Health Conditions Associated with Alzheimer’s Disease and Vascular Dementia

Abstract

Objective:

We examined medical records to determine health conditions associated with dementia at varied intervals prior to dementia diagnosis in participants from the Baltimore Longitudinal Study of Aging (BLSA).

Methods:

Data were available for 347 Alzheimer’s disease (AD), 76 vascular dementia (VaD), and 811 control participants without dementia. Logistic regressions were performed associating ICD-9 health codes with dementia status across all timepoints, at 5 and 1 year(s) prior to dementia diagnosis, and at the year of diagnosis, controlling for age, sex and follow-up length of the medical record.

Results:

In AD, the earliest and most consistent associations across all time points included depression, erectile dysfunction, gait abnormalities, hearing loss, and nervous and musculoskeletal symptoms. Cardiomegaly, urinary incontinence, non-epithelial skin cancer, and pneumonia were not significant until 1 year before dementia diagnosis. In VaD, the earliest and most consistent associations across all time points included abnormal EKG, cardiac dysrhythmias, cerebrovascular disease, non-epithelial skin cancer, depression, and hearing loss. Atrial fibrillation, occlusion of cerebral arteries, essential tremor, and abnormal reflexes were not significant until 1 year before dementia diagnosis.

Interpretation:

These findings suggest that some health conditions are associated with future dementia beginning at least 5 years before dementia diagnosis and are consistently seen over time, while others only reach significance closer to the date of diagnosis. These results also show that there are both shared and distinctive health conditions associated with AD and VaD. These results reinforce the need for medical intervention and treatment to lessen the impact of health comorbidities in the aging population.

INTRODUCTION

A variety of factors have been related to the onset of dementia in older individuals. The most common risk factors include older age, genetic factors such as apolipoprotein E (APOE) genotype and family history, education level, as well as health conditions including a variety of vascular-related illnesses ^{1, 2}. It remains difficult to accurately predict the potential for future dementia in older individuals, however, because no single factor appears to be the common rate-limiting step at the onset of the cascade leading to dementia. This suggests that the pathway to dementia is heterogeneous and multifaceted. To better understand one facet of this pathway, we examine health conditions associated with the onset of dementia in the years leading up to diagnosis.

Many studies have examined the relationship between illness and dementia. Studies investigating the effects of individual diseases on Alzheimer’s disease (AD) have shown that hypertension, diabetes, and dyslipidemia increase the risk for mild cognitive impairment (MCI) or dementia ^{3, 4}. Some studies investigating multiple illnesses or comorbidities have shown that hypertension, diabetes, cerebrovascular disease, and psychiatric disorders ^5–7 predict dementia, while others show that depression, osteoporosis/arthritis, fractures, and anemia are associated with cognitive impairment and dementia ^7–9. These findings suggest that while hypertension, cerebrovascular disease, and depression are most commonly associated with dementia in the literature, there is some variability in the health conditions linked to AD. Further, the timing of the onset of these conditions may be especially important ^{10, 11}, but less is known about the years immediately preceding dementia diagnosis.

In vascular dementia (VaD), the etiology is more precise, as vascular-related health conditions are thought to be the primary cause of VaD. These conditions include cardiovascular disease, cerebrovascular disease, metabolic disorders, and stroke ^{12, 13}. However, the timing of these conditions in relation to dementia onset has not been fully studied. From a clinical standpoint, a better understanding of the temporal trends of health risks is critical in determining the likelihood of future dementia in those whose physical health is deteriorating. A better grasp of health risks is also important for understanding common health-related contributors and the timing of their occurrence along the pathway to dementia in older individuals.

To understand the temporal associations of health in the trajectory of dementia, we use a phenome-disease association study (PheDAS) approach to determine the most common health conditions associated with of future dementia in the years preceding onset. Closely related to the genome-wide association study (GWAS) method ¹⁴, this approach allows us to determine the association between a defined disease state (AD or VaD) and its medical record (health condition) phenotype to identify associations that are both statistically significant and easily interpretable ^{15, 16}. Here, we examine both AD and VaD to assess health contributors in two differing forms of dementia common in the older population, with the expectation that these two forms of dementia will vary in their health record phenotypes based on the etiology of the disease. Using medical record data from participants in the Baltimore Longitudinal Study of Aging (BLSA), we investigate the overall associations between health conditions and dementia onset over a period of 15 years in AD and 13 years in VaD. We also examine the temporal associations of health conditions beginning at 5 years prior, 1 year prior, and at the year of diagnosis to further our understanding of both stable and emerging health conditions over time in both dementia subtypes.

METHODS

Participants

This study used data from the BLSA, an ongoing longitudinal study started in 1958 to assess age-changes in physical and psychological health in a cohort of community-dwelling volunteers ¹⁷. Participants currently are evaluated every 1–4 years depending on age (every 4 years for age <60 years; every 2 years for age 60–79 years; every year for age ≥80 years). Data used in the current analyses were collected from the study inception through 2015.

Participants were included in the study if they had valid medical record data. Participants with Alzheimer’s disease (AD) and vascular dementia (VaD) were included if they had data at any timepoint(s) prior to the onset of dementia. Control participants with available medical record data intervals that matched those of the participants with dementia were selected for the analyses. Data were available for 347 AD, 76 VaD, and 811 control participants without dementia. For both the AD and VaD studies, subjects were matched 1:n for the dementia:control participants based on age (age at diagnosis in the dementia groups, age at last visit for controls), sex, and follow-up length of the medical record. Data from 1006 participants were used in the Alzheimer’s disease (AD) analysis. This group included 659 Controls (CN) with no clinical diagnosis of cognitive impairment, and 347 participants with a clinical diagnosis of AD. Of the 347 AD participants, 316 had a dementia diagnosis of AD alone and 31 (9%) had a mixed dementia diagnosis that included AD and vascular dementia. For the VaD analysis, data from 228 BLSA participants were used. This included 152 Controls and 76 participants with a clinical diagnosis of VaD alone. Demographic information and sample sizes for each analysis are shown in Table 1. Differences in the numbers of participants used in each analysis resulted from the availability of health code data at specific visits and availability of matched controls for each analysis.

Table 1. AD and VaD Study Demographics.

Demographic data are shown for both the AD (left columns) and VaD (right columns) analyses. Age represents the mean age at the last follow-up visit for controls and age at diagnosis for AD and VaD participants. The interval reflects time from age 50 to the last visit. Last visit for the AD and VaD groups was defined as the date of diagnosis.

	Controls	AD	Controls	VaD
Overall
n	659	347	152	76
Age (mean (SD))	82.02 (7.62)	82.64 (7.71)	81.33 (7.66)	81.41 (7.77)
Sex (male n (%))	414 (63%)	207 (60%)	100 (66%)	50 (66%)
Interval (mean (SD))	15.85 (11.01)	15.50 (9.96)	20.46 (11.30)	13.42 (9.17)
5 years prior
n	444	291	111	66
Age (mean (SD))	73.22 (9.10)	76.63 (7.99)	73.74 (8.42)	75.87 (7.20)
Sex (male n (%))	289 (65%)	176 (60%)	80 (72%)	46 (70%)
1 year prior
n	538	325	139	74
Age (mean (SD))	77.61 (8.91)	79.75 (8.01)	78.19 (8.63)	78.76 (8.31)
Sex (male n (%))	344 (64%)	195 (60%)	95 (68%)	50 (68%)
Diagnosis
n	579	327	147	75
Age (mean (SD))	79.84 (7.77)	80.62 (7.73)	80.15 (7.95)	80.17 (7.98)
Sex (male n (%))	356 (62%)	195 (60%)	98 (67%)	50 (67%)

Clinical diagnoses of AD and VaD were determined by consensus case conference. At each BLSA visit, participants were screened with the Blessed Information Memory Concentration (BIMC) test, the Clinical Dementia Rating (CDR) Scale, and/or a dementia questionnaire. If the BIMC was ≥4, the CDR was ≥0.5, or if the questionnaire revealed concerns about cognitive status, participants were reviewed at a consensus conference. Diagnosis of AD was based on DSM-III-R (1987) and the National Institute of Neurological and Communication Disorders and Stroke: Alzheimer’s Disease and Related Disorders Association criteria ¹⁸. Diagnosis of VaD was based on a clinical diagnosis of dementia in conjunction with the presence of vascular risk factors (smoking, hypertension, diabetes, atrial fibrillation) and clinical confirmation of stroke on review of medical records, or the presence of lacunar, subcortical or cortical strokes on MRI.

This study was approved by the local Institutional Review Boards. All participants provided written informed consent prior to each assessment.

ICD-9 Health Codes

International Classification of Diseases, 9th Revision (ICD-9) ¹⁹ codes were extracted for each participant from the BLSA medical records database. The ICD-9 data were based on self-report, physical examinations, and medical record history. All available data were collected for all participants and censored to begin at age 50. The last visit for participants with AD or VaD was censored at the year of diagnosis.

Statistical Analyses

For each individual, ICD-9 codes associated with each BLSA visit were collected. All available data from visits prior to and including the last follow-up visit were included in the analyses. For each visit, all ~15,000 ICD-9 codes were then consolidated and mapped to one of 1,866 phenotype codes (phecodes) ^{20, 21}. For example, after consolidation, the “depression” phecode 296.2 included a grouping of ICD-9 depression codes of “major depressive disorder, single episode, mild degree” (ICD-9 = 296.21), “major depressive disorder, recurrent episode, mild degree” (ICD-9 = 292.31), and “depressive disorder NEC” (ICD-9 = 311). The longitudinal phecodes were then aggregated across each participant’s record, yielding a 1×1,866 binary vector or string for each subject in which each column corresponded to a unique phecode. The string was updated to indicate the presence or absence of each phecode over time across the length of the medical record. A value of 1 in this vector indicated that column’s phecode was present in the participant’s record, while a 0 indicated that phecode was absent from the participant’s record. These phecodes, which will be referred to as ‘codes’ throughout the paper, were further grouped into health categories for interpretation. 17 major physiologic or systems categories were examined: circulatory, congenital abnormalities, dermatologic, digestive, endocrine/metabolic, genitourinary, hematopoietic, infectious disease, injuries & poisonings, mental disorders, musculoskeletal, neoplasms, neurological, pregnancy complications, respiratory, sense organs, and symptoms.

The number of repeated measures varied between subjects. To address this, repeated measures were aggregated across each person’s record as follows. After mapping ICD-9 events to phecodes, longitudinal events for each subject were compressed to a single binary vector of length 1×1866, where each of the 1866 columns corresponded to a unique phecode. A value of 1 in a phecode’s column of this vector indicated that the subject had at least one instance of that phecode in their record, while a 0 indicated that the subject did not have an instance of that phecode in their record.

The PheDAS method ^{16, 22} was used to determine significant differences in the prevalence of reported health codes between the participants with dementia (AD or VaD) and respective control groups. Subjects were matched 1:n for the dementia:control participants based on maximum age (age at diagnosis for AD and VaD; age at last follow-up visit for controls), sex, and follow-up length of medical records. For each participant, all available data from visits prior to and including the last follow-up visit were used in the analyses.

Overall differences between the groups were assessed across all available timepoints beginning at age 50, with the last visit censored at the date of diagnosis in the participants with dementia. Timed differences were also assessed at 5 years and 1 year prior to dementia diagnosis, and at the year of dementia diagnosis. Timed differences were assessed by censoring the dementia cohort at 5 years and 1 year prior to dementia diagnosis, and at the year of dementia diagnosis, and included all available datapoints prior to the time of censoring.

Using a series of logistic regressions, each code was assessed for the difference in prevalence between the dementia and control groups. Each regression included cognitive diagnosis as the dependent variable and the ICD code as a binary (presence/absence) independent variable. Age at last visit age, sex and length of follow-up record were included as covariates. To control for multiple comparisons, the second-generation p-value (SGPV) measure described by Blume et al ¹⁵ was used (significant null interval = [0.3, 1.1] (odds ratio) or [−1.20, 0.10] (log odds ratio)).

The odds ratio for each code was determined by the regression coefficient (beta) from the models. The analyses were also stratified by sex to assess differences in the associations between health and dementia in males and females.

RESULTS

Alzheimer’s Disease

Overall differences in AD

The main effect of diagnosis represents the overall difference between AD (n=347) and controls ((CN) n=659) across all timepoints up to the date of diagnosis in the AD group (mean interval AD=15.50 (9.96) years; controls=15.85 (11.01 SD) years). The results showed significant differences for 19 codes (Figure 1). Significant codes were observed in the following categories: dermatologic, digestive, endocrine/metabolic, genitourinary, injuries & poisonings, mental disorders, neurological, respiratory, sense organs, and symptoms. These group differences were positive associations indicating a higher prevalence in the AD group. The categories and codes with higher prevalence (≥5% of participants) in the AD group included mental disorders (delirium dementia/amnestic/other cognitive disorders, depression), genitourinary (urinary incontinence, erectile dysfunction, elevated PSA), neurological (cerebral degeneration, gait abnormality), sense organs (vertiginous syndromes, hearing loss), symptoms (nervous/musculoskeletal, syncope), and respiratory (other dyspnea).

Figure 1. Overall Health Conditions Associated with AD.

Codes with a significantly higher prevalence in the AD group relative to controls over a span of 15 years prior to dementia diagnosis. The log of the odds ratio (dot) and confidence intervals (bars) are shown.

Significant health conditions associated with AD were observed for codes predominately involving genitourinary and mental disorder categories in males. Associations were seen for codes predominantly involving endocrine/metabolic, neurological and symptoms categories in females (Table 2). Health conditions associated with AD were seen for urinary incontinence and gait abnormality codes in both males and females.

Table 2. Overall Health Conditions Associated with AD.

Categories and codes with a significantly higher prevalence in the AD group relative to controls across a span of 15 years prior to diagnosis. The Odds Ratio [95% Confidence Interval (CI)] for each code is shown for the whole AD group. Highlighting in the ‘males’ and ‘females’ columns indicate significant sex effects; missing odds ratios represent codes with <6 positive participants needed for model stability. The percentage of AD and control (CN) participants with each code is also shown.

CATEGORY	CODE	CONDITION	ODDS RATIO All [95% CI]	%AD (%CN)	ODDS RATIO Males	ODDS RATIO Females
Dermatologic	681.5	Cellulitis and abscess of leg, except foot	4.80 [1.49, 15.49]	2.9 (0.6)	13.48 [1.64, 110.77]	2.06 [0.40, 10.58]
Digestive	574.3	Cholecystitis without cholelithiasis	7.72 [1.63, 36.62]	2.3 (0.3)	9.82 [1.14, 84.83]	•
Endocrine/Metabolic	240	Simple and unspecified goiter	2.39 [1.24, 4.62]	6.0 (2.6)	1.72 [0.61, 4.83]	2.95 [1.24, 7.04]
Genitourinary	599.4	Urinary incontinence	1.83 [1.28, 2.62]	22.5 (13.7)	2.05 [1.22, 3.45]	1.84 [1.1, 3.05]
Genitourinary	605	Erectile dysfunction [ED]	2.50 [1.43, 4.38]	8.6 (3.9)	2.24 [1.29, 3.88]	•
Genitourinary	796	Elevated prostate specific antigen [PSA]	2.01 [1.12, 3.58]	7.5 (4.1)	1.97 [1.10, 3.53]	•
Injuries & Poisonings	800.1	Fracture of neck of femur	2.88 [1.32, 6.26]	4.9 (1.7)	7.47 [1.53, 36.41]	1.73 [0.98, 4.06]
Mental Disorders	290	Delirium dementia, amnestic, other cognitive disorders	3.95 [1.81, 8.64]	5.5 (1.5)	3.59 [1.59, 8.09]	•
Mental Disorders	291.8	Alteration of consciousness	3.74 [1.12, 12.55]	2.3 (0.6)	5.01 [0.95, 26.28]	•
+Mental Disorders	295.3	Psychosis	11.78 [2.61, 53.12]	3.5 (0.3)	6.05 [1.61, 22.72]	•
Mental Disorders	296.2	Depression	2.30 [1.40, 3.77]	11.0 (5.0)	4.39 [1.85, 10.39]	•
Neurological	331	Other cerebral degenerations	36.75 [4.88, 277.04]	5.2 (0.2)	15.87 [0.309, >499.99]	21.66 [2.7, 173.86]
Neurological	350.2	Abnormality of gait	2.76 [1.62, 4.69]	10.4 (3.9)	2.52 [1.30, 4.92]	3.26 [1.32, 8.09]
Respiratory	512.9	Other dyspnea	2.30 [1.20, 4.41]	6.1 (2.7)	1.20 [0.48, 3.02]	6.02 [2.05, 17.65]
Respiratory	513.3	Hypoventilation	3.88 [1.16, 13.03]	2.3 (0.6)	4.44 [1.09, 18.09]	•
Sense Organs	386	Vertiginous syndromes, disorders of vestibular system	4.26 [1.46, 12.43]	3.2 (0.8)	3.35 [0.96, 11.62]	•
Sense Organs	389	Hearing loss	1.76 [1.33, 2.34]	38.9 (26.7)	1.41 [0.99, 2.01]	2.57 [1.6, 4.13]
Symptoms	781	Symptoms involving nervous and musculoskeletal systems	2.25 [1.40, 3.62]	11.5 (5.5)	1.65 [0.93, 2.96]	4.29 [1.76, 10.42]
Symptoms	788	Syncope and collapse	1.80 [1.16, 2.79]	12.4 (7.3)	1.38 [0.81, 2.35]	3.28 [1.44, 7.49]

Other commonly associated health risk factors for AD, such as essential hypertension (prevalence=47% AD, 44% CN) and diabetes mellitus (prevalence=12% AD, 15% CN), did not show significant differences in prevalence in this sample.

These results show that health conditions within several categories are associated with AD in the years leading up to dementia diagnosis. The results also show that men and women have different patterns of associations with these health conditions in the years preceding dementia diagnosis.

Timed Analyses in AD

Whereas the overall analysis illustrates health codes associated with AD over a span of 15 years, the timed analyses focus on those health diagnoses that were present at least 5 years prior to dementia diagnosis (AD n=291, Control n=444; Table 3). The earliest codes indicating a higher prevalence in the AD group at 5 years prior to diagnosis included circulatory, dermatologic, digestive, genitourinary, hematopoietic, injury & poisoning, mental disorders, musculoskeletal, neurological, respiratory, and sense organs categories. Codes with higher prevalence (>5%) in the AD group included first degree AV block, right bundle branch block, cardiac dysrhythmias, diverticulosis/diverticulitis, erectile dysfunction, iron deficiency and other anemias, allergy/adverse effect of penicillin, depression, spondylosis, abnormal gait, influenza, hearing loss, tinnitus, and nervous and musculoskeletal symptoms. Odds ratios and the percentage of AD and CN participants with these codes are listed in Supplementary Table S1.

Table 3. Health Conditions Associated with AD at Different Time Intervals.

Categories and codes with a significantly higher prevalence in the AD group relative to controls at 5 years prior, 1 year prior, and year of diagnosis. The Odds Ratio [95% Confidence Interval (CI)] for each significant code is highlighted.

CATEGORY	CODE	CONDITION	5 Years Odds Ratio [95% CI]	1 Year Odds Ratio [95% CI]	Diagnosis Odds Ratio [95% CI]
Circulatory	416	Cardiomegaly		2.13 [1.27, 3.57]	1.89 [1.18, 3.03]
Circulatory	426.21	First degree AV block	2.00 [1.13, 3.53]
Circulatory	426.31	Right bundle branch block	2.12 [1.20, 3.74]
Circulatory	426.7	Abnormal electrocardiogram [ECG] [EKG]			1.52 [1.11, 2.08]
Circulatory	427.3	Other specified cardiac dysrhythmias	1.66 [1.15, 2.39]	1.54 [1.11, 2.15]
Circulatory	427.8	Sinoatrial node dysfunction (Bradycardia)		6.04 [1.24, 29.39]
Dermatologic	681.5	Cellulitis and abscess of leg, except foot		4.96 [1.29, 19.14]	3.81 [1.12, 12.97]
Dermatologic	686	Other local infections of skin and subcutaneous tissue	12.27 [1.44, 104.84]	11.69 [1.37, 99.51]	5.93 [1.17, 30.14]
Dermatologic	686.3	Pilonidal cyst	4.81 [1.21, 19.07]	4.67 [1.18, 18.43]	4.75 [1.20, 18.75]
Dermatologic	695.3	Rosacea	3.03 [1.16, 7.88]
Digestive	530.5	Disorders of esophageal motility		5.47 [1.41, 21.22]	5.40 [1.40, 20.78]
Digestive	562	Diverticulosis and diverticulitis	2.17 [1.25, 3.78]
Endocrine/Metabolic	240	Simple and unspecified goiter			2.17 [1.12, 4.23]
Genitourinary	599.4	Urinary incontinence		1.67 [1.11, 2.52]
Genitourinary	605	Erectile dysfunction [ED]	2.27 [1.16, 4.44]	2.30 [1.26, 4.21]	2.41 [1.34, 4.34]
Genitourinary	614.52	Vaginitis and vulvovaginitis		9.84 [1.16, 83.33]	10.94 [.129, 92.87]
Genitourinary	624.9	Stress incontinence, female			1.82 [1.26, 2.98]
Hematopoietic	280.1	Iron deficiency anemias, unspecified, not blood loss	4.61 [1.76, 12.08]	2.50 [1.15, 5.45]
Hematopoietic	285	Other anemias	1.83 [1.13, 2.97]
Injury & Poisoning	800.1	Fracture of neck of femur			3.06 [1.16, 8.42]
Injury & Poisoning	960.2	Allergy/adverse effect of penicillin	1.91 [1.18, 3.11]	1.84 [1.19, 2.86]	1.73 [1.14, 2.64]
Injury & Poisoning	971	Poisoning by drugs affecting autonomic nervous system			5.98 [1.12, 31.99]
Injury & Poisoning	979	Adverse drug events and drug allergies			1.64 [1.17, 2.30]
Mental Disorders	292.1	Aphasia/speech disturbance	5.36 [1.41, 20.38]	4.85 [1.51, 15.57]	4.00 [1.36, 11.78]
Mental Disorders	296.2	Depression	2.43 [1.24, 4.79]	2.34 [1.29, 4.25]	2.19 [1.25, 3.86]
Musculoskeletal	721	Spondylosis and allied disorders	1.89 [1.21, 2.94]	1.95 [1.32, 2.90]	1.75 [1.20, 2.56]
Musculoskeletal	721.1	Spondylosis without myelopathy		2.42 [1.28, 4.58]
Musculoskeletal	737.3	Kyphoscoliosis and scoliosis	2.71 [1.25, 5.89]
Neoplasms	172.2	Other non-epithelial cancer of skin		1.75 [1.26, 2.44]	1.78 [1.29, 2.45]
Neurological	350.2	Abnormality of gait	3.45 [1.31, 9.13]	2.43 [1.27, 4.66]	2.17 [1.18, 3,96]
Respiratory	480	Pneumonia		1.59 [1.11, 2.28]
Respiratory	481	Influenza	1.82 [1.26, 2.62]	1.57 [1.14, 2.19]
Respiratory	513.3	Hypoventilation		5.04 [1.31, 19.50]
Sense Organs	368.4	Visual field defects	2.98 [1.15, 7.73]
Sense Organs	379.2	Disorders of vitreous body	7.10 [1.49, 33.91]
Sense Organs	389	Hearing loss	1.98 [1.40, 2.81]	1.99 [1.46, 2.71]	1.98 [1.46, 2.69]
Sense Organs	389.4	Tinnitus	1.76 [1.11, 2.77]
Symptoms	781	Symptoms involving nervous/ musculoskeletal systems	3.45 [1.72, 6.82]	2.36 [1.34, 4.13]	2.19 ]1.30, 3.70]

Significant categories observed at 5 years prior were also seen at 1 year prior and at the date of diagnosis (AD n=325, Control n=538). New codes that reached significance at 1 year prior included cardiomegaly, urinary incontinence, non-epithelial skin cancer, and pneumonia (Supplementary Table S2). Codes that reached significance at the year of diagnosis included abnormal electrocardiogram, simple and unspecified goiter, and adverse drug events/allergies (AD n=327, Control n=579; Supplementary Table S3).

For those codes with higher prevalence in the AD group, the earliest and most consistent associations with AD across the three timepoints were depression, erectile dysfunction, gait abnormality, hearing loss, nervous/musculoskeletal symptoms, spondylosis, and allergy/adverse effect of penicillin. These codes were significant at all three timepoints.

Sex differences in patterns of health conditions associated with AD were observed (Supplementary Tables S1–S3), showing increased odds for erectile dysfunction, other anemias, and depression codes in males. Increased odds for cardiomegaly, goiter, iron deficiency anemias, abnormal gait, hearing loss, and tinnitus were seen in females. Spondylosis and nervous and musculoskeletal symptoms associated with AD were seen in both males and females.

Vascular Dementia

Overall differences in VaD

The overall difference between VaD (n=76) and controls (n=152) across all timepoints (mean interval VaD=13.42 (9.17 SD) years; CN=20.46 (11.30 SD) years), showed significant differences for 16 codes (Figure 2), indicating a higher prevalence in the VaD group. The codes included circulatory (abnormal electrocardiogram, atrial fibrillation, cardiac dysrhythmias, occlusion of cerebral arteries, acute but ill-defined cerebrovascular disease, late effects of cerebrovascular disease), genitourinary (erectile dysfunction, other disorders of male genital organs), injury & poisoning (fracture of foot), musculoskeletal (spondylosis without myelopathy), neurological (Parkinson’s disease, hemiplegia, abnormality of gait, abnormal reflex), sense organs (hearing loss), and respiratory (hypoventilation) categories.

Figure 2. Overall Health Conditions Associated with VaD.

Codes with a significantly higher prevalence in the VaD group relative to controls. These overall results are across a span of 13 years. The log of the odds ratio (dot) and 95% confidence intervals (bars) are shown.

Significant conditions associated with AD were observed in the circulatory category, genitourinary (erectile dysfunction), neurological (abnormality of gait), and sense organs (hearing loss) in males. Associations for injury & poisoning (fracture of foot) were seen in females. Associations with circulatory (atrial fibrillation), and musculoskeletal (spondylosis without myelopathy) codes. (Table 4) were observed in both males and females.

Table 4. Overall Health Conditions Associated with VaD.

Categories and codes with a significantly higher prevalence in the VaD group relative to controls across 13 years prior to diagnosis. The Odds Ratio [95% Confidence Interval (CI)] for each code is shown for the whole VaD group. Highlighting in the ‘males’ and ‘females’ columns indicate significant sex effects; missing odds ratios represent codes with <6 positive participants needed for model stability. The percentage of VaD and control (CN) participants with each code is also shown.

CATEGORY	CODE	CONDITION	ODDS RATIO All [95% CI]	%AD (%CN)	ODDS RATIO Males	ODDS RATIO Females
Circulatory	426.7	Abnormal electrocardiogram [ECG] [EKG]	2.42 [1.24, 4.72]	40.7 (32.2)	4.28 [1.82, 10.07]	0.42 [0.12, 1.51]
Circulatory	427.21	Atrial fibrillation	3.44 [1.66, 7.14]	30.2 (11.8)	2.70 [1.19, 6.12]	8.73 [1.64, 46.57]
Circulatory	427.3	Other specified cardiac dysrhythmias	2.48 [1.30, 4.73]	36.8 (21.7)	2.92 [1.34, 6.34]	1.32 [0.40, 4.43]
Circulatory	433.2	Occlusion of cerebral arteries	3.57 [1.36, 9.37]	17.1 (5.2)	3.20 [1.12, 9.12]	•
Circulatory	433.6	Acute, but ill-defined cerebrovascular disease	5.74 [2.18, 15.09]	21.0 (4.6)	5.62 [1.79, 17.60]	6.62 [1.01, 43.39]
Circulatory	433.8	Late effects of cerebrovascular disease	12.51 [2.49, 62.82]	11.8 (1.3)	20.48 [2.44, 171.97]	•
Genitourinary	605	Erectile dysfunction [ED]	7.88 [2.32, 26.81]	13.1 (3.2)	6.03 [1.83, 19.85]	•
Genitourinary	608	Other disorders of male genital organs	4.59 [1.21, 17.37]	7.8 (3.2)	3.42 [0.93, 12.51]	•
Injury & Poisoning	801.1	Fracture of foot	9.05 [2.19, 37.40]	10.5 (1.9)	•	12.48 [1.61, 96.76]
Musculoskeletal	721.1	Spondylosis without myelopathy	6.08 [1.94, 19.02]	13.1 (3.9)	4.42 [1.11, 17.60]	11.89 [1.19, 118.73]
Neurological	332	Parkinson’s disease	4.43 [1.34, 14.73]	10.5 (3.2)	3.91 [1.01, 15.14]	•
Neurological	342	Hemiplegia	6.12 [1.56, 24.09]	11.8 (1.9)	7.56 [1.49, 38.40]	•
Neurological	350.2	Abnormality of gait	5.39 [1.64, 17.70]	10.5 (3.9)	6.45 [1.41, 29.51]	•
Neurological	350.5	Abnormal reflex	19.10 [2.13, 171.61]	7.8 (0.6)	•	•
Sense Organs	389	Hearing loss	2.88 [1.57, 5.30]	48.6 (27.6)	3.24 [1.55, 6.75]	1.64 [0.53, 5.02]
Respiratory	513.3	Hypoventilation	14.54 [1.58, 133.61]	7.8 (0.6)	•	•

These results show that VaD is associated with health conditions predominately within circulatory and neurological categories in the years leading up to dementia diagnosis. The results also show that many codes were significantly associated with VaD both men and women, yet men show associations with a majority of codes observed in the years preceding dementia diagnosis.

Timed Analyses in VaD

At 5 years prior to diagnosis (VaD n=66, Control n=111), the VaD group showed significant differences in circulatory, genitourinary, injuries & poisonings, mental disorders, musculoskeletal, neoplasms, neurological, and sense organs categories (Table 5). Significant codes included first degree AV block, abnormal electrocardiogram, cardiac dysrhythmias, supraventricular premature beats, acute but ill-defined cerebrovascular disease, skull/face fracture and other intercranial injury, depression, spondylosis, non-epithelial cancer of skin, peripheral nerve disorders, dizziness/ giddiness, and hearing loss. Odds ratios and the percentage of participants with these codes are listed in Supplementary Table S4.

Table 5. Health Conditions Associated with VaD at Different Time Intervals.

Categories and codes with a significantly higher prevalence in the VaD group relative to controls at 5 years prior, 1 year prior, and year of diagnosis. The Odds Ratio [95% Confidence Interval (CI)] for each significant code is highlighted.

CATEGORY	CODE	CONDITION	5 Years Odds Ratio [95% CI]	1 Year Odds Ratio [95% CI]	Diagnosis Odds Ratio [95% CI]
Circulatory	426.21	First degree AV block	4.00 [1.31, 12.22]
Circulatory	426.7	Abnormal electrocardiogram [ECG] [EKG]	2.46 [1.23, 4.99]	2.25 [1.18, 4.30]	2.11 [1.11, 4.02]
Circulatory	427.21	Atrial fibrillation		2.39 [1.11, 5.12]	2.45 [1.20, 5.01]
Circulatory	427.3	Other specified cardiac dysrhythmias	2.95 [1.39, 6.25]	2.40 [1.24, 4.64]	2.44 [1.26, 4.69]
Circulatory	427.61	Supraventricular premature beats	4.70 [1.34, 16.44]	5.45 [1.82, 16.33]	4.98 [1.80, 13.77]
Circulatory	429.3	Symptoms involving cardiovascular system	4.26 [1.30, 14.01]
Circulatory	433.2	Occlusion of cerebral arteries		5.62 [1.68, 18.86]	4.05 [1.41, 11.59]
Circulatory	433.6	Acute, but ill-defined cerebrovascular disease	10.01 [2.09, 48.02]	7.73 [2.38, 25.08]	5.62 [2.02, 15.65]
Genitourinary	605	Erectile dysfunction [ED]			4.73 [1.49, 14.98]
Genitourinary	608	Other disorders of male genital organs	9.11 [1.58, 52.43]
Genitourinary	611.3	Lump or mass in breast		4.47 [1.18, 16.95]
Genitourinary	627.3	Postmenopausal atrophic vaginitis	13.29 [1.45, 122.06]	5.52 [1.27, 23.92]	6.85 [1.62, 28.97]
Hematopoietic	695	Erythematous conditions			9.80 [1.12, 86.11]
Injury & Poisoning	819	Skull/face fracture, other intercranial injury	6.74 [1.27, 35.80]	5.83 [1.48, 22.90]	4.31 [1.24, 15.00]
Mental Disorders	296.2	Depression	6.42 [1.20, 34.47]	8.90 [1.81, 43.85]	10.79 [2.22, 52.35]
Musculoskeletal	716.9	Arthropathy NOS		4.06 [1.17, 14.01]
Musculoskeletal	721.1	Spondylosis without myelopathy	5.57 [1.34, 23.27]	5.19 [1.52, 17.77]
Neoplasms	172.2	Other non-epithelial cancer of skin	2.93 [1.35, 6.36]	2.53 [1.29, 4.96]	2.16 [1.12, 4.18]
Neurological	332	Parkinson’s disease			14.20 [1.66, 121.19]
Neurological	333.1	Essential tremor		5.33 [1.90, 15.01]	3.31 [1.36, 8.08]
Neurological	342	Hemiplegia			7.32 [1.45, 36.82]
Neurological	350.5	Abnormal reflex		6.58 [1.27, 34.08]	6.70 [1.30, 34.55]
Neurological	351	Other peripheral nerve disorders	6.47 [1.45, 28.85]	6.75 [1.66, 27.49]	4.76 [1.48, 15.33]
Sense Organs	386.9	Dizziness, giddiness (Light-headedness/vertigo)	9.78 [2.14, 44.56]	7.69 [2.14, 27.65]	6.69 [2.08, 21.46]
Sense Organs	389	Hearing loss	3.32 [1.63, 6.73]	3.52 [1.87, 6.61]	3.64 [1.93, 6.88]

New codes that reached significance at 1 year prior to dementia diagnosis included atrial fibrillation, occlusion of cerebral arteries, essential tremor, and abnormal reflex (VaD n=74, Control n=139; Supplementary Table S5). Codes that reached significance at the year of diagnosis (VaD n=75, Control n=147) included erectile dysfunction, hemiplegia, and Parkinson’s disease (Supplementary Table S6).

For those codes with higher prevalence (≥5% of participants), the earliest and most consistent associations with VaD across the three timepoints were abnormal electrocardiogram, cardiac dysrhythmias, supraventricular premature beats, acute but ill-defined cerebrovascular disease, skull/face fracture and other intercranial injury, depression, other non-epithelial cancer of skin, peripheral nerve disorders, dizziness/giddiness, and hearing loss.

Sex differences in patterns of health conditions associated with VaD were observed (Supplementary Tables S4–S6), with significant associations for circulatory (abnormal electrocardiogram, atrial fibrillation, cardiac dysrhythmias, occlusion of cerebral arteries, acute but ill-defined cerebrovascular disease), neoplasms (other non-epithelial cancer of skin), and neurological (essential tremor, hemiplegia, abnormal reflex) codes were observed in males. Associations with cardiac dysrhythmias and depression codes were observed in females. Associations with dizziness/giddiness and hearing loss codes with VaD were observed in both males and females.

Temporal Trends in AD and VaD

There were different patterns in the number and type of category codes over time in AD and VaD as illustrated in Figure 3. In AD, circulatory, dermatologic, genitourinary, mental disorder, and sense organ categories exhibited the highest number of codes at the three timepoints. In VaD, the highest number of codes were seen in circulatory and neurological categories.

Figure 3. Significant Category Codes Over Time.

These stacked bar plots illustrate the number of significant codes in each health category across the three timepoints of the timed analyses. Significant codes for AD are shown in the top graph, and VaD codes are shown in the bottom graph. The darkest color represents significant codes at 5 years prior, the medium color at 1 year prior, and the lightest color at the date of diagnosis.

DISCUSSION

The results of this PheDAS study show that specific health conditions are significantly associated with the onset of dementia in older individuals. Significant health conditions associated with both AD and VaD are observed across all years leading up to dementia diagnosis, and in a time-dependent manner. These time-dependent patterns show early and consistent associations with dementia, as well as associations that occur closer to the year of diagnosis. The results illustrate that time-dependent associations of health with AD involve several categories including circulatory, dermatologic, genitourinary, mental disorder, and sense organ disorders suggesting that AD is associated with global health status. In contrast, VaD associations predominately involve circulatory and neurological disorders. Our findings suggest that most health relationships are unique to the type of dementia diagnosis, while some are shared between AD and VaD.

There are several data-driven methods designed to examine large datasets ^23–25. For example, machine learning ²⁶, convolutional network ²⁷, and transformer approaches ²⁸ have been used to study encoding spaces for electronic health records. The ability to interpret the findings produced by these methods is a fundamental problem for data-driven approaches, however. Here, we explore association of known disease phenotypes associated with AD and VaD through a massively univariate regression method known as PheDAS ²⁹. Our approach presents compact signatures that ease interpretation and hold promise for future for hypothesis driven methods and other techniques such as novel AI-learning network methods.

In AD, health conditions associated with dementia involving several categories were observed over a 15-year interval before dementia diagnosis. These categories included genitourinary, mental disorders, neurological, respiratory, and sense organs, with hearing loss (39% of AD participants), urinary incontinence (23% AD) and depression (11% AD) among the most common disorders. Many of these results support previous studies of health conditions associated with the diagnosis of AD ^{7, 8, 30}. Depression ^{8, 9, 11, 12} and urinary incontinence ⁸ and have been associated with AD in comorbidity studies, and hearing loss ^{31, 32} and gait abnormalities ³³ are known to be a significant risk factors for dementia. We also found associations between AD and less commonly reported conditions such as erectile dysfunction, syncope, and dyspnea which add to our understanding of health and dementia. We did not, however, observe other common risk factors for dementia such as hypertension and diabetes in this sample. The literature examining comorbidities in relation to AD is mixed with regard to the contributions of these conditions, with about half of studies finding associations with hypertension ^{5–7, 11, 34} and about a third showing associations with diabetes ^{5, 6, 8}. Here, the lack of greater prevalence of vascular conditions in AD may be because only a small percentage of the AD sample (9%) had a clinical diagnosis of mixed dementia which included a combination of both AD and VaD. This interpretation is supported by studies that found a decreased risk for AD in relation to hypertension in patients without other cardiovascular risk factors ³⁴ and in those receiving antihypertensive treatment ¹². Although mid-life data were not available for all participants in our study, another important aspect to consider is midlife versus late-life onset of these health conditions, as studies have shown that vascular and metabolic disorders which begin in midlife have a greater impact on the risk for future dementia ^{10, 11}.

Whereas the overall analysis in AD illustrated health conditions that are associated with dementia diagnosis across a broad time interval, temporal trends in health conditions were also observed when examining the data in a timed fashion. When examining associations at 5 and 1 year(s) prior to diagnosis and at the year of diagnosis, we found that the earliest and most consistent associations with AD across the three timepoints included depression, erectile dysfunction, gait abnormality, hearing loss, nervous/musculoskeletal symptoms, and spondylosis and allied disorders. Codes that became significant closer to diagnosis included cardiomegaly, and other non-epithelial cancer of skin at 1 year prior to diagnosis, and incontinence and adverse drug events/allergies at the year of diagnosis. These results suggest that there is a temporal component related to the health conditions associated with AD. Importantly, these temporal trends may play a role in variations in health comorbidities reported in the literature ^{5, 8, 9, 30}, due to the assessment of different time intervals across studies and the time-dependent nature of comorbidities illustrated here.

In VaD, health conditions associated with dementia were observed over a 13-year interval before dementia diagnosis. These conditions included circulatory, genitourinary, injury & poisoning, musculoskeletal, neurological, sense organs, and respiratory categories. The most common conditions included hearing loss (49% VaD), abnormal electrocardiogram (41% VaD), cardiac dysrhythmias (37% VaD), and atrial fibrillation (30% VaD). These results show that health conditions associated with with VaD occur primarily within circulatory and neurological categories, which is in line with previous studies and the etiology of the disease ^{12, 13}.

When examining associations at 5 and 1 year(s) prior to diagnosis and at the year of diagnosis, we found that the earliest and most consistent associations with VaD across the three timepoints included abnormal electrocardiogram, cardiac dysrhythmias, supraventricular premature beats, acute but ill-defined cerebrovascular disease, depression, other non-epithelial cancer of skin, light-headedness and vertigo, and hearing loss. Codes that became significant closer to diagnosis in the VaD group included occlusion of cerebral arteries, essential tremor, and abnormal reflex at 1 year prior to diagnosis. At the year of diagnosis, significance was seen for erectile dysfunction, erythematous conditions, Parkinson’s disease likely of vascular etiology, and hemiplegia. Unlike AD which had a high number of significant codes for circulatory, dermatologic, genitourinary, mental disorders and sense organs categories, the majority of associations in VaD involve the circulatory category including both cardiovascular and cerebrovascular codes, and neurological health conditions. Cardiovascular conditions and acute cerebrovascular disease appear to emerge earlier in vascular dementia, with neurologic conditions emerging closer to the date of diagnosis. These findings highlight differences between the two dementia types and add to our knowledge of the timing of health comorbidities in relation to the progression of vascular-related dementia.

Several health codes were also significant in both the AD and VaD groups. Across all timepoints spanning 13–15 years, both groups had significant codes for erectile dysfunction, abnormality of gait, and hypoventilation. In the timed analyses beginning 5 years prior to diagnosis, both groups also showed significant codes for first degree AV block, abnormal electrocardiogram, cardiac dysrhythmias, depression, other non-epithelial cancer of the skin, and hearing loss. These associations also involve several health categories, including genitourinary, sense organs, mental disorders and circulatory codes, suggesting that widespread physical dysfunction may be associated with dementia in older individuals. This implies that physiological dysregulation or change in homeostasis may play a role in the eventual cognitive decline of these individuals, as a loss of homeostasis can lead to general physical decline, frailty, and accelerated aging which is associated with dementia ^35–37.

Few studies have examined sex differences in health conditions associated with dementia ^{8, 9}. Here, sex differences in patterns of associations with both AD and VaD were observed. In males, dermatologic, digestive, genitourinary, and mental disorders codes were associated with AD, whereas in females circulatory, endocrine/metabolic, hematopoietic, respiratory, and sense organ (hearing loss) codes were associated with AD. In both males and females, urinary incontinence and gait abnormality codes were associated with AD. In VaD, circulatory, genitourinary, and neurological codes were associated with dementia in males, mental disorders and injury & poisoning codes were associated with dementia in females. Atrial fibrillation, cardiac dysrhythmias, spondylosis without myelopathy, and hearing loss were associated with VaD in both males and females. These results illustrate that there are sex differences in health conditions associated with AD which support previous research ^{8, 9}, and add to our knowledge of the relationships in VaD. The findings also illustrate that the patterns of health relationships vary across dementia types. For example, in females, circulatory codes show significant association with AD, whereas in both males and females, circulatory codes have a significant association with VaD. This suggests that the health code associations with dementia could vary by sex, and the pattern of sex-related associations depends on the type of dementia diagnosis.

This study has both advantages and limitations. One advantage is that our participants received a clinical diagnosis of dementia based on consensus case conferences, instead of relying on a medical record diagnosis often used in other studies of comorbidity. Our participants also receive thorough medical exams at each BLSA visit, the results of which are included in their medical records. Limitations include data availability for a relatively small number of participants with dementia. This may be due to the characteristics of the BLSA. Our participants are generally highly educated and receive regular medical exams, so they are aware of health comorbidities and seek treatment when warranted. Both characteristics can reduce the risk of dementia. Those who are developing cognitive issues may also seek more medical attention than those who are not, resulting in increased reporting of less common dementia-related health conditions, such as the dermatologic conditions observed in the AD group. Further, it should also be noted that our definition of AD is based on a clinical diagnosis, as opposed to a biologically based diagnosis which relies on ante- or postmortem assessment of the presence of neuropathology ³⁸. A clinical diagnosis of dementia generally occurs after the onset of symptoms, so health conditions associated with dementia observed closer to the date of diagnosis can also occur concurrently or subsequent to the onset of the dementia process. The BLSA also uses the ICD-9 coding system, whereas current coding systems rely on ICD-10. While the ICD-9 system provides fewer details related to a medical condition (e.g. ICD-9 Fracture of the forearm versus ICD-10 Fracture of lower end of radius), primary medical conditions are still included in the ICD-9 system. Finally, disease severity likely plays an important role in the impact of comorbidities on health and well-being, yet information on the severity of disease was unavailable.

Health conditions associated with dementia onset are of great importance, as they often represent modifiable factors that offer an avenue for intervention. Our results show that health conditions associated with AD and VaD include both common and distinctive health disorders. We further show that there are time-dependent patterns in the significance of these relationships, with some health conditions associated with dementia appearing early and consistently over time and others becoming significant closer to the time of diagnosis. These time-dependent findings are important not only from a clinical standpoint, but may help explain the variability of results from studies examining different intervals prior to dementia diagnosis ^{5, 30}. We also find that males and females show different patterns in the health conditions associated with dementia. Together, these novel findings advance our knowledge of the relationship between health and dementia, and reinforce the need for medical intervention and treatment to lessen the impact of health comorbidities, which could potentially reduce the risk of future dementia in our aging population.

Data Availability:

Data from the BLSA are available on request by proposal submission through the BLSA website (blsa.nih.gov). All requests are reviewed by the BLSA Data Sharing Proposal Review Committee and are also subject to approval from the NIH institutional review board.