Abstract
Background
The Montreal Cognitive Assessment (MoCA) is widely used in Parkinson's disease (PD) research to measure cognition, but published cutoffs for mild cognitive impairment (PD‐MCI) and dementia (PDD) vary.
Objective
To identify context‐specific MoCA cutoffs for PD‐MCI, PDD, and impairment on formal neuropsychological testing, and to assess whether combining MoCA with routinely collected clinical measures improves discrimination.
Methods
Data from 1094 people with PD were analyzed. 39.0% were diagnosed with PD‐MCI, 6.2% with PDD. Normative ranges for five neuropsychological tests were derived from healthy controls. Mixed‐effects logistic regression accounted for repeated measures and predicted (1) impairment on formal neuropsychological testing and (2) clinician‐derived cognitive diagnoses. Cutoffs were derived using Youden's J, a prevalence‐matching approach (cutoff where the proportion below threshold equals outcome prevalence), and standard screening and diagnostic criteria.
Results
Education‐adjusted MoCA predicted impairment in two or more cognitive domains with an AUC of 0.86 (95% CI: 0.84–0.88). A cutoff of ≤24 maximized Youden's J (sensitivity 74.7% [70.5–79.3], specificity 83.1% [82.0–84.2]); the prevalence‐matched cutoff was ≤21. For PD‐MCI, the Youden's J cutoff was ≤25 and the prevalence‐matched cutoff ≤24; PDD cutoffs were ≤23 and ≤17. Functional impairment (MDS‐UPDRS 1.1) showed greater concordance with clinician‐assigned diagnosis than MoCA alone.
Conclusion
Optimal MoCA cutoffs vary by intended use. For research stratification, prevalence‐matched cutoffs can align classification with observed outcome prevalence. Generalizability may be limited, particularly to routine clinical populations. Cutoffs derived in the PPMI cohort may over‐identify impairment in broader populations.
Keywords: dementia, diagnosis, functional impairment, neuropsychology, PPMI
An accurate assessment of the cognitive state of people with Parkinson's is crucial not only in clinic, but also in research. 1 , 2 , 3 , 4 , 5 In PD, a neuropsychological test battery comprising two tests for each of the five major cognitive domains has been recommended (“Level‐II” assessment). 6 However, detailed testing is time‐consuming and requires specialist training, so briefer bedside tests are often used instead. One of the currently most widely used tests is the Montreal Cognitive Assessment (MoCA). 7 Although the MoCA assesses domains commonly affected in PD such as executive function, visuospatial function, and attention/working memory, its design and scoring were developed for general population screening rather than PD‐specific cognitive profiles. Numerous studies have derived PD‐specific MoCA cutpoints for mild cognitive impairment (PD‐MCI) and/or Parkinson's disease dementia (PDD). 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 Cutoff derivation faces a two‐fold challenge: the question of generalizability beyond the population in which a cutoff was optimized and the need for context‐specific calibration. Optimal cutoffs are not universal but depend on population demographics and their intended application. Screening typically prioritizes sensitivity, diagnostic use specificity, and research applications such as epidemiological studies, clinical trials, or risk‐modeling frameworks often require particular trade‐offs. Combined with statistical uncertainty, these factors contribute to the wide range of reported MoCA thresholds in the literature, spanning from ≤17 to ≤26 for PD‐MCI and ≤9 to ≤25 for PDD. 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 We therefore aimed to derive context‐specific MoCA cutoffs for multiple cognitive outcomes in PD using complementary optimization criteria, including a prevalence‐matching approach, and to evaluate whether incorporating demographic characteristics and functional impairment improves discrimination.
Methods
Data
We utilized data from the Parkinson's Progression Markers Initiative (PPMI) cohort study. Details on recruitment, inclusion/exclusion criteria and follow‐up are described elsewhere https://www.ppmi-info.org. 21 , 22 Data was downloaded from the PPMI data portal on July 12, 2024 https://www.ppmi-info.org/access-data-specimens/download-data.
Routine variables and assessments include basic participant demographics, the MDS‐Unified Parkinson's Disease Rating Scale (MDS‐UPDRS), 23 and the MoCA. 7 Part 1 of the MDS‐UPDRS assesses non‐motor experiences of daily living. Its question 1 assesses functional impairment due to cognitive deficits as perceived by the patient and/or caregiver on a scale from 0 (no functional impairment) to 4 (severe functional impairment). MoCA total scores were adjusted for education by adding one point for individuals with ≤12 years of education (up to a maximum of 30). The added discriminative value of a more flexible education adjustment was explored within the multivariable prognostic modeling framework.
In the PPMI cohort, detailed testing of executive function, memory and visuospatial function has been conducted since the study began in 2010. Memory has been assessed with the Hopkins Verbal Learning Test‐Revised (HVLT‐R), 24 of which we consider the scores for immediate recall (ie, total recall) and recognition discrimination. Executive function has been assessed with the animal fluency test, 25 Symbol Digit Modalities Test (SDMT) 26 , 27 and Letter‐Number‐Sequencing (LNS). 28 Visuospatial functioning has been assessed with the 15‐item (odd/even) version of the Benton Judgment of Line Orientation (BJL). 29
As part of the PPMI study protocol, each PD participant is categorized as having normal cognition, mild cognitive impairment (PD‐MCI), or dementia (PDD). A diagnosis of PD‐MCI is made if there exists (i) a decline from prior usual cognitive abilities, (ii) impairment in at least one cognitive domain, but (iii) no significant impact on daily functioning. A diagnosis of PDD is made if there exists (i) a decline from prior usual cognitive abilities, (ii) impairment in more than one cognitive domain, and (iii) significant impact of cognitive impairment on daily function. To determine the categorization, the site investigator may review cognitive test performance, consider functional impact reported by the participant, an informant, or observed during the research clinic, as well as their overall clinical impression. The corresponding variable in the PPMI dataset is COGSTATE, which we refer to as the “research clinic diagnosis” to distinguish it from diagnoses made in routine clinical care and from categorizations based solely on cognitive test scores.
Evaluating Neuropsychological Testing
Initially, we attempted to evaluate the formal neuropsychological testing data, utilizing normative ranges from existing literature, but we found the percentages of patient visits with test scores <1.5 standard deviations (SDs) below norm to vary greatly across the different tests when applying respective published norms. For the BJL, for example, the percentage of PD <1.5 SDs below norm 30 , 31 was less than the 6.5% expected in a healthy population. Similar findings were recently reported by Weintraub et al. 32 We therefore opted to utilize the neuropsychological testing data available for the control arm of the PPMI cohort to derive our own age‐, sex‐ and education‐adjusted normative ranges for LNS, SDMT, animal fluency, BJL, HVLT‐immediate recall and HVLT‐discrimination using mixed‐effects linear and tobit regression. The latter was used if floor or ceiling effects were evident in the statistical distribution of the test scores. Details are provided in the Supplementary Material.
Prognostic Prediction Models
We consider four different outcomes: (i) impairment in ≥1 domain in formal neuropsychological testing, (ii) impairment in ≥2 domains, (iii) research clinic diagnosis of cognitive impairment (PD‐MCI/PDD vs. PD‐NC) and (iv) research clinic diagnosis of dementia (PDD vs. PD‐NC/PD‐MCI).
We note that defining impairment as <1.5 SDs in one or two domains corresponds to a one‐domain vs. multi‐domain impairment distinction, rather than an MCI versus dementia analogue, as dementia typically requires more severe deficits (often ≥2 SDs below norm) and functional impairment.
We first assess the discriminative ability of the education adjusted MoCA totals. We then proceed to study if discriminative ability can be improved by (i) incorporating routinely collected information on basic demographics (age, sex, and years of education), (ii) incorporating functional impairment as assessed in MDS‐UPDRS 1.1, or (iii) utilizing MoCA subscores for visuospatial/executive, naming, attention, language, abstraction, delayed recall, and orientation. 7 , 33 Models including education years as a covariate used unadjusted MoCA totals. The MDS‐UPDRS 1.1 responses are treated as a factor variable with categories 3 (moderate) and 4 (severe) merged into a single factor due to severe impairment being rarely recorded (only 10 times in a total 4700 patient visits, 0.2%, Table 1). For the research clinic diagnosis outcomes we further assess the discriminative ability of the neuropsychology test scores for the three tested domains s executive , s memory and s visuospatial in comparison to the MoCA. Details on the derivation of the domain test scores are provided in the Supplementary Material. As we use longitudinal data and therefore non‐independent observations, we employ mixed‐effects logistic regression with a random intercept for each patient. Details including software packages used are provided in the Supplementary Material.
TABLE 1.
Demographics and cognition in the PD arm of the PPMI cohort stratified by the number of domains impaired according to neuropsychological testing (top) and by research clinic diagnosis (bottom)
| Impaired domains (neuropsychology) | ||||
|---|---|---|---|---|
| 0 | 1 | 2 | 3 | |
| N [patient visits] | 3431 | 881 | 297 | 91 |
| Age | 65.1 (9.4) | 66.5 (10.2) | 68.5 (10.4) | 67.3 (10.9) |
| Female | 37.5% (1286) | 40.2% (354) | 35.7% (106) | 38.6% (35) |
| Education years | 15.9 (3.2) | 15.3 (3.5) | 14.8 (4.2) | 14.3 (4.7) |
| PD duration [years] | 5.4 (3.3) | 5.4 (3.3) | 6.2 (3.5) | 6.5 (3.3) |
| MoCA total (adj) a | 27.4 (2.4) | 25.3 (3.2) | 22.4 (4.0) | 18.0 (4.8) |
| MDS‐UPDRS 1.1 (functional impairment) | ||||
| 0 (none) | 65.5% (2244) | 50.3% (442) | 37.4% (111) | 15.4% (14) |
| 1 (slight) | 27.7% (948) | 34.1% (299) | 34.7% (103) | 39.6% (36) |
| 2 (mild) | 5.6% (192) | 11.5% (101) | 16.9% (50) | 25.3% (23) |
| 3 (moderate) | 1.2% (40) | 3.8% (33) | 10.4% (31) | 14.3% (13) |
| 4 (severe) | 0.0% (0) | 0.3% (3) | 0.7% (2) | 5.5% (5) |
| Research clinic diagnosis | ||||
| PD‐NC | 85% (2829) | 67.8% (576) | 41.9% (121) | 20.9% (19) |
| PD‐MCI | 14.5% (484) | 29.7% (252) | 44.6% (129) | 49.5% (45) |
| PDD | 0.5% (17) | 2.5% (21) | 13.5% (39) | 29.7% (27) |
| Research clinic diagnosis | |||
|---|---|---|---|
| PD‐NC | PD‐MCI | PDD | |
| N [patient visits] | 3545 | 910 | 104 |
| Age | 64.5 (9.5) | 70.0 (8.5) | 72.4 (9.2) |
| Female | 40.1% (1420) | 31.2% (284) | 32.7% (34) |
| Education years | 15.8 (3.4) | 15.5 (3.3) | 14.7 (4.3) |
| PD duration [years] | 5.3 (3.2) | 6.6 (3.5) | 7.6 (3.1) |
| MoCA total (adj) a | 27.1 (2.6) | 24.6 (3.7) | 19.4 (5.1) |
| MDS‐UPDRS 1.1 (functional impairment) | |||
| 0 (none) | 71.5% (2532) | 19.6% (178) | 4.8% (5) |
| 1 (slight) | 25.9% (916) | 47.0% (427) | 10.6% (11) |
| 2 (mild) | 2.4% (85) | 26.1% (237) | 36.5% (38) |
| 3 (moderate) | 0.3% (10) | 7.2% (65) | 40.4% (42) |
| 4 (severe) | 0.0% (0) | 0.2% (2) | 7.7% (8) |
| Impaired domains (neuropsychology) | |||
| 0 | 80.0% (2827) | 53.3% (484) | 16.4% (17) |
| 1 | 16.3% (576) | 27.6% (251) | 20.2% (21) |
| 2 | 3.5% (121) | 14.2% (129) | 37.5% (39) |
| 3 | 0.5% (19) | 4.9% (45) | 26.0% (27) |
Note: The table summarizes the follow‐up visits from 1094 different patients. Shown are means and standard deviations or percentages and counts.
The MoCA scores were adjusted for education by adding one point for individuals with ≤12 years of education (up to a maximum of 30).
Abbreviations: MoCA = Montreal Cognitive Assessment; PD‐NC = PD‐Normal Cognition; PD‐MCI = PD‐Mild Cognitive Impairment; PDD = Parkinson's Disease Dementia.
The models first combine an individual's scores and demographics (the predictor values x m ) with the corresponding model coefficients (β M,m ) into a single score, the linear predictor lp. This score is then converted into a predicted probability of an outcome (eg, PDD) between 0 and 1 using the logistic function:
Cutoffs
Youden's J
Youden's J is a widely used criterion for selecting an optimal cutoff and is defined as sensitivity plus specificity minus one. 34 Youden's J implicitly assigns equal weight to false positives and false negatives. This allows for a convenient single‐number summary of statistical test performance, but is rarely aligned with real‐world clinical or research needs.
Prevalence‐Matched
The prevalence‐matched cutoff is defined as the cutoff at which the proportion of individuals scoring below the threshold matches the prevalence of the respective outcome. In this study, the reference prevalence was the observed prevalence in the PPMI cohort. Because the data are longitudinal, prevalence for cutoff calibration was defined at the observation level rather than the participant level. The MoCA is a discrete scale, thus the equality is typically only approximate.
The prevalence matching approach is particularly suited for studies aiming to estimate the incidence, prevalence, or risk of PD–MCI or PDD using the MoCA as a proxy outcome. Although prevalence matching is not a standard cutoff selection method, it is conceptually equivalent to equipercentile equating. 35 At the prevalence‐matched cutoff, false positives and false negatives occur in approximately equal numbers. Like Youden's J, this approach is not intended for clinical screening or diagnosis, where the consequences of false‐positive and false‐negative decisions typically differ.
Screening and Diagnosis
Screening prioritizes minimizing missed cases, whereas diagnosis prioritizes minimizing false positives. Following conventions from previous MoCA studies in PD, 8 , 9 , 14 , 19 we report the highest cutoff with sensitivity ≥80% and negative predictive value (NPV) ≥80% as a potential screening threshold and the lowest cutoff with specificity ≥80% and positive predictive value (PPV) ≥80% as a potential diagnostic threshold. PPV and NPV were calculated using the observed prevalence of each outcome within the PPMI cohort. For clinical decision‐making, the choice of cutoff has direct implications for patients, and the preferred balance between false‐positives and false‐negatives depends on the specific clinical context, including the consequences of misclassification for the patient and subsequent care pathways. 36 To support context‐specific decision‐making, we provide a range of cutoffs together with their respective empirical properties in the Supplementary Material. Predictive values associated with the proposed screening and diagnostic thresholds will vary according to the prevalence of the outcome in the target population. Therefore, guidance and equations for calculating PPV and NPV at any user‐specified prevalence are provided in the Supplementary Methods.
Results
Demographics and Cognitive Outcomes
We analyzed data from 1094 participants with PD in the main analyses, ie, diagnostic prediction modeling and cutoff derivation, and 267 healthy controls (HC) in the development of the normative ranges. Basic demographics and cognitive test scores are summarized in Table 1 and Supplementary Table 1. PD participant and observation counts differ between these tables because different inclusion criteria were applied. Table 1 summarizes the cohort used for the main analyses, for which visits were required to have a complete MoCA assessment, demographic data, and the relevant cognitive outcome measure (neuropsychology or clinician diagnosis). Supplementary Table 1 summarizes a larger cohort, which required only complete neuropsychological and demographic data but not a MoCA assessment. Because the MoCA was not collected at baseline in PPMI, some participants were included in Supplementary Table 1 but not in the main analyses. Detailed information on the normative ranges for the neuropsychological test battery, which were derived from the PPMI controls data is provided in the Supplementary Methods and Supplemental Table 2–4.
427 participants with PD (39.0%) were diagnosed with PD‐MCI, 62 participants (6.2%) with PDD during follow‐up. Across PD visits, neuropsychology indicated impairment in one domain at 881 visits (18.7%), two domains at 297 (6.3%) visits, and all three domains tested at 91 visits (1.9%). PD‐MCI was diagnosed at 920 (20%) visits, PDD at 104 (2.3%) visits (Table 1). As expected, MoCA scores are decreased and functional impairment increased with less favorable cognitive outcomes (Table 1, Fig. 1). Notably, whereas nearly half of PDD cases (47.1%) were reported to experience moderate to severe functional impairment, only 19.8% of cases with impairment in all three tested domains experienced moderate to severe functional impairment, underscoring that research clinic diagnosis of PDD incorporated functional information beyond the test battery. Observation‐level prevalences used to derive prevalence‐matched cutoffs were 2.3% for PDD versus PD without dementia, 22.2% for PD‐MCI/PDD versus PD with normal cognition, 8.3% for ≥2 versus <2 impaired domains, and 27.0% for ≥1 versus 0 impaired domains.
Figure 1.

Education‐adjusted MoCA scores stratified by (A) number of domains impaired in neuropsychological testing, (B) cognitive research clinic diagnosis and (C) functional impairment. Means, standard deviations and numbers of observations are listed in Table 1.
Predicting Neuropsychology Outcomes
The education‐adjusted MoCA total scores discriminate patients with ≥1 impaired domain from those with no impairment in formal neuropsychological testing with an AUC of 0.76, and patients with ≥2 impaired domains from those with one or none with an AUC of 0.86 (Table 2, Fig. 2A,B). Incorporating demographics, MDS‐UPDRS 1.1 or using MoCA subscores only marginally improves discrimination, increasing the AUC by ≤0.01 (Table 2, Fig. 3A,B). Model coefficients are listed in Supplemental Table 5, 6.
TABLE 2.
Area under the receiver operating characteristic curve (AUC) of prediction models for four different outcomes utilizing different sets of covariates. Model optimism (Bias) and root‐mean‐square‐error (RMSE) were estimated from n = 200 bootstrap iterations
| Outcome | Covariates | AUC (corr.) a | Bias | RMSE | AUC (orig.) b |
|---|---|---|---|---|---|
| ≥1 impaired domain (neuropsychology) | MoCA total (adj) | 0.757 | −0.000 | 0.012 | 0.756 |
| MoCA total, demographics | 0.755 | 0.004 | 0.012 | 0.759 | |
| MoCA subscores | 0.763 | 0.001 | 0.011 | 0.764 | |
| MoCA subscores, demographics | 0.760 | 0.005 | 0.012 | 0.765 | |
| MDS‐UPDRS 1.1 | 0.622 | −0.001 | 0.013 | 0.622 | |
| MoCA total (adj), MDS‐UPDRS 1.1 | 0.763 | 0.002 | 0.012 | 0.765 | |
| MoCA total, demogr., MDS‐UPDRS 1.1 | 0.766 | 0.002 | 0.012 | 0.769 | |
| ≥2 impaired domains (neuropsychology) | MoCA total (adj) | 0.861 | −0.000 | 0.014 | 0.861 |
| MoCA total, demographics | 0.862 | 0.001 | 0.014 | 0.864 | |
| MoCA subscores | 0.869 | 0.002 | 0.011 | 0.871 | |
| MoCA subscores, demographics | 0.870 | 0.003 | 0.012 | 0.873 | |
| MDS‐UPDRS 1.1 | 0.687 | −0.002 | 0.018 | 0.685 | |
| MoCA total (adj), MDS‐UPDRS 1.1 | 0.866 | 0.001 | 0.013 | 0.867 | |
| MoCA total, demogr., MDS‐UPDRS 1.1 | 0.865 | 0.004 | 0.013 | 0.869 | |
| Impairment (research clin. diag.) | MoCA total (adj) | 0.738 | −0.002 | 0.014 | 0.736 |
| MoCA total, demographics | 0.756 | 0.000 | 0.012 | 0.756 | |
| MoCA subscores | 0.740 | 0.002 | 0.013 | 0.742 | |
| MoCA subscores, demographics | 0.762 | 0.004 | 0.012 | 0.766 | |
| Domain scores | 0.728 | 0.001 | 0.013 | 0.729 | |
| Domain scores, demographics | 0.775 | 0.001 | 0.012 | 0.776 | |
| MDS‐UPDRS 1.1 | 0.812 | 0.000 | 0.008 | 0.812 | |
| MoCA total (adj), MDS‐UPDRS 1.1 | 0.863 | 0.000 | 0.008 | 0.863 | |
| MoCA total, demogr., MDS‐UPDRS 1.1 | 0.870 | 0.001 | 0.008 | 0.871 | |
| Dementia (research clin. diag.) | MoCA total (adj) | 0.897 | 0.001 | 0.015 | 0.898 |
| MoCA total, demographics | 0.883 | 0.004 | 0.019 | 0.887 | |
| MoCA subscores | 0.897 | 0.007 | 0.016 | 0.903 | |
| MoCA subscores, demographics | 0.891 | 0.011 | 0.022 | 0.901 | |
| Domain scores | 0.887 | 0.002 | 0.020 | 0.889 | |
| Domain scores, demographics | 0.915 | 0.006 | 0.014 | 0.921 | |
| MDS‐UPDRS 1.1 | 0.914 | 0.003 | 0.017 | 0.917 | |
| MoCA total (adj), MDS‐UPDRS 1.1 | 0.963 | 0.001 | 0.008 | 0.964 | |
| MoCA total, demogr., MDS‐UPDRS 1.1 | 0.961 | 0.004 | 0.010 | 0.964 |
Bias corrected AUC (AUC corr. = AUC orig–Bias).
Original AUC.
Figure 2.

Discriminative ability of education‐adjusted MoCA scores. ROC curves for four outcomes: (A) ≥1 impaired neuropsychological domain, (B) ≥2 domains, (C) diagnosis of any impairment vs. none, (D) dementia vs. no‐dementia. Youden's J in blue; Prevalence‐matched cutoff in red.
Figure 3.

ROC curves for education‐adjusted MoCA scores, MDS‐UDRDS 1.1 and selected prediction models (see legend in panels) for four outcomes: (A) ≥1 impaired domain, (B) ≥2 domains, (C) any impairment vs. none, (D) dementia vs. no‐dementia. AUCs are listed in Table 2.
Internal bootstrap validation shows no indication of overfitting for any of the models (Table 2) and reasonable calibration with some tendency towards too moderate estimates (calibration slopes >1, Supplemental Table 7, 8, Supplemental Figure 1).
Cutoffs
Table 3 summarizes selected cutoffs for patient stratification. As no model leads to a substantial improvement over the MoCA, we consider cutoffs on the original MoCA scale (Fig. 3A,B). Youden's J is maximized at a MoCA cutpoint of ≤26 for stratifying ≥1 impaired domain; ≤ 24 for stratifying ≥2 impaired domains. Prevalence‐matched cutoffs are ≤26 and ≤21, respectively, with sensitivities of 57.4% and 49.7%, meaning that, when stratifying to match outcome prevalence, about half of the cases with the respective outcome are correctly identified. (See Supplemental Table 9, 10 for a comprehensive list of MoCA cutoffs standalone and in conjunction with functional impairment).
TABLE 3.
Selected cutoffs for four outcomes and their sensitivity, specificity, positive and negative predictive values and accuracy with 95% confidence intervals
| Criterion | Sensitivity | Specificity | PPV | NPV | Accuracy | |
|---|---|---|---|---|---|---|
| ≥1 Impaired domain (neuropsychology) | ||||||
| MoCA (adj) | ||||||
| Youden's J | ≤26 | 68.0 (65.4, 70.6) | 69.5 (68.0, 71.0) | 45.2 (43.0, 47.4) | 85.5 (84.1, 86.7) | 69.1 (67.8,70.3) |
| Prev‐matched | ≤25 | 57.4 (54.6, 60.2) | 80.1 (78.6, 81.4) | 51.6 (49.0, 54.2) | 83.5 (82.3, 84.8) | 73.9 (72.6, 75.2) |
| Screening | ≤28 | 88.7 (86.9, 90.5) | 39.5 (37.8, 41.1) | 35.1 (33.5, 36.8) | 90.5 (88.9, 91.9) | 52.8 (51.3, 54.2) |
| Diagnostic | ≤21 | 23.7 (21.4, 26.1) | 97.8 (97.3, 98.3) | 80.0 (76.1, 84.0) | 77.6 (76.4, 78.8) | 77.8 (76.7, 79.0) |
| ≥2 impaired domains (neuropsychology) | ||||||
| MoCA (adj) | ||||||
| Youden's J | ≤24 | 74.7 (70.5, 79.3) | 83.1 (82.0, 84.2) | 28.5 (25.8, 31.2) | 97.3 (96.8, 97.9) | 82.5 (81.3, 83.5) |
| Prev‐matched | ≤21 | 49.7 (44.7, 55.0) | 95.8 (95.2, 96.3) | 51.3 (46.2, 56.5) | 95.5 (94.8, 96.1) | 92.0 (91.2, 92.7) |
| Screening | ≤25 | 81.4 (77.6, 85.3) | 74.6 (73.2, 75.9) | 22.4 (20.3, 24.5) | 97.8 (97.3, 98.3) | 75.2 (73.8, 76.4) |
| Diagnostic | ≤15 | 10.3 (7.5, 13.4) | 99.8 (99.7, 99.9) | 81.6 (70.2, 92.0) | 92.5 (91.8, 93.3) | 92.4 (91.7, 93.2) |
| Impairment (research clinic diagnosis) | ||||||
| MoCA (adj) | ||||||
| Youden's J | ≤25 | 58.0 (55.1, 61.0) | 78.1 (76.7, 79.4) | 43.1 (40.5, 45.6) | 86.7 (85.5, 87.8) | 73.6 (72.3, 74.9) |
| Prev‐matched | ≤24 | 47.0 (43.9, 50.0) | 85.7 (84.6, 86.8) | 48.5 (45.4, 51.6) | 85.0 (83.8, 86.1) | 77.1 (75.9, 78.3) |
| Screening | ≤28 | 87.2 (85.1, 89.2) | 37.5 (35.9, 39.1) | 28.5 (26.9, 30.0) | 91.1 (89.6, 92.5) | 48.5 (47.1, 49.9) |
| Diagnostic | ≤16 | 5.4 (4.1, 6.8) | 99.6 (99.4, 99.8) | 80.9 (70.7, 89.3) | 78.7 (77.5, 79.8) | 78.7 (77.5, 79.9) |
| MDS‐UPDRS 1.1 ≥1 and MoCA (adj) | ||||||
| Youden's J | (≤30) | 81.9 (79.7, 84.2) | 71.5 (69.9, 73.0) | 45.1 (42.9, 47.3) | 93.3 (92.4, 94.2) | 73.8 (72.5, 75.1) |
| Prev‐matched | ≤26 | 57.2 (54.3, 60.4) | 88.9 (87.8, 89.9) | 59.6 (56.5, 62.8) | 87.9 (86.8, 89.0) | 81.8 (80.7, 83.0) |
| Screening | (≤30) | 81.9 (79.7, 84.2) | 71.5 (69.9, 73.0) | 45.1 (42.9, 47.3) | 93.3 (92.4, 94.2) | 73.8 (72.5, 75.1) |
| Diagnostic | ≤21 | 19.8 (17.4, 22.3) | 98.6 (98.2, 99.0) | 80.4 (75.3, 85.2) | 81.1 (80.0, 82.3) | 81.1 (80.0, 82.2) |
| Model‐estimated probability P* | ||||||
| Youden's J | ≥0.19 | 78.4 (75.9, 80.9) | 77.4 (76.1, 78.8) | 49.8 (47.5, 52.2) | 92.6 (91.7, 93.5) | 77.7 (76.5,78.8) |
| Prev‐matched | ≥0.32 | 63.3 (60.5, 66.1) | 89.6 (88.7, 90.6) | 63.6 (60.7, 66.5) | 89.5 (88.5, 90.5) | 83.8 (82.8, 84.8) |
| Screening | ≥0.17 | 82.6 (80.3, 84.8) | 71.6 (70.2, 73.1) | 45.4 (43.3, 47.7) | 93.5 (92.6, 94.4) | 74.0 (72.9, 75.3) |
| Diagnostic | ≥0.57 | 42.2 (39.2, 45.2) | 97.0 (96.5, 97.5) | 80.1 (77.0, 83.4) | 85.4 (84.3, 86.5) | 84.8 (83.8, 85.8) |
| Dementia (research clinic diagnosis) | ||||||
| MoCA (adj) | ||||||
| Youden's J | ≤23 | 77.9 (70.1, 85.6) | 85.8 (84.8, 86.8) | 11.4 (9.2, 13.7) | 99.4 (99.2, 99.6) | 85.6 (84.6, 86.6) |
| Prev‐matched | ≤17 | 32.7 (24.0, 41.9) | 98.6 (98.3, 99.0) | 35.4 (26.9, 45.5) | 98.4 (98.1, 98.8) | 97.1 (96.6, 97.6) |
| Screening | ≤25 | 91.3 (85.7, 96.3) | 71.5 (70.2, 72.8) | 7.0 (5.7, 8.3) | 99.7 (99.5, 99.9) | 72.0 (70.6, 73.2) |
| Diagnostic | no MoCA cutoff satisfied the PPV ≥80%; specificity ≥80% criterion | |||||
| MDS‐UPDRS 1.1 ≥2 and MoCA (adj) | ||||||
| Youden's J | ≤29 | 84.6 (78.2, 91.2) | 91.6 (90.8, 92.3) | 19.0 (15.7, 22.7) | 99.6 (99.4, 99.8) | 91.4 (90.6, 92.2) |
| Prev‐matched | ≤20 | 46.1 (36.9, 55.9) | 99.0 (98.7, 99.3) | 52.2 (42.4, 62.5) | 98.8 (98.4, 99.1) | 97.8 (97.4, 98.2) |
| Screening | 28 | 83.7 (76.9, 90.4) | 92.3 (91.5, 93.0) | 20.2 (16.7, 24.0) | 99.6 (99.4, 99.8) | 92.1 (91.3, 92.8) |
| Diagnostic | no MoCA cutoff satisfied the PPV ≥80%; specificity ≥80% criterion | |||||
| Linear predictor lp** | ||||||
| Youden's J | ≥−3.41 | 91.3 (85.4, 96.2) | 89.5 (88.7, 90.4) | 16.9 (13.9, 20.0) | 99.8 (99.6, 99.9) | 89.6 (88.7, 90.4) |
| Prev‐matched | ≥−1.27 | 52.9 (43.3, 61.7) | 98.8 (98.5, 99.1) | 51.4 (41.8, 61.1) | 98.9 (98.6, 99.2) | 97.8 (97.4, 98.2) |
| Screening | ≥−2.31 | 81.7 (74.3, 88.7) | 96.0 (95.5, 96.6) | 32.6 (27.0, 38.3) | 99.6 (99.4, 99.7) | 95.7 (95.2, 96.3) |
| Diagnostic | ≥1.82 | 4.8 (1.0, 9.3) | 100.0 (99.9, 100.0) | 83.3 (50.0, 100.0) | 97.8 (97.4, 98.2) | 97.8 (97.4, 98.2) |
Note: Domain impairment defined as <1.5 SDs below norm reflects a one‐domain vs. multi‐domain impairment distinction and is not intended as an MCI vs. dementia analogue. *P probability of any impairment estimated by model with covariates MoCA total (adj) and MDS‐UPDRS 1.1; **lp linear predictor of dementia model with covariates MoCA total (adj) and MDS‐UPDRS 1.1.
Abbreviations: PPV: Positive Predictive Value; NPV: Negative Predictive Value; Prev‐matched: Prevalence‐matched.
Predicting Research Clinic Diagnoses
The education‐adjusted MoCA total scores discriminate patients with research clinic diagnosis of any cognitive impairment (PD‐MCI/PDD) from those with normal cognition with an AUC of 0.74; those with research clinic diagnosis of dementia from those with no dementia diagnosis with an AUC of 0.90 (Table 2, Fig. 2C,D). Incorporating demographic covariates improves discrimination of any cognitive impairment (AUC = 0.76; P = 0.001) with age being the most relevant demographic covariate. Of two patients with the same MoCA total score, the older patient is more likely diagnosed with cognitive impairment or dementia at a research clinic visit (P ≤ 0.001, Supplementary Table 11, 12).
The greatest improvement in discrimination is achieved by adding the MDS‐UPDRS 1.1 as a model covariate. This holds for both comparisons, discriminating any cognitive impairment from normal cognition (AUC = 0.86) as well as discriminating dementia from no dementia (AUC = 0.96). Clinician‐assigned diagnosis is not an independent reference standard, hence AUC values are likely inflated due to diagnostic incorporation. The level of functional impairment appears to be a more important factor in research clinic diagnosis than MoCA performance (Table 2, Fig. 3C,D). For predicting any impairment all factor levels are relevant (OR M ≥5, P < 0.001), for predicting dementia diagnosis only MDS‐UDPRS 1.1 scores of 2 or greater are predictive (OR M >5, P < 0.001). The domain scores derived from the neuropsychological test battery were not superior compared to the MoCA in predicting research clinic diagnosis.
Internal bootstrap validation shows no indication of overfitting for any of the models. The models predicting any cognitive impairment show reasonable calibration. The models predicting dementia diagnosis are not reasonably calibrated (Supplementary Table 7, 8 and Supplementary Figure 2). In this case model‐estimated probabilities should not be interpreted as such.
Cutoffs
Youden's J is maximized at a MoCA cutpoint of ≤25 for discriminating any cognitive impairment from normal cognition, and≤23 for discriminating dementia from no dementia (Table 3). Prevalence‐matched cutoffs are ≤24 and ≤17, respectively, with sensitivities of 47.0% and 32.7%. Table 3 further lists the cutoffs for the best‐performing model which combines education‐adjusted MoCA totals and MDS‐UPDRS 1.1.
Also a simpler approach of combining a MoCA cutoff with a MDS‐UPDRS 1.1 cutoff improves discrimination compared to a pure MoCA cutoff. These improvements should, however, be interpreted with caution, as discrimination may be inflated by diagnostic incorporation and reflects approximation of clinician judgment rather than validation against an independent reference standard. For discriminating any cognitive impairment from none, a combination of MDS‐UPDRS 1.1 ≥1 and MoCA ≤26 optimizes with respect to our prevalence‐matching criterion with a sensitivity of 57.2%, an increase of 10% compared to MoCA totals only. For discriminating dementia from no dementia, a combination of MDS‐UPDRS 1.1 ≥2 and MoCA ≤20 optimizes with respect to the prevalence‐matching criterion with a sensitivity of 46.1%, an increase of about 13% compared to MoCA totals only. If in favor of a slightly higher sensitivity, one might alternatively choose the combination of MDS‐UPDRS 1.1≥2 and MoCA ≤21, as the ‘true’ optimum is estimated to lie in between.
Discussion
The MoCA predicts neuropsychological testing outcomes and cognitive diagnoses with AUCs between 0.73–0.90, a range commonly regarded “acceptable” at its lower end and “excellent” at its upper end. 37 Our finding that the domain scores derived from the neuropsychological test battery are not superior predictors of research clinic diagnosis compared to the MoCA totals may seem surprising at first. However, this is no indication that formal neuropsychology is in any way inferior to the MoCA, but suggests that the performance on the MoCA plays an equally great role in the research clinic diagnosis. Reviewing performance on 5 (or more) tests is more time‐consuming and might require re‐consideration of the respective normative ranges, while the interpretation of MoCA scores is comparably straightforward. Similarly, our finding that functional impairment, as recorded in MDS‐UPDRS 1.1, outperforms the MoCA in predicting the research clinic diagnosis should not be interpreted as evidence of true superiority. Rather, it reflects diagnostic incorporation and the relative weighting that clinicians give to functional impairment. These findings are consistent with a broader literature demonstrating that assessment of functional impairment provides complementary information to cognitive screening and can improve diagnostic and prognostic accuracy in neurodegenerative disorders. 38 , 39 , 40
In the PPMI cohort, in the absence of a clinician‐derived diagnostic categorization, functional impairment serves as a stronger proxy than the MoCA alone, and discriminative performance is optimized when both measures are combined. The finding of better discrimination and larger AUC for the more severe cognitive outcomes is, from a statistical perspective, not surprising. The statistical distribution of MoCA scores at the corresponding lower end is less dense, contributing to reduced overlap between impaired and non‐impaired groups in that region.
As a sensitivity analysis, we examined the stability of discriminative performance in a cross‐sectional dataset limited to the most recent measurement for each participant. Discriminative performance was broadly similar to that obtained from the full dataset, indicating that the use of repeated measurements did not materially alter our main findings (Supplementary Table 13, 14, 15). The largest differences emerged for the clinician‐diagnosed outcome of any cognitive impairment (PD‐MCI or PDD) versus no impairment, for which AUC, sensitivity, and specificity were modestly higher when using only the last observation. This pattern likely reflects the wider spread and lower values of MoCA scores at the latest visits, which enhances separation between impaired and non‐impaired groups.
A core emphasis of our study is that there is no single optimal cutoff; rather, optimality depends on the specific purpose and context of use. We would like to emphasize that for in‐clinic use with direct effect on patients, such as decisions on specialist referrals, counseling or treatments, cutoff selection should be made in a situation‐specific manner taking into consideration benefits and potential harms to patients in the situation at hand. In pure research applications, on the other hand, for instance when harmonizing data from distinct cohorts for incidence or prevalence studies, or when developing prediction models for future cognitive decline or dementia we recommend considering a prevalence‐matched cutoff. It is explicitly selected such that the percentage of observations below cutoff equals the prevalence of the respective outcome, which provides an intuitive interpretation. The Supplementary Material provides the discriminative properties of a wide range of cutoffs to inform decisions in individual clinical or research contexts. Ongoing work builds on the derived cutoffs to define outcomes for multivariable prognostic models of individualized risk of cognitive impairment and dementia, incorporating clinical and genetic risk factors.
The number of PDD cases in PPMI is lower than meta‐analytic incidence estimates of 4.5% per 100 person‐years, 2 in line with the cohort's younger, drug‐na¨ıve profile at baseline, its highly educated participants, and the high level of motivation required to meet the demands of the study. This underscores the need to carefully consider the generalizability of our findings. Given the variety of literature on PD‐specific MoCA cutoffs, we here only explicitly compare our findings with the two studies with the greatest sample sizes known to us. In Kim et al, 16 cutoffs were identified by optimizing Youden's J. For discriminating PDD from PD without dementia, they found the same cutoff of ≤23 as the present study. The optimal cutoff to discriminate PD‐MCI of ≤26 reported in Kim et al is one point higher than in the present study. Thus, all in all, our findings are consistent with Kim et al.
The study with the largest sample size we are aware of is by Fiorenzato et al. 19 They derived screening and diagnostic cutoffs using the same criteria as the present study, but concluded throughout lower cutoff values, for instance a screening cutoff of 17 for dementia. The mean MoCA score in their PD‐NC group was 24.5, 20.2 in the PD‐MCI group, and 13.4 in the PDD group, which is several points lower than in our study. Their reference, a clinical diagnosis, is based on the MDS diagnostic criteria for MCI and dementia 41 , 42 including neuropsychological testing of all five cognitive domains evaluated using respective Italian norms. While not fully equivalent in terms of study design, the magnitude of the differences between our findings and Fiorenzato et al is surprising. In the original publication of the MoCA, a score of ≤25 was described to signify mild cognitive impairment. A mean MoCA score below that threshold in a group classified as normal cognition therefore appears counterintuitive at first glance. However, several studies 43 , 44 , 45 found strong associations between MoCA scores and education levels and in particular, a more pronounced decline with age in those who spent fewer years in education, suggesting that the standard one‐point adjustment may not fully account for educational effects. These findings further suggest that the difference in education levels between the study population of Fiorenzato et al and the PPMI cohort might explain some of the differences in derived cutoffs. However, Fiorenzato et al reported that, just like in the regression models of the present study, demographic variables, including years of education and age, were not among the key features of their decision trees.
These discrepancies highlight the challenges of defining PD‐MCI and PDD across different geographical and cultural contexts, as well as between research cohorts and broader clinical populations. The PPMI cohort is predominantly drawn from North America and Western Europe, with relatively high educational attainment, whereas the cohort studied by Fiorenzato et al represents a large nationwide Italian clinical sample recruited across three specialist centers and characterized by lower and more heterogeneous educational levels and older age distributions. MoCA performance is known to vary across regions due to differences in language versions, educational systems, and cultural familiarity with test content 46 , 47 Our cutoffs, empirically derived from the PPMI cohort, may therefore not reliably generalize to populations with distinct demographic, cultural, or clinical characteristics, particularly when applied beyond selected research cohorts to broader clinical populations, where they may over‐identify impairment.
A diagnosis made by a clinician or the results of higher‐order cognitive testing are respected references to validate the MoCA against, but neither provides a perfectly accurate representation of a patient's true cognitive state. While diagnostic criteria are outlined in great detail in the literature, 41 , 42 , 48 they include passages that inherently leave room for interpretation, such as “Impairment on neuropsychological tests may be demonstrated in several ways: performance between 1 to 2 standard deviations (SDs) below age, education, sex, and culturally appropriate norms, […].” 42 Already a choice of a cutoff of either 1 or 1.5 SDs below normative expectations will systematically shift patient classifications, and consequently influence the MoCA cutoffs that best align with such categorization. Further, as we have found when working with different published norms, classification outcomes substantially depend on the choice of normative data sets. Deriving our own internally consistent normative ranges for all tests using data from the control arm of the PPMI cohort and employing Tobit regression, where necessary, to account for ceiling effects constitutes one of the strengths of our study. Although the PPMI control cohort does not apply a cognitive exclusion criterion, inspection of the COGSTATE variable indicated that only 3.5% of observations were categorized as MCI and 0.2% as dementia, which is well below the proportion of healthy individuals expected to fall below common impairment thresholds in neuropsychological testing. Any resulting distortion of percentile‐based norms is therefore likely to be minimal. However, because MoCA performance and functional impairment contribute to the PPMI research diagnoses, associations with diagnostic outcomes and discrimination estimates may be inflated due to criterion contamination.
Taken together, our findings highlight that there is no universally optimal MoCA cutoff for cognitive impairment in PD; the appropriate cutoff depends on the context in which it is applied. In clinical practice, where decisions directly affect patients, cutoffs should be chosen in a situation‐specific manner, weighing potential benefits and harms. For research applications, particularly when estimating incidence or prevalence, or developing prediction models, a prevalence‐matched cutoff offers a transparent and interpretable choice. When drawing on the literature for suitable cutoffs, we recommend assessing not only sample sizes but also how closely the population of interest resembles the cohort from which a cutoff was derived, as the divergence of proposed cutoffs across studies underscores their population‐specific nature and limited generalizability. Given the heterogeneity of published PD‐specific MoCA cutoffs, a systematic comparison and meta‐analysis is warranted, and future work should examine the generalisability of MDS‐UPDRS 1.1–based cutoffs relative to the MoCA across diverse populations.
Author Roles
(1) Research Project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript Preparation: A. Writing of the First Draft, B. Review and Critique.
T.Z.: 2A, 2B, 2C, 3A, 3B.
S.G.M.: 2A, 2C, 3B.
J.R.: 1B, 1C, 3B.
F.A.H.: 1C, 3B.
K.G.: 1C, 3B.
J.W.: 1B, 1C, 3B.
T.E.: 2C, 3B.
D.W.: 1A, 1B, 1C, 3B.
J.K.: 2C, 3B.
Y.B.S.: 2A, 2C, 3B.
M.T.H.: 1B, 1C, 2A, 2C, 3B.
M.L.: 2A, 2C, 3B.
Disclosures
Ethical Compliance Statement: The PPMI study was conducted in accordance with the Declaration of Helsinki and the Good Clinical Practice (GCP) guidelines, and approved by the local ethics committees of the participating sites. 22 Participants have given written consent. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this work is consistent with those guidelines.
Financial Disclosures and Conflicts of Interest: The authors have no conflict of interest to declare. Author disclosures are available in the Supporting Information.
Supporting information
Data S1. Supporting Information
Supplementary Table 1. Basic demographics and cognitive test scores in the PPMI cohort.
Supplementary Table 2. Model coefficients for age, sex and education correction of HVLT‐immediate recall, HVLT‐discrimination, animals fluency, SDMT, LNS, BJL scores.
Supplementary Table 3. Mean and standard deviation for z‐scoring age, sex and education corrected test scores.
Supplementary Table 4. Cut‐offs defining impairment in each cognitive domain tested.
Supplementary Table 5. Model coefficients for predicting impairment according to neuropsychological testing (≥1 impaired domain).
Supplementary Table 6. Model coefficients for predicting impairment according to neuro‐ psychological testing (≥2 impaired domains).
Supplementary Table 7. Multi‐level model calibration.
Supplementary Table 8. Calibration of the marginal predictions.
Supplementary Table 9. MoCA cut‐offs for discriminating impairment according to neuropsychological testing and cognitive research clinic diagnoses: Specificity, sensitivity, PPV, NPV and accuracy.
Supplementary Table 10. MoCA cut‐offs in combination with an MDS‐UPDRS 1.1 cut‐off for discriminating impairment according to neuropsychological testing and cognitive research clinic diagnoses: Specificity, sensitivity, PPV, NPV and accuracy.
Supplementary Table 11. Model coefficients for predicting research clinic diagnosis of any cognitive impairment (PD‐MCI or PDD).
Supplementary Table 12. Model coefficients for predicting research clinic diagnosis of PDD.
Supplementary Table 13. Sensitivity analysis: Discriminative ability evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Table 14. Sensitivity analysis: MoCA‐based cutoffs evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Table 15. Sensitivity analysis: Model‐based cutoffs evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Figure 1. Calibration curves for the models predicting impairment according to neuropsychological testing from MoCA subscores.
Supplementary Figure 2. Calibration curves for the models predicting research clinic diagnoses from MoCA totals and MDS‐UPDRS 1.1.
Data S2. ICMJE Disclosure Statement form.
Acknowledgments
This work was funded by Parkinson's UK: grant number J‐2101 “Understanding Parkinson's Progression.” JCK acknowledges support from the NIHR Oxford Biomedical Research Centre, NIHR Oxford Health Clinical Research Facility, and Parkinson's UK (PRO‐25‐01). This analysis uses data openly available from PPMI (Tier 1). Data used in the preparation of this article were obtained on December 7, 2024 from the Parkinson's Progression Markers Initiative.
PPMI database (https://www.ppmi-info.org/access-data-specimens/ download‐data), RRID: SCR 006431. For up‐to‐date information on the study, visit http://www.ppmi-info.org. PPMI‐a. Public‐private partnership‐is funded by the Michael J. Fox Foundation for Parkinson's Research and funding partners, including 4D Pharma, Abbvie, AcureX, Allergan, Amathus Therapeutics, Aligning Science Across Parkinson's, AskBio, Avid Radiopharmaceuticals, BIAL, Biogen, Biohaven, BioLegend, BlueRock Therapeutics, Bristol‐Myers Squibb, Calico Labs, Celgene, Cerevel Therapeutics, Coave Therapeutics, DaCapo Brainscience, Denali, Edmond J. Safra Foundation, Eli Lilly, Gain Therapeutics, GE HealthCare, Genentech, GSK, Golub Capital, Handl Therapeutics, Insitro, Janssen Neuroscience, Lundbeck, Merck, Meso Scale Discovery, Mission Therapeutics, Neurocrine Biosciences, Pfizer, Piramal, Prevail Therapeutics, Roche, Sanofi, Servier, Sun Pharma Advanced Research Company, Takeda, Teva, UCB, Vanqua Bio, Verily, Voyager Therapeutics, the Weston Family Foundation and Yumanity Therapeutics. The authors wish to thank all participants of the PPMI study and all staff involved in data collection.
Data Availability Statement
The data that support the findings of this study are openly available from the Parkinson's Progression Markers Initiative (PPMI) database (Tier 1): https://www.ppmi-info.org/access-data-specimens/download-data, RRID:SCR 006431. Data used in the preparation of this article were obtained on 2024‐07‐12.
References
- 1. Schrag A, Siddiqui UF, Anastasiou Z, Weintraub D, Schott JM. Clinical variables and biomarkers in prediction of cognitive impairment in patients with newly diagnosed Parkinson's disease: a cohort study. Lancet Neurol 2017;16:66–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Gibson LL, Weintraub D, Lemmen R, Perera G, Chaudhuri KR, Svenningsson P, Aarsland D. Risk of dementia in Parkinson's disease: a systematic review and meta‐analysis. Mov Disord 2024;39:1697–1709. [DOI] [PubMed] [Google Scholar]
- 3. Morales DA, Vives‐Gilabert Y, Gómez‐Ansón B, et al. Predicting dementia development in Parkinson's disease using Bayesian network classifiers. Psychiatry Res 2013;213:92–98. [DOI] [PubMed] [Google Scholar]
- 4. Phongpreecha T, Cholerton B, Mata IF, et al. Multivariate prediction of dementia in Parkinson's disease. npj Park Dis 2020;6:20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Coughlin DG, Irwin DJ. Fluid and biopsy based biomarkers in Parkinson's disease. Neurotherapeutics 2023;20:932–954. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Aarsland D, Batzu L, Halliday GM, Geurtsen GJ, Ballard C, Ray Chaudhuri K, Weintraub D. Parkinson disease‐associated cognitive impairment. Nat Rev Dis Primers 2021;7:47. [DOI] [PubMed] [Google Scholar]
- 7. Nasreddine ZS, Phillips NA, B'edirian V, et al. The Montreal cognitive assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc 2005;53:695–699. [DOI] [PubMed] [Google Scholar]
- 8. Hoops S, Nazem S, Siderowf A, Duda J, Xie S, Stern M, Weintraub D. Validity of the MoCA and MMSE in the detection of MCI and dementia in Parkinson disease. Neurology 2009;73:1738–1745. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Dalrymple‐Alford J, MacAskill M, Nakas C, et al. The MoCA: well‐suited screen for cognitive impairment in Parkinson disease. Neurology 2010;75:1717–1725. [DOI] [PubMed] [Google Scholar]
- 10. Marras C, Armstrong MJ, Meaney CA, et al. Measuring mild cognitive impairment in patients with Parkinson's disease. Mov Disord 2013;28:626–633. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Biundo R, Weis L, Facchini S, Formento‐Dojot P, Vallelunga A, Pilleri M, Antonini A. Cognitive profiling of Parkinson disease patients with mild cognitive impairment and dementia. Parkinsonism Relat Disord 2014;20:394–399. [DOI] [PubMed] [Google Scholar]
- 12. Kandiah N, Zhang A, Cenina AR, Au WL, Nadkarni N, Tan LC. Montreal cognitive assessment for the screening and prediction of cognitive decline in early Parkinson's disease. Parkinsonism Relat Disord 2014;20:1145–1148. [DOI] [PubMed] [Google Scholar]
- 13. Lucza T, Karádi K, Kállai J, et al. Screening mild and major neurocognitive disorders in Parkinson's disease. Behav Neurol 2015;2015:983606. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Federico A, Maier A, Vianello G, et al. Screening for mild cognitive impairment in Parkinson's disease: comparison of the Italian versions of three neuropsychological tests. Park Dis 2015;2015:681976. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Federico A, Trentin M, Zanette G, et al. Diagnosing mild cognitive impairment in Parkinson's disease: which tests perform best in the Italian population? Neurol Sci 2017;38:1461–1468. [DOI] [PubMed] [Google Scholar]
- 16. Kim HM, Nazor C, Zabetian CP, et al. Prediction of cognitive progression in Parkinson's disease using three cognitive screening measures. Clin Parkinsonism Relat Disord 2019;1:91–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Yu RL, Lee WJ, Li JY, et al. Evaluating mild cognitive dysfunction in patients with Parkinson's disease in clinical practice in Taiwan. Sci Rep 2020;10:1014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. D'Iorio A, Aiello EN, Amboni M, et al. Validity and diagnostics of the Italian version of the Montreal cognitive assessment (MoCA) in non‐demented Parkinson's disease patients. Aging Clin Exp Res 2023;35:2157–2163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Fiorenzato E, Cauzzo S, Weis L, et al. Optimal MMSE and MoCA cutoffs for cognitive diagnoses in Parkinson's disease: a data‐driven decision tree model. J Neurol Sci 2024;466:123283. [DOI] [PubMed] [Google Scholar]
- 20. O'Caoimh R, Foley MJ, Timmons S, Molloy DW. Screening for cognitive impairment in movement disorders: comparison of the Montreal cognitive assessment and quick mild cognitive impairment screen in Parkinson's disease and Lewy body dementia. J Alzheimers Dis Rep 2024;8:971–980. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Marek K, Jennings D, Lasch S, et al. The Parkinson progression marker initiative (PPMI). Prog Neurobiol 2011;95:629–635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Marek K, Chowdhury S, Siderowf A, et al. The Parkinson's progression markers initiative (PPMI)–establishing a PD biomarker cohort. Ann Clin Transl Neurol 2018;5:1460–1477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Goetz CG, Tilley BC, Shaftman SR, et al. Movement Disorder Society‐sponsored revision of the unified Parkinson's disease rating scale (MDS‐UPDRS): scale presentation and clinimetric testing results. Mov Disord 2008;23:2129–2170. [DOI] [PubMed] [Google Scholar]
- 24. Benedict RHB, Schretlen D, Brandt J. Hopkins verbal learning test – revised: normative data and analysis of inter‐form and test‐retest reliability. Clin Neuropsychol 1998;12:43–55. [Google Scholar]
- 25. Ardila A, Ostrosky‐Sol'ıs F, Bernal B. Cognitive testing toward the future: the example of semantic verbal fluency (ANIMALS). Int J Psychol 2006;41:324–332. [Google Scholar]
- 26. Smith A. Symbol digit modalities test. Western Psychological Services 1982. [Google Scholar]
- 27. Benedict RH, DeLuca J, Phillips G, LaRocca N, Hudson LD, Rudick R, Consortium MSOA . Validity of the symbol digit modalities test as a cognition performance outcome measure for multiple sclerosis. Mult Scler J 2017;23:721–733. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Wechsler D III. WMS‐III Administration and Scoring Manual. San Antonio, TX: The Psychological Corporation; 1997. [Google Scholar]
- 29. Venderploeg RD, LaLone LV, Greblo P, Schinka JA. Odd‐even short forms of the judgment of line orientation test. Appl Neuropsychol 1997;4:244–246. [DOI] [PubMed] [Google Scholar]
- 30. Ivnik RJ, Malec JF, Smith GE, Tangalos EG, Petersen RC. Neuropsychological tests' norms above age 55: COWAT, BNT, MAE token, WRAT‐R reading, AMNART, STROOP, TMT, and JLO. Clin Neuropsychol 1996;10:262–278. [Google Scholar]
- 31. Calderón‐Rubio E, Oltra‐Cucarella J, Bonete‐López B, In~esta C, Sitges‐Maci'a E. Regression‐based normative data for independent and cognitively active spanish older adults: free and cued selective reminding test, rey–osterrieth complex figure test and judgement of line orientation. Int J Environ Res Public Health 2021;18:12977. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Weintraub D, Brumm MC, Kurth R, York MK, Initiative PPM . Use of robust norming to create a sensitive cognitive summary score in de novo Parkinson's disease: an illustrative example. Mov Disord 2025;40:468–477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Nazem S, Siderowf AD, Duda JE, et al. Montreal cognitive assessment performance in patients with Parkinson's disease with “normal” global cognition according to mini‐mental state examination score. J Am Geriatr Soc 2009;57:304–308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Hassanzad M, Hajian‐Tilaki K. Methods of determining optimal cut‐point of diagnostic biomarkers with application of clinical data in ROC analysis: an update review. BMC Med Res Methodol 2024;24:84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Kolen MJ, Brennan RL. Test Equating: Methods and Practices. New York: Springer; 2013. [Google Scholar]
- 36. Vickers AJ, Van Calster B, Steyerberg EW. Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests. BMJ 2016;352:i6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Hosmer DW Jr, Lemeshow S, Sturdivant RX. Applied Logistic Regression. Hoboken, New Jersey: John Wiley & Sons; 2013. [Google Scholar]
- 38. nand DP, Liu X, Brown PJ. Impact of functional deficits in instrumental activities of daily living in mild cognitive impairment: a clinical algorithm to predict progression to dementia. Alzheimer Dis Assoc Disord 2017;31:55–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. González DA, Gonzales MM, Jennette KJ, Soble JR, Fongang B. Cognitive screening with functional assessment improves diagnostic accuracy and attenuates bias. Alzheimers Dement Diagn Assess Dis Monit 2021;13:e12250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Becker S, Bode M, Brockmann K, et al. Cognitive‐driven activities of daily living impairment as a predictor for dementia in Parkinson disease: a longitudinal cohort study. Neurology 2022;99:e2548–e2560. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Dubois B, Burn D, Goetz C, et al. Diagnostic procedures for Parkinson's disease dementia: recommendations from the movement disorder society task force. Mov Disord 2007;22:2314–2324. [DOI] [PubMed] [Google Scholar]
- 42. Litvan I, Goldman JG, Tröster AI, et al. Diagnostic criteria for mild cognitive impairment in Parkinson's disease: Movement Disorder Society task force guidelines. Mov Disord 2012;27:349–356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Rossetti HC, Lacritz LH, Cullum CM, Weiner MF. Normative data for the Montreal cognitive assessment (MoCA) in a population‐based sample. Neurology 2011;77:1272–1275. [DOI] [PubMed] [Google Scholar]
- 44. Sachs BC, Steenland K, Zhao L, et al. Expanded demographic norms for version 3 of the Alzheimer disease centers' neuropsychological test battery in the uniform data set. Alzheimer Dis Assoc Disord 2020;34:191–197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Amitrano NR, Lord AR, González DA. Montreal cognitive Assessment's auditory items (MoCA‐22): normative data and reliable change indices. Appl Neuropsychol Adult 2026;33(3):708–715. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. O'Driscoll C, Shaikh M. Cross‐cultural applicability of the Montreal cognitive assessment (MoCA): a systematic review. J Alzheimer's Dis 2017;58:789–801. [DOI] [PubMed] [Google Scholar]
- 47. Gilad T, Algon AL, Vakil E, et al. Breaking geographic and language barriers in neuropsychology: online administration of MoCA in diverse populations. NPJ Digit Med 2025;8:379. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Emre M, Aarsland D, Brown R, et al. Clinical diagnostic criteria for dementia associated with Parkinson's disease. Mov Disord 2007;22:1689–1707. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. Supporting Information
Supplementary Table 1. Basic demographics and cognitive test scores in the PPMI cohort.
Supplementary Table 2. Model coefficients for age, sex and education correction of HVLT‐immediate recall, HVLT‐discrimination, animals fluency, SDMT, LNS, BJL scores.
Supplementary Table 3. Mean and standard deviation for z‐scoring age, sex and education corrected test scores.
Supplementary Table 4. Cut‐offs defining impairment in each cognitive domain tested.
Supplementary Table 5. Model coefficients for predicting impairment according to neuropsychological testing (≥1 impaired domain).
Supplementary Table 6. Model coefficients for predicting impairment according to neuro‐ psychological testing (≥2 impaired domains).
Supplementary Table 7. Multi‐level model calibration.
Supplementary Table 8. Calibration of the marginal predictions.
Supplementary Table 9. MoCA cut‐offs for discriminating impairment according to neuropsychological testing and cognitive research clinic diagnoses: Specificity, sensitivity, PPV, NPV and accuracy.
Supplementary Table 10. MoCA cut‐offs in combination with an MDS‐UPDRS 1.1 cut‐off for discriminating impairment according to neuropsychological testing and cognitive research clinic diagnoses: Specificity, sensitivity, PPV, NPV and accuracy.
Supplementary Table 11. Model coefficients for predicting research clinic diagnosis of any cognitive impairment (PD‐MCI or PDD).
Supplementary Table 12. Model coefficients for predicting research clinic diagnosis of PDD.
Supplementary Table 13. Sensitivity analysis: Discriminative ability evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Table 14. Sensitivity analysis: MoCA‐based cutoffs evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Table 15. Sensitivity analysis: Model‐based cutoffs evaluated using a cross‐sectional dataset containing only the most recent observation per participant.
Supplementary Figure 1. Calibration curves for the models predicting impairment according to neuropsychological testing from MoCA subscores.
Supplementary Figure 2. Calibration curves for the models predicting research clinic diagnoses from MoCA totals and MDS‐UPDRS 1.1.
Data S2. ICMJE Disclosure Statement form.
Data Availability Statement
The data that support the findings of this study are openly available from the Parkinson's Progression Markers Initiative (PPMI) database (Tier 1): https://www.ppmi-info.org/access-data-specimens/download-data, RRID:SCR 006431. Data used in the preparation of this article were obtained on 2024‐07‐12.
