We searched the PubMed database using and expanding the MeSH terms “antihypertensive agents” and “depression” until August 30, 2021. The full search terms were illustrated in Supplementary Appendix S1. Literature retrieval was limited to human studies published in English. Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines for systematic reviews were followed and fulfilled (Supplementary Appendix S2) (Page et al., 2021). Publication that simultaneously fulfilled the following criteria were included in our study: 1) control groups were users of other classes of medication and/or nonmedicated participants; 2) studies that used a validated method to assess depression or symptoms of depression, and the measure of depression was used as a categorical variable rather than a continuous variable; and 3) studies with sample size of more than 100, so as to avoid selection bias. The selection of relevant literature was independently conducted by two researchers, and disagreements were resolved by consulting a third reviewer.

For all eligible studies, two researchers independently extracted the following data: the first author, PMID, the year of publication, country, study type, sample age range, percentage of males, the methods used to define depression, odds ratio (OR) or relative risk (RR) and the corresponding 95% CI, as well as control variables for adjustment. For the analysis model used in the studies, we gave priority to multivariate analysis or adjusted OR/RR values over univariate analysis or crude results. If the studies failed to report OR/RR, the raw data were reviewed to determine whether the OR/RR could be calculated. Different antihypertensive medication stratifications were treated as several independent results with corresponding populations separately. If there was stratification by the number of antihypertensive agents, the effect estimates were regarded as independent results separately.

The quality evaluation was assessed by two researchers independently. The Agency for Healthcare Research and Quality (AHRQ) was used to evaluate cross-sectional studies and the Newcastle-Ottawa Scale (NOS) was used for cohort studies (Rostom et al., 2004; Wells et al., 2014). The AHRQ comprised 11 items, with a scale ranging from 0 to 11 (Supplementary Appendix S3). Scores of 8–11 were regarded as high quality, and scores of 4–6 were regarded as moderate quality. For the NOS, an overall quality score ranged from 0 to 9 stars (Supplementary Appendix S4). When a study obtained more than six scores, it was regarded as high quality. When a study obtained scores of 4–6, it was regarded as moderate quality. According to the AHRQ and NOS, all studies involved were of high and moderate quality (Supplementary Table S1).

In meta-analysis, the degree of heterogeneity was assessed using the I ² statistic (Higgins et al., 2003). An I ² value < 50% was considered an acceptable level of heterogeneity and we used the fixed effects meta-analysis model to pool data; otherwise, the random-effects model was used. The combined effect estimates were shown as pooled ORs with 95% confidence intervals (CIs) and p values. The antihypertensive drugs were divided into four groups: angiotensin antagonists, beta blockers, calcium channel blockers and diuretics. ACEI and ARB were grouped together as angiotensin antagonists. The incidence of depression was pooled based on the single drug category for all subsequent analysis. Sensitivity analysis was conducted by omitting one study each time. Publication bias was investigated using funnel plots. Subgroup analysis was performed according to the type of study (cohort/cross-sectional). To investigate the impact of confounding factors in the control group, subgroup analysis was also conducted by dividing the control group into two subgroups, with the first subgroup not taking antihypertensive medication (NoAntiHTN) as a control and the second subgroup receiving other classes of antihypertensive drugs (AntiHTN) as a reference.

The network meta-analysis was performed using the frequentist approach in the package “netmeta” (version 1.5–0) in R (version 4.1.0; https://www.r-project.org/). The random-effects model was conducted in network meta-analysis to minimize the influence of heterogeneity. Treatments were ranked using P-score which were based solely on the point estimates and standard errors of the network estimates (Rücker and Schwarzer, 2015). Heterogeneity between studies was assessed using Cochran Q-statistics, and p value < 0.05 was considered suggestive of significant heterogeneity. Moreover, a net heat plot was employed to locate inconsistency in network meta-analysis. A comparison-adjusted funnel plot was used to assess publication bias. All statistical analyses were conducted using R version 4.1.0, with a 0.05 significance level.

The workflow of study selection is illustrated in Figure 1. In brief, a total of 9,557 publications were retrieved from the PubMed database, and 9 studies consisting 6 cross-sectional studies and 3 cohort studies were finally involved in this study (Gerstman et al., 1996; Feng et al., 2008; Simonson et al., 2011; Johansen et al., 2012; Michal et al., 2013; Ringoir et al., 2014; Boal et al., 2016; Cao et al., 2019; Agustini et al., 2020).

Characteristics of selected publications are presented in Table 1. Overall 414,873 individuals were included in the final analysis. The sample sizes ranged from 573 to 181,709 individuals. Subjects were recruited from a total of 10 countries, including Australia, China, Germany, Netherlands, Norway, Singapore, the United Kingdom and the United States. Of note, the methods used to assess depression or symptoms of depression varied across the nine studies. The two main diagnostic classification systems were Diagnostic and Statistical Manual of Mental Disorders and International Classification of Diseases (Malhi and Mann, 2018). In all nine studies, three studies identified depression based on these two criteria (Gerstman et al., 1996; Simonson et al., 2011; Boal et al., 2016).

Calcium channel blockers use was significantly associated with an increased risk of depression (OR 1.09, 95% CI 1.06–1.13; Figure 2A). Other antihypertensive medications, including angiotensin antagonists, beta blockers and diuretics, showed no significant associations with the incidence of depression or symptoms of depression (Figures 2B–D). The pooled ORs and 95% CI of depression were 1.09 (95% CI 0.96–1.25) for angiotensin antagonists, 1.18 (95% CI 0.99–1.41) for beta blockers and 1.03 (95% CI 0.95–1.10) for diuretics, respectively.

Given the obvious heterogeneity shown in the forest plots, the sensitivity analysis was conducted by omitting one study at a time (Supplementary Figure S1). In line with the results earlier mentioned, associations between angiotensin antagonists and diuretics use and depression were little, and the result was not excessively influenced by any single study (Supplementary Figures S1A,B). After removing the study by Simonson et al., a significant association between use of beta blockers and depression was observed (pooled OR: 1.26; 95% CI, 1.12–1.42) with the heterogeneity (I ² value) decreasing from 85 to 40% (Supplementary Figure S1C). When the study by Cao et al. was excluded, no association was found between calcium channel blockers usage and depression (Supplementary Figure S1D).

To evaluate publication bias, funnel plots were constructed. No publication bias was found in the beta blockers group (Supplementary Figure S2A), while apparent biases were found in the other three groups (Supplementary Figures S2B–D).

A total of five studies, involving 263,025 participants, were included in the network meta-analysis. The network plot depicts the direct comparison between the treatment groups (Supplementary Figure S3). With reference to diuretics, except for the NoAntiHTN group, all antihypertensive treatments were significantly associated with higher risk of depression: beta blockers (OR 1.53, 95% CI 1.22–1.91), calcium channel blockers (OR 1.40, 95% CI 1.12–1.75) and angiotensin antagonists (OR 1.30, 95% CI 1.04–1.63) (Figure 5). Using NoAntiHTN group as comparison group, the odds ratios of network meta-analysis results were as follows: beta blockers 1.25 (95% CI 0.96–1.61), calcium channel blockers 1.14 (95% CI 0.88–1.48), angiotensin antagonists 1.06 (95% CI 0.81–1.39) and diuretics 0.82 (95% CI 0.60–1.11) (Supplementary Figure S4). The ranking P-score based on network meta-analysis was 0.056 for beta blockers, 0.305 for calcium channel blockers, 0.529 for angiotensin antagonists, 0.638 for NoAntiHTN, and 0.972 for diuretics (Supplementary Table S2).

There was moderate heterogeneity in the whole network meta-analysis (p = 0.011), and the between-designs Q statistic supported global consistency (p = 0.614). Local inconsistency in network meta-analysis is shown in Supplementary Table S3 and Supplementary Figure S5. The comparison-adjusted funnel plot analysis demonstrated no publication bias (Supplementary Figure S6). Taken together, these results demonstrate the reliability of the net meta-analysis.