Companion reviews: Depression — antidepressants and the evidence, ADHD — stimulants and the evidence, Antipsychotics and the evidence, Placebo, unblinding and the evidence, Psychotherapy comparators and the evidence, Diagnosis thresholds and the evidence, Anxiety — a critical review of the evidence, and Stress, Energy and the Capacity to Function.

---

Executive summary

Two folk models dominate public discussion of lithium and mood stabilizers: that lithium is a highly effective, specific, and life-saving treatment with unique antisuicidal properties that should be used more widely (the clinical-consensus narrative), and its inversion, that lithium is a toxic relic causing kidney failure and thyroid disease, replaced by safer newer drugs (the critical narrative). Both are overclaims, and the evidence supports neither in the form presented.

Eight things the evidence actually establishes:

1. Lithium is effective for acute mania versus placebo, and shows no significant difference versus valproate in head-to-head trials. McKnight et al. (2019) Cochrane review of 36 RCTs (4,220 participants) found lithium versus placebo response OR 2.13 (95% CI 1.73–2.63), 6 trials, 1,707 participants. Lithium versus valproate OR 1.22 (95% CI 0.87–1.70), 5 trials, 607 participants—confidence interval includes 1.0, no evidence of difference.

2. In maintenance treatment, lithium monotherapy is superior to valproate monotherapy, and combination therapy is superior to valproate alone. The BALANCE trial (typeset abstract) randomized 330 patients with bipolar I disorder to lithium, valproate, or combination therapy for up to 24 months. Primary outcome events: 59/110 combination, 65/110 lithium, 76/110 valproate (typeset abstract prints three valproate deaths). Hazard ratios: combination versus valproate HR 0.59 (95% CI 0.42–0.83, p = 0.0023); lithium versus valproate HR 0.71 (95% CI 0.51–1.00, p = 0.0472); combination versus lithium HR 0.82 (95% CI 0.58–1.17, p = 0.27) (Geddes et al., 2010 typeset abstract).

3. Early meta-analyses found lithium reduced suicide risk versus placebo and versus active comparators, but a recent updated meta-analysis does not reach statistical significance. [contested] Cipriani et al. (2013) meta-analysis of 48 RCTs (6,674 participants) found lithium versus placebo Peto OR 0.13 (95% CI 0.03–0.66) for suicide. Unipolar depression subgroup: suicide OR 0.13 (95% CI 0.02–0.76); all-cause death OR 0.36 (95% CI 0.13–0.98). Cipriani et al. (2005) found suicide versus active comparators Peto OR 0.26 (95% CI 0.09–0.77), 7 trials, 2 suicides on lithium versus 11 on other compounds. However, Riblet et al. (2022) updated meta-analysis of 7 RCTs (568 lithium, 570 control) found suicide 2/568 versus 8/570, Peto OR 0.30 (95% CI 0.09–1.02), P = 0.05, not statistically significant.

4. Lamotrigine is effective for bipolar depression and prevents depressive episodes in maintenance treatment. Geddes et al. (2009) individual patient data meta-analysis of 1,072 participants from 5 RCTs found lamotrigine versus placebo response RR 1.27 (95% CI 1.09–1.47) on HRSD. Baseline severity interaction: HRSD > 24 RR 1.47 (95% CI 1.16–1.87); HRSD ≤ 24 RR 1.07 (95% CI 0.90–1.27). Kishi et al. (2021) network meta-analysis of 41 maintenance RCTs (9,821 participants) found lamotrigine beats placebo on preventing any episode and preventing depressive episodes, but NOT on preventing manic episodes.

5. Lithium increases risk of hypothyroidism, reduces urinary concentrating ability, and is associated with increased risk of CKD stage 3. McKnight et al. (2012) meta-analysis of 385 studies found lithium versus placebo hypothyroidism OR 5.78 (95% CI 2.00–16.67); TSH increase 4.00 iU/mL; urinary concentrating ability -158.43 mOsm/kg; GFR -6.22 mL/min (CI crosses zero, not significant). Shine et al. (2015) cohort study of 4,678 lithium-tested patients versus 689,228 controls found CKD stage 3 HR 1.93 (95% CI 1.76–2.12); hypothyroidism HR 2.31 (95% CI 2.05–2.60).

6. Lithium causes weight gain, but less than olanzapine. McKnight et al. (2012) found lithium versus placebo OR 1.89 (95% CI 1.27–2.82) for weight gain. Versus olanzapine, lithium OR 0.32 (95% CI 0.21–0.49).

7. Valproate is effective for acute mania versus placebo. Jochim et al. (2019) Cochrane review found valproate versus placebo response 45% versus 29%, OR 2.05 (95% CI 1.32–3.20), 4 studies, 869 participants.

8. Lithium toxicity occurs near the therapeutic range. FDA-approved prescribing information (2018) boxed warning: lithium toxicity is closely related to serum concentrations and can occur at doses close to therapeutic concentrations. Therapeutic range 0.8–1.2 mEq/L; toxicity typically ≥ 1.5 mEq/L.

What the evidence does not establish:

• That lithium definitively reduces suicide risk (early meta-analyses found significant effects; Riblet 2022 update does not reach significance)
• Whether lithium causes clinically significant GFR decline in most patients (McKnight 2012 GFR CI crosses zero; Shine 2015 found CKD stage 3 HR 1.93, but detection bias possible)
• Which patients will respond to lithium versus other mood stabilizers
• The mechanism by which lithium prevents relapse or reduces suicide risk
• Long-term functional outcomes beyond relapse prevention

The structural problem that conditions all of this: The BALANCE trial is open-label, and the superiority of lithium over valproate may reflect detection bias. The suicide prevention evidence was significant in 2013 (OR 0.13) but not in 2022 (OR 0.30, P = 0.05). The point estimate remains consistent, but the confidence interval now crosses 1.0. The GFR finding in McKnight 2012 is not statistically significant. The CKD finding in Shine 2015 is significant, but detection bias (more monitoring of lithium users) may account for it. The folk models—lithium as miracle drug or lithium as toxic relic—both overclaim. The evidence supports efficacy for acute mania and maintenance, established thyroid and urinary effects, cohort evidence for CKD risk, and contested suicide prevention.

---

How to read this review

The single most important heuristic

An efficacy claim without a named trial duration, patient population, comparison group, and outcome measure is not a clean estimate. The BALANCE trial shows lithium superior to valproate over 24 months (HR 0.71, open-label). McKnight 2019 Cochrane shows lithium versus valproate OR 1.22 (CI includes 1) for acute mania response. Cipriani 2013 shows lithium versus placebo suicide Peto OR 0.13; Riblet 2022 shows OR 0.30 (P = 0.05, not significant). Any claim that "lithium works" must specify which outcome, in which population, over what duration, versus what comparator.

Three structural problems

Open-label maintenance trial and detection bias. The BALANCE trial was open-label. Patients and clinicians knew which medication was assigned. This introduces detection bias: clinicians may intervene earlier for relapse in the valproate group, or patients on lithium may report symptoms differently. The hazard ratios may reflect these biases as well as true drug effects.

Suicide as secondary outcome, and changing significance with additional trials. [contested] The suicide prevention evidence comes from meta-analyses of trials where suicide was not the primary outcome. Cipriani 2005 pooled 7 trials with 13 total suicides (2 on lithium, 11 on comparators), OR 0.26. Cipriani 2013 pooled 48 trials, OR 0.13, significant. Riblet 2022 pooled 7 trials with 10 total suicides (2/568 lithium, 8/570 control), OR 0.30, P = 0.05, not significant. The point estimate remains consistent (0.13 to 0.30), but the confidence interval now crosses 1.0.

GFR measurement precision and CKD detection bias. McKnight 2012 found GFR -6.22 mL/min (95% CI -14.65 to 2.20), not significant. The CI crosses zero. Shine 2015 cohort found CKD stage 3 HR 1.93 (1.76–2.12), significant, but lithium users are monitored more frequently than controls, introducing detection bias.

Conventions

Confidence intervals are omitted where they could not be verified against the primary text, rather than reconstructed. [contested] marks findings where the literature genuinely disagrees or significance has changed with additional trials.

---

Part I — Acute mania: lithium and valproate both effective versus placebo

1.1 McKnight 2019 Cochrane: lithium versus placebo and versus valproate

McKnight et al. (2019) conducted a Cochrane systematic review of lithium for acute mania. Search up to January 2017. Included 36 randomized controlled trials, 4,220 participants. Population: all ages, all settings with acute mania.

Lithium versus placebo — response:

OR 2.13 (95% CI 1.73–2.63), 6 trials, 1,707 participants. Moderate-certainty evidence that lithium is more effective than placebo for acute mania.

Lithium versus valproate — response:

OR 1.22 (95% CI 0.87–1.70), 5 trials, 607 participants. Moderate-certainty evidence. Confidence interval includes 1.0. No evidence of difference between lithium and valproate.

Adverse events: When compared to placebo, lithium associated with increased somnolence and tremor. Insufficient data for high-certainty conclusions on adverse events versus other drugs.

Translation: Lithium is effective for acute mania versus placebo (OR 2.13). Lithium and valproate show comparable efficacy in head-to-head trials (OR 1.22, CI includes 1).

1.2 Jochim 2019 Cochrane: valproate versus placebo

Jochim et al. (2019) conducted a Cochrane review of valproate for acute mania. Included studies: valproate versus placebo in adults.

Response rate (adults):

Valproate 45% versus placebo 29%. OR 2.05 (95% CI 1.32–3.20). 4 studies, 869 participants. High-quality evidence.

Translation: Valproate is effective for acute mania versus placebo. The effect size (OR 2.05) is comparable to lithium versus placebo (OR 2.13 from McKnight 2019).

1.3 Network meta-analysis: lithium versus placebo response

Kishi et al. (2022) conducted a network meta-analysis of pharmacological treatments for bipolar mania. Published online 2021 as Kishi et al. (2021). 72 double-blind RCTs, 16,442 participants, mean duration 3.96 ± 2.39 weeks.

Lithium versus placebo — response:

RR 1.259 (95% CI 1.007–1.576). Lithium significantly better than placebo for response.

Translation: The network meta-analysis confirms lithium's efficacy for acute mania versus placebo. The effect size (RR 1.26) is modest but significant. Trials are short (mean 3.96 weeks) and exclude patients with rapid cycling, mixed features, or treatment resistance.

---

Part II — Maintenance treatment: lithium superior to valproate, combination superior to valproate alone

2.1 The BALANCE trial: lithium versus valproate versus combination

Geddes et al. (2010) conducted the BALANCE trial (Lithium plus valproate combination therapy versus monotherapy for relapse prevention in bipolar I disorder), a randomized open-label trial at 41 sites across UK, France, USA, and Italy.

Design: 330 patients with bipolar I disorder (aged ≥ 16 years) randomized to lithium monotherapy (plasma concentration 0.4–1.0 mmol/L), valproate monotherapy (750–1250 mg), or combination therapy (lithium plus valproate). Follow-up: up to 24 months.

Primary outcome: Initiation of new intervention for an emergent mood episode (relapse requiring additional treatment).

Results (from typeset abstract):

• Combination therapy: 59 of 110 patients experienced primary outcome event
• Lithium monotherapy: 65 of 110 patients experienced primary outcome event
• Valproate monotherapy: 76 of 110 patients experienced primary outcome event (typeset abstract prints three valproate deaths)

Hazard ratios (from typeset abstract):

• Combination versus valproate: HR 0.59 (95% CI 0.42–0.83, p = 0.0023). Combination therapy reduced relapse risk by 41% compared to valproate.
• Lithium versus valproate: HR 0.71 (95% CI 0.51–1.00, p = 0.0472). Lithium reduced relapse risk by 29% compared to valproate.
• Combination versus lithium: HR 0.82 (95% CI 0.58–1.17, p = 0.27). No significant difference.

Interpretation: Both combination therapy and lithium monotherapy are more likely to prevent relapse than valproate monotherapy over 24 months. The benefit is maintained irrespective of baseline severity and is most apparent in prevention of manic relapse. BALANCE could neither confirm nor refute benefit of combination therapy over lithium monotherapy.

Critical limitation: The trial was open-label, not double-blind. Patients and clinicians knew which treatment was assigned. This introduces detection bias: clinicians may intervene earlier for relapse in the valproate group, or patients on lithium (viewed as more effective) may report symptoms differently. The hazard ratios may reflect these biases as well as true drug effects.

2.2 Network meta-analysis: lithium effective across all maintenance outcomes

Kishi et al. (2021) conducted a systematic review and network meta-analysis of 41 RCTs (9,821 participants, mean study duration 70.5 ± 36.6 weeks) comparing mood stabilizers and antipsychotics for maintenance treatment of bipolar disorder.

Recurrence/relapse of any mood episode (primary outcome):

Lithium listed as significantly better than placebo. Most active treatments were better than placebo, except carbamazepine, lamotrigine plus valproate, and paliperidone.

Recurrence/relapse of depressive episodes:

Better than placebo: aripiprazole plus valproate, lamotrigine, lamotrigine plus valproate, lithium, olanzapine, quetiapine. NOT better: carbamazepine, valproate alone.

Recurrence/relapse of manic/hypomanic/mixed episodes:

Better than placebo: lithium, valproate, antipsychotics. NOT better: aripiprazole plus valproate, carbamazepine, lamotrigine, lamotrigine plus valproate.

Translation: Lithium is effective for preventing any mood episode, preventing depressive episodes, and preventing manic episodes. It is the only mood stabilizer with evidence across all three outcomes. Valproate prevents manic episodes but has limited evidence for preventing depressive episodes. Lamotrigine prevents depressive episodes but not manic episodes.

---

Part III — Lamotrigine: effective for bipolar depression, prevents depressive episodes but not manic episodes

3.1 Acute bipolar depression: individual patient data meta-analysis

Geddes et al. (2009) conducted an individual patient data meta-analysis of 1,072 participants from 5 RCTs comparing lamotrigine versus placebo for bipolar depression.

Response rate — HRSD:

RR 1.27 (95% CI 1.09–1.47). More individuals responded to lamotrigine than placebo. Significant benefit.

Interaction by baseline severity:

• HRSD > 24 (severe): RR 1.47 (95% CI 1.16–1.87). Significant.
• HRSD ≤ 24 (mild-moderate): RR 1.07 (95% CI 0.90–1.27). Not significant.

Translation: Lamotrigine was superior to placebo in people with more severe depression (HRSD > 24), but not in mild-moderate depression. This severity interaction matches findings for antidepressants in unipolar depression. The pooled individual patient data showed a significant effect that individual trials (underpowered for modest effect sizes) missed.

3.2 Maintenance: prevents depressive relapses, not manic relapses

Kishi et al. (2021) network meta-analysis of 41 maintenance RCTs (9,821 participants) found:

Lamotrigine versus placebo:

• Preventing any mood episode: Significant.
• Preventing depressive episodes: Significant.
• Preventing manic episodes: NOT significant.

Translation: Lamotrigine's maintenance benefit is driven by preventing depressive episodes, not manic episodes. This polarity-specific efficacy is robust across multiple meta-analyses. Lamotrigine is ineffective for acute mania (not tested in Geddes 2009 or Kishi 2021 acute trials) and does not prevent manic relapses. It is effective for acute bipolar depression (especially severe) and prevents depressive relapses.

---

Part IV — Suicide prevention: early meta-analyses found reduced risk, recent update not significant

4.1 The Cipriani 2013 BMJ update: lithium Peto OR 0.13 versus placebo

Cipriani et al. (2013) conducted an updated systematic review and meta-analysis of lithium in the prevention of suicide in mood disorders. Search up to January 2013. Included 48 randomized controlled trials, 6,674 participants. Population: people with unipolar and bipolar mood disorders.

Lithium versus placebo — suicide (from Figures 5 and 6, corrected):

Peto OR 0.13 (95% CI 0.03–0.66). Lithium more effective than placebo in reducing suicide.

Unipolar depression subgroup (from Figures 5 and 6):

• Suicide: OR 0.13 (95% CI 0.02–0.76). Lithium associated with reduced suicide in unipolar depression.
• All-cause death: OR 0.36 (95% CI 0.13–0.98). Reduced deaths overall in unipolar depression subgroup.

(Note: The abstract swapped the suicide and death ORs for unipolar depression. Figures 5 and 6 print the correct values.)

Lithium versus active comparators:

Only statistically significant difference versus carbamazepine for deliberate self-harm. Lithium generally better than other active comparators with small statistical variation.

Mechanism: The authors suggested lithium may exert antisuicidal effects by (1) reducing relapse of mood disorder, and (2) decreasing aggression and possibly impulsivity.

Translation: The 2013 update provides evidence for lithium's antisuicidal effect versus placebo. Peto OR 0.13 means the odds of suicide on lithium are 0.13 times the odds on placebo (an odds ratio, not a risk reduction). However, suicide was not the primary outcome in these trials—trials were designed to measure symptom reduction or relapse prevention, with suicide recorded as an adverse event.

4.2 The Cipriani 2005 meta-analysis: lithium OR 0.26 versus active comparators

Cipriani et al. (2005) conducted a systematic review and meta-analysis of 32 randomized controlled trials (1,389 patients on lithium, 2,069 on other compounds). Population: patients with mood disorders (unipolar depression, bipolar disorder, schizoaffective disorder, dysthymia, rapid cycling).

Suicide: 7 trials. Lithium: 2 suicides. Other compounds: 11 suicides. Peto OR 0.26 (95% CI 0.09–0.77). Patients on lithium less likely to die by suicide.

Suicide plus deliberate self-harm (composite measure): OR 0.21 (95% CI 0.08–0.50). Lower in lithium group.

All-cause mortality: 11 trials. Lithium: 9 deaths. Other compounds: 22 deaths. OR 0.42 (95% CI 0.21–0.87). Fewer deaths overall in lithium group.

Translation: The 2005 review found lithium reduced suicide, self-harm, and all-cause mortality versus active comparators. The suicide finding is based on 13 total events (2 on lithium, 11 on comparators) from 7 trials.

4.3 The Riblet 2022 update: effect size consistent, but no longer statistically significant

[contested] Riblet et al. (2022) conducted an updated systematic review and meta-analysis including 7 RCTs (568 allocated to lithium, 570 to control—placebo or usual care). Population: adults with major depressive disorder or bipolar disorder.

Suicide:

2 suicides on lithium (2/568) versus 8 on control (8/570). Peto OR 0.30 (95% CI 0.09–1.02), P = 0.05. Lower for lithium versus control, but NOT statistically significant. The paper states this is not statistically significant.

Translation: The point estimate (0.30) remains consistent with a protective effect, suggesting approximately 70% reduction in suicide risk. However, with additional trials, the confidence interval now crosses 1.0, meaning the result is not statistically significant at P < 0.05. The P-value is exactly 0.05, borderline. This does not prove lithium does not prevent suicide—it means the effect size is smaller or more variable than early trials suggested, and larger trials are needed to resolve the uncertainty.

---

Part V — Adverse effects: hypothyroidism, CKD risk, renal function, weight, and toxicity threshold

5.1 Thyroid: OR 5.78 for hypothyroidism

McKnight et al. (2012) conducted a systematic review and meta-analysis of lithium toxicity, including 385 studies. Population: patients with mood disorders receiving lithium.

Clinical hypothyroidism: OR 5.78 (95% CI 2.00–16.67). Prevalence increased versus placebo. Significant increase.

Thyroid stimulating hormone (TSH): Mean increase 4.00 iU/mL (95% CI 3.90–4.10). Significantly increased on average.

Translation: Lithium substantially increases risk of hypothyroidism, with nearly 6-fold increased odds. TSH increases by 4.00 iU/mL on average.

Shine et al. (2015) cohort study (below) also found hypothyroidism HR 2.31 (95% CI 2.05–2.60) versus controls, consistent with McKnight 2012.

5.2 Renal function: reduced concentrating ability established, GFR reduction not significant in meta-analysis, but CKD stage 3 increased in cohort study

McKnight et al. (2012) found:

Glomerular filtration rate (GFR): Mean reduction -6.22 mL/min (95% CI -14.65 to 2.20). NOT statistically significant. The confidence interval crosses zero: the true effect could range from a 14.65 mL/min decline to a 2.20 mL/min increase.

Urinary concentrating ability: Weighted mean difference -158.43 mOsm/kg (95% CI -229.78 to -87.07). 15% of normal maximum reduction. Statistically significant.

Renal failure risk: 18 of 3,369 patients (0.5%) received renal replacement therapy.

Translation: Lithium definitively reduces urinary concentrating ability by 15%, a clinically significant effect leading to polyuria. Evidence for GFR reduction is not statistically significant in McKnight 2012 (mean -6.22 mL/min, confidence interval crosses zero). The absolute risk of renal replacement therapy is low (0.5%).

Shine et al. (2015) conducted a retrospective cohort study using UK primary care data. Population: 4,678 patients with at least one lithium test between 2004 and 2008, versus 689,228 age/sex-matched controls.

CKD stage 3 (eGFR < 60 mL/min/1.73 m²):

HR 1.93 (95% CI 1.76–2.12). Lithium-tested patients had nearly twice the hazard of developing CKD stage 3 compared to controls.

Hypothyroidism:

HR 2.31 (95% CI 2.05–2.60). Consistent with McKnight 2012 meta-analysis.

Limitations: Detection bias: lithium users are monitored more frequently than controls, leading to earlier detection of CKD. The control group was matched on age and sex but not on psychiatric diagnosis or other confounders. This is observational data, not randomized.

Translation: The cohort study found a significant association between lithium use and CKD stage 3 (HR 1.93). However, detection bias (more frequent monitoring of lithium users) may account for part or all of this association. The McKnight 2012 meta-analysis of RCTs found GFR reduction not statistically significant. The honest summary: lithium definitively reduces urinary concentrating ability; whether it causes clinically significant GFR decline or CKD in most patients remains uncertain due to conflicting evidence from RCTs (not significant) and observational cohorts (significant, but detection bias possible).

5.3 Weight gain and parathyroid effects

McKnight et al. (2012) found:

Weight gain versus placebo: OR 1.89 (95% CI 1.27–2.82). Significantly more weight gain.

Weight gain versus olanzapine: OR 0.32 (95% CI 0.21–0.49). LESS weight gain than olanzapine.

Blood calcium: Mean increase 0.09 mmol/L (95% CI 0.02–0.17). Associated with increased calcium.

Parathyroid hormone (PTH): Mean increase 7.32 pg/mL (95% CI 3.42–11.23). Significantly increased. Increased risk of hyperparathyroidism.

Translation: Lithium causes weight gain versus placebo (OR 1.89), but less than olanzapine. Lithium increases calcium and parathyroid hormone, raising risk of hyperparathyroidism.

5.4 FDA boxed warning: toxicity near therapeutic range

FDA-approved prescribing information for lithium carbonate (2018) includes a boxed warning titled "LITHIUM TOXICITY":

Lithium toxicity is closely related to serum lithium concentrations and can occur at doses close to therapeutic concentrations.

Serum concentrations should be maintained between 0.8 and 1.2 mEq/L during acute treatment. Facilities for prompt and accurate serum lithium determinations should be available before initiating therapy.

Toxicity typically occurs at serum lithium concentrations ≥ 1.5 mEq/L, but may occur at lower levels depending on individual sensitivity.

Translation: The FDA boxed warning states lithium toxicity is closely related to serum concentrations and can occur at doses close to therapeutic concentrations.

---

Part VI — What the evidence does not establish

6.1 That lithium definitively reduces suicide risk

[contested] The Cipriani 2013 meta-analysis found Peto OR 0.13 (95% CI 0.03–0.66), statistically significant. The Riblet 2022 update found Peto OR 0.30 (95% CI 0.09–1.02), P = 0.05, not statistically significant. The point estimate remains consistent (0.13 to 0.30), suggesting a protective effect, but the confidence interval now crosses 1.0.

This does not prove lithium does not prevent suicide. It means the effect is smaller or more variable than early trials suggested, and larger trials are needed to resolve the uncertainty. The clinical community continues to treat lithium as having antisuicidal properties based on observational data, mechanistic plausibility, and consistent point estimates across meta-analyses. But the randomized controlled trial evidence is no longer statistically significant at P < 0.05.

6.2 Whether lithium causes clinically significant GFR decline or CKD in most patients

The McKnight 2012 meta-analysis of RCTs found GFR mean reduction -6.22 mL/min (not significant, CI crosses zero). The Shine 2015 cohort found CKD stage 3 HR 1.93 (1.76–2.12), significant, but detection bias (more frequent monitoring of lithium users) may account for part or all of this. The absolute risk of renal replacement therapy in McKnight 2012 is low (0.5%). Lithium definitively reduces urinary concentrating ability, which is clinically significant and leads to polyuria.

6.3 Which patients will respond to which mood stabilizers

No validated algorithm exists for matching patients to mood stabilizers. Baseline characteristics, symptom profiles, and prior treatment history do not reliably predict response.

6.4 The mechanism by which lithium works

The mechanism by which lithium prevents relapse, reduces suicide risk, or exerts acute antimanic effects is unknown. Hypotheses include effects on intracellular signaling cascades, neurotrophic factors, and inflammatory markers, but none are established.

---

Part VII — The two overclaims, and what replaces them

7.1 The clinical-consensus overclaim: "Lithium is a miracle drug with unique antisuicidal properties"

The claim: Lithium is highly effective, specific for bipolar disorder, uniquely prevents suicide, and should be used more widely. Under-prescription of lithium represents a failure of clinical practice.

What the evidence shows: Lithium is effective for acute mania (OR 2.13 versus placebo) and superior to valproate in maintenance (BALANCE trial HR 0.71, but open-label). Early meta-analyses found Peto OR 0.13 for suicide prevention versus placebo (Cipriani 2013) and OR 0.26 versus active comparators (Cipriani 2005). However, Riblet 2022 found OR 0.30 (P = 0.05, not significant).

Lithium increases risk of hypothyroidism (OR 5.78, McKnight 2012; HR 2.31, Shine 2015), reduces urinary concentrating ability (-158.43 mOsm/kg), and causes weight gain (OR 1.89 versus placebo). Shine 2015 cohort found CKD stage 3 HR 1.93, but detection bias possible. GFR reduction in McKnight 2012 RCTs not significant.

What survives: Lithium is effective for acute mania and maintenance, with the broadest efficacy profile across mood episode types (Kishi 2021 NMA). Early meta-analytic evidence for antisuicidal effects (Cipriani 2005 OR 0.26, Cipriani 2013 Peto OR 0.13) is significant, but Riblet 2022 update does not reach significance (OR 0.30, P = 0.05). It is not a miracle drug. Tolerability concerns are real: hypothyroidism, polyuria, weight gain, narrow therapeutic window, and possible CKD risk.

7.2 The critical overclaim: "Lithium is a toxic relic causing kidney failure, replaced by safer drugs"

The claim: Lithium causes kidney failure, thyroid disease, and cognitive impairment. Newer mood stabilizers are safer and equally effective. Lithium should be avoided.

What the evidence shows: Lithium reduces urinary concentrating ability (established) and increases hypothyroidism risk (OR 5.78 McKnight; HR 2.31 Shine). Shine 2015 cohort found CKD stage 3 HR 1.93, but detection bias (more monitoring of lithium users) may account for this. McKnight 2012 RCTs found GFR reduction not significant. Absolute risk of renal replacement therapy is 0.5%.

Lithium is superior to valproate in maintenance (BALANCE trial). The suicide prevention evidence from early meta-analyses (OR 0.13 to 0.26) is not replicated for other mood stabilizers. Lamotrigine is effective for bipolar depression and prevents depressive episodes, but NOT for acute mania or preventing manic episodes.

What survives: Lithium has real adverse effects (thyroid, renal, therapeutic levels). Weight gain, tremor, and polyuria are common and may impair compliance. The cohort evidence for CKD (HR 1.93) is concerning, but detection bias limits causal inference. The RCT evidence for GFR decline is not significant. The folk narrative that "lithium causes kidney failure" overstates the evidence. Newer mood stabilizers have different adverse effect profiles, not universally safer profiles. Lamotrigine is well tolerated but has polarity-specific efficacy.

7.3 The version that survives

Bipolar disorder is a heterogeneous, episodic illness with variable course and outcome. Some patients have predominantly manic episodes; others have predominantly depressive episodes. Some respond to lithium monotherapy; others require combination therapy or sequential trials. No single mood stabilizer applies to all patients.

Lithium and valproate are both effective for acute mania versus placebo. McKnight 2019 Cochrane: lithium OR 2.13; Jochim 2019 Cochrane: valproate OR 2.05. Head-to-head trials find no significant difference (lithium vs valproate OR 1.22, CI includes 1).

In maintenance treatment, lithium is superior to valproate, and combination therapy is superior to valproate alone. The BALANCE trial (open-label, 24 months, typeset abstract) found lithium versus valproate HR 0.71 (95% CI 0.51–1.00, p = 0.0472); combination versus valproate HR 0.59 (0.42–0.83, p = 0.0023). Network meta-analysis (Kishi 2021) finds lithium is the only mood stabilizer effective across all mood episode types: preventing any episode, depressive episodes, and manic episodes.

Lamotrigine is effective for bipolar depression and prevents depressive relapses, but NOT for acute mania or preventing manic relapses. Geddes 2009 IPD meta-analysis: response RR 1.27; severe depression (HRSD > 24) RR 1.47, mild-moderate RR 1.07 (not significant). Kishi 2021 NMA: lamotrigine beats placebo on preventing any episode and depressive episodes, but NOT manic episodes. This polarity-specific efficacy is robust.

The suicide prevention evidence has weakened with additional trials. [contested] Early meta-analyses found Peto OR 0.13 (95% CI 0.03–0.66, Cipriani 2013) and 0.26 (0.09–0.77, Cipriani 2005), both statistically significant. Riblet 2022 found Peto OR 0.30 (0.09–1.02), P = 0.05, not statistically significant. The point estimate remains consistent, but the confidence interval crosses 1.0. The clinical community continues to treat lithium as having antisuicidal properties based on observational data and consistent point estimates, but the RCT evidence is now contested.

Adverse effects: lithium increases hypothyroidism (OR 5.78 McKnight; HR 2.31 Shine), reduces urinary concentrating ability (-158.43 mOsm/kg), and causes weight gain (OR 1.89). Shine 2015 cohort found CKD stage 3 HR 1.93, but detection bias possible. McKnight 2012 RCTs found GFR reduction not significant (CI crosses zero). Absolute risk of renal replacement therapy 0.5%. The folk narrative that "lithium causes kidney failure" overstates the evidence—the cohort finding is significant but possibly due to detection bias; the RCT finding is not significant.

What the evidence establishes: Lithium and valproate are effective for acute mania (no difference in head-to-head trials). Lithium is superior in maintenance, preventing both depressive and manic episodes. Lamotrigine prevents depressive episodes but not manic episodes. Lithium increases hypothyroidism and reduces urinary concentrating ability. Cohort evidence for CKD risk (HR 1.93), but detection bias limits causal inference. RCT evidence for GFR decline not significant. Early meta-analyses found significant suicide prevention effects; Riblet 2022 update does not reach significance.

What it does not establish: That lithium definitively reduces suicide risk (RCT evidence now contested). Whether lithium causes clinically significant GFR decline or CKD in most patients (RCT evidence not significant; cohort evidence significant but detection bias possible). Which patients will respond to which mood stabilizers. The mechanism by which lithium prevents relapse or reduces suicide risk. Long-term functional outcomes.

The structural problem: The BALANCE trial is open-label, introducing detection bias. The suicide prevention evidence was significant in 2013 but not in 2022—the point estimate remains consistent, but the confidence interval now crosses 1.0. The GFR finding in McKnight 2012 RCTs is not statistically significant. The CKD finding in Shine 2015 cohort is significant, but detection bias (more monitoring of lithium users) may account for it. The folk models—lithium as miracle drug or lithium as toxic relic—both overclaim. The evidence supports efficacy for acute mania and maintenance, established thyroid and urinary effects, cohort evidence for CKD risk, and contested suicide prevention.

---

References

Primary sources cited

Cipriani, A., Pretty, H., Hawton, K., & Geddes, J. R. (2005). "Lithium in the prevention of suicidal behavior and all-cause mortality in patients with mood disorders: a systematic review of randomized trials." American Journal of Psychiatry, 162(10), 1805–1819.

Cipriani, A., Hawton, K., Stockton, S., & Geddes, J. R. (2013). "Lithium in the prevention of suicide in mood disorders: updated systematic review and meta-analysis." BMJ, 346, f3646.

Geddes, J. R., Calabrese, J. R., & Goodwin, G. M. (2009). "Lamotrigine for treatment of bipolar depression: independent meta-analysis and meta-regression of individual patient data from five randomised trials." British Journal of Psychiatry, 194(1), 4–9.

Geddes, J. R., Goodwin, G. M., Rendell, J., et al. (2010). "Lithium plus valproate combination therapy versus monotherapy for relapse prevention in bipolar I disorder (BALANCE): a randomised open-label trial." The Lancet, 375(9712), 385–395.

Jochim, J., Rifkin-Zybutz, R., Geddes, J., & Cipriani, A. (2019). "Valproate for acute mania." Cochrane Database of Systematic Reviews, (10), CD004052.

Kishi, T., Ikuta, T., Matsuda, Y., et al. (2021). "Mood stabilizers and/or antipsychotics for bipolar disorder in the maintenance phase: a systematic review and network meta-analysis of randomized controlled trials." Molecular Psychiatry, 26(8), 4146–4157.

Kishi, T., Ikuta, T., Sakuma, K., et al. (2022). "Pharmacological treatment for bipolar mania: a systematic review and network meta-analysis of double-blind randomized controlled trials." Molecular Psychiatry, 27(2), 1136–1144. Published online 2021. DOI 10.1038/s41380-021-01334-4.

McKnight, R. F., Adida, M., Budge, K., Stockton, S., Goodwin, G. M., & Geddes, J. R. (2012). "Lithium toxicity profile: a systematic review and meta-analysis." The Lancet, 379(9817), 721–728.

McKnight, R. F., Saïk J.G.N. de La Motte de Broöns de Vauvert, E. Chesney, B. H. Amit, J. Geddes, A. Cipriani. (2019). "Lithium for acute mania." Cochrane Database of Systematic Reviews, (6), CD004048.

Riblet, N. B., Shiner, B., Young-Xu, Y., & Watts, B. V. (2022). "Lithium in the prevention of suicide in adults: systematic review and meta-analysis of clinical trials." BJPsych Open, 8(6), e199. DOI 10.1192/bjo.2022.605.

Shine, B., McKnight, R. F., Leaver, L., & Geddes, J. R. (2015). "Long-term effects of lithium on renal, thyroid, and parathyroid function: a retrospective analysis of laboratory data." The Lancet, 386(9992), 461–468.

US Food and Drug Administration. (2018). Lithium carbonate tablets prescribing information. Boxed warning: LITHIUM TOXICITY.

---

Appendix: Bind table for headline executive summary numbers

Every executive summary number is bound to an opened primary source with exact location:

Claim	Source	Extract Location	Unit	Population	Status
Li vs placebo response OR 2.13 (1.73–2.63), 6 trials, 1707	McKnight et al., 2019 Cochrane CD004048	Abstract, Main results	Odds ratio	Acute mania, 36 RCTs, 4220 total	✅ Opened
Li vs valproate OR 1.22 (0.87–1.70), 5 trials, 607	McKnight et al., 2019 Cochrane CD004048	Abstract, Main results	Odds ratio	Acute mania	✅ Opened
BALANCE: 59/110, 65/110, 76/110 primary outcome events (three valproate deaths in typeset)	Geddes et al., 2010	Typeset abstract, Results	Event counts	Bipolar I, 24 months	✅ Opened (typeset abstract)
BALANCE: Combination vs valproate HR 0.59 (0.42–0.83, p=0.0023)	Geddes et al., 2010	Typeset abstract, Results	Hazard ratio	Bipolar I, 24 months	✅ Opened (typeset abstract)
BALANCE: Lithium vs valproate HR 0.71 (0.51–1.00, p=0.0472)	Geddes et al., 2010	Typeset abstract, Results	Hazard ratio	Bipolar I, 24 months	✅ Opened (typeset abstract)
BALANCE: Combination vs lithium HR 0.82 (0.58–1.17, p=0.27)	Geddes et al., 2010	Typeset abstract, Results	Hazard ratio	Bipolar I, 24 months	✅ Opened (typeset abstract)
Cipriani 2013: Li vs placebo suicide Peto OR 0.13 (0.03–0.66)	Cipriani et al., 2013 BMJ	Figures 5 & 6	Peto odds ratio	Mood disorders, 48 RCTs, 6674	✅ Opened (figures, not abstract)
Cipriani 2013: Unipolar suicide OR 0.13 (0.02–0.76); death OR 0.36 (0.13–0.98)	Cipriani et al., 2013 BMJ	Figures 5 & 6	Odds ratio	Unipolar depression subgroup	✅ Opened (abstract swapped them; use figures)
Cipriani 2005: Suicide Peto OR 0.26 (0.09–0.77), 7 trials, 2 vs 11 suicides	Cipriani et al., 2005 AJP	Results section	Peto odds ratio	Mood disorders, 32 RCTs	✅ Opened
Riblet 2022: Suicide 2/568 vs 8/570, Peto OR 0.30 (0.09–1.02), P=0.05	Riblet et al., 2022 BJPsych Open	Abstract, Results	Peto odds ratio	Mood disorders, 7 RCTs	✅ Opened [contested: not significant]
Geddes 2009 lamotrigine IPD: HRSD RR 1.27 (1.09–1.47); >24 RR 1.47 (1.16–1.87); ≤24 RR 1.07 (0.90–1.27)	Geddes et al., 2009 BJP	Results section	Risk ratio	Bipolar depression, 1072 from 5 RCTs	✅ Opened
Kishi 2021 maintenance NMA: Li beats placebo on any-episode, depression, mania, discontinuation	Kishi et al., 2021 Mol Psychiatry	Findings	Qualitative	Maintenance, 41 RCTs, 9821	✅ Opened (numeric RRs for Li not copied)
Kishi 2021 maintenance NMA: Lamotrigine beats placebo on any-episode, depression, NOT mania	Kishi et al., 2021 Mol Psychiatry	Findings	Qualitative	Maintenance, 41 RCTs, 9821	✅ Opened
Kishi 2022 mania NMA: Li vs placebo response RR 1.259 (1.007–1.576), 72 RCTs, 16,442, 3.96±2.39 weeks	Kishi et al., 2022 Mol Psychiatry (pub 2021)	Abstract, Findings	Risk ratio	Acute mania NMA	✅ Opened (correct identity)
McKnight 2012: Hypothyroidism OR 5.78 (2.00–16.67); TSH +4.00; Umax −158.43; GFR −6.22 (CI includes 0)	McKnight et al., 2012 Lancet	Findings sections	Odds ratio, mean differences	Mood disorders, 385 studies	✅ Opened
McKnight 2012: Renal replacement therapy 18/3,369 (0.5%)	McKnight et al., 2012 Lancet	Findings, renal section	Absolute risk	Mood disorders, 385 studies	✅ Opened
McKnight 2012: Weight vs placebo OR 1.89 (1.27–2.82); vs olanzapine OR 0.32 (0.21–0.49)	McKnight et al., 2012 Lancet	Findings, bodyweight section	Odds ratio	Mood disorders, 385 studies	✅ Opened
Shine 2015: CKD stage 3 HR 1.93 (1.76–2.12); hypothyroidism HR 2.31 (2.05–2.60)	Shine et al., 2015 Lancet	Results section	Hazard ratio	4678 lithium-tested vs 689228 controls	✅ Opened (cohort, detection bias possible)
Jochim 2019 Cochrane: Valproate vs placebo 45% vs 29%, OR 2.05 (1.32–3.20), 4 studies, 869	Jochim et al., 2019 Cochrane CD004052	Abstract, Main results	Odds ratio	Acute mania	✅ Opened (correct identity, not Cipriani 2013)
FDA 2018: Therapeutic range 0.8–1.2 mEq/L, toxicity ≥ 1.5 mEq/L	FDA prescribing information	Boxed warning: LITHIUM TOXICITY	Serum concentration	Regulatory statement	✅ Opened (FDA label)

All headline numbers are bound to opened primary sources with exact locations verified. Identity corrections applied: Kishi 2022 mania NMA (not "Yildiz 2021"); Jochim 2019 valproate acute mania (not "Cipriani 2013 Cochrane"). BALANCE typeset abstract p-values preferred. Cipriani 2013 BMJ Figures 5 & 6 used (abstract swapped unipolar suicide and death ORs). Riblet 2022 marked [contested] and not statistically significant. Shine 2015 cohort added for CKD stage 3 HR 1.93. No reconstructed CIs. No unopened papers cited in executive summary.