Diagnostic Precision in the Detection of Mild Cognitive Impairment: A Comparison of Two Approaches

Abstract

OBJECTIVE:

This study compared diagnostic rates and clinical predictors of discrepancies between diagnoses conferred via: 1) a comprehensive neuropsychological evaluation and National Institute on Aging−Alzheimer’s Association (NIA-AA) criteria versus 2) a cognitive screener and Diagnostic Statistical Manual of Mental Disorders (DSM-5) criteria.

DESIGN:

Cross-sectional examination of baseline data from the Prevention of Alzheimer’s dementia (AD) using Cognitive remediation and transcranial direct current stimulation in Mild Cognitive Impairment (MCI) and Depression (PACt-MD; ClinicalTrials.gov Identifier: NCT02386670) trial.

SETTING:

Five geriatric psychiatry and memory clinics mild located at academic hospitals affiliated with the Department of Psychiatry, University of Toronto.

PARTICIPANTS:

Older adults (N = 431) with a history of major depressive disorder (MDD) in remission, MCI, or both.

MEASUREMENTS:

Main outcome was a comparison of NIA-AA diagnostic rates of MCI or dementia versus DSM-5 rates of mild or major neurocognitive disorder. Secondary analyses examined demographic, race, gender, premorbid intellectual ability, psychosocial, health-related, and genetic predictors of discrepancy between DSM-5 and NIA-AA diagnoses.

RESULTS:

There were 103 (23.8%) discrepant cases, with most (91; 88.3%) of these discrepant cases reflecting more impairment with the detailed neuropsychological testing and NIA-AA criteria. Discrepancies were more likely in individuals with a history of MDD or who had at least one ApoE4 allele.

CONCLUSION:

The NIA-AA criteria, in conjunction with comprehensive neuropsychological testing, identified a greater prevalence of cognitive impairment than DSM-5 criteria, in conjunction with the Montreal Cognitive Assessment. Detailed neuropsychological evaluations are recommended for older adults who have a history of MDD or a genetic vulnerability to dementia.

INTRODUCTION

Accurate classification of early cognitive decline is needed to inform intervention efforts to prevent or delay progression to dementia. Multiple diagnostic classification systems for detecting cognitive decline have been developed, contributing to variability in the operationalization of early cognitive decline. The term Mild Cognitive Impairment (MCI) has commonly been used to describe the intermediary stage between normal cognitive aging and dementia. Current research criteria for a clinical diagnosis of MCI are based on the 2011 National Institute of Aging and Alzheimer’s Association (NIA-AA) workgroup guidelines.¹ Criteria include 1) evidence of a concern regarding change in cognition; 2) objective evidence of mild impairment in one or more cognitive domains (1−1.5 standard deviations [SD] below normative expectations); 3) relative preservation of functional independence; and 4) no dementia. The 2013 fifth edition of the Diagnostic Statistical Manual of Mental Disorders (DSM-5) contains criteria for the diagnosis of mild neurocognitive disorder (NCD) that are similar to, but distinct from, the NIA-AA MCI criteria. A diagnosis of mild NCD requires evidence of: 1) concern regarding change in cognition; 2) objective cognitive impairment using a broader window of cognitive performance relative to NIA-AA criteria (1−2 SD below normative expectations); and 3) relative preservation of functional independence.

Diagnosis of MCI or mild NCD using a brief cognitive screening tool (e.g., Montreal Cognitive Assessment [MoCA],² Mini Mental State Examination [MMSE])³ and qualitative clinical information available from the individual and possibly an informant, (e.g., concerns regarding cognitive decline with grossly preserved functioning) is efficient and increases access to diagnosis. However, this method can result in both overdiagnosis (i.e., false-positives) and under diagnosis (i.e., false-negatives), especially when only one test is used for diagnosis.^4,5 Brief cognitive assessments are sensitive to significant impairment but may not detect early and/or subtle deficits. Brief cognitive assessments are also limited in distinguishing domains of cognitive impairment or etiology.

Comprehensive neuropsychological evaluations are considered the gold standard for the diagnosis of MCI or mild NCD and typically involve hours of testing across multiple cognitive domains. Test results are interpreted in the context of sociocultural factors including first or primary language, years of formal education, premorbid intellectual abilities, self- and informant-reports of cognitive change, functional ability, medical illness burden, medications ingested, and potential reversible causes for cognitive impairment (e.g., via blood tests or neuroimaging, when available). Comprehensive neuropsychological evaluations can help identify etiologies for cognitive impairment based on the pattern of cognitive deficits. Cultural considerations such as language of origin are imperative when interpreting assessment results, especially if tests are administered in a language other than an individual’s primary language.⁶ Psychiatric history (e.g., depression) and genetic factors (e.g., apolipoprotein E [ApoE] genotype) are now recognized as significant risk factors for cognitive decline and can inform differential diagnosis.^5,7 However, the extent to which these individual-level characteristics affect the diagnostic process is unclear.

Despite being a gold standard when assessing MCI or mild NCD, comprehensive neuropsychological evaluations are typically not available in community-based or primary care settings. In these cases, cognitive screeners provide an alternative method for detecting cognitive impairment. Thus, it is important to understand 1) discrepancy rates between approaches using NIA-AA criteria with comprehensive evaluations versus DSM-5 criteria with brief cognitive screening, and 2) whether there are subgroups for whom a comprehensive neuropsychological evaluation may be paramount because brief cognitive assessments are more likely to result in a misdiagnosis. The present analysis contrasts the results of using these two approaches of assessing cognitive impairment. We used data from the Prevention of Alzheimer’s dementia (AD) using Cognitive remediation and transcranial direct current stimulation in MCI and Depression (PACt-MD; ClinicalTrials.gov Identifier: NCT02386670).⁸ PACt-MD is an AD prevention trial in individuals with a history of major depressive disorder (MDD), MCI, or both. We hypothesized that a higher proportion of individuals would be diagnosed with MCI using a full cognitive evaluation and the NIA-AA criteria than with mild NCD using a brief cognitive screener (i.e., Montreal Cognitive Assessment (MoCA)) and DSM-5 criteria. The MoCA, rather than the Mini-Mental State Exam (MMSE), was used for diagnosis of mild NCD since the MoCA is more sensitive to mild cognitive impairment.⁹ Our second aim was to explore clinical predictors of discrepancy between these two approaches, including demographic (age, education, race, sex, premorbid intellectual ability), psychosocial (English as a primary language, history of depression), health-related (severity of physical illness burden), and genetic (ApoE e4 carrier status) factors.

METHODS

Setting

The current analysis uses PACt-MD baseline data.¹⁰ Recruitment was conducted in geriatric psychiatry and memory clinics located at five academic hospitals in Toronto affiliated with the Department of Psychiatry, University of Toronto: Baycrest Health Sciences Center; Center for Addiction and Mental Health (CAMH; coordinating site); Saint Michael’s Hospital; Sunnybrook Health Sciences Center; and University Health Network. Eligibility criteria included: Age ≥ 60 if diagnosed with MCI or age ≥ 65 if diagnosed with a history of MDD in remission (with or without MCI); Montgomery-Asberg Depression Rating Scale (MADRS) ¹¹ score ≤ 10; availability of a study partner who has regular contact with the participant; ability to read and communicate in English. The goal was to recruit 500 participants in anticipation of enrolling 375 for the trial, as well as up to 200 healthy controls.

Screening, Clinical Assessment or Diagnosis, Referral, and Enrollment

At screening, a geriatric psychiatrist completed a clinical assessment including a psychiatric history, an interview, and a mental status examination. All patients were administered the MMSE and MoCA; some also completed a clinical neuropsychological assessment - e.g., the Toronto Cognitive Assessment (TorCA).¹² Patients had to score 24 or higher on the MMSE (unless an exception was granted by a committee comprising four geriatric psychiatrists, for instance in the case of a functionally independent participant with very low education).

All participants provided written informed consent in accordance with the procedures of the CAMH Research Ethics Board. Eligibility was confirmed after completion of baseline assessment including the MADRS and Clinical Dementia Rating (CDR) Scale.¹³ Eligible participants were assigned to one of three groups based on their initial clinical diagnosis (made by the geriatric psychiatrist), their MoCA score, or a brief clinical neuropsychological assessment such as the TorCA: 1) cognitively normal with history of MDD; 2) MCI and history of MDD; 3) MCI and no history of MDD. For a clinical diagnosis of MCI, a MoCA score of 26 or below was required except in the case of 33 participants for whom the committee granted an exception. These exceptions were granted for cases where a participant had a very high level of education, high level of professional attainment or general functioning, or a participant showing cognitive deficits using another cognitive scale. In these cases, concern regarding cognitive decline was expressed by the participant, informant, or medical provider, and was essential for the clinical diagnosis of MCI.

Research Neuropsychological Assessment

All participants completed a comprehensive neuropsychological test battery consisting of standardized clinical measures that have established reliability and are sensitive to cognitive changes in both MCI and MDD. The battery assessed the following five DSM-5 cognitive domains: Attention or speed of processing (WAIS-III Digit Symbol Substitution Test;¹⁴ Trail Making Test Part A;¹⁵ computerized Paced Auditory Serial Addition Test,¹⁶ Continuous Performance Test, N-Back Task, executive functioning (clock drawing; Trail Making Test Part B;^15,17 Stroop Neuropsychological Screening Test)¹⁸, verbal memory (California Verbal Learning Test, Second Edition)¹⁸, visual memory (Brief Visual Memory Test, Revised)¹⁹, language (Boston Naming Test;²⁰ Category Fluency; Letter Fluency²¹), and visuospatial processing (Judgment of Line Orientation,²² Brief Visual Memory Test, Revised, copy condition).¹⁹ Premorbid verbal intellectual ability was estimated by the Wide Range Achievement Test − 4th edition Reading Subtest (WRAT-Reading).²³ Concern regarding cognitive decline was established quantitatively using participant and informant reports of the everyday cognition questionnaire (E-COG), based on established cut-off scores.²⁴ Independence in everyday activities was evaluated quantitatively using select tasks from the Performance Assessment of Self-care Skills (PASS) (shopping, bill paying, checkbook balancing, mailing, medication management) that have been shown to discriminate between normal cognition and MCI.²⁵

Other Demographic, Psychosocial, and Health-Related Assessments

We recorded age, education, race, ethnicity, occupational history, and languages spoken. Psychiatric diagnoses were characterized with the Structured Clinical Interview for DSM-IV (SCID).²⁶ Physical illness burden was measured with the Cumulative Illness Rating Scale-Geriatric (CIRS-G).²⁷ ApoE genotype was determined using a combination of variants rs7412 and rs429358 that were genotyped at CAMH using standard TaqMan protocols (Life Technologies, Burlington, ON) in accordance with manufacturer’s directions. ApoE status was unknown at the time of the consensus conference.

Consensus Conference Procedure to Confer Diagnoses Based on the NIA-AA or DSM-5 Criteria

Weekly consensus conference calls were conducted to adjudicate research diagnoses. These calls included the study geriatric psychiatrist and the research associate who completed the clinical and research assessments; at least two of four senior geriatric psychiatrists; a senior neuropsychologist; and the study coordinator.

First, the research associate summarized the clinical, (including medical history and current medication list) as well as MADRS and CDR scores. Then, the study coordinator queried the study geriatric psychiatrist about DSM-5 diagnostic criteria for mild NCD, starting with criteria (A1) “concerns of the individual, knowledgeable informant, or the clinician that there has been a mild decline in cognitive function”. Next, the research associate reported the MoCA total score, where points had been lost, and the corresponding impaired DSM-5 cognitive domains.

Next, the neuropsychologist described the results of the neuropsychological tests, E-COG questionnaire, and PASS and made a neuropsychological diagnosis based on the 2011 NIA-AA recommendations for the classification of normal cognitive status, MCI (“NIA-AA diagnosis”), or dementia. The neuropsychologist made a NIA-AA diagnosis of MCI based on the following criteria: 1) concern regarding a change in cognition indicated by the E-COG or clinical concerns from the study geriatric psychiatrist; 2) objective evidence of cognitive impairment in one or more cognitive domains, indicated by performance at 1−1.5 SD below expectation based on culturally appropriate norms on either two tests within the same cognitive domain or three tests scattered across cognitive domains; 3) relatively preserved functional independence, with at most mild difficulty on cognitively challenging IADLs (C-IADLs), indicated by performance on the PASS C-IADL subtests (i.e., holistic assessment of independence, adequacy, and level of assistance required for completing C-IADLs). Participants who did not have a NIA-AA diagnosis of MCI or dementia received a NIA-AA classification of “normal cognitive status.”

After the NIA-AA diagnosis was established, the study geriatric psychiatrist reviewed whether the participant met DSM-5 criteria A2 and B-D for mild NCD: (A2) “a modest impairment in cognitive performance [. . .] documented by [. . .] quantified clinical assessment” (in this case, the MoCA); (B) “the cognitive deficits do not interfere with capacity for independence in everyday activities”; (C) “the cognitive deficits do not occur exclusively in the context of a delirium”; and (D) “the cognitive deficits are not better explained by another mental disorder”. If these criteria were met, the participant was given a DSM-5 diagnosis of mild NCD. Otherwise, the participant was classified as either having a normal cognitive status or a diagnosis of major NCD based on the DSM-5 diagnostic criteria (considering the cognitive screening results). Rare exceptions to these rules were made. For instance, participants with a very high level of education (e.g., Master’s degree) or high occupational attainment (e.g., company CEO) were assumed to have a high level of cognitive reserve; thus, all 33 participants with a MoCA score between 27 and 29 who had been given a clinical diagnosis of MCI were given a DSM-5 diagnosis of mild NCD (rather than normal cognitive status).

The goal of the present study was to compare cognitive impairment diagnostic rates using DSM-5 and NIA-AA approaches. Since the two diagnoses were both completed during the consensus conference, thus not completely independent, we excluded the following cases where DSM-5 diagnosis of MCI may have been influenced by the results of the comprehensive neuropsychological testing: 1) participants who received a clinical diagnosis of history of MDD but no MCI at study entry but then received a DSM-5 diagnosis of MCI during the consensus conference (N = 21); or 2) participants with a MoCA score of 30 (N = 3), as this “perfect score” could not be used to establish “a modest impairment in cognitive performance [. . .] documented by [. . .] quantified clinical assessment.”

Statistical Analysis

We used baseline data of 431 participants who met study criteria and completed neuropsychological testing (See Fig. 1). We compared NIA-AA diagnostic rates of MCI or dementia versus DSM-5 rates of mild or major NCD. We then used multivariate binary logistic regression models to examine demographic (age, education [<12 years, 12−15 years, 16 years, 17 + years], race [White, African descent, Asian descent, Other], gender [man, woman]), premorbid intellectual ability (WRAT reading raw score), psychosocial (English as a primary language, history of depression), health-related (burden of physical illness; CIRS-G), and genetic (ApoE e4 carrier status) predictors of discrepancy between DSM-5 and NIA-AA diagnoses. These characteristics were chosen due to established associations with cognitive function⁶. We did not examine the PASS, E-COG, or cognition as predictors, since these variables were used in the diagnostic process. Analyses were conducted separately for two groups of participants for whom there was a diagnostic discrepancy: 1) those for whom the NIA-AA diagnosis reflected more cognitive impairment than DSM-5 diagnosis; and 2) those for whom the NIA-AA diagnosis reflected less cognitive impairment than DSM-5 diagnosis.

FIGURE 1.

Flow chart of process resulting in final study sample. CC: consensus conference; MoCA: montreal cognitive assessment; MDD: major depressive disorder; MCI: mild cognitive impairment

RESULTS

Participant characteristics (N = 431) are displayed in Table 1, and Table 2 shows the results from the two diagnostic approaches. The two approaches were congruent in 328 (76.1%) participants (89 with “normal cognitive status”; 234 with MCI or mild NCD; 5 with dementia or major NCD). Cohen’s κ was run to determine if there was agreement between these diagnostic approaches. There was only moderate agreement between the NIA-AA diagnostic approach and DSM-5 diagnostic approach, κ = .507 (95% CI, .429−.585), p<0.0005. When there was discrepancy, the NIA-AA diagnoses most frequently reflected greater cognitive impairment than DSM-5 diagnoses (McNemar’s χ² (2,431) = 64.290, p<0.0005): of the 103 (23.8%) cases of discrepant diagnoses, 50 had a MCI NIA-AA diagnosis and no DSM-5 diagnosis (i.e., normal cognitive status); 41 had a dementia NIA-AA diagnosis and a mild NCD DSM-5 diagnosis; 12 had no NIA-AA diagnosis (i.e., normal cognitive status) and a DSM-5 diagnosis of mild NCD.

TABLE 1.

Participant Characteristics for Pact-MD Participants Included in Analysis (N = 431). Numbers Indicate Mean (Standard Deviation) Or Percent Frequency. Data Were Not Available from All Participants for Some Measure; Available Sample Sizes are Indicated in Parentheses If It Differs from the Overall Sample Size.

Characteristic	N=431
Age (years)	71.13 (6.42)
Sex (% female)	62.6%
Race (% White)	78.2% (N = 429)
Education (% at least 4-year college graduate)	74.2%
CIRS-G (total score)	4.63 (3.12) (N = 414)
ApoE4 carrier	23.4% (N = 326)
History of a major depressive disorder	37.6%
MADRS total score	3.74 (3.07)
English as primary language	81.9% (N = 428)
WRAT Reading raw score	63.50 (5.03) (N = 428)
MoCA total score	24.72 (3.49) (N = 427)

MoCA: montreal cognitive assessment; WRAT: wide range achievement test – 4^th edition; ApoE4: apolipoprotein epsilon 4 allele carrier; CIRS-G: cumulative illness rating scale- geriatric; MADRS: montgomery-asberg depression rating scale

TABLE 2.

Cognitive Diagnoses Yielded by A Comprehensive Neuropsychological Evaluation and NIA-AA Criteria Versus Moca and DSM-5 Criteria (N = 431)

		DSM-5 Classification
		Normal Cognitive Status	Mild NCD	Major NCD
NIA-AA Classification	Normal cognitive status	89	12	0
	MCI	50	234	0
	Dementia	0	41	5

Numbers in cells indicated frequency of diagnosis. NIA-AA: National Institute of Aging and Alzheimer’s Association 2011 workgroup guidelines; DSM-5: Diagnostic and Statistical Manual of Mental Disorders criteria; MoCA: Montreal Cognitive Assessment; MCI: Mild Cognitive Impairment; NCD: neurocognitive disorder.

When examining predictors of diagnostic discrepancies (N = 308; 52 discrepant cases with complete data), binary logistic regression revealed that cases in which NIA-AA diagnosis reflected worse cognitive function than DSM-5 diagnosis (χ² (13) = 31.097, p = 0.003) were slightly older (73 years versus 72 years), had a history of MDD (60% versus 40%), and were more likely to be an ApoE4 carrier (50% versus27%) (Table 3). MDD history and ApoE4 status were not significantly associated (χ²(1,326)=0.151, p = 0.697); other variables did not predict these diagnostic discrepancies (Table 3). For this binary logistic regression model examining predictors of cases in which NIA-AA diagnosis reflected worse cognitive function than DSM-5 diagnosis, sensitivity was 5.8%, specificity was 98.8%, positive predictive value (PPV) was 50% and negative predictive value (NPV) was 83.8%. There were no variables predictive of cases for which DSM-5 diagnosis reflected greater cognitive impairment relative to NIA-AA (χ² (13) = 15.034, p = 0.305) (Table 4).

TABLE 3.

Logistic Regression Using Participant Characteristics to Predict Cases for Whom the Comprehensive Neuropsychological Evaluation and NIA-AA Criteria Yielded A Diagnosis Reflecting A More Severe Cognitive Impairment Than the Moca and DSM-5 Criteria (N = 52)

Characteristic	df	B	SE	Wald χ2	Odds Ratio	p-value
Age	1	0.062	0.027	5.268	1.064	0.022
Sex	1	0.055	0.347	0.025	1.057	0.873
Race	3
White		ref	ref	ref	ref	ref
African descent		−0.841	0.827	1.034	0.431	0.309
Asian descent		−0.351	0.861	0.166	0.704	0.684
Other		−0.494	0.702	0.495	0.610	0.482
Education	3
Less than high school		ref	ref	ref	ref	ref
High school and/or some college		−0.484	0.687	0.496	0.617	0.481
College graduate		−0.480	0.660	0.528	0.619	0.468
Graduate level education		−0.014	0.699	0.000	0.986	0.984
WRAT reading raw score	1	−0.057	0.037	2.363	0.945	0.124
English as primary language	1	0.565	0.617	.841	1.760	0.359
ApoE4 carrier	1	1.175	0.349	11.310	3.238	0.001
CIRS-G total score	1	−0.041	0.056	0.542	0.690	0.462
History of MDD	1	1.208	0.357	11.415	3.346	0.001

Binary logistic regression results examining participant characteristics that were associated with a case receiving a more severe cognitive diagnosis when using the NIA-AA criteria than when using DSM-5 criteria. NIA-AA: national institute of aging and alzheimer’s association 2011 workgroup guidelines; DSM-5: diagnostic and statistical manual of mental disorders criteria; df: degrees of freedom; SE : standard error; MoCA: montreal cognitive assessment; WRAT: Wide Range Achievement test – 4^th edition; ApoE4: apolipoprotein epsilon 4 allele carrier; CIRS-G: cumulative illness rating scale- geriatric

TABLE 4.

Logistic Regression Using Participant Characteristics to Predict Cases for Whom the Comprehensive Neuropsychological Evaluation and NIA-AA Criteria Yielded A Diagnosis Reflecting A Less Severe Cognitive Impairment Than the MoCA and DSM-5 Criteria (N = 12)

Characteristic (df)	df	B	SE	Wald χ2	Odds Ratio	p-value
Age	1	−0.084	0.057	2.143	0.921	0.143
Sex	1	0.268	0.672	0.159	1.248	0.690
Race	3
White		ref	ref	ref	ref	ref
African descent		−18.006	8802.561	0.000	0.000	0.998
Asian descent		0.835	0.970	0.742	2.051	0.389
Other		0.411	1.142	0.130	1.519	0.719
Education	3
Less than high school		ref	ref	ref	ref	ref
High school and/or some college		−0.109	1.367	0.006	0.900	0.937
College graduate		−0.492	1.366	0.130	0.609	0.719
Graduate level education		−0.857	1.521	0.318	0.393	0.573
WRAT reading raw score	1	0.006	0.062	0.009	1.005	0.926
English as primary language	1	1.598	1.176	1.845	4.604	0.174
ApoE4 carrier	1	−0.208	0.734	0.080	0.829	0.777
CIRSG total health burden	1	0.015	0.119	0.015	1.022	0.902
History of depressive episode	1	−2.260	1.075	4.419	0.089	0.036

Binary logistic regression results examining participant characteristics that were associated with a case receiving a more severe cognitive diagnosis when using DSM-5 criteria than when using the NIA-AA criteria. NIA-AA: national institute of aging and alzheimer’s association 2011 workgroup guidelines; DSM-5: diagnostic and statistical manual of mental disorders criteria; df: degrees of freedom; SE : standard error; MoCA: montreal cognitive assessment; WRAT: wide range achievement test – 4^th edition; ApoE4: apolipoprotein epsilon 4 allele carrier; CIRS-G: cumulative illness rating scale- geriatric

Since DNA samples were not available or genotype could not be obtained for 105 participants, we re-examined predictors of diagnostic discrepancies using participants (N = 406; 83 discrepant cases) who had complete data excluding APOE status. In this sub-analysis, cases in which NIA-AA diagnosis reflected worse cognitive function than DSM-5 diagnosis were more likely to have a history of MDD (64% versus36%; B = 0.588, Odds Ratio [OR] = 1.80, Wald χ² (1) = 4.994, p = 0.02; sensitivity = 0%, specificity = 100%, PPV = 0% and NPV = 79.6%).

DISCUSSION

We found a 23.9% discrepancy rate between two diagnostic classification approaches, one using a comprehensive neuropsychological evaluation and NIA-AA criteria, the other using a brief cognitive screener and DSM-5 criteria. The neuropsychological evaluation and NIA-AA criteria more frequently yielded diagnoses reflecting greater cognitive impairment. Diagnostic discrepancy in these cases was predicted by history of MDD, ApoE4 status, and older age, although the difference in age between discrepant and non−discrepant cases was only one year and may not be clinically meaningful. While no variables were indicative of cases for which DSM-5 diagnosis reflected greater cognitive impairment relative to NIA-AA, it should be noted that this pattern only occurred in a small number of cases, thereby limiting the reliability of this analysis.

If we consider the comprehensive neuropsychological evaluation as a “gold standard” for cognitive diagnosis, a diagnostic approach relying primarily on a cognitive screening tool and clinical history from the participant and an informant resulted in missed diagnosis of MCI in 11% of cases and missed diagnosis of dementia in 9% of cases. Nearly half of the discrepant cases (N = 41) received a diagnosis of dementia with a comprehensive evaluation and NIA-AA criteria (our “gold standard”), while they received a diagnosis of mild NCD with a cognitive screener and DSM-5 guidelines. While this discrepancy does not invalidate the utility of cognitive screeners in the diagnosis of cognitive impairment, it does highlight the need for in-depth evaluations for individuals whose abilities are not as well captured on cognitive screeners (e.g., people with higher cognitive reserve). Misclassification is meaningful, given the serious immediate and long-term clinical implications of a dementia diagnosis.

Our results are consistent with previous research showing that assessing cognition with multiple measures allows for more sensitive characterization of cognitive impairment, especially in individuals with mild impairment ^4,7. This is particularly important to consider when comparing the range of cognitive impairment used in the NIA-AA (1−1.5 SD below expectation) versus DSM-5 (1−2 SD below expectation). Brodaty and colleagues (2017) found that diagnosing cognitive impairment with two tests per domain with a 1-SD cutoff or one test per domain with a 1.5-SD cutoff predicted similar progression to dementia over 6 years in a community sample. One single measure of one cognitive domain was only predictive at more severe levels of impairment (>1.5 SDs), suggesting that using DSM-5 range of −1 − 2 SD may lead to false negatives when using a summary score from a single cognitive test, such as the MoCA.

Many communities do not have access to geriatric neuropsychologists, so identifying individuals who would most benefit from a comprehensive neuropsychological evaluation is critical. We identified two potential characteristics for identifying patients for whom a comprehensive neuropsychological assessment may identify cognitive impairment missed by a cognitive screener: A history of MDD or having an ApoE4 allele. Clinicians can readily assess history of MDD, which is already a well-established risk factor for cognitive decline, even after symptoms remission.²⁸ MDD both in earlier life and in late-life, even post−remission, are associated with at least a two-fold increased risk for dementia.²⁹ Executive impairment is consistently observed in most individuals with a history of MDD even after depression is successfully treated.²⁸ Executive impairment can be subtle and difficult to detect using cognitive screeners; thus, comprehensive neuropsychological evaluations may be critical for capturing cognitive deficits in older adults with a history of MDD. While determination of ApoE4 status is not a part of routine healthcare, clinicians may use proxies such as family history of AD to determine who could benefit from a neuropsychological referral.³⁰

Our study was conducted in the greater Toronto area and our sample was predominantly white, female, and highly educated, which limits generalizability of these results to other sociodemographic groups. In addition, we do not know whether variability in sensorimotor acuity, which may impact cognitive performance, was predictive of diagnostic discrepancy. Further, this study was unable to address likely cognitive etiologies given the lack of biomarker data. In addition, while the two diagnostic procedures were completed separately, they were both conducted within the same consensus conference and thus the diagnosing clinicians were not blinded. The parent study recruited participants with a history of MDD, current MCI, or both, which may have introduced systematic bias in the importance of MDD as a predictor. However, this oversampling allows for greater generalizability of results since MDD is a leading cause of disability worldwide.³¹ In addition, our analyses of participant characteristics that predicted diagnostic discrepancy were exploratory in nature and are especially susceptible to Type 1 error due to the small sample size (at least 50% chance per discrepancy calculation). Future studies powered to examine predictors of diagnostic discrepancy will be needed to follow-up on these findings. Finally, this study compared DSM-5 criteria for cognitive impairment to 2011 NIA-AA criteria; however, the 2011 criteria were updated for defining and staging AD across the diagnostic spectrum.³² These updated criteria defined MCI as cognitive performance below expectation based on clinical judgment and/or neuropsychological test performance that may be adjusted by normative comparison. This updated definition is closely aligned with DSM-5 classification. However, the 2018 NIA-AA framework focuses on a biologic definition of AD through presence and grouping of beta-amyloid plaques, presence of fibrillar tau, and neurodegeneration. The 2018 NIA-AA framework specifically states that it is not intended for clinical practice, did not consider clinical criteria, and did not provide guidelines for diagnosis; as such, we chose to use the original 2011 guidelines as a clinical comparison to DSM-5.

Despite these limitations, our study is one of the few that was designed to assess for discrepancy between diagnostic classification approaches. Many studies of cognitive impairment exclude participants with a psychiatric history; we directly addressed the importance of a history of MDD on the diagnostic process. Since MDD is a strong risk factor for dementia, understanding how it impacts the diagnostic process can inform clinicians on whom to refer for specialty evaluations. Finally, we analyzed only baseline data of a clinical trial. In the future, analysis of follow-up data can examine which diagnostic approach is more closely associated with cognitive decline. Our study highlights the discrepancy that occurs when using different diagnostic criteria to identify cognitive impairment. These results also speak to the need for more in-depth cognitive assessments and consideration of a broad array of clinical characteristics when diagnosing cognitive impairment in older adults.

Author contributions: Weinstein, Gujral: Study concept and design, analysis and interpretation of data, preparation of manuscript. Butters: Study concept and design, interpretation of data, preparation of manuscript. Bowie: Study concept and design, interpretation of data, critical review of manuscript. Fischer, Flint, Kennedy, Mah, Pollock: Interpretation of data, critical review of manuscript. Herrmann: Study concept and design, interpretation of data, critical review of manuscript. Ovaysikia: Acquisition of participants and data, critical review of manuscript. Rajji, Mulsant: Study concept and design, acquisition of participants and data, interpretation of data, critical review of manuscript.

Highlights.

• The primary question addressed by this study: This study compared diagnostic rates and clinical predictors of discrepancies between diagnoses conferred via: 1) a comprehensive neuropsychological evaluation and National Institute on Aging−Alzheimer’s Association (NIA-AA) criteria versus 2) a cognitive screener and Diagnostic Statistical Manual of Mental Disorders (DSM-5) criteria.

• The main finding of this study: There were 103 (23.8%) discrepant cases, with most (91; 88.3%) of these discrepant cases reflecting more impairment with the detailed neuropsychological testing and NIA-AA criteria. Discrepancies were more likely in individuals with a history of major depressive disorder (MDD) or who had at least one ApoE4 allele.

• The meaning of the finding: The NIA-AA criteria, in conjunction with comprehensive neuropsychological testing, identified a greater prevalence of cognitive impairment than DSM-5 criteria, in conjunction with the Montreal Cognitive Assessment. Detailed neuropsychological evaluations are recommended for older adults who have a history of MDD or a genetic vulnerability to dementia.