Overview

How accurate are brief cognitive tests for identifying clinical Alzheimer’s-type dementia (CATD) in individuals with suspected cognitive impairment? Brief cognitive tests alone cannot diagnose CATD. However, this review found that among individuals with suspected cognitive impairment, many of these tests are highly sensitive and specific for case finding, and appear more accurate for distinguishing CATD from normal cognition than for distinguishing CATD from mild cognitive impairment (MCI) or distinguishing mild CATD from normal cognition. Therefore, these tests may help providers decide which patients warrant a more comprehensive diagnostic evaluation. Brief cognitive tests may also provide a less burdensome way than comprehensive neuropsychological testing to objectively document cognitive impairment when patients have already undergone an appropriate cognitive history and have evidence of functional impairment.

Which brief cognitive tests, test combinations, and test cut points are best for distinguishing between CATD, MCI, and normal cognition in individuals with suspected cognitive impairment? We found few direct comparisons of different tests and cut points within studies, and different studies rarely compared the same test cut points. However, results from this review suggest that in individuals with suspected cognitive impairment, brief instruments commonly used as individual stand-alone tests, brief multidomain batteries, and, among domain-level tests typically part of a larger battery, memory, and verbal fluency tests may have the most potential for distinguishing between CATD, normal cognition and MCI. Within these test categories, the individual tests with the most available accuracy data were clock drawing and Mini-Mental Status Exam (MMSE) among the tests commonly used as individual stand-alone tests, Dementia Rating Scale (DRS) for brief multidomain batteries, list learning delayed recall and retention for memory tests, and semantic (category) fluency for language. However, the optimal version and cut point for each of these tests is uncertain. Test combinations may be more accurate for distinguishing between diagnostic categories than individual tests, but the additional administration time may not be feasible in primary care settings.

What is the evidence that brief, web-based cognitive tests can accurately distinguish between CATD, MCI, and normal cognition? Our review identified no eligible studies that evaluated the classification accuracy of web-based cognitive testing, alone or in comparison to a different testing approach. Though the Cogstate Brief Battery evaluated in this review is available for web-based administration, the web version was not used in the reported study.

Does the classification accuracy of brief cognitive tests for CATD vary based on participant characteristics, and should different tests and cut points be used in different individuals? In clinical practice, cognitive tests administered as part of a neuropsychological test battery are normed for age, gender, and education. Norms for brief cognitive tests also vary by these factors. However, the small set of studies that examined whether classification accuracy of brief cognitive tests varies by participant characteristics reported that accuracy did not vary by age, sex, race/ethnicity, or education. These studies were likely too small to rule out such associations and inadequate for concluding whether or not different cut points should be used for classifying between CATD and normal cognition or MCI in different clinical populations. By comparison, sensitivity was consistently high for distinguishing participants with moderate CATD from normal cognition, and lower for distinguishing between mild CATD and normal cognition. These findings suggest that brief cognitive testing may be sufficient to categorize more moderately impaired patients, but insufficient to categorize more mildly impaired patients. For the latter group, inconclusive brief cognitive testing may warrant followup with more comprehensive neuropsychological testing and possibly specialty referral.

Are there harms of brief cognitive testing for CATD? We identified no studies that reported data on harms of brief cognitive testing for CATD. We also found no such data in prior systematic reviews of brief cognitive testing for CATD published since 2013. Nevertheless, potential harms following brief cognitive testing that incorrectly classifies someone as having CATD (false positive) include stigma from being incorrectly labeled with CATD and harms of unnecessary interventions (e.g., restrictions on independence, physical or psychological harms of further diagnostic testing, medication adverse effects). Potential harms following brief cognitive testing that incorrectly fails to identify CATD (false negative) include psychological distress from unexplained symptoms and lack of appropriate interventions (e.g., future planning, taking safety precautions, initiation of medications).

In individuals with CATD, which brain imaging test, cerebrospinal fluid (CSF) test, or combination is most accurate for distinguishing Alzheimer’s disease (AD) from non-AD dementia? Our review could not determine which brain imaging test is most accurate for distinguishing neuropathologically-confirmed AD from non-AD dementia, or from individual types of non-AD dementia, because the only study that compared different brain imaging tests (amyloid positron emission tomography [PET] plus computed tomography [CT] versus amyloid PET alone) reported no data.⁴ However, single studies suggested that amyloid PET and fluorodeoxyglucose (FDG)-PET may increase classification accuracy when added to clinical evaluation compared with clinical evaluation only. By comparison, in two studies, single-photon emission computed tomography (SPECT) added to clinical evaluation had lower sensitivity and higher specificity than clinical evaluation alone, and no studies compared magnetic resonance imaging (MRI) medial temporal atrophy (MTA) plus clinical evaluation versus clinical evaluation alone.

Evidence regarding whether certain individual CSF tests or combinations of CSF tests are better than others for distinguishing neuropathologically-confirmed AD from non-AD is inconclusive. Three eligible studies directly compared accuracy of selected CSF biomarkers. Collectively, they suggested that abnormally phosphorylated tau (p-tau), the Aβ42/p-tau ratio, and the t-tau/Aβ42 ratio may be more accurate and beta amyloid 42 (Aβ42) and total tau (t-tau) may be less accurate. Analyses comparing CSF tests for distinguishing between AD and individual types of non-AD showed different patterns. However, all cut points for defining abnormal versus normal CSF biomarker levels were specific to individual studies and virtually none were validated in other studies.

A 2018 systematic review reported much greater diagnostic accuracy for Aβ42, Aβ42/p-tau, and Aβ42/t-tau ratio.¹⁷ That review was conducted to provide the evidence base to support Alzheimer’s Association appropriateness criteria on the accuracy of CSF testing for AD. Using an autopsy reference, authors reported sensitivity ranging from 0.89 to 0.92 and specificity ranging from 0.82 to 0.88. However, their analysis focused on the ability of CSF markers to distinguish AD from patients without dementia, for which testing performance would be expected to be higher. Our analyses, in contrast, examined the ability of CSF markers to distinguish AD from non-AD dementias in patients with CATD.

In patients with CATD, how likely are brain imaging or CSF tests to add to a clinical evaluation by correcting an incorrect clinical diagnosis between AD and non-AD dementia? In two studies, SPECT plus clinical evaluation had lower sensitivity and higher specificity than clinical evaluation alone. By comparison, in one study (with AD prevalence of 74 percent and clinical diagnostic accuracy for distinguishing AD from non-AD dementia of 80 percent), amyloid PET increased diagnostic accuracy to 98 percent by correctly reclassifying 92 percent of clinical false negatives, while falsely reclassifying 4 percent of clinical true positives. In a second study (with AD prevalence of 57 percent, and clinical diagnostic accuracy for distinguishing between AD and non-AD of 68 percent), FDG-PET increased diagnostic accuracy to 80 percent, due to small reductions in both clinical false negatives and clinical false positives. This second study showed that FDG-PET also modestly increased accuracy distinguishing between AD and frontotemporal lobar degeneration (FTLD) when added to a clinical evaluation. However, the gains in diagnostic accuracy with amyloid PET and FDG-PET reported in research studies may not be achievable in clinical settings. One reason is that the high prevalence of AD in research studies may have increased diagnostic vigilance and test sensitivity. Second, in clinical settings, both individuals with and without AD dementia may be more likely to have multiple etiologies and thus be harder to distinguish. Third, it is not likely that image interpretation in clinical settings will be able to rely on the consensus of multiple readers (five in the amyloid PET study and two in the FDG-PET study).

The recent IDEAS study on amyloid PET brain imaging was not included in this review because biomarker testing was not compared with a neuropathological reference standard.³⁸⁴ Nevertheless, its findings are pertinent to the discussion about the feasibility of testing outside of research settings and the potential for test results to change clinical diagnoses and subsequent clinical management. IDEAS enrolled 11,409 Medicare beneficiaries with dementia or MCI of uncertain etiology after they had completed a comprehensive evaluation by a dementia specialist (most based in private practice). For eligibility, AD must have been a diagnostic consideration and knowledge of amyloid PET status must have been expected to alter diagnosis and management. Amyloid PET imaging was performed in accredited imaging facilities and scans were interpreted by imaging specialists following approved reading methodologies. Prior to imaging, AD was the leading suspected etiology in 83 percent of participants with dementia and 73 percent of those with MCI, and 59 percent of participants with dementia were taking AD drugs compared with 35 percent of those with MCI. Amyloid PET results were rated positive in 70.1 percent of participants with dementia and 55.3 percent of those with MCI. Among individuals with previously suspected AD, based on amyloid PET imaging, the suspected etiological diagnosis changed to non-AD in 32.6 percent (2,860 of 8,770), whereas among individuals with previously suspected non-AD, the suspected etiological diagnosis changed to AD in 45.5 percent (1,201 of 2,639). Reclassification results were not reported separately for participants with dementia versus MCI. Results also were not compared to a neuropathological reference. Clinical management (use of AD drugs, use of non-AD drugs, or counseling about safety or future planning) changed within 90 days after amyloid PET in 63.5 percent of participants with dementia and 60.2 percent of participants with MCI. These changes included large increases in AD drug use after positive amyloid PET results in both patients with AD and MCI, and small reductions in AD drug use after negative amyloid PET results. However, the ability to attribute clinical management changes to amyloid PET results was limited by the lack of a non-PET control group and inclusion only of participants for whom amyloid PET results were expected to alter management. In addition, clinical management was not evidence-based (many patients with MCI or with negative amyloid PET results were treated with AD drugs) and the association of amyloid PET evaluation with clinical outcomes was not examined.

By comparison, SPECT combined with clinical evaluation in two studies did not appear to increase classification accuracy compared with clinical evaluation alone. Further, in one study, clinical evaluation alone and MRI medial temporal atrophy alone had similar diagnostic accuracy and no studies directly compared MRI plus clinical evaluation versus clinical evaluation alone.

Only one eligible study reported data on the diagnostic accuracy of CSF testing added to clinical evaluation compared with clinical evaluation alone. In this study, clinical evaluation had a sensitivity of 0.80 and specificity of 0.80 for distinguishing between neuropathologically-confirmed AD and FTLD. However, a regression model including CSF Aβ42 and p-tau levels measured using a Luminex assay had a sensitivity of 0.98 and specificity of 0.93. CSF testing improved accuracy primarily by reclassifying participants who were incorrectly classified by clinical evaluation as having FTLD. To a lesser extent, CSF testing improved accuracy by reclassifying participants who were incorrectly classified by clinical evaluation as having AD. However, CSF testing also falsely reclassified 3 percent of participants who were correctly classified by clinical evaluation. Authors also created a regression model of CSF markers using results from an enzyme-linked immunosorbent assay (ELISA), but did not report how this affected classification accuracy compared with clinical evaluation.

Does the classification accuracy of brain imaging or CSF tests for AD vary based on patient characteristics, and should different tests and cut points be used in different populations? Available data appears inadequate to guide targeting of different brain imaging or CSF tests, or use of different test cut points for classification of patients between AD and non-AD dementias in different populations. We found little information in eligible studies about whether accuracy of brain imaging or CSF tests for distinguishing autopsy-confirmed AD from non-AD varies based on study participant characteristics, including by age, race/ethnicity, sex or CATD severity. One study reported similar sensitivity and specificity for FDG-PET between participants with milder versus more severe cognitive impairment. A second study reported that SPECT diagnostic accuracy did not vary by age at dementia symptom onset, MMSE score, disease duration, or interval between imaging and death. However, no studies reported tests of interaction for potential subgroup differences, or whether accuracy of any brain imaging test differed by other participant characteristics. For CSF testing, one study reported that classification accuracy of t-tau levels for distinguishing autopsy-confirmed AD from non-AD did not vary by age or sex. Otherwise, no studies reported on whether classification accuracy of brain imaging or CSF tests or the optimal test cut points vary by participant characteristics.

In patients with CATD, how likely are prescription drugs to prevent worsening of cognition and function, how likely are they to improve these outcomes, and how long will any benefits last? Trials of about 6 months showed benefits for cholinesterase inhibitors compared with placebo regardless of baseline CATD severity. However, average differences for cognition and function between treatment groups were small, with standardized mean differences generally between 0.20 to 0.40 for cognition and about 0.20 for function. Responder analyses showed that compared with placebo, for approximately every 5 to 9 participants assigned cholinesterase inhibitors, one additional individual was stable at 6 months on a brief multidomain cognitive battery (ADAS-Cog) or a global change measure (Clinician’s Interview-Based Impression of Change with caregiver input [CIBIC-Plus] or Clinical Global Impression of Improvement [CGIC]). Further, for approximately every 5 to 13 participants assigned cholinesterase inhibitors compared with placebo, one additional individual was improved at 6 months on ADAS-Cog (≥4-point improvement), CIBIC-Plus or CGIC. Whether improvements meeting these thresholds are clinically meaningful is unclear. Data on moderate or marked improvement for cognition or function were not reported and moderate or marked improvement on the global change measure was rare and no more likely with cholinesterase inhibitor treatment than placebo. Because no eligible cholinesterase inhibitor trials with low or medium risk of bias reported efficacy outcomes beyond 6 months, it is uncertain whether any benefits compared with placebo are sustained beyond this duration. For memantine, one trial suggested that compared with placebo, for approximately every six participants who were taking a cholinesterase inhibitor and were assigned add-on memantine compared with placebo, one additional individual improved on a global measure of change. However, this trial did not report responder analyses for cognition or function and no other eligible memantine trials reported responder analyses for any efficacy outcomes. Similar to the cholinesterase inhibitor results, it is unclear whether the 6-month improvement in this single outcome compared with placebo is clinically meaningful and whether there are any sustained benefits beyond 6 months.

In patients with CATD, how likely are supplements to prevent worsening of cognition and function, how likely are they to improve these outcomes, and how long will any benefits last? Available evidence was mostly insufficient about whether supplements are more effective or safe than placebo for treatment of patients with CATD, let alone about the magnitude or duration of any benefits. Among eligible trials with low or medium risk of bias, compared with placebo, omega-3 fatty acids did not improve cognition and the nutritional drink Souvenaid^® did not improve function. Although single trials showed statistically significant benefit for at least one cognitive outcome for melatonin, choline alfoscerate, and the combination of omega-3 fatty acid and alpha lipoic acid, respectively, strength of evidence for these three treatments was insufficient. We found additional eligible trials that compared gingko biloba, acetyl-l-carnitine, vitamin E, ginseng, curcumin, lecithin, and other supplements versus placebo, but all were rated as having high risk of bias (ROB), most often due to high attrition. Therefore, evidence from these trials was a priori considered insufficient and was not analyzed. We found no eligible studies for Prevagen (apoaequorin), huperzine or phosphatidylserine.

Our findings are mostly consistent with those reported in prior systematic reviews. A 2000 Cochrane review on lecithin reported no clear benefit for AD,³⁹ a 2003 Cochrane review on acetyl-l-carnitine for AD suggested that isolated favorable findings for cognition were likely due to chance in the context of many other negative results,³⁸⁵ and a 2009 Cochrane review concluded that evidence for whether gingko improved cognition or function compared to placebo in people with AD was inconsistent and unreliable.³⁸⁶ In addition, a 2016 Cochrane review on omega-3 fatty acids in patients with mild to moderate AD found no convincing evidence of a benefit compared with placebo in trials of 6 to 18 months for cognition, function, staging, or quality of life. ³⁸⁷ However, our findings differed from those in a 2017 Cochrane review on vitamin E.³⁸⁸ We excluded a moderately sized (n=304), long-term trial of vitamin E versus placebo due to high ROB from high attrition (41% death, withdrawal, or loss to followup at 4 years; analyzed by a longitudinal repeated-measures mixed-effects model assuming missing at random). In contrast, the prior review, for which this was the only trial with extractable data in individuals with CATD, rated its attrition bias as low, stating that missing data were balanced across groups.³⁰⁰ This trial reported that among individuals with mild to moderate AD, participants assigned vitamin E for 6 to 48 months had a small improvement in function compared with placebo, but that treatments did not differ for cognition or serious adverse events.

In patients with CATD, how do prescription drugs and supplements compare for effects on cognition, function, behavioral and psychological symptoms of dementia (BPSD), and safety? We identified several eligible trials that directly compared supplements with prescription drugs for treatment of individuals with CATD. Respectively, these trials compared gingko biloba to donepezil, gingko biloba to rivastigmine, saffron extract to donepezil, vitamin E to donepezil, vitamin E to memantine, and Yishen Huazhuo decoction group to donepezil. For all these comparisons, evidence was judged insufficient to draw conclusions about relative differences for cognition, function, and safety, and none of the trials reported on BPSD. Although evidence from direct comparisons is insufficient, indirect evidence from placebo controlled trials suggests supplements may be less effective but is less clear regarding safety. Compared with placebo, cholinesterase inhibitors modestly improved cognition and function, while possibly increasing serious adverse events and withdrawals due to adverse events. Memantine inconsistently improved cognition in individuals with moderate to severe CATD when added to a cholinesterase inhibitor, but did not improve function. However, supplements compared with placebo either showed no difference in efficacy (Souvenaid^®, omega-3 fatty acids), or insufficient evidence about efficacy and safety (e.g., gingko biloba, ginseng, curcumin, vitamin E, resveratrol).

In patients with CATD, is there evidence to support targeting different prescription drugs or supplements to different patients for cognition and function outcomes? Limited trial data indicate that memantine may have a small benefit in individuals with moderate to severe CATD, but not in those with mild to moderate CATD. Trials of memantine compared with placebo did not stratify results by baseline CATD severity, but those limited to study participants with moderate to severe CATD showed inconsistent improvement in cognition, while those that enrolled patients with mild to moderate CATD showed no benefit in cognition or function. Cholinesterase inhibitors compared with placebo showed small improvements in cognition and function in both participants with mild to moderate and moderate to severe CATD. Three cholinesterase inhibitor trials reported that efficacy compared with placebo, or of high- versus standard-dose donepezil (i.e., 23 mg/day vs. 10 mg/day), may be greater in participants with lower compared with higher baseline MMSE. However, these findings were based on post hoc analyses, no tests for interaction were performed, and they may be due to chance. One trial reported that omega-3 fatty acids were no better than placebo in any of several groups defined by baseline MMSE. No eligible studies reported results stratified by age, race/ethnicity or sex.

In patients with CATD and BPSD, how effective are prescription drugs and supplements for agitation, aggression, and psychosis, and how long do benefits last? In eligible trials of individuals with CATD and BPSD of at least 2 weeks duration, antipsychotics and the antidepressant citalopram statistically significantly improved a minority of BPSD outcomes compared with placebo. However, due in part to imprecise and inconsistent results, evidence about the efficacy of these treatments for these outcomes was insufficient to draw conclusions. For antipsychotics, data were limited to small trials with little statistical power. One 6-week trial reported moderate-sized improvements in two agitation measures with standard-dose haloperidol compared with placebo, but neither of these differences was statistically significant and standard-dose haloperidol and placebo did not differ for either of two aggression measures. In another 6-week trial, quetiapine and placebo did not differ for agitation. Two antipsychotic trials showed inconsistent effects for psychosis, with statistically significant improvement in one of four measures for standard-dose haloperidol after 6 weeks, and in two of three measures for aripiprazole after 10 weeks. No evidence addressed the effects of antipsychotics on these outcomes beyond 10 weeks. For antidepressants, citalopram compared with placebo statistically significantly improved four of seven agitation measures after 9 weeks. Citalopram compared with placebo also statistically significantly improved one of five measures for psychosis. However, both citalopram trials evaluated a dose of 30 mg/day that exceeds the current maximum recommended dose of 20 mg/day. The effects of the this lower dose on BPSD outcomes is unknown. In people receiving open-label cholinesterase inhibitor, sertraline compared with placebo did not improve agitation. No eligible trials with low or medium risk of bias compared different antipsychotics, different antidepressants, or antipsychotics with antidepressants and reported data on agitation, aggression, or psychosis.

Trials investigating donepezil, divalproex sodium, and the traditional Japanese herbal mixture, Yokukansan, for 4 to 12 weeks collectively, reported no statistically significant improvements in agitation, aggression, or psychosis. Evidence was mostly insufficient to draw conclusions about these findings. Two trials of memantine for BPSD were rated high ROB and not analyzed. In one trial in which individuals living in care homes and receiving antipsychotics were randomized to either continue antipsychotics or switch to memantine, treatments did not statistically differ for agitation after 24 weeks, but evidence was graded insufficient.

In patients with CATD and BPSD, how well do prescription drugs and supplements work for treating disinhibited sexual behavior? Antipsychotics, selective serotonin reuptake inhibitor (SSRI) antidepressants, and less often hormones are sometimes used in clinical practice to treat disinhibited sexual behavior in patients with CATD and BPSD. However, no eligible trials with low or medium ROB reported on the effects of antipsychotics, SSRIs, other prescription drugs, or supplements on disinhibited sexual behavior. One small trial (n=14) compared estrogen versus placebo for 4 weeks and reported on sexual aggressiveness as a secondary outcome, but was rated high ROB and therefore not analyzed. Although trials in individuals with CATD and BPSD provide insufficient strength evidence about differences in risk of harms compared with placebo, trials in broader populations suggest potential risks of somnolence, confusion, falls, fractures, stroke, mortality, and other adverse outcomes. Given these potential harms, and the absence of evidence from eligible trials for benefits (not the same as definitive evidence that there is no benefit), use of these treatments for disinhibited sexual behavior in patients with CATD may not be warranted.

Which patients with CATD and BPSD are most likely to experience improvements in BPSD or harms from drug treatments? No eligible studies in individuals with CATD and BPSD reported on whether differences in BPSD or harms vary by participant characteristics.

In patients with CATD and BPSD, how do prescription drugs and nondrug treatments compare for BPSD outcomes and safety? Current guidelines for managing BPSD in patients with CATD recommend nondrug interventions as first line therapy.²⁴ However, a 2016 AHRQ report found that psychosocial interventions were not superior to usual care for managing agitation and aggression.²⁵ Our review found insufficient evidence about the comparative effectiveness of drug treatment and nondrug interventions for BPSD in patients with CATD and BPSD because no eligible trials addressed this question. We also found little indirect evidence. Eleven trials compared prescription drugs with placebo for BPSD. However, only one reported that enrolled patients must have had an inadequate response to a prior psychosocial intervention, and one reported that all participants in both treatment groups received a concomitant psychosocial intervention during the drug trial. Both tested the question of whether prescription medication improved BPSD beyond the uncertain benefit of psychosocial intervention.

AD is a qualifying condition for medical marijuana in many U.S. States, but what are the efficacy and safety of cannabinoids in patients with CATD? We identified no eligible trials of cannabinoids lasting at least 2 weeks for outcomes of agitation, aggression, psychosis, or disinhibited sexual behavior, or at least 24 weeks for cognition, function, and other efficacy outcomes and harms in individuals with CATD. Therefore, evidence was insufficient to draw conclusions about the efficacy or safety of cannabinoids for treatment of CATD and more specifically for treatment of BPSD in patients with CATD. Future trials are needed to examine the scientific evidence on the efficacy and harms of cannabinoids for these indications.

Applicability

This review aimed to evaluate the accuracy of brief cognitive tests for distinguishing CATD from normal cognition and MCI among individuals with suspected cognitive impairment (i.e., case finding). By intent, this review did not evaluate the role of formal neuropsychological testing for clinical diagnosis of CATD. This review also did not address the accuracy of brief cognitive tests for predicting future clinical progression to CATD or for distinguishing CATD from other types of dementia, though accuracy distinguishing between different types of dementia would be expected to be lower than that for distinguishing CATD from either normal cognition or MCI. We calculated true positive, false positive, true negative and false negative rates by applying the sensitivity and specificity reported in each analyzed study to its prevalence of CATD, normal cognition and MCI. These rates may not be generalizable to populations with a lower CATD prevalence or greater variability in the causes of cognitive impairment. In addition, participants in studies of brief cognitive testing were younger (mean 73 to 74 years) than many who present with suspected cognitive impairment in typical clinical settings, potentially limiting generalizability to these older patients. In the minority of cognitive testing studies that reported race, most participants were white; therefore, generalizability of study findings to other race/ethnic groups is unknown.

This review aimed to evaluate the accuracy of brain imaging and CSF tests to distinguish autopsy-confirmed AD from non-AD in individuals with CATD. Our review did not examine the accuracy of these biomarker tests to distinguish AD from either normal cognition or MCI, and excluded studies that used a biomarker as the reference gold standard rather than autopsy. This review also did not address the accuracy of biomarker tests for predicting clinical progression to AD over time. In the minority of studies that reported race, most participants were white; therefore, generalizability of study findings to other race/ethnic groups is unknown. Because studies examining biomarker classification accuracy included mostly participants whose symptoms began in their early 60s to early 70s, who were followed for years in research settings, and had low life expectancy after biomarker collection to limit the time interval between testing and autopsy, applicability to patients in better health, with later onset disease, and earlier in their disease course may be limited. This is in part because abnormalities detected by these tests and neuropathology may be less earlier in the disease course. In addition, studies examining the classification accuracy of brain imaging and CSF tests compared to a neuropathological reference sometimes used methods not feasible in clinical settings (e.g., complex analytical techniques for scan interpretation, gradings based on the consensus of multiple readers). Also, studies did not use standardized imaging techniques, assays, and cut points to categorize normal versus abnormal test results. The applicability of study findings on the accuracy of brain imaging and CSF tests for AD also may be limited because many of these biomarker tests may not be easily available in typical clinical settings-though the IDEAS study suggested that amyloid PET imaging at least may be accessible to dementia specialists outside of academic research settings. Further, CSF tests are invasive, and the association of all these biomarker test results with clinical patient outcomes is unknown.

It was the aim of the review on CATD drug treatment efficacy and harms to apply to older adults with CATD in typical clinical settings whose clinical dementia is not secondary solely to traumatic brain injury, frontotemporal dementia (FTD), Parkinson’s disease, Lewy body disease (LBD), stroke, or another non-AD etiology. Therefore, review findings may not apply to populations with these non-AD causes of dementia. Because few drug treatment trials reported data on race or ethnicity and those enrolled predominately white participants, the applicability of review findings to other racial or ethnic groups is unknown. Also, comorbidity was infrequently reported, so applicability of results to patients with multiple comorbid conditions is unknown. In addition, this review does not address the question of stopping drug treatment for CATD towards the end of life or the efficacy and harms of acute drug treatment for BPSD.

Limitations

The first limitation of this review on brief cognitive tests and biomarker tests is that it did not identify eligible studies that linked test accuracy to patient or caregiver outcomes, including cognitive, functional, psychological, quality of life and others. Moreover, eligible studies did not report on the association of these tests with process outcomes like changes in drug or nondrug management, including lifestyle changes or changes in life planning that may or may not affect patient or caregiver outcomes. Although the IDEAS study reported on changes in management, it is likely the frequency of management changes investigators reported after amyloid PET testing exceeded what would be expected in typical clinical practice. This is both because participating clinicians expected to change management following the test and a substantial portion of the management changes made were not evidence based (e.g., starting AD drug treatment in patients with MCI).

Evidence on the accuracy of brief cognitive tests for distinguishing CATD from normal cognition and MCI in patients with suspected cognitive impairment was limited in several ways. Many studies had small sample sizes. There were few eligible studies in this population for most individual cognitive tests, fewer of test combinations, and none for several common tests (e.g., Mini-Cog, St. Louis University Mental Status [SLUMS], Telephone Interview for Cognitive Status [TICS], Cambridge Neuropsychological Test Automated Battery [CANTAB]) or for any web-based tests. Studies on the classification accuracy of these and other brief cognitive tests were excluded for various reasons, including non-English test administration, not completing a diagnostic evaluation in many participants who completed brief cognitive testing, not using a acceptable definition for CATD (e.g., basing it only on a single cognitive test). ³⁸⁹ Evidence on the accuracy of brief cognitive tests also was limited because studies used variable standards to define CATD, MCI and normal cognition, none defined CATD using National Institute on Aging-Alzheimer’s Association (NIA-AA) clinical criteria,¹⁵ and none directly compared whether test classification accuracy varied as a function of which definitions were used. Cognitive test studies also used variable test scoring metrics and most often used cut points for distinguishing normal from abnormal that were defined from the analyzed cohort rather than being prespecified. Test performance was assessed when distinguishing CATD from MCI or NC, and is expected to be less accurate when distinguishing CATD from any other cause of cognitive impairment.

Limitations of biomarker classification studies included that there were few brain imaging and CSF biomarker studies with autopsy-confirmed reference standards, and none for blood tests. Many biomarker studies were limited by small sample sizes. Biomarker studies were methodologically heterogeneous, with sources of heterogeneity including composition of non-AD comparison groups, interval between biomarker collection and autopsy, methods of image acquisition or CSF assay and analysis, autopsy reference standards, and use of test cut points unique to their individual study cohorts. Biomarker studies also were limited because many study participants with biomarker measures did not complete autopsy and weren’t included in analyses, potentially introducing a selection bias. Further, no eligible biomarker studies evaluated accuracy of MRI hippocampal atrophy, CT, tau PET, or functional magnetic resonance imaging (fMRI) brain imaging; Aß42/Aß40 ratio or neurofilament light protein CSF tests; or of any blood tests compared to a reference standard of autopsy-confirmed AD. Few studies examined the classification accuracy of test combinations.

In both cognitive and biomarker test accuracy studies, the high prevalence of CATD and AD, respectively, could have increased diagnostic vigilance and led to sensitivity results higher than what would be expected in typical clinical populations,³⁹⁰ even those in whom CATD is suspected. Where reported, most participants in these studies were white, and little data evaluated whether accuracy varied by patient characteristics. Lastly, no cognitive testing studies or brain imaging or CSF testing studies reported data on clinical patient outcomes and only a few brain imaging or CSF testing studies reported on harms.

There are several limitations of the evidence on the efficacy and harms of CATD drug treatments. Few trials examined individual drug treatments, especially for supplements, BPSD treatments, and for prescription drugs when results were stratified by CATD severity. Many trials were limited by small sample size and short follow-up times, resulting in low statistical power for even somewhat common events and large mean differences between groups. For example, only 14 total deaths occurred in the 451 participants in the three eligible antipsychotic trials that reported mortality. This review limited prescription drug classes evaluated for cognition and function to cholinesterase inhibitors and memantine, and prescription drug classes evaluated for BPSD to cholinesterase inhibitors, memantine, antipsychotics, antidepressants, anxiolytics, antiepileptics/mood stabilizers, hormonal agents, and cannabinoids. Consequently, dextromethorphan-quinidine for treatment of individuals with CATD and agitation was not included, though one phase 2 trial suggests it may improve agitation, aggression and some caregiver measures compared with placebo.³⁹¹ By design, this review required studies of cognition and function to be at least 24 weeks in length, and studies of agitation, aggression, and psychosis to be at least 2 weeks in length. Trials reporting only on shorter-term treatment effects were excluded. Few included trials were longer than 26 weeks, so longer-term drug effects were unclear. Because trial populations were predominately white, generalizability to other racial/ethnic groups is uncertain. Few studies directly compared different drug treatments. Few trials reported results for CATD staging, individual cognitive domains, quality of life, or caregiver outcomes, and no eligible studies without high ROB reported results for disinhibited sexual behavior. Harms reporting was poor. For example, no eligible antipsychotic trials reported data on incident stroke. Many eligible trials were excluded from analyses due to high ROB, often because of high attrition, especially trials longer than 26 weeks and some that compared two active treatments. In at least one example, a trial we rated high ROB and excluded from analysis was not rated high ROB and was analyzed in a prior systematic review; this highlighted the potential sensitivity of systematic review results to the details of eligibility criteria. Many studies analyzed results using methods of accounting for missing data that may overestimate treatment benefit. It was difficult to interpret the relevance of small between-group differences in continuous outcomes, and most studies did not report data on between-group differences in the likelihood of experiencing clinically important treatment effects (i.e., responder analyses). Few trials evaluated whether treatment efficacy and harms varied by participant characteristics. Last, because we analyzed studies grouped by participant CATD severity and graded SOE for treatment effects within these severity categories, it is possible that SOE grades would have been different in cases when lumping studies regardless of baseline CATD severity may have been clinically reasonable (e.g., for harms).

Future Research