I have read the journal’s policy and the authors of this manuscript have the following competing interests: SC, and JRL declare no conflict of interest with this work. PBJ is a scientific advisory board member for Janssen and Recordati. BRU is clinical director for older people’s and adult community services at CPFT. He is clinical director of the Windsor Unit at Fulbourn Hospital (CPFT), which delivers clinical trials in dementia/mild cognitive impairment for academic and commercial organisations without personal benefit, and is the clinical lead for dementia for the NIHR Clinical Research Network (CRN) in the East of England. His salary is part-funded by the NIHR CRN. His lectureship is funded by Gnodde Goldman-Sachs Giving. He has been principal investigator on trials for Axovant, Lilly, and EIP Pharma; his institution has benefited from payment for research carried out but he has not personally received any money. His wife is the lead for mental health for Suffolk Clinical Commissioning Group. RNC consults for Campden Instruments Ltd and receives royalties from Cambridge University Press, Cambridge Enterprise, and Routledge.
Dementia is the leading cause of death in elderly Western populations. Preventative interventions that could delay dementia onset even modestly would provide a major public health impact. There are no disease-modifying treatments currently available. Lithium has been proposed as a potential treatment. We assessed the association between lithium use and the incidence of dementia and its subtypes.
We conducted a retrospective cohort study comparing patients treated between January 1, 2005 and December 31, 2019, using data from electronic clinical records of secondary care mental health (MH) services in Cambridgeshire and Peterborough NHS Foundation Trust (CPFT), United Kingdom (catchment area population approximately 0.86 million). Eligible patients were those aged 50 years or over at baseline and who had at least 1 year follow-up, excluding patients with a diagnosis of mild cognitive impairment (MCI) or dementia before, or less than 1 year after, their start date. The intervention was the use of lithium. The main outcomes were dementia and its subtypes, diagnosed and classified according to the International Classification of Diseases-10th Revision (ICD-10).
In this cohort, 29,618 patients (of whom 548 were exposed to lithium) were included. Their mean age was 73.9 years. A total of 40.2% were male, 33.3% were married or in a civil partnership, and 71.0% were of white ethnicity. Lithium-exposed patients were more likely to be married, cohabiting or in a civil partnership, to be a current/former smoker, to have used antipsychotics, and to have comorbid depression, mania/bipolar affective disorder (BPAD), hypertension, central vascular disease, diabetes mellitus, or hyperlipidemias. No significant difference between the 2 groups was observed for other characteristics, including age, sex, and alcohol-related disorders. In the exposed cohort, 53 (9.7%) patients were diagnosed with dementia, including 36 (6.8%) with Alzheimer disease (AD) and 13 (2.6%) with vascular dementia (VD). In the unexposed cohort, corresponding numbers were the following: dementia 3,244 (11.2%), AD 2,276 (8.1%), and VD 698 (2.6%). After controlling for sociodemographic factors, smoking status, other medications, other mental comorbidities, and physical comorbidities, lithium use was associated with a lower risk of dementia (hazard ratio [HR] 0.56, 95% confidence interval [CI] 0.40 to 0.78), including AD (HR 0.55, 95% CI 0.37 to 0.82) and VD (HR 0.36, 95% CI 0.19 to 0.69). Lithium appeared protective in short-term (≤1-year exposure) and long-term lithium users (>5-year exposure); a lack of difference for intermediate durations was likely due to lack of power, but there was some evidence for additional benefit with longer exposure durations. The main limitation was the handling of BPAD, the most common reason for lithium prescription but also a risk factor for dementia. This potential confounder would most likely cause an increase in dementia in the exposed group, whereas we found the opposite, and the sensitivity analysis confirmed the primary results. However, the specific nature of the group of patients exposed to lithium means that caution is needed in extending these findings to the general population. Another limitation is that our sample size of patients using lithium was small, reflected in the wide CIs for results relating to some durations of lithium exposure, although again sensitivity analyses remained consistent with our primary findings.
We observed an association between lithium use and a decreased risk of developing dementia. This lends further support to the idea that lithium may be a disease-modifying treatment for dementia and that this is a promising treatment to take forwards to larger randomised controlled trials (RCTs) for this indication.
Shanquan Chen and colleagues investigate the association between lithium use and the incidence of dementia and its subtypes.
Dementia is the leading cause of death and disability in elderly Western populations.
Preventative interventions that could delay dementia onset even modestly would provide a major public health impact.
Lithium has been proposed as a potential treatment, but limited population-level research has been conducted to support the idea that lithium might delay the onset of dementia.
We identified 548 patients who were exposed to lithium and 29,070 patients who were not exposed to lithium between January 1, 2005 and December 31, 2019, using data from electronic clinical records of secondary care mental health (MH) services in Cambridgeshire and Peterborough NHS Foundation Trust (CPFT), UK.
We assessed the association between lithium use and the incidence of dementia and its subtypes, including Alzheimer disease (AD) and vascular dementia (VD).
We found that lithium was associated with a lower risk of receiving a diagnosis of dementia, including a lower risk of being diagnosed with either AD or VD.
The extended analysis by the duration of lithium treatment indicated that both short- (≤1-year exposure) and long-term (>5-year exposure) lithium exposure (but not medium term, 1- to 5-year lithium exposure) was associated with decreased incidence of dementia and its subtypes. Power was less for the intermediate durations. There was some evidence for additional benefit with longer exposure durations.
Our findings support the possibility that lithium treatment could decrease the risk of developing dementia and supports the need for further randomised controlled trials (RCTs) to test the efficacy of lithium as a disease-modifying drug in dementia.
The main limitation is that 73% of the patients in lithium-exposed group had mania/bipolar affective disorder (BPAD), which is a significant risk factor for dementia. Our results were in the opposite direction, however, and were supported by sensitivity analyses.
Dementia is the leading cause of death and disability in elderly Western populations: Approximately 47 million people had dementia worldwide in 2015, and this number is projected to triple by 2050 [
One major consideration in observational studies is the need to account for the influence of mania/bipolar affective disorder (BPAD), or unipolar depression, the 2 most common indications for lithium. BPAD and depression may both themselves increase the risk of dementia [
In this study, our primary aim was to assess the association of lithium use with the incidence of dementia and its subtypes via a retrospective cohort study, with analysis of potential confounding factors, in routinely collected clinical data covering a 15-year period. Our second aim was to examine the degree of association by the duration of lithium treatment. We hypothesised that lithium would be associated with a reduced risk of dementia and its subtypes, and this association would be seen in both short- and long-term lithium exposure.
We performed a retrospective cohort study using data from the electronic clinical records of Cambridgeshire and Peterborough NHS Foundation Trust (CPFT), UK, deidentified to create the CPFT Research Database [
We examined data between January 1, 2005 and December 31, 2019. For the cohort of patients exposed to lithium (group Li+), each patient’s origin time (start date) was defined as their earliest recorded exposure to lithium (defined below). For the unexposed cohort (group Li−), the origin time was the latest of their CPFT registration date or January 1, 2005. Follow-up was until the patients’ final record, death, or the first record of dementia, whichever occurred first. Eligible patients were those aged ≥50 years at baseline and had at least 1 year of follow-up. We excluded patients with a preexisting diagnosis of MCI or dementia or those diagnosed <1 year after their origin time.
The CPFT Research Database operates on an opt-out basis and contains data from most CPFT patients (at present totalling approximately 270,000 across all ages). Natural language processing (NLP) tools were used to extract smoking status, medication [
Dementia was diagnosed and classified according to the World Health Organization’s (WHO) International Classification of Diseases-10th Revision (ICD-10), using codes F00 (AD), F01 (vascular dementia, VD), F02 (dementia in other diseases), F03 (unspecified dementia), and G30 (AD). We also subdivided dementia into AD (F00, G30) and VD (F01) as independent outcomes.
Exposure to lithium was judged if patients were prescribed lithium (via NLP tools detecting medication mentions) or had a serum lithium level of ≥0.2 mmol/L recorded (via NLP tools detecting lithium test results). The duration of lithium exposure was calculated as the range between the earliest and latest date of lithium events (defined as above) and was treated as a categorical variable. There is no uniform standard to categorise lithium exposure duration; we chose 1 year as the minimum duration of treatment likely to have a function on dementia outcomes [
We investigated several sociodemographic variables: age at baseline (years), sex (male versus female), marital status (married, cohabiting, or civil partnership versus not), and ethnicity (white versus others/unknown). We investigated smoking status (current or past smoker versus not). We also investigated coprescription and comorbidity as potential confounders and treated them as binary time-constant variables (i.e., whether the medications had ever been prescribed to the patient or whether they had ever had the condition). Medication/comorbidity variables were the following: taking
MCI was identified with ICD-10 codes starting F06.7. Death was ascertained by weekly linkage to national NHS Spine mortality data for all patients known to CPFT MH services.
We report categorical variables as number (percentage) and continuous variables as mean (standard deviation). Differences between exposure groups were assessed via 2-tailed
Cox proportional hazard models were used to examine the corrected association between lithium exposure and risk of incident dementia, covarying for sociodemographic variables, smoking status, coprescription, and comorbidities. Unlike RCTs, clinical decisions to prescribe lithium are often based on prognostic factors. Therefore, the association estimate for lithium might be confounded by treatment selection. To further control for possible selection bias between lithium exposure and nonexposure, the Cox models were adjusted by inverse probability weights (IPWs). IPW (or inverse probability of treatment weighting) is an extension of the propensity score method used to summarise the conditional probability of assignment for a treatment [
We repeated the Cox regression to analyse associations by duration of lithium exposure.
Data were complete for outcomes and predictors except for ethnicity and any unmeasured diagnostic undercoding in clinical practice. We used multiple imputation with chained equations [
We compared models using partial likelihood ratio tests (as models were not nested) and present also Akaike information criterion (AIC) values.
We performed sensitivity analyses as follows. (1) We used a longer criterion (up to 2 years before the end of follow-up) to identify concomitant medications or morbidities. Since the follow-up duration was long and medications/morbidities were treated as constant (time-invariant) variables, then (for example) if hypertension develops only in the last year of follow-up, it would be treated as present throughout for that patient. This sensitivity analysis tests the potential bias of including late-onset medication/conditions or not. (2) We required a longer follow-up (at least 2 years), in order to exclude patients seen only transiently (e.g., registered in CPFT temporarily, such as visitors, or people discharged after a single assessment). (3) We added the assumption that all lithium users had BPAD, to account for diagnostic undercoding of BPAD. (The primary UK indications for lithium are mania/BPAD and recurrent depression. BPAD is a more common indication; note also that the use for unipolar depression is off-licence e.g., in the USA [
Analyses were performed using R (v3.5.0), including the packages survival (v3.1.7), ipw (v1.0.11), timereg (v1.9.4), dplyr (v0.8.0.1), mice (v3.11.0), weights (v1.0.4), forestplot (v1.9), and nonnestcox (v0.0.0.9). Statistical significance was defined as
The analysis plan was made prior to the start of all analyses and agreed upon among coauthors. No data-driven changes to the analysis plan were made. Four additional sensitivity analyses were included in response to peer review, including controlling only for confounders known at baseline, treating the duration of exposure to lithium as a time-varying variable, pooling intermediate exposure durations, and adding the lithium exposure × exposure duration interaction.
This study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (
Deidentified data were extracted from the electronic clinical records of Cambridgeshire and Peterborough NHS Foundation Trust (CPFT) under NHS Research Ethics approvals (references 12/EE/0407 and 17/EE/0442). No patients were involved in the development of the research question or the outcome measures or in developing plans for the design and analysis of the study.
We followed 548 patients exposed to lithium (Li+) and 29,070 patients unexposed to lithium (Li−) (
CPFT, Cambridgeshire and Peterborough NHS Foundation Trust; MCI, mild cognitive impairment; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.
Exposed group (Li+) | Unexposed group (Li−) | SMD before IPW | SMD after IPW | ||
---|---|---|---|---|---|
( |
( |
||||
73.05 (11.82) | 73.93 (12.72) | 0.0857 | 0.0004 | 0.0003 | |
224 (40.9) | 11,668 (40.1) | 0.7601 | 0.0150 | 0.0021 | |
Married/civil partnership | 231 (42.2) | 9,634 (33.1) | <0.0001 | 0.4210 | 0.0860 |
Cohabiting | 2 (0.4) | 52 (0.2) | |||
Single | 69 (12.6) | 1,881 (6.5) | |||
Divorced/separated | 48 (8.8) | 3,140 (10.8) | |||
Widowed | 62 (11.3) | 5,791 (19.2) | |||
Unknown | 136 (24.8) | 8,576 (29.5) | |||
White | 409 (74.6) | 20,616 (70.9) | 0.0005 | 0.1670 | 0.0930 |
Asian | 12 (2.2) | 233 (0.8) | |||
Black | 3 (0.5) | 87 (0.3) | |||
Other | 57 (10.4) | 3,790 (13) | |||
Not known | 67 (12.2) | 4,344 (14.9) | |||
34 (6.2) | 699 (2.4) | <0.0001 | 0.1880 | 0.0762 | |
6 (1.1) | 271 (0.9) | 0.8666 | 0.0160 | 0.0330 | |
230 (42) | 1,998 (6.9) | <0.0001 | 0.8950 | 0.0371 | |
355 (64.8) | 6,090 (20.9) | <0.0001 | 0.9010 | 0.0830 | |
399 (72.8) | 971 (3.3) | <0.0001 | 0.9880 | 0.4208 | |
154 (28.1) | 3,915 (13.5) | <0.0001 | 0.3670 | 0.0523 | |
236 (43.1) | 5,681 (19.5) | <0.0001 | 0.5240 | 0.0813 | |
50 (9.1) | 987 (3.4) | <0.0001 | 0.2380 | 0.0402 | |
140 (25.5) | 3,082 (10.6) | <0.0001 | 0.3960 | 0.0021 | |
No exposure | 29,070 (100) | ||||
>0, < = 1 years | 323 (58.9) | ||||
>1, < = 2 years | 47 (8.6) | ||||
>2, < = 3 years | 32 (5.8) | ||||
>3, < = 4 years | 22 (4) | ||||
>4, < = 5 years | 23 (4.2) | ||||
>5 years | 101 (18.4) | ||||
Dementia (yes) | 53 (9.7) | 3,244 (11.2) | 0.2776 | 0.0490 | 0.1760 |
AD (yes) | 36 (6.8) | 2,276 (8.1) | 0.2613 | 0.0500 | 0.1330 |
VD (yes) | 13 (2.6) | 698 (2.6) | 0.8612 | 0.0050 | 0.2370 |
4.77 (3.32) | 4.31 (2.89) | <0.0001 | 0.0008 | 0.0114 |
Data are shown as mean (SD) for age and time to event or number (percentage) for others.
AD, Alzheimer disease; BPAD, bipolar affective disorder/mania; IPW, inverse probability weighting; SMD, standardised mean difference between groups; VD, vascular dementia.
In the exposed cohort, there were 53 (9.7%) patients diagnosed with dementia, including 36 (6.8%) with AD and 13 (2.6%) with VD. The cumulative hazard of dementia and its subtypes in lithium users versus nonusers is shown in
After controlling for confounding by sociodemographic factors (age, sex, marital status, and ethnicity), smoking status, other medications, other mental comorbidities, and physical comorbidities, compared with patients who never took lithium, exposure to lithium was associated with a lower risk of dementia (adjusted hazard ratio [HR] 0.56, 95% confidence interval [CI] 0.40 to 0.78;
HRs, 95% CIs, and
HRs, 95% CIs, and
Sensitivity analyses using a longer medication/comorbidity discovery criteria (Figs A and B in
Analysis of the effects of exposure duration, over and above the fact of exposure, suggested a significant decrease in risk associated with additional exposure to lithium, for all dementias and AD (Fig Q in
Using a retrospective cohort study based on a large and comprehensive clinical record database, and after controlling for sociodemographic factors (age, sex, marital status, and ethnicity), smoking status, comedications, co-mental morbidities, and co-physical morbidities, we found that lithium was associated with a lower risk of dementia and its subtypes (AD and VD). Analysis by the duration of lithium treatment indicated that both short (≤1-year exposure) and long-term (>5-year exposure) lithium exposure were associated with a decreased incidence of dementia; effects at intermediate durations were not significant, likely due to lesser power (discussed further below), but there was evidence that the risk of dementia reduced further with longer durations of lithium exposure.
Our findings supported the hypothesis that lithium is associated with a reduced risk of dementia and its subtypes. These associations are consistent with a hypothesised protective effect of lithium on dementia from cellular and rodent models [
To the best of our knowledge, our study is the first to examine the association between lithium and VD in people prescribed lithium. Two studies from Denmark [
While the potential for lithium to slow progression in AD has been studied experimentally [
We have further expanded on the existing literature by examining the influence of the duration of lithium treatment, as previously suggested [
In our study, medium-term treatment with lithium (1 to 5 years) was not associated with a significantly decreased incidence of dementia. A true U-shaped or bimodal association of duration would be hard to explain if the function of lithium on cognitive function is cumulative and chronic [
One strength of this study is that we examined subtypes of dementia, not only focusing on AD, for which the biological plausibility of a lithium effect is perhaps strongest, but also including others such as VD, for which corresponding evidence is more limited. Another advantage of our study is that we examined our results by the duration of lithium treatment.
One limitation is the handling of BPAD, which is a significant risk factor for dementia [
Another limitation is that our sample size of patients using lithium was smaller than other studies conducted in Denmark [
It is possible that there may have been selection bias if clinicians avoided prescribing lithium to patients with possible dementia. In previous research [
Finally, mental disorders (including dementias) and physical comorbidities may have been undercoded. This possibility always exists with the use of routinely collected clinical data and is an unmeasured potential source of power reduction and/or bias.
The main unanswered question from this work is the dose–response association between lithium within its therapeutic range and the incidence of dementia. The clinical context means that lithium levels primarily lay within its therapeutic range of 0.4 to 1.0 mmol/L (mean 0.61 mM, interquartile range 0.49 to 0.80 mM, as above), but at times, lithium levels in some patients may have been >1.0 mmol/L, with resultant potential for neurotoxicity, or subtherapeutic, either might alter the estimate of the protective effects of lithium, and the optimal level for any such protective effect is unknown. The nature of our data set, with incomplete lithium level data, did not allow us to explore the relationship between lithium levels and outcomes.
Caution should be exercised when drawing conclusions from the current study with regard to the general population. Our cohort differs from the general population, as our database was of patients treated for MH conditions, although the age structure of our database was similar to that of England. The frequency of dementia in our control cohort was higher than in the general population [
In conclusion, prescription lithium use was associated with a decreased incidence of dementia and its subtypes, including AD and VD, after accounting for confounding variables. This finding was robust to sensitivity analyses, with evidence for duration-dependent effects. This adds to the growing preclinical, epidemiological, and trial evidence that lithium may be a disease-modifying agent in dementia. The present study adds to this evidence base in people with preexisting mental disorders, primarily BPAD. Since lithium is already established as a treatment of choice for BPAD [
STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.
(DOCX)
(DOCX)
We thank Dr. Chuoxin Ma in the department of public health at the University of Cambridge, for her help in statistical issues.
The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care.
Alzheimer disease
Akaike information criterion
bipolar affective disorder/mania
confidence interval
Cambridgeshire and Peterborough NHS Foundation Trust
hazard ratio
International Classification of Diseases-10th Revision
inverse probability weights/weighting
mild cognitive impairment
mental health
National Institute for Health and Care Excellence
natural language processing
randomised controlled trial
standardised mean difference
Strengthening the Reporting of Observational Studies in Epidemiology
vascular dementia
World Health Organization