The incidence risk of breast and gynecological cancer by antidepressant use: A systematic review and dose–response meta-analysis of epidemiological studies involving 160,727 patients

Background and objective Antidepressants are widely prescribed to treat depression and anxiety disorders that may become chronic conditions among women. Epidemiological studies have yielded inconsistent results on the correlation between antidepressant use and the incidence risk of female breast and gynecological cancer, along with uncertain dose–response relationship. Therefore, we performed a systematic review and dose–response meta-analysis to investigate the association. Methods Web of Science, Embase, PubMed, The Cochrane Library, and PsycINFO were systematically searched in January 2022, with no language limits. Random-effect models were used to calculate pooled effect sizes and 95% confidence intervals between studies. Linear and non-linear dose–response analyses were performed to evaluate the dose or duration of antidepressant use affecting the incidence risk of female breast and gynecological cancer. Further subgroup analyses were systematically performed by stratifying almost all study characteristics and important potential confounders, in order to further clarify and validate the important potential hypotheses regarding the biological mechanism underlying this association. Results Based on a systematic literature search, 34 eligible studies (27 case–control studies and 7 cohort studies) involving 160,727 female breast and gynecological cancer patients found that antidepressant use did not increase the incidence risk of female breast and gynecological cancer (pooled OR: 1.01; 95% CI: 0.97, 1.04, I² = 71.5%, p < 0.001), and even decreased the incidence risk of ovarian cancer (pooled OR: 0.91; 95% CI: 0.83, 1, I² = 17.4%, p = 0.293). There were a non-linear dose–response relationship (p non-linearity < 0.05) between the duration of antidepressant use and incidence risk of female breast cancer, and an inverse linear dose–response relationship between antidepressant use and the incidence risk of gynecological cancer, specifically with an increase of cumulative defined daily dose or duration to a high level, like 25,550 doses (OR: 0.91, 95% CI: 0.85–0.98, p linearity < 0.05) or 4,380 days (OR: 0.82; 95% CI: 0.7, 0.96, p linearity < 0.05), compared to never antidepressant users. Conclusion This systematic review and dose–response meta-analysis found that antidepressant use did not increase the incidence risk of female breast and gynecological cancer and even decreased the incidence risk of ovarian cancer, along with a non-linear or linear dose–response relationship. Systematic Review Registration PROSPERO https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=313364, identifier CRD42022313364.


Introduction
Female breast and gynecological cancers account for 18.7% of all reported new cancer cases in all cancer cases worldwide in 2020, representing 38.9% new incidence and approximately 27.9% mortality (1) in women. Current projections indicate that, by 2070, the worldwide number of new breast and cervix cases diagnosed reach 5.03 million annually (2).
Prevalence of antidepressant (AD) use has been rising for decades (3,4), in parallel with increased diagnosis of mental disorders, expanding indications for use, and longer treatment duration (5)(6)(7). Furthermore, women use AD at a rate approximately twofold higher than men (6), as the incidence of depression is 70% higher among women (8). Approximately 23% of women have an episode of depression throughout life and high recurrences (9,10).
Recently, preclinical in vivo studies from the US Food and Drug Administration (FDA) have found that 63.6% (7/11) of examined ADs were associated with carcinogenicity (11). Actually, since the early 1990s, several studies, in both tumor cell cultures and animal models, have raised a possible association between AD and cancer risk (12). Several potential hypotheses have been proposed regarding the biological mechanism underlying this association. First, the specific tricyclic ADs have been found to be genotoxic and carcinogenic by using somatic mutation and recombination test (SMART) in wing cells of Drosophila melanogaster because of the nitrogen atom at position 5 in the seven-membered ring of the tricyclic molecule (13). Second, fluoxetine, a selective serotonin reuptake inhibitor (SSRI), and amitriptyline, a tricyclic AD (TCA), have been shown to promote the growth of mammary tumors in animal models by binding to growth regulatory intracellular histamine receptors and perturbing cellular growth and differentiation (12,14). Third, the more important finding is that the growth and normal function of these organs are controlled by gonadotropin and female sex hormones, and it has been postulated that these hormones have an important role in the development of cancer of these organs. Certain monoaminergic medications may play a role in breast and gynecological carcinogenesis by affecting the release of gonadotropins and prolactin (15)(16)(17)(18), as the human cervix contains functional gonadotropin receptors as do other parts of female genital tract (19). Both estrogens and prolactin have been shown to increase the incidence of spontaneous and chemically induced mammary tumors in rodents (20,21). Raised endogenous estrogen levels and exogenous estrogen use are associated with the risk of breast and endometrial cancer in humans (22,23). In addition, higher plasma prolactin levels are associated with breast cancer risk in a prospective study (24). Furthermore, prolactin, as a tumor-promoter, has been shown to stimulate proliferative activity in the mammary gland, suppress apoptosis (normal process of cell selfdestruction), and upregulate the BRCA1 (breast cancer 1) gene (25,26). Furthermore, the results of subsequent epidemiologic studies suggest that AD prescriptions are associated with increased risk of hormone-related cancer, including breast, endometrial, ovarian cancer (relative risk [RR] : 1.8; 95% CI: 1.15-2.81) (27) and cervical cancer (odds ratio [OR]: 1.22, 95% CI : 1.03-1.43) (28). However, this relationship has not been validated subsequently by in vivo or in vitro studies (29,30). Furthermore, the antiproliferative effects of AD use have been supported by subsequent studies (31)(32)(33)(34)(35). Moreover, the results of epidemiologic studies have been inconclusive and demonstrated high (27,(36)(37)(38)(39)(40)(41)(42)(43), low (44-47), or no change in risk (28,(48)(49)(50)(51)(52)(53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63). Thus, the association of breast and gynecological cancer risk with the use of AD is highly conflicted, leaving an open question that will impede clinical practice. Given the increasing and widespread usage of AD, even a small increase in cancer risk associated with their use could translate into a large number of cancer cases at the population level. Furthermore, ADs are commonly prescribed after the diagnosis of cancer, not only for the treatment of depression but also for pain management (64). On the other hand, if the inverse association between AD use and breast and gynecological cancer risk proves to be causal, this may have major implications for the indications and prescribing of ADs. Thus, the hypothesis that ADs could promote or inhibit breast and gynecological cancer growth has important implications in terms of the etiology and treatment of breast and gynecological cancer and certainly merits further investigation.
Although a few systemic reviews have been carried out to elaborate this association, they all have limitations in some aspects. Most of them have been published mostly before 2012 (65) and have no new evidence published recently (52,66). Some of them have performed neither a formal meta-analysis (67-72) nor a non-linear or linear dose-response analysis (73), and some of them have either no subgroup analysis or only incomplete subgroup analysis (73). Some of them have only focused on single tumor sites, such as breast (73) or ovarian cancer, respectively (74). However, as the majority of gynecological cancer share morphological and molecular features and familial breast and ovarian cancers have common genetic predisposition (75), a systemic review is needed that puts all hormone-related cancers, such as endometrial (45,49), corpus uteri, and cervical cancer (28), together as an integral part of hormone-related cancer in female patients.
Therefore, in this study, we performed a comprehensive systematic review and meta-analysis, and a comprehensive subgroup analysis stratified by almost all study characteristics and important potential confounders to investigate the association between AD use and the incidence risk of female breast and gynecological cancer. Moreover, we performed a dose-response meta-analysis to evaluate the dose or duration of AD use affecting the incidence risk of female breast and gynecological cancer, and further clarify and validate the several important potential hypotheses regarding the biological mechanism underlying this association.

Methods
This meta-analysis was registered in the International Prospective Register of Systematic Reviews (PROSPERO; registration No.: CRD42022313364: https://www.crd.york.ac. uk/PROSPERO/display_record.php?RecordID=313364) and adhered to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) checklist (Supplementary Material  Table 1) and Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline for all processes (76,77).

Search strategy
We searched the following electronic databases for studies published from their inception until January 2022: Web of Science, Embase, MEDLINE (PubMed), The Cochrane Library (CENTRAL), and PsycINFO. The search strategy was implemented using combined index terms (Medical Subject Headings, and Emtree) and free-text keywords, including (e.g., neoplasms or cancer) AND (e.g., antidepressive agents or antidepress*) AND (e.g., morbidity or incidence or risk or occurrence) AND (e.g., case-control studies or cohort studies). A full description of the initial and supplementary search strategies is available in Supplementary Material Table 2. We combined search results using a bibliographic management tool (EndNote, version X9).

Selection criteria
The included studies were limited to (1) observational studies that (2) explored the relationship between AD use and the incidence risk of female breast and gynecological cancer, and (3) provided maximum adjusted risk estimates [risk ratios (RRs), odds ratios (ORs), and hazard ratios (HRs) with 95% confidence intervals (CIs)] or data allowing the calculation of the risk estimates and 95% CIs, and (4) studies were conducted in humans. Exclusion criteria were as follows: studies assessing the relationship between AD use and recurrence and mortality of treated breast and gynecological cancer; studies lacking a control group; specified population (e.g., HIV-infected patient); nonpeer-reviewed reports (e.g., oral presentations, posters, dissertations, and conference abstracts) and ongoing studies; and studies with insufficient information. Additionally, only the latest and/or complete one was used in the meta-analysis for duplicate articles.

Data extraction
Two reviewers independently extracted and cross-checked the data from the included studies. The following details were presented in this review: first author's name, year of publication, country, study design, participant number, age, type of ADs, exposure definition, exposure assessment, statistical indicators, outcome with the maximum covariateadjusted ORs, RRs, HRs and 95% CIs, and adjusted/matched factors. All of the selected articles were examined by two researchers independently in terms of quality according to a Newcastle-Ottawa Scale (NOS; range 1-9 with 1-3 indicating low quality, 4-6 indicating moderate quality, and 7-9 indicating high quality) (78). Any conflicts were handled by consensus with a senior author.

Statistical analysis
Given the lower than 10% incidence of cancer, we approximated RR and HR as OR when pooling the estimates across the studies (79). Summary ORs (along with their corresponding 95% CIs) were calculated by performing random-effect meta-analysis for the overall relationship between the AD use and incidence risk of female breast and gynecological cancer on comparisons of the ever AD use group versus the never AD use group, and forest plots were used to summarize the pooled estimates and effect sizes. The Cochrane Q test and I² statistic were employed to evaluate heterogeneity between studies, and I² > 50% was considered statistically significant. In order to examine its source, we performed subgroup analyses according to almost all relevant factors that may lead to significant heterogeneity. Moreover, we used the G-L method (80) to carry out a doseresponse meta-analysis using the levels of cumulative defined daily dose (81) (CDDD) or duration of AD use (days) and the adjusted natural log of the ORs with their SE. If the dose or duration was reported by range, we assigned the midpoint of the upper and lower boundaries in each category as the average duration or dose. If the highest category was open-ended, we considered the width of the category to be the same as that of the adjacent category. If the lowest category was open-ended, the lowest boundary was assumed to be zero. We included studies for this dose-response analysis only if they reported the distributions of cases and total persons or person-years, as well as the ORs and 95% CI with the variance estimates for at least three quantitative exposure categories. Step 1, we performed a non-linear dose-response meta-analysis by restricted cubic splines with three or four knots of the distribution, then based on the c² and p-value calculated in step 1, we determined whether a linear (p > 0.05) or non-linear (p < 0.05) doseresponse meta-analysis should be adopted.
The funnel plots and Egger's test were used to detect potential publication bias (82). To assess the stability of the results, a leave-one-out sensitivity analysis was carried out. All statistical tests were two-sided using a significance level of p < 0.05. Analyses were performed with Stata 13.1 (StataCorp LP, College Station, TX, USA).

Results
Eligible studies and study characteristics After identifying 7,875 references, 1,703 duplicate publications, 6,019 irrelevant studies, 13 reviews/case reports, 10 dissertations/commentaries/conference abstracts, and 77 other cancers were excluded after screening the titles and abstracts. The remaining 53 potentially studies were carefully read in full, and 18 of them were excluded for shorting the data of risk estimates and one was excluded for duplicate study data. Finally, 34 observational studies (27 case-control studies and 7 cohort studies) (27, 28, 36-57, 59-61, 63, 66, 83-87) were included in the present meta-analysis, including 22 female breast cancer studies, 7 ovarian cancer studies, 2 endometrial cancer studies, 1 cervical cancer study, and 2 female breast and gynecological cancer studies ( Figure 1).
A total of 160,727 patients with female breast and gynecological cancer were involved, and the cases ranged in size from 20 to 45,147 participants. Overall, the studies were published between 1995 and 2021, involving participants from North America (n = 22), Europe (n = 8), and Asia (n = 4). Drug exposure was assessed and collected by questionnaires in 16 studies and databases in 18 studies, respectively. The quality of the included studies is assessed in Supplementary Material Table 3. The mean NOS score was 7 (median, 6.5; range, [5][6][7][8], indicating that the overall quality of articles was good. Additional characteristics of the included studies are shown in Table 1.
In the light of the remarkable between-study heterogeneity, we investigated its potential sources by subgroup analyses of the study characteristics and confounders. The detailed results showed in Table 2. The results showed that the incidence risk of female breast and gynecological cancer was not statistically different across great majority strata, but the incidence risk appeared to be stronger in exposures that occurred less than 1 year prior to the index date ( Dose-response meta-analysis between the CDDD or duration of antidepressant use and incidence risk of female breast and gynecological cancer There was a significant linear dose-response association between the CDDD of AD use and incidence risk of female breast cancer in two eligible studies (66,86) involving 48,852 cases and 2,141,294 participants (p linearity < 0.05, Figure 3A). The OR kept a slightly increasing trend without breaking 1.2 until the CDDD increased to 1,520. By comparison, a non-linear doseresponse meta-analysis with 13 eligible studies (37, 38, 42, 47, 48, 55-57, 60, 61, 66, 83, 84) involving 109,215 cases and 398,024 controls (p non-linearity < 0.05, Figure 3B) showed a very slight increase in incidence risk of female breast cancer along with the increase of duration of AD use until 1,460 days (OR: 1.05; 95% CI: 1.01, 1.09), then followed a subsequent statistically significant inverse trend, specifically when an increase of duration to 6,570 days (OR: 0.9, 95% CI:0.79-1.02), compared to never AD users.

Sensitivity analysis and publication bias
Applying the leave-one-out sensitivity analysis showed that none of the eligible studies had considerable effect on the overall estimate (Supplementary Material Figure 2). No publication bias was observed based on funnel plot symmetry ( Figure 5) and results of Egger's test (p = 0.354).

Discussion
The pooled results of this systematic review and metaanalysis found that AD use did not increase the incidence risk of female breast and gynecological cancer, either by tricyclic ADs (pooled OR: 1; 95% CI: 0.97, 1.03) or by selective serotonin reuptake inhibitors (pooled OR: 0.99; 95% CI: 0.95, 1.04), and even decreased the incidence risk of ovarian cancer. Further subgroup analyses of confounders found that AD use was associated with higher risk in almost all of subgroups of unadjusted confounders; thus, the incomplete adjustment for these potential confounders was likely to bias our results toward a positive relationship. Additionally, further subgroup analyses of study characteristics found that the incidence risk of female breast and gynecological cancer was not statistically different across great majority strata, but it appeared to be stronger in exposures that occurred less than 1 year prior to the index date (1.25 for less than 1 year vs. 0.89 for more than 1 year) for ovarian cancer, as protopathic bias and reverse causation might drive the estimates towards an increased risk (88). It was also stronger in questionnaire for exposure assessment (1.02 for questionnaire vs. 0.89 for database) for ovarian cancer due to recall bias and selection bias by questionnaire for collecting information (88).
However, several potential hypotheses have been proposed regarding the biological mechanism underlying this positive association and have been supported by subsequent epidemiologic studies. However, most of them are limited for some reason, such as exposure misclassifications (27,43), reverse causation bias (27, 37-40, 42, 43), recall bias, selection bias (27,(38)(39)(40)(41), and unadjusted confounders (27,42,43). Moreover, the weight of epidemiologic evidence does not support the hypothesis that AD use increases the overall risk of breast and gynecological cancer (28,(44)(45)(46)(47)(48)(49)(50)(51)(52)(53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63), which are consistent with our finding. Although the incidence risk of ovarian cancer is stronger in genotoxic TCA (pooled OR: 1.38; 95% CI: 0.78, 2.45) than that in nongenotoxic TCA, and in dopamine/norepinephrine reuptake inhibitors (pooled OR: 1.38; 95% CI: 0.79, 2.42) than that in serotonin alone or mixed with norepinephrine reuptake inhibitors in our finding, the results are only based on one (44) or three studies (36,46,50). Nevertheless, exposure to either genotoxic or nongenotoxic TCAs is not associated with a significant increase in the incidence of female breast cancer. In addition, Ashbury et al. (48) grouped SSRIs as higher and lower inhibitors dependent on the dissociation constant (K d ) in order to accurately assess for levels of prolactin secreted by the secondary pituitary gland, and found that neither higher nor lower inhibitor of serotonin reuptake increased the risk for breast cancer, which was consistent with our result.
If AD medications work through changes in the secretion of gonadotropins and female sex hormones, the observed association may be more pronounced in premenopausal women who have functioning ovaries than that in postmenopausal women. Our results in this study support this hypothesis, but they are only based on limited two studies (40, 44) with possible bias for ovarian cancer. Conversely, prior prospective studies have linked prolactin with increased postmenopausal women breast cancer risk (89), but our results in this comprehensive study shows that these associations are weak or null as a whole. Moreover, different from several studies that have found that prolactin may encourage the development of estrogen receptor (ER)-positive tumors (89) and higher risk in in situ (52,85), our study shows reverse results and/or weak associations, but they are based on limited studies and possible bias, respectively. Additionally, previous studies have shown a significant relationship between depression and the risk of cancer incidence (90). Our subgroup analyses in this study have also found that after adjusting the depressive symptoms, OR tends to slightly decrease for female breast cancer. It cannot be ruled out that depression itself may have an impact on cancer incidence. Thus, it is possible that the positive association observed between AD use and female breast cancer risk is due to depressive symptoms rather than AD use itself. Therefore, further studies are needed to figure it out.
Our stratified analyses for specific types of ADs indicated that the results were more pronounced in serotonin and norepinephrine reuptake inhibitors (SNRIs) for female breast, ovarian, endometrial, and cervical cancer, and stronger in noradrenergic and specific serotonergic antidepressants (NaSSas) for ovarian cancer. We also found a reduced risk of female breast cancer with the use of monoamine oxidase inhibitors (MAOIs) and serotonin antagonist and reuptake inhibitors (SARIs), and a reduced risk of endometrial, corpus uteri, and cervical cancer with the use of norepinephrine and dopamine reuptake inhibitors (NDRIs) and NaSSas. Amerio Funnel plot for identifying the publication bias. S.E., standard error. The circles alone are real studies. The vertical lines represent the summary effect estimates, and the dashed lines represent pseudo-95% confidence interval limits. et al. reviewed the US Food and Drug Administration (FDA) preclinical in vivo evidence and found that 63.6% (7/11) of examined ADs were associated with carcinogenicity, including duloxetine (SNRIs), mirtazapine (NaSSas), and NDRI (bupropion), but the agents unassociated with carcinogenicity were trazodone (SARIs) and venlafaxine (SNRIs) (11). Meanwhile, several previous studies (91, 92) had reported the mechanisms on how mirtazapine (NaSSas) and MAOIs acted to inhibit tumor growth by enhancing immune function and causing neurotoxicity and repressing BHC110/LSD1, respectively. Conversely, a previous study has provided preliminary data of the possible association of trazodone (SARIs) and invasive cervical cancer (28). However, information pertaining to breast, uterine, and ovarian carcinogenesis clearly highlights that cervical cancer carcinogenesis is very different to the others, and is almost exclusively HPV driven and vastly different to that of the other organs analyzed, which is supported by the results of epidemiologic study that TCAs, MAOIs, and SSRIs are not associated with increased risk of invasive cervical cancer by Chan et al. (28). Furthermore, our results are based on limited one or two studies for each cancer site and short of adjusting the HPV infection as the critical potential confounders (28), and the incomplete adjustment for this potential confounder is likely to bias final results in an epidemiologic study. Thus, the above results are based on preliminary analysis and has so far proved inconclusive. Further large-scale prospective cohort studies adjusting for all possible confounding factors (including HPV infection) or animal and in vitro studies are needed to clarify the tumor-inhibiting or growth-promoting effect by different types of ADs and the biological mechanism underlying this association.
Furthermore, there is a non-linear dose-response relationship between the duration of AD use and incidence risk of female breast cancer, in which the bi-phasic phenomenon is characterized by "low-dose stimulation and high-dose inhibition" (12, 93) of malignant cell proliferation. This biphasic phenomenon shows that short-term use and/or lowdose AD may increase the risk of breast cancer in women or exacerbate cancer cell growth in women in the early stages of breast cancer, and long-term use and/or high dose inhibit tumor growth, which may help explain the high between-study heterogeneity for female breast cancer based on distinct dose or duration from different studies. Nevertheless, there is a positive linearity association between the CDDD of AD use and incidence risk of female breast cancer based on limited studies with possible bias (86). Additionally, an inverse linear dose-response association exists between the CDDD or duration of AD use and incidence risk of gynecological cancer, as well as between the CDDD and the incidence risk of endometrial, corpus uteri, and cervical cancer. Furthermore, the antiproliferative effects of AD use have been supported by previous studies (34, 44, 45) with a possible biological mechanism (31-33). However, this negative linearity association does not exist in individual cancer sites partly due to the limited numbers of studies and the strong non-linearity or linearity phenomenon that only happens with an increase of CDDD or duration to a high level. Thus, further large-scale prospective cohort studies specifying dose or duration are needed to accurately assess and clarify the protective or biphasic effect and biological mechanism.

Strength and limitation
Notably, our meta-analysis has the following advantages. First, to the best of our knowledge, this is the first study to systematically perform a qualitative dose-response metaanalysis of the relationship between AD use and the incidence of female breast and gynecological cancer as a whole of hormone-related cancer in female patients, along with comprehensive subgroup analyses stratified by almost all study characteristics and important potential confounders. Furthermore, this is the first study to demonstrate shapes of non-linear or linear association between the CDDD or duration of AD use and female breast or gynecological cancer, and further clarify and validate the several important potential hypotheses regarding the biological mechanism underlying this association.
Inevitably, this study also has some limitations. First, moderate or high heterogeneity among studies was observed when pooling estimates for female breast cancer. Furthermore, a small portion of subgroup analyses of study characteristics and important potential confounders was based on a limited number of existing studies. Second, a linearity association was not observed in the relation between the CDDD or duration of AD use and the incidence risk of ovarian, endometrial, corpus uteri, and cervical cancer, respectively, partly due to the limited number of existing studies for each cancer site. Finally, the interpretation criteria for exposure were inconsistent, and misclassification bias might affect the results.

Conclusion
Overall, the results of the present updated meta-analysis involving the largest sample size to date and mostly included comprehensive observational studies show that AD use does not increase the incidence risk of female breast and gynecological cancer, either by TCA or by SSRI, and even decreases the incidence risk of ovarian cancer, compared to never AD users. There is a non-linear dose-response relationship between the duration of AD use and incidence risk of female breast cancers, with a very slight increase in incidence risk of female breast cancer on short-term usage. An inverse linearity association exists between the CDDD or duration of AD use and incidence risk of gynecological cancer, and also between the CDDD of AD use and incidence risk of endometrial, corpus uteri, and cervical cancer. More future studies specifying dose or duration are needed in order to accurately assess and clarify this protective or bi-phasic effect and biological mechanism.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.