Key Findings

Nineteen randomized controlled trials (RCTs) and two observational studies constituted the evidence base for this review. Six therapies were compared with placebo: selective serotonin reuptake inhibitors (SSRIs) (citalopram, escitalopram, fluoxetine and sertraline), serotonin – norepinephrine reuptake inhibitors (SNRIs) (duloxetine and venlafaxine), bupropion extended release (XR), mirtazapine, trazodone and vortioxetine. Fewer direct comparisons of antidepressants exist: SSRI vs. tricyclic antidepressants (TCAs), within-class comparisons of the SSRIs, SNRI vs. SSRI, mirtazapine vs. paroxetine and vortioxetine vs. duloxetine. None of the RCTs were designed to evaluate adverse events and were not powered to do so, thus our confidence in the findings were attenuated in some circumstances, as reflected in the associated strength of evidence (SOE). Interpretation of these findings was based on statistical significance, thus potentially missing small differences in outcome. Suspected selective outcome reporting was an additional domain that was commonly downgraded, again contributing to lower SOE ratings.

SNRIs, but not SSRIs, were statistically significantly associated with adverse effects when used as treatment during the acute phase of major depressive disorder (MDD), although both classes led to more study withdrawals due to adverse events compared with placebo. SOE was relatively lower for SSRIs than for SNRIs because of imprecision and suspected selective outcome reporting. Unfortunately when studies reported the contributing adverse events they were mostly nonspecific and those most commonly expected according to prescribing information (e.g., nausea, dizziness). Observational data suggests increased adverse events with longer treatment durations for SSRIs and venlafaxine, although SOE was low given the observational design and residual confounding. Serious adverse events may be less frequent with duloxetine (low SOE) compared with placebo during treatment of the acute phase of MDD but not with longer treatment into the continuation phase (moderate SOE). SOE was low and moderate, respectively, owning to study risk of bias and imprecision. In addition, the details of the serious adverse events were not always provided.

Not surprising, we found SSRIs to have fewer adverse events or withdrawal due to adverse events compared with TCAs. Within the SSRI class comparisons, (paroxetine, escitalopram, and sertraline versus fluoxetine) data do not suggest a difference in evaluated harms although any given outcome was usually represented by a single trial with few events. Similarly, comparisons of SNRIs with SSRIs were usually based on a single trial; hence, outcomes did not differ significantly between these two classes. Compared with paroxetine, mirtazapine increased the risk of withdrawal due to adverse events. Vortioxetine was compared with duloxetine in a single trial and decreased the risk of any adverse events.

Clinically it is more informative to understand specific harms associated with antidepressants although we found specific harms to be less frequently reported than general outcomes (i.e. any adverse event or study withdrawals). In older adults, clinicians are often concerned with prescribing therapies that may increase the risk of falls or fractures, in part based on recommendations made in the Beers Criteria.²⁸ Trial data supported an increased risk of falls with duloxetine and a cohort study suggested an association of venlafaxine with falls. The same cohort study found SSRIs as a class to be associated with falls although this outcome hasn’t been studied in RCTs to date; thus, confidence in the association of falls with SSRIs was lower than with SNRIs. Data directly comparing SNRIs with SSRIs were insufficient regarding outcomes of falls or fractures.

An additional concern regarding prescribing of antidepressants in the elderly is the risk of syndrome of inappropriate antidiuretic hormone (SIADH).²⁸ We found no evidence regarding SIADH for any of the included antidepressants.

Data regarding subgroups of interest (KQ 2) were scarce. Current data suggest that an age greater than 75 years is associated with a greater risk of serious adverse events and that the risk of falls with duloxetine is influenced by the presence of cardiopulmonary disease.

Findings in Relationship to What Is Already Known

Comparing our findings with those from prior systematic reviews is difficult for several reasons. First, many earlier reviews in MDD included populations ineligible in our review because their age thresholds were lower (less than 65), thus in this way our review is unique. In addition, earlier systematic reviews^12,27 that included any assessment of harms tended to focus on general outcomes such as overall tolerability or discontinuation rates due to adverse events rather than any specific adverse events of more concern in the older population (i.e. falls, fractures, SIADH).

One prior systematic review and network meta-analysis⁹ in patients 60 years and older in age with MDD found falls to be rare. Three RCTs reported four falls, three in the SSRI arm and one in SNRI arm. Other systematic reviews on SSRIs in older adults allow inclusion of broad indications^79,80 One review found SSRIs to be associated with fractures even when adjusted for presence of depression, based on observational studies.⁸⁰ The second found no experimental study data regarding falls and SSRIs.⁷⁹ Similarly, we did not find trial data for SSRIs and falls or fractures, although a single cohort study suggested an association with low SOE. This cohort study was not included in these prior reviews.

Recent systematic reviews in younger patients (<65y old) can inform how our findings compare to a younger population. Cipriani et al.⁸¹ evaluated safety as part of a large systematic review of 21 antidepressants, in patients 18y and older. Each of the 21 antidepressants were associated with increased drop outs due to adverse events versus placebo during treatment of the acute phase of MDD, including all of the therapies we reviewed in this report. Specific harms were not evaluated. A Cochrane review⁸² of antidepressants in primary care of patients under the age of 65 found the SSRIs citalopram and escitalopram were not associated with greater risk of adverse events versus placebo [RR 1.08 (0.96 to 1.22)] but did lead to more withdrawals due to harms [RR 2.05 (1.11 to 3.75). These findings were consistent with those in our review. Other than TCAs, additional antidepressants were not studied.

The Beers Criteria recommend that clinicians avoid prescribing SSRIs and TCAs in patients 65 years and older with a history of falls or fractures although note there may be situations where clinicians may decide use to be appropriate.²⁸ The evidence base supporting this particular recommendation is not focused on prescribing SSRIs or TCAs for a specific disease state but rather the use of the class of drugs in the older population generally.^83–86 Depression is a known risk factor for falls⁸⁷ in older adults thus confounding by indication may influence results of analyses evaluating treatment of depression on the outcome of falls. Our review only included studies of patients diagnosed with MDD thus baseline risk of falls due to depression presence should be similar across compared treatment arms. Clinicians should balance risks identified on treatment with risks that may remain present, such as falls, with untreated depression.

Applicability

This review exclusively included studies that required an age of 65 years or older. The included studies were consistent in excluding patients with multiple comorbidities or other psychological conditions, particularly patients with high suicide risk. None of the studies were specific to nursing facility residents. Unfortunately this limits applicability of results given that older adults commonly have multiple comorbidities and are taking several therapies concurrently. Resulting drug-drug interactions and pharmacokinetic changes must be taken into consideration when prescribing antidepressants.

The doses of antidepressants studied in this evidence base were rarely reflective of the full range cited in guidelines² or regulatory documents⁸⁸ as the usual dose range in older adults and was most often reflective of the lower half of that range. For example, in 30 active antidepressant arms of the 19 included RCTs, only 6 arms allowed doses that reflect the guideline suggested usual range for older adults. The rest of the treatment arms either limited dose to the lower limit of this range or allowed dosing in the lower half the this dosage range. Therefore, the data in this report does not reflect higher usual antidepressant doses that may be clinical utilized for effective treatment of MDD in this population.

Studies diagnosed major depression mostly using Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria and the severity of MDD in the population was moderate based on the mean Hamilton Depression Rating Scale (HAM-D) or Montgomery and Asberg Depression Rating Scale (MADRS) scores. The majority of trials evaluated the acute treatment period up to 12 weeks. Although we aimed to evaluate some therapies on a class basis (SSRI and SNRI), evidence for each drug within the class was not found thus results should not be extrapolated to the class. Concurrent treatments, when described, were usually as-needed therapies for sleep. It should be noted that the setting of focus was outpatient and did not include inpatient or urgent care scenarios.

Limitations

There are several limitations that pertain to the literature base of this review. No evidence was found for a number of the interventions of interest in this review, nor for many of the adverse events we aimed to analyze. Most of the available data featured comparison to placebo and few direct data were found to inform comparative harms of antidepressants. Even when studies were eligible for this review, the small number of trials and limited samples sizes posed an analytic challenge. As an example, the largest literature base was found for the comparison of SSRI to placebo (7 trials and 1 observational study) although for any given outcome, at most three trials were pooled.

None of the studies were powered to evaluate harms as they were all designed to assess efficacy. Interpretations of findings were made based on statistical significance, which may miss small differences due to inadequate power. Many outcomes suffered from the rareness of events where, for example, only one or two events occurred in one arm and zero in the other arm. In several other instances no events were reported in the literature base at all. It should not be assumed that a failure to find a difference means the given interventions are similar in adverse event profiles. The issue of sparse data throughout the evidence base was further complicated by the treatment phase which was being evaluated as most studies were specific to treatment of the acute phase of MDD (<12 weeks), but others evaluated only the continuation or maintenance periods. The least amount of data were available for these longer treatment periods. Furthermore, when studies did evaluate continuation or maintenance, they were considered to have higher risk of bias because open-label acute treatment periods were used and subjects experiencing adverse events were withdrawn prior to randomization into the longer treatment period. Thus, events were less likely to occur during the randomized period. A majority of the included RCTs, 11 of the 19 RCTs, disclosed industry sponsorship which has potential to introduce bias.⁸⁹

Most studies relied on spontaneous reporting of adverse events rather than active surveillance and it was difficult to determine if adverse outcomes were defined or pre-specified. Commonly we suspected selective outcome reporting because studies to state that certain measurements were part of the routine clinical monitoring (e.g. vitals, electrocardiogram) although none of these related outcomes were reported in the results. Little data exist regarding subgroups that are of interest in this field and although we sought to collect and analyze such data, only data regarding the impact of age and comorbidities were found.

A single, retrospective, population-based cohort study⁵⁷ was the single source of data identified for some intervention/outcome combinations and suggested associated harms. Although this study was very large and methodologically sound, residual confounding after adjustment for a considerable list of patient characteristics cannot be ruled out. For example, SSRIs and SNRIs were associated with falls. Although adjustments were made for dementia, antihypertensives, sedatives and hypnotics, and prior falls other factors such as hypotension were not included. Comparator subjects had depression diagnosed at some point although differences in depression severity, concomitant medical illnesses, and prior medication history between the populations compared cannot be excluded. Authors of this cohort study also stated that further biases inherent to observational designs such as channeling bias, confounding by indication, and residual confounding could have resulted in differences in patients that informed prescribing different antidepressants which could account for some of the associations seen in the study. Effect sizes for the reported harms were not large and dose-response relationships were not adjusted for. In many cases, this study was the only source of data (e.g. mirtazapine and trazodone) thus consistency of results is unknown. Authors of this cohort study themselves suggested that results should be confirmed in a long-term trial or meta-analysis of RCTs.

Research Gaps

There are several research gaps to address in order to more fully understand the adverse events associated with antidepressants in older patients with MDD. Other than SSRIs and SNRIs, we found no evidence for several therapies of interest. Even within the classes of SSRIs and SNRIs, some evidence is specific to a single drug within the class because others have not been studied. There were many outcomes (e.g. SIADH) that we sought to analyze that were not reported in the eligible studies, yet these are important to clinicians and descisionmakers according to the Key Informants, Technical Expert Panelists and partners on this project who helped shape the list of outcomes of interest. Aside from subgroup data based on age and one study that looked at influence of comorbidities, there were no data to evaluate the other subgroups of interests. Again, since these subgroups were identified largely by the stakeholders involved in this review, information about their influence is highly important doe the care of older depressed patients. Future studies should include these outcomes and subgroups important to the care of older adults and also account for other important factors such as nursing facility residence.

Aside from comparisons to placebo, limited data were available for direct comparisons among antidepressants. While a decision must first be made as to whether or not to treat MDD with antidepressants, with more severe depression the more telling decision is likely to be which antidepressant to prescribe, requiring assessment of comparative harms in addition to comparative efficacy. Thus, we believe this literature base overall would benefit from additional research to further characterize comparative harms of antidepressants.

Conclusions

In patients 65 years of age or older with MDD, treatment of the acute phase of MDD with SNRIs (duloxetine and venlafaxine) led to a greater number of adverse events compared with placebo while adverse events were statistically similar to placebo with SSRIs (escitalopram, fluoxetine), vortioxetine and bupropion. SSRIs (citalopram, escitalopram and fluoxetine) and SNRIs (duloxetine and venlafaxine) led to a greater number of study withdrawals due to adverse events compared with placebo and duloxetine increased the risk of falls. Further characterization of the comparative safety of antidepressants is difficult because few studies were identified, comparisons were based on statistical significance, trials were not powered to identify small difference in adverse events and observational studies may be confounded. Comparative, long-term, well-designed studies that report specific adverse events are needed to better inform decision making in this population.