High prevalence of moderate and severe depressive and anxiety symptoms in polycystic ovary syndrome: a systematic review and meta-analysis

Abstract

STUDY QUESTION

Do women with polycystic ovary syndrome (PCOS) have an increased prevalence of moderate and severe depressive and anxiety symptoms compared with control women, and do these symptoms correlate with age, BMI, testosterone, hirsutism or insulin resistance (IR)?

SUMMARY ANSWER

Women with PCOS have significantly increased odds of moderate and severe depressive and anxiety symptoms, independent of obesity, and the symptoms are weakly associated with age, BMI, elevated testosterone, hirsutism and IR.

WHAT IS KNOWN ALREADY

Previous studies have reported that women with PCOS have an increased prevalence of mild depressive and anxiety symptoms or an increase in mean depression and anxiety scores, although these scores are usually within the normal range. Thus, it is therefore not clear whether these findings are clinically significant. The prevalence of moderate and severe depressive and anxiety symptoms, which require follow-up and would benefit from treatment, is not known in this population.

STUDY DESIGN, SIZE, DURATION

A comprehensive systematic review (SR) was performed up to January 2016 and included 30 cross-sectional studies, representing 3050 subjects with PCOS and 3858 controls, from 10 different countries. The meta-analysis (MA) on depressive symptoms included 18 studies and the MA on anxiety symptoms included 9 studies. A separate SR identified 15 studies for the meta-regression examining the associations with PCOS-related symptoms or comorbidities.

PARTICIPANTS/MATERIALS, SETTING, METHODS

All studies included adult women with PCOS, defined by the National Institutes of Health or Rotterdam criteria, and a control group without PCOS. Ovid, Embase, PsychInfo and Cochrane were searched up to January 2016. Included studies used a validated screening tool to compare the prevalence or mean scores of depressive and/or anxiety symptoms. Random effects MA was used to estimate the pooled odds ratio (OR) of depressive and anxiety symptoms. Sensitivity analyses of methodological characteristics and a meta-regression of the pooled standardized mean difference (SMD) to evaluate PCOS-related clinical and laboratory associations were performed.

MAIN RESULTS AND THE ROLE OF CHANCE

Women with PCOS had increased odds of any depressive symptoms (OR: 3.78; 95% CI: 3.03–4.72; 18 studies) and of moderate/severe depressive symptoms (OR: 4.18; 95% CI: 2.68–6.52; 11 studies). Women with PCOS had increased odds of any anxiety symptoms (OR: 5.62; 95% CI: 3.22–9.80, nine studies) and of moderate/severe anxiety symptoms (OR: 6.55; 95% CI: 2.87, 14.93; five studies). When subjects were matched on BMI, women with PCOS still had higher odds of both depressive (OR: 3.25; 95% CI 1.73–6.09; four studies) and anxiety symptoms (OR: 6.30, 95% CI: 1.88–21.09; three studies). There was no substantial heterogeneity among studies in the overall MA on depressive symptoms (I² = 22.4%, P = 0.19), but there was significant heterogeneity among studies in the analysis on anxiety symptoms (I² = 59.6%, P= 0.01). In the meta-regression evaluating pooled SMDs between groups, women with PCOS and concurrent depression had higher mean values of age, BMI, hirsutism score and IR, while women with PCOS and concurrent anxiety had higher mean values of BMI, hirsutism score and free testosterone (P < 0.05 for all comparisons).

LIMITATIONS, REASON FOR CAUTION

All studies were cross-sectional, thus we can only hypothesize that the diagnosis of PCOS precedes the diagnosis of depression and anxiety. There were large variations in methodological characteristics especially in the studies screening for anxiety; however, they only partly explained effect size variation.

WIDER IMPLICATIONS OF THE FINDINGS

This evidence-synthesis analysis shows that PCOS diagnosis is associated with an increased risk of moderate and severe depressive and anxiety symptoms and suggests that providers should consider screening women with PCOS for both depression and anxiety. Although age, obesity, hyperandrogenism and IR do not explain the entire association, well-designed studies are needed to assess the impact of treatment of these factors on depressive and anxiety symptoms in women with PCOS.

STUDY FUNDING/COMPETING INTEREST(S)

No funding was used for this study. There are no conflicts of interest.

REGISTRATION NUMBER

N/A.

PCOS, polycystic ovary syndrome, depression, anxiety, meta-analysis

Introduction

Polycystic ovary syndrome (PCOS) is an endocrine disorder affecting between 6 and 10% of reproductive age women (Bozdag et al., 2016). Characteristic clinical features include irregular menses, clinical or laboratory evidence of elevated androgens and polycystic-appearing ovaries on ultrasound (Rotterdam Consensus, 2004). Women with PCOS have an increased risk of concurrent cardiovascular risk factors and comorbidities such as insulin resistance (IR), dyslipidemia and diabetes (Behboudi-Gandevani et al., 2016; Zhao et al., 2016); however, there is also increasing evidence that women with PCOS have an increased risk of mood and anxiety disorders (Barry et al., 2011b; Dokras et al., 2011, 2012; Veltman-Verhulst et al., 2012). The etiology of this association has not been well established, but potential proposed mechanisms include obesity, IR, elevated androgens, abnormalities in the hypothalamic pituitary axis and increased inflammatory markers (Farrell and Antoni, 2010).

Previously published meta-analyses in women with PCOS have shown an increase in the prevalence of both depressive and anxiety symptoms in women with PCOS compared with controls; however, these meta-analyses collectively included 10 studies evaluating the prevalence of depressive symptoms (Dokras et al., 2011) and only 4 studies evaluating the prevalence of anxiety symptoms (Dokras et al., 2012). In addition, these results were based on a single cutoff value on a screening test and therefore do not allow identification of moderate and severe depressive or anxiety symptoms, which require higher cutoffs and subsequent evaluation by a clinical psychologist or psychiatrist. Two other meta-analyses have compared mean depression and anxiety scores between PCOS subjects and controls (Barry et al., 2011b; Veltman-Verhults et al., 2012). Although they reported higher mean scores in women with PCOS, most of the scores were in the normal or mild range. Data on moderate and severe symptoms for both depression and anxiety are therefore lacking in this population. One study performed sensitivity analyses (Veltman-Verhulst et al., 2012); however, they did not examine the use of medications or the impact of location of recruitment of controls compared with PCOS subjects, a significant source of selection bias in observational studies.

The underlying factors that may predispose women with PCOS to an increased risk of depression or anxiety are not clear. Recent publications can provide a larger sample size to examine whether any PCOS-related symptoms or comorbidities, such as obesity, elevated testosterone, hirsutism or IR, are associated with depressive or anxiety symptoms in these women. In addition, few studies have examined the impact of PCOS-related treatments, such as oral contraceptive pills or metformin, on the prevalence of depressive and anxiety symptoms (Cinar et al., 2012; Sirmans et al., 2012; Dokras et al., 2016). Thus, we performed a meta-analysis (MA) to examine the prevalence of moderate and severe depressive and anxiety symptoms in women with PCOS, with a priori sensitivity analyses. We also performed a separate systematic review (SR) and meta-regression to examine the association between depressive or anxiety symptoms and age, BMI, clinical/biochemical hyperandrogenism and IR. This evidence synthesis will allow implementation of appropriate guidelines for routine screening followed by early intervention, and thereby have a favorable impact on the health and financial burden associated with depression and anxiety.

Materials and Methods

Literature search

This review includes the results of four SRs and meta-analyses (Dokras et al., 2011, 2012; Barry et al., 2011b; Veltman-Verhulst et al., 2012), with literature searches performed up to November 2011. We searched all articles included in these meta-analyses and evaluated them for inclusion in our study. We also performed a new search in a similar manner to these studies from January 2011 to the third week of January 2016. Computerized literature searches for PCOS (MeSH: polycystic ovary syndrome, PCOS, Stein Leventhal Syndrome) and depression/anxiety (MeSH: anxiety, anxiety disorder, depression, depressive disorder, mental disorder, mood disorder) (Supplementary Table SI) were conducted in the following electronic databases: Ovid (MEDLINE and MEDLINE in-process and other non-indexed citations), EMBASE, PsycINFO and the Cochrane Reviews database. The Cochrane Reviews database did not produce any published reviews of PCOS and anxiety or depression. Retrieval was limited to human studies. We did not apply any restrictions on language. We excluded secondary publications of the same patient populations. Abstracts were considered if they included the necessary information.

Inclusion and exclusion criteria

The articles were searched by L.C and A.D. for inclusion and exclusion criteria.

Criteria for the SR

In our primary SR, we included all articles with cross-sectional comparisons of adult women with PCOS and controls who were screened for depression and/or anxiety using a validated screening tool. PCOS was diagnosed using either the National Institutes of Health (NIH) (Zawadski and Dunaif, 1992) or Rotterdam (Rotterdam Consensus, 2004) criteria. Controls were women who did not have a diagnosis of PCOS, either excluded as part of study protocol or subject self-report. No restrictions were placed on the location of PCOS or control recruitment (community versus clinic). Studies were included if they used a validated screening tool for depressive or anxiety symptoms (Table I). To reduce heterogeneity, studies that specified recruitment of only adolescents were excluded. In addition, as obesity and age are known to be associated with depression and anxiety (Kessler et al., 2003; Lieb et al., 2005; Simon et al., 2008), studies that did not report the mean BMI and age of both groups were excluded.

Table I

Demographics of studies included in systematic review.

References, country	PCOS criteria/confirmation/location of subject recruitment	Variables on which PCOS subjects and controls were matched	Reported demographic characteristics and significant differences between groups	Number of subjects		Age (years, mean ± SD)		BMI (kg/m2, mean ± SD)		Depression screening tool	Anxiety screening tool	Current meds	Included in meta-analysis; reason for exclusion
				PCOS	Control	PCOS	Control	PCOS	Control
Acmaz et al. (2013), Turkey#	Rotterdam, both: confirmed PCOS: PCOS clinic; controls: community volunteers	None	Low income, employed (C)*, university education (C)*, married, intent to have a child (P)**	86	47	25.6 ± 5.5	27.8 ± 6.5	27.5 ± 5.2**	23.3 ± 3.1	BDI Min ≤11 Mild: 12–≤19 Moderate: >19	BAI None: ≤22 Moderate: >22	Excluded those on hormonal meds	Yes
Adali et al. (2008), Turkey	Rotterdam, both: confirmed Both: gyn clinic, PCOS subjects were consecutive	Age; subjects with infertility excluded	Employment, education, married	42	42	23.5 ± 3.1	24.5 ± 2.5	28.4 ± 4.3**	24.1 ± 4.1	BDI: ≥11	NA	Subjects stopped hormonal meds and Metformin for 6 months	Yes
Altinkaya et al. (2014), Turkey	Rotterdam, both: confirmed Not stated	Age	None reported	50	50	23.7 ± 5.3	24.4 ± 5.7	25.1 ± 3.7**	22.1 ± 2.7	BDI Mild: 10–16 Moderate: 17–29	BAI Mild: 8–15 Moderate: 16–25 Severe: 26–63	Subjects stopped meds × 3 months	Yes
Annagür et al. (2014), Turkey	Rotterdam, PCOS: confirmed; controlsb PCOS: gyn clinic, consecutive controls: hospital employees	Age	Low economic status, employed (C)** education, married (C)*	83	64	22.3 ± 1.8	22.9 ± 2.1	23.9 ± 4.7**	22.0 ± 2.4	BDI	NA	Subjects stopped hormonal or psych meds × 4 weeks	No, prevalence not reported
Arshad et al. (2012), Iran	Rotterdam, PCOS: confirmed; controlsa PCOS: endocrine clinic controls: hospital personnel and university students and staff	None	Unknown	71	134	25.5 ± 5.6**	29.6 ± 6.7	27.2 ± 5.7**	24.9 ± 4.8	BDI	NA	No med exclusion	Yes
Asik et al. (2015), Turkey#	Rotterdam, PCOS: confirmed; controlsa PCOS: endocrine clinic, new diagnosis controls: internal medicine clinic	Age	Infertility (P)*	71	50	22 (IQR: 18–32)	24 (IQR: 18–37)	27 (IQR:17–47)**	21 (IQR:16–37)	HADS: >10	HADS: >10	No med exclusion	Yes
Barnard et al. (2007), UK#	Rotterdam, PCOS: self-report of prior PCOS diagnosis by a medical practitioner; controlsc Both: community: internet, email, posters, snowball technique	None	None reported	439	935	30.1 ± 6.1	28.9 ± 6.9	31.5 ± 9.8*	24.2 ± 4.8	Zung self-rating depression scale Mild: 50–59, Moderate: 60–69 Severe: >70	NA	Excluded if taking hormonal meds	No, self report of PCOS diagnosis
Barry et al. (2011a), UK	Rotterdam, both: confirmed Both: gyn clinic	Infertile controls	Lowest job classification, race, infertility	76	49	28.8 ± 4.8**	35.1  ± 4.4	27.9 ± 7.4**	24.7  ± 7.1	HADS Mild: 8–10 Moderate: 11–14 Clinical: 15–21	HADS Mild: 8–10 Moderate: 11–14 Clinical: 15–21	Excluded if taking hormonal meds	No, prevalence not reported
Battaglia et al. (2008), Italy	NIH, PCOS: confirmedt; controlsa Not stated	Age, BMI, Caucasian	Race, stable heterosexual relationship	25	18	27.7  ± 5.4	30.7  ± 3.9	21.6  ± 2.4	21.2  ± 2.0	BDI Mild: 10–18 Moderate: 19–29	NA	Excluded if taking hormonal meds	Yes
Benson et al. (2008), Germany	NIH, both: confirmed PCOS: endocrine clinics controls: community advertisement	None	None reported	57	28	28.9  ± 0.7	29.9 ± 1.2	29.6 ± 1.0	23.6  ± 0.7	BDI: ≥11	NA	No med exclusion	Yes
Bhattacharya and Jha (2010), India	Rotterdam, PCOS: confirmed; controlsa Both: gyn clinic. PCOS: PCOS related complaints; controls: routine care	None	Employment, married	117	84	21.6  ± 3.4	22.4  ± 2	26.2 ± 4.1*	24.5  ± 3.5	PHQ-9	NA	Subjects stopped hormonal meds x 3 months	Yes
Cinar et al. (2011), Turkey	Rotterdam, both: confirmed PCOS: endocrine clinic; controls: community	BMI	None reported	226	85	23.2 ± 5.2*	24.4 ± 4.0	24.7 ± 5.7	23.4 ± 5.4	BDI Moderate: ≥17	HADS Moderate: ≥11	Subjects stopped hormonal meds × 3 months	Yes
Davari-Tanha et al. (2014), Iran	Rotterdam, both: confirmed Both: infertility clinic	Age, all infertile	Infertility	110	110	29.6 ± 5.6	29.0 ± 3.4	30.5 ± 4.1	29.0	Evaluation by Psychiatrist	NA	Excluded if taking hormonal meds	No; control depression rates not representative of general population
Deeks et al. (2011), Australia#	Rotterdam, PCOS: self-report of prior diagnosis by a medical practitioner, confirmed by detailed phone interview; controlsc Both: community and internet advertising	None	Income, employed (C)*, education, infertility (P)**	177	109	32.8 ± 7.8**	41.9 ± 15.4	31.5 ± 7.9**	24.5 ± 5.4	HADS Mild: 8–10 Moderate: 11–14 Severe: 15–21	HADS Mild: 8–10 Moderate: 11–14 Severe: 15–21	No med exclusion	No; self-report of PCOS diagnosis
Ercan et al. (2013), Turkey#	Rotterdam, confirmed in PCOS; controlsd Both: gyn clinic; controls presented for routine check-up	Age, all married with active sexual life	Low income, race, employed, education, married	32	32	27.4  ±  3.3	27.0 ± 3.2	25.5 ± 2.0	24.4 ±  3.6	BDI: Mild: 12–18 Moderate >19	NA	Subjects stopped hormonal meds x 3 months	Yes
Elsenbruch et al. (2003), Germany	NIH, confirmed in both PCOS: endocrine clinics; controls: health screening program for employees	Age	Employed, married, unfulfilled wish to conceive (P)**	50	50	28.4  ± 5.0	29.9 ± 5.7	30.1 ± 9.8 **	24.4 ± 5.3	SCL-90	SCL-90	No med exclusion	No; prevalence not reported
Harmanci et al. (2013), Turkey	Rotterdam, confirmed in both Both: gyn clinic: PCOS diagnosis made in the last month	None	Education	42	42	22.5 ±  3.6	22.9  ± 3.7	22.8 ± 5.5	22.5  ± 3.7	BSI	BSI	Subjects stopped hormonal meds x 6 months	No, prevalence not reported
Himelein and Thatcher (2006), USA	Rotterdam, PCOS: considered confirmed as only PCOS subjects seen in author's clinic were included; controlsc PCOS: Reproductive endocrinology clinic: controls: community advertising (N = 60), infertility clinic (N = 40)	None	Low income, race, education, married	40	100	32.1 ± 5.5(mean age for entire group)		34 ± 9*	23 ± 4	Short form BDI Moderate: ≥12	NA	No med exclusion	Yes
Hollinrake et al. (2007), USA	Rotterdam, PCOS: confirmed; controlsa PCOS: Reproductive endocrinology clinic controls; gyn clinic; annual exam	None	Race, employed, education, married, currently attempting pregnancy (P)**	103	103	29.8  ± 6.2	30.7  ± 8.5	34.9  ± 8.5*	25.4  ± 4.7	PRIME-MD PHQ	PRIME-MD PHQ	No med exclusion	Yes
Jedel et al. (2010), Sweden	Rotterdam, both: confirmed Both: community advertising	Age, weight	Race, employed, married or cohabitating, have biological children	30	30	28.0 (IQR: 21.0–37.0)	27.8 (IQR: 22.0–35.0)	24.8 (IQR: 18.2–40.3)	24.7 (IQR: 19.3–41.6)	Montgomery Asberg Depression Rating Scale (MADRS-S) – cutoff ≥11	BSA-S: sum total > 11	Subjects stopped hormonal meds x 3 months	Yes
Mansson et al. (2008), Sweden	Rotterdam, PCOS: confirmed; controlsd PCOS: gyn and medicine clinics Controls: population registry	Age (DOB)	Employed or student (C)*, education, married or cohabitating, have children	49	49	35.9  ± 10.4	35.9 ± 10.4	29 ± 7**	23.5 ± 3	MINI NPI by Psychiatrist: Any Major Depressive Episodes	MINI NPI by Psychiatrist: Generalized Anxiety	No med exclusion No difference between hormone use in groups	Yes
Moran et al. (2010), Australia#	Rotterdam, PCOS: self-report of prior PCOS diagnosis by a medical practitioner confirmed by detailed phone interview; controlsc Both: community advertising	None	None reported	24	22	22.4  ± 0.4	22.0 ± 0.5	29.2  ± 1.5**	22.1  ± 0.8	HADS Mild: 8–10 Moderate: 11–14 Severe: 15–21	HADS Mild: 8–10 Moderate: 11–14 Severe: 15–21	No med exclusion	No; self-report of PCOS diagnosis
Moran et al. (2012), Australia#	Rotterdam, both: confirmed Both: community advertising	None	None reported	52	24	32.6 ± 1.3**	36.4  ± 1.7	34.4  ± 2.3**	28.7  ± 0.8	HADS Mild: 8–10 Moderate: 11–14 Severe: 15–21	HADS Mild: 8–-10 Moderate: 11–14 Severe: 15–21	Subjects stopped hormonal meds and insulin sensitizing meds x 3 months	Yes
Moran et al. (2015), Australia	Rotterdam, both: confirmed Both: birth cohort	None	Education, living with a partner (C)*, at least one child	87	637	30.2 (IQR: 29.9–-30.8)	30.2 (IQR: 29.9–-30.9)	30.1 (IQR: 25.1–-38.6)**	25.4 (IQR: 22.4–29.9)	CES-D Clinical depression: >16	NA	No med exclusion	Yes
Ozenli et al. (2008), Turkey	Rotterdam PCOS: confirmedtt; controlsc PCOS: gyn clinic; controls: friends and family of staff	None; subjects with infertility excluded	Low income, education, married, infertility	35	35	27.6 ± 7.7	26.5 ± 5.2	25.4 ± 5.5	24.8 ± 5.4	BDI	STAI	Subjects stopped hormonal meds x 3 months	No, prevalence not reported
Sayyah-Melli et al. (2015), Iran	Rotterdam, both: confirmed Both: academic center clinics	BMI, day of menstruation	University education (P)**, married (C)**	742	798	23.5 ± 5.2**	27.1 ± 5.9	27.4 ± 8.2	26.8 ± 7.1	Screened with MMPI. Confirmed by psychologist –DSM IV	Screened with MMPI. Confirmed by psychologist – DSM IV ‘Anxiety Disorder’	No med exclusion	Yes
Shi et al. (2011), China	Rotterdam, both: confirmed Both: infertility clinic	Infertility, all newly diagnosed	Education, married, infertility	30	30	25.4	26.7	Weight: 59.4 ± 6.88 kg	Weight: 53.9 ± 6.50 kg	SCL-90	SCL-90	No med exclusion	No, prevalence not reported
Soyupek et al. (2008), Turkey	Rotterdam, both: confirmed PCOS: not stated, controls: hospital staff and other volunteers	Age	None reported	37	35	24.1 ± 6.1	26.1 ± 5.7	24.7 ± 6.5	22.5 ± 2.6	BDI: ≥11	NA	No med exclusion	Yes
Soyupek et al. (2010), (overlap with Soyupek et al., 2008) Turkey	Rotterdam, both: confirmed PCOS: not stated, controls: hospital staff and other volunteers	Age	Employed (C)**	40	39	24.0 ± 6.0	26.2 ± 5.5	24.2  ± 5.6*	21.8  ± 3.8	BDI: ≥11	BAI: ≥11	No med exclusion	No, substantial overlap with Soyupek et al. (2008)
Weiner et al. (2004), Germany	NIH, PCOS: confirmed in some subjects, others self-reported prior PCOS diagnosis by blood tests; controlsd PCOS: endocrine clinic and community advertising, controls: community advertising	BMI, age, education, race	Socioeconomic level, education, married	27	27	28.1  ± 6.5	30.1 ± 6.4	38  ±  8	37  ±  7	DACL State and DACL Trait Depression	STAI	Subjects stopped hormonal and psych meds x 60 days	No, prevalence not reported

The following tools were used in included studies: Beck Depression/Anxiety Inventory (BDI/BAI), Brief Scale for Anxiety (BSA-S), Brief Symptom Inventory (BSI), Depression Adjective Checklist (DACL), Hospital Anxiety Depression Scale (HADS), Minnesota Multiphasic Personality Inventory (MMPI), Montgomery Asberg Depression Rating Scale (MADRS-S), Primary Care Evaluation of Mental Disorders Patient Health Questionnaire (PRIME-MD PHQ), State-Trait Personality Inventory by Spielberger (STPI), State-Trait Anger Expression Inventory (STAXI), Symptom Check List 90 (SCL-90), Quick Inventory of Depressive Symptomatology Self Report (QIDSSR), and Zung self-rating depression scale.

^#Author provided unpublished information.

* p < 0.05 compared to controls ** p < 0.01 compared to controls.

(P): greater prevalence of demographic characteristic in PCOS subjects.

(C): greater prevalence of demographic characteristic in controls.

NA: Analysis not performed.

^tIn addition to NIH criteria, PCOS subjects also had infertility, LH/FSH > 2, color Doppler findings of decreased stromal ovarian artery resistance.

^ttAll PCOS subjects had oligo or amenorrhea.

Rule out of PCOS in controls by:

^aRegular menses and no clinical evidence of hyperandrogenism.

^bRegular menses and no known hormonal disturbance.

^cSelf-report of no prior POCS diagnosis by a physician and no PCOS symptoms.

^dRegular menses.

Criteria for the MA

For our MA on the prevalence of depression and anxiety symptoms in women with PCOS, articles that reported only mean depression and anxiety scores, and not prevalence, were excluded. In addition, to only include high-quality studies, only studies where diagnosis of PCOS was confirmed by the investigators, rather than self-report, were included. One other study that met inclusion criteria (Davari-Tanha et al., 2014) was excluded after final review as the prevalence of depression, diagnosed by a psychiatrist, in the control group was 87%.

The Meta-Analysis of Observational Studies in Epidemiology guidelines (MOOSE) (Stroup et al., 2000) were applied to this study.

Quality assessment

Studies were assessed for quality using the Newcastle-Ottawa Quality Assessment Scale (NOS) (Wells et al., 2010) with one adaptation. Under our modified scale, one point was awarded in the ‘selection of controls’ category if the subjects were all recruited from the same location (i.e. both controls and PCOS subjects were recruited from the clinic). This represents a change from the original scale, which awards one point for community controls. This was changed as other meta-analyses on this subject have suggested that recruitment from the same location may be superior to having controls recruited from the community if PCOS subjects are recruited from a hospital clinic (Barry et al., 2011b; Veltman-Verhults et al., 2012). Studies were categorized into low (0–3 points), medium (4–6 points) and high (7–9 points) quality studies.

Data extraction

The following data were extracted from each paper when possible: diagnostic criteria for PCOS and whether this was confirmed by the study or self-reported by subjects, any characteristics that subjects with PCOS and controls were matched on, comparison of demographic characteristics between groups, location of subject recruitment, total number of subjects, mean age and mean BMI for both groups, screening tool and cutoffs used, study exclusion criteria including current use of hormonal medication or Metformin, non-response rate, origin of study, prevalence of mild/moderate/severe depressive or anxiety symptoms, and mean depression or anxiety scores. If any data were missing, the authors were contacted for additional information.

Data analysis

STATA version 14.1 was used to perform random effects meta-analyses to estimate the pooled odds of elevated depression and anxiety scores in women with PCOS compared with control women without PCOS and to construct forest plots. Because most scales categorize symptom severity into none, mild, moderate or severe, with moderate and severe considered clinically significant, a separate analysis was also performed of studies comparing the odds of moderate and severe depressive or anxiety symptoms. When a psychiatrist diagnosed depression or anxiety, this was put in the ‘moderate and severe’ category.

Chi-squared tests were used to test the significance of the pooled odds ratio (OR); I² tests of heterogeneity were also applied. Publication bias was assessed with visual inspection of the funnel plots.

Sensitivity analysis

Because of the anticipated study heterogeneity, the following sensitivity analyses were planned to assess the robustness of our results: level of NOS quality, criteria used for PCOS diagnosis, studies that matched on BMI, age, marital status or infertility status, whether the groups were comparable in terms of baseline demographic characteristics (income, race, employment, education, marital status, infertility/number of children), tool used for the diagnosis of depression or anxiety symptoms, whether hormonal medications or Metformin were stopped prior to the study, continent of origin, location of PCOS subject recruitment, location of control recruitment, agreement between location of PCOS subject and control recruitment including whether the study specified that controls presented for routine care or an annual exam, whether PCOS was completely excluded in the controls, and whether these studies were included in the prior meta-analyses assessing prevalence of depression or anxiety symptoms (Dokras et al., 2011, 2012).

Meta-regression

Literature search, inclusion/exclusion criteria and data extraction

For our secondary SR and meta-regression on the association between depressive and anxiety symptoms in women with PCOS and PCOS-related symptoms or comorbidities, we used the same key words to perform our literature search as above (Supplementary Table S1); however, we had additional inclusion and exclusion criteria. We included all articles with cross-sectional comparisons of clinical and laboratory characteristics between women who had both PCOS and concurrent depressive or anxiety symptoms and women with PCOS without depressive or anxiety symptoms. Positive depressive and anxiety symptoms were diagnosed if a subject scored as mild or above in the symptom severity for a validated screening test. The data extracted included the same study characteristics as extracted in the main SR in addition to the following variables: testosterone (mean values: total testosterone, free testosterone (FT) and free androgen index (FAI)), hirsutism (evaluated both as mean Ferriman Galleway (FG) score and prevalence of hirsutism, defined as FG >8; Ferriman and Gallwey, 1961) and IR (mean values: homeostatic model assessment–insulin resistance (HOMA-IR)). These variables were chosen because they each have been shown to have significant associations with depression and/or anxiety in other populations, although the strength and magnitude of these associations are unclear (Holt et al., 2009; Bromberger et al., 2010; Ekback et al., 2013; Kan et al., 2013; Milman et al., 2015; Pasch et al., 2016). In addition, an elevated testosterone and an elevated FG score are part of the diagnostic criteria of PCOS and are thus measured and reported by most studies. The articles were searched by L.C. and I.L.

Data analysis

The pooled standardized mean difference (SMD) was used to compare the following variables: age, BMI, total testosterone, FT, FG score and HOMA-IR between those with and without depressive or anxiety symptoms. The SMD represents the group difference relative to the pooled variability observed among groups. Statistically, the SMD magnitude is interpreted as 0.20 (small), 0.50 (moderate) and 0.80 (large) (Cohen, 1992). The SMD was chosen rather than comparing mean values directly because of the heterogeneity in assays used for measurement of each of the laboratory variables. Some studies reported FAI rather than FT; both measures were included together as markers of FT when calculating the SMD. The pooled odds of hirsutism between those with and without depressive/anxiety symptoms were calculated using random effects MA.

Results

Search results

Figure 1 demonstrates the flow chart for articles selected. A search of published SRs or meta-analyses (prior to 2012) identified 28 studies, and a new search of the literature from 2011 to 2016 identified 515 articles that met our current inclusion criteria. Of these, 85 articles were evaluated in detail. A total of 30 studies from 2003 to 2016 met our inclusion criteria for our SR (Table I) (Elsenbruch et al., 2003; Weiner et al., 2004; Himelein and Thatcher, 2006; Barnard et al., 2007; Hollinrake et al., 2007; Adali et al., 2008; Battaglia et al., 2008; Benson et al., 2008; Mansson et al., 2008; Ozenli et al., 2008; Soyupek et al., 2008, 2010; Bhattacharya and Jha, 2010; Jedel et al., 2010; Moran et al., 2010, 2012, 2015; Barry et al., 2011a; Cinar et al., 2011; Deeks et al., 2011; Shi et al., 2011; Arshad et al., 2012; Acmaz et al., 2013; Ercan et al., 2013; Harmanci et al., 2013; Altinkaya et al., 2014; Annagür et al., 2014; Davari-Tanha et al., 2014; Asik et al., 2015; Sayyah-Melli et al., 2015). All 30 reported depressive symptoms and 18 reported anxiety symptoms. In total, there were 3050 subjects with PCOS and 3858 controls from 10 different countries in the depressive symptoms SR and 1910 subjects with PCOS and 1639 controls from 7 different countries in the anxiety symptoms SR.

Figure 1

Flow chart of study inclusion in systematic review and meta-analysis.

Of the 30 studies in the SR, 26 used Rotterdam for PCOS diagnosis and 4 used NIH criteria. There were 17 studies that matched PCOS and control patients on a demographic characteristic, most commonly on age (n = 13) and/or BMI/weight (n = 5). Most studies recruited their PCOS patients from a hospital clinic (n = 20). The mean BMI in the majority of studies was in the overweight range (BMI: 25–29.9 kg/m²) for the PCOS group but in the normal range (18.5–24.9 kg/m²) for the control group. The mean age for both groups in most studies was between 20 and 30 years old. The most common screening tools used were the Beck Depression/Anxiety Inventory (BDI/BAI) and the Hospital Anxiety and Depression Scale (HADS), although two studies used diagnosis by a psychiatrist. There were 17 studies which had some type of medication exclusion, most commonly for subjects who took hormonal medication for at least the prior month. Most studies (n = 24) were of medium quality (Supplementary Table SII).

Systematic review

Comparisons between the prevalence of depressive and anxiety symptoms and the mean depression/anxiety scores in PCOS subjects versus controls are presented in Supplementary Table SIII. The majority of studies reported significantly higher depressive symptom prevalence and/or scores (25/30) and significantly higher anxiety symptom prevalence and/or scores (16/18) in the PCOS subjects compared with controls.

Meta-analysis

There were 18 studies included in the MA to estimate a pooled association between PCOS and depressive symptoms and 9 studies met the criteria for inclusion in the MA to estimate a pooled association between PCOS and anxiety symptoms. There was only one study in each of the following categories: NIH diagnostic criteria, matched on infertility or matched on marital status; thus, these could not be evaluated in our sensitivity analyses.

Depressive symptoms

The median prevalence of depression in the 18 included studies was 36.6% (IQR: 22.3, 50.0%) in the PCOS group and 14.2% (IQR: 10.7, 22.2%) in the control group. Using a random effects model, the MA showed a significantly increased odds of depressive symptoms in women with PCOS compared with controls (OR: 3.78; 95% CI: 3.03, 4.72) (Fig. 2A). The 11 studies that reported the prevalence of moderate or severe depressive symptoms showed a 4.18-increased odds in women with PCOS compared with controls (95% CI: 2.68, 6.52) (Fig. 2B).

Figure 2

Results of meta-analysis on depressive symptoms. Forrest plot comparing odds of depressive symptoms in women with polycystic ovary syndrome (PCOS) compared with controls. (A) Overall odds of depressive symptoms, random effects; (B) odds of moderate and severe depressive symptoms and (C) odds of depressive symptoms in studies that did and did not match PCOS subjects and controls on BMI.

There was no substantial heterogeneity among studies in the overall MA (I² = 22.4%, P= 0.19). Our pre-specified sensitivity analyses are reported in Table II. Studies where PCOS subjects and controls were matched on BMI reported an OR of 3.25 for depressive symptoms (95% CI: 1.73, 6.09; four studies) (Fig. 2C). There were no differences among subgroups when stratified by any of the other methodological characteristics except the analysis on hormonal medication, where studies that stopped hormonal medication ahead of time had higher odds of depressive symptoms (Table II). Homogeneous effect sizes (tests of heterogeneity with P> 0.05) were achieved in all subgroup analyses, except when studies were not matched on age. All effect sizes within individual subgroups remained statistically significant.

Table II

Sensitivity analyses of methodological characteristics.

	Depression					Anxiety
	Number of studies	Total number of subjects (PCOS/controls)	OR (95% CI)	I2 (%)	P-valuet	Number of studies	Total number of subjects(PCOS/controls)	OR (95% CI)	I2 (%)	P-value
Meta-analysis
Random effects	18	1917/2346	3.78 (3.03, 4.72)	22.40		9	1409/1236	5.62 (3.22, 9.80)	59.6*
Fixed effects	18	1917/2346	3.56 (2.99, 4.24)	22.40			1409/1236	4.71 (3.49, 6.36)	59.6*
Moderate/severe symptoms	11	1385/1919	4.18 (2.68, 6.52)	56.4*		5	1137/993	6.55 (2.87, 14.93)	69.6*
Sensitivity analyses
NOS quality assessment
Low	1	117/84	5.71 (3.05, 10.71)		0.6					0.43
Medium	16	1770/2232	3.63 (2.87, 4.59)	21.10		8	1379/1206	5.20 (2.90, 9.32)	60.2*
High	1	30/30	4.00 (1.27, 12.58)			1	30/30	11.23 (3.10, 40.71)
Matched on BMI
Yes	4	1023/931	3.25 (1.73, 6.09)	49.30	0.33	3	998/913	6.30 (1.88, 21.09)	82.7*	0.92
No	14	895/1415	3.94 (3.13, 4.96)	4.50		6	411/323	5.32 (2.85, 9.90)	33.30
Matched on age
Yes	8	336/306	3.83 (2.63, 5.59)	0.00	0.91	4	200/179	6.94 (3.78, 12.73)	0.00	0.3
No	10	1581/2040	3.99 (2.91, 5.46)	47.0*		5	1209/1057	4.96 (2.18, 11.29)	72.5*
Compared demographic characteristics between groupsa
All similar	7	373/943	3.75 (2.36, 5.96)	45.40	0.82	1	30/30	11.23 (3.10, 40.71)		0.68
At least one different	6	1277/1132	4.35 (2.89, 6.56)	46.30		6	1277/1132	5.69 (2.84, 11.40)	60.6*
None reported	5	267/271	3.50 (2.24, 5.48)	0.00		2	102/74	4.22 (0.84, 21.2)	79.6*
Scale used
Psychiatrist	2	791/847	2.84 (2.11, 3.81)	0.00	0.63	2	791/847	2.59 (1.63, 4.12)	0.00	0.35
HADS	2	123/74	4.11 (1.92, 8.81)	0.00		3	349/159	4.47 (1.54, 12.98)	72.6*
BDI/BAI	10	666/571	4.53 (3.17, 6.47)	14.00		2	136/97	8.72 (4.02, 18.89)	0.00
Other	4	337/854	3.72 (2.25, 6.13)	52.70		2	133/133	12.42 (4.17, 36.97)	0.00
Hormonal medication
Stopped	9	660/412	5.19 (3.70, 7.26)	0.00	0.044	5	444/236	6.96 (3.40, 14.24)	55.40	0.69
Not stopped	5	1021/1687	3.07 (2.26, 4.17)	28.20		3	894/950	4.55 (1.40, 14.77)	47.80
Not reported	4	236/247	3.78 (3.03, 4.72)	0.00		1	71/50	3.99 (1.49, 10.64)
Metformin
Stopped	5	1021/1687	3.07 (2.26, 4.17)	28.20	0.09	3	894/950	4.55 (1.40, 14.77)	47.80
Not stopped	6	432/263	5.09 (3.20, 8.09)	0.00		4	358/189	6.83 (2.74, 17.04)	66.4*
Not reported	7	464/396	4.11 (2.73, 6.20)	28.60		2	157/97	5.33 (2.56, 11.13)	0.00
Continent of origin
America	2	143/203	5.18 (2.78, 9.66)	0.00	0.37	1	103/103	16.04 (2.07, 124.46)		0.41
Asia	10	1474/1357	4.21 (3.10, 5.71)	32.00		5	1175/1030	5.69 (2.81, 11.53)	69.24*
Australia	2	139/661	2.36 (1.54, 3.63)	0.00		1	52/24	1.89 (0.71, 5.03)
Europe	4	161/125	3.45 (2.01, 5.95)	0.00		2	79/79	9.80 (3.24, 29.62)	0.00
Recruitment of PCOS
Reproductive endocrinology clinic	2	143/203	5.18 (2.78, 9.66)	0.00	0.53	1	103/103	16.04 (2.07, 124.46)		0.8
Gyn/PCOS clinic	6	1068/1052	4.36 (2.87, 6.60)	47.40		3	877/894	4.05 (1.69, 9.72)	50.20
Internal med/endocrine clinic	4	425/297	4.23 (2.69, 6.65)	0.00		2	297/135	6.86 (2.37, 19.87)	59.10
Community	3	169/691	2.52 (1.69, 3.77)	0.00		2	82/54	4.38 (0.76, 25.06)	78.60
Unknown	3	112/103	2.77 (1.17, 6.56)	35.60		1	50/50	9.75 (3.32, 28.65)
Recruitment of controls
Clinic	6	1107/1109	3.86 (2.80, 5.33)	24.20	0.75	3	916/951	3.57 (1.68, 7.59)	44.80	0.52
Community	9	649/1122	3.32 (2.48, 4.45)	4.90		4	357/188	6.17 (2.21, 17.27)	64.0*
Not stated	3	161/115	4.20 (1.41, 12.51)	64.00		2	136/97	8.72 (4.02, 18.89)	0.00
Subject recruitment agreement (PCOS/controls)
Clinic/clinic [routine care]#	3	252/219	5.33 (3.43, 8.30)	0.00	0.11	1	103/103	16.04 (2.07, 124.5)	0.00	0.3
Clinic/clinic [not specified]	3	855/890	2.91 (2.18, 3.88)	0.00		2	813/848	2.71 (1.76, 4.15)
Community/community	3	169/691	2.52 (1.69, 3.77)	0.00		2	82/54	4.38 (0.76, 25.06)	78.6*
Clinic/community	5	443/396	4.39 (2.91, 6.62)	0.00		2	275/134	10.63 (4.48, 25.18)	0.00
Community/clinic	0					0
Either unknown	4	198/150	3.81 (1.67, 8.71)	54.00		2	136/97	8.72 (4.02, 18.89)	0.00
Inclusion in prior MA
Yes	9	500/489	4.27 (3.12, 5.86)	0.00	0.54	3	182/182	10.95 (4.14, 28.99)	0.00	0.25
No	9	1417/1857	3.73 (2.69, 5.16)	41.40		6	1227/1054	4.75 (2.53, 8.95)	67.5*

MA, meta analysis; OR, odds ratio. I², test for heterogeneity.

*P-value for Chi² test for heterogeneity is <0.05.

^aIncluded demographic characteristics are income, employment, education, marital status and infertility/number of children.

^#Among the studies where both PCOS subjects and controls were recruited from the clinic, the reason for control presentation was divided into presenting for routine care/annual exam versus not specified.

^tP-value is Chi² test for subgroup differences.

When we specifically looked at the location of subject recruitment, the most common site of recruitment for PCOS subjects was from a clinic (N = 12) and for controls, it was from the community (N = 9). To identify whether this was a source of bias, we looked at agreement of recruitment site between PCOS subjects and controls and further examined the reason for presentation to the clinic. Homogeneous effect sizes with I² values of 0.0% were achieved in all subgroups where location of subject recruitment was specified by the study. There were no differences among subgroups (P= 0.11).

If we repeated our MA including the four studies that were excluded from our MA because of other study factors (Barnard et al., 2007; Moran et al., 2010; Deeks et al., 2011; Davari-Tanha et al., 2014), the increased odds of depressive symptoms in women with PCOS remained similar (OR: 3.60; 95% CI: 2.81, 4.62), although the heterogeneity increased (I² = 57.1%; P = 0.001).

A funnel plot of the log OR of each study plotted against the standard error of the log OR (with smaller standard error representing larger studies) shows some evidence of publication bias (Supplementary Fig. S1A), as there were fewer published studies that are small and report negative findings.

Anxiety symptoms

The median prevalence of anxiety in the nine included studies was 41.9% (IQR: 13.6, 52.0%) in the PCOS group and 8.5% (IQR: 3.3, 12.0%) in the control group. Using a random effects model, the MA showed a significantly increased odds of anxiety symptoms in women with PCOS compared with controls (OR: 5.62; 95% CI: 3.22, 9.80) (Fig. 3A). The odds of moderate and severe anxiety symptoms were similarly high in this group (OR: 6.55; 95% CI: 2.87, 14.93; five studies) (Fig. 3B).

Figure 3

Results of meta-analysis on anxiety symptoms. Forrest plot comparing odds of anxiety symptoms in women with PCOS compared with controls. (A) Overall odds of anxiety symptoms, random effects; (B) odds of moderate and severe anxiety symptoms and (C) Odds of anxiety symptoms in studies that did and did not match PCOS subjects and controls on BMI.

There was substantial heterogeneity among studies in the overall MA (I² = 59.6%; P= 0.01); however, this was not explained by any of our sensitivity analyses, as there were no significant differences among subgroups when stratified by methodological characteristics (Table II). Studies where PCOS subjects and controls were matched on BMI reported 6.30 increased odds of anxiety symptoms in women with PCOS (95% CI: 1.88, 21.09), although this only included three studies. The majority of subgroup analyses resulted in homogeneous effect sizes (tests of heterogeneity with P> 0.05).

When we specifically looked at the location of subject recruitment for the anxiety analysis, the distribution of recruitment was similar to what was seen in the depressive symptom analysis with the clinic being the most common site of PCOS subject recruitment (N = 6) and the community being the most common site of control recruitment (N = 4). When we looked at agreement of subject recruitment between PCOS subjects and controls, the overall number of studies in each subgroup was small, making comparisons of different subgroups less robust. Despite visual differences between the ORs in these subgroups, there was no statistical difference among subgroups (P= 0.30).

Inclusion of the two studies that were excluded from our MA because of other study factors (Moran et al., 2010; Deeks et al., 2011) resulted in a similarly increased odds of anxiety symptoms in women with PCOS (OR: 4.45; 95% CI: 2.91, 6.80, I² = 57.8%).

A funnel plot of the studies included in the anxiety MA shows some evidence of publication bias (Supplementary Fig. S1B), as there were fewer published studies that are small and report negative findings.

Meta-regression

Supplementary Figure S2 demonstrates the flow chart to identify articles that reported comparisons between women with PCOS with concurrent depressive or anxiety symptoms and women with PCOS without such symptoms. There were 15 studies that reported a least one of the following comparisons: age, BMI, total testosterone, FT/FAI, FG score, prevalence of hirsutism or HOMA-IR (Hollinrake et al., 2007; Mansson et al., 2008; Soyupek et al., 2008; Bhattacharya and Jha, 2010; Jedel et al., 2010; Cinar et al., 2011; Livadas et al., 2011; Moran et al., 2012; Annagur et al., 2013; Bazarganipour et al., 2013; Rahiminejad et al., 2014; Asik et al., 2015, Greenwood et al., 2015; Klimczak et al., 2015; Naqvi et al., 2015). Of these, 14 studies reported depressive symptoms and 6 reported anxiety symptoms (Supplementary Table SIV).

In the meta-regression evaluating the pooled SMDs (Table III), women with PCOS and concurrent depression had a higher mean age, BMI, FG score and HOMA-IR than women with PCOS who did not have depression, although the effect sizes were small. In addition, women with depressive symptoms had a 53% increased odds of hirsutism (FG score ≥8) compared with those without depressive symptoms (OR: 1.53; 95% CI: 1.10, 2.12; six studies). There was no substantial heterogeneity in the analysis on hirsutism (I² = 1.4%, P= 0.41). Women with PCOS and concurrent anxiety had a higher mean BMI, FG score and FT/FAI, but not HOMA- IR, than women with PCOS who did not have anxiety. Women with anxiety did not have an increased odds of hirsutism (OR: 1.17; 95% CI: 0.43, 3.17, two studies).

Table III

Meta-regression of associations between laboratory and clinical measurements and depressive or anxiety symptoms in women with PCOS.

	Depression			Anxiety
	Number of studies	SMD (95% CI)	I2(%)	Number of studies	SMD (95% CI)	I2(%)
Age	12	0.11 (0.01, 0.22)*	0.0	6	0.09 (−0.07, 0.24)	0.0
BMI	14	0.23 (0.07, 0.40)*	52.4*	7	0.26 (0.04, 0.47)*	41.2
Total testosterone	9	0.06 (−0.10, 0.22)	28.5	4	0.16 (−0.13, 0.45)	37.2
Free testosterone/FAI	9	0.01 (−0.15, 0.16)	30.4	6	0.27 (0.02, 0.51)*	47.1
Mean FG score	6	0.18 (0.04, 0.31)*	0.0	4	0.20 (0.04, 0.37)*	0.0
Insulin resistance: HOMA-IR	5	0.26 (0.08, 0.44)*	24.3	3	0.29 (−0.10, 0.67)	50.7

FAI, free androgen index; FG, Ferriman Galleway; I², test for heterogeneity; SMD, standardized mean difference.

*P < 0.05.

Discussion

Our MA of 18 studies from different regions of the world showed that women with PCOS have over three times the odds of depressive symptoms and over five times the odds of anxiety symptoms than controls, supporting prior research on this topic (Dokras et al., 2011, 2012; Barry et al., 2011b; Veltman-Verhulst et al., 2012). More importantly, we showed for the first time that this association remained significant when we restricted our analysis to the prevalence of moderate and severe depression and anxiety scores. Our findings suggest a worldwide increased risk of depressive and anxiety symptoms of a significant magnitude in the young PCOS population emphasizing the importance of screening and appropriate follow-up in this population.

Obesity, a common morbidity associated with PCOS (ACOG Practice Bulletin No. 108: Polycystic ovary syndrome, 2009), is also associated with an increased risk of depression (Luppino et al., 2010). Restricting our analysis to the four studies where PCOS and controls subjects (~2000 subjects) were matched on BMI, we report a significantly increased odds of depressive symptoms, suggesting that women with PCOS have an increased prevalence of depressive symptoms independent of BMI. These findings are also supported by three studies that performed logistic regression adjusting for BMI and each found a significantly elevated odds of depressive symptoms in women with PCOS (Hollinrake et al., 2007; Bhattacharya and Jha, 2010; Deeks et al., 2011).

In our meta-regression, we found that women with PCOS and depressive symptoms had higher mean values of both HOMA-IR and BMI compared with PCOS women without depression. In the general population, meta-analyses have shown associations between depression and IR (18 studies) (Kan et al., 2013), obesity (15 studies) (Luppino et al., 2010) and diabetes (7 studies) (Mezuk et al., 2008); however, most studies are cross-sectional and cannot establish the direction of association. The small magnitude of the differences observed in our meta-regression (SMD for IR: 0.26; BMI: 0.23) suggests that these factors are unlikely to be responsible for the entire relationship between PCOS and depressive symptoms. There is little information on the association between anxiety and IR in the general population. A large MA of 12 studies showed a small, but significant, positive association between diabetes and both anxiety disorders (OR: 1.20; 95%CI: 1.10–1.31) and elevated anxiety symptoms (OR: 1.48; 95% CI: 1.02–1.93); however, the direction of the association could not be assessed (Smith et al., 2013). In our MA, only three studies evaluated the risk of anxiety in BMI-matched subjects, and in the meta-regression, the magnitude of the difference in mean BMI between anxiety groups was small (SMD: 0.26), making it difficult to draw conclusions about the associations between anxiety and BMI. Further, we found no significant association between anxiety and HOMA-IR.

We also found that women with PCOS and depressive or anxiety symptoms had higher odds of hirsutism and/or increased FG scores. High patient-rated FG scores have been associated with elevated HADS depression and anxiety scores (r = 0.209; P< 0.05 and r = 0.230; P< 0.01, respectively) in women who present to dermatology clinics to discuss hair removal treatments (Ekback et al., 2013). Another study also showed that patient-rated FG scores, but not clinician-rated FG scores, were significantly associated with the risk of depression, indicating some influence of a patient's perceived level of hirsutism on these results (Pasch et al., 2016). This distinction could not be evaluated in our meta-regression as only two studies (Jedel et al., 2010; Greenwood et al., 2015) described whether the subject or physician performed FG scoring.

We examined infertility as a potential confounder. Although most studies did not match on infertility status, when PCOS subjects were recruited from general medicine or endocrinology clinics (likely not seeking fertility care), they had increased odds of both depressive (OR: 4.2) and anxiety (OR: 6.9) symptoms. PCOS subjects who were recruited from the community also had increased odds of depressive symptoms (OR: 2.5), again suggesting that this increase may be independent of complaints of infertility. In addition, in our SR, three of the four studies that either recruited all infertile subjects or excluded all infertile subjects showed significantly higher depression and/or anxiety scores in the PCOS group (Adali et al., 2008; Ozenli et al., 2008; Shi et al., 2011; Davari-Tanha et al., 2014). The prevalence of depression and anxiety in the infertile population has been reported to range between 7–26% and 14–23%, respectively (Chen et al., 2004; Volgsten et al., 2008; Biringer et al., 2015; Gameiro et al., 2016; Sethi et al., 2016). These numbers appear to be lower than the prevalence in women with PCOS in our MA (median: 36.6 and 41.9%, respectively). Collectively, the results of our sensitivity analyses and review of other studies suggest that infertility likely does not explain the entire relationship between PCOS and an increased risk of depression/anxiety symptoms.

The strengths of our study include the evaluation of a large number of studies from several geographical regions, the ability to analyze the effect of important confounders such as obesity and medication use, and the evaluation of the impact of site of subject recruitment. We performed a separate SR for our meta-regression and thus included a large number of subjects in the meta-regression. Prior meta-analyses had limitations such as inclusion of studies where PCOS was not confirmed (Dokras et al., 2011, 2012; Barry et al., 2011b; Veltman-Verhulst et al., 2012) or studies that included emotional quality of life symptoms in addition to depressive and anxiety symptoms (Barry et al., 2011b; Veltman-Verhulst et al., 2012). When our strict inclusion criteria were applied and we evaluated the prevalence of depressive and anxiety symptoms rather than mean scores, our MA included twice as many studies meeting these criteria compared with any of the prior publications. Inclusion of many different screening tools, all of which demonstrated an increased risk of depressive and/or anxiety symptoms in women with PCOS, makes our findings generalizable and supports the use of an easy to implement tool (Siu et al., 2016).

There are many methodological considerations for cross-sectional studies that could lead to a false elevation in the association between PCOS and depressive and anxiety symptoms. Another strength of our study is that we performed numerous sensitivity analyses to evaluate different biases that could affect our association. We specifically assessed the recruitment methods of PCOS subjects and controls to overcome biases introduced by site and type of visit. If PCOS subjects are recruited from a clinical population, whereas controls are recruited from the community, where they are more likely to be healthy, the resulting increase in depressive and anxiety symptoms in those with PCOS could reflect selection bias. In all scenarios for subject recruitment, women with POCS had higher odds of depressive symptoms. Although recruitment from clinical sites showed an increased odds of anxiety symptoms in women with PCOS, there were very few subjects in the community-based studies, limiting our ability to evaluate this comparison.

One limitation of our MA is the large variation in methodological characteristics among the studies in the anxiety analysis. Although we attempted to address this with our sensitivity analyses, the heterogeneity among studies remained high in most subgroups, specifically the analyses on moderate/severe symptom severity and studies that matched on BMI. Individual patient data were not available for a more accurate estimation of risks, control of confounders or evaluation of the association between different PCOS phenotypes and psychiatric symptoms. We were also unable to evaluate other potential factors that might contribute to the development of depressive and anxiety symptoms in women with PCOS, such as higher inflammation or elevated cortisol (Gallinelli et al., 2000; Benson et al., 2009; Farrell and Antoni, 2010), as there are no studies that have looked at the association between these factors and depression or anxiety in women with PCOS.

Unlike mild anxiety or depression due to a specific occasion, these conditions in a person with anxiety or depressive disorder do not go away and may get worse over time as compared with temporary worry or fear. A psychiatrist made the diagnosis of either depressive disorder or generalized anxiety disorder in only two studies included in our MA (Mansson et al., 2008; Sayyah-Melli et al., 2015). However, even in this small subgroup analysis, women with PCOS had higher odds of a diagnosis of depression and anxiety compared with controls, supporting our overall findings. There are limited data on the longitudinal follow-up of women with PCOS, and this was not evaluated in our MA. Both a Taiwanese study using a national health insurance database (median follow-up: 5 years) and an Australian study using hospital records (median follow-up: 8 years) showed that women with PCOS, who did not have a psychiatric disorder at baseline, had higher hazards of developing depression and anxiety during follow-up (Hung et al., 2014; Hart and Doherty, 2015). In addition, a US study showed that women with PCOS had high rates of both persistence of existing depression and development of new depression over 2 years of follow-up (Kerchner et al., 2009).

Given the high prevalence of depression in the general population and the positive effects of treatment, in 2016 the US Preventative Task Force recommended that all adults be screened for depression (Siu et al., 2016). Our MA demonstrates a significantly increased risk of moderate and severe depressive and anxiety symptoms in women with PCOS. Further, a large population-based study from Sweden (Cesta et al., 2016) and data from hospital admissions in Western Australia (Hart and Doherty, 2015) confirm that women with PCOS are at an increased risk of depression and anxiety disorders. The Australian PCOS guidelines recommend that women with PCOS be routinely screened for depression and anxiety and assessed by appropriately qualified health professionals (PCOS Australian Alliance, 2015). Our data support the consideration of implementation of these guidelines to women with PCOS around the world. Future studies should focus on the longitudinal follow-up of women with PCOS and the long-term impact of PCOS-related treatments on depressive and anxiety symptoms in this high-risk population.

Supplementary data

Supplementary data are available at Human Reproduction online.

Authors’ roles

L.G.C. reviewed articles, performed analysis and wrote manuscript. I.L. reviewed articles and reviewed manuscript. M.D.S. (biostatistician) reviewed manuscript. A.D. reviewed articles and wrote and reviewed manuscript.

Funding

No specific funding was provided for this project. L.G.C is supported by a NIH Reproductive Epidemiology Training Grant (T32 - HD007440).

Conflict of interest

None declared.