Blood metal signatures and their association with Alzheimer’s disease risk and cognitive decline in an Egyptian cohort

Abstract

Background

Alzheimer’s disease (AD) is a multifactorial disease in which environmental exposures may contribute to disease risk. This study aimed to investigate associations between blood concentrations of arsenic, lead, mercury, and lithium with AD risk and cognitive performance in an Egyptian cohort.

Methods

We conducted a cross-sectional study in 50 AD patients and 50 cognitively healthy controls recruited across multiple centres in Egypt. Blood metal concentrations were measured by inductively coupled plasma optical emission spectrometry. A tiered analytical strategy addressed variable left-censoring: group-stratified Kaplan-Meier imputation was applied for arsenic, lead, and lithium, followed by logistic and linear regression adjusted for age and sex, with a binary detection sensitivity analysis for lithium, lead, and mercury. Multiple testing was corrected using the Benjamini-Hochberg false discovery rate method.

Results

Here we show that higher arsenic concentrations are strongly associated with increased odds of AD (OR = 1321.61, p = 3.48×10⁻⁶) and lower cognitive scores (β = −16.44, p = 1.46×10⁻⁸), both surviving false discovery rate correction. Higher lithium is associated with reduced disease odds (OR = 0.634, FDR-adjusted p = 0.022) and better cognitive performance (β = 1.21, FDR-adjusted p = 0.020), corroborated in sensitivity analyses. Mercury detection is markedly more frequent in patients than controls (24% vs. 2%, p = 0.002) and associates with lower cognitive scores (β = −7.08, p = 0.0003). Lead shows a borderline association with disease (OR = 1.53, p = 0.046) that does not survive correction.

Conclusions

These findings support arsenic neurotoxicity and lithium deficiency as potentially important contributors to AD risk in an Egyptian population, with implications for environmental public health in low- and middle-income settings.

Plain language summary

Alzheimer's disease (AD) appears to be caused by complex interactions. Environment, genetics, lifestyle, educational level, are all factors that can affect disease occurrence and progression. Of the multiple environmental factors that may play a role in AD, heavy metals exposure has been shown to have an important correlation to this disease. In this research we evaluated the blood levels of four heavy metals, arsenic, lead, mercury, and lithium in AD patients and healthy subjects. Our results showed that Arsenic is correlated to the disease while lithium could play a protective role! These findings need to be validated on other populations then translated into public health policies.

Othman et al. measure blood concentrations of arsenic, lead, mercury, and lithium in 50 Alzheimer’s disease patients and 50 cognitively healthy controls across multiple centres in Egypt. They show that blood arsenic is associated with increased Alzheimer’s disease risk and lower cognitive scores surviving false discovery rate correction.

Introduction

Alzheimer’s disease (AD) is the most common form of dementia, accounting for 60–80% of dementia cases in individuals aged 65 and above^1,2. It is a prevalent, progressive neurodegenerative disorder characterized pathologically by the accumulation of amyloid-beta plaques and hyperphosphorylated tau protein tangles in the brain³. These pathological hallmarks lead to synaptic dysfunction, memory loss, cognitive decline, and behavioral changes³. The pathogenesis of AD is complex and is believed to have a multifactorial etiology involving genetic, epigenetic, and environmental factors^4,5.

Heavy metals cause damage to the central nervous system, leading to cognitive and behavioral impairments⁶. Recent research suggests a potential association between environmental exposure to these metals and the development or progression of AD⁷. The exact mechanisms by which these elements cause AD are complex but are believed to involve oxidative stress, inflammation, and neurotransmitter disruption^7–9.

Toxic metals such as lead, arsenic, and mercury are pervasive environmental pollutants⁷. A body of evidence suggests that chronic exposure to these elements is associated with an increased risk of cognitive impairment and may contribute to the pathological processes underlying AD^8,10. These metals are thought to cause neuronal damage primarily by promoting oxidative stress and inflammation, which can disrupt normal brain function^10,11. In contrast to these harmful metals, lithium, an essential trace element, appears to play a different role¹². Emerging research indicates that lithium may have neuroprotective properties¹². Studies have reported lower levels of lithium in the brains of individuals with AD, and experimental data suggests that lithium deficiency can exacerbate features of the disease, while supplementation may offer protective benefits¹².

This study investigated the association between blood concentrations of arsenic, lead, mercury, and lithium and Alzheimer’s disease risk and cognitive performance in an Egyptian cohort. While the neurotoxic effects of these metals and their roles in AD pathogenesis are well established in studies from high-income settings, their associations remain largely unexplored in low- and middle-income countries. We report that arsenic is universally detected across all participants and shows the strongest associations with disease, with higher concentrations associated with markedly increased odds of Alzheimer’s disease and substantially lower cognitive scores surviving false discovery rate correction. Lithium concentrations are lower in Alzheimer’s disease patients, and higher levels are consistently associated with reduced disease odds and better cognitive performance across both primary and sensitivity analyses, supporting a potential neuroprotective role. Mercury is detectable in a significantly greater proportion of patients than controls, and its detection associates with lower cognitive scores, though these findings are considered preliminary given analytical constraints. Lead shows a borderline positive association with disease that does not survive multiple testing correction. Together, these findings address a critical gap in understanding environmental contributions to dementia risk in an underrepresented population and demonstrate the importance of applying rigorous censored-data methods when biomonitoring data contain high proportions of measurements below the limit of detection.

Methods

Participants recruitment and clinical workup

A total of fifty patients diagnosed with Alzheimer’s disease (AD) (n = 50) were recruited from the Egyptian Dementia Network (EDN) registry¹³. Clinical evaluations were conducted by qualified neurologists, geriatricians, or geriatric psychiatrists at participating university hospitals. Clinicians diagnosed Alzheimer’s disease (AD) patients according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5)¹⁴ and the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10)¹⁵. Functional status was assessed using the Katz Activities of Daily Living (ADL) scale for basic self-care and the Lawton–Brody Instrumental Activities of Daily Living (IADL) scale for complex independent living activities^16–19. Clinicians obtained comprehensive medical histories, and a series of laboratory tests—including liver and kidney function tests, complete blood count (CBC), blood glucose curve, lipid profile, and thyroid function tests (free T3, free T4, and TSH)—were performed to identify comorbidities and rule out systemic conditions, such as hypothyroidism, that may cause pseudodementia. They further assessed cognitive function using standardized psychometric testing, specifically, the Mini-Mental State Examination (MMSE)²⁰.

Additionally, fifty cognitively healthy individuals (n = 50) were enrolled as a reference control group from the same hospital settings and catchment areas as the AD cases — specifically, from companions of patients visiting neurology outpatient clinics, or from community members identified through the Egyptian Dementia Network registry who were willing to participate and met the eligibility criteria. Cognitively healthy status was confirmed by clinical evaluation and an MMSE score ≥ 24. The same laboratory workup (liver and kidney function tests, CBC, blood glucose, lipid profile, and thyroid function tests) was performed in all participants, including controls, to rule out systemic conditions that might affect metal metabolism or cognitive function.

Ethics statement

All participants or their legal guardians provided written informed consent in accordance with the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the American University in Cairo on February, 2025 (IRB-AUC #2024-2025-141).

Blood sample collection and metal concentration measurements

Venous whole blood samples were collected from 100 participants (aged ≥ 50 years) in 2025 using trace-metal–free EDTA tubes. Participants with confirmed cognitive and functional assessments were included, while those with occupational metal exposure, chronic kidney disease, active inflammatory or infectious conditions, prior stroke, or incomplete data were excluded. Samples were stored at −80 °C and analysed within six months of collection. Blood concentrations of arsenic, lead, mercury, and lithium were quantified by inductively coupled plasma–optical emission spectrometry (ICP-OES; Thermo Scientific iCAP) following validated nitric acid digestion. Analytical quality was confirmed by matrix-matched calibration (R² ≥ 0.99), certified reference materials, and spike recoveries of 80–120% with relative standard deviations < 10%. Concentrations are expressed in parts per billion (ppb, equivalent to µg/L for whole blood). Limits of detection were: arsenic 0.0069 ppb, lead 0.0691 ppb, mercury 0.1223 ppb, and lithium 0.0053 ppb.

Statistics and reproducibility

All statistical analyses were performed in R (version 2026.04.0 + 497) using the following packages: survival (version 3.5) for Kaplan-Meier imputation; stats (base R) for regression and non-parametric testing; broom for model output extraction; dplyr and tidyr for data manipulation; and ggplot2 for visualisation²¹. A two-sided p-value threshold of 0.05 was applied for all tests, and p-values are reported to four decimal places with values below 0.0001 expressed in scientific notation. Multiple comparisons were accounted for using the Benjamini-Hochberg false discovery rate (FDR) correction applied via p.adjust(method = “BH”), separately within each analysis block. For Kaplan-Meier imputation, left-censored survival objects were constructed using Surv(time, event, type = “left”) and fitted with survfit(); the restricted mean (rmean) from the summary table was used as the imputed value for below-detection-limit observations. Logistic regression was performed using glm() with family = binomial(link = “logit”) and linear regression using lm(), with age and sex entered as covariates via reformulate(). Between-group differences in detected concentrations were assessed using wilcox.test(exact = FALSE), reporting asymptotic p-values with continuity correction to account for ties; results are reported as Mann-Whitney U statistics.

Prior to analysis, extreme values were identified and removed within each metal-group stratum. Outliers were defined as detected values exceeding three standard deviations from the stratum-specific mean, assessed independently within the Control and AD groups. Values below the limit of detection (BDL) were not considered in outlier assessment, as their distributional properties differ from detected measurements.

Descriptive statistics

Continuous variables (age and MMSE score) are summarised as mean ± standard deviation and median (interquartile range), stratified by phenotype group. Categorical variables are reported as frequencies and percentages. For each metal, the proportion of samples with concentrations above and below the limit of detection (LOD) was computed separately for Control and AD groups. Between-group differences in detection frequency were assessed using Fisher’s exact test, which was preferred over the chi-square test given the small sample size and low cell counts observed for several metals. For samples with detectable concentrations, descriptive statistics (mean ± SD, median IQR, minimum, maximum) were computed per metal per group. Differences in detected concentrations between Control and AD groups were evaluated using the Mann-Whitney U test, appropriate given the non-normal distribution of blood metal concentrations and the small numbers of detected values.

Handling of Left-Censored Data

Blood metal concentrations are subject to left-censoring when measurements fall below the instrument-specific LOD. Given the variable and in some cases high degree of censoring observed across metals, imputation strategies were selected based on the censoring characteristics of each metal. For Lithium, Lead, and Arsenic, group-stratified Kaplan-Meier (KM) imputation was applied, implemented separately within each phenotype group to preserve group-specific distributional characteristics. The KM estimator treats BDL observations as left-censored survival times, using the restricted mean of the fitted survival function as the imputed value for censored observations. Imputation was not performed when all observations within a group were censored or when fewer than two detected values were available; in these cases imputed values were set to missing. Following imputation, metal concentrations were log-transformed to address right skewness prior to regression modelling. Mercury was excluded from KM imputation due to extreme censoring ( > 80%) combined with only a single detected value in the Control group, rendering distributional estimation unreliable. Mercury was therefore retained only in the sensitivity analysis using binary detection status. Arsenic was detected in all samples across both groups (100% detection rate), precluding binary detection analysis; it was therefore included only in the primary continuous analysis.

Primary association analysis

Associations between imputed log-transformed metal concentrations (Lithium, Lead, Arsenic) and AD status were examined using logistic regression (GLM with a binomial link function), with AD diagnosis as the binary outcome and Control as the reference group. Associations between metal concentrations and cognitive function, measured by the Mini-Mental State Examination (MMSE) score, were examined using linear regression (LM). All models were adjusted for age and sex as covariates. Results are reported as odds ratios (ORs) with 95% confidence intervals for the logistic models, and as beta coefficients (β) with 95% confidence intervals for the linear models.

Sensitivity analysis

Given the high degree of censoring in several metals, a sensitivity analysis was conducted using a binary detection variable (detected vs. non-detected) in place of continuous imputed concentrations, for Lithium, Lead, and Mercury. This approach avoids reliance on imputation assumptions by treating detectable exposure itself as the exposure of interest. The same logistic and linear regression models were applied, adjusted for age and sex, with the non-detected category as the reference group. Results are reported as ORs and β coefficients with 95% confidence intervals. Arsenic was excluded from this analysis due to universal detection, and Mercury was excluded from the primary analysis as described above.

Results

Study population

The study included 100 participants: 50 controls and 50 individuals diagnosed with Alzheimer’s disease (AD). The mean age was 61.96 ± 6.74 years (median 63, IQR 57–65) in the Control group and 69.82 ± 8.04 years (median 69.5, IQR 65–75) in the AD group, reflecting the expected older age distribution in AD. The Control group comprised 56% males and 44% females, while the AD group was evenly distributed by sex (50% each). Mean MMSE score was markedly lower in the AD group (13.96 ± 4.79, median 14, IQR 10.5–17) compared to controls (27.2 ± 1.55, median 27, IQR 26–28.75), confirming the expected difference in cognitive function between groups (refer to Table 1).

Table 1.

Demographic and clinical characteristics of study participants

Variable	Control (n = 50)	AD (n = 50)
	Mean ± SD / n (%)	Mean ± SD / n (%)	Median (IQR)
Age (years)	61.96 ± 6.74	69.82 ± 8.04	63 (57–65) vs 69.5 (65–75)
MMSE score	27.2 ± 1.55	13.96 ± 4.79	27 (26–28.75) vs 14 (10.5–17)
Sex
Female	22 (44%)	25 (50%)	—
Male	28 (56%)	25 (50%)	—

Continuous variables are presented as mean ± standard deviation (SD) and median (interquartile range, IQR). Categorical variables are presented as frequency and percentage. AD Alzheimer’s disease, MMSE Mini-Mental State Examination. n = 50 per group.

Blood metal detection frequencies

Detection rates varied substantially across metals and between groups (refer to Table 2 and Fig. 1). Lithium was detected in 58% of controls (n = 29/50) compared to only 28% of AD patients (n = 14/50), a difference that was statistically significant by Fisher’s exact test (OR = 0.285, 95% CI: 0.112–0.700, p = 0.004). Lead was detected in the majority of participants in both groups (90% controls vs. 96% AD), with no significant difference between groups (OR = 2.642, 95% CI: 0.407–29.071, p = 0.436). Arsenic was detected in 98% of controls and 100% of AD patients, with no between-group difference (p = 1.000); given universal detection, Arsenic was excluded from binary detection analyses. Mercury showed a strikingly different detection pattern, with only 2% of controls (n = 1/50) having detectable concentrations compared to 24% of AD patients (n = 12/50), a highly significant difference (OR = 15.143, 95% CI: 2.074–672.372, p = 0.002).

Table 2.

Blood Metal Detection Frequencies and Between-Group Comparisons

Metal	LOD (ppb)	Control Detected n (%)	Control Non-detected n (%)	AD Detected n (%)	AD Non-detected n (%)	Fisher’s OR (95% CI)	p-value
Lithium	0.0053	29 (58%)	21 (42%)	14 (28%)	36 (72%)	0.285 (0.112–0.700)	0.0044 *
Lead	0.0691	45 (90%)	5 (10%)	48 (96%)	2 (4%)	2.642 (0.407–29.071)	0.436
Arsenic	0.0069	49 (98%)	1 (2%)	50 (100%)	0 (0%)	—	1.000 †
Mercury	0.1223	1 (2%)	49 (98%)	12 (24%)	38 (76%)	15.143 (2.074–672.372)	0.0018 *

Detection frequencies are reported as the number and percentage of samples with concentrations at or above the instrument-specific limit of detection (LOD). Between-group differences in detection frequency were assessed using Fisher’s exact test. OR odds ratio, CI confidence interval. *p < 0.05. †Arsenic was detected in all samples; OR could not be estimated. LOD values are expressed in parts per billion (ppb).

The bold values are the statistically significant P values, as described p < 0.05.

Fig. 1. Blood Metal Detection Patterns Across Groups.

A Stacked bar chart showing the proportion of detected (solid) and non-detected (light) samples for each metal in Control and AD groups. Percentages of detected samples are annotated within bars. Significance markers indicate Fisher’s exact test results. B Lollipop plot of Fisher’s exact test odds ratios (log₁₀ scale) with 95% CIs for metals with computable ORs. The dashed line marks OR = 1. Arsenic is omitted (OR undefined due to near-universal detection). Red = p < 0.05; grey = p ≥ 0.05.

Concentrations among detected values

Among participants with detectable concentrations, mean Lithium was higher in controls (24.40 ± 26.03 ppb, median 17.91) than in AD patients (11.35 ± 12.45 ppb, median 7.02). Mean Lead was higher in AD patients (138.96 ± 115.59 ppb, median 118.30) compared to controls (68.98 ± 78.77 ppb, median 49.13). Mean Arsenic was also higher in AD patients (13.41 ± 2.26 ppb, median 13.93) than controls (10.21 ± 2.09 ppb, median 9.83). Mercury concentrations among detected values were substantially higher in AD patients (84.15 ± 55.84 ppb, median 85.14) compared to the single detected control value (20.83 ppb); however, the very limited number of detected values, particularly in controls, precludes a reliable group comparison for Mercury.

Mann-Whitney U tests on detected concentrations confirmed significantly higher Lead concentrations in AD patients compared to controls (U = 634, p = 0.0006), and significantly higher Arsenic concentrations in AD patients (U = 342, p < 0.0001). Lithium concentrations did not differ significantly between groups among detected values (U = 253, p = 0.200). Mercury could not be formally tested due to insufficient detected values in the Control group (n = 1) (refer to Table 3 and Fig. 2).

Table 3.

Blood Metal Concentrations Among Detected Samples by Group

Metal	n	Mean ± SD (ppb)	Median (IQR) (ppb)	Min (ppb)	Max (ppb)	MWU p-value
Control
Lithium	29	24.40 ± 26.03	17.91 (0.99–39.87)	0.191	79.998	0.1995
Lead	45	68.98 ± 78.77	49.13 (23.56–70.62)	0.935	385.823	6.0 × 10−4
Arsenic	49	10.21 ± 2.09	9.83 (8.90–10.80)	6.543	15.908	< 0.0001 *
Mercury	1	20.83 ± —	20.83 (—)	20.825	20.825	‡
AD
Lithium	14	11.35 ± 12.45	7.02 (3.22–16.12)	0.012	47.137	0.1995
Lead	48	138.96 ± 115.59	118.30 (45.15–179.86)	1.760	474.527	6.0 × 10−4 *
Arsenic	50	13.41 ± 2.26	13.93 (11.40–14.91)	9.815	18.833	< 0.0001 *
Mercury	12	84.15 ± 55.84	85.14 (41.04–130.39)	0.963	181.285	‡

Concentrations are reported in parts per billion (ppb) among samples with detectable levels only. Between-group differences were assessed using the Mann-Whitney U test (MWU). MWU p-values are shared across both groups for each metal. *p < 0.05. ‡ Formal statistical testing was not performed for Mercury due to insufficient detected values in the Control group (n = 1). IQR interquartile range, SD standard deviation.

The bold values are the statistically significant P values, as described p < 0.05.

Fig. 2. Blood Metal Concentrations Among Detected Samples.

Box plots showing the distribution of detected concentrations (ppb) per metal, stratified by group. The central line represents the median; boxes show the IQR; whiskers extend to the full range. Individual group sample sizes (n) are shown below. Mann-Whitney U test p-values are displayed above each panel. ‡ Mercury: formal statistical comparison was not performed due to insufficient detected values in the Control group (n = 1).

Imputation of left-censored values

Kaplan-Meier imputation was applied to Lithium, Lead, and Arsenic. Censoring rates for Lithium were 42% in controls and 72% in AD patients; for Lead, 10% in controls and 4% in AD patients; and for Arsenic, 2% in controls with complete detection in the AD group. Mercury was excluded from KM imputation due to 98% censoring in controls with only a single detected value, rendering distributional estimation unreliable; Mercury was retained only in the sensitivity analysis using binary detection status (refer to Table 4).

Table 4.

Imputation method log for left-censored blood metal concentrations

Metal	Group	Imputation Method	n Censored	n Total	% Censored
Lithium	Control	Kaplan-Meier	21	50	42%
Lithium	AD	Kaplan-Meier	36	50	72%
Lead	Control	Kaplan-Meier	5	50	10%
Lead	AD	Kaplan-Meier	2	50	4%
Arsenic	Control	Kaplan-Meier	1	50	2%
Arsenic	AD	Complete data (no censoring)	0	50	0%
Mercury	All	Excluded — binary detection analysis only (98% censored in controls; n = 1 detected)	87	100	87%

Group-stratified Kaplan-Meier (KM) imputation was applied to Lithium, Lead, and Arsenic based on the proportion of left-censored (below limit of detection) values within each phenotype group. Mercury was excluded from KM imputation due to extreme censoring in the Control group (98%) with only one detected value; Mercury was retained in the sensitivity analysis using binary detection status only. n censored number of samples below the limit of detection within each group.

Primary association analysis: KM-imputed concentrations

Association with AD diagnosis

In logistic regression models adjusted for age and sex, Arsenic concentration was strongly and positively associated with AD status (OR = 1321.61, 95% CI: 83.50–38830.29, p = 3.48 × 10⁻⁶; FDR-adjusted p = 2.09 × 10⁻⁵). As this OR reflects a per-unit increase in log-transformed concentration, it is more intuitively expressed per doubling of blood arsenic concentration (OR ≈ 6.97), indicating that each two-fold increase in blood arsenic was associated with approximately seven-fold greater odds of AD. Lead concentration was also positively associated with AD (OR = 1.53, 95% CI: 1.03–2.40, p = 0.046), though this did not survive FDR correction (FDR p = 0.062). Lithium concentration was inversely associated with AD (OR = 0.63, 95% CI: 0.43–0.90, p = 0.015; FDR p = 0.022), indicating that higher Lithium concentrations were associated with lower odds of AD (refer to Table 5A and Fig. 3A).

Table 5.

Primary Association Analysis: KM-Imputed Log-Transformed Metal Concentrations and AD Outcomes; A Logistic Regression — AD Diagnosis (OR, reference = Control); B Linear Regression — MMSE Score (β Coefficient)

A
Metal	OR	95% CI (Lower)	95% CI (Upper)	p-value	FDR-adjusted p
Arsenic	1321.61	83.50	38830.29	3.48 × 10−6 *	2.09 × 10−5 *
Lead	1.533	1.026	2.401	0.0461 *	0.0615
Lithium	0.634	0.429	0.897	0.0145 *	0.0218 *

B
Metal	β Coefficient	95% CI (Lower)	95% CI (Upper)	p-value	FDR-adjusted p
Arsenic	−16.444	−21.715	−11.173	1.46 × 10−8 *	1.75 × 10−7 *
Lead	−1.014	−2.144	0.116	0.0781	0.0937
Lithium	1.208	0.278	2.137	0.0114 *	0.0195 *

Models are adjusted for age and sex. Metal concentrations were log-transformed following Kaplan-Meier imputation. Mercury was excluded from primary analysis (see Table 4). OR odds ratio, β regression coefficient, CI = 95% confidence interval. FDR-adjusted p-values were calculated using the Benjamini-Hochberg method. *p < 0.05. For Arsenic, the OR reflects a per-unit increase in log-transformed concentration (log ppb); a one-unit increase on the log scale corresponds to an approximately 2.7-fold increase in raw concentration.

The bold values are the statistically significant P values, as described p < 0.05.

Fig. 3. Primary Association Analysis: KM-Imputed Metal Concentrations and AD Outcomes.

Forest plots showing associations between log-transformed KM-imputed metal concentrations and (A) AD diagnosis (logistic regression, ORs) and (B) MMSE score (linear regression, β coefficients). Error bars represent 95% confidence intervals. The dashed vertical line indicates the null value (OR = 1 or β = 0). All models adjusted for age and sex. Red = p < 0.05; grey = p ≥ 0.05. OR for Arsenic is plotted on a log₁₀ scale due to its magnitude.

Association with MMSE score

In linear regression models adjusted for age and sex, higher Arsenic concentration was significantly associated with lower MMSE scores (β = −16.44, 95% CI: −21.72 to −11.17, p = 1.46 × 10⁻⁸; FDR p = 1.75 × 10⁻⁷), representing the strongest association observed across all analyses. Conversely, higher Lithium concentration was significantly associated with higher MMSE scores (β = 1.21, 95% CI: 0.28–2.14, p = 0.011; FDR p = 0.020), consistent with its inverse association with AD diagnosis. Lead concentration showed a negative but non-significant trend with MMSE (β = −1.01, 95% CI: −2.14 to 0.12, p = 0.078; FDR p = 0.094) (refer to Table 5B and Fig. 3B).

Sensitivity analysis: binary detection

Association with AD diagnosis

In sensitivity analyses using binary detection status (detected vs. non-detected), Lithium detection was significantly inversely associated with AD (OR = 0.167, 95% CI: 0.055–0.453, p = 0.0008; FDR p = 0.002), indicating that detectable Lithium was associated with nearly six-fold lower odds of AD. Mercury detection was significantly positively associated with AD (OR = 28.41, 95% CI: 4.16–591.84, p = 0.004; FDR p = 0.008), indicating that detectable Mercury was associated with markedly higher odds of AD. Lead detection was not significantly associated with AD (OR = 3.16, 95% CI: 0.56–25.45, p = 0.218) (refer to Table 6A and Fig. 4A).

Table 6.

Sensitivity Analysis: Binary Metal Detection Status and AD Outcomes; A Logistic Regression — AD Diagnosis (OR, reference = Non-detected); B Linear Regression — MMSE Score (β Coefficient, reference = Non-detected)

A
Metal	OR	95% CI (Lower)	95% CI (Upper)	p-value	FDR-adjusted p
Arsenic	—	—	—	—	Excluded §
Lead	3.162	0.557	25.449	0.2176	0.2176
Lithium	0.167	0.055	0.453	0.0008 *	0.0024 *
Mercury	28.414	4.163	591.838	0.0040 *	0.0080 *

B
Metal	β Coefficient	95% CI (Lower)	95% CI (Upper)	p-value	FDR-adjusted p
Arsenic	—	—	—	—	Excluded §
Lead	−3.556	−8.766	1.653	0.1786	0.1948
Lithium	3.964	1.369	6.560	0.0031 *	0.0074 *
Mercury	−7.075	−10.844	−3.306	0.0003 *	0.0012 *

Binary detection variable: Detected (concentration ≥ LOD) vs. Non-detected (concentration < LOD). Models are adjusted for age and sex. § Arsenic was excluded from binary detection analysis due to 100% detection in both groups (no variation). OR odds ratio, β regression coefficient, CI 95% confidence interval. FDR-adjusted p-values were calculated using the Benjamini-Hochberg method. *p < 0.05.

The bold values are the statistically significant P values, as described p < 0.05.

Fig. 4. Sensitivity Analysis: Binary Metal Detection Status and AD Outcomes.

Forest plots showing associations between binary metal detection (Detected vs. Non-detected) and (A) AD diagnosis and (B) MMSE score. Reference category: Non-detected. Arsenic excluded (100% detection in both groups). Models adjusted for age and sex. Error bars = 95% CI. Red = p < 0.05; grey = p ≥ 0.05.

Association with MMSE score

Detection of Lithium was significantly associated with higher MMSE scores (β = 3.96, 95% CI: 1.37–6.56, p = 0.003; FDR p = 0.007), corroborating the primary analysis. Detection of Mercury was significantly associated with lower MMSE scores (β = −7.08, 95% CI: −10.84 to −3.31, p = 0.0003; FDR p = 0.001), consistent with its neurotoxic properties. Lead detection was not significantly associated with MMSE (β = −3.56, 95% CI: −8.77 to 1.65, p = 0.179). Arsenic was excluded from the sensitivity analysis due to universal detection in both groups (refer to Table 6B and Fig. 4B).

Summary of findings

Across both primary and sensitivity analyses, Lithium and Arsenic emerged as the most consistently associated metals with AD status and cognitive function, as illustrated by the association significance heatmap (refer to Fig. 5). Lower Lithium concentrations — and non-detectable Lithium — were associated with higher odds of AD and lower MMSE scores across both analytical approaches, suggesting a potential neuroprotective role; this consistency is reflected in Fig. 5 by significant cells in both the primary and sensitivity rows for Lithium. Higher Arsenic concentrations were strongly associated with greater odds of AD and markedly lower MMSE scores, with the strongest −log₁₀(p) values in the entire heatmap, consistent with its neurotoxic properties. Mercury detection was significantly associated with AD and lower cognitive scores in the sensitivity analysis, though interpretation is limited by the very low detection rate in controls; notably, Mercury shows significance only in the sensitivity analysis rows of Fig. 5, absent from the primary analysis where it could not be modelled. Lead showed borderline associations with AD in the primary analysis that did not withstand multiple testing correction, and no significant associations with MMSE, reflected in Fig. 5 by consistently pale cells across all four model rows.

Fig. 5. Association Significance Heatmap Across All Models and Metals.

Heatmap displaying −log₁₀(p-values) for all metal–outcome associations across primary (KM-imputed) and sensitivity (binary detection) analyses. Darker red cells indicate stronger statistical associations. Exact p-values and significance stars are annotated within each cell. Cells marked ‘—(excl.)’ indicate metals excluded from that analysis. The colorbar dashed line marks the p = 0.05 threshold. *p < 0.05; **p < 0.01; ***p < 0.001.

Discussion

This study investigated the association between blood concentrations of toxic heavy metals (arsenic, lead, mercury) and the essential trace element lithium with Alzheimer’s disease (AD) and cognitive performance in an Egyptian cohort. Our principal findings reveal a robust and significant association between elevated blood arsenic levels and increased AD risk and cognitive decline. We also found elevated levels of lead and mercury in AD patients compared to controls, although these associations were less consistent across analytical approaches. Conversely, lithium levels were lower in the AD group and were associated with reduced odds of AD and better cognitive performance, suggesting a potential protective role. These findings contribute to the growing body of evidence linking environmental metal exposure to AD pathogenesis, particularly within an understudied LMIC context, and highlight the critical importance of statistical methods for handling data with high proportions of measurements below the detection limit.

The most pronounced association in our study was with arsenic. AD patients had significantly higher blood arsenic concentrations than controls (median 13.93 ppb vs. 9.83 ppb), arsenic was detected in all participants in both groups, and higher arsenic was strongly associated with lower MMSE scores (β = −16.44, 95% CI: −21.72 to −11.17, p = 1.46 × 10⁻⁸) and increased odds of AD (OR = 1321.61 per log-unit increase, 95% CI: 83.50–38830.29, p = 3.48 × 10⁻⁶; FDR-adjusted p = 2.09 × 10⁻⁵) in the primary KM-imputed analysis. Because arsenic concentrations were log-transformed prior to modelling, the OR reflects a per-unit increase on the natural log scale, equivalent to an approximately 2.7-fold increase in raw concentration; re-expressed per doubling of blood arsenic, this corresponds to an OR of approximately 6.97, indicating that even modest increases in arsenic burden are associated with substantially elevated AD risk. These associations survived FDR correction and were consistent across both outcomes, confirming arsenic as the most robust environmental risk factor identified in this cohort. This aligns with a body of experimental literature demonstrating that chronic arsenic exposure can drive key AD pathological processes^10,11,22. Preclinical studies demonstrated that arsenic induces oxidative stress, promotes tau hyperphosphorylation, and enhances the amyloidogenic processing of APP, resulting in a dose-dependent increase in Aβ production^23,24. Our results translate these mechanistic insights to a human population, providing epidemiological evidence that environmental arsenic exposure may be a significant modifiable risk factor for AD. In Egypt, potential sources of heavy metal exposure include contaminated drinking water and industrial pollution, underscoring the need for public health interventions to mitigate exposure in vulnerable populations. Because arsenic was universally detected across all 100 participants, a binary detection approach was not applicable for this metal; the continuous KM-imputed analysis therefore represents the sole analytical approach and its findings should be interpreted accordingly.

While the neurotoxicity of lead is well-established, its specific role in AD has been inconsistent across epidemiological studies. In our primary analysis using KM-imputed log-transformed concentrations, lead showed a significant positive association with AD diagnosis (OR = 1.53, 95% CI: 1.03–2.40, p = 0.046), though this did not survive FDR correction (FDR-adjusted p = 0.062), and its association with MMSE score was negative but non-significant (β = −1.01, 95% CI: −2.14 to 0.12, p = 0.078). In the sensitivity analysis using binary detection status, lead detection was not significantly associated with AD (OR = 3.16, 95% CI: 0.56–25.45, p = 0.218) or MMSE (β = −3.56, 95% CI: −8.77 to 1.65, p = 0.179), though the wide confidence intervals reflect the small number of non-detected values (5 controls, 2 AD patients) and thus limited discriminatory power of the binary approach for this metal. Among detected samples, lead concentrations were significantly higher in AD patients than controls (median 118.30 vs. 49.13 ppb; Mann-Whitney U = 634, p = 6×10⁻⁴), providing descriptive evidence of group differences even where the regression models were underpowered. Lead is known to disrupt synaptic function, inhibit the NMDA receptor, and increase the expression of APP and BACE1, thereby promoting the formation of amyloid plaques30. The attenuation of its association after FDR correction may reflect limited statistical power in multivariable models given the sample size, the high variability in lead concentrations, or residual confounding by age. Nonetheless, the consistent directional signal across analytical approaches suggests that lead exposure remains a relevant environmental risk factor worthy of further investigation.

The interpretation of mercury data requires considerable caution and is shaped entirely by the extreme censoring observed in this cohort. Only 1 of 50 controls (2%) had detectable mercury compared with 12 of 50 AD patients (24%), a difference that was highly significant by Fisher’s exact test (OR = 15.14, 95% CI: 2.07–672.37, p = 0.002). Because the Control group had only a single detected mercury value, KM imputation was not feasible for that group and mercury was therefore excluded from the primary continuous analysis entirely. In the binary detection sensitivity analysis, mercury detection was significantly associated with greater odds of AD (OR = 28.41, 95% CI: 4.16–591.84, p = 0.004; FDR-adjusted p = 0.008) and lower MMSE scores (β = −7.08, 95% CI: −10.84 to −3.31, p = 0.0003; FDR-adjusted p = 0.001). Whilst the wide confidence intervals reflect the small number of detected controls, the consistency of significant findings across both outcomes and the large effect sizes suggest that measurable mercury exposure is meaningfully associated with AD risk and cognitive decline in this population. This pattern is itself informative: that mercury exposure above a detectable threshold appears concentrated in the AD group is consistent with the known neurotoxicity of mercury, which binds to thiol groups, disrupts glutathione balance, and promotes both amyloid plaque and neurofibrillary tangle formation in model systems²⁵. These findings should nonetheless be considered preliminary and hypothesis-generating, given the analytical constraints imposed by near-universal non-detection in controls.

Our study adds to the emerging literature on the potential neuroprotective role of lithium in AD. We found that lithium was detected in only 28% of AD patients compared with 58% of controls (Fisher’s exact OR = 0.285, 95% CI: 0.112–0.700, p = 0.004), and that among detected samples, median lithium concentrations were lower in the AD group (7.02 ppb vs. 17.91 ppb). In the primary KM-imputed analysis, higher lithium concentration was significantly associated with reduced odds of AD (OR = 0.634, 95% CI: 0.429–0.897, p = 0.015; FDR-adjusted p = 0.022) and higher MMSE scores (β = 1.21, 95% CI: 0.28–2.14, p = 0.011; FDR-adjusted p = 0.020). These primary findings were corroborated and strengthened in the binary sensitivity analysis, in which lithium detection was associated with substantially lower odds of AD (OR = 0.167, 95% CI: 0.055–0.453, p = 0.0008; FDR-adjusted p = 0.002) and higher MMSE scores (β = 3.96, 95% CI: 1.37–6.56, p = 0.003; FDR-adjusted p = 0.007). The consistency of these associations across both continuous imputed and binary detection approaches, and across both outcomes, represents the strongest evidence in this dataset for lithium as a neuroprotective factor. This is consistent with biological evidence showing that lithium bioavailability is reduced in the AD brain, possibly due to sequestration within Aβ plaques¹². Experimental lithium depletion accelerates AD-like pathology in mice, while lithium supplementation has been shown to inhibit glycogen synthase kinase-3β (GSK3β), a key kinase involved in tau phosphorylation and amyloidogenesis^12,26,27. Epidemiological studies have reported an inverse association between trace levels of lithium in drinking water and dementia incidence²⁸. Our findings from blood measurements in a clinical cohort strengthen the hypothesis that lithium deficiency may be a feature of AD pathogenesis and that maintaining adequate lithium levels could be protective.

The reported lead and mercury concentrations in this cohort may appear elevated relative to reference values from large biomonitoring programs such as NHANES; however, this is consistent with documented environmental conditions in Egypt. Studies from the Nile Delta and peri-urban industrial areas have reported substantially higher blood lead and mercury burdens in Egyptian populations compared with high-income-country reference values (El-Kowrany et al. 2016)²⁴, reflecting differences in environmental exposure sources including contaminated water, agricultural use of heavy-metal-containing pesticides, and proximity to industrial zones. All concentrations in this study are expressed in parts per billion (ppb). Analytical calibration was performed using matrix-matched multi-element standards (R² ≥ 0.99), spike recoveries ranged from 80 to 120%, and relative standard deviations were < 10%, confirming analytical validity. Given the absence of a comprehensive Egyptian population-level biomonitoring dataset, direct comparison with local reference intervals remains limited, and future work should prioritize establishing such normative data.

The interpretation of our results must be considered within the methodological framework employed. Given the highly variable and in several cases extreme censoring observed across metals — ranging from 0% for arsenic in the AD group to 98% for mercury in controls — a tiered imputation strategy was applied rather than a single uniform method. For metals with censoring rates amenable to distributional estimation (Lithium, Lead, Arsenic), group-stratified Kaplan-Meier (KM) imputation was used as the primary approach, as it uses the empirical survival function to estimate values below the LOD without imposing parametric assumptions, and is considered more appropriate than simple substitution methods for data with moderate-to-high censoring²⁹. Mercury was excluded from the KM-imputed primary analysis due to near-complete censoring in the Control group (98%, n = 1 detected), rendering distributional estimation unreliable. Arsenic was excluded from the binary detection sensitivity analysis due to universal detection across both groups, precluding any variation to model. A binary detection sensitivity analysis (Detected vs. Non-detected) was conducted for Lithium, Lead, and Mercury, which avoids imputation assumptions entirely by treating detectable exposure as the exposure metric. The convergence of significant findings across both analytical approaches for Lithium — and the novel significant findings for Mercury solely from the binary approach — underscores the value of complementary analytical strategies in environmental epidemiology, where non-detectable values are common, and the choice of handling method can materially influence conclusions. Multiple testing was addressed throughout using the Benjamini-Hochberg FDR correction, and only associations surviving FDR adjustment are treated as primary findings.

Our study has several limitations. First, the cross-sectional design precludes causal inference; it is possible that AD-related changes in diet, metabolism, or behavior influence metal levels, rather than vice versa. Second, the AD group was significantly older than the control group (mean 69.82 vs. 61.96 years). While we adjusted for age in all regression models, residual confounding remains a possibility. Third, regression analyses were adjusted only for age and sex. While these are major determinants of both metal exposure and AD risk, other potentially important confounders — including educational level, occupational history, medication use, family history of AD, alcohol and tobacco consumption, and dietary habits — were not uniformly available for all participants and therefore could not be included in the models. Fourth, a single measurement of blood metal levels may not reflect long-term exposure history, which is more relevant for a chronic neurodegenerative disease such as AD. Fifth, the high censoring rates for Lithium and Mercury introduce uncertainty and reduce statistical power, and the binary detection findings for these metals should be considered exploratory pending replication. Sixth, co-exposure analyses between metals were not performed in the current pipeline; whilst the observed detection patterns suggest partially overlapping environmental sources, a full mixture analysis — for example using weighted quantile sum (WQS) regression or Bayesian kernel machine regression (BKMR) — is warranted in future studies with larger sample sizes to assess joint and interactive neurotoxic effects. Finally, our cohort was recruited from a single country and the generalizability of findings requires verification in other populations.

Despite these limitations, our study possesses notable strengths. These include a well-characterised cohort with clinically confirmed AD diagnoses and comprehensive cognitive assessments, a multi-centre design enhancing sample representativeness, the application of a principled tiered imputation strategy with a complementary sensitivity analysis to handle a complex censored dataset, and FDR correction for multiple comparisons. Most importantly, this study addresses a critical gap in the literature by providing data from an LMIC setting in North Africa, a region disproportionately affected by environmental pollution yet severely underrepresented in dementia research.

Conclusion

Our findings from an Egyptian cohort provide compelling evidence linking blood metal signatures to Alzheimer’s disease. The strong association of arsenic with AD risk and cognitive decline underscores the potential impact of environmental toxins on brain health. The suggestive findings for lead and mercury warrant further investigation in larger studies. The potential protective role of lithium offers a promising avenue for further research into nutritional and therapeutic interventions. Future longitudinal studies with repeated metal measurements are needed to confirm these associations and elucidate the causal pathways. Ultimately, reducing exposure to toxic heavy metals could represent a critical, actionable strategy for mitigating the growing burden of AD in Egypt and other LMICs.

Author contributions

Conceptualization: M.O., M.S.; Data curation: E.K., G.F., A.I., A.Q., E.A.; Formal analysis: M.O., M.S.; Funding acquisition: M.S.; Investigation: E.K., G.F., A.I., A.Q., E.A.; Methodology: M.O., S.H., S.M., M.S.; Project administration: M.O., S.H., S.M., M.S.; Resources: M.S.; Software: M.O.; Supervision: M.S.; Validation: M.O., S.H., S.M., M.S.; Visualization: M.O.; Writing – original draft: M.O.; Writing – review and editing: M.O., M.S.

Peer?review

Peer review information

Communications Medicine thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported in part by funding from the Davos Alzheimer’s Collaborative (DAC): Mohamed Salama and Mai Othman through the DAC Global Cohorts program. Mohamed Salama and Shimaa Heikal were funded by the 100 PhDs for Africa programme under the UM6P- EPFL Excellence in Africa Initiative. The funding sources were not involved in the study design; collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

Data availability

Anonymized clinical and blood metal concentration data can be provided to bona fide researchers upon request from the corresponding author(s). The source data for Figs. 1–5 is in Supplementary Data.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s43856-026-01844-8.