Association between enrollment factors and incident cognitive impairment in Blacks and Whites: Data from the Alzheimer’s Disease Center

1. STRUCTURED ABSTRACT

INTRODUCTION:

We examined the influence of enrollment factors demonstrated to differ by race on incident mild cognitive impairment (MCI) and dementia using Alzheimer’s Disease Center (ADC) data.

METHODS:

Differences in rates of incident impairment between non-Latino Whites and Blacks (n=12,242) were examined with age-at-progression survival models. Models included race, sex, education, source of recruitment, health factors, and family history of dementia.

RESULTS:

No significant race differences in progression were observed in cognitively unimpaired participants. In those with MCI at baseline, Whites evidenced greater risk for progression than Blacks. Enrollment factors e.g., referral source, were significantly related to progression.

DISCUSSION:

The finding that Blacks demonstrated lower rate of progression than Whites is contrary to extant literature. Nested-regression analyses suggested that selection-related factors, differing by race, may account for these findings and influence our ability to accurately estimate risk for progression. It is potentially problematic to make racial comparisons using ADC datasets.

2. BACKGROUND

Recruitment for Alzheimer’s disease and related dementias (ADRD) research is challenging, especially for communities in which research participation connotes mistreatment and abuse from scientists[1, 2]. Research teams recruiting for ADRD research are encouraged to devise creative recruitment strategies to enlist adequate numbers of research volunteers, particularly when recruiting individuals from under-represented groups (URG) [3].

Successful programs employ various enrollment approaches including clinic and community-based approaches[4], both of which represent departures from random sampling and introduce potential threats to external and internal validity. Moreover, non-random sampling methods introduce systematic-biases associated with both independent variables (e.g., race) and outcomes (e.g., incident cognitive impairment)[5].

While selection bias and the resultant challenges with generalizability are often acknowledged as limitations in cognitive aging studies, the extent to which enrollment strategies contribute to selection bias and influence ADRD datasets is relatively unexamined. A recent exception highlighted the influence of sampling strategy on ADRD-related risk estimation. Chen et al[6]. found that rates of progression from cognitive health to mild cognitive impairment (MCI) differed based on recruitment setting with individuals recruited from clinic demonstrating an annual rate of conversion to MCI six times that of individuals recruited from the community.

We explored the effects of enrollment factors on incident MCI and dementia in a sample of Blacks and Whites from a large ADRD database. We hypothesized that referral source and a factor likely to influence enrollment, known family history of ADRD, would associate with primary clinical outcomes including rates of incident MCI and dementia and with racial identity in a sample of non-Latino Blacks and Whites. To investigate this hypothesis, we first depicted rates of diagnostic progression in Blacks and Whites, with separate survival analyses based on baseline cognitive status (cognitively unimpaired or MCI). We then used nested models to assess stability of race-associated risk for diagnostic progression associated with enrollment factors, e.g., referral source and family history of dementia.

3. METHODS

3.1. Participants

Data were collected between January 2005 and February 2017 at US-based Alzheimer’s Disease Centers (ADC) and complied by the National Alzheimer’s Coordinating Center (NACC). All participants provided informed consent. The NACC dataset (dated 5/10/17) used here included 33,444 subjects from 37 Centers with visits occurring on or before 2/24/2017.

Exclusion criteria included 1) Down syndrome, known autosomal dominant mutations for Alzheimer’s disease (AD) and frontotemporal lobar dementias (FTLD), 2) self-identified race other than solely Black or White, or Hispanic/Latino ethnicity, 3) baseline age <60 years. Subsequently, we excluded participants 4) with baseline cognitive diagnosis other than cognitively unimpaired or MCI, 5) with baseline MCI due to FTLD, given the relatively young age of onset for this dementia, and 6) those without follow-up. The resulting analytical sample included 8,182 subjects, cognitively unimpaired at baseline (15.7% Black), and 4,060 subjects with MCI at baseline (15.2% Black). Figure 1 depicts analytic sample derivation.

Figure 1.

CONSORT Diagram

3.2. Primary Outcomes

We examined the influence of (1) race, cardiovascular and MCI etiology and (2) enrollment factors on age-to-incident progression. Incident progression in cognitive impairment was defined as the first visit with a diagnosis of MCI or dementia, if the participant was cognitively unimpaired at baseline. For those with MCI at baseline, incident progression was denoted at first diagnosis of dementia. ADC clinicians determined cognitive status based on clinical examination data obtained at a research visit. Age of progression represents age when progression was ascertained.

3.3. Additional Variables

We investigated the relationship between enrollment factors and incident cognitive impairment using a nested modeling strategy. After examining the contributions of underlying differences in risk including cardiovascular conditions and etiology of impairment, enrollment factors were added to models.

3.3.1. Examination of cardiovascular factors and etiology of cognitive impairment:

Model 1 incorporated a number of cardiovascular conditions including diabetes, hypertension, history of myocardial infarction (MI), atrial fibrillation (AFib), and congestive heart failure (CHF), ascertained by clinical interview and physical examination. All conditions were categorized as (absent, recent/active, remote/inactive, or unknown) based on clinical interview. In the NACC database, recent/active events were conditions occurring in the year prior to study visit or requiring continued management. Conditions no longer needing management or events occurring >1 year ago were designated as remote/inactive. To simplify models, three cardiovascular events or conditions, MI, AFib, and CHF were combined into one covariate, categorized as: “recent/active” if any of the three conditions were “recent/active”, “absent” if all three were “absent” or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

For participants diagnosed with baseline MCI, we included suspected etiology of MCI at baseline in our regression models. Etiologic determinations were made by ADC clinician(s) using research visit data. Primary etiology was determined at the time of MCI diagnosis and categorized as either AD, vascular disease (VaD), Lewy Body dementia (DLB), other cause (Other), and unassessed/unknown (Unknown).

3.3.2. Examination of enrollment processes:

Factors judged to influence the participant’s decision to enroll in an ADC’s Clinical Core included (1) referral source and (2) family history of dementia. Data from individuals referred by a research study clinician or investigator, a nurse, doctor or other health-care provider were denoted as referral by a “Health Professional”. The health professional referral source here is presumed to reflect recruitment from a clinical or clinical research setting. In the case of ADCs, this is most likely neurology, psychiatry or geriatrics clinics. Family history, captured as one or more first degree relative with dementia, no first degree relative, or unknown, is included as an enrollment factor. Although family history is most often interpreted as a biological risk factor, we propose that self-reported family history represents knowledge of a family history – information not equally accessible across racial groups. The awareness of family history encompasses access to diagnostic services and awareness of dementia symptoms as a diagnosable condition as opposed to a phenomenon of aging. Dementia is frequently undiagnosed[7–9], especially for non-White populations[10]. Moreover, referral patterns differ between non-Hispanic Blacks and Whites, such that White patients diagnosed with dementia are more frequently evaluated in Neurology clinics than are Black patients[11].

This interpretation of known family history considers motivations for participation in intensive research protocols as a potential source of selection bias. As such, known family history may not only be a strong motivating factor for participation in intensive cognitive aging protocols[12], but in Black populations may correlate with protective factors such as socioeconomic advantage and health literacy.

3.4. Analyses

Characteristics of Black and White participants included in the analytic sample were compared using Mann-Whitney U tests and Chi-squared tests. Adjusted age-to-progression within Black and White subjects was compared using Cox proportional hazard models. All survival models used a variable entry setup, wherein baseline age was entry age paired with participants’ progression or censoring age. Participants diagnosed with incident MCI or dementia did not re-enter the at-risk pool if the diagnosis was later reversed. Differences in unadjusted age-to-progression between Blacks and Whites were compared visually using Kaplan-Meier curves.

Separate nested regression models were fitted for two subgroups: (1) participants cognitively unimpaired at baseline, and (2) participants with baseline MCI. For each subgroup, the base model included covariates of sex, education category (<high school, high school, >high school, bachelors, masters, doctorate, or unknown), race (black or white), diabetes, hypertension and cardiac events/conditions, and baseline etiology for MCI subgroup (Model 1). Subsequent models added covariates – referral source and family history of dementia (Model 2) allowing us to assess robustness of the hazard ratios for racial groups. Assumption checks yielded no concerns for non-proportionality, outliers, and goodness-of-fit. All analyses and graphics were performed in R version 3.4.0[13]. Survival analyses were performed using ‘survival’ package[14].

4. RESULTS

4.1. Group differences at baseline

Table 1 depicts the analytic sample characteristics. Racial comparisons revealed numerous baseline differences. Compared to Whites, Blacks were less likely to die during follow-up (e.g., cognitively unimpaired subsample: 9.3% vs 14.1%, p<0.001). Blacks had more cardiovascular and metabolic risk factors than Whites. Regarding enrollment factors, Blacks were less likely to have been referred by health professionals (e.g., baseline MCI subgroup: 25.0% vs 43.4%, p<0.001), and less likely than Whites to report 1^st degree family history of cognitive impairment (45.9% vs 57.4%, p<0.001).

Table 1.

Racial comparisons of analytic (regression) sample by baseline cognitive status. Regression sample included only non-Hispanic Whites and Blacks. Data were collected between June 2005 and February 2017.

	Cognitively Normal at Baseline				MCI at Baseline
	White	Black	p		White	Black	p
N	6894	1288			3444	616
Converted, N(%)	1559 (22.6)	267 (20.7)	0.146		1598 (46.4)	189 (30.7)	<0.001
Entry age in yrs, mean(SD)	74.33 (8.16)	72.66 (7.08)	<0.001		75.27 (7.75)	73.58 (7.59)	<0.001
Female Sex, N(%)	4272 (62.0)	1014 (78.7)	<0.001		1526 (44.3)	423 (68.7)	<0.001
Died, N(%)	971 (14.1)	120 (9.3)	<0.001		775(22.5)	67 (10.9)	<0.001
Age at censoring /conversion, mean(SD)	78.73 (8.37)	76.97 (7.68)	<0.001		78.30 (7.92)	76.85 (7.74)	<0.001
Education Category			<0.001				<0.001
Under High School, N(%)	116 (1.7)	124 (9.6)			114 (3.3)	92 (14.9)
High School, N(%)	896 (13.0)	282 (21.9)			565 (16.4)	172 (27.9)
Over High School, N(%)	1229 (17.8)	341 (26.5)			599 (17.4)	145 (23.5)
Bachelor degree, N(%)	1738 (25.2)	186 (14.4)			888 (25.8)	82 (13.3)
Masters degree, N(%)	2064 (29.9)	276 (21.4)			835 (24.2)	95 (15.4)
Doctorate degree, N(%)	822 (11.9)	76 (5.9)			437 (12.7)	30 (4.9)
Unknown, N(%)	29 (0.4)	3 (0.2)			6 (0.2)	0 (0.0)
Diabetes			<0.001				<0.001
Absent, N(%)	6307 (91.5)	956 (74.2)			3066 (89.0)	420 (68.2)
Recent/active, N(%)	548 (7.9)	316 (24.5)			346 (10.0)	184 (29.9)
Remote/inactive, N(%)	27 (0.4)	6 (0.5)			23 (0.7)	8 (1.3)
unknown, (N%)	12 (0.2)	10 (0.8)			9 (0.3)	4 (0.6)
Hypertension			<0.001				<0.001
Absent, N(%)	3623 (52.6)	316 (24.5)			1672 (48.5)	132 (21.4)
Recent/active, N(%)	3080 (44.7)	941 (73.1)			1658 (48.1)	460 (74.7)
Remote/inactive, N(%)	171 (2.5)	29 (2.3)			104 (3.0)	22 (3.6)
Unknown, (N%)	20 (0.3)	2 (0.2)			10 (0.3)	2 (0.3)
Cardiac event/Condition*			0.009				0.022
Absent, N(%)	5997 (87.0)	1143 (88.7)			2887 (83.8)	541 (87.8)
Recent/active, N(%)	514 (7.5)	66 (5.1)			295 (8.6)	34 (5.5)
Unknown, (N%)	383 (5.6)	79 (6.1)			262 (7.6)	41 (6.7)
Primary Etiology of MCI at Baseline		NA	NA				<0.001
Alzheimer’s disease, N(%)	NA	NA			1436 (41.7)	214 (34.7)
Vascular disease, N(%)	NA	NA			137 (4.0)	58 (9.4)
Lewy Body dementia, N(%)	NA	NA			176 (5.1)	3 (0.5)
Mixed or Other, N(%)	NA	NA			1695 (49.2)	341 (55.4)
Referral Source			<0.001				<0.001
Self/relative/friend, N(%)	2786 (40.4)	550 (42.7)			876 (25.4)	146 (23.7)
Health professional, N(%)	1004 (14.6)	105 ( 8.2)			1494 (43.4)	154 (25.0)
Other, N(%)†	2846 (41.3)	593 ( 46.0)			961 (27.9)	288 (46.8)
Unknown, N(%)	258 ( 3.7)	40 ( 3.1)			113 (3.3)	28 (4.5)
Family History of dementia			<0.001				<0.001
No 1st degree relative, N(%)	2548 (37.0)	544 (42.2)			1168 (33.9)	262 (42.5)
≥ One 1st degree relative, N(%)	3773 (54.7)	569 (44.2)			1976 (57.4)	283 (45.9)
unknown, (N%)	573 (8.3)	175 (13.6)			300 (8.7)	71 (11.5)
Years of follow-up, mean(SD)	4.92 (2.93)	4.77 (2.87)	0.158		3.88 (2.49)	3.79 (2.65)	0.08

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent”. A label of “unknown” selected if there was insufficient data to determine the status of the condition.

^*Cardiac event/condition includes history of myocardial infarction (MI), atrial fibrillation (AFib), and congestive heart failure (CHF). “Recent/active” selected if any of the three conditions were “recent/active”; categorization “absent” if all three were “absent” or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non-ADC media appeal.

For participants with MCI at baseline, there were racial differences in primary etiology. Whites were more often diagnosed with MCI due to AD pathology (41.7% vs 34.7%) and DLB (5.1% vs 0.5%), and less likely than Blacks to have Vascular disease denoted as primary MCI etiology (4.0% vs. 9.4%). In both subgroups (cognitively unimpaired and MCI), total follow-up time was similar between Blacks and Whites (cognitively unimpaired: 4.77 vs 4.92 years, p=0.158; MCI: 3.79 vs 3.88 years, p=0.080). (Table 1.)

Appendix A contains outcome and covariate comparisons for the full sample before applying final exclusion criteria, i.e., baseline dementia status, MCI etiology of FTLD, no follow-up. Appendix B provides comparisons of individuals in the final analytic sample to those satisfying all inclusion criteria, with the exception of having follow-up data.

4.2. Unadjusted racial differences in rate of incident cognitive impairment

Kaplan-Meier survival curves portray progression over time by race (Figure 2). In the cognitively unimpaired group (Table 1) progression proportions between Blacks and Whites did not differ (20.7% and 22.6%, p=0.146). In contrast, differences in progression by race were noted in the MCI subgroup, with more Whites progressing than Blacks (46.4% vs 30.7%, p<0.001).

Figure 2.

Survival analyses for incident cognitive impairment. Kaplan-Meier curves portrayed for baseline cognitive status subgroups.

Figure 2a. Kaplan-Meier Curves for non-Hispanic Whites and Blacks cognitively unimpaired at Baseline

Figure 2b. Kaplan-Meier Curves for non-Hispanic Whites and Blacks diagnosed with MCI at Baseline

4.3. Model 1: Cardiovascular factors and baseline MCI etiology

4.3.1. Demographic factors:

In participants who were cognitively unimpaired at baseline (Table 2, Model 1), women demonstrated lower risk than men for incident cognitive impairment as did individuals with advanced education. For those with MCI at baseline, education conferred a reduced risk for progression (Table 3, Model 1). Black race had no effect in those who were cognitive healthy at baseline and was associated with a reduced hazard for progression for those with MCI at baseline (HR=0.71; p<0.0001).

Table 2.

Regression models predicting incident MCI or Dementia in participants who were cognitively unimpaired at Baseline. Model 1 included demographic and cardiovascular risk factors; Model 2 adds recruitment factors.

	Model 1				Model 2
Coefficient	HR	HR 95% CI	p value		HR	HR 95% CI	p value
African American (reference: White)	0.99	0.86 — 1.14	0.8963		1.05	0.91 — 1.21	0.4864
Female (reference: Male)	0.79	0.72 — 0.87	<0.0001		0.79	0.72 — 0.87	<0.0001
Education: (reference: HS education)
under HS	1.21	0.94 — 1.56	0.1335		1.25	0.97 — 1.60	0.0870
over HS	0.93	0.80 — 1.08	0.3471		0.93	0.80 — 1.09	0.3843
Bachelor	0.92	0.79 — 1.06	0.2461		0.91	0.79 — 1.06	0.2265
Masters	0.83	0.72 — 0.96	0.0146		0.83	0.72 — 0.97	0.0166
Doctorate	0.80	0.66 — 0.97	0.0221		0.79	0.65 — 0.96	0.0178
unknown	1.37	0.71 — 2.67	0.3515		1.38	0.71 — 2.69	0.3401
Diabetes: (reference: non-diabetic)
recent/active	1.20	1.03 — 1.40	0.0172		1.21	1.04 — 1.41	0.0137
remote/inactive	1.27	0.66 — 2.45	0.4779		1.21	0.62 — 2.33	0.5798
unknown	1.65	0.73 — 3.73	0.2278		1.68	0.74 — 3.81	0.2122
Hypertension: (reference: non hypertensive)
recent/active	0.99	0.90 — 1.09	0.8102		0.99	0.90 — 1.09	0.8364
remote/inactive	1.28	0.97 — 1.68	0.0852		1.29	0.98 — 1.70	0.0717
unknown	0.96	0.39 — 2.36	0.9348		0.91	0.37 — 2.24	0.8394
Cardiac Event/Condition:* (reference: no event/condition)
recent/active	1.15	0.98 — 1.34	0.0971		1.15	0.98 — 1.35	0.0940
unknown	1.25	1.04 — 1.49	0.0146		1.22	1.02 — 1.46	0.0277
Referral: (reference: referred by self/relative/friend)
health professional					1.39	1.21 — 1.60	<0.0001
other†					1.20	1.08 — 1.33	0.0005
unknown					1.29	1.03 — 1.62	0.0282
Family History of Dementia: (reference: no family history)
≥1 1st degree relative					1.22	1.11 — 1.35	0.0001
unknown					0.87	0.73 — 1.04	0.1239

Reference for Baseline cognitively unimpaired: White, Male, High School education, non-diabetic, non-hypertensive, no history of cardiac event/condition, referred by self/relative/friend, and no family history of dementia.

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent”. A label of “unknown” selected if there was insufficient data to determine the status of the condition.

^*Cardiac event/condition includes history of myocardial infarction (MI), atrial fibrillation (AFib), and congestive heart failure (CHF). “Recent/active” selected if any of the three conditions were “recent/active”; categorization “absent” if all three were “absent” or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non-ADC media appeal.

NOTE: As is evident in the table, variables of interest included a number of “unknown” responses. Sensitivity analyses were performed, wherein an “unknown” response for any of the covariates in the table resulted in the subject being excluded from analyses. Although sample size was markedly reduced, race results were practically unchanged

Table 3.

Regression models predicting incident Dementia in participants who were diagnosed with MCI at Baseline. Model 1 included demographic, cardiovascular risk factors and baseline MCI etiology; Model 2 adds recruitment factors.

	Model 1				Model 2
Coefficient	HR	HR 95% CI	p value		HR	HR 95% CI	p value
African American (reference: White)	0.66	0.56 — 0.77	<0.0001		0.71	0.61 — 0.84	<0.0001
Female (reference: Male)	1.01	0.91 — 1.11	0.8642		1.03	0.94 — 1.14	0.5172
Education: (reference: HS education)
under HS	1.02	0.81 — 1.28	0.8859		0.99	0.79 — 1.25	0.9542
over HS	0.79	0.68 — 0.93	0.0035		0.80	0.68 — 0.93	0.0049
Bachelor	0.93	0.81 — 1.08	0.3359		0.94	0.82 — 1.09	0.4149
Masters	0.85	0.73 — 0.98	0.0298		0.85	0.74 — 0.99	0.0343
Doctorate	0.81	0.68 — 0.97	0.0236		0.79	0.66 — 0.95	0.0112
unknown	2.14	0.79 — 5.77	0.1332		2.13	0.79 — 5.77	0.1350
Diabetes: (reference: non-diabetic)
recent/active	1.04	0.89 — 1.22	0.5856		1.07	0.91 — 1.25	0.4100
remote/inactive	1.77	1.12 — 2.79	0.0148		1.96	1.24 — 3.10	0.0040
unknown	1.33	0.60 — 2.94	0.4785		1.33	0.59 — 2.98	0.4885
Hypertension: (reference: non hypertensive)
recent/active	0.98	0.89 — 1.09	0.7567		0.98	0.89 — 1.08	0.6952
remote/inactive	1.01	0.77 — 1.32	0.9447		0.97	0.74 — 1.28	0.8546
unknown	0.91	0.39 — 2.15	0.8351		0.95	0.40 — 2.28	0.9086
Cardiac Event/Condition:* (reference: no event/condition)
recent/active	0.95	0.80 — 1.13	0.5712		0.95	0.80 — 1.14	0.5954
unknown	0.88	0.73 — 1.06	0.1770		0.87	0.72 — 1.05	0.1374
Primary etiology for MCI (reference: Alzheimer’s disease)
Vascular disease	0.60	0.46 — 0.79	0.0003		0.63	0.48 — 0.83	9.00E-04
Lewy body	1.17	0.93 — 1.47	0.1716		1.10	0.88 — 1.38	0.4079
other	0.64	0.52 — 0.79	<0.0001		0.63	0.51 — 0.77	<0.0001
missing/unknown	0.59	0.54 — 0.66	<0.0001		0.63	0.57 — 0.70	<0.0001
Referral: (reference: referred by self/relative/friend)
health professional					1.46	1.29 — 1.64	<0.0001
other†					0.88	0.77 — 1.00	0.0574
unknown					0.80	0.61 — 1.06	0.1225
Family History of Dementia: (reference: no family history)
≥1 1st degree relative					1.12	1.01 — 1.25	0.0256
unknown					1.00	0.83 — 1.21	0.9727

Reference for Baseline cognitively unimpaired: White, Male, High School education, non-diabetic, non-hypertensive, no history of cardiac event/condition, referred by self/relative/friend, and no family history of dementia.

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent”. A label of “unknown” selected if there was insufficient data to determine the status of the condition.

^*Cardiac event/condition includes history of myocardial infarction (MI), atrial fibrillation (AFib), and congestive heart failure (CHF). “Recent/active” selected if any of the three conditions were “recent/active”; categorization “absent” if all three were “absent” or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non-ADC media appeal.

NOTE: As is evident in the table, variables of interest included a number of “unknown” responses. Sensitivity analyses were performed, wherein an “unknown” response for any of the covariates in the table resulted in the subject being excluded from analyses. Although sample size was markedly reduced, race results were practically unchanged.

4.3.2. Cardiovascular conditions and risk factors:

In those cognitively unimpaired at baseline, cardiovascular conditions were associated with risk of progression. Compared to those without a cardiac history, individuals with an unknown history for cardiac events demonstrated increased risk for progression to cognitive impairment. Additionally, recent/active diabetes conferred elevated risk for incident progression (Table 2, Model 1). In contrast, among those with baseline MCI, only remote/inactive diabetes was associated with risk for progression to dementia (Table 3, Model 1), although only a few participants reported remote/inactive diabetes (N=31; Table 1).

4.3.3. Etiology of MCI at Baseline:

For participants with MCI at baseline, primary etiology was included in the regression model (Table 3, Model 1). Non-AD etiologies were associated with reduced risk for progression. Specifically, VaD, “other”, and “missing/unknown” etiologies for MCI demonstrated an approximately 40% reduced hazard for progression compared to cases where AD was considered the primary cause for MCI.

4.4. Model 2: Examination of recruitment processes

With few exceptions, when enrollment factors were added to regression models, model parameters were generally unchanged. Notably, for those with MCI at baseline, the protective effect associated with being Black decreased slightly with the addition of enrollment factors.

In general, being referred by a clinician, and referral sources other than self, family or friend, was associated with increased risk for progression to cognitive impairment (Tables 2 and 3, Model 2). For those with baseline MCI, any referral source other than self, family or friend was associated with greater hazard for progression (Table 3, Model 2). Referral to the ADC by a health professional was associated with a 39% increased hazard for progression in those who were cognitively unimpaired at baseline and 46% increased hazard for those with MCI at baseline. Likewise, compared to individuals reporting no family history, those with a known family history of dementia exhibited an elevated risk for incident cognitive impairment – 22% increased hazard for those who were cognitively unimpaired at baseline, and 12% increased hazard in those with MCI at baseline.

Of note, sensitivity analyses excluding participants with ‘unknown’ responses for variables of interest were performed. The influence of race was essentially unchanged.

5. DISCUSSION

Employing a widely-used ADRD database, we derived a subsample of 12,242 participants from 37 US-based ADCs, and found that (1) non-Latino Blacks with MCI at baseline demonstrated a lower hazard for progression to dementia than Whites with MCI, and (2) cognitively unimpaired Blacks had a similar hazard for dementia or MCI compared to Whites. These findings contradict numerous reports of increased risk for incident cognitive impairment and dementia in Blacks compared to non-Latino Whites[15–22]. Published reports – predominantly population-based samples – include a meta-analysis suggesting that rate of progression to AD dementia was 64% higher in Blacks compared to non-Latino Whites[23] and a systematic review depicting an incident dementia rate 1.6 times higher in Blacks than in non-Latino Whites[24].

Considering findings from nested regression models including (1) subject characteristics, cardiovascular and metabolic risk factors, primary etiology for those with MCI at baseline, and (2) enrollment-related factors such as referral source and knowledge of family history of dementia, alongside differences in enrollment practices between Blacks and Whites, we suggest that selection bias could be among the factors contributing to results divergent from the extant literature. Specifically, survival models including the above factors revealed that outside of expected factors (female sex, education and diabetes), referral source and known family history of dementia were prominent predictors of incident cognitive impairment. Cognitively healthy participants referred to the ADCs Clinical Core by a health professional had approximately 39% higher hazard for MCI or dementia than those who were self- or family/friend-referred. Similarly, in the MCI subgroup, health professional referral was associated with 46% elevation in hazard. Importantly, enrollment factors differed between Blacks and Whites with Blacks being less likely than Whites to be recruited by a health professional.

These findings extend those of Chen et al[6] and others[25, 26]. Examining participants from a single ADC, Chen et al. found rate of progression to MCI differed based on recruitment source. Clinic-based recruitment was associated with a 25% higher rate of progression than community-based recruitment. The number of Blacks included in Chen et al.’s analysis was small, precluding an examination by race. It is nonetheless notable that all Blacks (N=8) in their analyses were recruited from non-clinical settings.

Individual Centers devise recruitment strategies to meet enrollment needs of Center-affiliated projects. Participant recruitment at ADCs typifies that of other similarly reputable multi-site ADRD cohort studies, e.g., Alzheimer’s Disease Neuroimaging Studies. These datasets include research volunteers recruited using a variety of strategies: recruitment from clinics, community outreach, among others. Data derived from ADCs serve to establish standards and models of disease progression, such as the recently published Research Framework for AD[27]. Moreover, seminal papers describing racial differences in cerebrospinal fluid proteins[28–30] and neuropathological findings have utilized ADC data[31, 32].

This study examined simple markers of sampling strategy and found preliminary evidence that sampling strategies may introduce selection bias at an aggregate level and likely at the level of individual ADCs. Selection bias occurs when sampling methods – in this case, our enrollment factors - are systematically associated with both an exposure of interest such as race and with outcomes of interest, e.g., incident dementia. Subsequently, our sampling strategies result in bias analogous to what is termed Healthy Worker Bias (HWB) [33] and obscure true relationships between race and cognitive aging trajectories. See Figure 3. for an elaboration of the potential effects of enrollment differences in reference to the HWB.

Figure 3.

Potential downstream influence of the enrollment factors, illustrated using the Healthy Worker bias as an example.

The degree to which this bias occurs likely varies by Center, as centers employ systematically different enrollment strategies. In the short-term, these findings suggest that including sampling strategy-related covariates may be important for analyses examining cognitive progression. In the longer-term, using community-based approaches for all groups could reduce both sampling and selection bias. Finally, the data support the idea that well-powered within-group analyses may provide more valid information on risk estimation than racial comparison studies[34].

It is possible that the present findings accurately reflect racial patterns in risk for cognitive impairment and dementia. Some studies report comparable rates for prevalence and incidence of dementia for non-Latino Blacks and Whites[35–37], or that adjusting for social factors such as education or literacy eliminates observed racial disparities in dementia prevalence and incidence[38, 39]. Data suggesting reductions in racial disparities for dementia, possibly related to improved education and healthcare, support the latter position[40].

To explore alternative contributing factors, we examined whether rate of incident dementia and cognitive impairment was influenced by underlying differences in (1) vascular risk and (2) clinically assessed baseline MCI etiology. Racial differences in vascular disease burden between non-Latino Whites and Blacks are hypothesized to underlie racial disparities in dementia prevalence. Data from the Atherosclerosis Risk in Communities (ARIC) Cohort study revealed that midlife vascular risk factors increased risk for dementia[41, 42] and amyloid deposition[43, 44], finding no interaction between race and vascular risk, i.e., elevated dementia risk in Blacks reflects differences in underlying vascular disease. In our analytic sample, Black participants had more cardiovascular disease than Whites. In total, our data did not suggest that elevated vascular burden in Whites accounted for Whites elevated risk for progression.

We also explored whether the lower rate of incident cognitive impairment in non-Latino Blacks occurred because their underlying etiology was less progressive, e.g., MCI due to VaD, as opposed to AD or DLB. In the present study, MCI due to VaD was associated with decreased hazard rate compared to MCI due to AD, supporting this supposition. On the other hand, MCI due to VaD accounted for fewer than 10% of diagnoses; the vast majority were AD or Mixed pathology. The limited number of autopsy studies including Black decedents precludes a full understanding of race-based etiological differences for MCI and dementia. Moreover, data from published studies were obtained from ADCs[31, 32], and may be subject to similar enrollment-based bias. The limited autopsy data available suggest that Blacks with a clinical diagnosis of dementia were more likely than Whites to have a mixed pathological load at autopsy; however, a mixed pathology including AD.

5.1. Limitations

5.1.1. The present study’s limitations include our inability to dissociate enrollment factors from other multi-dimensional factors, most obvious with family history. Family history is a dimensional risk factor, encompassing genetics, exposures, etc. We propose it also represents an enrollment factor, i.e., access to diagnostic services and/or awareness of risk – likely differentially associated with outcomes in Whites and Blacks. This position is based on the following reasoning: (1) the disease is frequently undiagnosed, especially in non-White populations[10]. Recent data suggest that ¹/₂ to ³/₄ of dementia cases go undiagnosed[7–9], and specialty-clinic referral patterns differ between non-Hispanic Blacks and Whites[11]; (2) Those affected by the disease would be inclined to participate in AD research[45]. A survey found that individuals with a family history of AD were both fearful of developing the disease and highly motivated to be in ADRD research [12].

We present a secondary analysis of variables collected for reasons other than assessing the effects of selection practices. Although imperfect, referral source and family history provide an estimate of how the participant was initially encouraged to join an ADC Clinical Core. Moreover, persons attending and enrolling at ADCs likely have markedly different factors influencing their cognitive health and in different proportions from individuals from the general population. Appendix B highlights the differences between the analytic sample and those excluded due to lack of follow-up.

Finally, regression models predicting incident cognitive impairment cannot fully account for participant attrition and events occurring before monitoring. Systematic differences between comparator groups in attrition and age of events would alter results. These concerns along with the sample’s racial imbalance make generalization beyond this population problematic.

5.2. Conclusions

Altogether, we suggest that it may be problematic to draw conclusions based on racial comparisons derived from datasets with racial differences in enrollment strategies. The bias introduced by enrollment strategies resulting in selection bias would be difficult to address in statistical models, as discussed by investigators wrestling with the Healthy Worker Bias[46]. Many ADCs rely on clinic-based recruitment, especially for MCI cases. Blacks are less likely to be evaluated in clinic[10]. Thus, URG recruitment often relies on community-based strategies. If clinic-based recruitment is the predominant source for White participants and community-based recruitment the primary source of Black participants, as is the case for many ADCs, the bias introduced by design would be difficult to overcome.

Socio-cultural factors including lack of access to quality education[47], neighborhood factors[48], and higher lifetime stress[49] are hypothesized to contribute mechanistically, either driving disease and comorbid risk factors (e.g., weathering, elevated cardiovascular disease), or lowering reserve (See Brewster et al[50]. for review). Blacks and other historically marginalized groups are more likely to carry the burden of these social-contextual inequities. It is imperative that we address study design flaws that may obscure accurate understanding of these factors’ influence on ADRD for both Blacks and Whites.

We argue that the approach to reducing the influence of selection bias in ADC datasets is two-fold. ADCs should (1) use community-based approaches applied within URG populations to recruit and retain both White and URG samples; and (2) improve clinic access for Blacks, e.g., encouraging primary care providers to refer Black patients with concerns, and empowering Black families to advocate for diagnostic services. In the meantime, it will be important to exercise caution when using ADC data to derive new frameworks and cut-points based on cohorts, subject to selection bias[50].

Appendix A.

Appendix Table A.

Racial comparisons of full sample.

Full Sample, included only non-Hispanic Whites and Blacks. N of 26,304 includes some individual excluded from analyses after additional exclusion criteria were applied. Data were collected between June 2005 and February 2017.

	Overall		White	Black	P
N	26304	22225		4079
African American or Black, N(%)	4079 (15.5)
Entry age in yrs, mean(SD)	74.65 (8.09)	74.79 (8.13)		73.90 (7.84)	<0.001
Female Sex, N(%)	14834 (56.4)	11857 (53.3)		2977 (73.0)	<0.001
Died, N(%)	6314 (24.0)	5696 (25.6)		618 (15.2)	<0.001
Education Category					<0.001
Under High School, N(%)	1513 (5.8)		817 (3.7)	696 (17.1)
High School, N(%)	5141 (19.5)		4091 (18.4)	1050 (25.7)
Over High School, N(%)	4809 (18.3)		3866 (17.4)	943 (23.1)
Bachelor degree, N(%)	5976 (22.7)		5426 (24.4)	550 (13.5)
Masters degree, N(%)	6051 (23.0)		5408 (24.3)	643 (15.8)
Doctorate degree, N(%)	2666 (10.1)		2491 (11.2)	175 (4.3)
Unknown, N(%)	148 (0.6)		126 (0.6)	22 (0.5)
Diabetes					<0.001
Absent, N(%)	22988 (87.4)		20018 (90.1)	2970 (72.8)
Recent/active, N(%)	3074 (11.7)		2024 (9.1)	1050 (25.7)
Remote/inactive, N(%)	165 (0.6)		128 (0.6)	37 (0.9)
unknown, (N%)	77 (0.3)		55 (0.2)	22 (0.5)
Hypertension					<0.001
Absent, N(%)	12189 (46.3)		11208 (50.4)	981 (24.1)
Recent/active, N(%)	13145 (50.0)		10200 (45.9)	2945 (72.2)
Remote/inactive, N(%)	885 (3.4)		745 (3.4)	140 (3.4)
Unknown, (N%)	85 (0.3)		72 (0.3)	13 (0.3)
Cardiac event/condition*					<0.001
Absent, N(%)	22435 (85.3)		18880 (84.9)	3555 (87.2)
Recent/active, N(%)	2044 (7.8)		1795 (8.1)	249 (6.1)
Unknown, (N%)	1825 (6.9)		1550 (7.0)	275 (6.7)
Diagnosis at Baseline					<0.001
Cognitively unimpaired	10249 (39.0)		8556 (38.5)	1693 (41.5)
Impaired-not-MCI	1126 (4.3)		894 (4.0)	232 (5.7)
MCI	5721 (21.7)		4768 (21.5)	953 (23.4)
Dementia	9208 (35.0)		8007 (36.0)	1201 (29.4)
Referral Source					<0.001
Self/relative/friend, N(%)	8134 (30.9)		6745 (30.3)	1389 (34.1)
Health professional, N(%)	9527 (36.2)		8475 (38.1)	1052 (25.8)
Other, N(%)†	7937 (30.2)		6393 (28.8)	1544 (37.9)
Unknown, N(%)	706 (2.7)		612 (2.8)	94 (2.3)
Family History of dementia					<0.001
No 1st degree relative, N(%)	9855 (37.5)		8218 (37.0)	1637 (40.1)
≥One 1st degree relative, N(%)	13677 (52.0)		11880 (53.5)	1797 (44.1)
unknown, (N%)	2772 (10.5)		2127 (9.6)	645 (15.8)
≥ One follow-up visit, N(%)	18812 (71.5)		16083 (72.4)	2729 (66.9)	<0.001
Years of follow-up, mean (SD)	2.94 (2.94)		2.97 (2.94)	2.75 (2.96)	<0.001

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent”. A label of “unknown” selected if there was insufficient data to determine the status of the condition.

^*Cardiac event/condition includes history of myocardial infarction (MI), atrial fibrillation (AFib), and congestive heart failure (CHF). “Recent/active” selected if any of the three conditions were “recent/active”; categorization “absent” if all three were “absent” or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non- ADC media appeal.

Appendix B.

Appendix Table B.1.

Comparison of analytic sample to group meeting all inclusion criteria with the exception of having follow-up data. Data were collected between June 2005 and February 2017.

	Analytic Sample	Excluded - No Follow-up data	P
N	12,242	3,622
African American or Black, N(%)	1904 (15.6)	742 (20.5)	<0.001
Entry age in yrs, mean(SD)	74.38 (7.94)	73.78 (8.15)	<0.001
Female Sex, N{%)	7235 ( 59.1)	2148 ( 59.3)	0.841
Died, N(%)	1933 (15.8)	397 (11.0)	<0.001
Education Category			<0.001
Under High School, N{%)	446 (3.6)	182 (5.0)
High School, N(%}	1915 (15.6)	647 (17.9)
Over High School, N(%)	2314 (18.9)	692 (19.1)
Bachelor degree, N(%)	2894 (23.6)	801 (22.1)
Masters degree, N(%)	3270 (26.7)	923 (25.5)
Doctorate degree, N(%)	1365 (11.2)	356 (9.8)
Unknown, N(%)	38 (0.3)	21 (0.6)
Diabetes			<0.001
Absent, N(%)	10749 (87.8)	3079 (85.0)
Recent/active, N(%)	1394 (11.4)	499 (13.8)
Remote/inactive, N(%)	64 (0.5)	33 (0.9)
unknown, (N%)	35 (0.3)	11 (0.3)
Hypertension			0.006
Absent, N(%)	5743 (46.9)	1608 (44.4)
Recent/active, N(%)	6139 (50.1)	1878 (51.8)
Remote/inactive, N(%)	326 (2.7)	119 (3.3)
Unknown, (N%)	34 (0.3)	17 (0.5)
Cardiac event/condition*			0.058
Absent, N(%)	10568 (86.3)	3070 (84.8)
Recent/active, N(%)	909 (7.4)	299 (8.3)
Unknown, (N%)	765 (6.2)	253 (7.0)
Diagnosis at Baseline
Cognitively unimpaired	8182 (66.8)	2067 (57.1)
Impaired-not-MCI	0(0.0)	0(0.0)
MCI	4060 (33.2)	1555 (42.9)
Dementia	0 (0.0)	0 (0.0)
Referral Source			<0.001
Self/relative/friend, N(%)	4358 (35.6)	1248 (34.5)
Health professional, N(%)	2757 (22.5)	1181 (32.6)
Other, N(%)†	4688 (38.3)	1106 (30.5)
Unknown, N{%)	439 (3.6)	87 (2.4)
Family History of dementia			<0.001
No 1st degree relative, N(%)	4522 (36.9)	1501 (41.4)
≥ One 1st degree relative, N(%}	6601 (53.9)	1661 (45.9)
unknown, (N%)	1119 (9.1)	460 (12.7)
≥ One follow-up visit, N(%)	12.242 (100.0)	0 (0)	<0.001
Years of follow-up, mean (SD)	4.55 (2.83)	0(0)	<0.001

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent” A label of “unknown” selected if there was insufficient data to determine the status of the condition.

^*Cardiac event/condition includes history of myocardial infarction (Ml), atrial fibrillation (AFib), and congestive heart failure {CHF). “Recent/active” selected if any of the three conditions were “recent/active’’; categorization “absent” if all three were “absent or “remote/inactive”, and “unknown” for combinations of “unknown”, “absent”, and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non-ADC media appeal.

Appendix Table B.2.

Comparison of inclusion status by race. Analytic sample is compared to group meeting all inclusion criteria with the exception of having follow-up data. Data were collected between June 2005 and February 2017.

	Analytic Sample			Excluded - No Follow-up data		p
	Blacks	Whites		Blacks	Whites
N	1904	10338		742	2880
Entry age in yrs, mean(SD)	72.96 {7.26)	74.64 (8.03)		72.15(7.34)	74.19 (8.29)	<0.001
Female Sex, N(%)	1437 (75.5)	5798 (56.1)		539 (72.6)	1609 (55.9)	<0.001
Died, N(%)	187 (9.8)	1746 (16.9)		55 (7.4)	342 (11.9)	<0.001
Education Category						<0.001
Under High School, N(%)	216(11.3)	230(2.2)		101(13,6)	81 (2.8)
High School, N(%)	454(23.8)	1461(14,1)		185 (24,9)	462 (16.0)
Over High School, N(%)	486 (25.5)	1828 (17.7)		179 (24.1)	513 (17.8)
Bachelor degree, N(%)	268(14.1)	2626 (25.4)		124(16.7)	677 (23.5)
Masters degree, N(%)	371(19.5)	2899 (28.0)		127(17.1)	796 (27.6)
Doctorate degree, N(%)	106 (5.6)	1259(12.2)		26 (3.5)	330 (11.5)
Unknown, N(%)	3 (0.2)	35 (0.3)		0 (0.0)	21(0.7)
Diabetes						<0.001
Absent, N(%)	1376 (72.3)	9373 (90.7)		535(72.1)	2544 (88.3)
Recent/active, N(%)	500(26.3)	894 (8.6)		196(26.4)	303 (10.5)
Remote/inactive, N(%)	14 (0.7)	50 (0.5)		10 (1.3)	23 (0.8)
unknown, (N%)	14 (0.7)	21(0.2)		1 (0.1)	10 (0.3)
Hypertension						<0.001
Absent, N(%)	448 (23.5)	5295 (51.2)		190(25.6)	1418 (49.2)
Recent/active, N(%)	1401 (73.6)	4738 (45.8)		519 (69.9)	1359 (47.2)
Remote/inactive, N(%)	51 (2.7)	275 (2.7)		27 (3.6)	92(3.2)
Unknown, (N%)	4 (0.2)	30 (0.3)		6 (0.8)	11(0.4)
Cardiac event/condition*						<0.001
Absent, N(%)	1684 (88.4)	8884 (85.9)		645 (86.9)	2425 (84.2)
Recent/active, N(%)	100 (5.3)	809 (7.8)		45 (6.1)	254 (8.8)
Unknown, (N%)	120 (6.3)	645 (6.2)		52 (7.0)	201 (7.0)
Diagnosis at Baseline
Cognitively unimpaired	1288 (67.6)	6894 (66.7)		405 (54.6)	1662 (57.7)
Impaired-not-MCI	0 (0.0)	0 (0.0)		0 (0.0)	0 (0.0)
MCI	616 (32.4)	3444 (33.3)		337 (45.4)	1218 (42.3)
Dementia	0 (0.0)	0 (0.0)		0 (0.0)	0 (0.0)
Referral Source						<0.001
Self/relative/friend, N(%)	696 (36.6)	3662 (35.4)		286 (38.5)	962 (33.4)
Health professional, N(%)	259 (13.6)	2498 (24.2)		169 (22.8)	1012 (35.1)
Other, N(%)†	881 (46.3)	3807 (36.8)		274(36.9)	832 (28.9)
Unknown, N(%)	68 (3.6)	371 (3.6)		13 (1.8)	74 (2.6)
Family Flistory of dementia						<0.001
No 1st degree relative, N(%)	806 (42.3)	3716 (35.9)		347 (46.8)	1154 (40.1)
≥ One 1st degree relative, N(%)	852 (44.7)	5749 (55.6)		272 (36.7)	1389 (48.2)
unknown, (N%)	246 (12.9)	873 (8.4)		123 (16.6)	337 (11.7)
≥ One follow-up visit, IM(%)	1904 (100.0)	10338 (100,0)		0(0)	0(0)	<0.001
Years of follow-up, mean (SD)	4.45 (2.84)	4.57 (2.83)		0 (0)	0 (0)	<0.001

The designation of “recent/active” was given if the condition occurred within the year prior to the study visit or required active management in the case of chronic conditions and the report was consistent with information obtained during the subject and study partner interview. The label “remote/inactive” was indicated if the condition existed or occurred in the past or more than one year ago, but had resolved or no longer needed treatment or management. As implied, if a condition was reported by either the participant or the study partner or indicated from other records, it was designated as “absent”. A label of “unknown” selected it there was insufficient data to determine the status of the condition.

^*‘Cardiac event/condition includes history of myocardial infarction (Ml), atrial fibrillation (AFib), and congestive heart failure (CHF). “Recent/active” selected if any of the three conditions were “recent/active”; categorization “absent” if all three were “absent or “remote/inactive”, and “unknown” for combinations of “unknown” “absent” and “remote/inactive”.

^†The category of “Other” for Referral Source represents a mixture of categories, including Alzheimer’s Disease Center (ADC) solicitation, non-ADC study, population sample, or non-ADC media appeal.