Skip to content

Maintenance and relapse prevention

TL;DR — Continuing effective maintenance treatment materially reduces recurrence, but no option eliminates it. Across 22 discontinuation trials, maintenance reduced six-month recurrence of any episode from the discontinuation comparator by RR 0.61 (95% CI 0.54–0.70; NNTB 5), with stronger protection against manic/hypomanic/mixed recurrence than depressive recurrence (Kishi 2021, PMID 33046156). Lithium has the broadest and highest-quality evidence across both poles; lamotrigine is more depression-weighted, while several antipsychotics prevent predominantly the polarity on which participants were stabilized (Miura 2014, PMID 26360999; Bowden 2003, PMID 12695317; Calabrese 2003, PMID 14628976). BALANCE found lithium alone and lithium–valproate combination superior to valproate alone, but could not establish combination superiority over lithium (BALANCE 2010, PMID 20092882). Most pivotal trials use enrichment designs, last no more than two years and understate the complexity of adherence, residual symptoms, comorbidity and patient-defined recovery.

Maintenance is an active phase

Maintenance is not simply “continue everything.” It requires explicit choices about the episode just treated, prior predominant polarity, past response, adverse effects, monitoring, reproductive context, adherence feasibility and the cost of another recurrence. Acute efficacy and maintenance efficacy overlap imperfectly.

The most useful outcomes are time to recurrence, polarity-specific recurrence, intervention for a new episode, hospitalization, all-cause discontinuation and function. These endpoints differ: “intervention” depends on clinician behavior, while syndromal recurrence depends on a formal threshold.

What continuation prevents

Six-month outcome Maintenance vs discontinuation RR (95% CI) NNTB Interpretation
Any mood episode 0.61 (0.54–0.70) 5 One recurrence prevented per about five maintained patients
Depressive episode 0.72 (0.60–0.87) 13 Smaller average protective effect
Manic/hypomanic/mixed episode 0.45 (0.36–0.57) 6 Stronger protection against elevated-polarity recurrence
All-cause discontinuation 0.71 (0.61–0.82) 6 Maintenance also improved trial retention

These estimates come from 22 double-blind randomized trials with 5,462 participants across lithium, anticonvulsants, antipsychotics and long-acting formulations (Kishi 2021, PMID 33046156). They do not mean everyone who stops immediately relapses: 47.3% of discontinued participants remained recurrence-free at six months. That denominator matters for shared decisions and for designing safer taper studies.

Comparative evidence

Strategy Any-episode prevention Polarity pattern Major evidence limitation
Lithium Reduced any recurrence; strongest cross-polarity evidence in network synthesis Particularly strong for mania; also depression prevention Tolerability and monitoring burden (Miura 2014, PMID 26360999)
Lamotrigine Reduced intervention for any episode vs placebo Stronger for depressive than manic recurrence Lamotrigine-enriched stabilization phase (Bowden 2003, PMID 12695317; Calabrese 2003, PMID 14628976)
Valproate Withdrawal due to any mood episode RR 0.68 (95% CI 0.49–0.93) vs placebo Broad but less certain than lithium Six trials, n=876; limited blinding details (Cipriani 2013, PMID 24132760)
Lithium + valproate Primary event 54% over up to two years Broad Open label after combination run-in; not superior to lithium alone (BALANCE 2010, PMID 20092882)
Quetiapine adjunctive Overall relapse RR 0.38 (0.32–0.46) Only adjunctive SGA in one review reducing both manic and depressive episodes Responders pre-stabilized on quetiapine (Lindström 2017, PMID 28222360)
Aripiprazole adjunctive Overall relapse RR 0.65 (0.50–0.85) More mania-weighted Enrichment and limited depressive protection (Lindström 2017, PMID 28222360)
Ziprasidone adjunctive Overall relapse RR 0.62 (0.40–0.96) Primarily manic/mixed evidence Enrichment; wide interval (Lindström 2017, PMID 28222360)
Olanzapine / risperidone LAI Better than placebo for overall relapse in selected studies Usually mania-weighted Metabolic or prolactin/EPS burdens; enriched populations (Lindström 2017, PMID 28222360)

The 2014 network meta-analysis included 33 RCTs, 6,846 participants and 17 strategies. All assessed treatments significantly reduced any relapse versus placebo except aripiprazole RR 0.62 (95% credible interval 0.38–1.03), carbamazepine 0.68 (0.44–1.06), imipramine 0.95 (0.66–1.36) and paliperidone 0.84 (0.56–1.24) (Miura 2014, PMID 26360999). The 2021 network update found that most active treatments reduced any recurrence, but polarity-specific efficacy remained uneven (Kishi 2021, PMID 33177610).

Lithium, valproate and their combination

BALANCE randomized 330 people with bipolar I disorder after a four-to-eight-week combination run-in. The primary event—new intervention for an emergent mood episode—occurred in 54% on combination, 59% on lithium and 69% on valproate over up to 24 months (BALANCE 2010, PMID 20092882).

BALANCE comparison Hazard ratio (95% CI) Conclusion
Lithium + valproate vs valproate 0.59 (0.42–0.83) Combination superior
Lithium + valproate vs lithium 0.82 (0.58–1.17) No reliable difference
Lithium vs valproate 0.71 (0.51–1.00) Borderline advantage for lithium

The open-label design and combination run-in favor tolerators and responders, but the direction is clinically coherent: valproate monotherapy was the weakest of the three studied strategies. The Cochrane valproate review similarly found lithium–valproate combination more protective than valproate alone, RR 0.78 (95% CI 0.63–0.96), while valproate and lithium did not differ significantly in pooled direct comparisons, RR 1.02 (0.87–1.20) (Cipriani 2013, PMID 24132760).

Lamotrigine and polarity-specific prevention

Two 18-month trials illustrate why “maintenance efficacy” must be split by polarity.

Entry state Randomized sample Any-episode result Polarity signal
Recently manic/hypomanic 175 Lamotrigine P=.02 and lithium P=.006 vs placebo for time to intervention Lamotrigine delayed depression; lithium delayed mania/hypomania/mixed (Bowden 2003, PMID 12695317)
Recently depressed 463 Median time to intervention: lamotrigine 200 d, lithium 170 d, placebo 93 d Depression-free at 1 y 57%, 46%, 45%; mania-free 77%, 86%, 72% (Calabrese 2003, PMID 14628976)

Both trials first treated everyone openly with lamotrigine and randomized only those who stabilized. This enrichment improves assay sensitivity but answers “what happens in lamotrigine stabilizers?” more directly than “which drug should every patient start?”

An enriched-trial meta-analysis quantified the difference: lithium reduced any relapse RR 0.52 (95% CI 0.41–0.66), NNT 2.3 (1.6–4.2), based on two studies and 218 participants; lamotrigine RR 0.81 (0.70–0.93), NNT 8.3 (5.0–25.0), based on four studies and 706 participants (Oya 2019, PMID 31026388).

Antipsychotic maintenance

A review of 15 RCTs, 6,142 participants, found adjunctive aripiprazole, quetiapine and ziprasidone reduced overall relapse after successful stabilization. Quetiapine was the only adjunct that reduced both manic and depressive episodes in that analysis (Lindström 2017, PMID 28222360).

The same review identified a structural ceiling: nearly every trial pre-stabilized participants on the study drug, almost all enrolled bipolar I disorder, and no long-term RCT extended beyond two years (Lindström 2017, PMID 28222360). Apparent efficacy and tolerability therefore apply most directly to people who already responded and tolerated the drug.

An earlier 20-trial meta-analysis found no monotherapy significantly reduced both manic/mixed and depressive relapse; only quetiapine plus lithium/divalproex did so against placebo plus lithium/valproate (Vieta 2011, PMID 21733231). Differences in review eligibility and later evidence explain part of the apparent discrepancy.

Residual risk on treatment

Maintenance lowers risk; it does not create immunity. A single-group meta-analysis of 21 randomized lithium trials, n=1,415 and mean duration 78.4 weeks, estimated:

Outcome while continuing lithium Event rate (95% CI)
Any mood recurrence 39.8% (32.8–47.1)
Depressive recurrence 25.6% (18.8–34.0)
Manic/hypomanic/mixed recurrence 18.5% (13.7–24.7)
All-cause discontinuation 67.0% (57.2–75.5)
Discontinuation due to adverse events 8.7% (5.1–14.7)

These event rates should not be compared naively with untreated natural history because the component trials differed and often used enriched designs (Kishi 2020, PMID 32701902). They are useful for setting realistic expectations and justifying relapse plans even during adherent treatment.

Discontinuation and restart

Stopping a maintenance drug changes recurrence risk. Across drug classes, discontinuation for at least one month significantly increased recurrence, with divergence visible by six months and persisting in observations through 24 months (Kishi 2021, PMID 33046156).

Whether lithium becomes less effective after stopping and restarting is less certain. A review found two studies suggesting decreased effectiveness and three not; pooled odds of at least one relapse after interruption/restart versus continuous treatment were 1.40 (95% CI 0.85–2.31), P=.19, across 212 cases (de Vries 2013, PMID 23911110). The result does not prove safety of abrupt discontinuation; it says evidence for pharmacologic “refractoriness” after restart is inconclusive.

Psychosocial relapse prevention

Medication and psychosocial maintenance are complementary. Five-year follow-up of a 120-person randomized group-psychoeducation trial found 3.86 versus 8.37 recurrences and 154 versus 586 days acutely ill, favoring psychoeducation (Colom 2009, PMID 19252157). A broader psychological-therapy meta-analysis estimated relapse RR 0.74 (95% CI 0.64–0.85), with I²=43.3% (Lam 2009, PMID 19624386).

Relapse planning can therefore include:

Component Function Evidence link
Personal early-warning signs Detect polarity-specific prodromes Psychoeducation trials reduced recurrence (Colom 2009, PMID 19252157)
Sleep and routine monitoring Identify destabilization before syndromal threshold Embedded across psychoeducation and rhythm-focused models
Written action thresholds Convert warning signs into timely contact or treatment review Reduces reliance on impaired judgment during escalation
Family/support involvement by consent Adds observation and communication pathways Family-focused therapy reduced relapse HR to 0.38 in one RCT (Miklowitz 2003, PMID 12963672)
Adverse-effect monitoring Preserves tolerability and persistence Discontinuation is common even in trials (Kishi 2020, PMID 32701902)

See psychotherapy and self-management for intervention-specific evidence.

Evidence limitations

  • Enrichment designs preferentially randomize acute responders and tolerators.
  • “Time to intervention” reflects clinician and system behavior as well as illness recurrence.
  • Most antipsychotic maintenance RCTs stop by two years (Lindström 2017, PMID 28222360).
  • Bipolar II disorder, high comorbidity and reproductive transitions are underrepresented.
  • Trial adherence exceeds ordinary care, while discontinuation remains high.
  • Polarity-specific benefits are often obscured by a composite any-episode endpoint.

Absolute recurrence and enrichment

A 22-trial review (n=7,773; 24–104 weeks) found maintenance monotherapy collectively reduced new episodes versus placebo (OR 0.42, 95% CI 0.34–0.51). Aripiprazole, asenapine, lithium, olanzapine, quetiapine and long-acting risperidone each separated from placebo, but most trials randomized only people first stabilized on the study drug and active treatment favored mania prevention over depression prevention (Nestsiarovich 2022, PMID 34489127). Enrichment improves assay sensitivity while narrowing generalizability.

Valproate's maintenance estimate is favorable but smaller in evidence volume: across 11 trials (n=1,063), relapse of any mood episode was lower than placebo (RR 0.63, 95% CI 0.48–0.83), with no clear difference from lithium, olanzapine or lamotrigine (Mari 2024, PMID 39500601). Overlapping systematic reviews and reproductive toxicity must remain visible alongside efficacy.

For people stabilized on a mood stabilizer plus a second-generation antipsychotic, eight trials found continued combination therapy reduced six-month recurrence of any episode (RR 0.51, 95% CI 0.39–0.86), manic/hypomanic/mixed episodes (RR 0.42, 0.30–0.59) and depression (RR 0.39, 0.28–0.54) versus withdrawing the antipsychotic (Kishi 2021, PMID 33561884). This does not establish indefinite combination treatment because metabolic harms and follow-up beyond one year remain thin.

In a review identifying seven randomized trials (n=1,016; six risperidone, one flupenthixol), the risperidone-long-acting-injectable-versus-placebo comparison reduced study-defined relapse (RR 0.63) and manic relapse (RR 0.42) but increased prolactin-related events (RR 4.82) and weight gain (RR 3.80); the report gives P values rather than confidence intervals for these estimates. LAIs did not outperform oral medication overall; an advantage appeared only in rapid-cycling/high-relapse sensitivity analysis (RR 0.58), with different comparators (Kishi 2016, PMID 27207910). A 2026 synthesis of 34 heterogeneous studies estimated pooled relapse at 43.6% and associated sleep disturbance (OR 3.69), residual symptoms (OR 2.70) and rapid discontinuation (OR 1.98) with relapse, but these are prognostic associations rather than randomized treatment effects (Bai 2026, PMID 41176250).

Open questions

  • Which acute-response features identify people who can safely simplify combination therapy during maintenance (BALANCE 2010, PMID 20092882)?
  • Can non-enriched head-to-head trials reproduce the large quetiapine adjunctive relapse RR 0.38 (Lindström 2017, PMID 28222360)?
  • What taper rate minimizes recurrence while separating withdrawal phenomena from illness relapse (Kishi 2021, PMID 33046156)?
  • Can maintenance selection be prospectively matched to predominant polarity rather than retrospectively inferred (Bowden 2003, PMID 12695317; Calabrese 2003, PMID 14628976)?
  • Which digital or human-supported relapse plans preserve psychoeducation’s five-year benefit at lower delivery burden (Colom 2009, PMID 19252157)?

References

  1. Kishi T, et al. Recurrence rates in stable bipolar disorder patients after drug discontinuation versus drug maintenance: a systematic review and meta-analysis. Psychological Medicine. 2021. PMID 33046156
  2. Miura T, et al. Comparative efficacy and tolerability of pharmacological treatments in maintenance treatment of bipolar disorder. Lancet Psychiatry. 2014. PMID 26360999
  3. Kishi T, et al. Mood stabilizers and/or antipsychotics for bipolar disorder in the maintenance phase. Molecular Psychiatry. 2021. PMID 33177610
  4. BALANCE investigators and collaborators. Lithium plus valproate combination therapy versus monotherapy for relapse prevention in bipolar I disorder (BALANCE). Lancet. 2010. PMID 20092882
  5. Bowden CL, et al. A placebo-controlled 18-month trial of lamotrigine and lithium maintenance treatment in recently manic or hypomanic patients. Archives of General Psychiatry. 2003. PMID 12695317
  6. Calabrese JR, et al. A placebo-controlled 18-month trial of lamotrigine and lithium maintenance treatment in recently depressed patients. Journal of Clinical Psychiatry. 2003. PMID 14628976
  7. Oya K, et al. Efficacy and safety of lithium and lamotrigine for maintenance treatment: systematic review and meta-analysis. Neuropsychopharmacology Reports. 2019. PMID 31026388
  8. Lindström L, et al. Maintenance therapy with second generation antipsychotics for bipolar disorder: systematic review and meta-analysis. Journal of Affective Disorders. 2017. PMID 28222360
  9. Cipriani A, et al. Valproic acid, valproate and divalproex in maintenance treatment of bipolar disorder. Cochrane Database of Systematic Reviews. 2013. PMID 24132760
  10. Vieta E, et al. Effectiveness of psychotropic medications in maintenance phase of bipolar disorder: meta-analysis. International Journal of Neuropsychopharmacology. 2011. PMID 21733231
  11. Kishi T, et al. Recurrence of mania or depression among adult bipolar patients who continued using lithium. Journal of Clinical Psychopharmacology. 2020. PMID 32701902
  12. de Vries C, et al. Effectiveness of restarted lithium treatment after discontinuation. Bipolar Disorders. 2013. PMID 23911110
  13. Colom F, et al. Group psychoeducation for stabilised bipolar disorders: 5-year outcome of a randomised clinical trial. British Journal of Psychiatry. 2009. PMID 19252157
  14. Lam DH, et al. Psychological therapies in bipolar disorder: effect of illness history on relapse prevention. Bipolar Disorders. 2009. PMID 19624386
  15. Miklowitz DJ, et al. A randomized study of family-focused psychoeducation and pharmacotherapy in outpatient management of bipolar disorder. Archives of General Psychiatry. 2003. PMID 12963672
  16. Nestsiarovich A, et al. Preventing new episodes of bipolar disorder in adults: systematic review and meta-analysis of randomized controlled trials. Eur Neuropsychopharmacol. 2022;54:75–89. PMID 34489127
  17. Mari J, et al. The efficacy of valproate in acute mania, bipolar depression and maintenance therapy for bipolar disorder. BMJ Open. 2024;14:e087999. PMID 39500601
  18. Kishi T, et al. Effects of a conventional mood stabilizer alone or in combination with second-generation antipsychotics on recurrence in bipolar I disorder. Bipolar Disord. 2021;23:789–800. PMID 33561884
  19. Kishi T, et al. Long-acting injectable antipsychotics for prevention of relapse in bipolar disorder: a systematic review and meta-analysis. Int J Neuropsychopharmacol. 2016;19:pyw038. PMID 27207910
  20. Bai C, et al. Meta-analysis of recurrence rate and influencing factors of bipolar disorder. J Affect Disord. 2026;394:120570. PMID 41176250