Risk of hypertension in women with polycystic ovary syndrome: a systematic review, meta-analysis and meta-regression

In this meta-analysis, PubMed, Scopus, ScienceDirect, web of science, and Google scholar were searched for retrieving observational studies published from inception to April 2019 investigating the HTN in patients with PCOS.

Before initiation of the study, we conducted the search strategy with the assistance of a professional healthcare librarian. All reviewers performed searches separately. At first, search in the PubMed was performed based on control vocabularies (MESH) using the following formula: (“Polycystic ovary syndrome” OR “PCOS”) AND (“cardiovascular” OR “cardio-metabolic” OR “metabolic” OR “hypertension” OR “hypertensive” OR “blood pressure”).

We also searched PubMed and other databases using free-text terms. Search criteria were humans, and English language. Search strategy was almost similar for all databases. The searches were done based on the ‘all fields’ in the PubMed and ‘titles, abstracts and keywords’ in other databases. A ‘pearl growing’ strategy was employed, whereby, after obtaining the full text articles, the reference lists of all included studies were reviewed for additional publications that could be used in this review.

Analytic observational studies of all types including cross-sectional, case-control, and cohort designs that assessed hypertension in women with PCOS were eligible to be included in the meta-analysis. In addition, studies needed to report a risk ratio (odds ratio [OR], relative risk [RR], or hazard ratio) or should have provided sufficient information to allow calculation of a relevant effect estimate. We included studies using National Institutes of Health (NIH) [22], Rotterdam [23], Androgen Excess Society (AES) [24], laparoscopic [25] and International Classification of Diseases (ICD) [26] as diagnostic criteria for PCOS.

Exclusion criteria included: (1) studies that did not differentiate between women and men, (2) studies that did not assess any cardiovascular events, (3) studies with unreliable and incomplete results, and (4) studies with unknown or invalid PCOS diagnostic criteria.

We included all relevant studies assessing hypertension in women with PCOS. The results of the searches were screened based on predefined eligibility criteria. All references were entered to EndNote X7 software (Thomson Reuters, New York, NY, USA). Initial selection was performed based on their titles, followed by a second selection performed by one reviewer (M.A), who deleted duplicates and reviewed the abstracts of all remaining records. Any disagreement in the selection of abstracts was resolved by consensus or by two other reviewers (F.R.T and E.C). Full text articles for review and data processing were obtained for all selected abstracts.

Two reviewers (M.A and S.B.G), in close consultation with another reviewer (F.R.T), extracted data from full text articles; they rechecked all precisely extracted data to minimize errors. For each study, the following information was extracted: Authors, year of publication, title, study design (cross-sectional, case-control, cohort), characteristics of study population, number of events, and the unadjusted or adjusted risk ratios provided (OR, RR, or hazard ratio) by each outcome. To prevent extraction errors, all reviewers performed a quality control check between the final data used in the meta-analysis and the original publications.

All studies included for the meta-analysis were appraised for the quality of their methodological and result presentation. Two reviewers (M.A and S.B.G) assessed the quality of the studies separately. They were blinded to study author, institution, and journal name. Disagreement was resolved and adjusted by other reviewers (F.R.T). All observational studies including cross-sectional, case-control, and cohort were appraised according to the Newcastle–Ottawa scale [27]. In this respect, 3 domains were scored for selection and comparability of study cohorts, and determination of the outcome of interest. If a study obtained ≥70% of the highest level of the Newcastle–Ottawa scale, it was considered to be of high quality, 40–70% as moderate, and 20 to 40% as low and < 20% as very low quality (Supplementary File 1).

We assessed the risk of bias in each study included, using the Cochrane Collaboration’s tools, which have been designed for various methodological studies including cross-sectional, case-control, and cohorts. Review authors’ judgments were categorized as “low risk,” “high risk,” and “unclear risk” of bias (either low or high risk of bias) [28].

The primary outcome of interest was hypertension, which was defined as systolic blood pressure (SBP) ≥ 140 mmHg and diastolic blood pressure (DBP) ≥ 90 mmHg or current use of anti-hypertensive medicine [29].

We performed a meta-analysis to estimate the pooled P / RRs of HTN. Heterogeneity was evaluated using the I-squared (I²) statistics; values above 50% were interpreted as heterogeneity. Both random and fixed effect models were used for heterogeneous and non-heterogeneous results, respectively. Publication bias was assessed using the Begg’s test [30]; bias was found to be significant for P-values < 0.05. The trim and fill method was not used because of non-significant results [31].

Pooled P and pooled RR (Pooled RR) were used for reporting results of the meta-analysis. The “Meta-prop” method was applied for the pooled estimation of the prevalence of HTN [32]; pooled RR was also estimated by the “Metan” method, using the normal distribution to estimate confidence intervals. Mantel–Haenszel method was used to estimate pooled data [33].

Subgroup analyses were performed to assess the pooled P / RR of HTN, based on age groups (reproductive vs. menopausal/aging) and study design (population- vs. non-population-based studies). Forest plots were drawn for RR of HTN in the mentioned subgroups as well.

Furthermore, the random effect meta-regression model was fitted to estimate the association between age groups and outcome of interest (here RR of HTN) in subgroups of study design. Bubble plots were drawn to illustrate the fitted models for each covariate.

We also adjust BMI and diabetes mellitus as a confounding variable to decrease the source of heterogeneity. Sensitivity analyses were performed to explore the source of heterogeneity with detecting the influence of any single study on the prevalence or relative risk of outcomes. Statistical analysis was performed using STATA software (version 10; STATA, INC., College Station, TX, USA).

Figure 1 presents the search strategy and study selection. Of 5236 records retrieved through searching databases 30 studies including 17 population based [2, 3, 7, 19, 34‐48] and 11 non-population based studies [4, 5, 10, 34, 49‐56] were selected for the final analyses. Twenty-four studies assessed a population of reproductive age patients with PCOS [2, 3, 5, 7, 10, 34‐38, 40, 41, 43‐47, 49‐51, 53‐56], 4 studies- menopausal/aging women [19, 39, 52, 57], and two studies- both age groups of patients (reproductive and menopausal/aging women) [4, 48]. Table 1 shows characteristics of studies included. Details of quality assessment are presented as supplementary file 1.

The review showed that the pooled prevalence of HTN in reproductive and menopausal/aging women with PCOS was higher than in the general population [(Pooled P: 0.15, 95% CI: 0.12–0.18 vs. Pooled P: 0.09, 95% CI: 0.08–0.10) and (Pooled P: 0.49, 95% CI: 0.28–0.70 vs. Pooled P: 0.40, 95% CI: 0.22–0.57), respectively]. The same results were obtained when only population-based studies were included in the meta-analysis [(Pooled P: 0.12, 95% CI: 0.08–0.15 vs. Pooled P: 0.08, 95% CI: 0.06–0.09) and (Pooled P: 0.60, 95% CI: 0.52–0.68 vs. Pooled P: 0.44, 95% CI: 0.40–0.48), respectively] (Table 2).

Figures 2, 3 and 4 illustrate the forest plots of pooled RRs of HTN in patients with PCOS, compared to control population. The RR of HTN in patients with PCOS was 1.60-fold (95% CI: 1.36–1.87) higher than in control population. Subgroup analysis based on the age groups revealed that the pooled RR of HTN in reproductive age PCOS patients was 1.70-fold (95% CI: 1.43–2.07) higher than in control populations of similar ages, but that the RR of HTN in menopausal/aging patients was not significantly different compared to the control population (Table 3, Fig. 2). Subgroup analysis of population-based studies revealed the same results; the pooled RR of HTN in reproductive age PCOS patients was 1.87-fold (95% CI: 1.51–2.33) higher than control population, while no significant difference was observed for menopausal/aging group (Table 3, Fig. 3). Figure 4 illustrates the forest plot of pooled RR of HTN for non-population based studies.

Meta-regression analysis based on all studies, population-based and non-population-based studies is presented in Fig. 5. Meta-regression analysis of population-based studies showed that the RR of HTN in reproductive age PCOS patients was 1.76-fold (95% CI: 0.65–5.30) higher than those of HTN in menopausal/aging PCOS patients (P = 0.262). After adjusting the model with BMI and diabetes mellitus result did not significantly change (Table 4).

There were significant heterogeneities in prevalence and RR of HTN among studies included in most subgroups of the study (Tables 2 and 3).

Egger’s test did not show any significant publication bias among studies included for HTN; therefore, trim and fill correction were not required (Tables 2 and 3).

Most cross-sectional and case-control studies had a low risk of bias in domains of assessment of exposure, development of outcome of interest in cases and controls, selection of cases, and selection of controls, and a high risk of bias in control of the prognostic variable. In the cohort studies, a low risk of bias selection of exposed and non-exposed cohorts, assessment of exposure, presence of outcome of interest at start of study, outcome assessment, and adequate follow up of cohorts were observed; however, we found a high risk of bias in control of prognostic variables and assessment of the presence or absence of prognostic factors (Supplementary File 1).

Sensitivity analysis suggested that the pooled RR and prevalence of hypertension were stable and excluding a single study did not change the significance of the pooled RR (Supplementary File 1). Since sensitivity analysis detected no significant heterogeneous study, the study type had no effect on the prevalence of hypertension. We also excluded the Rotterdam criteria and re-sun analyses; the results were analogues with the previous findings, indicating that the PCOS criteria had no significant effect on our results (supplementary File 1).