Hyponatremia Associated with the Use of Common Antidepressants in the All of Us Research Program

Abstract

Selective serotonin reuptake inhibitor (SSRI), serotonin‐norepinephrine reuptake inhibitor (SNRI), and norepinephrine–dopamine reuptake inhibitor (NRI) antidepressants can cause hyponatremia through syndrome of inappropriate antidiuretic hormone secretion (SIADH). This study assesses the differential risks of hyponatremia associated with commonly prescribed SSRIs (fluoxetine, paroxetine, sertraline, citalopram, escitalopram), SNRIs (duloxetine, venlafaxine) and NRI (bupropion), as well as omeprazole as a reference, with a retrospective observational cohort study in the All of Us Research Program, a national multicenter research cohort containing de‐identified electronic health records (EHR). Participants who had been prescribed monotherapy with any of eight common antidepressants were included, with each drug considered as a separate arm indexed with a start date. Events were defined as the first occurrence of a low plasma sodium measurement or a clinical diagnosis recorded for either hyponatremia or SIADH. Those who did not have events were censored at their last plasma sodium measurement. A total of 17,439 individuals were exposed to one of the eight antidepressants as monotherapy. The overall incidences for hyponatremia were 0.87% in the first 30 days and 10.5% in the first 3 years in the antidepressant arms. Compared to sertraline, duloxetine (hazard ratio [HR] = 1.37 [1.19–1.58]) and escitalopram (HR = 1.16 [1.01–1.33]) were associated with the highest overall risk of hyponatremia, and bupropion (HR = 0.83 [0.73–0.94]) and paroxetine (HR = 0.78 [0.65–0.93]) were associated with the lowest risk. The risks were unchanged after adjusting for comorbidity and polypharmacy. Such information could help guide providers in managing patients and their risks of hyponatremia when on common antidepressants.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Common antidepressants can cause hyponatremia. However, the relative risks among different drugs over long‐term follow‐up are unknown.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Are there differential risks of hyponatremia when prescribed one of eight common antidepressants?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

With this large national cohort, this study demonstrated differential short‐ and long‐term risks among common antidepressants. Duloxetine was associated with hyponatremia at a greater rate than the other studied antidepressants. Bupropion and paroxetine were associated with the lowest risks of hyponatremia.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Our study suggests that patients on duloxetine and other antidepressants with higher risks of hyponatremia might benefit from more rigorous monitoring for hyponatremia.

Several classes of commonly used antidepressants can cause hyponatremia via the syndrome of inappropriate antidiuretic hormone secretion (SIADH) through mechanisms involving the central nervous system. Selective serotonin reuptake inhibitors (SSRIs) and serotonin‐norepinephrine reuptake inhibitors (SNRIs) can lead to increased release of ADH from the posterior pituitary by potentiating the activation of adrenergic receptors within the hypothalamus. ¹ With SSRIs, the incidence of drug‐induced SIADH has been reported in 0.5 to 32% of patients using these drugs. ² , ³

SSRIs, SNRIs, and norepinephrine–dopamine reuptake inhibitors (NRIs) are widely used for the treatment of mood disorders (such as major depressive disorder), ⁴ , ⁵ anxiety disorders, ⁶ post‐traumatic stress disorder (PTSD), ⁷ and chronic pain. ⁸ In clinical practice, the selection of antidepressants is based on optimizing efficacy and minimizing toxicity. Other common side effects of these medications can include changes in eating habits, changes in sleep patterns, and changes in sexual function. ⁹

Hyponatremia can often accompany the use of SSRIs, SNRIs, or NRIs in routine clinical practice, but these laboratory abnormalities may not be identified until patients develop alterations in central nervous system function. ¹⁰ , ¹¹ Mild hyponatremia (serum sodium < 135 mEq/L) can lead to gait disturbances, falls, and fractures, especially in the advanced elderly. ¹² If hyponatremia is severe (< 125 mEq/L), cerebral edema can lead to seizures, disability, or death. ¹³ Most cases of very severe hyponatremia (< 116 mEq/L) are centrally mediated. ¹⁴

Hyponatremia can be caused by a variety of mechanisms. Patients with hypovolemia or congestive heart failure often exhibit hyponatremia resulting from maladaptive neurohormonal processes attempting to expand volume. ¹⁵ The most common form of euvolemic hyponatremia is SIADH. In addition to medications, SIADH can also be due to health conditions including severe infections such as pneumonia, or physiological stressors such as pain. Other commonly used drugs (e.g., neuroleptics or opioid analgesics) can cause drug‐induced SIADH as well. ¹⁶

The All of Us Research Program (All of Us) is a National Institutes of Health (NIH) initiative which aims to enroll one million diverse participants across the United States. The database contains de‐identified longitudinal healthcare data derived from EHRs from multiple healthcare institutions, baseline health surveys, biospecimens, and biometrics from wearable monitoring devices. ¹⁷ Over 80% of the All of Us participants are underrepresented in biomedical research. ¹⁸

Retrospective studies have evaluated the risks of hyponatremia associated with antidepressant use. The designs of these studies have acknowledged the limitations of observational data. However, most of these studies used matched population‐based controls, who often appeared to be either healthier ¹⁹ , ²⁰ or potentially less frequently evaluated in the healthcare systems following arbitrary index dates. ²¹ These studies also focused primarily on early events after initial prescription, and did not report long‐term risks. Furthermore, the relative risks of antidepressants in comparison with each other is not clear. In this study, we used a large aggregate dataset in All of Us to test the specific hypothesis that the proximal and long‐term risks of hyponatremia differ among SSRIs, SNRIs, and NRIs within the context of routine care.

METHODS

In this study, we constructed a cohort of participants from within the All of Us Research Program who had used SSRI, SNRI, or NRI antidepressants in the course of routine clinical care, and we assigned them to study arms defined by the individual antidepressants that they had used. We then compared the risk of hyponatremia between each of these different antidepressants, using sertraline as a reference medication with and without adjustment for polypharmacy and clinical comorbidity. We then further compared the risk of hyponatremia with these antidepressants to omeprazole, a commonly used non‐psychiatric maintenance medication prescribed at a frequency similar to antidepressants. Lastly, we used a phenome‐wide association study (PheWAS) to explore potential health factors that could contribute to a differential risk of hyponatremia for patients using these antidepressants in the context of routine clinical care.

Study population

This study was performed using All of Us Curated Data Repository (CDR) Version 6, which contained de‐identified EHR data from 230,369 participants. The overall study design is illustrated in Figure 1 . The demographic data (sex at birth, gender, race, and ethnicity) of the participants were either derived from EHR or collected through the All of Us enrollment survey.

Figure 1.

The study design. (a) A flowchart for inclusions and exclusions in the cohort building. ESRD: end‐stage renal diseases. (b) A design diagram to illustrate the temporality of variables. Years on the timeline are for example purposes only; actual timelines of exposure vary by individual. Drug 2: The prescription of a second antidepressant. Hypo‐Na: A hyponatremia event (defined by an abnormal laboratory sodium measurement or a diagnostic code. Rx: An EHR mention (such as prescription) of the study drug.

For SSRIs, we included fluoxetine, paroxetine, sertraline, citalopram, and escitalopram. For SNRIs, we included duloxetine and venlafaxine. The NRI bupropion was also included in this study. Our primary reference drug was sertraline, the most used antidepressant. In addition, we included omeprazole (a non‐psychiatric medication) as a secondary reference drug. Omeprazole is a commonly prescribed acid suppression medication used with a similar frequency, a similar age distribution, and a similar pattern of comorbidities in All of Us (Table 1 , Figure S6A,B ). Hyponatremia with omeprazole (or with any of the proton pump inhibitors) occurs infrequently. ²² Participants were assigned to study arms defined by the index drug that they had used: fluoxetine, paroxetine, sertraline, citalopram, escitalopram, duloxetine, venlafaxine, bupropion, or omeprazole. Participants who were prescribed more than one SSRI, SNRI, or NRI were excluded from our analysis (Supplementary Result S1 ) to avoid misclassification as well as potential analytical complications related to the timing of drug epochs, outcome ascertainment assignment, and related dosage adjustments. Participants prescribed both an antidepressant and omeprazole were assigned to the antidepressant arms if they otherwise met the criteria. All the included participants were indexed with the first recorded date of the study drugs in their EHR (time zero for the survival analysis).

Table 1.

Participant demographics

	Fluoxetine	Paroxetine	Sertraline	Citalopram	Escitalopram	Duloxetine	Venlafaxine	Bupropion	Overalla	Omeprazole
n	2,182	955	3,773	2,353	2,206	2,035	1,416	2,519	17,439	10,532
Index month/yearb	6/2015 (3/2011–1/2018)	7/2014 (6/2009–7/2017)	6/2016 (1/2013–8/2018)	6/2014 (5/2011–4/2017)	4/2017 (5/2014–3/2019)	9/2017 (4/2015–5/2019)	10/2015 (2/2012–6/2018)	10/2015 (7/2011–4/2018)	2/2016 (6/2012–7/2018)	7/2015 (2/2012–3/2018)
Ageb	48 (35–58)	54 (44–61)	49 (34–60)	52 (39–61)	51 (36–61)	55 (46–64)	53 (46–61)	50 (30–59)	51 (38–61)	57 (45–66)
Follow‐Up (Months)b	57.0 (24.9–108.3)	65.7 (29.2–125.2)	48.5 (21.1–90.3)	68.5 (33.9–112.3)	35.4 (14.9–73.2)	35.0 (14.6–62.2)	53.1 (22.0–100.4)	55.7 (23.7–107.3)	50.6 (21.5–95.8)	59.1 (25.4–103.0)
Sex at birthc
Female	685 (71.73%)	685 (71.73%)	2,683 (71.11%)	1,663 (70.68%)	1,582 (71.71%)	1,450 (71.25%)	1,132 (79.94%)	1,580 (62.72%)	12,312 (70.60%)	6,218 (59.04%)
Male	571 (26.17%)	250 (26.18%)	985 (26.11%)	639 (27.16%)	572 (25.93%)	530 (26.04%)	247 (17.44%)	871 (34.58%)	4,665 (26.75%)	4,015 (38.12%)
Race/ethnicityd
White	1,413 (64.76%)	576 (60.31%)	2,350 (62.28%)	1,530 (65.02%)	1,485 (67.32%)	1,227 (60.29%)	1,027 (72.53%)	1,664 (66.06%)	11,272 (64.64%)	6,472 (61.45%)
Black/African American	324 (14.85%)	162 (16.96%)	645 (17.10%)	407 (17.30%)	280 (12.69%)	425 (20.88%)	178 (12.57%)	426 (16.91%)	2,847 (16.33%)	1,587 (15.07%)
Hispanic/Latino	295 (13.52%)	147 (15.39%)	538 (14.26%)	266 (11.30%)	296 (13.42%)	271 (13.32%)	132 (9.32%)	279 (11.08%)	2,224 (12.75%)	1733 (16.45%)
Cum. Inc. at first 3 yearse	0.108 (0.0942–0.123)	0.0802 (0.0636–0.101)	0.103 (0.0929–0.115)	0.0923 (0.0806–0.106)	0.124 (0.109–0.141)	0.143 (0.126–0.162)	0.106 (0.0892–0.125)	0.0873 (0.0758–0.100)	0.105 (0.101–0.111)	0.0925 (0.0866–0.0987)
Cum. Inc. first 30 dayse	0.00882 (0.00563–0.0138)	0.0127 (0.00721–0.0222)	0.00675 (0.00456–0.00997)	0.00813 (0.00520–0.0127)	0.0115 (0.00776–0.0169)	0.0105 (0.00683–0.0160)	0.0100 (0.00596–0.0169)	0.00601 (0.00363–0.00994)	0.00871 (0.00742–0.0102)	0.00949 (0.00780–0.0116)
Use of AAP	351 (16.09%)	96 (10.05%)	475 (12.59%)	260 (11.05%)	230 (10.43%)	243 (11.94%)	188 (13.28%)	290 (11.51%)	2,133 (12.23%)	313 (2.97%)
DM	449 (20.58%)	252 (26.39%)	804 (21.31%)	510 (21.67%)	384 (17.41%)	575 (28.26%)	297 (20.97%)	546 (21.68%)	3,817 (21.89%)	2,404 (22.83%)
CHF	90 (4.12%)	51 (5.34%)	173 (4.59%)	96 (4.08%)	84 (3.81%)	136 (6.68%)	58 (4.10%)	133 (5.28%)	821 (4.71%)	573 (5.44%)

AAP, atypical antipsychotics; CHF, congestive heart failure; Cum. Inc., cumulative incidence; DM, diabetes mellitus.

^a Overall for antidepressant arms (omeprazole arm not included).

^b Median (interquartile range).

^c In the antidepressant arms combined, 62 females at birth were identified as male gender, transgender, or non‐binary, while 37 males at birth were identified as female gender, transgender, or non‐binary. In the omeprazole arm, 21 females at birth were identified as male gender, transgender, or non‐binary, while ≤ 20 males at birth were identified as female gender, transgender, or non‐binary.

^d Unlisted races and ethnicities (~8% combined and similar in all arms) include Asians, Native Hawaiians or other Pacific Islanders, Middle Eastern or North Africans, and those who preferred not to answer.

^e Estimate (95% confidence intervals).

Participants who ever had diagnostic codes for end‐stage renal disease, acidosis, diabetic ketoacidosis and nonketotic hyperosmolar coma, sepsis, hypovolemia, renal dialysis, and other conditions that suggest alternative causes for hyponatremia were excluded from all cohorts (Supplementary Method S1 ). Also, participants who ever had hyponatremic events prior to the start of antidepressants or omeprazole were excluded. Covariates adjusted included age at time zero (reported as in decades), sex, and self‐reported race and ethnicity, as defined in Supplementary Method S2 .

Outcome definition

For all of the survival analyses in this study, we defined outcome events as the first presence of decreased plasma sodium (≤ 134 mEq/L), adjusted using the highest glucose reading of the same day with Katz's formula: Measured sodium + 0.016 × (Serum glucose − 100). ²³ We also accepted diagnostic codes for hyponatremia/hypoosmolality (ICD‐9‐CM 276.1 and ICD‐10‐CM E87.1) or diagnostic codes for syndrome of inappropriate antidiuretic hormone (ICD‐10‐CM E22.2) as outcome events. Individuals who did not experience the events were right censored at the date of last plasma sodium laboratory measurement. In this study, the relative percentages of participants meeting event criteria (using 134 mEq/L as cut‐off) were 92% by laboratory result only, compared to 5% by diagnostic code only, and 3% by both on the same day.

We reported the cumulative incidence of outcome events within the first 30 days (early events) and within the first 3 years.

Polypharmacy and comorbidity

As a secondary analysis, we studied the role of polypharmacy and comorbidity by examining the influence of several other classes of medications and coexisting medical conditions as covariates. Concomitant medication classes included as covariates were drugs that can alter hemodynamics (antihypertensives) or electrolyte homeostasis (diuretics), because these factors can influence the secretion of ADH from the hypothalamus and posterior pituitary. We included β‐ or αβ‐blockers, angiotensin‐converting enzyme inhibitors (ACEi), angiotensin II receptor blockers (ARB), calcium channel blockers (CCB), and two common diuretics (hydrochlorothiazide and furosemide), as well as atypical antipsychotics. Relevant comorbidities included as covariates were congestive heart failure (CHF), which can directly cause SIADH, and diabetes mellitus, which can cause a constellation of electrolyte disorders including hyponatremia, as well as an osmotic laboratory‐based artifact. ¹⁵ , ²⁴ (For full lists of medications explored, please see Supplementary Method S3 ) Due to the chronic nature of these two comorbidities, we extended the period for capturing these covariates to “ever had” in an effort to minimize the impact of the false negatives in our secondary analyses.

Statistics

To compare the risks of hyponatremia events between different drugs, we used Kaplan–Meier survival analysis and the Cox proportional hazards (CPH) model from the Lifelines Python package. ²⁵ The CPH models have been adjusted according to age of time zero, self‐reported race/ethnicity, and sex. We conducted pairwise comparisons between each investigated drug class. To control for potential confounding from comorbidity and polypharmacy, we also performed similar CPH model tests that included these covariates and reported this in Figure S4B .

To study the interactions between SSRIs, SNRIs, NRIs, and relevant covariates, we also tested the impact of each covariate individually by introducing an interaction term between Drug 1 (compared with Drug 2) and that covariate into the CPH model. The P‐value and effect size of the interaction term of each test were reported.

To identify differences in disease profiles between arms that might contribute to the differential risk of hyponatremia, we utilized PheWAS ²⁶ , ²⁷ to conduct comparisons between the study arms. PheWAS is a set of EHR‐wide parallelized logistic regressions, each of which uses the presence or absence of a series of functionally grouped ICD codes (termed a phecode) as the outcome, and this analysis was performed as previously described. ²⁸ , ²⁹

All statistics assumed a two‐tailed distribution. All statistical tests were performed in the All of Us Researcher Workbench using Python version 3.7.

Ethics statement

The All of Us Research Program is overseen and approved by the All of Us institutional review board (IRB) and participants were consented through All of Us. Given that all data available to researchers in All of Us have personal identifiers removed, the All of Us IRB deemed studies in the Researcher Workbench not to be human participant research and waived the need for additional IRB approval for studies performed in the Researcher Workbench. ³⁰

RESULTS

Demographics

We identified 17,439 participants exposed to any SSRI, SNRI, or NRI drugs as monotherapy indexed over a time window reflecting almost 3 decades (date range 1989 and 2021) with median EHR duration of 10 years (IQR: 6–16 years). The demographics are shown in Table 1 . The median ages of the participants included within each arm were similar across the different index drug classes (median: 48–55 years old), although the omeprazole arm was slightly older (median: 57 years old). Overall, there were more females (antidepressants 73%; omeprazole 62%) than males. Most individuals self‐identified as White (60–70%) across study arms. Twelve percent of participants in the antidepressant arms had also been prescribed atypical antipsychotics, compared with 3% in the omeprazole arm. About 22% of the included participants had at least one diagnostic code for diabetes, which was similar to the overall prevalence of diabetes in All of Us. Congestive heart failure diagnoses were rare (about 5%) across arms.

Demographic parameters including age (HR = 1.06 per decade), race (Black/African American, HR = 1.32), ethnicity (Hispanic/Latino, HR =1.11), and male sex at birth (HR = 1.09) were all significantly associated with increased risk of hyponatremia. Similarly, polypharmacy with other relevant psychiatric medications (e.g., atypical antipsychotics [HR = 1.53]), medications impacting hemodynamics (e.g., CCBs [HR = 1.39], and β‐blockers [HR = 1.52]), and medications impacting electrolyte homeostasis (e.g., furosemide [HR = 1.86], hydrochlorothiazide [HR = 1.25], and ACEi/ARBs [HR = 1.50]), and comorbidity diabetes (HR = 1.67) and congestive heart failure (HR = 1.69) were all associated with increased risks (Figure S1 ).

Comparisons between different SSRIs, SNRIs, and NRIs

The differential risks of hyponatremia for all study drugs are summarized in Figure 2 and the onset of hyponatremia is plotted as a function with time in Figure 3 and detailed in Table S1 . Among the investigated drugs over the follow‐up, the order of risks for hyponatremia from high to low was: duloxetine (HR = 1.56, 95% confident interval [CI]: [1.38–1.76], compared with omeprazole, or 1.37 [1.19–1.58], compared with sertraline), escitalopram, venlafaxine, fluoxetine, sertraline, citalopram, bupropion (0.94 [0.84–1.06] compared with omeprazole or 0.83 [0.73–0.94] compared with sertraline), and paroxetine (0.86 [0.73–1.02] compared with omeprazole and 0.78 [0.65–0.93] compared with sertraline). Escitalopram had a higher risk of hyponatremia than citalopram (HR = 1.24 [1.07–1.44]). Results of the proportional hazard assumption tests are described in Supplementary Result S2 , Figure S3 , and Table S2 . At the end of the first 3 years, the cumulative incidence for hyponatremia was 10.5% in the overall combined antidepressant arms, with highest in the duloxetine arm (14.3%) and lowest in the paroxetine (8.0%) or bupropion (8.7%) arms (Table 1 ).

Figure 2.

A forest plot to compare the risks of hyponatremia among the SSRI, SNRI, and NRI drugs with reference drug omeprazole and sertraline, with a glucose‐adjusted sodium value of ≤ 134 mEq/L as the cut‐off for hyponatremia. CI., confidence interval; HR, hazard ratio.

Figure 3.

Differential risks of hyponatremia among the SSRI, SNRI, and NRI drugs. (a, b) The cumulative incidence of hyponatremia with glucose‐adjusted sodium ≤ 134 mEq/L (a) and ≤ 130 mEq/L (b) as cut‐offs; the red vertical dash lines are the 3‐year marks as a reference; y axis: Cumulative Hazard. (c, d) The chronological change of differential risks for hyponatremia between different antidepressants with Nelson Aalen Hazard plots (y axis: Hazard within the timeframe). (c) Early events since the initiation of the medications (bandwidth = 15 days). (d) Events during long‐term follow‐up when duloxetine and sertraline were compared (bandwidth = 12 months, Shades: 95% confidence interval).

The Nelson Aalen hazard plot for hyponatremia showed peaks in the risk for hyponatremia during the first 30 days from the initiation of the investigated drugs. The overall first 30‐day incidence for hyponatremia was 0.87% in the antidepressant arms (Table 1 ). Interestingly, although paroxetine had a lower risk of hyponatremia during the entire follow‐up, it had a high spike within the first 30 days (1.27%) (Figure 3 c ). Escitalopram (1.15%) and duloxetine (1.05%) also had a higher incidence of events within the first 30 days. Bupropion (0.601%) had the lowest first 30‐day hyponatremia incidence.

In contrast to previous studies limited to only proximal events (the first 30 days of initiation of SSRIs, SNRIs, or NRIs), ¹⁹ , ²¹ , ³¹ we observed a higher risk of hyponatremia for duloxetine users compared with sertraline users even 10 years after the first EHR mention of their usage (Figure 3 a,d ).

Next, we used the first occurrence of plasma sodium level below or equal to 130 mEq/L or hyponatremia billing code as events to investigate hyponatremia outcomes of increased severity. As expected, the outcome events for each arm are markedly reduced when the criteria are more stringent (Figure 3 b, Figure S4B ). However, we again observed that duloxetine had the highest risk for clinically significant hyponatremia when compared with reference drug omeprazole (HR = 1.74 [1.37–2.20]) or with sertraline (HR = 1.65 [1.25–2.19]), while paroxetine had the lowest risk (HR = 0.72 [0.50–1.03] compared with omeprazole or 0.67 [0.45–0.99] compared with sertraline).

Role of comorbidity and polypharmacy

To control for potential confounding due to comorbidity and commonly used hemodynamic‐altering medications within the older adult population, we performed multivariable regression analyses with additional covariates including the presence of diabetes and chronic heart failure, as well as interacting medications including hydrochlorothiazide, furosemide, CCBs, β‐blockers, ACEi/ARBs, and atypical antipsychotics. After adjusting for these factors, the differential risks among investigated antidepressants and the reference drug omeprazole were almost unchanged (Figure S4B ).

We then introduced interaction terms between investigated antidepressants and covariates (one in each model). None of the P‐values reported in the interaction studies passed multiple‐comparison‐adjusted thresholds for significance. Here, we report positive interactions with agents with increased risks at nominal significance levels (P < 0.05). We observed that escitalopram had nominally significant interaction signals in participants who also took hydrochlorothiazide (e.g., interaction HR [HRi] = 1.46 [1.08–1.97], P = 0.015, compared with sertraline) or ACEi/ARBs (HRi = 1.37 [1.02–1.84], P = 0.037, compared with citalopram; Figure S5A,B ). Fluoxetine appeared also to interact with older age (per decade), especially when compared with sertraline and citalopram (HRi = 1.15 [1.04–1.27], P = 0.006; Figure S5C,E ). The interaction terms between all study drugs and diabetes were not statistically significant, while the only significant interaction with congestive heart failure was between venlafaxine and paroxetine (Figure S5D ). Table S3 provides further details of potential interactions.

Phenome scanning for additional relevant comorbidities

PheWAS analyses provided further understanding of the health backgrounds of the participants of each study drug. For example, compared with the sertraline arm, the duloxetine arm shows enrichment for type 2 diabetes with neurological manifestations, obesity, chronic pain syndrome, other peripheral nerve disorders, peripheral vascular disease, gastroparesis, spondylosis and allied disorders, spinal stenosis, myalgia and myositis unspecified (Supplementary Result S4 , Figure S6 ). This is consistent with the common use of duloxetine to treat chronic pain. Participants taking bupropion were less likely to be diagnosed with mood disorders and epilepsy but more likely to be diagnosed with smoking‐associated lung disease phenotypes, as well as treatments for obesity.

DISCUSSION

We report a retrospective cohort study quantifying the differential risks of drug‐induced hyponatremia for SSRIs, SNRIs, and NRIs in All of Us, a diverse research cohort representing EHR data from across the United States. Duloxetine had the highest risk of hyponatremia. The risk of hyponatremia with duloxetine and escitalopram was both measurably greater than the risk of hyponatremia with sertraline, our primary reference drug. Duloxetine, escitalopram, venlafaxine, fluoxetine, and sertraline (in order of decreasing risk) had an elevated risk of hyponatremia when they were compared to the non‐psychiatric reference drug omeprazole. Bupropion and paroxetine were associated with the lowest frequency of hyponatremia, with risks similar to the risk of hyponatremia observed with omeprazole.

From among the eight study drugs, duloxetine was associated with the highest frequency of hyponatremia, (i) during the first 30 days of treatment, and (ii) throughout more than a decade of follow‐up. In terms of absolute risk, 14.3% of participants on duloxetine and 12.4% on escitalopram developed hyponatremia at < 134 mEq/L in the first 3 years, compared to 10.3% on sertraline and 8.7% on bupropion. When compared with our reference drug omeprazole, duloxetine was 56% more likely to cause hyponatremia. The relationship between duloxetine and hyponatremia has previously been reported within a month of starting the drug. ³² , ³³ , ³⁴ , ³⁵ Our results show that this increased risk persists over the course of a decade.

Although escitalopram (the active stereoisomer of citalopram) has been shown to have superior efficacy compared to citalopram, ³⁶ we found that escitalopram had a higher risk of hyponatremia than citalopram in this study (HR = 1.24 [1.07–1.44]).

The long‐term risk of hyponatremia with paroxetine was among the lowest of the antidepressants we studied, although we noted a higher rate within the first 30 days of paroxetine start (Figure 3 c ). In previous studies, paroxetine was often reported to be associated with higher hyponatremia risks. The Canadian ²¹ and Japanese ³¹ studies suggested that paroxetine is among the antidepressants with higher risks; both of these studies only evaluated events soon after the initiation of treatment. Prospective longitudinal studies have also found an increased frequency of hyponatremia within the first 14 days after initiation of paroxetine. ³⁷ , ³⁸

Bupropion was also associated with a lower risk of hyponatremia in this study. In the literature, a case report that described sertraline‐associated hyponatremia specifically highlighted a patient who was subsequently able to tolerate bupropion treatment without any electrolyte abnormalities. ³⁹ However, there are also case reports of bupropion‐induced hyponatremia. ⁴⁰ , ⁴¹

The population‐based nature of our study allows us to assess the potential impact of relevant comorbidities and drug interactions. For example, the risk of hyponatremia for escitalopram appears to be increased when on diuretics as well as ACEi and ARBs. This observation supports findings from a previous study reporting a higher risk for hyponatremia when SSRIs were used at the same time as diuretics and ACEi. ⁴² Fluoxetine, a potent CYP2D6 inhibitor, ⁴³ is also known to be associated with a high rate of drug interactions. ⁴⁴ In our dataset, the risk of hyponatremia with fluoxetine increased with age, especially when compared with sertraline and citalopram. We also assessed the impact of two key clinical covariates, diabetes mellitus and congestive heart failure, both of which were associated with higher risks for hyponatremia, as expected. However, when we assessed these two variables in our models, the interaction terms between our study drugs and these two comorbidities were not statistically significant. The P‐values for the reported possible drug–drug interactions did not pass the multiple‐comparison corrected threshold for statistical significance. Since clinical modifiers are difficult to study, we describe our observations to encourage further investigations.

As an EHR‐based retrospective observational study, this study is subjected to limitations. First, the actual durations of medication usage may be imprecise and it is possible medications were stopped before events were recorded. However, the drugs we studied here were maintenance medications and commonly used long‐term. We focused on participants who had only used a single agent to avoid interference by drug switching or overlaps, and the results were similar when we included people who used multiple medications. We also acknowledge that, while the design of our study (based on arm‐by‐arm comparisons) increases the likelihood that the effects of incomplete data would be similar across arms, these effects may not be identical. For outcome misclassification bias (e.g., due to incomplete data capture), non‐differential and independent misclassification may bias toward the null. ⁴⁵ Incomplete data capture in EHR may impact exposure, outcome, and confounder measurements. It is reassuring to note, however, that when we ran a more conservative analysis with censoring at +90 days of the last mention of the study drug (Figure S4C ), the HR and P‐values were similar to those presented in our initial analysis (Figure 2 ). Ultimately, prospective studies with adequate monitoring for hyponatremia will establish rates of hyponatremia with higher confidence, especially within the context of dose–response.

We also acknowledge that, although hyponatremia is known to occur infrequently with our secondary reference drug, omeprazole, rare SIADH events have been documented with the use of proton pump inhibitors and the rate is, therefore, non‐zero. ²² The purpose of the omeprazole arm in our study was not as a “normal” or “null” control, but as a commonly prescribed medication of a different class not typically thought of as causing hyponatremia. ⁴⁶ , ⁴⁷ Therefore, antidepressants that did not show significantly higher risks than omeprazole in this study might still have risks for hyponatremia. Lastly, we acknowledge that the studied antidepressants have different clinical indications for prescriptions. For example, we found that duloxetine was more commonly prescribed to treat chronic pain, and bupropion was more commonly used for tobacco cessation and avoided in patients at risk for seizures, concordant with the indications and contraindications ⁴⁸ , ⁴⁹ PheWAS provided a clinical context to interpret our findings and it highlighted specific patient subsets (Supplementary Result S4 ).

CONCLUSION

Within this large national cohort, our study demonstrates differential short‐ and long‐term risks among common SSRIs, SNRIs, and NRIs. Duloxetine (an SNRI) is associated with hyponatremia at a greater rate than the other studied antidepressants. As a retrospective study, the clinical interpretation and generalization of these findings should be viewed with caution. Without further evidence from prospective randomized clinical trials, our findings should not be used to justify switching duloxetine to sertraline to lower the risk; rather, our study suggests that patients on duloxetine might benefit from more rigorous monitoring for hyponatremia given their increased risks either from the medication or the comorbidities related to the prescription.

FUNDING

This research was supported by the Intramural Research Program of the National Human Genome Research Institute (NHGRI), National Institutes of Health. Members of the Cohort Analytics Core were funded by ZIC grant number HG200420, and members of the Precision Health Informatics Section were funded by ZIA grant number HG200417. The All of Us Research Program is funded by grants through the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA no.: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085 U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276.

CONFLICTS OF INTEREST

The authors declared no competing interests in this work.

AUTHOR CONTRIBUTIONS

H.M., Y.C., T.M.F., B.J.W., E.A.L., R.A.W., and J.C.D. wrote the manuscript; H.M., T.M.F., D.J.S., J.M.K., E.A.L., R.A.W., and J.C.D. designed the research; H.M., Y.C., D.J.S., A.H.A., S.B.G., A.W., A.B., and O.S., performed the research; H.M., D.J.S., and A.H.A. analyzed the data; H.M., D.J.S., T.C.T., and J.M.K. contributed new analytical tools.

Supporting information

Data S1