Skip to main content
Download PDF
  • Research article
  • Open access
  • Published:

SSRI use during acute COVID-19 and risk of long COVID among patients with depression

BMC Medicine volume 22, Article number: 445 (2024) Cite this article

Abstract

Background

Long COVID, also known as post-acute sequelae of COVID-19 (PASC), is a poorly understood condition with symptoms across a range of biological domains that often have debilitating consequences. Some have recently suggested that lingering SARS-CoV-2 virus particles in the gut may impede serotonin production and that low serotonin may drive many Long COVID symptoms across a range of biological systems. Therefore, selective serotonin reuptake inhibitors (SSRIs), which increase synaptic serotonin availability, may be used to prevent or treat Long COVID. SSRIs are commonly prescribed for depression, therefore restricting a study sample to only include patients with depression can reduce the concern of confounding by indication.

Methods

In an observational sample of electronic health records from patients in the National COVID Cohort Collaborative (N3C) with a COVID-19 diagnosis between September 1, 2021, and December 1, 2022, and a comorbid depressive disorder, the leading indication for SSRI use, we evaluated the relationship between SSRI use during acute COVID-19 and subsequent 12-month risk of Long COVID (defined by ICD-10 code U09.9). We defined SSRI use as a prescription for SSRI medication beginning at least 30 days before acute COVID-19 and not ending before SARS-CoV-2 infection. To minimize bias, we estimated relationships using nonparametric targeted maximum likelihood estimation to aggressively adjust for high-dimensional covariates.

Results

We analyzed a sample (n = 302,626) of patients with a diagnosis of a depressive condition before COVID-19 diagnosis, where 100,803 (33%) were using an SSRI. We found that SSRI users had a significantly lower risk of Long COVID compared to nonusers (adjusted causal relative risk 0.92, 95% CI (0.86, 0.99)) and we found a similar relationship comparing new SSRI users (first SSRI prescription 1 to 4 months before acute COVID-19 with no prior history of SSRI use) to nonusers (adjusted causal relative risk 0.89, 95% CI (0.80, 0.98)).

Conclusions

These findings suggest that SSRI use during acute COVID-19 may be protective against Long COVID, supporting the hypothesis that serotonin may be a key mechanistic biomarker of Long COVID.

Peer Review reports

Background

SARS-CoV-2 infection can have debilitating long-term consequences. Long COVID, also known as post-acute sequelae of COVID-19 (PASC), includes symptoms across a range of biological systems that can occur following SARS-CoV-2 infection. Millions of adults in the United States may be experiencing Long COVID, the majority of whom only experienced mild to moderate COVID-19 [1, 2]. Even though more than 10% of COVID-19 patients develop Long COVID, we have few insights regarding options for treatment and prevention [3]. Insights regarding treatments that may prevent Long COVID are crucial to preventing this condition and understanding its etiology.

Investigators have hypothesized several biological mechanisms that drive Long COVID and lead to clusters of symptoms. These explanations include [1] persistent COVID-19 viral load, [2] chronic hyperinflammation, [3] platelet and coagulation issues, and [4] central nervous system dysfunction [4, 5]. Previous studies have clustered these symptoms and speculated that these pathways may be distinct disorders caused by different components of acute COVID-19 [6]. On the other hand, recent investigations have highlighted reduced serotonin as a driver of all four of these symptom clusters [4]. A metabolomics investigation found that persistent COVID-19 viral load led to sustained interferon response, decreased tryptophan (a serotonin precursor) uptake, hypercoagulation, and subsequent decrease in serotonin [4]. This peripheral serotonin deficiency leads to reduced vagus nerve activity, which subsequently contributes to decreased hippocampal activity, which can result in memory loss and cognitive dysfunction (Fig. 1) [4].

Fig. 1
figure 1

Hypothesized mechanism of the relationship between serotonin and Long COVID [4, 7]

Full size image

Selective serotonin reuptake inhibitors (SSRIs) are the first-line medication class used to treat depression. They have high tolerability and are considered safe and effective [8, 9]. SSRI’s mechanism of action is to prevent serotonin reuptake by inhibiting serotonin transporter at the presynaptic axon terminal. The prevention of this reuptake allows for a greater concentration of serotonin in the synaptic cleft that can bind to receptors [8]. Compared with other classes of antidepressants, such as tricyclic antidepressants or monoamine oxidase inhibitors, SSRIs have fewer side effects due to fewer effects on other neurotransmitters and receptors [8]. Given SSRI’s specific targeting of serotonin, it is an ideal candidate to evaluate the role of serotonin in the development of Long COVID.

Several studies have investigated the relationship between SSRI use and acute SARS-CoV-2 infection as well as Long COVID. The TOGETHER trial found that early treatment with the SSRI fluvoxamine improved COVID-19 patient recovery [10]. On the other hand, the COVID-OUT trial found that fluvoxamine treatment during acute COVID-19 did not reduce the cumulative incidence of Long COVID (1.36, 95% CI (0.78–2.34)), although this analysis included a relatively small sample size (334 patients assigned to fluvoxamine) and may have been underpowered [11]. More broadly, previous studies have found that SSRI use may reduce the probability of hospitalization or mortality due to SARS-CoV-2 infection [12, 13]. A 2022 study evaluated the relationship between SSRI use and the predicted PASC and found that SSRI use was associated with 0.75 (95% CI, 0.62, 0.90) times the risk of predicted PASC compared to non-use [14]. While this observational study provided evidence that SSRI use may be protective against Long COVID, this study used predicted PASC diagnosis (via XGBoost machine learning) as its primary endpoint, rather than actual PASC diagnosis. This predicted PASC status did not directly include SSRI use in its prediction, but it did use a myriad of other diagnoses and medications, which may be correlated with SSRI use and may have induced bias. Furthermore, the study used a general sample of SSRI users and nonusers rather than restricting to conditions that may yield SSRI use, leading to the possibility of residual confounding by indication, as a recent study found that personality and psychiatric disorders were associated with Long COVID [15]. In addition, another observational study found that patients experiencing Long COVID experienced improvement in self-reported symptoms following treatment with SSRIs [16].

Several investigators have evaluated the impact of individual types of SSRIs on Long COVID. A multi-system study of the relationship between serotonin and Long COVID hypothesized that fluoxetine may be particularly effective in preventing and treating Long COVID, and they found that treating mice with fluoxetine improved cognitive function and restored tryptophan levels [4]. Furthermore, animal models involving fruit flies have demonstrated that the specific SSRI types fluoxetine, escitalopram, citalopram, and paroxetine may differentially impact serotonin reuptake [17]. A systematic review of studies evaluating the use of fluvoxamine for COVID-19 and Long COVID suggested that baseline use of fluvoxamine may reduce the risk of Long COVID due to the drug’s sigma 1 receptor agonist activity and the role of sigma 1 receptor activity in acute SARS-CoV-2 infection [18]. Observational analyses of human electronic health record (EHR) data did not find a significant difference in the relationship between moderate to high-affinity sigma 1 receptor agonist SSRIs (fluvoxamine, fluoxetine, escitalopram, and citalopram) versus non-high affinity SSRIs (sertraline and paroxetine) in their impact on Long COVID [14].

The purpose of this study is to evaluate the impact of SSRI use during acute COVID-19 on subsequent Long COVID risk. This study evaluates a potential pharmaceutical intervention to prevent Long COVID while testing a hypothesis regarding a mechanistic pathway of Long COVID. Identifying interventions that prevent Long COVID is crucial for clinical applications as well as our understanding of underlying biological mechanisms. Nationally sampled electronic health record (EHR) databases, such as the National COVID Cohort Collaborative (N3C), provide an excellent opportunity to evaluate these hypotheses but require analytic methods that can aggressively adjust for high-dimensional confounders without making bias-inducing parametric assumptions [19,20,21,22,23,24]. While randomized controlled trials may eventually provide definitive evidence regarding the benefit of SSRI use to prevent or treat Long COVID, observational analyses, using appropriate methods that are designed to leverage the complexity, including missing data, and large sample sizes characteristic of EHR real-world data (RWD) can provide early insights regarding the relationship between SSRI use and Long COVID. Thus, to evaluate the relationship between SSRI use during COVID-19 and Long COVID cumulative incidence, we conducted an observational analysis of individuals in N3C with an acute SARS-CoV-2 infection and comorbid depression diagnosis using a machine-learning-based method targeted to reduce bias due to confounding and missing data (Targeted Machine Learning) [19, 20, 22,23,24].

Methods

Study sample, data source, and study design

Our primary study sample included individuals with a diagnosis of acute SARS-CoV-2 infection between September 1, 2021 (ensuring that all patients were eligible to be diagnosed with PASC during person-time at risk, as PASC ICD-10 code U09.9 was released October 1, 2021) and December 1, 2022, as well as a comorbid diagnosis of a depressive disorder (see concept IDs listed in Additional file 1: Supplemental Table 1) [25, 26]. This sample was drawn from patients in N3C (DUR-80D09B6), which includes 22 million patients from 83 healthcare institutions [21]. N3C provides high-dimensional, longitudinal data on these patients, which enables researchers to conduct evaluations of a wide range of factors related to Long COVID and acute COVID-19 while rigorously adjusting for factors related to medical history and sociodemographic information.

We constructed a retrospective cohort of patients in N3C who were diagnosed with a depressive disorder (depression) before their acute SARS-CoV-2 infection, and we excluded patients with a prior diagnosis of bipolar disorder. As SSRI prescription is often indicated by depression, we restricted our sample to only include those with depression to limit confounding by indication. We evaluated SSRI use (as a time-invariant, binary variable) at the time of acute COVID-19, and we assessed patients’ cumulative incidence of Long COVID (PASC) between 1 and 12 months (i.e., day 31 to 365) following acute SARS-CoV-2 infection, comparing SSRI users to nonusers. We included patients from 80 data partners (contributors of patient data) in N3C. We found that 23% of data partners did not report PASC diagnosis, and 6% of data partners did not report SSRI use in this study sample.

N3C inclusion criteria for identifying COVID-19 patients includes either [1] at last one laboratory diagnostic positive result (either PCR or antigen) or [2] a provider diagnosis (ICD-10-CM U07.1). We used the earliest of the two dates as the index date for SARS-CoV-2 infection [27].

Key covariates

Exposures

We defined the exposure of interest as a binary variable that represents SSRI use (fluoxetine, sertraline, paroxetine, citalopram, escitalopram, fluvoxamine, and vilazodone) during incident COVID-19. We defined SSRI users as patients who began using an SSRI at least 30 days before COVID-19 and continued through acute COVID-19 (binary, time-invariant), and we defined all other patients as nonusers.

Outcomes

Our outcome of interest was observed PASC diagnosis, which was defined by ICD code U09.9, between 1 and 12 months following acute SARS-CoV-2 infection [28]. We included observed PASC (U09.9) diagnosis as our outcome of interest, as it provides a standardized metric of Long COVID incidence across diagnostic settings. In contrast, using the predicted probability of PASC diagnosis (e.g., via machine learning methods) may induce bias if the predictions are generated using the same EHR data as the exposures of interest [14]. We ensured that all patients would have 12 months of follow-up by restricting to patients who were diagnosed with COVID-19 between September 1, 2021 (1 month before the implementation of ICD code U09.9) [25] and December 1, 2022, and including PASC diagnosis data within 12 months of SARS-CoV-2 infection (i.e., up to December 1, 2023). We will describe PASC (ICD code U09.9) as “Long COVID” hereafter.

Subgroups of interest

We created subgroups of individuals with specific SSRI drug type exposures for SSRIs with a sufficient sample size, which included fluoxetine, sertraline, paroxetine, citalopram, and escitalopram. Vilazodone and fluvoxamine had insufficient sample sizes and, therefore, were excluded from subgroup analyses. We constructed separate models for each SSRI of interest to assess potential effect heterogeneity. Furthermore, we conducted exploratory analyses of potential dose–response relationships by analyzing subgroups defined by SSRI dosage among fluoxetine users, given fluoxetine’s large sample size and hypothesized relationship with Long COVID [14]. Finally, to evaluate the possibility of residual bias due to history of SSRI use, we conducted a subgroup analysis comparing new SSRI users (new prescription for an SSRI between 1 and 4 months before acute COVID-19 and no prior history of SSRI prescription) to SSRI nonusers.

Confounders and other covariates

We extracted extensive medical histories from patients in N3C to adjust for a rich history of patient data and thus avoid unmeasured confounding. Our set of baseline covariates included the following: healthcare utilization rate (number of healthcare interactions pre-SARS-CoV-2 infection and healthcare interactions per month before SARS-CoV-2 infection), sex, age at acute SARS-CoV-2 infection, race/ethnicity, common data model format, region of residence, body mass index (BMI), tobacco smoking status, obesity, diabetes, chronic lung disease, heart failure, hypertension, use of systemic corticosteroids, depression severity, anxiety, antipsychotic medication use, benzodiazepine medication use, whether the patient was immunocompromised, and the number of COVID-19 vaccination doses before infection [29]. We defined a healthcare interaction as a single interaction, or cluster of interactions, with a healthcare provider that was associated with a given medical condition, diagnosis, or procedure. We included county-level socioeconomic variables that included the percent of the county with an income level below the poverty line and the county’s social deprivation index score. We also used methods that can minimize bias due to differential monitoring among patients, by including an indicator variable for whether the patient had a documented healthcare interaction between 1 and 12 months following acute SARS-CoV-2 infection (the outcome observation period). For additional covariate information, see Additional file 1: Appendix 1.

Negative control outcome

We sought to evaluate a negative control outcome to evaluate the possibility of bias. We evaluated bone fracture diagnosis between 1 and 12 months after acute COVID-19 diagnosis as a negative control outcome.

Analysis

To accomplish the goals of using nonparametric statistical methods that could adjust for rich, messy patient history and monitoring data, we applied a Targeted Learning approach, which is well-suited for this context of observational analyses of electronic health record data [19, 20, 23, 30]. Traditional parametric analyses make assumptions regarding model form and relationships between covariates, and these assumptions will inevitably be violated in this high-dimensional setting. This potential model misspecification would increase bias and the probability of type 1 error, particularly given our large sample size [19, 20, 23, 30]. On the other hand, Targeted Learning utilizes advances in machine learning and causal inference by capitalizing on the extensive data to minimize bias introduced by arbitrary modeling assumptions, which can result in improper under-adjustment of confounders. In addition, Targeted Learning methods provide robust statistical inference despite data-adaptive, machine learning methods being used to estimate the statistical relationships of interest.

Our goal was to estimate the impact of SSRI use at the time of acute COVID-19 on the probability of developing Long COVID by comparing the predicted distribution of Long COVID under universal (i.e., all patients using SSRIs) versus no use of SSRIs among our target population, patients with a depressive disorder, under a scenario of universal monitoring of patients between 1 and 12 months after acute COVID-19. Our analysis approach first used Super Learner, an ensemble machine learning algorithm, to model Long COVID status given individual covariate status (diagnoses, treatment, demographics, and other history) [31, 32]. Super Learner uses cross-validation to determine the optimal weighting of candidate algorithms to maximize a parameter of interest. Next, we used targeted maximum likelihood estimation to estimate the causal parameter of interest (the risk ratio) comparing Long COVID incidence in the exposed versus unexposed population [19, 20, 23, 30]. Targeted maximum likelihood estimation allows us to generate interpretable measures of association, such as a risk ratio while reducing bias. In addition, targeted maximum likelihood estimation is doubly robust, meaning that it guarantees consistent estimation as long as the outcome regression or propensity score is estimated consistently [19, 20, 23, 30].

As Super Learner guarantees that the ensemble will perform at least as well as the best-performing candidate learner, given sufficient sample size, we sought to include a diverse library of parametric and nonparametric candidate algorithms to ensure optimal performance [31, 32]. We included the following candidate algorithms: generalized linear models (SL.glm), Bayesian Additive Regression Trees (tmle.SL.dbarts2), Generalized Linear models net (SL.glmnet), XGBoost (SL.xgboost), Caret (SL.caret), Caret Recursive Partitioning and Regression Trees (SL.caret.rpart), K Nearest Neighbors (SL.knn), Neural Net (SL.nnet), Random Forest (SL.randomForest), and Recursive Partitioning and Regression Trees (SL.rpart) [31, 32]. We also used cross-validated (cross-fitted) targeted maximum likelihood estimation (TMLE), which avoids overfitting and adds robustness [19, 20, 23, 30].

We applied a W, A, \(\Delta\), \(\Delta Y\) data structure, where W referred to our baseline confounders and covariates of interest, A referred to our exposure of interest, \(\Delta\) referred to participant observation during the outcome period (months 1–12), and \(\Delta Y\) referred to our observed outcome. If a participant did not have a healthcare interaction during the outcome window (months 1–12 following SARS-CoV-2 infection), which could be due to lack of healthcare engagement or patient death before observation, \(\Delta\) would be equal to 0. We defined our causal parameter of interest as \(E(Y(\text{1,1})-Y(\text{0,1}))\), where \(Y(a,\Delta =1)\) is defined as the counterfactual outcome if SSRI status is set to A = a, and the person was monitored during the at-risk period (\(\Delta =1\)). We intervened on \(\Delta\) to ensure that all patients were observed (had at least one healthcare visit) during the outcome window (between 1 and 12 months following SARS-CoV-2 infection). We make the assumption that the subset of confounders that are observed for each subject was sufficient to adjust for confounding; operationally, this was done by adding new basis functions for confounders with missing values, which were indicators that the variable was observed, and imputed values for the missing covariate. This allows us to aggressively adjust for confounding and keep observations with missing covariate information (W) [24, 29].

Sensitivity analyses

In order to evaluate underlying biases in our analysis and data, we conducted a nonparametric sensitivity analysis [33]. This nonparametric sensitivity analysis allows us to compare the theoretical bias that would nullify our results to benchmarks, such as the difference between our observed adjusted estimate and unadjusted estimate, that could explain the magnitude of our observed association. Furthermore, we evaluated the relationship between SSRI use during SARS-CoV-2 infection and bone fracture between 1 and 12 months following SARS-CoV-2 infection as a negative control outcome analysis. We compared our observed, adjusted result to the [1] unadjusted association and [2] the negative control outcome association.

Results

Descriptive statistics

We analyzed EHR data from a sample of 302,626 patients who were diagnosed with a depressive disorder before COVID-19 diagnosis. Among these patients, 100,803 (33%) were using an SSRI at the time of SARS-CoV-2 infection and 201,823 (67%) were not (Table 1, see Additional file 1: Supplemental Fig. 1). We found that SSRI users generally had a greater burden of disease and more markers of poor health than SSRI nonusers. Among SSRI users, 17% were morbidly obese compared to 16% of nonusers, 14% had experienced heart failure compared to 11% of nonusers, 34% had experienced lung disease compared to 31% of nonusers, and 64% used systemic corticosteroids compared to 54% of nonusers. We found that 27% of both groups were smokers. We observed that 47% of SSRI users were diagnosed with an anxiety-related condition and 17% were prescribed benzodiazepines, while 60% of nonusers were diagnosed with an anxiety-related condition and 22% were prescribed benzodiazepines. We also found that 8% of SSRI users had severe major depressive disorder, compared with 6% of non-SSRI users. SSRI users had a healthcare utilization rate of 3.0 healthcare interactions per month, while nonusers had 2.3 healthcare interactions per month. We found that 30% of SSRI users had at least one dose of a COVID-19 vaccination, and 34% of nonusers had at least one vaccination dose.

Table 1 Patient characteristics
Full size table

Relationship between SSRI use and long COVID

We found that SSRI users had a lower risk of Long COVID (adjusted risk ratio (aRR) 0.922, 95% confidence interval (CI) (0.863, 0.986)) compared to nonusers (Fig. 2). Adjustment for baseline confounders shifted the estimate a fair distance from the unadjusted association, which failed to detect relationship between SSRI use and Long COVID (unadjusted RR 1.042, 95% CI (0.984, 1.104)) (Table 2).

Fig. 2
figure 2

Relationship between SSRI use (overall and by SSRI type) and Long COVID among patients with depression

Full size image
Table 2 Relationships between SSRI use during acute COVID-19 infection and 12-month long COVID risks
Full size table

We evaluated the relationship between individual SSRI types and Long COVID (fluoxetine, sertraline, paroxetine, escitalopram, and citalopram). In our subgroup analysis, comparing the use of each of the five SSRIs to no SSRI use, we did not detect an association between the use of fluoxetine (aRR 0.897, 95% CI (0.752, 1.071)), sertraline (aRR 0.954, 95% CI (0.849, 1.073)), escitalopram (aRR 0.912, 95% CI (0.804, 1.034)), paroxetine (aRR 0.858, 95% CI (0.581, 1.267)), or citalopram (aRR 0.949, 95% CI (0.778, 1.157)) and the risk of Long COVID, although all point estimates indicated a protective (i.e., RR < 1), albeit not significant, relationship. We did not find evidence of a dose–response relationship between fluoxetine dose and risk of Long COVID (60 mg vs. 10 mg aRR 1.421, 95% CI (0.656, 3.080)) (see Additional file 1: Supplemental Table 2).

Sensitivity analyses and confounding

We found that the relationship between SSRI use and Long COVID was strongly and qualitatively confounded, as the unadjusted estimate indicated a positive (harmful) correlation, but the adjusted estimate indicated a negative (protective) correlation. We observed the change in estimate following the backward exclusion of each covariate, where we defined “confounder RR” as the risk ratio in the fully adjusted model divided by the risk ratio of the partially adjusted model (with the covariate excluded) (see Additional file 1: Supplemental Table 3). We found that the strongest confounders of the relationship between SSRI use and Long COVID were baseline systemic corticosteroid use (confounder RR 0.983), monitoring during the outcome window (binary indicator of healthcare interactions between 1 and 12 months after acute COVID-19) (confounder RR 0.989), healthcare utilization at baseline (confounder RR 0.995), and social deprivation index (confounder RR 1.005). We also evaluated the impact of excluding two groups of covariates, healthcare utilization (number of healthcare interactions before baseline, healthcare interaction rate before baseline, and monitoring during the outcome window) and baseline general health and comorbidities general health and comorbidities (BMI, chronic lung disease, diabetes, obesity, immunocompromised status, smoking, corticosteroid use, hypertension, and COVID-19 vaccinations). We found that excluding factors related to healthcare utilization led to a confounder RR of 0.969 while excluding factors related to baseline comorbidities and health led to a confounder RR of 0.979.

In a subgroup analysis comparing new SSRI users (first SSRI prescription 1 to 4 months before acute COVID-19 with no prior history of SSRI use) to SSRI nonusers, we found a protective association similar to the primary analysis (aRR 0.886, 95% CI (0.780, 0.985)) (Table 2).

We conducted a nonparametric sensitivity analysis to evaluate the potential impact of bias on our results (Fig. 3). We found that 0.65 units of bias, where one unit corresponds to the difference between our adjusted and unadjusted estimate, could lead to a value as extreme as our observed estimate, due to random variation alone.

Fig. 3
figure 3

Nonparametric sensitivity analysis depicting the observed, adjusted risk ratio (TMLE + SL) as well as the unadjusted risk ratio (unadjusted) and the results of an analysis of a negative control outcome (bone fracture)

Full size image

Discussion

We found a protective effect of SSRI use at the time of acute SARS-CoV-2 infection on subsequent 12-month risk of Long COVID among patients with depression. These results are consistent with the hypothesis that SSRIs may be an effective intervention to prevent Long COVID, which also supports the hypothesis that serotonin may play a role in the development of Long COVID. Randomized controlled trials are currently underway to evaluate the ability of SSRIs to prevent or treat Long COVID (NCT05874037, NCT06128967). With ongoing COVID-19 transmission, the risk of Long COVID remains prevalent, and finding interventions to prevent Long COVID remains prudent. SSRIs may serve as an important tool in preventing this condition and limiting the rippling effects of the COVID-19 pandemic.

Our findings regarding the protective effect of SSRI use on Long COVID risk are consistent with previous studies. This observed treatment effect, a risk ratio of 0.922 (95% CI 0.863, 0.986), is less protective than a previous analysis, which found a risk ratio of 0.76 (95% CI 0.62, 0.90) [14]. The difference in these effects may be attributed to several potential factors, including the previous study’s use of predicted Long COVID status rather than observed diagnosis (yielding a prevalence of 15% rather than 2%) as well as our restriction to only include patients with a diagnosis of a depressive disorder [14]. These considerations may avoid bias and confounding due to indication, respectively.

The observed unadjusted and adjusted estimates varied. The unadjusted association indicating a non-significant relationship between SSRI use and Long COVID may be explained by strong confounding due to various factors and is supported by our finding of imbalance and confounding due to healthcare utilization, medication usage, and socioeconomic factors (see Additional file 1: Appendix 2 for details). Our finding that factors related to healthcare utilization rate (number of healthcare interactions before baseline and outcome monitoring indicator) were strong confounders of our observed relationship highlights the importance of addressing differential healthcare utilization rates and other causal considerations in observational studies that rely on Long COVID diagnosis as an outcome of interest [29]. We observed similar estimates in our overall analysis (aRR 0.92, 95% CI (0.86, 0.99)) and subgroup analysis comparing new SSRI users to SSRI nonusers (aRR 0.89, 95% CI (0.80, 0.98)). This finding supports that our observed associations were minimally biased by patient history of SSRI use, although the small difference in observed point estimates indicates that our observed association may be conservative (i.e., the true protective effect of SSRIs on Long COVID may be even stronger).

These findings provide support for the hypothesis that low serotonin may be a driver of Long COVID incidence and that SSRIs may prevent or treat Long COVID. This finding merits further exploration of the hypothesis of Wong et al. regarding Long COVID etiology via hypoactivity in the serotonin system [4]. This hypothesis posits that remnants of the SARS-CoV-2 virus leads to sustained release of viral RNA-induced type 1 interferons, which decreases tryptophan uptake and prevents cortisol production. According to this hypothesis, SSRI use may interrupt this causal pathway of disease etiology [4]. As this hypothesis posits that low serotonin is a downstream effect of lingering SARS-CoV-2 virus and sustained interferon 1 response, these findings also hint at interventions that aim to detect or treat persistent viral load of SARS-CoV-2 or viral RNA-induced type 1 interferon.

These findings indicate the need for several future studies to further explore these hypotheses. The impact of SSRIs on Long COVID risk in other populations, such as premenstrual dysphoric disorder or generalized anxiety disorder, may provide additional insights regarding the generalizability of these findings. In addition, future investigations should evaluate the impact of other serotonergic drugs, such as serotonin-norepinephrine reuptake inhibitors (SNRIs), on Long COVID risk. Finally, given the phenotypic overlap between Long COVID and other post-infectious chronic somatoform disorders, such as chronic Lyme’s disease (patients with both conditions frequently exhibit post-exertional malaise, chronic pain, brain fog, etc.), investigators should investigate the ability of SSRIs to treat or prevent these related conditions [34,35,36]. It should be noted that investigators have found mixed results regarding long-term effects of SSRIs on the serotonin system, with some studies indicating that long-term SSRI use may lead to decreased serotonin signaling (i.e., a negative feedback loop) [37,38,39,40,41].

We did not find evidence of heterogeneity of the relationship between SSRI use and Long COVID depending on SSRI type, which is consistent with previous findings [14]. A previous study did not find a differential impact of moderate to high-affinity sigma 1 receptor agonist SSRIs (fluvoxamine, fluoxetine, escitalopram, and citalopram) versus non-high affinity SSRIs (sertraline and paroxetine) in their relationship with Long COVID [14]. We caution readers to consider this finding in the context of a few limitations. Residual confounding due to indication, as different depressive symptomatology, comorbidities, side effects, and tolerance may lead providers to prescribe a specific SSRI over another SSRI. For instance, citalopram and paroxetine are often prescribed for obsessive–compulsive disorder, which may be associated with Long COVID symptoms [42,43,44,45]. Furthermore, our analysis of paroxetine was limited by a small sample size of users (n = 7189). We also did not find evidence of a dose–response relationship between fluoxetine and Long COVID. This may be explained due to a large proportion of missingness of dose information leading to a small functional sample size. Future studies should further explore the possibility of a dose–response relationship [9].

Strengths and limitations

A strength of this study is its large, national sample size of patients and the broad range of high-dimensional data that we included via N3C. This rich data source allows us to construct a cohort of patients with a diagnosis of a depressive condition, assess their SSRI use at the time of SARS-CoV-2 infection, and evaluate their probability of Long COVID diagnosis. Furthermore, the documentation of comorbidities, sociodemographic information, and other medical history allows for rigorous multivariate adjustment.

A second strength of this study is the analytic methods that we applied. A Targeted Learning approach, involving Super Learner and targeted maximum likelihood estimation, allows for efficient and flexible estimation while making minimal parametric assumptions [19, 20, 23, 31, 46]. With this large sample size of high-dimensional data, this allows us to aggressively reduce bias due to measured, potentially high-dimensional confounding and to do so with nearly no model assumptions. These methods allowed us to intervene on participant observation during the outcome window, which is an important driver of differential outcome ascertainment [29, 33]. Furthermore, nonparametric sensitivity analyses allowed us to determine the extent to which our results are vulnerable to bias. Cumulatively, these methods provide a replicable framework for investigators to conduct rigorous observational analyses using electronic health record databases such as N3C.

A third strength of this study was its ability to flexibly account for and intervene on the missingness of the outcome and heterogeneous monitoring [33]. There is significant heterogeneity in N3C’s documentation of Long COVID diagnoses (our outcome of interest), as is common with electronic health record databases. Previous studies have found that Long COVID diagnosis is strongly correlated with healthcare utilization rate [29, 47]. We sought to control for healthcare utilization rate by adjusting for multiple factors related to healthcare utilization, including healthcare visits per month before SARS-CoV-2 infection. In addition, we were able to use novel causal inference framing to define our parameter of interest at the ratio of probabilities of PASC under universal monitoring, i.e., by “intervening” on whether an individual had a healthcare visit between 1 and 12 months following acute SARS-CoV-2 infection (the period at-risk for Long COVID), to observe the counterfactual impact of SSRI exposure given that all patients were observed during the period at-risk for the outcome [48]. It remains possible that residual confounding due to healthcare utilization rate remains, although this would likely bias our estimate towards the null, indicating that our observed measure of association is likely conservative [29].

This study had several limitations. We defined the exposure of interest as a binary, time-invariant variable based on SSRI use during COVID-19. It remains possible that factors related to the duration of SSRI use, timing of SSRI use, or SSRI dosage may modify this relationship, although these factors are poorly documented (i.e., high missingness) in EHR databases such as N3C and should be explored in a future study. Furthermore, PASC diagnosis (ICD code U09.9) has limited sensitivity and low clinical utilization, which may lead to outcome misclassification. Furthermore, the binary definition of Long COVID may fail to reflect heterogeneity within Long COVID subtypes (e.g., neurological versus gastrointestinal symptoms). Finally, as an observational study, this analysis may be subject to residual bias and investigators should conduct randomized controlled trials to corroborate these findings. The generalizability of N3C patients has been described previously [6, 47, 49]. N3C is a broad, national sample of patients, as it relies on electronic health record data, but it skews towards patients who engage more with healthcare systems. This yields a study population that is generally older, has more comorbidities than the general population, and underrepresents un- or underinsured patients [47].

Conclusions

This study suggests that the use of SSRIs during acute COVID-19 is associated with a lower risk of Long COVID among patients with depression. These results support the hypothesis that serotonin may be a mechanistic biomarker of Long COVID and that SSRIs may be an effective intervention to prevent Long COVID.

Supplementary information.

Additional file 1: Supplemental Table 1 List of depressive conditions. Appendix 1 Covariate information. Supplemental Fig. 1 CONSORT diagram. Supplemental Table 2 Exploratory analysis of dose–response relationship between fluoxetine use and Long COVID. Supplemental Table 3 Variable importance of included covariates. Appendix 2 Nonparametric sensitivity analysis.

Availability of data and materials

All analytic code and data are available in the N3C Enclave by request. Access to the N3C Data Enclave is managed by NCATS (https://ncats.nih.gov/research/research-activities/n3c/resources/data-access). Interested researchers must first complete a data use agreement, and next a data use request, in order to access the N3C Data Enclave. Once access is granted, the N3C data use committee must review and approve all use of data and the publication committee must approve all publications involving N3C data.

Abbreviations

SSRI:

Selective serotonin reuptake inhibitor

COVID-19:

Coronavirus disease 2019

PASC:

Post-acute sequelae of COVID-19

EHR:

Electronic health record

RR:

Risk ratio

References

  1. Robertson MM, Qasmieh SA, Kulkarni SG, Teasdale CA, Jones HE, McNairy M, et al. The epidemiology of long coronavirus disease in US adults. Clin Infect Dis. 2023;76(9):1636–45.

    Article  PubMed  Google Scholar 

  2. Raveendran AV, Jayadevan R, Sashidharan S. Long COVID: an overview. Diab Metab Syndr. 2021;15(3):869–75.

    Article  CAS  Google Scholar 

  3. Davis HE, McCorkell L, Vogel JM, Topol EJ. Long COVID: major findings, mechanisms and recommendations. Nat Rev Microbiol. 2023;21(3):133–46.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  4. Wong AC, Devason AS, Umana IC, Cox TO, Dohnalová L, Litichevskiy L, et al. Serotonin reduction in post-acute sequelae of viral infection. Cell. 2023. Available from: https://doi.org/10.1016/j.cell.2023.09.013. Cited 2023 Oct 19.

  5. Belluck P. Scientists offer a new explanation for long COVID. The New York Times. 2023. Available from: https://www.nytimes.com/2023/10/16/health/long-covid-serotonin.html.

  6. Reese JT, Blau H, Casiraghi E, Bergquist T, Loomba JJ, Callahan TJ, et al. Generalisable long COVID subtypes: findings from the NIH N3C and RECOVER programmes. eBioMedicine. 2023;87:104413 Available from: https://linkinghub.elsevier.com/retrieve/pii/S2352396422005953.. Cited 2023 Oct 26

    Article  PubMed  Google Scholar 

  7. Created with Biorender.com. 2024. Available from: ‍ https://biorender.com.

  8. Chu A, Wadhwa R. Selective serotonin reuptake inhibitors. In: StatPearls. Treasure Island (FL): StatPearls Publishing; 2023. Available from: http://www.ncbi.nlm.nih.gov/books/NBK554406/. Cited 2023 Nov 29.

  9. Preskorn S. Clinical pharmacology of serotonin selective reuptake inhibitors. Caddo: Professional Communications; 1996.

  10. Reis G, Dos Santos Moreira-Silva EA, Silva DCM, Thabane L, Milagres AC, Ferreira TS, et al. Effect of early treatment with fluvoxamine on risk of emergency care and hospitalisation among patients with COVID-19: the TOGETHER randomised, platform clinical trial. Lancet Glob Health. 2022;10(1):e42–51.

    Article  CAS  PubMed  Google Scholar 

  11. Bramante CT, Buse JB, Liebovitz DM, Nicklas JM, Puskarich MA, Cohen K, et al. Outpatient treatment of COVID-19 and incidence of post-COVID-19 condition over 10 months (COVID-OUT): a multicentre, randomised, quadruple-blind, parallel-group, phase 3 trial. Lancet Infect Dis. 2023;23(10):1119–29 Available from: https://linkinghub.elsevier.com/retrieve/pii/S1473309923002992. Cited 2023 Oct 13. Cited 2023 Oct 13

    Article  PubMed  PubMed Central  Google Scholar 

  12. Foletto VS, da Rosa TF, Serafin MB, Hörner R. Selective serotonin reuptake inhibitor (SSRI) antidepressants reduce COVID-19 infection: prospects for use. Eur J Clin Pharmacol. 2022;78(10):1601–11.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Stingl JC. Antidepressant drug treatment protecting from COVID-19: one more piece in the repurposing puzzle. BJPsych Open. 2021;8(1):e20.

    Article  PubMed  Google Scholar 

  14. Sidky H, Sahner DK, Girvin AT, Hotaling N, Michael SG, Gersing K. Assessing the effect of selective serotonin reuptake inhibitors in the prevention of post-acute sequelae of COVID-19. medRxiv : the preprint server for health sciences. United States; 2023. p. 2022.11.09.22282142.

  15. Slurink IAL, van den Houdt SCM, Mertens G. Who develops long COVID? Longitudinal pre-pandemic predictors of long COVID and symptom clusters in a representative Dutch population. Int J Infect Dis. 2024;10(144):107048.

    Article  Google Scholar 

  16. Rus CP, de Vries BEK, de Vries IEJ, Nutma I, Kooij JJS. Treatment of 95 post-COVID patients with SSRIs. Sci Rep. 2023;13(1):18599 https://doi.org/10.1038/s41598-023-45072-9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  17. Dunham KE, Venton BJ. SSRI antidepressants differentially modulate serotonin reuptake and release in Drosophila. J Neurochem. 2022;162(5):404–16.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  18. Hashimoto K. Overview of the potential use of fluvoxamine for COVID-19 and long COVID. Discov Ment Health. 2023;3(1):9.

    Article  PubMed  PubMed Central  Google Scholar 

  19. van der Laan MJ, Rose S. Targeted learning: causal inference for observational and experimental data. New York: Springer; 2011. p. 626 Springer series in statistics.

    Book  Google Scholar 

  20. van der Laan M, Coyle J, Hejazi N, Malenica I, Phillips R, Hubbard A. Targeted learning in R: causal data science with the tlverse software ecosystem. 2023. Available from: https://tlverse.org/tlverse-handbook/optimal-individualized-treatment-regimes.html.

  21. National center for advancing translational sciences. N3C dashboards. 2023. Available from: https://covid.cd2h.org/dashboard/.

  22. Coyle JR, Hejazi NS, Malenica I, Phillips RV, Arnold BF, Mertens A, et al. Targeted learning. In: Wiley StatsRef: Statistics Reference Online. 2023:1–20. Available from: https://doi.org/10.1002/9781118445112.stat08414. Cited 2023 May 17.

  23. Van der Laan MJ, Rose S. Targeted learning in data science: causal inference for complex longitudinal studies. New York, NY: Springer Berlin Heidelberg; 2017.

  24. Gruber S, Lee H, Phillips R, Ho M, van der Laan M. Developing a targeted learning-based statistical analysis plan. Stat Biopharmaceut Res. 2022;23:1–8. https://doi.org/10.1080/19466315.2022.2116104.

    Article  Google Scholar 

  25. ICD10 Data. 2023 ICD-10-CM diagnosis code U09.9. ICD10data.com. 2023. Available from: https://www.icd10data.com/ICD10CM/Codes/U00-U85/U00-U49/U09-/U09.9. Cited 2023 Sep 12.

  26. Applicable Data Methods & Standards Domain Team. N3C concept set - 38249145 (depression). 2024 Jan 30. Available from: https://zenodo.org/doi/10.5281/zenodo.7685710. Cited 2024 May 14.

  27. Beasley W. Phenotype data acquisition. Github; Available from: https://github.com/National-COVID-Cohort-Collaborative/Phenotype_Data_Acquisition/wiki/Latest-Phenotype.

  28. McGrath LJ, Scott AM, Surinach A, Chambers R, Benigno M, Malhotra D. Use of the postacute sequelae of COVID-19 diagnosis code in routine clinical practice in the US. JAMA Netw Open. 2022;5(10):e2235089.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Zachary Butzin-Dozier, Yunwen Ji, Haodong Li, Jeremy Coyle, Junming (Seraphina) Shi, Rachael V. Philips, et al. Predicting long COVID in the National COVID Cohort Collaborative using super learner. medRxiv. 2023;2023.07.27.23293272. Available from: http://medrxiv.org/content/early/2023/08/04/2023.07.27.23293272.abstract.

  30. Van Der Laan M. Why we need a statistical revolution. Sense about Science USA. 2015. Available from: https://senseaboutscienceusa.org/super-learning-and-the-revolution-in-knowledge/.

  31. van der Laan MJ, Polley EC, Hubbard AE. Super learner. Stat Appl Genet Mol Biol. 2007;6:Article25.

    PubMed  Google Scholar 

  32. Phillips RV, van der Laan MJ, Lee H, Gruber S. Practical considerations for specifying a super learner. Int J Epidemiol. 2023:dyad023. Available from: https://doi.org/10.1093/ije/dyad023 . Cited 2023 Jun 15.

  33. Diaz Munoz I, van der Laan MJ. Sensitivity analysis for causal inference under unmeasured confounding and measurement error problems. Division of Biostatistics, UC Berkeley; 2012. http://www.bepress.com/ucbbiostat/paper303.

  34. Hassett AL, Radvanski DC, Buyske S, Savage SV, Sigal LH. Psychiatric comorbidity and other psychological factors in patients with “chronic Lyme disease.” Am J Med. 2009;122(9):843–50.

    Article  PubMed  PubMed Central  Google Scholar 

  35. Hassett AL, Radvanski DC, Buyske S, Savage SV, Gara M, Escobar JI, et al. Role of psychiatric comorbidity in chronic Lyme disease. Arthritis Care Res. 2008;59(12):1742–9. https://doi.org/10.1002/art.24314. Cited 2024 May 9.

    Article  Google Scholar 

  36. Liu Y, Zhao J, Fan X, Guo W. Dysfunction in serotonergic and noradrenergic systems and somatic symptoms in psychiatric disorders. Front Psychiatry. 2019;10: 286.

    Article  PubMed  PubMed Central  Google Scholar 

  37. Hensler JG, Artigas F, Bortolozzi A, Daws LC, De Deurwaerdère P, Milan L, et al. Catecholamine/serotonin interactions: systems thinking for brain function and disease. Adv Pharmacol. 2013;68:167–97.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Mansari ME, Manta S, Oosterhof C, El Iskandrani KS, Chenu F, Shim S, et al. Restoration of serotonin neuronal firing following long-term administration of bupropion but not paroxetine in olfactory bulbectomized rats. Int J Neuropsychopharmacol. 2015;18(4):pyu050. https://doi.org/10.1093/ijnp/pyu050. Cited 2024 May 14.

    Article  CAS  PubMed Central  Google Scholar 

  39. Hahn A, Lanzenberger R, Wadsak W, Spindelegger C, Moser U, Mien LK, et al. Escitalopram enhances the association of serotonin-1A autoreceptors to heteroreceptors in anxiety disorders. J Neurosci. 2010;30(43):14482–9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Vidal C, Herzog C, Haeberle AM, Bombarde C, Miquel MC, Carimalo J, et al. Early dysfunction of central 5-HT system in a murine model of bovine spongiform encephalopathy. Neuroscience. 2009;160(4):731–43.

    Article  CAS  PubMed  Google Scholar 

  41. Wielpuetz C, Kuepper Y, Grant P, Munk AJL, Hennig J. Acute responsivity of the serotonergic system to S-citalopram and positive emotionality-the moderating role of the 5-HTTLPR. Front Hum Neurosci. 2013;7:486.

    Article  PubMed  PubMed Central  Google Scholar 

  42. Montgomery SA, Kasper S, Stein DJ, Bang Hedegaard K, Lemming OM. Citalopram 20 mg, 40 mg and 60 mg are all effective and well tolerated compared with placebo in obsessive-compulsive disorder. Int Clin Psychopharmacol. 2001;16(2):75–86.

    Article  CAS  PubMed  Google Scholar 

  43. Pallanti S, Quercioli L, Koran LM. Citalopram intravenous infusion in resistant obsessive-compulsive disorder: an open trial. J Clin Psychiatry. 2002;63(9):796–801.

    Article  CAS  PubMed  Google Scholar 

  44. Kamijima K, Murasaki M, Asai M, Higuchi T, Nakajima T, Taga C, et al. Paroxetine in the treatment of obsessive-compulsive disorder: randomized, double-blind, placebo-controlled study in Japanese patients. Psychiatry Clin Neurosci. 2004;58(4):427–33.

    Article  CAS  PubMed  Google Scholar 

  45. Alkhamees AA. Obsessive-compulsive disorder post-COVID-19: a case presentation. Egypt J Neurol Psychiatr Neurosurg. 2021;57(1):150.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Schuler MS, Rose S. Targeted maximum likelihood estimation for causal inference in observational studies. Am J Epidemiol. 2017;185(1):65–73.

    Article  PubMed  Google Scholar 

  47. Pfaff ER, Girvin AT, Bennett TD, Bhatia A, Brooks IM, Deer RR, et al. Identifying who has long COVID in the USA: a machine learning approach using N3C data. Lancet Digit Health. 2022;4(7):e532–41.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Chen DG (Din), Sun J, Peace KE, editors. Interval-censored time-to-event data: methods and applications. 0 ed. Chapman and Hall/CRC; 2012. Available from: https://www.taylorfrancis.com/books/9781466504288. Cited 2024 Jan 10.

  49. Pfaff ER, Madlock-Brown C, Baratta JM, Bhatia A, Davis H, Girvin A, et al. Coding long COVID: characterizing a new disease through an ICD-10 lens. BMC Med. 2023;21(1):58.

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

N3C attribution

The analyses described in this manuscript were conducted with data or tools accessed through the NCATS N3C Data Enclave https://covid.cd2h.org and N3C Attribution & Publication Policy v 1.2-2020-08-25b supported by NCATS U24 TR002306, Axle Informatics Subcontract: NCATS-P00438-B, the Bill & Melinda Gates Foundation: OPP1165144, and the National Institutes of General Medical Sciences: U54GM115458 and 5U54GM104942-04. This research was possible because of the patients whose information is included within the data and the organizations (https://ncats.nih.gov/n3c/resources/data-contribution/data-transfer-agreement-signatories) and scientists who have contributed to the on-going development of this community resource [https://doi.org/https://doi.org/10.1093/jamia/ocaa196].

Disclaimer

The N3C publication committee confirmed that this manuscript (MSID:1784.118) is in accordance with N3C data use and attribution policies; however, this content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the N3C program.

IRB

The N3C data transfer to NCATS is performed under a Johns Hopkins University Reliance Protocol # IRB00249128 or individual site agreements with NIH. The N3C Data Enclave is managed under the authority of the NIH; information can be found at https://ncats.nih.gov/n3c/resources.

This research project was approved by the University of California, Berkeley Committee for the Protection of Human Subjects (CPHS protocol number 2022-01-14980). This approval is issued under University of California, Berkeley Federalwide Assurance #00006252.

Individual acknowledgements for core contributors

We gratefully acknowledge the following core contributors to N3C: Adam B. Wilcox, Adam M. Lee, Alexis Graves, Alfred (Jerrod) Anzalone, Amin Manna, Amit Saha, Amy Olex, Andrea Zhou, Andrew E. Williams, Andrew Southerland, Andrew T. Girvin, Anita Walden, Anjali A. Sharathkumar, Benjamin Amor, Benjamin Bates, Brian Hendricks, Brijesh Patel, Caleb Alexander, Carolyn Bramante, Cavin Ward-Caviness, Charisse Madlock-Brown, Christine Suver, Christopher Chute, Christopher Dillon, Chunlei Wu, Clare Schmitt, Cliff Takemoto, Dan Housman, Davera Gabriel, David A. Eichmann, Diego Mazzotti, Don Brown, Eilis Boudreau, Elaine Hill, Elizabeth Zampino, Emily Carlson Marti, Emily R. Pfaff, Evan French, Farrukh M. Koraishy, Federico Mariona, Fred Prior, George Sokos, Greg Martin, Harold Lehmann, Heidi Spratt, Hemalkumar Mehta, Hongfang Liu, Hythem Sidky, J. W. Awori Hayanga, Jami Pincavitch, Jaylyn Clark, Jeremy Richard Harper, Jessica Islam, Jin Ge, Joel Gagnier, Joel H. Saltz, Joel Saltz, Johanna Loomba, John Buse, Jomol Mathew, Joni L. Rutter, Julie A. McMurry, Justin Guinney, Justin Starren, Karen Crowley, Katie Rebecca Bradwell, Kellie M. Walters, Ken Wilkins, Kenneth R. Gersing, Kenrick Dwain Cato, Kimberly Murray, Kristin Kostka, Lavance Northington, Lee Allan Pyles, Leonie Misquitta, Lesley Cottrell, Lili Portilla, Mariam Deacy, Mark M. Bissell, Marshall Clark, Mary Emmett, Mary Morrison Saltz, Matvey B. Palchuk, Melissa A. Haendel, Meredith Adams, Meredith Temple-O’Connor, Michael G. Kurilla, Michele Morris, Nabeel Qureshi, Nasia Safdar, Nicole Garbarini, Noha Sharafeldin, Ofer Sadan, Patricia A. Francis, Penny Wung Burgoon, Peter Robinson, Philip R. O. Payne, Rafael Fuentes, Randeep Jawa, Rebecca Erwin-Cohen, Rena Patel, Richard A. Moffitt, Richard L. Zhu, Rishi Kamaleswaran, Robert Hurley, Robert T. Miller, Saiju Pyarajan, Sam G. Michael, Samuel Bozzette, Sandeep Mallipattu, Satyanarayana Vedula, Scott Chapman, Shawn T. O’Neil, Soko Setoguchi, Stephanie S. Hong, Steve Johnson, Tellen D. Bennett, Tiffany Callahan, Umit Topaloglu, Usman Sheikh, Valery Gordon, Vignesh Subbian, Warren A. Kibbe, Wenndy Hernandez, Will Beasley, Will Cooper, William Hillegass, Xiaohan Tanner Zhang. Details of contributions available at covid.cd2h.org/core-contributors.

Data partners with released data

The following institutions whose data is released or pending:

Available: Advocate Health Care Network — UL1TR002389: The Institute for Translational Medicine (ITM) • Boston University Medical Campus — UL1TR001430: Boston University Clinical and Translational Science Institute • Brown University — U54GM115677: Advance Clinical Translational Research (Advance-CTR) • Carilion Clinic — UL1TR003015: iTHRIV Integrated Translational health Research Institute of Virginia • Charleston Area Medical Center — U54GM104942: West Virginia Clinical and Translational Science Institute (WVCTSI) • Children’s Hospital Colorado — UL1TR002535: Colorado Clinical and Translational Sciences Institute • Columbia University Irving Medical Center — UL1TR001873: Irving Institute for Clinical and Translational Research • Duke University — UL1TR002553: Duke Clinical and Translational Science Institute • George Washington Children’s Research Institute — UL1TR001876: Clinical and Translational Science Institute at Children’s National (CTSA-CN) • George Washington University — UL1TR001876: Clinical and Translational Science Institute at Children’s National (CTSA-CN) • Indiana University School of Medicine — UL1TR002529: Indiana Clinical and Translational Science Institute • Johns Hopkins University — UL1TR003098: Johns Hopkins Institute for Clinical and Translational Research • Loyola Medicine — Loyola University Medical Center • Loyola University Medical Center — UL1TR002389: The Institute for Translational Medicine (ITM) • Maine Medical Center — U54GM115516: Northern New England Clinical & Translational Research (NNE-CTR) Network • Massachusetts General Brigham — UL1TR002541: Harvard Catalyst • Mayo Clinic Rochester — UL1TR002377: Mayo Clinic Center for Clinical and Translational Science (CCaTS) • Medical University of South Carolina — UL1TR001450: South Carolina Clinical & Translational Research Institute (SCTR) • Montefiore Medical Center — UL1TR002556: Institute for Clinical and Translational Research at Einstein and Montefiore • Nemours — U54GM104941: Delaware CTR ACCEL Program • NorthShore University HealthSystem — UL1TR002389: The Institute for Translational Medicine (ITM) • Northwestern University at Chicago — UL1TR001422: Northwestern University Clinical and Translational Science Institute (NUCATS) • OCHIN — INV-018455: Bill and Melinda Gates Foundation grant to Sage Bionetworks • Oregon Health & Science University — UL1TR002369: Oregon Clinical and Translational Research Institute • Penn State Health Milton S. Hershey Medical Center — UL1TR002014: Penn State Clinical and Translational Science Institute • Rush University Medical Center — UL1TR002389: The Institute for Translational Medicine (ITM) • Rutgers, The State University of New Jersey — UL1TR003017: New Jersey Alliance for Clinical and Translational Science • Stony Brook University — U24TR002306 • The Ohio State University — UL1TR002733: Center for Clinical and Translational Science • The State University of New York at Buffalo — UL1TR001412: Clinical and Translational Science Institute • The University of Chicago — UL1TR002389: The Institute for Translational Medicine (ITM) • The University of Iowa — UL1TR002537: Institute for Clinical and Translational Science • The University of Miami Leonard M. Miller School of Medicine — UL1TR002736: University of Miami Clinical and Translational Science Institute • The University of Michigan at Ann Arbor — UL1TR002240: Michigan Institute for Clinical and Health Research • The University of Texas Health Science Center at Houston — UL1TR003167: Center for Clinical and Translational Sciences (CCTS) • The University of Texas Medical Branch at Galveston — UL1TR001439: The Institute for Translational Sciences • The University of Utah — UL1TR002538: Uhealth Center for Clinical and Translational Science • Tufts Medical Center — UL1TR002544: Tufts Clinical and Translational Science Institute • Tulane University — UL1TR003096: Center for Clinical and Translational Science • University Medical Center New Orleans — U54GM104940: Louisiana Clinical and Translational Science (LA CaTS) Center • University of Alabama at Birmingham — UL1TR003096: Center for Clinical and Translational Science • University of Arkansas for Medical Sciences — UL1TR003107: UAMS Translational Research Institute • University of Cincinnati — UL1TR001425: Center for Clinical and Translational Science and Training • University of Colorado Denver, Anschutz Medical Campus — UL1TR002535: Colorado Clinical and Translational Sciences Institute • University of Illinois at Chicago — UL1TR002003: UIC Center for Clinical and Translational Science • University of Kansas Medical Center — UL1TR002366: Frontiers: University of Kansas Clinical and Translational Science Institute • University of Kentucky — UL1TR001998: UK Center for Clinical and Translational Science • University of Massachusetts Medical School Worcester — UL1TR001453: The UMass Center for Clinical and Translational Science (UMCCTS) • University of Minnesota — UL1TR002494: Clinical and Translational Science Institute • University of Mississippi Medical Center — U54GM115428: Mississippi Center for Clinical and Translational Research (CCTR) • University of Nebraska Medical Center — U54GM115458: Great Plains IDeA-Clinical & Translational Research • University of North Carolina at Chapel Hill — UL1TR002489: North Carolina Translational and Clinical Science Institute • University of Oklahoma Health Sciences Center — U54GM104938: Oklahoma Clinical and Translational Science Institute (OCTSI) • University of Rochester — UL1TR002001: UR Clinical & Translational Science Institute • University of Southern California — UL1TR001855: The Southern California Clinical and Translational Science Institute (SC CTSI) • University of Vermont — U54GM115516: Northern New England Clinical & Translational Research (NNE-CTR) Network • University of Virginia — UL1TR003015: iTHRIV Integrated Translational health Research Institute of Virginia • University of Washington — UL1TR002319: Institute of Translational Health Sciences • University of Wisconsin-Madison — UL1TR002373: UW Institute for Clinical and Translational Research • Vanderbilt University Medical Center — UL1TR002243: Vanderbilt Institute for Clinical and Translational Research • Virginia Commonwealth University — UL1TR002649: C. Kenneth and Dianne Wright Center for Clinical and Translational Research • Wake Forest University Health Sciences — UL1TR001420: Wake Forest Clinical and Translational Science Institute • Washington University in St. Louis — UL1TR002345: Institute of Clinical and Translational Sciences • Weill Medical College of Cornell University — UL1TR002384: Weill Cornell Medicine Clinical and Translational Science Center • West Virginia University — U54GM104942: West Virginia Clinical and Translational Science Institute (WVCTSI)

Submitted: Icahn School of Medicine at Mount Sinai — UL1TR001433: ConduITS Institute for Translational Sciences • The University of Texas Health Science Center at Tyler — UL1TR003167: Center for Clinical and Translational Sciences (CCTS) • University of California, Davis — UL1TR001860: UCDavis Health Clinical and Translational Science Center • University of California, Irvine — UL1TR001414: The UC Irvine Institute for Clinical and Translational Science (ICTS) • University of California, Los Angeles — UL1TR001881: UCLA Clinical Translational Science Institute • University of California, San Diego — UL1TR001442: Altman Clinical and Translational Research Institute • University of California, San Francisco — UL1TR001872: UCSF Clinical and Translational Science Institute

Pending: Arkansas Children’s Hospital — UL1TR003107: UAMS Translational Research Institute • Baylor College of Medicine — None (Voluntary) • Children’s Hospital of Philadelphia — UL1TR001878: Institute for Translational Medicine and Therapeutics • Cincinnati Children’s Hospital Medical Center — UL1TR001425: Center for Clinical and Translational Science and Training • Emory University — UL1TR002378: Georgia Clinical and Translational Science Alliance • HonorHealth — None (Voluntary) • Loyola University Chicago — UL1TR002389: The Institute for Translational Medicine (ITM) • Medical College of Wisconsin — UL1TR001436: Clinical and Translational Science Institute of Southeast Wisconsin • MedStar Health Research Institute — UL1TR001409: The Georgetown-Howard Universities Center for Clinical and Translational Science (GHUCCTS) • MetroHealth — None (Voluntary) • Montana State University — U54GM115371: American Indian/Alaska Native CTR • NYU Langone Medical Center — UL1TR001445: Langone Health’s Clinical and Translational Science Institute • Ochsner Medical Center — U54GM104940: Louisiana Clinical and Translational Science (LA CaTS) Center • Regenstrief Institute — UL1TR002529: Indiana Clinical and Translational Science Institute • Sanford Research — None (Voluntary) • Stanford University — UL1TR003142: Spectrum: The Stanford Center for Clinical and Translational Research and Education • The Rockefeller University — UL1TR001866: Center for Clinical and Translational Science • The Scripps Research Institute — UL1TR002550: Scripps Research Translational Institute • University of Florida — UL1TR001427: UF Clinical and Translational Science Institute • University of New Mexico Health Sciences Center — UL1TR001449: University of New Mexico Clinical and Translational Science Center • University of Texas Health Science Center at San Antonio — UL1TR002645: Institute for Integration of Medicine and Science • Yale New Haven Hospital — UL1TR001863: Yale Center for Clinical Investigation

Other acknowledgements

We are grateful for the reflections and recommendations from Dr. Robert Kulkarni.

Data access

Investigators can access all data and analytic code used in this study through the National COVID Cohort Collaborative (N3C) Data Enclave (https://ncats.nih.gov/research/research-activities/n3c/data-overview/access). Access to the N3C data enclave is granted and managed by the National Center for Advancing Translational research and the N3C Data Access Committee.

Inclusion and ethics statement

All co-authors and collaborators included in this manuscript have fulfilled the criteria for authorship required by BMC Medicine

Funding

This research was financially supported by a global development grant (OPP1165144) from the Bill & Melinda Gates Foundation to the University of California, Berkeley, CA, USA.

Author information

Authors and Affiliations

  1. School of Public Health, University of California, Berkeley, Berkeley, CA, USA

    Zachary Butzin-Dozier, Yunwen Ji, Sarang Deshpande, Jeremy Coyle, Junming Shi, Andrew Mertens, Mark J. van der Laan, John M. Colford Jr & Alan E. Hubbard

  2. University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

    Eric Hurwitz

  3. University of Nebraska Medical Center, Omaha, NE, USA

    A. Jerrod Anzalone

  4. University of Alabama at Birmingham, Birmingham, AL, USA

    Rena C. Patel

Authors
  1. Zachary Butzin-Dozier
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Yunwen Ji
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Sarang Deshpande
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Eric Hurwitz
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. A. Jerrod Anzalone
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Jeremy Coyle
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Junming Shi
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Andrew Mertens
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Mark J. van der Laan
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. John M. Colford Jr
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Rena C. Patel
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Alan E. Hubbard
    View author publications

    You can also search for this author in PubMed Google Scholar

Consortia

the National COVID Cohort Collaborative (N3C) Consortium

Contributions

Authorship was determined using ICMJE recommendations. ZB: generated research question, drafted manuscript, managed project timeline, and coordinated analysis. YJ, SD, EH, JC, JA, and JS: reviewed manuscript, provided feedback, and conducted analysis. AM, ML, JC, RP, and AH: provided oversight on study design and analysis plan, reviewed manuscript, provided feedback, and supported interpretations. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Zachary Butzin-Dozier.

Ethics declarations

Ethics approval and consent to participate

This study was approved by the UC Berkeley Office for Protection of Human Subjects (2022–01-14980). The N3C data transfer to NCATS is performed under a Johns Hopkins University Reliance Protocol # IRB00249128 or individual site agreements with NIH. N3C received a waiver of consent from the NIH Institutional Review board and allows the secondary analysis of these data without additional consent.

Consent for publication

The authors consent to the publication of this manuscript in its entirety.

Competing interests

The authors declare no competing interests related to this study.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

12916_2024_3655_MOESM1_ESM.pdf

Additional file 1: Supplemental Table 1 – List of depressive conditions, Appendix 1 – Covariate information, Supplemental Figure 1 – CONSORT diagram, Supplemental Table 2 - Exploratory analysis of dose-response relationship between fluoxetine use and Long COVID, Supplemental Table 3 - Variable importance of included covariates, Appendix 2 – Nonparametric sensitivity analysis.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Butzin-Dozier, Z., Ji, Y., Deshpande, S. et al. SSRI use during acute COVID-19 and risk of long COVID among patients with depression. BMC Med 22, 445 (2024). https://doi.org/10.1186/s12916-024-03655-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12916-024-03655-x

Keywords

BMC Medicine

ISSN: 1741-7015

Contact us