Preventing Cognitive Decline in Black Individuals With Mild Cognitive Impairment: A Randomized Clinical Trial

Key Points

Question  Does increasing cognitive, physical, and/or social activities prevent cognitive decline in black individuals with mild cognitive impairment?

Findings  This 2-year randomized clinical trial included 221 black individuals with mild cognitive impairment and found that behavioral activation (designed to increase cognitive, physical, and/or social activity) significantly reduced the risk of cognitive decline compared with supportive therapy (an attention control treatment).

Meaning  Behavioral activation prevents cognitive decline in black individuals with mild cognitive impairment over and above social interaction.

Importance  Mild cognitive impairment (MCI) is a transition state between normal cognitive aging and dementia that increases the risk for progressive cognitive decline. Preventing cognitive decline is a public health priority.

Objective  To determine whether behavioral activation prevents cognitive and functional decline over 2 years in black individuals with MCI.

Design, Setting, and Participants  Single-center, single-masked, attention-controlled randomized clinical trial. Participants were enrolled from June 21, 2011, to October 3, 2014, and follow-up ended December 13, 2016. Community-based recruitment and treatment of black individuals older than 65 years with amnestic MCI. Volunteer sample of 1390 persons with memory complaints were screened. Overall, 536 individuals had baseline assessment, and 315 (58.8%) were ineligible, most often owing to normal cognition (205 of 315 [65%]) or dementia (59 of 315 [18.7%]); 221 fully eligible participants were randomized. Analyses were intention to treat.

Interventions  Participants were randomized to behavioral activation, which aimed to increase cognitive, physical, and social activity (111 [50.2%]), or supportive therapy, an attention control treatment (110 [49.8%]).

Main Outcomes and Measures  The prespecified primary outcome was a decline of 6 or more recalled words on the total recall score of the Hopkins Verbal Learning Test–Revised assessed at 6, 12, 18, and 24 months. The secondary outcome was functional decline.

Results  Of 221 randomized participants (mean [SD] age, 75.8 [7.0] years, 175 women [79%]), 77 behavioral activation participants (69.4%) and 87 supportive therapy participants (79.1%) had 2-year outcome assessments. After baseline, behavioral activation participants engaged in significantly more cognitive activities than supportive therapy participants. The 2-year incidence of memory decline was 1.2% (95% CI, 0.2-6.4) for behavioral activation vs 9.3% (95% CI, 5.30-16.4) for supportive therapy (relative risk, 0.12; 95% CI, 0.02-0.74; P = .02). Behavioral activation was associated with stable everyday function, whereas supportive therapy was associated with decline (difference in slopes, 2.71; 95% CI, 0.12-5.30; P = .04). Rates of serious adverse events for behavioral activation and supportive therapy, respectively, were: falls (14 [13%] vs 28 [25%]), emergency department visits (24 [22%] vs 24 [22%]), hospitalizations (36 [32%] vs 31 [28%]), and deaths (7 [5%] vs 3 [4%]).

Conclusions and Relevance  Behavioral activation prevented cognitive and functional decline, but this finding requires further investigation. Black individuals have almost twice the rate of dementia as white individuals; behavioral activation may reduce this health disparity.

Trial Registration  ClinicalTrials.gov Identifier: NCT01299766

Mild cognitive impairment (MCI) is a transition state between normal cognitive aging and dementia that increases the risk for progressive cognitive decline.^1,2 There is no effective pharmacologic treatment for MCI, but randomized clinical trials (RCTs) and observational studies have suggested that cognitive, physical, and/or social activity may prevent decline.^2-4 However, these studies have included few black individuals who may differ from white individuals in risk profile (eg, cognitive reserve, hypertension, diabetes), mechanisms of decline (ie, microvascular disease, amyloid deposition), and rates of dementia (ie, black individuals have almost twice the rate).^5-9 These differences reflect variances in education, literacy, access to care, and genetic, social, financial, and health status and provide theoretical support for culturally relevant interventions.^10-13 There are now 3.5 million older black individuals in the United States, and the number is expected to triple by 2050.¹⁴ This projected growth increases the potential burden of dementia in this high-risk population and necessitates preventive interventions. This RCT compared the efficacy of behavioral activation, which aimed to increase participation in cognitive, physical, and social activities, vs supportive therapy, an attention control treatment, to prevent cognitive decline.

Institutional review board approval was obtained for this RCT, and all participants provided written informed consent. Race-concordant community health workers recruited the sample, delivered the interventions, and conducted clinical assessments masked to treatment assignment. Potential participants were recruited when they sought memory screening at senior centers, senior housing sites, churches, and primary care clinics in Philadelphia, Pennsylvania. The community health workers administered trial 1 (immediate recall) of the Hopkins Verbal Learning Test–Revised (HVLT-R).¹⁵ Interested participants who recalled fewer than 5 of 12 words had a comprehensive in-home baseline assessment to determine eligibility. The inclusion criteria were self-identified black race, 65 years and older, and amnestic-multiple or single-domain MCI according to National Institute on Aging and Alzheimer’s Association criteria.¹ The exclusion criteria were axis I Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) psychiatric diagnoses (eg, major depression, dementia), taking a US Food and Drug Administration–approved antidementia medication, and severe sensory deficits. The full trial protocol is available in Supplement 1.

The study statistician (B.L.) randomized eligible participants using a random numbers table, sealed envelopes containing treatment assignments, and a fixed randomization scheme with a 1:1 allocation ratio to the 2 study groups. The RCT was single-masked in that outcome assessments were masked to treatment assignment, but participants were aware of their assigned treatments. Behavioral activation (experimental treatment) and supportive therapy (control treatment) consisted of 5 in-home 60-minute treatment sessions over 4 months and 6 in-home 60-minute follow-up maintenance sessions over the next 20 months.

Behavioral activation uses goal setting and action plans to reinforce the practice of healthy activities.¹⁶ We used this approach to increase cognitive, physical, and/or social activity and incorporated concepts from the health belief model to enhance cultural relevance, including respect for personal values, self-selected treatment goals, racially sensitive education materials, race-concordant community health workers to build trust, and in-home delivery to increase access.¹⁷ Action plans relied on visual cues, written schedules, step-by-step sequencing, and procedural memory (ie, learning behaviors that are enacted automatically) to compensate for cognitive deficits. Reminders were embedded in the steps (eg, set timer to ring 30 minutes before social event; record upcoming tasks on calendar). An action plan for exercise, such as walking with a friend, might be to call a friend after lunch, pick a time to meet, and record the date on the calendar to cue the action. The community health workers anticipated negative perceptions (eg, “This will never work.”) and encouraged participants to allow these thoughts while still following the action plan (ie, “Follow the plan instead of the feeling.”). Subsequent treatment sessions assessed treatment progress. If goals were not met, barriers and ways to overcome them were discussed. If goals were unrealistic, similar yet easier activities were scheduled (eg, puzzle with fewer pieces, less strenuous exercise).

Supportive therapy controlled for the nonspecific effects of social interaction. This is a structured, nondirective psychological treatment that facilitates personal expression and conveys empathy, respect, and optimism.¹⁸ Its techniques include active listening, open questioning, reflecting back, and summation. Discussions focused on the experience of aging, memory loss, illness, disability, and social isolation but did not include any of behavioral activation’s goal-setting strategies.

All treatment sessions were audiotaped, and an experienced psychotherapy researcher (M.T.H.) reviewed one-third of randomly selected tapes to rate adherence to both treatment protocols. On a scale from 1 to 5 (better), the mean (SD) global fidelity ratings were 4.1 (0.9) for behavioral activation and 4.4 (0.8) for supportive therapy.

Data collected at baseline included age, sex, marital status, years of education, and literacy (assessed using the reading recognition subtest of the Wide Range Achievement Test-Version 3).¹⁹ Other study measures are described below.

Cognition was assessed with the HVLT-R and the National Alzheimer Coordinating Center’s uniform data set (UDS) neuropsychological battery.²⁰ The HVLT-R is a word list learning test that consists of a 12-item word list presented in 3 learning trials. The total recall score is the sum of the 3 learning trials (from 0 to 36) and represents learning and working memory. The UDS includes the Wechsler Memory Scale–Revised logical memory tests (immediate and delayed), digit span forward/backward, Wechsler Adult Intelligence Scale–Revised digit symbol, Trail Making Test A and B, animal fluency, and Boston Naming Test. Most UDS tests have age-, race/ethnicity–, and education-adjusted norms.^19,21-23

At baseline, diagnoses of amnestic MCI were based on the following criteria: (1) subjective memory complaint; (2) scores 1.5 SD below published norms for older black individuals on the HVLT-R (amnestic single-domain MCI) or 1.5 SD below published norms on the HVLT-R and a second cognitive test in the UDS (amnestic multiple-domain MCI); (3) no substantial functional impairment (ie, scores ≤43 on the University of California Performance-Based Skills Assessment, which is the mean score in persons with dementia); and (4) no global cognitive impairment suggestive of dementia. After baseline, the HVLT-R was administered at months 6, 12, 18, and 24 using different test versions. The UDS was administered at months 12 and 24. The prespecified primary outcome was a decline of 6 or more recalled words on the HVLT-R total recall score at 24 months relative to baseline. A decline of this magnitude correlates with functional decline and is the reliable change score.^15,24 Progression to dementia was defined as a decline of 6 or more on the HVLT-R and a University of California Performance-Based Skills Assessment score indicative of dementia at 24 months.²⁵

Participants self-reported chronic medical conditions and current medications. All medications were used to calculate the Chronic Disease Score, which is a weighted medication-based score that represents overall medical comorbidity (higher scores indicate worse comorbidity).²⁶ The University of California Performance-Based Skills Assessment was used as an objective test of accuracy with writing checks, making change, using a telephone, and scheduling a physician appointment (higher scores indicate better function).²⁵

The Florida Cognitive Activities Scale is a 25-item measure of cognitive and social activities with known reliability and validity in older black individuals. The 10-item higher cognition subscale was used to assess frequency of participation in cognitive activities (eg, reading, cooking, crossword puzzles), which were rated as not done in past year (0); a few times a year (1); every couple of months (2); a couple times a month (3); a few times a week (4); and every day (5).²⁷ The 9-item US Health Interview Survey was used to assess frequency of physical activities (eg, number of days of walking for exercise, dancing, biking) over the previous 2 weeks.²⁸

Continuous baseline demographic and clinical characteristics were summarized using mean (SD) and categorical variables using counts and percentages. The primary analysis was performed on the modified intent-to-treat population including all available data from all participants with at least 1 follow-up visit. Poisson regression with robust standard errors (generalized estimating equation) was used to jointly model decline in HVLT-R total recall score of 6 or more words at 6, 12, 18, and 24 months by treatment group. The model included time, treatment, time-by-treatment interaction terms, and baseline HVLT-R total recall score. A compound symmetric structure was assumed for the working correlation structure. After adjusting for baseline HVLT-R total recall score, other potential baseline predictors of decline (eg, age, sex, diabetes, literacy) were evaluated for potential adjustment. Only age met the prespecified threshold of P value less than .10 for inclusion. The final model included time (treated as a categorical variable), treatment group, time-by-treatment interaction, baseline HVLT-R total recall score, and age. The primary hypothesis estimated the relative risk (RR) for decline in behavioral activation vs supportive therapy participants at 24 months. The RCT was powered to detect an overall 50% reduction in risk of decline for the primary outcome using a generalized linear model for binomial data with a log link. Assuming 40% of supportive therapy participants would decline and 20% attrition by 24 months, complete data on 160 participants at 24 months would yield 80% power to detect an RR of 0.5 using a 2-sided test with α = .05.

The primary analysis is valid under the assumption that missing data are missing completely at random, and results are subject to potential bias if missing data are more likely in 1 of the arms or otherwise not missing completely at random. To mitigate concerns about differential attrition and to assess sensitivity to the missing completely at random assumption, missing HVLT-R data at months 6 to 24 were imputed for all randomized individuals under the missing at random mechanism using the full conditional specification method in SAS PROC MI (SAS Institute Inc). Hopkins Verbal Learning Test–Revised scores were imputed and then categorized with respect to decline from baseline. Variables included in the imputation procedure were age, treatment group, and prior time HVLT-R scores. Two hundred data sets were imputed, and the same analysis as above was applied to these 200 complete data sets with estimates summarized using SAS PROC MIANALYZE (SAS Institute Inc).

In supporting analysis for the primary outcome, Poisson regression was used to model the number of assessments (ie, at 6, 12, 18, and 24 months) with a decline of 6 or more words over 2 years as a function of treatment group, baseline HVLT-R score, and age, with number of assessments as the offset.

Mixed-effects linear regression was used to model the trajectory of neuropsychological test scores (continuous outcomes) measured repeatedly over time. Fixed effects for time (as a continuous variable), treatment group, time-by-treatment group interaction, and age at baseline were included. Random intercepts and slopes were included to account for within-individual correlation. This analysis was applied separately to the modified intent-to-treat population and to all randomized participants.

From June 2011 to October 2014, 536 individuals had baseline assessments. Of them, 315 (58.8%) were ineligible, most often because of normal cognition (205 [65.1%]) or dementia (59 [18.7%]). A total of 221 participants with amnestic MCI (177 [80%] with amnestic-multiple domain MCI) were randomized to behavioral activation (111 [50.2%]) or supportive therapy (110 [49.8%]). The treatment groups had similar baseline demographic and clinical characteristics (Table 1 and eTable 1 in Supplement 2). The mean (SD) age of participants was 75.8 (7.0) years, and 175 (79%) were women. The Figure depicts the CONSORT flowchart. Compared with participants with any follow-up data after baseline (ie, the modified intent-to-treat population, n = 190), participants with no follow-up data (ie, those who withdrew or died before the 6-month follow-up; 31 [14%]) were more likely to be men (11 [35.5%] vs 35 [18.4%]; χ² = 4.71; P = .03) and had lower literacy scores (mean [SD], 24.68 [7.45] vs 27.32 [6.69]; F_1,219 = 4.02; P = .05). The 2-year RCT was completed by 77 behavioral activation participants (69.4%) and 87 supportive therapy participants (79.1%) (χ² = 2.73; P = .10). Attrition was related to sex such that 19 noncompleters (33%) (ie, those who withdrew from the study or died by 24 months) vs 27 completers (16%) were men (χ² = 7.30; P = .007); otherwise, there were no differences in the characteristics (including severity of MCI) of randomized participants and those who were evaluated at 24 months.

Behavioral activation participants received a mean (SD) of 4.2 (1.7) initial treatment sessions and 5.2 (1.5) maintenance sessions; supportive therapy participants received a mean (SD) of 4.4 (1.4) initial treatment sessions and 5.6 (1.0) maintenance sessions. After baseline, behavioral activation participants engaged in significantly more cognitive activities at 6, 12, 18, and 24 months than supportive therapy participants (Table 2). There were no differences in physical or social activity (eTables 2 and 3 in Supplement 2).

Incidence rates of a decline of 6 or more on HVLT-R total recall (primary outcome) at each time by treatment group appear in Table 3. After 6 months, the RR for cognitive decline at months 12, 18, and 24 was progressively lower in behavioral activation vs supportive therapy participants. At 24 months, the model-estimated rate of decline was 1.2% (95% CI, 0.2-6.4) in behavioral activation vs 9.3% (95% CI, 5.3-16.4) in supportive therapy (RR, 0.12; 95% CI, 0.02-0.74; P = .02). The multiple imputation analysis yielded a similar result (RR, 0.17; 95% CI, 0.03-1.11; P = .06). A sensitivity analysis comparing treatment group differences on a decline of 5 or more in HVLT-R total recall yielded similar results to the primary analysis (RR, 0.16; 95% CI, 0.04-0.61; P = .01).

The mean number of assessments (ie, at 6, 12, 18, and 24 months) at which behavioral activation vs supportive therapy participants showed a decline of 6 or more on HVLT-R total recall was 0.16 (95% CI, 0.10-0.27) vs 0.33 (95% CI, 0.23-0.47), respectively (incidence rate ratio, 0.49; 95% CI, 0.27-0.87; P = .02). The incidence rate ratio indicates that behavioral activation participants had half the number of assessments showing decline as supportive therapy participants over 2 years. eTables 4 and 5 in Supplement 2 illustrate the patterns of decline for each treatment group. Three participants in the behavioral activation group and 5 in the supportive therapy group met the outcome criterion and subsequently withdrew, went missing, or died. The number of participants with an improvement of 6 or more on the HVLT-R was higher in the behavioral activation vs the supportive therapy group at 12 months and later (although not statistically significant), mirroring the findings from the analysis of decline (eTable 6 in Supplement 2).

When HVLT-R scores were considered as a continuous variable in the modified intent-to-treat population, the mean change per year (slope) in HVLT-R total recall scores in behavioral activation participants was 0.74 (95% CI, 0.25-1.23; P = .003), representing improvement, whereas supportive therapy participants had no change (0.04 [95% CI, −0.42 to 0.50; P = .87]). The difference in slopes was 0.70 (95% CI, 0.03-1.37; P = .04). The mixed-effects analysis of all randomized participants (eTable 7 in Supplement 2) yielded a similar result (difference in slopes, 0.68; 95% CI, 0.01-1.35; P = .05). Also, behavioral activation participants had stable University of California Performance-Based Skills Assessment (function) scores over time (slope, −0.13; 95% CI, −2.05 to 1.79; P = .89), whereas supportive therapy participants declined (−2.60; 95% CI, −4.36 to −0.83; P = .004). The difference in slopes was 2.47 (95% CI, −0.14 to 5.07; P = .06). The mixed-effects analysis of all randomized participants (eTable 7 in Supplement 2) yielded a similar result (difference in slopes, 2.71; 95% CI, 0.12-5.30; P = .04).

Four of 90 behavioral activation participants (4.4%) and 8 of 95 supportive therapy participants (8.4%) progressed to dementia after month 6 (χ² = 1.45; P = .25). Among the neuropsychological test outcomes measured at 12 and 24 months, in all randomized participants, behavioral activation participants had significant annual improvement in executive function (ie, Trail Making Test, part B; slope, −13.5; 95% CI, −21.4 to −5.52; P < .001), while supportive therapy did not (slope, 3.18, 95% CI, −4.17 to 10.53; P = .40). The difference in slopes was significant (−16.6; 95% CI, −27.5 to −5.82; P = .003). This effect was nearly identical in the modified intent-to-treat population (difference in slopes, −16.6; 95% CI, −27.6 to −5.67; P = .003). No significant differences were found on other neuropsychological tests (eTables 8 and 9 in Supplement 2).

Rates of serious adverse events over the 2 years for participants in the behavioral activation and supportive therapy groups, respectively, were falls (14 [13%] vs 28 [25%]), emergency department visits (24 [22%] vs 24 [22%]), hospitalizations (36 [32%] vs 31 [28%]), and deaths (7 [5%] vs 3 [4%]).

To our knowledge, this is the first RCT to demonstrate that behavioral activation specifically prevents cognitive and functional decline in older black individuals with MCI. We found that 1 behavioral activation participant (1.2%) vs 9 supportive therapy controls (9.3%) declined significantly in memory over 2 years. This was a statistically significant reduction in relative risk (number needed to treat, 12.25), but its clinical significance requires further investigation. To provide context to this result, epidemiologic studies suggest that 12% to 41% of older black individuals decline in memory over 2 years.^29-32 Thus, behavioral activation’s treatment effect is notable, and supportive therapy’s positive effect suggests that social interaction itself may be beneficial but to a lesser extent.

Previous RCTs of behavioral interventions to prevent cognitive decline in at-risk (not MCI) populations have had mixed results. The Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability RCT (N = 1260) found that, compared with general health advice, a multidomain lifestyle intervention (diet, exercise, cognitive training, and vascular risk monitoring) improved or maintained scores on a comprehensive neuropsychological test battery over 2 years.³³ The Multidomain Alzheimer Preventive Trial (N = 1680) found that a multidomain intervention (cognitive training, physical activity, and nutrition, with and without ω-3 polyunsaturated fatty acids) provided no cognitive benefit over 3 years, but a secondary data analysis revealed positive treatment effects in participants with high cardiovascular risk scores.^34,35 The effectiveness of multidomain lifestyle/behavioral interventions remains uncertain in persons with MCI. A meta-analysis of 14 RCTs of exercise interventions for MCI revealed small effects on verbal fluency but no effects on executive function, memory, or information processing.³⁶ A review of 10 studies on cognitive training found that poor methodologic quality limited assessment of overall effectiveness.³⁷ These conflicting results likely reflect variation in diagnostic criteria and accuracy, study interventions, treatment fidelity and adherence, outcome measures, and/or lack of efficacy.

This RCT overcomes these limitations by using current diagnostic criteria and validated psychometric outcome measures, recruiting a large sample, using an intent-to-treat analysis, and having successful randomization, adequate power, masked outcome assessments, and high adherence to 2 standardized and credible interventions.¹ A unique strength is the prevention of cognitive decline in older black individuals with MCI, a high-risk and greatly understudied population. However, the racial characteristics of the sample and motivation to enroll an RCT limit generalizability. Also, nonrandom attrition (higher in men) may have skewed results; higher attrition rates in men have similarly been found in other RCTs.^38,39 The primary outcome was psychometrically rather than clinically defined to facilitate community-based assessment. This approach balanced the need for scientific rigor with the recognized difficulties recruiting, assessing, and retaining minority participants in a relatively long RCT. The use of a psychometrically defined outcome may also explain why different numbers of participants met the primary outcome at different times and likely reflects the effects of changes in medical status, practice effects, and attrition, as well as cumulative treatment effects. Despite these sources of variability, the treatment effect of behavioral activation was consistent, increased over time, was evident in other measures of cognition and function, and was associated with greater participation in cognitive activity rather than physical or social activities, perhaps because participants pursued cognitive activity to a greater extent than other activities. However, the clinical significance of the findings and underlying neuroprotective mechanisms remain uncertain; the latter may involve reduced vascular disease, oxidative stress, inflammation, and/or amyloid deposition.²

No effective pharmacotherapy exists to prevent cognitive decline in persons with MCI, and black individuals are twice as likely to decline cognitively, meet criteria for MCI, and develop dementia as white individuals.⁵ Black individuals also have less accurate knowledge of dementia and are less likely to be treated for it.¹² These facts highlight the need for culturally competent preventive treatments such as behavioral activation to reduce the burden of dementia and achieve health equity for everyone.

Accepted for Publication: June 21, 2018.

Corresponding Author: Barry W. Rovner, MD, Jefferson Hospital for Neuroscience, 900 Walnut St, Ste 200, Philadelphia, PA 19107 (barry.rovner@jefferson.edu).

Published Online: September 10, 2018. doi:10.1001/jamaneurol.2018.2513

Author Contributions: Dr Rovner had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: All authors.

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

Drafting of the manuscript: Rovner, Casten, Leiby.

Critical revision of the manuscript for important intellectual content: Casten, Hegel, Leiby.

Statistical analysis: Leiby.

Obtained funding: Rovner, Casten.

Administrative, technical, or material support: Rovner, Casten.

Supervision: Rovner, Casten, Hegel.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by the National Institute on Aging (grant 1R01AG035025).

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