Lower Serum Calcium as a Potentially Associated Factor for Conversion of Mild Cognitive Impairment to Early Alzheimer’s Disease in the Japanese Alzheimer’s Disease Neuroimaging Initiative

Sample datasets

This study is a sub-analysis of the J-ADNI study results. We obtained the J-ADNI dataset from the National Bioscience Database Center (NBDC), with approval from its data access committee (https://humandbs.biosciencedbc.jp/en/hum0043-v1). The J-ADNI was launched in 2007 as a public-private partnership, led by Principal Investigator Takeshi Iwatsubo, MD. The primary goal of J-ADNI was to test whether clinical and neuropsychological assessment, together with biological markers including, but not limited to, serial MRI and PET, could be combined to measure the progression of MCI and mild AD in the Japanese population. As such, the investigators within the J-ADNI contributed to the design and implementation of the J-ADNI and/or provided data, but they did not participate in analyses within, or in writing of, this report. A complete list of J-ADNI investigators can be found at: https://humandbs.biosciencedbc.jp/en/hum0043-j-adni-authors.

The inclusion/exclusion criteria used in the J-ADNI largely correspond to those for the NA-ADNI [2, 26] and they are described at: https://upload.umin.ac.jp/cgi-open-bin/ctr_e/ctr_view.cgi?recptno=R000001668. General inclusion criteria for MCI participants were as follows: participants themselves or participants’ family must have presented with complaints of memory disturbances, participants’ age must be within the range of 60 to 84 years at baseline, they must be native Japanese speakers, their total Mini-Mental State Examination (MMSE) score must fall in the range of 24 to 30 (inclusive), and their Clinical Dementia Rating (CDR) and memory box of CDR should be 0.5 and 0.5 or greater, respectively. Moreover, patients who had medical history of central nervous system disorders or trauma, who could not receive brain MRI examination, who had an acute brain-related event at the time of screening, who had recent history of depression or alcohol abuse, or who had etiology of treatable dementia (such as vitamin B₁₂ deficiency, folate deficiency, thyroid abnormality, or syphilis) were excluded from the study. The follow-up period was 3 years (36 months) for MCI.

In addition, we obtained the NA-ADNI datasets from ADNI-1, ADNI-GO, and ADNI-2 from the Laboratory of Neuro Imaging (LONI), with their approval. The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of the ADNI was to test whether clinical and neuropsychological assessments and biomarkers, such as serial MRI and PET, could be combined to measure the progression of MCI and early AD. Eligible participants whose diagnosis at baseline (column ‘Dx_bl’ in the ‘ADNIMERGE’ file within the dataset) corresponded to late MCI were included in the analysis. The data used in the preparation of this article were obtained from the ADNI database (http://adni.loni.usc.edu), under permission (downloaded on July 2018).

Assessments and clinical variables

We used the incidence of MCI conversion to AD as the primary measure in this study. In cases where participants presented conversion and subsequent reversion during the observation period, the participants were considered as converted. The clinical and laboratory data used for analysis included the participants’ sex, age, body weight, body height, family history of dementia, marital status, smoking habits, neuropsychological functional tests’ results, Geriatric Depression Scale (GDS) scores, fasting blood sugar level, total cholesterol (T-cho) level, uric acid (UA) level, electrolytes’ level (sodium, potassium, calcium), phosphate level, C-reactive protein (CRP) level, serum albumin (Alb) level, serum creatinine level, estimated glomerular filtration rate (eGFR) [28], creatinine clearance (calculated from Cockroft-Gault equation) [29], plasma cell counts (white blood cells, WBC; blood hemoglobin level, Hb; blood platelet level, Plt), CSF Aβ₄₂ level, CSF t-tau, and phosphorylated tau (p-tau) levels, Fazekas periventricular hyperintensity (PVH) score, deep and subcortical white matter hyperintensity (DWMH) scores from the screening MRI scan (scored by a trained radiologist blind to clinical information), APOE ɛ4 allele status, amyloid positivity (defined as either amyloid-PET positive or CSF Aβ₄₂ less than 333 pg/mL) [26], and total intracranial and hippocampal volumes of brain. From the several neuropsychological tests used in the J-ADNI cohort, we chose the Clinical Dementia Rating Sum of Boxes (CDR-SOB), Alzheimer’s Disease Assessment Scale – cognitive subscale 13 (ADAS-cog13), MMSE, and Functional Activities Questionnaire (FAQ) to assess the longitudinal cognitive and global function of participants. The progression speed of these scores is calculated by taking the difference between the baseline score and the last visit score and dividing by the number of months in the maximum follow-up period. Serum calcium level was corrected in cases of hypoalbuminemia (serum Alb less than 4 g/dL), where corrected serum calcium level (mg/dL) = raw calcium level (mg/dL) +  (4 – serum albumin level (g/dL)) [30]. Serum calcium was measured by arsenazo III method using Serotec Ca-AL kit (Serotec, Sapporo, Japan). Amyloid positivity was defined by visual inspection of amyloid-PET images, using Pittsburgh Compound B as a ligand, performed by trained radiologists who were blinded to any clinical information [31]. The ethnicity of participants in the J-ADNI cohort is almost entirely eastern-Asian, although ethnicity was not always recorded.

Corresponding data from the NA-ADNI datasets were acquired as follows: CSF biomarkers were obtained from the ‘UPENNBIOMK9_04_19_17.csv’ file on the database. In addition, AV45-using amyloid-PET scan results we obtained using the feature ‘SUMMARYSUVR_WHOLECEREBNORM_1.11CUTOFF’ in the ‘UCBERKELEYAV45_12_08_15.csv’ file on the database. As the renal function measure in NA-ADNI, we used the creatinine (CCr) levels alone.

Statistical analyses

All statistical analyses were performed using R version 3.3.3 (https://www.r-project.org). For categorical data, we used Fisher exact test and summarized the data as median and interquartile range (IQR). For continuous data, we used Wilcoxon rank sum test and summarized the data as frequency or %. A p-value of <0.05 were considered statistically significant.

In the conversion-free period survival analysis, we used log-rank test by regarding the conversion from MCI to AD as the event occurred. The censoring of observation was determined when the participant converted during the observation period, dropped off or finished the entire 36 months of observation. In adjusting multivariate factors, we used a binomial logistic regression model for the incidence of conversion.

Ethics

The study protocol was approved by the University of Tokyo ethics committee (11628).

Identifying risk factors for conversion to AD in the J-ADNI cohort

Out of 234 eligible MCI participants, 121 (51.7%) converted to AD during the 3 years of observational period (mean annual conversion rate: 21.54%). Among these 121 converters, 4 participants showed reversion to MCI during the observation. In the NA-ADNI results, 238 (42.3%) out of 563 eligible MCI participants converted to AD within the first 3 years of observation (mean annual conversion rate: 16.7%), thus showing a significantly lower conversion-rate compared to J-ADNI (p = 0.016, Fisher exact test).

Table 1 summarizes the univariate analysis results comparing the basic characteristics of AD-converted participants with those of non-converted participants in the J-ADNI cohort. Several factors were significantly associated with increased conversion risk, including: having APOE

ɛ4 allele(s) (Odds Ratio [OR] = 4.01), being amyloid-positive (as defined by amyloid-PET or CSF Aβ_42 ; OR = 3.92), being female (OR = 1.85), having a smaller number of education years, lower MMSE scores, higher ADAS-cog13, FAQ scores, and CDR-SOB scores at baseline, smaller hippocampal volume, or taking donepezil at baseline (OR = 2.20). Meanwhile, there were no significant associations between conversion and age or history of smoking.

In the corresponding basic characteristic analysis in the NA-ADNI results, a statistically significant association with AD conversion was seen for the APOE ɛ4 status (OR = 2.63), amyloid-PET positivity (OR = 10.6), medication status of donepezil (OR = 2.46), smaller hippocampal volume, and baseline cognitive and neuropsychiatric functional scales (MMSE, ADAS-cog13, FAQ, and CDR-SB) (Supplementary Table 1).

Medical history for the J-ADNI cohort is summarized in Table 2. There was a negative association between conversion risk and medical history of musculoskeletal (OR = 0.43) or dermatological disorders (OR = 0.20). These results are only partially reproduced in NA-ADNI participants, where only those with neurological medical history were less likely to convert to AD (OR = 0.65) (Supplementary Table 2).

A detailed medication history at baseline was available for the J-ADNI cohort, as shown in Table 3. Patients taking donepezil or not taking antihypertensive drugs were more likely to convert to AD (OR = 2.20 and OR = 0.47, respectively). In addition, there was a significant tendency in the difference of conversion incidence between those taking and not taking ‘Yokukansan’ (OR = 3.45, p = 0.052), one of the Kampo (or Chinese herbal medicine) drugs which is sometimes prescribed for neuropsychiatric symptoms in Japan [32]. There was no significant difference in the prevalence of conversion in those taking vitamin D supplements or peroral calcium supplements. Similarly, in the NA-ADNI results, those taking donepezil (OR = 2.46) or memantine (OR = 2.82) were more likely to develop conversion (Supplementary Table 3) (note that the medication items obtained from NA-ADNI does not correspond to that of J-ADNI).

Hematological results for the J-ADNI are summarized in Table 4. MCI converters in the J-ADNI cohort showed lower corrected serum calcium, higher serum T-cho, lower serum UA, higher total CSF t-tau and CSF p-tau, and lower CSF Aβ₄₂ (CSF results not shown; [26]). There was no difference in the baseline renal function or in its change in the 3 years of observation. In contrast, MCI converters in the NA-ADNI cohort showed no significant difference in any of the corresponding serum laboratory data (Supplementary Table 4). CSF biomarkers all had robust significant differences as in the J-ADNI.

Regarding serum calcium, where normal levels are in the rage of 8.7–10.2 mg/dL, patients with hypocalcemia (corrected calcium <8.7 mg/dL) amounted to 4.1% (5/121) of the total number of converters and 0.9% (1/113) of the total number of non-converters, without a statistically significant difference (p = 0.214). Patients with hypercalcemia (corrected calcium >10.2 mg/dL) were 0% (0/121) of the total number of converters and 0% (0/113) in non-converters in J-ADNI. When the MCI participants in the J-ADNI cohort were split into two groups according to their level of corrected serum calcium—higher or lower than its median value (cut-off: 9.2 mg/dL), there was a significant difference in the overall conversion rate: 58.7% (64/109) in the subgroup with lower serum calcium level (corrected serum calcium <9.2 mg/dL) against 45.6% (57/125) in the subgroup with higher serum calcium level (corrected serum calcium ≥9.2 mg/dL; p = 0.049). These two subgroups also showed a significant difference in the conversion-free period survival analysis (log-rank test, p = 0.011) (Fig. 1). However, we could not replicate these results with the NA-ADNI data: differences in the incidence of conversion to AD were non-significant between the higher calcium subgroup (conversion rate of 45.1% (130/288), corrected serum calcium ≥9.9 mg/dL) and the lower calcium subgroup (conversion rate of 38.3% (95/248), corrected serum calcium <9.9 mg/dL; p = 0.115). Moreover, conversion-free period survival analysis showed opposite results compared to the J-ADNI, with the higher calcium subgroup showing significantly longer periods until conversion compared to the lower calcium subgroup (log-rank, p = 0.031) (Fig. 1).

OPEN IN VIEWER

Fig.1

Kaplan-Meier conversion-free survival curves in accordance with higher/lower serum calcium level in each of J-ADNI (A) or NA-ADNI (B). In J-ADNI (A), MCI participants with lower serum calcium levels (corrected calcium <9.2 mg/dL) showed significantly shorter conversion-free period than those with higher serum calcium levels (corrected calcium ≥9.2 mg/dL; log-rank test, p = 0.011). Conversely, in NA-ADNI (B), MCI subjects with higher serum calcium levels showed significantly shorter conversion-free period than those with lower calcium levels (log-rank test, p = 0.031).

Multivariate adjustment of serum calcium effect on MCI conversion

Next, we performed multivariate analysis for MCI conversion, focusing mainly on the effect of serum calcium. Conversion was set as the objective variable and the effect of serum calcium was assessed while adjusting for confounding factors including: age, sex, education period, APOE ɛ4 status, blood laboratory data at baseline (serum phosphate level, serum UA, t-cho, eGFR at baseline, total neuropsychological scores at baseline (ADAS-cog13 and FAQ), medication history (donepezil, ‘Yokukansan’, calcium supplements or vitamin D supplements), past medical history (cardiovascular, neurological, musculoskeletal, or head disorders), and Fazekas PVH score and the total brain volume at baseline (intracranial volume and total hippocampal volume), assessed with MRI. The variables above were chosen by considering each variable’s medical or statistical significance in relation to the MCI conversion. It should be noted that we have not included the amyloid-positivity results into the multivariate analysis. Our rationale for this choice was based on a concern about the possibility of statistical instability due to the relatively small size of sample size of the amyloid data (n = 50 among converter and n = 47 among non-converters). A stepwise, backward and forward selection procedure (function ‘stepAIC’ in R) was used, and the derived final model included serum calcium level as shown in Table 5 (model A). Serum calcium level remained in the derived predictive model: adjusted OR = 1.132 per 0.1 mg/dL decrease (95% CI: 1.016–1.266). When we adjusted for the changes in renal function during the 3 years of observation (total n = 149), serum calcium level was still a significant factor: adjusted OR = 1.156 per 0.1 mg/dL decrease (95% CI: 1.015–1.329). Interestingly, the change in eGFR during the 3 years of observation was also significantly associated with conversion, with the magnitude of the decrease in eGFR being negatively correlated with the risk of conversion (adjusted OR = 0.939 per-1 mL/min/1.73 m² decrease from baseline eGFR; 95% CI: 0.889–0.987) (Supplementary Table 5, model B). The same result was obtained in an alternative model using CCr instead of eGFR (data not shown), in which body weight is accounted for in the calculation. There was no evidence for multicollinearity for any of the variables, with their variance inflation factor being less than 5. The logistic regression analysis was also performed on data from the NA-ADNI cohort, with results sharing similar features, including the finding of baseline cognitive scores and hippocampal volume as significant predictors. However, the findings regarding serum calcium level could not be reproduced (results not shown). Among factors shown in Table 5 (model A), higher baseline ADAS-cog13 had the best predictive performance for MCI conversion both in J-ADNI or NA-ADNI cohort (area under curve (AUC) = 0.746 in J-ADNI and AUC = 0.756 in NA-ADNI).

Since the above multivariate analysis did not included amyloid-positivity results among the covariates adjusted for, we assessed association between the serum calcium level and this factor. Results showed no significant difference in the serum calcium level between those with and without amyloid positivity (p = 1.000), amyloid-PET positivity (p = 0.562), higher/lower CSF Aβ₄₂ level (p = 0.522), or APOE ɛ4 status (p = 0.534). Furthermore, there was no association between the serum calcium level and the raw level of CSF Aβ₄₂, t-tau or p-tau (Spearman’s rank correlation: p = 0.987, p = 0.738 and p = 0.605, respectively). In addition, we checked for potential confounders of the calcium level by screened for factors having significant association with the serum calcium level; such factors were age (p = 0.002), serum phosphate (p < 0.001, Spearman’s rank correlation), Alb (p < 0.001), T-cho level (p < 0.001), and medication history of ‘Yokukansan’ (p = 0.002), all of which are corrected for in the above multivariate adjustment. In addition, participants with lower serum calcium level was slightly but significantly had higher Fazekas PVH than those with median or higher serum calcium level (p = 0.005, Wilcoxon rank sum test). We summarized basic clinical features between subgroups with lower and higher serum calcium level in the Supplementary Table 6.