The antidepressant class is a standing drug the wiki appraises for efficacy and its limitations,
not because it is a lifestyle lever but because a large share of the target population is on one, and it
is the default primary-care prescription for depression — so understanding what it does and does not do
IS a stratum-level decision. This page holds the class-level efficacy appraisal (the drug facet of the
depression cluster; the lifestyle levers — exercise, diet — and the exercise-vs-drug comparison live on
the cluster nucleus -> Depression and Modifiable Exposures). The limitations half — withdrawal /
discontinuation on stopping — is appraised below (Henssler 2024).
Scope: this is the efficacy/limitations appraisal of the class, IN scope as a standing drug. What is OUT (prescriber zone): which agent to prescribe for an individual, dosing/titration, screening. The per-agent ranking is reported below as a finding that it exists and is uncertain, never as a prescribing guide.
The headline finding — the class beats placebo, modestly
The landmark evidence is Cipriani 2018, a Bayesian network meta-analysis of 522 double-blind RCTs / 116,477 participants, 21 antidepressants (first- and second-generation) for the acute treatment of adults (>=18) with major depressive disorder, trials done 1979-2016. (Cipriani et al., 2018)
Every one of the 21 drugs was more effective than placebo on response rate: «all antidepressants were more effective than placebo, with ORs ranging between 2·13 (95% credible interval [CrI] 1·89–2·41) for amitriptyline and 1·37 (1·16–1·63) for reboxetine». (Cipriani et al., 2018)
But the effect size is modest. The pooled continuous effect across all antidepressants was «The random-effects summary SMD for all antidepressants was 0·30 (95% CrI 0·26–0·34; p<0·0001» (Cipriani et al., 2018) — a standardised mean difference of 0.30 (Cohen’s d) is a small-to-moderate effect. The authors state the motivating problem plainly: «short-term benefits are, on average, modest; and because long-term balance of benefits and harms is often understudied» (Cipriani et al., 2018).
Effect-estimate block
- effect_measure: response-rate OR 2.13 (1.89–2.41) [amitriptyline, largest] to 1.37 (1.16–1.63) [reboxetine, smallest], all 21 vs placebo; pooled continuous SMD 0.30 (0.26–0.34).
- population_and_comparator: adults with moderate-to-severe MDD (mean baseline HAM-D-17 25.7), vs placebo (the reference); the head-to-head sub-analysis compares drugs to each other.
- outcome: response rate = «a reduction of ≥50% of the total score on a standardised observer-rating scale for depression» — a scale-threshold surrogate, not remission, function, or a patient-important long-term endpoint -> Surrogate Outcomes.
- horizon: acute only — «We recorded the outcomes as close to 8 weeks as possible for all analyses», studied range 4–12 weeks. Says nothing about maintenance or long-term outcomes.
- certainty: «the certainty of evidence was moderate to very low» (GRADE); 46 (9%) of 522 trials high risk of bias, 380 (73%) moderate, 96 (18%) low.
- confidence: low — single (gold) source, a surrogate outcome, an acute horizon, and the class-beats- placebo direction far more secure than any per-agent magnitude.
Three load-bearing qualifications on “it works”
— the direction (class > placebo) is robust; each qualification narrows what that licenses.
- The outcome is a surrogate, measured acutely. Response = a ≥50% drop on a symptom rating scale at ~8 weeks. It is not sustained remission, restored function, or quality of life — the outcomes a person with depression most cares about. The authors could not even measure global functioning: «because of the paucity of information reported in the original studies, we were not able to quantify some outcomes, such as global functioning» (Cipriani et al., 2018). Treat the OR/SMD as an acute symptom-scale effect, not a life-outcome effect.
- The short horizon hides the harm side. «some of the adverse effects of antidepressants occur over a prolonged period, meaning that positive results need to be taken with great caution, because the trials in this network meta-analysis were of short duration» (Cipriani et al., 2018). An 8-week benefit estimate paired with adverse effects that accrue over months is an incomplete benefit–harm balance by construction.
- The certainty is moderate-to-very-low, and the between-drug ranking is the fragile part (next section).
Acceptability — the class is roughly as tolerable as placebo, and why that is a composite
Acceptability was measured as all-cause discontinuation — «treatment discontinuation measured by the proportion of patients who withdrew for any reason», chosen «because it encompasses efficacy and tolerability» (Cipriani et al., 2018).
On this composite, most antidepressants did not differ from placebo: «only agomelatine (OR 0·84, 95% CrI 0·72–0·97) and fluoxetine (0·88, 0·80–0·96) were associated with fewer dropouts than placebo, whereas clomipramine was worse than placebo (1·30, 1·01–1·68)» (Cipriani et al., 2018).
The composite nets a real harm against a real benefit — read it as a decomposition, not a clean bill. All-cause dropout mixes two opposing forces: patients leave less when a drug works (fewer inefficacy dropouts) and more when it causes side-effects (more tolerability dropouts). The separate adverse-event-dropout outcome shows the tolerability side is uniformly worse — all active drugs had higher withdrawal-for-adverse-events than placebo (ORs ~1.64–4.44), the sole exception agomelatine (Cipriani et al., 2018). So acceptability roughly equal to placebo does not mean as side-effect-free as placebo; it means the efficacy benefit roughly offsets the side-effect cost in the net stay-on-treatment measure. The harm is present; the composite absorbs it.
The per-agent ranking exists but is low-certainty — do not read it as a prescribing order
The robust finding is class-level (all 21 > placebo). The between-drug ordering is fragile, and Cipriani says so structurally:
- Differences between drugs are small with wide intervals: «differences in ORs between antidepressants ranged from 1·15 to 1·55 for efficacy and from 0·64 to 0·83 for acceptability, with wide CrIs on most of the comparative analyses» (Cipriani et al., 2018).
- A novelty-effect bias inflates newer drugs: «when a treatment was the novel or experimental drug of comparison, it appeared to be significantly more effective than when that same treatment was the older or control drug of comparison (difference 1·18-times, 95% CrI 1·09–1·27). Adjusting for this novelty effect diminished the differences between antidepressants» (Cipriani et al., 2018). Smaller and older studies also showed larger effects (a small-study/bias signal).
- Per-comparison GRADE certainty is low-to-very-low for many drugs.
Decision consequence: at the margin, the choice of which first-line antidepressant is weakly evidenced — the class effect is real and the ordering is mostly noise plus novelty bias. That is a per-agent selection question (prescriber zone), not a fabric claim; this page records only that the ranking is uncertain, and deliberately does not table the 21 drugs as a recommendation.
What this class-efficacy finding does NOT establish (named gaps)
from the source’s own stated limits — the frontier of the standing-drug appraisal:
- Withdrawal / discontinuation on stopping. Cipriani explicitly did not cover it: «We did not cover important clinical issues that might inform treatment decision making in routine clinical practice (eg, specific adverse events, withdrawal symptoms, or combination with non-pharmacological treatments)» (Cipriani et al., 2018). This is the limitations half of the standing-drug appraisal, now supplied by Henssler 2024 -> Limitations — discontinuation and withdrawal (Henssler 2024) below.
- Long-term efficacy and the full benefit–harm balance — the acute horizon leaves maintenance and long-accruing harms unmeasured (above).
- Individual-level effect modifiers. «We analysed only average treatment effects and were not able to investigate potentially important clinical and demographical modifiers of treatment response at the individual patient level (eg, age, sex, severity of symptoms, or duration of illness)» (Cipriani et al., 2018). So there is no basis here for route-(b) personalization of drug choice.
- Treatment-resistant depression / next step after failure. «Given the modest effect sizes, non-response to antidepressants will occur. Our information unfortunately cannot guide next-step choices after failure of such a first step (ie, they do not apply to treatment-resistant depression)» (Cipriani et al., 2018). The finding is for initial treatment only; the excluded populations (psychotic, treatment-resistant) bound transportability.
Limitations — discontinuation and withdrawal (Henssler 2024)
The limitations half of the standing-drug appraisal: what happens on stopping. Cipriani named this as uncovered; Henssler 2024 supplies it — a gold SR+MA (79 studies, 44 RCTs, 21 002 patients) that is the first to quantify discontinuation-symptom incidence and the first to include placebo-discontinuation arms as the nocebo comparator. (Henssler et al., 2024) This refines the efficacy page rather than contradicting it (type-F): the two sources answer different sub-questions — Cipriani, does the class work acutely; Henssler, how hard is it to stop — so their numbers are not commensurable and are never netted against each other (Cipriani’s efficacy SMD/OR is a symptom-scale effect vs placebo on treatment; Henssler’s figures are withdrawal-symptom incidence proportions on stopping — different quantities, different construct).
The load-bearing move — separate raw incidence from the drug-attributable fraction
The raw discontinuation-symptom rate massively over-states the pharmacological withdrawal burden, because it folds in nocebo/expectation effects and re-emerging depression. The placebo-discontinuation arm is what isolates the part attributable to the drug.
- Raw incidence (any symptom): ~1 in 3, but this is the inflated number. «Incidence of at least one antidepressant discontinuation symptom was 0·31 (95% CI 0·27–0·35) in 62 study groups after discontinuation of antidepressants, and 0·17 (0·14–0·21) in 22 study groups after discontinuation of placebo. Between antidepressant and placebo groups of included RCTs, the summary difference in incidence was 0·08 [0·04–0·12].» (Henssler et al., 2024) So ~1 in 6 people report discontinuation-like symptoms after stopping an inert placebo — the nocebo/non-specific floor.
- Drug-attributable (the decision-relevant number): ~15%, one in six to seven. «Subtracting non- specific effects, we estimate the frequency of antidepressant discontinuation symptoms to be in the range of approximately 15%, thus affecting about one in six to seven patients.» (Henssler et al., 2024) Two anchor points bracket it: the clean within-RCT drug-minus-placebo risk difference is 0.08 (0.04-0.12); the broader raw-minus-placebo gap is ~15% (the authors’ headline). Read ~8-15%, not 31%.
- The nocebo share is large — about half. «Considering the placebo results, approximately half of antidepressant discontinuation symptoms could be attributable to expectation or non-specific symptoms» — placebo-arm patients «might expect to be discontinuing from verum and thus would be susceptible to nocebo effects.» (Henssler et al., 2024) Same design logic as the statin-nocebo isolation -> The Observational-Trial Discordance (an inert-pill arm isolates the nocebo component of a side-effect).
- Severe symptoms are uncommon: ~1 in 35, drug-attributable ~1 in 45. «The incidence of severe antidepressant discontinuation symptoms after discontinuation of an antidepressant was 0·028 (0·014–0·057) compared with 0·006 (0·002–0·013) after discontinuation of placebo.» (Henssler et al., 2024) The 0.028 vs 0.006 gap is the drug-attributable severe fraction; the authors call the ~1-in-35 severe figure preliminary.
Three qualifications on the withdrawal numbers
— each narrows what the incidence figures license.
- The ~15% is possibly an over-estimate, not a floor. «we conservatively estimate that one out of six to seven patients has truly pharmacologically caused antidepressant discontinuation symptoms. This might still be an over-estimate, as it is difficult to factor in residual or re-emerging symptoms of depression or anxiety.» (Henssler et al., 2024) Re-emerging depression masquerades as withdrawal and cannot be cleanly subtracted, so the true pharmacological share sits at or below ~15%.
- The measurement instrument manufactures symptoms — a streetlight artifact. Structured checklists (e.g. DESS, >40 items) raised the count even in placebo arms: «the use of established instruments in placebo discon tinuation studies also resulted in higher antidepressant discontinuation symptoms event rates, indicating the risk of false positive signals.» (Henssler et al., 2024) A >40-symptom checklist over a non-specific symptom set (dizziness, headache, nausea, insomnia) catches base-rate complaints; the tool inflates the estimate independent of any drug. Heterogeneity was substantial throughout (I² ~94%, wide prediction intervals), so a single trial can land far from the pooled rate.
- Study-level tapering showed no effect here — but that does NOT refute individual hyperbolic taper. «We did not find a difference between studies that applied tapering of the drug and studies with abrupt cessation of the drug. The substantial heterogeneity in study designs (eg, duration of taper) and specific antidepressants used preclude firm conclusions; for example, all trials on venlafaxine and desvenlafaxine were among the studies that applied tapering.» (Henssler et al., 2024) This is an indirect, study-level comparison (ecological-fallacy risk — high-withdrawal agents happen to be the tapered ones); it does not overturn the separate within-patient evidence that prolonged/hyperbolic tapering reduces (not eliminates) withdrawal. Which taper regimen is prescriber-zone and not appraised here.
Decision consequence
For the counselling decision at stopping: the honest number to give a patient is ~1 in 6 to 7 will have genuinely drug-caused discontinuation symptoms and ~1 in 35 a severe one — not the 1-in-2-or-more figures from selection-biased surveys the authors set out to correct, and not the raw 1-in-3 (which counts nocebo and returning illness). Higher caution for imipramine, paroxetine, and venlafaxine/desvenlafaxine; the per-agent ordering is otherwise preliminary (wide CIs; mirtazapine, bupropion, amitriptyline unstudied). This is a stratum-level appraisal of a standing drug’s limitation, not a discontinuation protocol; tapering-regimen selection stays prescriber-zone.
Substitution — where the class sits against the realistic alternatives
The telos frames a drug as the realistic comparator to lifestyle levers. For depression the realistic alternatives to a first-line antidepressant are psychotherapy and exercise, both appraised on the cluster nucleus -> Depression and Modifiable Exposures. Two guards on any such comparison:
- Comparator mismatch — do not weld the numbers. Cipriani’s OR/SMD are vs placebo; the exercise network’s SSRI estimate (g ~−0.26) is vs an active control and comes from a review not designed to find all drug trials. These are different quantities on different reference groups, so the head-to-head exercise-beats-SSRIs reading lives on the nucleus page with its own caveats and must not be restated here as if Cipriani supported it.
- Non-substitutable where it is non-substitutable. A drug that delivers an acute symptom-scale response does not thereby substitute for the pleiotropic other-channel value of exercise (cardiometabolic, function); the substitution is outcome-specific. Conversely a modest acute drug effect is a real option for someone unwilling or unable to sustain a behavioural lever — the choice stays the person’s (layer 3).
Open loop: nothing here grades the class against a realized long-term patient outcome. The evidence is acute, symptom-scale, moderate-to-very-low certainty; the trajectory and functional outcomes depression most degrades are under-measured -> Surrogate Outcomes.