Psychiatric medications and the risk of autoimmune and immune-mediated inflammatory diseases: A systematic review and meta-analysis of observational studies

Background Pharmacovigilance reports have suggested that certain commonly used medications may trigger autoimmune diseases (ADs) and immune-mediated inflammatory diseases (IMIDs). We systematically reviewed the literature to evaluate whether psychiatric medication use is associated with ADs and IMIDs. Methods The protocol was registered in PROSPERO (CRD42022296524) before the start of the study. We searched Medline Ovid and Scopus up to November 28th, 2021, for comparative studies, with any psychiatric medication as exposure and ADs and IMIDs as outcomes. Meta-analysis was performed using DerSimonian-Laird random-effects modeling. The PRISMA 2020 guidelines were followed in reporting. Study-level risk of bias was assessed using the Newcastle-Ottawa Scale, and the overall certainty of evidence using GRADE. Results There were 7,265 citations from which 31 studies were eligible, all from high-income countries, covering 15 distinct immune diseases. The evidence for the association between selective serotonin reuptake inhibitor (SSRI) use and higher risk of microscopic colitis (meta-OR 2.60, 95% CI 1.05–6.39, I2 97.5%, 6 studies) was of low certainty. A subgroup analysis by the histological type of microscopic colitis showed a statistically significant association only with lymphocytic colitis (meta-OR 2.88, 95% CI 2.60–3.18, I2 00.00%, three studies). In two case-control studies, SSRI use had no significant association with psoriasis (meta-OR 0.80, 95% CI 0.58–1.10, I2 82.4%). The risk of acute pancreatitis was slightly increased with exposure to SSRIs (meta-OR 1.13, 95% CI 1.01–1.26, I2 00.0%), as was the risk of bullous pemphigoid after exposure to antipsychotics (meta-OR 1.79, 95% CI 1.17–2.73, I2 0%). Conclusions We reviewed the literature on whether psychiatric medications associate with the risk of ADs and IMIDs and concluded that, despite several signals, the credibility of evidence remains low at best. Prospective cohort studies would be needed as the next step to confirm the suggestions of increased risk.


Introduction
The causes of most autoimmune diseases (ADs) and immune-mediated inflammatory disorders (IMIDs) are unknown. The etiology and pathophysiology of most of them are probably unrelated, yet some share common genetic and environmental risk factors. The common denominator of these diseases is the dysregulation of immune pathways, leading to mistargeted or excessively activated responses, manifesting in subsequent organ-specific or multisystem damage [1]. Globally, the incidence of many of these diseases is increasing, including celiac disease, rheumatoid arthritis, inflammatory bowel diseases, and type 1 diabetes mellitus [2]. Numerous case reports, case series, and some pharmacovigilance risk signals suggest that certain commonly prescribed medications could trigger immune diseases [3]. Additionally, some observational studies suggest that exposure to antidepressants may be associated with a beneficial course in certain inflammatory diseases [4].
The most widely documented drug-induced autoimmune phenomenon is drug-induced lupus (DIL). A range of medications has been linked with it, including psychiatric drugs such as carbamazepine and levomepromazine [5]. Drug-induced immune thrombocytopenia characteristically occurs within two weeks after initiation of the causative medication and reoccurs rapidly after subsequent exposures [6]. Among inflammatory diseases without autoimmune mechanisms, the United European Gastroenterology organization and the European Microscopic Colitis Group have both stated that the use of selective serotonin reuptake inhibitors (SSRIs) is associated with an increased risk of developing an inflammatory colonic disease called microscopic colitis (MC) [3].
Animal models have associated mental stress with a proinflammatory state [7]. In humans, there is a relationship between having mental disorders and higher levels of proinflammatory cytokines (e.g., TNF-alpha, interleukin-6) that can activate the inflammatory response system [7]. This relationship appears bidirectional, as higher levels of these cytokines also increase the risk of depression [8].
We aimed to assess whether reports from case series could be supported by epidemiological evidence from comparative observational studies. Due to uncertainty in the etiology of the range of these diseases and finding no biological rationale to restrict the search to only a few of these diseases, we conducted a search of all major ADs and IMIDs covered in the literature.

Methods
The systematic review protocol was registered on PROSPERO (CRD42022296524) before commencing the study. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement was used to guide the reporting of the process [9]. Authors IL and ON searched MEDLINE Ovid and Scopus from inception until 28 November 2021 (non-duplicate process). After this initial search of titles and abstracts, full-texts were read independently and in parallel by IL and ON. Reference lists of eligible articles were browsed for potentially missed studies. Studies were eligible if they satisfied the following inclusion criteria, which are demonstrated under the PICO model for cohort studies in Table 1. The full search strategy and PRISMA checklist are available in the supplement. In short, we listed as exposures all psychiatric medications that have been widely used. We included studies having any immune diseases as outcomes after exposure to psychiatric medications. Primarily allergic immune diseases, immune deficiencies, or diseases with an infectious etiology such as viral hepatitis, were not included as outcomes.
We considered original articles published in any language but excluded conference abstracts and reviews. We managed the citations using the RefWorks legacy edition. All authors participated in the data collection and risk-of-bias assessments, which were performed in duplicate by two authors using a Microsoft Excel sheet. The collected data were checked and synthesized by one author (ON). Data items included basic study details, which diseases were studied as outcomes, which medications were studied as exposures, drug exposure definitions, study country and continent, age and gender distributions, years of recruitment, study design (case-control or cohort), and whether the study was prospective or retrospective. Studies were excluded, if an effect measure was not reported and it was not possible to calculate it based on the reported data. Individual studies were assessed with the Newcastle-Ottawa Scale (NOS) in three domains: selection, comparability, and outcome ascertainment [10]. The NOS score ranges from 0 to 9, where 9 indicates the lowest risk of bias. NOS scores are shown in the supplement, including which studies had a sufficient adjustment for confounders. To evaluate the overall strength of evidence of each meta-analysis, we applied the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) procedure [11].

Statistical analysis
The summary measure for binary outcomes is the odds ratio (OR) in case-control studies, the risk ratio (RR) in cohort studies, and the hazard ratio (HR) in cohort studies with time-toevent data. RStudio (version 2022.02.2) and the meta package with metagen and forest.meta functions were used to perform random-effects meta-analyses (DerSimonian and Laird) and create forest plots. Between-study heterogeneity in estimates is designated by the Tau 2 and I 2 statistics, and in the case of the latter, 0-40% indicates non-important heterogeneity, and 75-100% indicates very important heterogeneity [12,13]. We used the harvest plot to visualize the results when it was not possible to conduct a meta-analysis [14]. Reanalyzing the data using different statistical approach was used for the sensitivity analysis in the current study.

Table 1. Search question in terms of PICO (Population, intervention, comparison, and outcome).
Population (P) Human subjects who were exposed to any medication used to treat psychiatric disorders as indexed in the WHO ATC (S4 File). The exception was benzodiazepines, which were not considered. We accepted the drug exposure period as defined by the investigators. Exposure to the medication should have occurred prior to the immune disease diagnosis, and studies with an unclear sequence of events were not considered.
Intervention (I) We included only observational studies. Because immune diseases are rare, we did not expect to find interventional studies on these relationships.

Comparison (C)
The observational study should include a comparison group of individuals who have not been exposed to or have been exposed to a lesser degree (e.g., regular vs. non-regular use) to the medication. Case series were not included due to the lack of a comparison group.

Outcome (O)
The outcome is the diagnosis of any AD or IMID as has been listed in the search strategy (Supplement). Inflammatory conditions with a clear infectious etiology, such as viral hepatitis, were not included. https://doi.org/10.1371/journal.pone.0281979.t001

Microscopic colitis and its subtypes
Seven case-control studies investigated the odds of MC (or CC and LC) by exposure to SSRIs (any) or sertraline [16]. The overall association between SSRIs and MC had a very high heterogeneity with an OR of 2.60 (95% CI 1.06-6.39, I 2 97.45%, 6 studies, at least 1,322 exposed cases). The studies by Fernández-Bañares and Pascua had the highest risk of bias (NOS scores 5 and 6, respectively) [17,19]. In a sensitivity analysis of four studies with the lowest risk of bias, heterogeneity remained high while the meta-OR was only marginally lower, 2.26 (95% CI

PLOS ONE
Psychiatric medications and the risk of autoimmune and inflammatory diseases 1.37-3.73, I 2 92.01%, 4 studies) than in the analysis of all SSRI studies. All studies except Weimers et al. reported the numbers of individuals in the analysis [21]. A meta-analysis by histologic subtype of MC showed a statistically significant association between SSRIs and LC (meta-OR 2.88, 95% CI 2.60-3.18, I 2 00.00%, 3 studies) but not between SSRIs and CC (meta-OR 2.32, 95% CI 0.55-9.81, I 2 95.53%, 3 studies). Based on GRADE, the certainty of the evidence was low, with considerably high heterogeneity.

Fig 2. Forest plots for available meta-analyses of the risk of immune-mediated inflammatory diseases by the use of selective serotonin reuptake inhibitors (SSRI) and antipsychotics.
https://doi.org/10.1371/journal.pone.0281979.g002

Psoriasis
Psoriasis was investigated by Brauchli et al. in a case-control study based on the nationwide GPRD patient register in the UK and by Tzeng et al. in a cohort study based on the nationwide health insurance registry in Taiwan [37,38]. Both studies were considered to have a low risk of bias. The Taiwanese study found a lower psoriasis risk among the users of any kind of antidepressants analyzed together, and also when analyzed by the main antidepressant classes (SNRI, SSRI, TCA) (Fig 3). The incidence of psoriasis was 2.80 (95% CI 2.61-3.00) per 1,000 personyears among users of any antidepressants and 3.82 (95% CI 3.15-4.59) per 1,000 person-years among non-users. A meta-analysis was possible for SSRI use and psoriasis risk, and the difference was not statistically significant (meta-OR 0.80, 95% CI 0.58-1.10, I 2 82.39%, 2 studies). In individual studies, only lithium and phenothiazines were associated with a slightly increased psoriasis risk (Fig 3). All associations for psoriasis were considered to be of low certainty based on GRADE.

Pancreatitis
AP was the outcome in six case-control studies, all representing a low risk of bias according to NOS (score 9 for all) [22][23][24][25][26][27]. For reasonably similar medication-outcome pairs, a meta-analysis was possible for only two studies on the current use of SSRIs and acute pancreatitis (meta-OR 1.13, 95% CI 1.01-1.26, I 2 00.00%, 649 SSRI-exposed patients) with adjustment for major confounding factors, including alcohol-related diseases. Current SNRI use was associated with a marginally increased pancreatitis risk after the adjustment of relevant confounders (aOR 1.2, 95% CI 1.0-1.4) [25]. The estimates for both SSRIs and SNRIs and the development of pancreatitis represent low-certainty evidence.
Gasse et al. were the only ones to investigate the potency of conventional antipsychotics in blocking dopamine receptors with pancreatitis as the outcome [23]. The current use of preparations with high potency in blocking dopamine receptors was not associated with a statistically significant pancreatitis risk (aOR 1.2, 95% CI 0.7-2.0), whereas the use of intermediatepotency preparations (aOR 1.5, 95% CI 1.0-2.2) and low-potency preparations (aOR 2.8, 95% CI 2.0-3.8) were associated nearly 1.5 and three-fold odds, respectively. Neither the current use of atypical antipsychotics (OR 0.6, 95% CI 0.3-1.1) nor the previous use of clozapine or olanzapine (Boden 2012) had a statistically significant association with pancreatitis.
The current use of valproic acid increased the risk of pancreatitis in an analysis adjusted for, among other factors, alcohol-related diseases and gallstone disease (aOR 1.9, 95% CI 1.1-7.0). However, this estimate was based on only 16 valproic acid-exposed pancreatitis patients and thus constitutes very low-certainty evidence. Roberts et al. was the only cohort study on the risk of systemic lupus erythematosus (SLE), with a high risk of bias (NOS 5) [33]. They found similarly increased risks of SLE with the use of SSRIs (HR 2.63, 95% CI 1.66-4.14) and other antidepressants (HR 2.85, 95% CI 1.53-5.31) in an analysis adjusted for BMI, smoking, and oral contraceptive and postmenopausal hormone use. Schoonen et al. investigated both SLE and SCLE in a case-control study with a low risk of bias (NOS 9) [34]. Carbamazepine use increased the odds (aOR 1.88, 95% CI 1.09-3.22). Moreover, in a subgroup analysis on the number of prescriptions, the risk increase was related to longer use of carbamazepine (i.e., those with seven or more prescriptions) (aOR 3.04, 95% CI 1.34-6.85). The risk of SCLE was studied by Grönhagen et al., but the analysis involved only eight cases exposed to any antidepressants (OR 1.2, 95% CI 0.5-2.6) [32].

Other diseases
Two cohort studies assessed the risks of alopecia areata after exposure to any antidepressants with contrasting results. Vallerand

Discussion
We systematically searched an extensive list of ADs and IMIDs and evaluated whether the use of psychiatric medications can be considered a risk factor for their development. Published associations between any psychiatric medications and immune diseases were considered to represent low to very low certainty of evidence at best. This means that the available literature cannot reliably confirm or refute whether associations really exist, and the true effect sizes might be markedly different from the available estimates.
In a meta-analysis, we found that SSRIs are associated with a 2-3-fold increase in the odds of developing MC. Furthermore, in a subgroup analysis by histological type, this risk appeared to be related to the development of LC rather than CC. However, considering the large imprecision and overall uncertainty, these analyses represent low-certainty evidence, and the estimate is, therefore, likely to change with further research. Serotonin is an important neurotransmitter in the gut, and animal models have shown SSRIs to increase the release of serotonin in the gut (Minami 2003), and that serotonin is involved in the inflammatory reaction of experimental colitis (Ghia 2009). Although common adverse effects of SSRIs include stomach complaints such as diarrhea, it remains uncertain whether SSRIs could trigger chronic inflammation manifesting in diarrhea. Hundreds of medications have been associated with chronic diarrhea, mainly in case reports and uncontrolled case series [47]. In controlled observational studies, however, proton-pump inhibitors (especially lansoprazole) have been linked with moderate certainty of the development of a chronic inflammatory diarrhea disease, namely MC and its subtype CC [48].
There was a non-statistically significant suggestion in two case-control studies that SSRI use would associate with a lower risk of psoriasis. All of the available outcome diseases are rare, and even in hypothetical causal cases, the absolute number of new diseases would be low. For example, using the incidence figures by Tzeng et al. 2021 to illustrate the scale of these risk magnitudes, and assuming the incidences were true, would yield a number needed to treat of 980 (95% CI 656-1936) to prevent the development of one psoriasis case with any antidepressant treatment [38]. SSRIs have been shown to have some anti-inflammatory effects, and a case-control study from Sweden found that patients with psoriasis who used SSRIs had less often need for systemic treatments [49]. Although there are findings of SSRIs having antiinflammatory effects, how this would translate into changes in both psoriasis severity and incidence requires more epidemiological research preferably with a cohort study design.
BP is the most common bullous autoimmune disease affecting the subepidermal layer of the skin and mainly affects elderly people. Our meta-analysis of three studies found a 1.8-fold increase in the odds of BP after exposure to antipsychotics, representing low certainty of evidence due to these studies' relatively high risk of bias. Such a modest effect size could be explained by confounding and biases, and larger prospective studies would be needed before antipsychotics could be considered risk factors for BP. The case-control study by Varpuluoma et al. (2019) reported only statistically significant associations, therefore risking reporting bias [31]. They reported 21 psychiatric medications (and additionally seven benzodiazepines, which we did not include in this review) that increase the risk of BP. We consider it biologically implausible that so many psychiatric medications spanning different classes could cause BP. These associations, therefore, most likely reflect systematic bias due to the selection of basal cell carcinoma patients as controls instead of a random sample from the general population. Such distortion would be expected to occur with any exposure associated with a lower risk of basal cell carcinoma but not related to BP, thus appearing in a case-control design as an increase in the odds of BP [50].

Limitations
The limitations of this study mostly reflect limitations in the available literature and especially how results have been reported. Although some individual studies reported the number of prescriptions, the literature was too heterogenous and scarce for a meta-analysis on dose dependence.
Patients with immune diseases generally have higher than expected rates of psychiatric disorders, that are often treated with medications [51]. In studies on the associations between psychiatric medications and immune diseases, protopathic bias would occur if the exposure medication was used to manage the symptoms due to the yet unidentified AD or IMID. Because of the highly somatic manifestation of identified target diseases, we consider in the case of psychiatric medications, the possibility of protopathic bias to be less important than other major sources of bias, although we had no lag-time analyses available to support this assumption. In pharmacoepidemiological studies, confounding by indication is generally an important source of bias. However, it is probably less of an issue with our research question, as psychiatric medications are not used to manage the target diseases.
Assessing drug exposure independently of outcome data using nationwide registers eliminates recall bias and diminishes the impact of selection bias. Other approaches would introduce more uncertainty. For example, Pascua et al. (2010) had only a modest inter-rater agreement (kappa statistic 0.50) when patient-reported medications were compared to those found in electronic databases [19].

Conclusions
Currently, in the case of psychiatric medications, the entire literature of comparative studies represents, at best, low certainty of evidence linking them to any AD or IMID. We have reviewed important sources of bias in the field for future studies to be more reliable. Identified associations, particularly that of SSRIs and the risk of MC, should be investigated in prospective cohort studies.