MMSE–CDR‐SB residual as an exploratory indicator of deviation from an Alzheimer's disease–typical cognitive–functional pattern

Abstract

INTRODUCTION

In clinical practice, patients with Alzheimer's disease (AD) often present with cognitive and functional profiles that diverge from what is expected. We tested whether the Mini‐Mental State Examination Clinical Dementia Rating Sum of Boxes (MMSE–CDR‐SB) residual, defined as observed minus expected CDR‐SB from a published MMSE–CDR‐SB reference equation, reflects clinically meaningful deviation from the expected cognitive–functional relationship.

METHODS

Using National Alzheimer's Coordinating Center data, we analyzed an autopsy cohort (n = 1981) and a separate clinical diagnosis cohort (n = 3184). Negative residual values indicated less functional impairment than expected for a given cognitive score, and positive values indicated greater functional impairment than expected. Associations were examined using multivariable logistic regression adjusted for MMSE score range, age, sex, and education.

RESULTS

Lower residual values were associated with lower odds of amyloid and AD‐type tau pathology. Higher residual values showed a directional association with transactive response DNA binding protein 43 kDa; vascular burden was also associated with the residual index but across a broader range. In the clinical cohort, higher residual values were enriched in progressive supranuclear palsy and frontotemporal lobar degeneration (other), whereas AD cases clustered near the reference pattern.

DISCUSSION

The MMSE–CDR‐SB residual may help flag less AD‐typical presentations for further etiologic evaluation, but cannot be interpreted as a pathology‐specific marker due to substantial overlap across diagnostic groups.

Highlights

• Mini‐Mental State Examination Clinical Dementia Rating Sum of Boxes (MMSE–CDR‐SB) residual summarizes deviation from an Alzheimer's disease (AD)‐typical pattern.

• Lower residuals were associated with lower odds of amyloid and AD‐type tau.

• Positive residuals were clinically enriched in progressive supranuclear palsy and frontotemporal lobar degeneration (other).

• MMSE–CDR‐SB residual has potential as a triage‐level signal of a less AD‐typical presentation.

1. INTRODUCTION

Dementia affects > 55 million people worldwide and is a major cause of disability and dependency in older adults, with its prevalence expected to almost triple by 2050. ¹ , ² In clinical practice and therapeutic trials, cognitive and functional status are commonly assessed with two standardized tools: the Mini‐Mental State Examination (MMSE), a brief screening measure of global cognitive performance, ³ and the Clinical Dementia Rating Sum of Boxes (CDR‐SB), a composite measure of dementia severity integrating cognitive and functional impairment. ⁴ , ⁵ , ⁶ These measures are also important for patient selection for Alzheimer's disease (AD) disease‐modifying therapies, ⁷ , ⁸ including recent anti‐amyloid treatments such as lecanemab and donanemab. ⁹

In AD, the most common cause of dementia, ¹⁰ MMSE and CDR‐SB are generally correlated across disease stages. Brogaard et al. proposed a linear equation to estimate expected CDR‐SB from MMSE, with coefficients derived from a mapping spanning the AD clinical severity spectrum and externally validated across the full cognitive spectrum in the National Alzheimer's Coordinating Center (NACC) dataset: CDR‐SB = −0.68 × MMSE + 20.2. ¹¹ Earlier studies also showed that common dementia scales such as MMSE, CDR‐SB, and ADAS‐Cog can be statistically linked across overlapping severity ranges. ¹² The Brogaard equation provides an expected CDR‐SB value for a given MMSE score. Therefore, it can be used to quantify deviation from the expected MMSE–CDR‐SB relationship. In this study, we defined this deviation as the residual, calculated as the difference between observed and predicted CDR‐SB. A positive residual means that the observed CDR‐SB is higher than expected for a given MMSE score, suggesting greater overall clinical severity, encompassing daily functional activities, than predicted from cognition alone. A negative residual means that the observed CDR‐SB is lower than expected, suggesting relatively lower overall clinical severity, with relatively preserved daily functional abilities relative to the level of cognitive impairment.

Deviation from this reference relationship may not only reflect pathology. Part of this deviation may stem from measurement error and reliability limitations inherent to the MMSE and CDR‐SB, as well as from differences in assessment method, because CDR‐SB relies in part on informant report whereas MMSE reflects direct patient performance. Deviation may also partially reflect the fact that CDR‐SB captures both cognitive and functional domains, whereas MMSE captures cognition alone. Additionally, NACC Uniform Data Set (UDS) guidance instructs clinicians to synthesize all available information before scoring the CDR, without specifying that cognitive test results such as the MMSE be withheld from the rater. ¹³ Criterion contamination between MMSE and CDR‐SB therefore cannot be excluded. Despite these sources of measurement‐related variability, this residual may be clinically informative because different pathological processes may affect cognition and everyday function differently. ¹⁴ This idea is supported by studies showing that different dementia syndromes can show different levels of clinical or functional severity at similar levels of cognitive impairment. For example, frontotemporal lobar degeneration (FTLD) has been reported to show greater clinical severity than AD after matching for MMSE and disease duration. ¹⁵ , ¹⁶ Therefore, deviation from the MMSE–CDR‐SB reference relationship may serve as a practical summary of an atypical cognitive–functional presentation or possible mixed pathology, without implying that it provides information independent of MMSE and CDR‐SB themselves. In this sense, the residual should be viewed not as a marker of one specific pathology, but as a simple summary of deviation from the expected MMSE–CDR‐SB relationship, which may correspond to a less AD‐typical cognitive–functional pattern in the present cohorts.

This question is clinically important because AD often co‐occurs with other pathologies. ¹⁴ , ¹⁷ In older adults, dementia is often a multi‐morbid brain condition rather than the result of a single disease. ¹⁴ , ¹⁸ Frequent co‐pathologies accompanying AD neuropathological change include Lewy body pathology, vascular pathology, cerebral amyloid angiopathy, and limbic‐predominant age‐related transactive response DNA binding protein 43 kDa (TDP‐43) encephalopathy (LATE) neuropathological change. ¹⁹ These co‐pathologies are clinically relevant because they contribute to cognitive decline and may reduce the measurable effect of therapies targeting only one pathological substrate. ¹⁹ However, biomarkers for non‐AD neuropathological changes are still limited in routine clinical practice. ¹⁹ , ²⁰

In routine practice, clinicians often encounter patients whose MMSE score and overall clinical severity do not fit a typical AD pattern, while access to biomarkers for non‐AD co‐pathologies remains limited. This is partially because the MMSE has limited coverage of frontal and executive functions, processing speed, and visuospatial abilities, and may therefore better capture the cognitive profile of AD than that of other dementia syndromes. ²¹ Indeed, the MMSE shows a ceiling effect in progressive supranuclear palsy (PSP), ²² does not adequately reflect the attentional and visuospatial deficits typical of Lewy body disease (LBD), ²³ is largely insensitive to the executive dysfunction predominant in vascular dementia, ²⁴ and may not adequately track disease progression in some FTLD phenotypes. ²⁵ These limitations are relevant to the interpretation of the residual across non‐AD syndromes. In this setting, a simple bedside measure is unlikely to establish etiology by itself. However, it may still be useful as an exploratory triage‐level signal to identify presentations that appear less AD typical and may merit closer etiologic evaluation. Prospectively, one possible use case would be to treat marked positive residuals as a prompt for closer non‐AD syndromic or mixed‐pathology evaluation, rather than as a diagnostic label.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed PubMed and reference lists for studies on Mini‐Mental State Examination Clinical Dementia Rating Sum of Boxes (MMSE–CDR‐SB) crosswalks, mixed dementia pathology, and syndrome‐specific differences in cognitive–functional severity. Prior work mainly mapped scores across instruments and did not test whether deviation from an Alzheimer's disease (AD)–oriented MMSE–CDR‐SB reference relationship was associated with neuropathological or clinical heterogeneity.

2. Interpretation: In National Alzheimer's Coordinating Center data, the MMSE–CDR‐SB residual behaved as a summary of deviation from an AD‐typical cognitive–functional pattern rather than as a specific pathology marker. Negative residuals were associated with lower odds of core AD pathology, whereas positive residuals were enriched in progressive supranuclear palsy and frontotemporal lobar degeneration (other) and showed only directional, non‐specific signals for selected non‐AD co‐pathologies.

3. Future directions: Prospective validation should test generalizability, refine practical thresholds, and determine whether this simple bedside metric improves etiologic triage for mixed pathology and non‐AD syndromic features, especially when biomarker access is limited.

Against this background, we used the MMSE–CDR‐SB relationship as an interpretive benchmark and examined whether deviation from this benchmark differs across neuropathological and clinical subgroups. We did not treat the residual as an independent biological marker. Rather, we evaluated whether it may serve as a practical summary of a less AD‐typical cognitive–functional presentation. Using the NACC dataset, we had two aims: first, to examine the association between the residual and neuropathological findings in an autopsy cohort; and second, to assess its relationship with clinician‐assigned etiologic diagnoses in a separate clinical cohort.

2. METHODS

2.1. Study design and participants

This study used data from the NACC UDS, a multicenter longitudinal observational cohort that collects standardized clinical, cognitive, and neuropathological data from Alzheimer's Disease Research Centers across the United States. ²⁶ We included visit data collected between 2005 and 2023. To reduce potential confounding by major non‐neurodegenerative conditions that may affect cognition or function, participants were excluded if they had a history of stroke, Parkinson's disease, alcohol abuse, schizophrenia, depressive disorder of < 2 years’ duration, bipolar disorder, or a known autosomal dominant AD mutation.

For the cross‐sectional analyses, we defined two analytic subcohorts according to data availability and study aim. For neuropathological analyses, we selected the clinical visit closest to death and retained only visits occurring within 24 months before death, to maximize the temporal proximity between ante mortem clinical assessment and post mortem findings. For clinical diagnosis analyses, we selected the visit closest to death with a clinician‐reported etiologic diagnosis. We did not impose an upper limit on the interval to death in the main clinical analysis, primarily to preserve sample size. Because this design differs from the neuropathological cohort, we additionally evaluated the distribution of the visit‐to‐death interval in the clinical cohort.

Because the MMSE–CDR‐SB reference equation proposed by Brogaard et al. was reported to perform less well in more severe stages, the main analyses were restricted to visits with MMSE scores between 16 and 30. ¹¹ The upper end of this range is subject to known ceiling effects of the MMSE, which are addressed further in the Discussion section. After this restriction, the final analytic samples consisted of 1981 participants in the neuropathological cohort and 3184 participants in the clinical cohort (Figure S1 in supporting information). Cognitively normal individuals within this MMSE‐eligible range were retained in both cohorts to preserve the observed cognitive–functional spectrum. These single‐visit cohort definitions ensured that each participant contributed exactly one observation per cohort, yielding independent observations for the primary cross‐sectional analyses.

2.2. Discrepancy index

For each eligible visit, we calculated the difference between observed CDR‐SB and the value predicted from MMSE by the Brogaard equation (CDR‐SB = 20.2 − 0.68 × MMSE). ¹¹ This value was used as the discrepancy index (residual), defined as: residual = observed CDR‐SB − (20.2 − 0.68 × MMSE). For example, for a participant with an MMSE score of 25 and an observed CDR‐SB of 1.5, the expected CDR‐SB would be 3.2, yielding a residual of −1.7.

As in the example above, a negative residual indicates that the observed CDR‐SB is lower than expected, corresponding to less overall daily functional severity than expected. A positive residual indicates that the observed CDR‐SB is higher than expected for a given MMSE score, corresponding to greater overall daily functional severity than predicted from cognition alone.

For all descriptive, stratified, and graphical analyses, the discrepancy index was categorized a priori into four ordinal bins (≤ −4, −4 to 0, 0 to +4, and ≥ +4), and MMSE was categorized into three bands (16–21, 22–27, and 28–30). These categories were chosen pragmatically to support stratified visualization, preserve adequate cell sizes, and balance sample size across MMSE strata; detailed rationale is provided in the Supplementary Methods in supporting information. Categorical specifications were retained in regression models for consistency with the stratified figures, and a sensitivity analysis using continuous MMSE yielded materially unchanged conclusions.

2.3. Neuropathological and clinical variables

Neuropathological variables were defined from NACC autopsy form items using predefined study definitions informed by the NACC Neuropathology Coding Guidebook and contemporary consensus frameworks. The main neuropathological variables included amyloid pathology (A+), ²⁷ AD‐type tau pathology (T+), ²⁸ non‐AD tau pathology, ²⁹ neurodegeneration (N+), ³⁰ Lewy body pathology, ²³ TDP‐43 pathology, ³¹ hippocampal sclerosis, ³² cerebral amyloid angiopathy, ³³ and high vascular burden. ³⁴ High vascular burden was defined as a study‐specific composite across prespecified vascular lesion variables. The main variable TDP‐43 should not be interpreted as a broad age‐related TDP‐43 measure. Given its derivation from a single composite item without regional specificity, it predominantly reflects FTLD‐type TDP‐43 pathology in this dataset. Findings related to this variable should therefore be read as a directional signal rather than a robust prevalence estimate. Detailed coding rules and source variables are provided in Table S1 in supporting information.

For the clinical cohort, presumptive etiologic diagnoses were derived from the clinician‐reported primary etiologic diagnosis variable (NACCETPR) and converted into one‐versus‐rest binary flags for AD, LBD, vascular dementia (VaD), FTLD‐other, PSP, corticobasal degeneration (CBD), other neurological disorders, and cognitively normal status. These variables reflect NACC‐recorded etiologic diagnoses rather than diagnoses re‐adjudicated within the present study. Detailed coding is provided in Table S1.

2.4. Statistical analysis

All statistical analyses were conducted in R version 4.5.1. Analyses were performed separately in the neuropathological cohort and the clinical cohort using complete‐case data for the variables required by each model. Two‐sided p < 0.05 were considered statistically significant. No formal multiplicity adjustment was applied because the analyses were exploratory and hypothesis generating. Because both MMSE and residual were categorized for interpretability rather than empirical optimization, these analyses were intended primarily to describe pattern‐level associations rather than to establish decision thresholds for individual‐level use.

As descriptive analyses, we plotted subgroup‐specific smoothed MMSE–CDR‐SB trajectories with the published MMSE–CDR‐SB reference line overlaid (Figures 1 and 2) and examined prevalence of each diagnostic subgroup across MMSE‐band × residual‐bin strata (Figures 3 and 4). Adjusted prevalence was estimated using logistic‐regression standardization, with age at visit, sex, and years of education as covariates; crude proportions were reported when the adjusted model did not converge or analyzable sample size was too small. The primary cross‐sectional analyses used separate multivariable logistic regression models for each binary neuropathological and clinical outcome, with residual bin as the predictor of interest, adjusting for MMSE band, age at visit, sex, and years of education. The reference categories were the 0 to +4 residual bin and the 16 to 21 MMSE band. Odds ratios (ORs), 95% confidence intervals (CIs), and p values were estimated using cluster‐robust standard errors clustered by participant (NACCID). Full model outputs are provided in supporting information.

FIGURE 1.

MMSE–CDR‐SB relationship across neuropathological groups. Relationship between MMSE and CDR‐SB across neuropathological categories (MMSE 16–30). Each panel shows smoothed regression curves (spline, 95% CI) for positive (1, green) vs. negative (0, red) groups. The published MMSE–CDR‐SB reference trajectory from Brogaard et al. ¹¹ (CDR‐SB = 20.2 − 0.68 × MMSE) is overlaid as a dashed line for comparison. Shaded areas represent 95% CIs around the smoothed trajectories. Only categories included in the formal GLM analysis (Table 2A) are shown; n ≥ 10 in both groups required. A+, amyloid positive; T+, tau‐positive (AD pattern, Braak ≥ 3); Non‐AD tau+, tau‐positive (non‐AD pattern); N+, neurodegeneration positive; T class, tau− / AD‐tau+ / non‐AD‐tau+. AD, Alzheimer's disease; CDR‐SB, Clinical Dementia Rating Scale Sum of Boxes; CI, confidence interval; GLM, generalized linear model; MMSE, Mini‐Mental State Examination; TDP‐43, transactive response DNA binding protein 43 kDa.

FIGURE 2.

MMSE–CDR‐SB relationship across clinical diagnostic groups. Relationship between MMSE and CDR‐SB across clinical etiologic diagnoses (MMSE 16–30). Each panel shows smoothed regression curves (spline, 95% CI) for the target diagnosis (1, green) versus all others (0, red). Only categories with n ≥ 10 in both positive and negative groups are shown. The published MMSE–CDR‐SB reference trajectory from Brogaard et al. ¹¹ (CDR‐SB = 20.2 − 0.68 × MMSE) is overlaid as a dashed line for comparison. Shaded bands represent 95% CIs around the smoothed trajectories. AD, Alzheimer's disease; CBD, corticobasal degeneration; CDR‐SB, Clinical Dementia Rating Scale Sum of Boxes; CI, confidence interval; Cogn. normal, cognitively normal; FTLD‐other, frontotemporal lobar degeneration (other);LBD, Lewy body disease (clinical diagnosis; distinct from Lewy body pathology); MMSE, Mini‐Mental State Examination; PSP, progressive supranuclear palsy; VaD, vascular dementia.

FIGURE 3.

Prevalence of neuropathological co‐pathologies across discrepancy bins, stratified by MMSE band. Bar plots display the age‐, sex‐, and education‐adjusted proportion of participants positive for each neuropathological variable across categories of discrepancy between observed and expected CDR‐SB given MMSE (discrepancy bins: ≤ −4, −4 to 0, 0 to +4, ≥ +4), stratified by MMSE band (16–21, 22–27, 28–30). Adjusted proportions were estimated using logistic standardization based on models including discrepancy bin, MMSE band, age at visit, sex, and years of education. Predicted probabilities were averaged over the observed covariate distribution within each MMSE band to obtain adjusted prevalence estimates for each discrepancy bin. The number of participants in each bin is displayed within each bar. Only variables with n ≥ 10 cases overall and used in the GLM are shown. Sample sizes (n) are indicated within each bar. Analyses were conducted in the neuropathological cohort (n = 1981). A+, amyloid positive (Thal phase ≥ 3 or Consortium to Establish a Registry for Alzheimer's Disease score ≥ 2 or Alzheimer's disease neuropathologic change ≥ 2); T+, AD‐tau positive (Braak stage ≥ 3); Non‐AD tau+, tau‐positive in a non‐AD distribution; N+, neurodegeneration positive. AD, Alzheimer's disease; CDR‐SB, Clinical Dementia Rating Scale Sum of Boxes; GLM, generalized linear model; MMSE, Mini‐Mental State Examination.

FIGURE 4.

Prevalence of presumptive etiologic diagnoses across discrepancy bins, stratified by MMSE band. Bar plots display the age‐, sex‐, and education‐adjusted proportion of participants with each clinician‐reported etiologic diagnosis across categories of discrepancy between observed and expected CDR‐SB given MMSE  (discrepancy bins: ≤ −4, −4 to 0, 0 to +4, ≥ +4), stratified by MMSE band (16–21, 22–27, 28–30). Adjusted proportions were estimated using logistic standardization based on models including discrepancy bin, MMSE band, age at visit, sex, and years of education. Predicted probabilities were averaged over the observed covariate distribution within each MMSE band to obtain adjusted prevalence estimates for each discrepancy bin. Only diagnoses with n ≥ 10 cases overall are shown. Sample sizes (n) are indicated within each bar. Analyses were conducted in the clinical cohort (n = 3184). AD, Alzheimer's disease; CBD, corticobasal degeneration; CDR‐SB, Clinical Dementia Rating Scale Sum of Boxes; Cogn. normal, cognitively normal; FTLD‐other, frontotemporal lobar degeneration (other subtypes); LBD, Lewy body disease (clinical diagnosis); MMSE, Mini‐Mental State Examination; PSP, progressive supranuclear palsy; VaD, vascular dementia.

Sensitivity analyses for death/autopsy‐related selection bias, replacement of MMSE band with continuous MMSE as the adjustment variable, analyses of co_AD × residual‐bin interactions for selected pathologies, and exploratory longitudinal mixed‐effects analyses in an autopsy‐linked longitudinal subsample are described in the Supplementary Methods.

2.5. Ethics

This study was approved by the University of Tokyo Graduate School of Medicine Institutional Ethics Committee (ID: 2025264NI). Informed consent was not required because the study uses anonymized, publicly available data only.

3. RESULTS

3.1. Sample characteristics

The neuropathological cohort included 1981 participants (mean age 83.1 ± 9.5 years; 46.7% female; mean education 15.6 ± 5.9 years), with a median visit‐to‐death interval of 10.2 months (interquartile range [IQR] 5.9–14.8). The clinical cohort included 3184 participants, with a median visit‐to‐death interval of 13.3 months (IQR 6.8–30.1). Detailed participant characteristics are presented in Table 1.

TABLE 1.

Basic characteristics of participants.

Characteristic	Neuropathological cohort (n = 1981)	Clinical cohort (n = 3184)
Demographics
Age at visit, mean ± SD (years)	83.1 ± 9.5	81.5 ± 10.2
Female sex, %	46.7%	48.8%
Education, mean ± SD (years)	15.6 ± 5.9	15.8 ± 7.5
Visit characteristics
Time from visit to death, median [IQR] (months)	10.2 [5.9–14.8]	13.3 [6.8–30.1]
Cognitive and functional scores
MMSE, mean ± SD	24.8 ± 4.2	24.4 ± 4.4
CDR‐SB, mean ± SD	3.5 ± 4.1	3.7 ± 4
CDR Global Score, n (%)
0	647 (32.7%)	938 (29.5%)
0.5	667 (33.7%)	1070 (33.6%)
1	399 (20.1%)	732 (23%)
2	247 (12.5%)	412 (12.9%)
3	21 (1.1%)	32 (1%)
MMSE band, n (%)
16–21	491 (24.8%)	888 (27.9%)
22–27	746 (37.7%)	1200 (37.7%)
28–30	744 (37.6%)	1096 (34.4%)
Residual bin, n (%)
≤ −4	67 (3.4%)	132 (4.1%)
−4 to 0	976 (49.3%)	1556 (48.9%)
0 to +4	766 (38.7%)	1226 (38.5%)
≥ +4	172 (8.7%)	270 (8.5%)
Neuropathological variables, % positive (n) neuropathological cohort only
A+ (amyloid)	54.1% (n = 1023)	—
T+ (AD tau)	62.9% (n = 1010)	—
N+ (neurodegeneration)	77.5% (n = 511)	—
Lewy body pathology	22.5% (n = 1020)	—
TDP‐43 pathology	4.2% (n = 428)	—
LATE‐like (regional TDP)	26.5% (n = 366)	—
LATE stage ≥ 2	16.8% (n = 346)	—
Hippocampal sclerosis	11.4% (n = 507)	—
High vascular burden	46.4% (n = 1023)	—
CAA (any)	50.1% (n = 1001)	—
Presumptive etiologic diagnosis, n (%) clinical cohort only
Alzheimer's disease	—	1525 (47.9%)
Lewy body disease	—	119 (3.7%)
Vascular dementia	—	109 (3.4%)
FTLD‐other	—	161 (5.1%)
PSP	—	38 (1.2%)
CBD	—	16 (0.5%)
Other neurological	—	16 (0.5%)
Cognitively normal	—	935 (29.4%)

Note: Demographic, cognitive, and neuropathological characteristics are presented separately for the neuropathological cohort (n = 1981) and the clinical cohort (n = 3184). Time from visit to death is reported for both cohorts. The neuropathological cohort was restricted to visits within 24 months before death, whereas the clinical cohort had no upper time limit on this interval. A+, T+, and N+ were defined independently of one another. T+ was defined as Braak stage ≥ 3 regardless of amyloid status. N+ was defined as the presence of cortical atrophy, hippocampal atrophy, or hippocampal sclerosis, regardless of amyloid or tau status. These three biomarkers therefore reflect independent neuropathological dimensions and are not hierarchically nested. Percentages for neuropathological variables are based on variable‐specific non‐missing denominators, which vary substantially across variables due to differences in neuropathological form version availability. The total cohort n is provided in the column header for reference. The discrepancy index was computed as CDR‐SB − (20.2 − 0.68 × MMSE), following Brogaard et al.¹¹ Neuropathological variables were derived from standardized NACC autopsy data and are reported for the neuropathological cohort only.  Reference category for discrepancy bin: 0 to +4.

Abbreviations: A+, amyloid positive; CAA, cerebral amyloid angiopathy; CBD, corticobasal degeneration; CDR‐SB, Clinical Dementia Rating Sum of Boxes; FTLD, frontotemporal lobar degeneration; HS, hippocampal sclerosis; IQR, interquartile range; LATE, limbic‐predominant age‐related transactive response DNA binding protein 43 kDa encephalopathy; LBD, Lewy body disease; MMSE, Mini‐Mental State Examination; N+, neurodegeneration positive; PSP, progressive supranuclear palsy; SD, standard deviation; T+, AD‐tau positive (Braak ≥ 3); TDP‐43, TAR DNA‐binding protein 43.

3.2. Visual inspection of MMSE–CDR‐SB trajectories across pathological and clinical groups

Descriptive MMSE–CDR‐SB trajectories by subgroup are shown in Figures 1 and 2. Core AD‐related pathological markers, including A+, T+, and N+, largely followed the Brogaard reference line. In contrast, TDP‐43 pathology, and to a lesser extent hippocampal sclerosis, tended to lie above the reference line, corresponding to positive residual values. Clinically, FTLD‐other, PSP, and VaD tended to lie above the reference line, whereas cognitively normal participants fell well below it, corresponding to negative residual values. These descriptive patterns were evaluated formally in the regression analyses below.

3.3. Association of the residual index with neuropathological markers

In cross‐sectional regression analysis, the residual index showed significant overall associations with amyloid positivity, AD‐tau positivity, high vascular burden, and TDP‐43, whereas no significant overall association was observed for Lewy body pathology or neurodegeneration (Table 2A).

TABLE 2.

Adjusted odds ratio for residual bin.

Outcome	N	Pos	OR ≤ −4	OR −4 to 0	OR 0 to +4	OR ≥ +4	p
A. Neuropathological cohort
A+ (amyloid)	1023	553	0.63 (0.22–1.86)	0.62 (0.47–0.82)	(Reference)	0.70 (0.44–1.10)	0.007
T+ (AD tau)	1010	635	0.85 (0.25–2.92)	0.57 (0.42–0.77)	(Reference)	0.70 (0.42–1.16)	0.003
Non‐AD tau+	1018	31	NC*	0.52 (0.21–1.29)	(Reference)	2.01 (0.80–5.07)	0.060
N+ (neurodegeneration)	511	396	0.47 (0.07–2.98)	1.06 (0.68–1.66)	(Reference)	1.85 (0.66–5.16)	0.449
TDP‐43 pathology	428	18	NC*	0.33 (0.06–1.76)	(Reference)	2.90 (0.97–8.65)	0.011
Hippocampal sclerosis	507	58	NC*	0.63 (0.33–1.20)	(Reference)	1.31 (0.58–2.93)	0.089
Lewy body pathology	1020	229	0.73 (0.23–2.27)	0.93 (0.67–1.28)	(Reference)	0.94 (0.55–1.58)	0.929
High vascular burden	1023	475	3.32 (1.12–9.84)	1.34 (1.02–1.77)	(Reference)	1.29 (0.81–2.06)	0.030
B. Clinical cohort
AD	3072	1525	0.13 (0.09–0.21)	0.38 (0.32–0.46)	(Reference)	0.68 (0.47–0.99)	< 0.001
LBD (clinical)	3072	119	0.63 (0.26–1.53)	0.45 (0.29–0.70)	(Reference)	1.14 (0.67–1.95)	< 0.001
VaD	3072	109	1.64 (0.66–4.08)	1.43 (0.92–2.23)	(Reference)	1.39 (0.65–2.97)	0.396
FTLD‐other	3072	161	1.43 (0.65–3.18)	0.36 (0.22–0.58)	(Reference)	1.80 (1.09–2.98)	< 0.001
PSP	3072	38	0.87 (0.10–7.35)	0.27 (0.11–0.70)	(Reference)	2.69 (1.20–6.03)	< 0.001
CBD	3072	16	1.96 (0.22–17.22)	1.58 (0.52–4.75)	(Reference)	0.60 (0.07–5.45)	0.677
Cognitively normal	3072	935	47.62 (24.72–91.73)	5.59 (4.38–7.14)	(Reference)	0.03 (0.00–0.19)	< 0.001

Note: P (LRT): overall p value from a likelihood ratio test comparing the full model with residual‐bin categories to a reduced model without residual bin, testing the joint association of all residual categories with the outcome. ORs shown in this table are category‐specific estimates for each residual bin relative to the 0 to +4 reference bin. Models were adjusted for MMSE band, age at visit, sex, and years of education. Covariate terms not shown in this table are provided in Table S2 in supporting information. Part A presents adjusted ORs for the association between discrepancy index bins and the presence of neuropathological co‐pathologies in the neuropathological cohort. Cerebral amyloid angiopathy was not included in the primary regression analyses to maintain consistency with the pre‐specified outcome set; it is described in Table 1 and Table S1 in supporting information. Part B presents the corresponding associations with presumptive clinical diagnoses in the clinical cohort. ORs were estimated using logistic regression models with discrepancy bin as the primary predictor and the bin 0 to +4 as the reference category. All models were adjusted for MMSE band, age at visit, sex, and years of education. ORs > 1 indicate higher odds of the outcome relative to the reference discrepancy bin, whereas values < 1 indicate lower odds. The “other neurological disorders” was excluded from the primary regression analyses owing to insufficient positive cases (n = 16). The neuropathological cohort includes participants with autopsy data and a clinical visit within 24 months prior to death. The clinical cohort includes participants with a clinician‐reported etiologic diagnosis at their visit closest to death. NC* indicates “not calculable” due to zero cells / complete separation. Bold p values indicate statistical significance (p < 0.05).

Abbreviations: AD, Alzheimer's disease; CBD, corticobasal degeneration; CI, confidence interval; FTLD, frontotemporal lobar degeneration; LRT, likelihood ratio test; MMSE, Mini‐Mental State Examination; OR, odds ratio; PSP, progressive supranuclear palsy; VaD, vascular dementia.

For the core AD pathological markers, participants with residuals below the reference value were less likely to harbor amyloid or AD‐type tau pathology than those in the reference residual bin (Table 2A), consistent with lower overall functional severity than expected for a given level of cognitive impairment.

For TDP‐43 pathology, the direction of association was consistent with a priori hypothesis. Participants in the highest residual bin (≥ +4) had higher odds of TDP‐43 positivity relative to the reference bin (OR = 2.90, 95% CI: 0.97–8.65, category‐specific P = 0.057), although the overall likelihood ratio test for the residual bin term was significant (P = 0.011, Table 2A). Interpretation is limited by the small number of TDP‐43–positive cases and sparse data in the lowest residual bin, in which no TDP‐43–positive cases were observed (Figure 3). The corresponding raw counts are shown in Figure S2 in supporting information. High vascular burden was also associated with the residual index, but the pattern was spread across multiple residual categories rather than confined to the positive extreme, suggesting a broader and less specific relationship. Covariate terms not shown in Table 2 are provided in Table S2 in supporting information.

3.4. Association of the residual index with clinician‐assigned etiologic diagnosis

In the clinical cohort, the residual index was significantly associated with several clinician‐assigned etiologic diagnoses. The most clinically interpretable associations were observed for AD, PSP, and FTLD‐other (Table 2B).

Cognitively normal status was concentrated in the lower residual range, with substantially higher odds in the negative residual bins. Conversely, AD diagnosis was less common away from the reference bin, including in the highest residual category (OR for ≥ +4 bin vs. reference, 0.68, 95% CI: 0.47–0.99, p = 0.043), consistent with clinically typical AD clustering near the MMSE–CDR‐SB reference relationship in this cohort.

PSP showed a positive residual pattern, with higher odds in the ≥ +4 bin (OR = 2.69, 95% CI: 1.20–6.03, p = 0.016) and lower odds in the −4 to 0 bin (OR = 0.27, 95% CI: 0.11–0.70, p = 0.007). For FTLD‐other, odds were higher in the ≥ +4 bin (OR = 1.80, 95% CI: 1.09–2.98, p = 0.021) and lower in the −4 to 0 bin (OR = 0.36, 95% CI: 0.22–0.58, p < 0.001), relative to the reference bin. The ≤ −4 bin also showed an elevated point estimate (OR = 1.43, 95% CI: 0.65–3.18), although this was not statistically significant. The pattern across bins was therefore not strictly monotonic with residual value. No clear association was observed for VaD (overall p = 0.396). For LBD, the overall association reached significance, but this was mainly driven by lower odds in the negative residual range rather than enrichment in the positive range. CBD was not significantly associated with the residual index (overall p = 0.677). Overall, these findings suggest that the residual index captures clinical heterogeneity, with the highest residual enriched in selected non‐AD syndromic diagnoses such as PSP and FTLD‐other. The distribution of clinician‐assigned etiologic diagnoses across MMSE‐band × residual‐bin strata is shown in Figure 4. The corresponding raw counts are shown in Figure S3 in supporting information.

3.5. Sensitivity and supplementary analyses

Additional sensitivity and supplementary analyses are summarized in the Supplementary Results in supporting information. Death/autopsy‐related selection analyses indicated associations of death status and death × residual‐bin interactions for several diagnoses (Table S3 in supporting information). AD background did not significantly modify selected residual‐bin associations (Table S4 and Figure S4 in supporting information). In an autopsy‐linked longitudinal subsample, residual values increased over time, with steeper increases for amyloid, FTLD‐like TDP‐43, LATE‐like, and LATE stage ≥ 2 pathologies, although this may partially reflect MMSE limitations rather than accumulating pathology (Table S5 and Figure S5 in supporting information). Replacing MMSE band with continuous MMSE yielded materially consistent residual‐bin associations, with only minor changes in nominal significance (Table S6 in supporting information).

4. DISCUSSION

In this study, we examined whether deviation from the MMSE–CDR‐SB reference relationship can be used as a practical summary of a less AD‐typical cognitive–functional presentation. Consistent with the AD clinical severity spectrum used to calibrate the Brogaard equation, AD‐related pathological groups in the autopsy cohort and clinically diagnosed AD cases in the clinical cohort clustered closely around the reference line, whereas non‐AD subgroups deviated from it in differentiated ways: cognitively normal participants concentrated in the negative residual range, and PSP/FTLD‐other enriched in the positive range (Figures 1 and 2; Table 2). This empirical alignment supports interpreting deviation from this reference, in the present cohorts, as a signal of a less AD‐typical cognitive–functional profile. Our findings suggest that the residual is better understood as a summary of deviation from the expected MMSE–CDR‐SB relationship than as an independent pathology marker. Lower residual values were associated with lower odds of core AD pathology, whereas the highest residual bin was enriched in selected non‐AD syndromic diagnoses, especially PSP and FTLD‐other; the main residual patterns are summarized in Table S7 in supporting information. In this sense, the residual may be most useful as a practical triage‐level signal of a less AD‐typical cognitive–functional presentation.

Among the neuropathological findings, vascular burden showed a statistically significant association with the residual index overall, but the pattern was not directionally consistent. Higher odds were seen in several residual categories, but not in the highest positive bin. This pattern, with higher odds concentrated in the negative rather than the positive residual range, does not suggest a simple relationship between vascular burden and greater‐than‐expected functional severity, and remains difficult to interpret in a single biological direction. By contrast, TDP‐43 pathology showed a more consistent directional pattern. Participants in the highest residual bin had higher odds of TDP‐43 positivity, and no TDP‐43–positive cases were seen in the lowest residual bin. This result should be interpreted carefully because the number of positive cases was small and missing data were common. Moreover, the TDP‐43 variable used here is derived from a single composite NACC item and leans toward FTLD‐type TDP‐43 rather than reflecting the broader age‐related TDP‐43 construct associated with LATE. It should therefore not be generalized to age‐related TDP‐43 pathology. Still, the direction of the association is biologically plausible, because TDP‐43 pathology often affects limbic structures and may increase memory‐related and functional impairment more than would be expected from MMSE alone, ³⁵ , ³⁶ though this interpretation remains speculative given the alternative sources of residual deviation discussed earlier.

The clinical cohort gave an important complementary result. In this analysis, AD cases tended to stay near the MMSE–CDR‐SB reference line, whereas cognitively normal participants were concentrated in the negative residual range. In contrast, positive residual values were more common in non‐AD syndromic diagnoses such as PSP and FTLD‐other. These findings suggest that the residual index may capture atypical cognitive–functional profiles in clinical practice. ¹⁵ This interpretation is in line with earlier studies suggesting that functional or global clinical severity can differ across syndromes even at similar levels of cognitive impairment. ¹⁵ , ¹⁶ , ³⁷ In this sense, the residual may be useful not because it identifies one diagnosis, but because it highlights patients whose clinical profile looks less AD‐typical. However, there was still a large overlap between diagnostic groups. For this reason, the residual index alone is not enough for etiologic diagnosis at the individual level, and it should be considered only as an additional clinical signal. ³⁸ , ³⁹ This limitation is not surprising, because the MMSE is a brief global screening tool and the reference equation was calibrated using data spanning the AD clinical severity spectrum and externally validated in NACC. Its main practical value may be to help prioritize which patients warrant closer etiologic workup when the presentation is less typical of clinically typical AD. Relatedly, criterion contamination between MMSE and CDR‐SB cannot be excluded. In this dataset, MMSE correlated similarly with the CDR memory box (Spearman r = −0.62) and with other cognitively oriented CDR boxes (r = −0.56 to −0.59), and less strongly with the personal care box (r = −0.37), a pattern consistent with general construct overlap rather than disproportionate contamination of a single domain.

Several limitations should be noted. First, the neuropathological analyses were based on an autopsy‐defined cohort and were therefore vulnerable to selection bias. ⁴⁰ Our sensitivity analyses also suggested that death‐related selection may affect how residual categories relate to diagnosis. Second, the apparently low frequency of TDP‐43 positivity in this older AD‐predominant cohort may reflect not only biology but also the variable definition, because the main TDP variable mainly reflects FTLD‐type rather than age‐related TDP‐43, as well as substantial missingness, form–version dependence, and autopsy selection bias. Therefore, this finding should be interpreted as hypothesis generating rather than as a robust prevalence estimate. Third, MMSE is a brief global screening test and does not capture domain‐specific deficits in detail. ³⁸ This is particularly relevant for disorders such as LBD, in which visuospatial and neuropsychiatric features are prominent, ²³ and vascular pathology, which often affects executive function more than memory. ³⁴ Accordingly, the residual framework may be less sensitive for non‐AD conditions requiring multidimensional assessment. In addition, NACC data may not fully capture functional or psychiatric comorbidities that could independently increase CDR‐SB. Fourth, residual confounding remains possible, and the residual itself is a linear function of MMSE and CDR‐SB. It therefore does not provide information independent of the original measures and should be viewed as a practical summary index rather than a separate biological construct. ⁴¹ Accordingly, an extreme residual should first prompt consideration of measurement‐related explanations, including informant factors, administration differences, and construct overlap between MMSE and CDR‐SB, before being interpreted as evidence of an atypical syndrome or mixed pathology. Two methodological issues also require caution: CDR‐SB floor effects may partially explain the negative residual pattern in cognitively normal participants, and near MMSE ≈ 30 the linear equation can yield slightly negative predicted CDR‐SB values. Additionally, the MMSE also shows a ceiling effect and limited sensitivity for detecting MCI, ²¹ which, combined with the CDR‐SB floor effect described above, may further limit residual interpretability at the upper end of the MMSE range. Relatedly, the increase in residual values observed over time in the exploratory longitudinal analysis may also reflect MMSE insensitivity to longitudinal cognitive decline or practice effects from repeated administration. Finally, because the reference MMSE–CDR‐SB equation was calibrated using data spanning the AD clinical severity spectrum, ¹¹ the clustering of AD cases near the reference pattern is partially expected by construction and should not be overinterpreted as diagnostic specificity. In this study, the residual should be interpreted as deviation from a published MMSE–CDR‐SB benchmark, with AD typicality inferred only from the empirical alignment observed in the present cohorts, rather than as deviation from a universal dementia trajectory. ¹¹ , ¹⁴ Future work should examine whether analogous residual frameworks based on domain‐specific neuropsychological measures in NACC, rather than MMSE alone, better capture non‐AD cognitive–functional profiles.

In conclusion, the MMSE–CDR‐SB residual may serve as a simple summary of deviation from the expected MMSE–CDR‐SB relationship, which may help identify less AD‐typical cognitive–functional presentations in this cohort. In this study, lower residual values were associated with lower odds of core AD pathology, whereas positive residual values were enriched in selected non‐AD syndromic diagnoses such as PSP and FTLD‐other. These findings support the residual as an exploratory triage‐level signal rather than as a standalone diagnostic or prognostic marker. Future work should clarify its performance in time‐restricted clinical cohorts and prospectively evaluate whether it adds practical value for etiologic triage.

CONSENT STATEMENT

N/A.

CONFLICT OF INTEREST STATEMENT

A.M. and S.N. have no conflicts of interest to disclose. K.S. has no conflicts of interest related to the content of the manuscript, is involved in a joint research project with the MetLife Foundation, and had received a research grant from Eli Lilly for collaborative research unrelated to the current manuscript. Y.N. is involved in collaborative researches with NIPRO Corporation, CANON Medical Systems Corporation, and Eli Lilly & Company, and had received consultancy/speaker fees from Eisai, and Eli Lilly. M.K. received honoraria for lectures from Eisai, Eli Lilly, FUJIREBIO, and Nihon Medi‐Physics and a patent assignment fee from FUJIREBIO. R.I. received advisory fees from Eisai, Eli Lilly, and MSD; consultant fee from Chugai; and honoraria for lectures from Eisai, Eli Lilly, Nihon Medi‐Physics, PDR Pharma, FUJIREBIO, Sysmex, and IQVIA. A.I. received research grants from Eisai, FUJIREBIO, Janssen pharma, Sysmex, Kobayashi Pharma, Eli Lilly, Fujifilm, SONY, Biogen, and Chugai/Roche; advisory fees from Eisai, FUJIREBIO, Eli Lilly, Roche, GSK, Otsuka, Soundwave Innovation; honoraria for lectures from Eisai, Eli Lilly, Biogen, Chugai/Roche, HU frontier, FUJIREBIO, Kowa, Sysmex, Ono, Otsuka, Alnylam, Daiichi Sankyo, Tokio Marine & Nichido Fire Insurance, PDR pharma, IQVIA, Sumitomo Pharma, MSD, Janssen pharma, and Kyowa Kirin; a patent assignment fee from FUJIREBIO; and is involved in postmarketing surveillance of lecanemab in Japan. T.I. had received consultancy/speaker fees from Biogen, Eisai, Eli‐Lilly, and Roche/Chugai. This manuscript has been prepared in a neutral and objective manner, and all disclosed financial relationships are not relevant to the content of this work. Author disclosures are available in the Supporting Information.

Supporting information

Supporting Information

Supporting Information

DATA AVAILABILITY STATEMENT

The NACC study data are available from https://www.naccdata.org upon reasonable request.