Frontal atrophy as a marker for dementia conversion in Parkinson's disease with mild cognitive impairment

Abstract

This study aimed to investigate the cortical neural correlates of dementia conversion in Parkinson's disease with mild cognitive impairment (PD‐MCI). We classified 112 patients with drug‐naïve early stage PD meeting criteria for PD‐MCI into either PD with dementia (PDD) converters (n = 34) or nonconverters (n = 78), depending on whether they developed dementia within 4 years of PD diagnosis. Cortical thickness analyses were performed in 34 PDD converters and 34 matched nonconverters. Additionally, a linear discriminant analysis was performed to distinguish PDD converters from nonconverters using cortical thickness of the regions that differed between the two groups. The PDD converters had higher frequencies of multiple domain MCI and amnestic MCI with storage failure, and poorer cognitive performances on frontal/executive, memory, and language function domains than did the nonconverters. Cortical thinning extending from the posterior cortical area into the frontal region was observed in PDD converters relative to nonconverters. The discriminant analysis showed that the prediction model with two cortical thickness variables in the right medial superior frontal and left olfactory cortices optimally distinguished PDD converters from nonconverters. Our data suggest that cortical thinning in the frontal areas including the olfactory cortex is a marker for early dementia conversion in PD‐MCI.

1. INTRODUCTION

Cognitive impairment is commonly observed in patients with Parkinson's disease (PD), and approximately 30% of patients are diagnosed with mild cognitive impairment (MCI) at early stages of PD (Kandiah et al., 2009). The early presence of MCI in PD signals a highly increased risk of early incident dementia (Pedersen, Larsen, Tysnes, & Alves, 2013) as well as faster atrophy in various cortical and subcortical structures (Hanganu et al., 2014). However, a diagnosis of MCI does not always predict the progression to PD with dementia (PDD); a considerable number of patients with PD with MCI (PD‐MCI) remain clinically stable or revert to a cognitively normal status (Pedersen et al., 2013), and among PDD converters, the rate of progression to dementia or the severity of cognitive deficits is highly variable (Aarsland et al., 2007). Therefore, establishing validated biomarkers is important for identifying individuals at high risk of PDD conversion and for implementing appropriate therapeutic strategies for these patients (Eberling et al., 2014).

To date, several neuroimaging studies have suggested that the cortical atrophy pattern can serve as a marker for ongoing cognitive decline in PD. The few existing longitudinal neuroimaging studies have reported inconsistent findings; some stressed the importance of frontal atrophy in PDD conversion (Compta et al., 2013; Gasca‐Salas et al., 2019; Lee et al., 2014), while others demonstrated that the cortical atrophy in PDD converters was predominantly concentrated in the posterior cortical regions (Mak et al., 2015; Tropea et al., 2018; Weintraub et al., 2012). However, these studies were limited by their small sample sizes (Compta et al., 2013; Gasca‐Salas et al., 2019; Lee et al., 2014), short follow‐up periods (Lee et al., 2014; Mak et al., 2015; Weintraub et al., 2012), inclusion of patients in the advanced stages of PD (Gasca‐Salas et al., 2019; Tropea et al., 2018), and lack of baseline dichotomization into MCI and a cognitively normal status (Compta et al., 2013; Tropea et al., 2018). Therefore, in the present study, we investigated the cortical neural correlates of dementia conversion in a relatively large sample of patients with early stage PD‐MCI. Given that a considerable number of patients with PD eventually develop dementia (Hely, Reid, Adena, Halliday, & Morris, 2008; Lee et al., 2017), we employed a 4‐year time window based on previous studies (Pedersen, Larsen, Tysnes, & Alves, 2017; Wood et al., 2016) to determine whether the patients with drug‐naïve early stage PD meeting criteria for PD‐MCI were at imminent risk of PDD conversion during the follow‐up period. Then, we examined the cognitive profiles and cortical thinning patterns of patients who converted from PD‐MCI to PDD to elucidate whether such patterns have the potential to serve as neuroanatomical predictors for the development of dementia in PD. A linear discriminant analysis (LDA) was additionally performed to determine the prediction model for PDD conversion using the regional cortical thickness.

2. MATERIALS AND METHODS

2.1. Subjects

We retrospectively reviewed the Yonsei Parkinson Center database to identify patients with PD who visited the Movement Disorders outpatient clinic at Severance Hospital, Yonsei University Health System, from March 2008 to September 2016. During this time frame, a total of 563 patients with PD were diagnosed with MCI using a detailed neuropsychological test (Litvan et al., 2012). The PD and PD‐MCI were diagnosed according to the clinical diagnostic criteria of the UK PD Society Brain Bank (Hughes, Daniel, Kilford, & Lees, 1992) and the Movement Disorder Society Task Force guidelines (Litvan et al., 2012), respectively. To ensure clinical diagnostic accuracy, ¹⁸F‐FP‐CIT PET scans were performed on the subjects and all showed decreased ¹⁸F‐FP‐CIT binding in the posterior putamen. Of the 563 patients with PD‐MCI, 148 met the following criteria: (a) de novo subjects (i.e., drug‐naïve state in early stages of PD to reduce the confounding effects of advanced PD and chronic dopamine replacement on cognition); (b) subjects with a brain magnetic resonance imaging (MRI) scans available for the cortical thickness analyses (i.e., high‐resolution axial T1‐weighted images) at the baseline assessment; and (c) subjects who subsequently underwent serial neuropsychological assessments during the follow‐up period (two to five times, with a mean interval of 2 years). Parkinsonian motor symptoms were assessed using the Unified PD Rating Scale Part III (UPDRS‐III). Olfactory function was evaluated using the cross‐cultural smell identification test (CCSIT) and patients were classified into hyposmic PD if the CCSIT score was 6 or less (Doty, Marcus, & Lee, 1996; Lee et al., 2015). Depression was assessed using the Beck Depression Inventory (BDI). We complied with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and this study was approved by the Yonsei University Severance Hospital Institutional Review Board. The need for informed consent was waived because of the retrospective nature of the study.

2.2. Neuropsychological assessment

All subjects were administered the Seoul Neuropsychological Screening Battery, a comprehensive Korean language neuropsychological test battery (Ahn et al., 2010; Kang, J, & Na, 2012). The composite score of each cognitive domain was calculated by averaging the z‐scores of the subtests that constitute the domain. To diagnose PD‐MCI, two tests were designated to represent each of the four cognitive domains except language, as described in our previous work (Chung et al., 2018). Attention and working memory were tested using the digit span task (backward) and color‐word Stroop test. Executive function was evaluated using the controlled oral word association test and 10‐point clock‐drawing test. Memory was tested using the Seoul verbal learning test and Rey complex figure test. Visuospatial function was tested using the interlocking pentagon drawing test and Rey complex figure test copy. Language was examined using the Korean version of the Boston Naming Test. The scores on each cognitive domain were classified as abnormal when they were below the 16th percentile (1 SD) of the age‐ and education‐specific norms of 447 normal subjects. A diagnosis of PD‐MCI was made if the patient demonstrated abnormal performances on at least two tests within the five cognitive domains. PDD was diagnosed according to the clinical diagnostic criteria for probable PDD (Emre et al., 2007) with evidence of abnormalities in the activities of daily living (ADL), judged both clinically and on an instrumental ADL scale (Kang, 2002; Ku et al., 2004) while functional disabilities merely due to parkinsonian motor symptoms were not considered to be impairment of complex ADL. Abnormalities in ADL were determined based on two standardized measures of functional impairments (Korean Instrumental ADL, cutoff point of 0.43; Seoul Instrumental ADL, cutoff point of 8) (Kang, 2002; Ku et al., 2004).

2.3. Classification of PD‐MCI into the PDD converter and nonconverter groups

Given that the cumulative prevalence of PDD increases to 80% after a PD duration of 20 years (Halliday, Hely, Reid, & Morris, 2008), a defined time window within the follow‐up period is needed to determine whether a subject has a high risk of PDD conversion. Therefore, we set our time window at 4 years based on previous studies (i.e., progression to PDD before the end of the 4‐year period was treated as an a priori end point) (Pedersen et al., 2017; Wood et al., 2016), and patients who developed dementia after more than 4 years postdiagnosis were not regarded as having a high risk of converting to PDD and thus were assigned to the nonconverter group. Among the 148 patients with drug‐naïve early stage PD meeting criteria for PD‐MCI, 36 patients who were followed up for <4 years and did not progress to PDD were additionally excluded. Of the remaining 112 patients, 53 patients who did not develop dementia over the 4‐year period (follow‐up duration, 6.32 ± 1.60 years) were assigned to the nonconverter group. The other 59 patients were diagnosed with dementia during the follow‐up period; among them, 34 patients who developed dementia within 4 years after receiving the PD diagnosis (interval [mean ± SD], 2.20 ± 0.97 years) were assigned to the PDD converter group and the remaining 25 patients who progressed to PDD after more than 4 years postdiagnosis (interval, 5.44 ± 1.14 years) were classified as the nonconverter group (Figure 1).

Figure 1.

Flowchart of participants and enrollment

2.4. Neuroimaging analyses

To reduce the effects of potential confounding factors and provide the covariate balance, propensity scores were used to match the PDD converter group with a subset of the nonconverter group. The propensity score for the predicted probability of PDD conversion in each patient was estimated using a logistic regression model, which included the age at onset, sex, years of education, and PD duration as covariates. The propensity score represented a balancing score, and thus the distribution of the covariates would be the same between the groups if a set of subjects had the same propensity score (Austin, 2011). Based on the propensity scores, 34 patients with PD‐MCI in the converter group were matched to 34 of the 78 patients in the nonconverter group (i.e., 53 who never converted to dementia and 25 who developed dementia after more than 4 years postdiagnosis) using a nearest neighbor matching algorithm: At each matching step, nonconverter subject who was closest to the converter subject on the propensity score was selected. If multiple nonconverters had propensity scores that were equally close to that of the converter subject, one of these nonconverters was chosen at random (Austin, 2011; Ho, Imai, King, & Stuart, 2011). Propensity score matching was performed with the R software package, version 3.4.0 (http://www.r-project.org). We also included 20 healthy subjects without a history of neurological disease (age, 68.35 ± 5.13; female, 50.0%; years of education, 10.40 ± 4.94 years) as a control group.

2.4.1. MRI acquisition

MRI scans were acquired using a Philips 3.0 T scanner (Philips Achieva; Philips Medical Systems, Best, The Netherlands) with a SENSE head coil (SENSE factor = 2) as described in our previous work (Baik et al., 2014; Chung et al., 2013; Sunwoo et al., 2015). The high‐resolution axial T1‐weighted MRI data were obtained using a 3D T1‐TFE sequence with the following parameters: 224 × 224 axial acquisition matrix; 256 × 256 reconstructed matrix with 170 slices; voxel size, 0.859 × 0.859 × 1 mm³; field of view, 220 mm; echo time, 4.6 ms; repetition time, and 9.8 ms; flip angle, 8°.

2.4.2. Analysis of cortical thickness

We used the same methodology as that described in previous work to analyze the cortical thickness (Chung et al., 2017). We visually validated the quality of the T1‐weighted MR images before extracting the cortical surface model and cortical thickness. Structural MR images were registered to a standardized stereotaxic space using linear transformation (Collins, Neelin, Peters, & Evans, 1994). The N3 algorithm was used to correct images for intensity nonuniformity resulting from inhomogeneity in the magnetic field (Sled, Zijdenbos, & Evans, 1998). The nonbrain tissues of registered and corrected images were removed using the Brain Extraction Tool (Smith, 2002) and then classified into gray matter, white matter, cerebrospinal fluid, and background using the Intensity‐Normalized Stereotaxic Environment for Classification of Tissues algorithm (Zijdenbos et al., 1996). The surfaces of the inner and outer cortices, which consisted of 40,962 vertices, were automatically extracted using the Constrained Laplacian‐based Automated Segmentation with Proximities algorithm, which reconstructs the inner cortical surface by deforming a spherical mesh onto the white matter/gray matter boundary and then expanding the deformable model to the gray matter/cerebrospinal fluid boundary (Kim et al., 2005). Cortical thickness was defined using the t‐link method, which captures the Euclidean distance between the linked vertices of the inner and outer cortical surfaces (MacDonald, Kabani, Avis, & Evans, 2000). Each individual cortical thickness map was smoothed with a 20 mm full‐width at half‐maximum Gaussian smoothing kernel to increase the signal‐to‐noise ratio and aligned to an unbiased iterative surface template using vertex‐wise sphere‐to‐sphere nonlinear surface registration (Lyttelton, Boucher, Robbins, & Evans, 2007; Robbins, Evans, Collins, & Whitesides, 2004). Group comparisons of cortical thickness among the three groups were performed at the vertex‐wise level using an analysis of covariance (ANCOVA) with age, sex, and years of education as covariates. We conducted false discovery rate (FDR) correction with a threshold at p < .05 for multiple comparisons. Post hoc two sample t tests were performed to identify differences of cortical thickness between each pair of groups within a mask including the significant regions in the ANCOVA. We conducted the Bonferroni correction for multiple contrast tests with a threshold at p < .05. Additionally, we performed correlation analyses between cortical thickness and cognitive composite scores (see Supplementary Methods).

2.5. Prediction analysis of PDD conversion in patients with de novo PD‐MCI within a 4‐year window

The mean cortical thickness of regions of interest (ROIs) using a surface‐based automated anatomical labeling template (Tzourio‐Mazoyer et al., 2002) excluding the cerebellum and subcortical regions to segment the cortical regions into 78 areas was obtained for each individual subject. An LDA was performed to determine the prediction model for PDD conversion using the cortical thickness of the regions that differed significantly between the two groups. The LDA is a simple and intuitive classification algorithm that can be implemented under the assumption that the input variables follow the Gaussian distribution. To do this, we first compared the mean cortical thickness of each ROI between the PDD converter and matched nonconverter groups using an analysis of variance with age, sex, and years of education as covariates. For exploratory purposes, 10 regions were identified as possible discriminants at a relatively liberal threshold (uncorrected p value <.01). Following this, the LDA prediction was defined using discriminant coefficients, which were calculated by multiplying the inverse pooled covariance matrix by the mean vector of the two PD group differences. The best LDA prediction model was determined using a stepwise selection of these ROIs, that is, Wilk's lambda and the area under the receiver operating characteristic curve (AUC) were compared between the models using leave‐one‐out cross validation.

2.6. Statistical analyses

To compare the baseline demographic characteristics and the level of cognitive performance between the groups, Student's t tests and Pearson's χ² tests were conducted for continuous and categorical variables, respectively. The FDR‐controlling method was used for multiple comparisons correction. We used the LDA classification models to predict which patients were at a high risk on PDD conversion based on the cortical thickness in the frontal sub‐regions. The statistical analyses were performed with SPSS (version 23.0; IBM Corp., Armonk, NY), SAS (version 9.3; SAS Inc., Cary, NC), and R (version 3.4.0; http://www.r-project.org); statistical significance was set at p < .05 (two tailed).

2.7. Data availability

For purposes of replicating procedures and results, any qualified investigator can request anonymized data after ethics clearance and approval by all authors.

3. RESULTS

3.1. Baseline clinical characteristics of patients with PD‐MCI

Table 1 shows the baseline demographic characteristics of patients with drug‐naïve early stage PD meeting criteria for PD‐MCI. No significant differences were observed in age, sex, years of education, duration of PD, initial UPDRS‐III scores, BDI scores, and the presence of vascular risk factors were observed between the PDD converter (n = 34) and matched nonconverter groups (n = 34), while the 34 patients in the PDD converter group tended to be older than the 78 patients in the nonconverter group before the propensity score matching (p = .080). The PDD converter group also tended to have lower CCSIT scores and a higher prevalence of hyposmia than did the nonconverter group.

Table 1.

Baseline demographic characteristics of patients with de novo PD‐MCI

	Overall series			Propensity score‐matched pairs
	Converter (n = 34)	Nonconverter (n = 78)	p‐Value	Converter (n = 34)	Nonconverter (n = 34)	p‐Value
Age	71.12 ± 7.11	68.53 ± 7.16	.080	71.12 ± 7.11	69.62 ± 7.85	.412
Female, no. (%)	16 (47.1%)	32 (41.0%)	.553	16 (47.1%)	15 (44.1%)	.808
Education (years)	9.53 ± 5.10	8.66 ± 4.38	.360	9.53 ± 5.10	9.43 ± 4.47	.930
PD duration (months)	20.65 ± 19.43	16.26 ± 13.32	.236	20.65 ± 19.43	18.94 ± 13.98	.679
UPDRS‐III	24.32 ± 11.33	21.36 ± 10.43	.181	24.32 ± 11.33	21.18 ± 10.34	.236
CCSIT	5.30 ± 1.93	6.07 ± 2.05	.082	5.30 ± 1.93	6.30 ± 2.44	.078
Hyposmia, no. (%) a	26 (76.5%)	48 (61.5%)	.125	26 (76.5%)	16 (47.1%)	.013
BDI	13.61 ± 8.25	15.88 ± 10.43	.268	13.61 ± 8.25	16.32 ± 9.99	.230
Vascular risk factors, no. (%)
Hypertension	19 (55.9%)	34 (43.6%)	.231	19 (55.9%)	17 (50.0%)	.627
Diabetes mellitus	4 (11.8%)	22 (28.2%)	.058	4 (11.8%)	9 (26.5%)	.123
Dyslipidemia	3 (8.8%)	13 (16.7%)	.383	3 (8.8%)	6 (17.6%)	.476
Cardiac disease	3 (8.8%)	11 (14.1%)	.546	3 (8.8%)	5 (14.7%)	.709

Note. The values are expressed as mean ± SD or number (percentage).

Abbreviations: BDI, Beck Depression Inventory; CCSIT, the cross‐cultural smell identification test; PD, Parkinson's disease; PDD, PD with dementia; PD‐MCI, PD with mild cognitive impairment; UPDRS‐III, Unified PD Rating Scale Part III.

^aWe classified the patients into hyposmic PD if the CCSIT score was 6 or less.

Table 2 shows the detailed neuropsychological data of patients with drug‐naïve early stage PD meeting criteria for PD‐MCI. The frequency of nonamnestic and amnestic MCI types did not significantly differ between the PDD converter and nonconverter groups (amnestic MCI was present in 82.4 and 70.5% of patients in the PDD converter group and nonconverter group, respectively; p = .188). However, among patients with the amnestic MCI type, the storage failure subtype (i.e., abnormal performances on both delayed recall and recognition tests) was more prevalent in the PDD converter group (20 out of 28 patients) than in the nonconverter group (24 out of 55 patients; p = .016). In addition, the PDD converter group tended to have a higher frequency of the multiple domain MCI type (91.2%) than did the nonconverter group (75.6%, p = .057). The patients in the PDD converter group had lower Korean version of the Mini‐Mental State Examination scores (p = .044) and lower composite scores for frontal/executive function (p = .019), language (p = .014), and verbal and visual memory (p = .006 and .021) than did those in the nonconverter group, while patients' cognitive performance in the domains of attention/working memory (p = .119) and visuospatial function (p = .110) was comparable between the groups. Similar results were obtained for the neuropsychological data in the propensity score‐matched pairs.

Table 2.

Neuropsychological data of the patients with de novo PD‐MCI

	Overall series			Propensity score‐matched pairs
	Converter (n = 34)	Nonconverter (n = 78)	p‐Value	Converter (n = 34)	Nonconverter (n = 34)	p‐Value
MCI subtypes (1)
Nonamnestic	6 (17.6%)	23 (29.5%)	.188b	6 (17.6%)	10 (29.4%)	.253b
Amnestica	28 (82.4%)	55 (70.5%)		28 (82.4%)	24 (70.6%)
Retrieval failure	8 (23.5%)	31 (39.7%)	.016c	8 (23.5%)	14 (41.2%)	.030c
Storage failure	20 (58.8%)	24 (30.8%)		20 (58.8%)	10 (29.4%)
MCI subtypes (2)			.057			.031
Single domain	3 (8.8%)	16 (24.4%)		3 (8.8%)	10 (29.4%)
Multiple domain	31 (91.2%)	59 (75.6%)		31 (91.2%)	24 (70.6%)
Neuropsychological data
K‐MMSE	25.44 ± 2.46	26.54 ± 2.69	.044	25.44 ± 2.46	26.88 ± 2.38	.017
Attention and working memory	−0.76 ± 1.06	−0.31 ± 0.81	.119d	−0.76 ± 1.06	−0.19 ± 0.97	.157d
Frontal executive function	−1.04 ± 0.81	−0.50 ± 0.88	.019d	−1.04 ± 0.81	−0.38 ± 0.98	.017d
Language and related function	−1.14 ± 1.56	−0.27 ± 1.22	.014d	−1.14 ± 1.56	−0.10 ± 1.20	.016d
Memory function	−1.09 ± 0.78	−0.47 ± 0.67	<.001d	−1.09 ± 0.78	−0.36 ± 0.70	.001d
Verbal memory	−1.22 ± 0.91	−0.56 ± 0.95	.006d	−1.22 ± 0.91	−0.47 ± 0.92	.006d
Visual memory	−0.95 ± 0.89	−0.39 ± 0.90	.021d	−0.95 ± 0.89	−0.25 ± 1.00	.017d
Visuospatial function	−1.00 ± 2.37	0.09 ± 1.37	.110d	−1.00 ± 2.37	0.04 ± 1.00	.114d

Note. The values are expressed as mean ± SD or number (percentage).

Abbreviations: K‐MMSE, the Korean version of the Mini‐Mental State Examination; PD‐MCI, Parkinson's disease with mild cognitive impairment.

^aRetrieval failure represents a free‐recall memory deficit that improved with cues on recognition tests, while storage failure represents abnormal performances on both delayed recall and recognition tests.

^b p‐Value for the frequency of nonamnestic and amnestic types between the groups.

^c p‐Value for the frequency of retrieval and storage failure subtypes between the groups.

^dBonferroni correction method for multiple comparisons.

3.2. Analysis of cortical thickness

Compared to the healthy control group, the PDD converter group exhibited reduced cortical thickness in the bilateral medial frontal, left orbitofrontal, right inferomedial temporal, bilateral occipital, frontoparietal operculum, superior temporal, precuneus, and cingulum regions (Bonferroni‐corrected p < .05; Figure 2a), while regions of different cortical thickness between the nonconverter and control groups were confined to the bilateral occipital and precuneus regions (Bonferroni‐corrected p < .05; Figure 2b). In a direct comparison between the PD groups, the PDD converter group exhibited reduced cortical thickness in the bilateral medial frontal, left orbitofrontal, right inferomedial temporal, bilateral occipital, frontoparietal operculum, superior temporal, and cingulum regions compared with the matched nonconverter group (Bonferroni‐corrected p < .05; Figure 2c). However, nonconverters did not demonstrate any areas with reduced cortical thickness relative to PDD converters. Partial correlation analyses showed that the composites scores of the attention/working memory and frontal/executive function domain tests were positively correlated with the cortical thickness in the bilateral frontotemporal and right posterior parietal regions in patients with PD‐MCI, whereas the composite scores of the memory, visuospatial, and language function domains did not correspond well to the specific patterns of cortical thinning (see Supplementary Results and Figures S1 and S2).

Figure 2.

Analysis of cortical thickness. (a) Parkinson's disease with dementia (PDD) converter versus healthy control. (b) PDD nonconverter versus healthy control. (c) PDD converter versus PDD nonconverter. The results for the differences between the groups in cortical thickness were considered significant at corrected p < .05 [Color figure can be viewed at http://wileyonlinelibrary.com]

3.3. The LDA to identify a high‐risk group for PDD conversion

An ROI‐based analysis of cortical thickness using a surface‐based automated anatomical labeling template (Tzourio‐Mazoyer et al., 2002) demonstrated that the PDD converter group tended to exhibit cortical thinning mainly in the frontal subregions relative to the matched nonconverter group, including the medial part of right superior frontal gyrus, orbital part of the left superior frontal gyrus, right Rolandic operculum, left olfactory cortex, right median cingulate and paracingulate gyri, left gyrus rectus, medial orbital part of the left superior frontal gyrus, left anterior cingulate and paracingulate gyri, medial part of the left superior frontal gyrus, and opercular part of the right inferior frontal gyrus (uncorrected p < .01; Table 3). We tested the LDA prediction models using a stepwise selection of these ROIs and found that the prediction model with two cortical thickness variables in the medial part of the right superior frontal gyrus and left olfactory cortex optimally distinguished PDD converters from nonconverters (Wilk's lambda, 0.7643; AUC 0.7803, 95% confidence interval [0.6700–0.8905]; misclassification rate, 25.0%; cross‐validated misclassification rate, 29.4%). The cutoff point was set as D = 25.23768, where the discriminant (D) = 5.796743 × (cortical thickness in the medial part of the right superior frontal gyrus) + 2.095820 × (cortical thickness in the left olfactory cortex); if the calculated D value was less than 25.23768, then the subject was classified into the PDD converter group.

Table 3.

ROI‐based analysis of cortical thickness in patients with PD‐MCI

	Converter (n = 34)	Nonconverter (n = 34)	Uncorrected p
SFGmed.R	3.120 (0.024)	3.243 (0.024)	.001
ORBsup.L	3.083 (0.024)	3.192 (0.024)	.002
ROL.R	3.063 (0.029)	3.182 (0.029)	.005
OLF.L	3.151 (0.044)	3.334 (0.044)	.005
DCG.R	3.200 (0.023)	3.293 (0.023)	.005
REC.L	3.123 (0.033)	3.259 (0.033)	.005
ORBsupmed.L	3.096 (0.028)	3.208 (0.028)	.006
ACG.L	3.155 (0.022)	3.244 (0.022)	.007
SFGmed.L	3.167 (0.026)	3.269 (0.026)	.008
IFGoperc.R	3.091 (0.026)	3.189 (0.026)	.010
SMA.L	3.273 (0.033)	3.395 (0.033)	.013
SMA.R	3.201 (0.036)	3.332 (0.036)	.013
ORBsupmed.R	3.135 (0.028)	3.229 (0.028)	.020
ORBinf.L	3.225 (0.022)	3.298 (0.022)	.024
STG.L	3.134 (0.031)	3.235 (0.031)	.024
ORBsup.R	3.042 (0.020)	3.109 (0.020)	.024
ROL.L	3.076 (0.027)	3.162 (0.027)	.026
PHG.R	3.486 (0.049)	3.639 (0.049)	.031
PCUN.R	2.963 (0.027)	3.048 (0.027)	.033
SFGdor.R	3.005 (0.031)	3.100 (0.031)	.033
ANG.L	3.031 (0.037)	3.137 (0.037)	.045
MFG.R	2.973 (0.029)	3.056 (0.029)	.049

Note. The values are expressed as estimated mean (SE).

Abbreviations: ACG.L, left anterior cingulate and paracingulate gyri; ANG.L, left angular gyrus; DCG.R, right median cingulate and paracingulate gyri; IFGoperc.R, opercular part of right inferior frontal gyrus; MFG.R, right middle frontal gyrus; OLF.L, left olfactory cortex; ORBinf.L, orbital part of left inferior frontal gyrus; ORBsup.L, orbital part of left superior frontal gyrus; ORBsup.R, orbital part of right superior frontal gyrus; ORBsupmed.L, medial orbital part of left superior frontal gyrus; ORBsupmed.R, medial orbital part of right superior frontal gyrus; PCUN.R, right precuneus; PD‐MCI, Parkinson's disease with mild cognitive impairment; PHG.R, right parahippocampal gyrus; REC.L, left gyrus rectus; ROI, region of interest; ROL.L, left Rolandic operculum; ROL.R, right Rolandic operculum; SFGdor.R, dorsolateral part of right superior frontal gyrus; SFGmed.L, medial part of left superior frontal gyrus; SFGmed.R, medial part of right superior frontal gyrus; SMA.L, left supplementary motor area; SMA.R, right supplementary motor area; STG.L, left superior temporal gyrus.

4. DISCUSSION

The present study investigated the cortical neural correlates of early dementia conversion in patients with drug‐naïve early stage PD meeting criteria for PD‐MCI. The major findings were as follows: (a) PDD converters had higher frequencies of multiple domain MCI, amnestic MCI with storage failure and hyposmia, and lower cognitive composite scores on frontal/executive, memory, and language function domain tests than did nonconverters; (b) PDD converters exhibited cortical thinning in the frontal regions that extended into the temporal areas relative to the matched nonconverters; (c) the prediction model with two cortical thickness variables in the medial part of the right superior frontal gyrus and left olfactory cortex had fair discriminatory power to distinguish PDD converters from nonconverters. These data suggest that cortical thinning in the frontal areas as well as distinctive cognitive profile patterns and level of cognitive performance have the potential to serve as useful predictive markers for early dementia conversion in patients with newly diagnosed PD‐MCI.

Dementia greatly affects the morbidity and mortality in PD, and early identification of patients who are at the highest risk of PDD is important for the proper and rapid implementation of therapeutic and supportive strategies (Williams‐Gray, Foltynie, Brayne, Robbins, & Barker, 2007). Several studies have consistently reported the risk factors for PDD, which include increased age, male sex, long duration of PD, low education level, postural instability‐gait difficulty phenotype, and hallucinations (Hobson & Meara, 2004; Weintraub et al., 2012). Since cognitive deterioration in PD does not reliably follow the expected progression (i.e., PD‐MCI does not always progress to dementia) (Pedersen et al., 2017), a longitudinal study design identifying patients with PD‐MCI who progressed to dementia within a defined time window (i.e., at imminent risk of dementia) would provide more direct evidence for the neural basis of PDD conversion. To date, few longitudinal studies have reported inconsistent findings, and thus it remains unclear whether frontal atrophy (Compta et al., 2013; Gasca‐Salas et al., 2019; Lee et al., 2014) or posterior cortical atrophy (Mak et al., 2015; Tropea et al., 2018; Weintraub et al., 2012) is more important for PDD conversion. In the present study, compared to the healthy controls, cortical thinning in the nonconverter group was predominantly observed in the posterior cortical regions, while the regions of cortical atrophy in the converter group were more extensive and additionally involved the frontal regions. A direct comparison between the PD groups showed that the PDD converter group exhibited cortical thinning mainly in the frontal areas compared to the matched nonconverter group. Based on the pattern of cortical thickness in this study, atrophy in the posterior cortical areas may merely reflect the MCI status in PD. Along with this posterior pattern of atrophy, additional frontal atrophy could serve as an important marker of the imminent risk of PDD conversion in patients with newly diagnosed PD‐MCI, supporting the findings of our previous voxel‐based morphometry study (Lee et al., 2014) which demonstrated that PDD converters had significantly lower gray matter density in the left frontal areas, left insular cortex, and bilateral caudate nucleus compared with that in the nonconverters. Atrophic changes in the frontal region may act as the underlying neural correlates of PDD by affecting the level of performance on various cognitive domains via disrupting the reciprocal cortico‐cortical connections (Simons & Spiers, 2003) or important nodes of information integration (Achard, Salvador, Whitcher, Suckling, & Bullmore, 2006). Besides, frontal atrophy may indirectly reflect neurotransmitter dysregulation (Klein et al., 2010), cortical deafferentation secondary to white matter degeneration (Rae et al., 2012), or greater neuropathological burden such as cortical Lewy bodies (Hurtig et al., 2000) and/or β‐amyloid and tau pathologies (Aarsland et al., 2017), which likely contribute to the development of PDD.

In terms of the cognitive profile of PDD converters, our findings are in accordance with the results of previous literature. Indeed, ample evidence has suggested that patients with the multiple domain MCI subtype have an increased chance of converting to PDD (Broeders et al., 2013; Goldman & Litvan, 2011; Janvin, Larsen, Aarsland, & Hugdahl, 2006; Lyoo, Jeong, Ryu, Rinne, & Lee, 2010). Here, the amnestic MCI type with storage failure, which is a neuropsychological characteristic of Alzheimer's disease (AD), was more prevalent in PDD converters than in nonconverters, suggesting the possibility of coexistent AD‐related pathology (Petrou et al., 2012; Sabbagh et al., 2009). The severity of the impairments in cognitive performance, particularly in the frontal/executive, memory, and language function domains, was greater in the converter versus nonconverter group, which may also affect the cognitive prognosis (Valenzuela & Sachdev, 2006). The cognitive profile of the PDD converter group was also in line with the results of the cortical thickness analysis in the current study, that is, patients who converted to PDD exhibited cortical thinning that was more extensive, particularly in the frontal regions, relative to those who did not convert to PDD. Interestingly, the area where cortical atrophy was observed in PDD converters compared to nonconverters included the right medial temporal region. The medial temporal lobe is known to be associated with memory function (Squire & Zola‐Morgan, 1991), and atrophy of this region is often regarded as the hallmark of AD‐related pathology (Devanand et al., 2007).

Our structural neuroimaging findings appear to be inconsistent with the dual syndrome hypothesis proposed by the CamPaIGN study (Williams‐Gray et al., 2007; Williams‐Gray et al., 2009): Cognitive deficits with a posterior cortical basis (e.g., semantic fluency, picture copying, verbal memory, and language), which appear more related to multiple neurotransmitter deficits, are a predictor for incident dementia in PD, whereas frontostriatal deficits (e.g., working memory and set‐shifting), which may be related to dopaminergic deficits, are not (Martinez‐Horta & Kulisevsky, 2011). However, it should be noted that the dual syndrome hypothesis was established based only on neuropsychological data without validation using neuroimaging data. This discrepancy was also observed in other previous studies (Compta et al., 2013; Lee et al., 2014) which demonstrated that frontal cortical thinning was a significant PDD predictor, and it might be due to the clinical heterogeneity of PD‐MCI and to differences in the study designs (Lee et al., 2014). Additionally, differences in the ethnic backgrounds might have contributed to these discrepant results. Some evidence has suggested that the heterogeneity in patients' cognitive profiles is associated with their genotype, including variations in the catechol‐O‐methyltransferase (Williams‐Gray, Hampshire, Barker, & Owen, 2008), microtubule‐associated protein tau (Williams‐Gray et al., 2009), and β‐glucocerebrosidase genes (Oeda et al., 2015). It is also unclear whether posterior cortical dysfunction is a single pathological entity (Monchi, Hanganu, & Bellec, 2016). Alternatively, posterior cortical dysfunction might not correspond well to the specific patterns of cortical atrophy, particularly in early stage PD (Segura et al., 2014). Rather, the neural basis of these deficits might be related to the functional basis. Indeed, in our study, the composite scores of cognitive domain tests with a posterior cortical basis were not significantly correlated with the cortical thickness in any region. In this case, altered hypometabolism (Gonzalez‐Redondo et al., 2014), dysregulation of neurotransmitter systems (Chung et al., 2018; Ray & Strafella, 2012), and/or early axonal damage (Wen et al., 2015) may play more important roles in posterior‐cortical neuropsychological dysfunction than cortical atrophy. Nevertheless, the present study demonstrated that the specific pattern of frontal cortical thinning, which was obtained from the LDA, had fair discriminatory power for identifying the patients with PD‐MCI who are at a high risk of dementia. In particular, our findings that PDD converters exhibited cortical thinning in the olfactory cortex with a higher prevalence of hyposmia are in line with previous evidence of olfactory dysfunction as a risk of PDD (Baba et al., 2012; Bohnen et al., 2010; Domellof, Lundin, Edstrom, & Forsgren, 2017). Moreover, the high frequency of cognitive deficits with a posterior cortical basis in PD‐MCI (approximately 80% of patients in both the converter and nonconverter groups in this study) makes it difficult to use this cognitive profile as a specific predictive marker. In this regard, cortical thickness can potentially emerge as a reliable biomarker for PDD conversion, regardless of the discrepancy with previous neuropsychological data.

Our study had some limitations. First, a 4‐year time window to classify the patients into the PDD converter and nonconverter groups, which is important given that the majority of patients with PD eventually develop dementia (Lee et al., 2017), was based on previous studies (Pedersen et al., 2017; Wood et al., 2016), and may be an arbitrary time frame. Second, some studies have found that the best criterion to minimize the inclusion of cognitively normal patients as having MCI was to require cognitive deficits of at least 1.5 SD (Dalrymple‐Alford et al., 2011; Wood et al., 2016) or 2 SD (Goldman et al., 2013; Goldman et al., 2015; McDermott, Fisher, Bradford, & Camicioli, 2018) below the norms, and the 1 SD cutoff in this study might be less stringent. However, we have previously demonstrated that the 1 SD criteria would be sufficiently sensitive to identify patients with PD‐MCI at high risk for future cognitive decline than patients who do not meet any PD‐MCI criterion at baseline, at least in our database (Chung et al., 2019). In addition, stricter levels of impairment in cognitive function might be needed to meet criteria for PDD, although the consensus for the diagnostic cutoffs of PDD has not been established yet. Third, as this was a retrospective study, there is some individual variability in the follow‐up intervals between neuropsychological assessments, although we attempted to perform the neuropsychological tests at a regular interval of 2 years. Fourth, the effects of coexistent AD‐related pathology (Petrou et al., 2012; Sabbagh et al., 2009) or striatal dopamine depletion (Chung et al., 2018) were not considered in this study. Fifth, we cannot completely exclude the possibility of including some cases with dementia with Lewy bodies (DLB) in this study, which may affect the interpretation of the results. However, we applied a highly structured, systematized approach to thoroughly detect the core features of DLB including fluctuating cognition (Walker et al., 2000) and visual hallucination (Cummings et al., 1994), and clinically suspected DLB cases performed additional PET studies such as an ¹⁸F‐fludeoxyglucose PET scans to check whether they demonstrated supportive imaging biomarkers for DLB (McKeith et al., 2017). This effort may minimize the misclassification of prodromal DLB cases (Blanc et al., 2016) into the PD‐MCI cases. Finally, compared to our previous study (Lee et al., 2014), the current study has several strengths, including a more sophisticated study design, criteria for classifying PDD converters that are more stringent, a larger sample size, and longer follow‐up duration, as well as methodological improvements in the neuroimaging analyses.

5. CONCLUSION

In conclusion, our data suggest that cortical thinning, particularly in the frontal region with a clinical relevance to frontostriatal deficits, is a useful biomarker for identifying patients with PD‐MCI who are at a high risk of ongoing cognitive decline. These specific patterns of cortical atrophy may be complemented by the addition of other biomarkers to more accurately predict the progression to PDD.

CONFLICT OF INTEREST

The authors declare no potential conflict of interest.

Supporting information

Figure S1 Correlation analyses between cortical thickness and cognitive performance in patients with PD‐MCI. The composite scores of attention/working memory and frontal/executive function domain tests positively correlated with cortical thickness in the bilateral frontal, temporal, and right posterior parietal cortices (FDR‐corrected p < .05).

Figure S2 Correlation analyses between cortical thickness and cognitive performance in patients with PD‐MCI. The composite scores memory and visuospatial function domains tended to positively correlate with cortical thickness in the relatively small areas (uncorrected p < .001), while the composite scores of language function domain did not correlate with cortical thickness.

Chung SJ, Yoo HS, Lee YH, et al. Frontal atrophy as a marker for dementia conversion in Parkinson's disease with mild cognitive impairment. Hum Brain Mapp. 2019;40:3784–3794. 10.1002/hbm.24631