Association Between Groundwater Lithium and the Diagnosis of Bipolar Disorder and Dementia in the United States

Lithium, a naturally occurring trace element in groundwater, is a cornerstone therapy for bipolar disorder and may have a role in the treatment of dementia.¹ Kessing et al² found an inverse association between lithium in drinking water and dementia in Denmark. In the United States, lithium exposure has also been associated with lower rates of mental health disorders.³ However, mental health diagnosis rates vary substantially with local health care resources and demographics,⁴ potentially confounding the relationship with groundwater lithium concentrations. We examined the association between groundwater lithium and diagnoses of bipolar disorder and dementia in the United States, adjusting for local health care resources and demographics.

Groundwater lithium concentrations were collected by the US Geological Survey⁵ from more than 3000 drinking water wells from 1992 to 2003. Lithium concentrations vary widely in the United States because of diverse climates and geological compositions of aquifers.⁵ Diagnoses were identified from the inpatient hospital, long-term care, and other therapy claims files in the following sources: Truven Health MarketScan Commercial Claims and Encounters (CCAE) database (2003-2010), a claims database for privately insured patients; Truven Health MarketScan Medicare Supplemental database (2003-2010), a claims database for Medicare recipients with employer-sponsored supplemental insurance; and Medicaid Analytic eXtract (MAX) (2011-2012). As confirmed with the University of Chicago institutional review board, the secondary analysis of deidentified data was exempt from informed consent. Primary outcomes were the prevalence of bipolar disorder and dementia. To prevent spurious causal inference from inadequate adjustment for confounders, we repeated our analysis for 3 “negative control” outcomes that have no known link to groundwater lithium (major depressive disorder, myocardial infarction, and prostate cancer).

County-level health care resources and demographics were collected from the Health Resources & Services Administration (HRSA) 2010 Area Health Resources Files (AHRF).⁶ The ARHF county-level data were assembled from the American Medical Association, the American Hospital Association, and the American Community Survey by the HRSA. The ARHF is designed to measure geographic variation in “factors that may impact health status and health care in the United States.”⁶ From the ARHF, we extracted the number of hospital beds, primary care physicians per 100 000 persons, psychiatrists per 100 000 persons, and median household income for each county. County-level demographics included census population, median age, education, race, and ethnicity. Counties in the top and bottom deciles of census population were trimmed from the data set.

We fit a mixed-effects Poisson regression model with inverse probability of treatment weighting (IPTW), where treatment was defined as lithium exposure exceeding 40 μg/L (a natural break in the lithium distribution).^7,8 Weights were based on county-level health care resources, designed to give the low-lithium counties the same distribution of health care resources as the high-lithium counties. We also controlled for sex, payer, and county-level demographics. As a sensitivity analysis, we also examined the association of lithium as a continuous variable (restricted cubic spline with 5 knots), controlling for county-level demographics and health care resources.

Claims data for 4 227 556 adults living in 174 counties were analyzed, including 3 046 331 with private insurance, 261 461 with Medicare Supplemental, and 919 764 with Medicaid. Among them, 404 662 patients (9.6%) lived in 1 of 32 counties with high lithium (>40 μg/L). The mean and median lithium concentrations were 27.4 μg/L and 11.1 μg/L, respectively (IQR, 3.7-23.6 μg/L).

Unadjusted prevalence rates for all outcomes were significantly lower in high-lithium counties. However, high-lithium counties had fewer physicians and health care resources and had smaller, younger, less educated, and more Hispanic populations (Table).

After adjustment for county-level demographics and health care resources, high lithium did not confer any significant benefit for bipolar disorder, dementia, or the negative controls major depressive disorder, myocardial infarction, or prostate cancer. The Figure shows the lack of any association across the entire lithium distribution.

Despite the substantial variation in groundwater lithium exposure in the United States, we found no significant association between groundwater lithium exposure and risk of bipolar disorder or dementia after adjustment for county-level demographics and health care resource. This indicates the purported association of high-lithium concentrations in drinking water with mental health disorders is driven by unaccounted variation in demographics, health care resources, and diagnosis practices.

Therapeutic lithium doses are orders of magnitude larger than groundwater lithium concentrations, making a true causal relationship between groundwater lithium and mental health biologically dubious. In our study, the high-lithium group was exposed to a mean of 141.3 μg/L in their water supply. This means that a patient would need to drink more than 1000 L of water a day to ingest the lowest reported effective therapeutic lithium dose of 150 mg.¹

Accepted for Publication: March 22, 2018.

Corresponding Author: William F. Parker, MD, Department of Medicine, The University of Chicago, 5841 S Maryland Ave, Mail Code 6076, Chicago, IL 60637 (william.parker@uchospitals.edu).

Published Online: May 23, 2018. doi:10.1001/jamapsychiatry.2018.1020

Author Contributions: Drs Parker and Gibbons had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Parker, Gao, Gibbons.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Parker, Gao, Gibbons.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: All authors.

Administrative, technical, or material support: Gorges, Zhang.

Study supervision: Gibbons.

Conflict of Interest Disclosures: None reported.

Funding/Support: Dr Parker is supported by grant 5T32HL007605-32 from the National Institutes of Health. Ms Gorges is supported by grant T32HS000084 from the Agency for Healthcare Research and Quality. Ms Gao is supported by National Institutes of Health TL1 Linked Training awards 5TL1RR025001-05 and 2TL1TR000432-06 and by National Research Service awards 5T32GM007281-41 and 1T32AG051146-0.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Information: This article was the product of a class project of the Statistical Applications course at The University of Chicago taught by Dr Gibbons from September 25 to December 9, 2017. The Truven Health MarketScan Commercial Claims and Encounters (CCAE) database contains individuals who are employees, their covered spouses and dependents, Consolidated Omnibus Budget Reconciliation Act of 1985 (COBRA) continuees, and non-Medicare retirees. There are approximately 99 million unique enrollee records in the CCAE database from 2003 to 2010. The Truven Health MarketScan Medicare Supplemental database contains individuals who are Medicare-eligible retirees with employer-sponsored Medicare Supplemental plans. There are approximately 7.5 million unique enrollee records in this database from 2003 to 2010. The Medicaid Analytic eXtract (MAX) data are created by the Centers for Medicare & Medicaid Services (CMS) from data submitted through each state’s Medicaid Statistical Information System and then processed to provide uniform coding across states. Beneficiary enrollment status was determined using the person’s summary files and then used to exclude beneficiaries enrolled in comprehensive managed care plans and/or Medicare to ensure that full claims records were present. Diagnoses were identified from the inpatient hospital, long-term care, and other therapy claims files, and county diagnosis and population counts were pooled over the 2-year period (2011-2012). Because of CMS restrictions on minimum cell sizes, county-level sex cells with diagnosis or population counts less than 11 were omitted.

Additional Contribution: Prof Don Rubin, PhD (Department of Statistics, Harvard University), provided assistance with our statistical approach. No compensation was received for his contribution to the study.