used [see Fig., Supplemental Digital Content 2] The subgroup consisted of four studies (n = 280). This subgroup showed an SMD of −1.13 [−1.44, −0.82] test for overall effect: Z  = 7.23 ( P  < 0.00001) with substantial decrease in heterogeneity (Tau 2  = 0.03; χ 2  = 4.05, df = 3 [ P  = 0.26]; I 2  = 26%). The second subgroup consisted of studies which used extraction or chromatography before applying RIA. This consisted of five studies (n = 859). This subgroup showed an SMD of −0.66 [−0.99, −0.34], Z  = 4.00 ( P  < 0.0001). Heterogeneity was still substantial in this subgroup with Tau 2  = 0.09; χ 2  = 14.08, df = 4 ( P  = 0.007); I 2  = 72%. The source of this heterogeneity could be explained by the fact that three of the five studies included in this particular subgroup happened to be lower-quality studies and were not age or BMI-matched. We tried removing these three studies from this subgroup and saw that the heterogeneity had substantially reduced to about 4%. The subgroup consisted of studies measured using the chemiluminescent immunoassay, and it consisted of four studies (n = 455). This subgroup showed a SMD of −0.20 [−0.38, −0.01], Z  = 2.10 ( P  = 0.04). This subgroup showed no heterogeneity with Tau 2  = 0.00; χ 2  = 1.30, df = 3 ( P  = 0.73); I 2  = 0%. The subgroup consisted of only one study (n = 62). This subgroup showed an SMD of −1.88 [−2.48, −1.28], Z  = 6.10 ( P  < 0.00001). The POI/POF versus fertile controls subgroup showed a SMD (−0.73 [−0.99, −0.46]) with P value <0.05, whereas the POI/POF versus postmenopausal controls (n = 465) showed a SMD of −0.05 [−0.68, 0.58] test for overall effect: Z  = 0.15 ( P  = 0.88). Heterogeneity: Tau 2  = 0.27; χ 2  = 14.92, df = 2 ( P  = 0.0006); I 2  = 87%. The test for subgroup differences: χ 2  = 3.77, df = 1 ( P  = 0.05), I 2  = 73.5%, indicating substantial heterogeneity between these subgroups (see Fig., Supplemental Digital Content 3). Sensitivity analysis was first conducted using the leave-one-out method by systematically omitting one study at a time, and the results were robust. Then, four studies (n = 615) that scored 6 or greater on the Newcastle-Ottawa Scale, considered as the best-quality studies, were chosen out of the 14 studies. Seven studies had a score of more than 6, namely the studies by Kalantaridou et al, van der Stege et al, Gulhan et al, Doldi et al, Hartmann et al, Janse et al, and Ates et al. The studies by Ates et al, Gulhan et al, and Doldi et al were not included, because even though they scored 6 on the scale, the selection methodology for cases and controls were not clearly mentioned. The studies by Kalantaridou et al, van der Stege et al, and Janse et al scored 7 or more on the scale and were of the highest methodological quality. The study by Hartmann et al, while scoring only 6, mentioned the selection methodology; it was just that it was not clear if the participants were consecutively chosen in this study, but because it mentioned how they were chosen as opposed to no description in the other three studies, it was included. These studies used for the sensitivity analysis were the ones with the best methodological quality and considered having the least amount of bias, and all these studies were age and BMI-matched; this showed an SMD of −0.74 [−0.92, −0.55], Z  = 7.80 ( P  < 0.00001). With a drop in heterogeneity to 0%: Tau 2  = 0.00; χ 2  = 2.53, df = 3 ( P  = 0.47); I 2  = 0% (Fig. 3 ). Sensitivity analysis for total testosterone concentrations in women with POI/POF compared with fertile controls. Four studies (n = 615) that scored >6 on the Newcastle-Ottawa Scale, considered as the best quality studies, were chosen out of the 14 studies. POF, primary ovarian failure; POI, premature ovarian insufficiency. Women suffering from POI/POF have lower DHEA-S in comparison with the controls according to data retrieved from eight studies (n = 911) (Table 2 ) (SMD [IV, random, 95% CI] −0.65 [−0.92, −0.37]), with P value <0.05. There was significant heterogeneity observed in the meta-analysis ( I 2  = 71%), with heterogeneity: Tau 2  = 0.11; χ 2  = 24.44, df = 7 ( P  = 0.0010); so, random-effects model was used. The funnel plot was evaluated to check for publication bias. A subgroup analysis was done to verify the source of heterogeneity. The sensitively analysis showed a consistency among the studies (Fig. 4 ). Forest plot for meta-analysis of eight comparative studies on DHEA-S concentrations in women with POI/POF compared with fertile controls. Squares represent effect of each study, and the diamond represents the overall effect of the study at 95% confidence intervals. The values to the left of the solid line indicate SMD less than 1, decreased concentrations in POI/POF group. The values to the right of the solid line indicate SMD greater than 1, increased concentrations in POF/POI compared to the control group. DHEA-S, dehydroepiandrosterone sulfate; POF, primary ovarian failure; POI, premature ovarian insufficiency; SMD, standardized mean difference. The subgroups were divided according to the various assays used for measurement. For analysis of DHEA-S concentrations, there were two subgroups identified. The first subgroup consisted of six studies which directly applied RIA for the measurement of DHEA-S concentrations. The second subgroup was the studies in which the assays used were either not mentioned or needed more clarification. This consisted of two studies (see Fig., Supplemental Digital Content 4). The subgroup consisted of six studies (n = 398). This subgroup showed a SMD of −0.60 [−0.90, −0.29], Z  = 3.79 ( P  = 0.0002) with a decrease in heterogeneity: Tau 2  = 0.08; χ 2  = 10.94, df = 5 ( P  = 0.05); I 2  = 54%. The source of this heterogeneity could be explained by the fact that three of the six studies included in this particular subgroup happened to be studies with lower methodological quality. The study by Benetti-Pinto et al had no description of how cases and controls were selected. The study by Elias et al did not have age or BMI-matched cases and controls, nor did it have a description of how cases and controls were selected. The study by Falsetti et al had no description of how cases and controls were selected, nor did it have a description if both cases and controls used similar tests to check for exposure. We tried removing these studies from this subgroup and saw that the heterogeneity had substantially reduced to about 0%. The subgroup consisted of only two studies (n = 513). This subgroup showed a SMD of −0.74 [−1.35, −0.14], Z  = 2.39 ( P  = 0.02), with very high heterogeneity: Tau 2  = 0.17; χ 2  = 10.32, df = 1 ( P  = 0.001); I 2  = 90%. Both studies included in this subgroup were of lower methodological quality, and both were not age and BMI-matched in selecting cases and controls. The POI/POF versus fertile controls subgroup showed a SMD (IV, random, 95% CI) −0.65 [−0.92, −0.37], with P value <0.05. POI/POF versus postmenopausal controls (n = 465) showed a SMD 1.12 [0.10, 2.14], Z  = 2.15 ( P  = 0.03), heterogeneity: Tau 2  = 0.77; χ 2  = 35.54, df = 2 ( P  < 0.00001); I 2  = 94%, indicating that the DHEA-S levels are much reduced in POI/POF cases, but still show a higher concentration than DHEA-S in postmenopausal women. Test for subgroup differences showed χ 2  = 10.46, df = 1 ( P  = 0.001), I 2  = 90.4%, indicating substantial heterogeneity between these subgroups (Fig. 5 ). Subgroup analysis for DHEA-S concentrations in women with POI/POF compared with postmenopausal controls (n = 465), indicating that the DHEA-S levels are much reduced in POI/POF controls, but still show a higher concentration than DHEA-S in postmenopausal women. DHEA-S, dehydroepiandrosterone sulfate; POF, primary ovarian failure; POI, premature ovarian insufficiency. Sensitivity analysis was first conducted using the leave-one-out method by systematically omitting one study at a time, and the results were robust. Three studies (n = 210) that scored 6 or greater on the Newcastle-Ottawa Scale, considered as the best-quality studies, were chosen out of the eight studies, namely the studies by Doldi et al, Hartmann et al, and van der Stege et al. The study by van der Stege et al scored 7 ∗ s on the Newcastle-Ottawa Scale and was of the highest methodological quality. The studies by Doldi et al and Hartmann et al scored 6 ∗ s on the Newcastle-Ottawa Scale, even though the selection methodology for cases and controls was not clearly mentioned. The study by Doldi et al, was included here as we wanted to use at least three studies to conduct the sensitivity analysis and these three were the highest-quality studies comparing DHEA-S. These were the studies with the best methodological quality and considered having the least amount of bias; in all these studies cases, controls were age, and BMI-matched. This showed a SMD of −0.46 [−0.74, −0.17], Z  = 3.16 ( P  = 0.002), with a drop in heterogeneity to 0%: heterogeneity: Tau 2  = 0.00; χ 2  = 1.43, df = 2 ( P  = 0.49); I 2  = 0% (Fig. 6 ). Sensitivity analysis for DHEA-S concentrations in women with POI/POF compared with fertile controls. Three studies (n = 210) that scored >6 on the Newcastle-Ottawa Scale, considered as the best quality studies, were chosen out of the eight studies. DHEA-S, dehydroepiandrosterone sulfate; POF, primary ovarian failure; POI, premature ovarian insufficiency. Women suffering from POF have lower androstenedione in comparison with the controls according to data retrieved from 10 studies (n = 1,226) (Table 3 ), SMD (IV, random, 95% CI) −1.09 [−1.71, −0.48], with P value <0.05. There was significant heterogeneity observed in the meta-analysis ( I 2  = 95%). With heterogeneity: Tau 2  = 0.92; χ 2  = 180.64, df = 9 ( P  < 0.00001); so, a random-effects model was used. The funnel plot was evaluated to check for publication bias. A subgroup analysis was done to verify the source of heterogeneity. The sensitively analysis showed a consistency among the studies (Fig. 7 ). Forest plot for meta-analysis of 10 comparative studies androstenedione concentrations in women with POI/POF compared with fertile controls. The values to the left of the solid line indicate SMD less than 1, decreased concentrations in POI/POF group. The values to the right of the solid line indicate SMD greater than 1, increased concentrations in POF/POI when compared to the control group. POF, primary ovarian failure; POI, premature ovarian insufficiency; SMD, standardized mean difference. The subgroups were divided according to the various assays used for measurement. The first subgroup consisted of six studies that directly applied RIA for the measurement of DHEA-S concentrations. The second subgroup was the studies in which the assays used were either not mentioned or needed more clarification. This consisted of four studies (Fig., Supplemental Digital Content5). The subgroup consisted of six studies (n = 398). This subgroup showed a SMD of −1.10 [−1.61, −0.58], Z  = 4.17 ( P  < 0.0001) with heterogeneity: Tau 2  = 0.34; χ 2  = 27.45, df = 5 ( P  < 0.0001); I 2  = 82%. The source of this heterogeneity could be explained by the fact that three of the six studies included in this particular subgroup happened to be studies with lower methodological quality. The study by Benetti-Pinto et al had no description of how cases and controls were selected; the study by Elias et al did not have age or BMI-matched cases and controls, nor did it have a description of how cases and controls were selected; and the study by Falsetti et al had no description of how cases and controls were selected, nor did it have a description whether both cases and controls used similar tests to check for exposure. We tried removing these studies from this subgroup and saw that the heterogeneity had substantially reduced to about 2%. The subgroup consisted of only four studies (n = 828). This subgroup showed a SMD of −1.08 [−2.30, 0.15], Z  = 1.73 ( P  = 0.08) with very high heterogeneity: Tau 2  = 1.52; χ 2  = 150.25, df = 3 ( P  < 0.00001); I 2  = 98%. Two studies included in this subgroup were of lower methodological quality, and both were not age and BMI-matched in selecting cases and controls, namely the studies by Szlendak-Sauer et al and Daan et al; we tried removing these studies from this subgroup and saw that the heterogeneity had substantially reduced to about 5%. The POI/POF versus fertile controls subgroup showed a SMD (IV, random, 95% CI) −1.09 [−1.71, −0.48] with P value <0.05. POI/POF versus postmenopausal controls (n = 466) showed a SMD 0.27 [−0.13, 0.67], Z  = 1.34 ( P  = 0.18), heterogeneity: Tau 2  = 0.08; χ 2  = 5.97, df = 2 ( P  = 0.05); I 2  = 66%. Test for subgroup differences showed χ 2  = 13.41, df = 1 ( P  = 0.0003), I 2  = 92.5% (Fig., Supplemental Digital Content 6). Sensitivity analysis was first conducted using the leave-one-out method by systematically omitting one study at a time, and the results were robust. Then four studies (n = 463) that scored 6 or greater on the Newcastle-Ottawa Scale, considered as the best-quality studies, were chosen out of the 10 studies, namely the studies by Doldi et al, Hartmann et al, van der Stege et al, and Janse et al. The studies by van der Stege et al and Janse et al scored 7 ∗ s on the Newcastle-Ottawa Scale and were of the highest methodological quality. The studies by Doldi et al and Hartmann et al scored 6 ∗ s on the Newcastle-Ottawa Scale. This showed a SMD of −0.82 [−1.16, −0.47], Z  = 4.66 ( P  < 0.00001), with a drop in heterogeneity to 55%: heterogeneity: Tau 2  = 0.07; χ 2  = 6.65, df = 3 ( P  = 0.08); I 2  = 55% (Fig. 8 ). Sensitivity analysis for androstenedione concentrations in women with POI/POF compared with fertile controls. Three studies (n = 463) that scored >6 on the Newcastle-Ottawa Scale, considered as the best quality studies, were chosen out of the 10 studies. POF, primary ovarian failure; POI, premature ovarian insufficiency. Women suffering from POI/POF did not seem to have a statistically significant difference to fertile controls with regards to SHBG levels according to data retrieved from nine studies (n = 1,613) (Table 4 ); SMD (IV, random, 95% CI) −0.13 [−0.35, 0.09] with P value >0.05 (not significant) (see Fig., Supplemental Digital Content 7).

In conclusion, this meta-analysis demonstrated that women with POI/POF are at risk for decreased concentrations of total testosterone, DHEA-S, and androstenedione. SHBG concentrations did not show any statistical significance between POI/POF and controls. DHEA-S levels were more reduced in postmenopausal controls when compared with POI/POF cases. Testosterone and androstenedione concentrations did not show any statistical significance when POI/POF cases were compared with postmenopausal controls.

In this systematic review and meta-analysis, we analyzed the various androgen concentrations including both the principal ovarian and adrenal androgens such as serum concentrations of total T, A, DHEA-S, and SHBG, and compared them with fertile controls and postmenopausal controls. The pooled testosterone concentrations were found to be lower in women with POI/POF compared with fertile controls; SMD (IV, random, 95% CI) −0.73 [−0.99, −0.46] with P value 0.05. The pooled DHEA-S concentrations were found to be lower in women with POI/POF compared with fertile controls; SMD (IV, random, 95% CI) −0.65 [−0.92, −0.37] with P value <0.05. DHEA-S concentrations in POI/POF versus postmenopausal controls were statistically significant with a SMD 1.12 [0.10, 2.14], P value < 0.05, indicating that the DHEA-S levels are reduced in POI/POF, but still show a higher concentration than DHEA-S in postmenopausal controls. Women in the postmenopausal group are older than POI/POF women. It was only for DHEA-S that a significant decrease was noted in the postmenopausal group when comparing testosterone, DHEA-S, and androstenedione between POI/POF women and the postmenopausal group. The reasons for a significant decrease of DHEA-S seen in the postmenopausal group could be due to aging. The study by Labrie et al 44 has summarized that DHEA secretion will have already decreased by an average of 60% at time of menopause and will continue to decrease thereafter. We propose that this difference in decreasing pattern in serum DHEA-S with increasing age between POI/POF women and the postmenopausal group is mainly due to a higher compensatory mechanism in the relatively younger POI/POF women. In women of all ages, the inactive sex steroid precursor DHEA is mainly of adrenal origin. The study by Labrie et al 45 , 46 reported that the ovary is responsible for production of only 20% of circulating DHEA. The inactive DHEA will be transformed into the appropriate minute intracellular amounts of androgens to exert their physiological function. There is no biologically significant release into the serum for these intracellular androgens. This intracrine mechanism of androgen formation from DHEA could be responsible for why only for DHEA-S that a significant decrease was noted in the postmenopausal group, and why no significant difference was seen when comparing testosterone and androstenedione between POI/POF women and the postmenopausal group. The pooled A concentrations were found to be lower in women with POI/POF compared with fertile controls; SMD (IV, random, 95% CI) −1.09 [−1.71, −0.48] with P value 0.05. Women suffering from POI/POF did not seem to have a statistically significant difference compared to fertile controls with regards to SHBG levels according to data retrieved; SMD (IV, random, 95% CI) −0.13 [−0.35, 0.09], with P value >0.05. There was significant heterogeneity between studies. Each of these androgens decreases with age. Most of the studies chosen had age-matched cases and controls, and only three studies did not match for age or BMI. So, we conducted a sensitivity analysis using only the best quality studies for each androgen assessed. In these studies, cases and controls were age and BMI-matched, and results showed substantial decrease in heterogeneity. Age and BMI were important factors in quality assessment for these studies involved. Some studies did not adjust for these factors. This could be one of the reasons for the heterogeneity between these studies. Another reason for heterogeneity between studies could be due to the differences in selecting controls and methodologies used for recruitment. Finally, the most important factor contributing towards the heterogeneity between studies could be explained by the fact that the studies used different assays for the measurement of androgen concentrations. This was apparently evident in the subgroup analysis using different assays (see Fig., Supplemental Digital Content3, 4, and 5). There was a substantial decrease in heterogeneity within most subgroups, which conducted analysis using the same assay. Few of the subgroups undertaken using the same assay showed reduced but still a significant level of heterogeneity. This could be explained by the fact that even though they were using the same kind of assay for measuring the concentrations, the intra and interassay coefficients used were widely variable. And, also, the source of this heterogeneity could be explained by the fact that few of these studies included happened to be lower-quality studies and were not age or BMI-matched or had no description of how cases and controls were chosen or if the same tests were used to ascertain exposure in cases and controls. The study by Benetti-Pinto et al 21 had no description of how cases and controls were selected. The study by Elias et al 7 did not have information on age or BMI for cases and controls, nor did it have a description of how cases and controls were selected. The study by Falsetti et al 38 had no description of how cases and controls were selected, nor did it have description whether both cases and controls used similar tests to check for exposure. The studies by Szlendak-Sauer et al 39 and Daan et al 43 were not age and BMI-matched in selecting cases and controls. When these studies with lower methodological quality were removed from the respective subgroups, the heterogeneity substantially reduced. So we conducted a sensitivity analysis using only studies with the best methodological quality, in which the most important confounding factors were age and BMI, and we observed significant results and substantial reduction in heterogeneity in all of the androgens assessed. Taking all these factors contributing to heterogeneity into account, SMD was used to estimate the pooled concentrations and a random-effects model was used. SMD was preferred over a mean difference, because the methodologies used for estimation of the concentrations widely varied and this variation is already acknowledged by using SMD for estimation. Also, the sensitivity analysis using the best quality studies proved that the results obtained were robust. Therefore, we may interpret the results of this meta-analysis as being statistically significant. The major androgens found in women in descending order of their serum concentrations include the following: DHEA-S, A, and T. 47 Of these androgens, T is the most potent. 48 It is secreted by the adrenal zonafasciculata (25%) and the ovarian stroma (25%), with the remaining 50% being produced from circulating A. 49 DHEA is a secretory product of the adrenal zonareticularis (50%) and the ovarian theca (20%), and 30% is derived from circulating DHEA-S, catalyzed by steroid sulphatase. 49 Androstenedione is secreted by the adrenal zonafasciculata (50%) and the ovarian stroma (50%). 49 We have identified decreased concentrations of these androgens, namely DHEA, which is produced mostly by the adrenal glands; however, it is still not clear whether this reduction is solely due to the ovarian component or if the adrenal component is also affected in idiopathic POI/POF. The consequences of androgen deficiency have been discussed in very few studies. Androgens are thought to be one of the basic prerequisites for a healthy woman. 7 - 14 A lack of androgens may lead to symptoms of sexual dysfunction, such as decreased libido, loss of sexual responsiveness, or decreased sexual arousal. 7 - 14 Other clinical symptoms are a diminished sense of well-being, dysphoric mood, cognitive dysfunction, and persistent, unexplained loss of energy. 7 - 14 The role of androgens in maintaining a woman's health has been receiving increasing attention, but still, there is considerable controversy relating to the role of androgens in women, and its clinical significance. 4 , 13 - 21 Testosterone at supraphysiological, but not at physiological, levels enhance the effectiveness of low-dose estrogen therapies at increasing women's sexual desire; however, the mechanism by which supraphysiological testosterone increases women's sexual desire in combination with an estrogen remains unknown. 12 Clinical evidence suggests that testosterone has anxiolytic and antidepressant benefits, with the potential to promote improved mood and mental health in women. 50 However, the neurobiological mechanisms underlying the protective effects of testosterone in males and females remain poorly understood. Selective androgens appear capable of improving early stages of folliculogenesis. Androgens, like T, appear effective in improving functional ovarian reserve in women with diminished ovarian reserve. 51 A study found that a low-testosterone status is a potentially important step in the development of POI/POF in women with endometriosis. 52 In women with diminished ovarian reserve, the basal T level presented a positive association with pregnancy outcome in in vitro fertilization. 52 This meta-analysis has a few limitations. Very few studies were available for the topic of interest that would meet the eligibility criteria. Inclusion criteria were somewhat limited in databases assessed, some studies with lower methodological quality were included; this lead to significant heterogeneity between the studies. We have rectified these issues of including lower methodological quality studies by conducting a sensitivity analysis with higher methodological quality studies only. Each of these androgens decrease with age, most of the studies chosen had age-matched cases and controls. Only three studies did not match for age or BMI. Again, we conducted a sensitivity analysis using the best quality studies for each androgen assessed, in these studies cases and controls were age and BMI-matched; and the results seemed to be consistent and robust. Further, it is very difficult to measure and compare these sex steroids at lower ranges with accuracy using various assays. This was another challenge we faced while conducting this meta-analysis. Based on this limitation in assessment, we had to standardize the units of measurement and use a SMD and random-effects model to account for these differences. Currently, liquid chromatography-tandem mass spectrometry (LC-MS/MS) is a widely accepted and accurate assay for measuring low levels of steroids, such as androgens. Further comparative studies which employ LC-MS /MS, or similar assays, are needed to make sure the results obtained using various other assays can be depended upon. Our meta-analysis has found that there is significant androgen deficiency in women with POI/POF. A significant reduction in androgen levels could lead to various complaints. Using androgen replacement therapy as an adjunct to pre-existing estrogen-based replacement therapy could be advantageous in relieving health complaints and promoting a more qualitative life in these women. Also, we suggest that, because the assays used for measurement of various androgens are widely variable, a standardized method for measurement needs to be developed for accurate assessment of these indices.