Data were derived from the MRI300 study, part of the Clinical Research for Dementia of South Korea (CREDOS) study, a nationwide multicenter study starting in 2005 of elderly individuals with dementia in South Korea. This study developed common protocols, including its own ischemic scale, and recruited patients with subjective memory impairment, mild cognitive impairment (MCI), MCI of a subcortical vascular type, Alzheimer's Disease (AD), and subcortical vascular dementia (SVaD). The MRI300 project, which was devoted to evaluate the relationship between various clinical variables and WMLs in patients with mild-to-moderate dementia, recruited approximately 300 patients with newly diagnosed mild-to-moderate dementia (clinically probable AD or SVaD) from March 2007 to April 2008, and was performed at 13 dementia clinics in university or general hospitals throughout the country.

Patients included in the MRI300 study were as follows: 1) aged 50-85 years; 2) had mild to moderate dementia, as determined by evaluations such as a complete medical history including vascular risk factors (such as hypertension, diabetes, hyperlipidemia, previous stroke history, and smoking), physical and neurologic examinations, neuropsychological tests; vascular risk factors were regarded as positive either if the patient had previously been diagnosed with associated diseases or if he or she was currently under medications for those diseases. 3) had undergone MRI scans; 4) had a standard laboratory dementia screen (complete blood counts, blood chemistry panels, vitamin B₁₂/folate levels, syphilis serology, and thyroid function tests); 5) provided a written informed consent; and 6) had a primary caregiver who could supervise evaluations assessing the condition of the patient throughout the study, and to provide input into efficacy assessments in accordance with all protocol requirements. Mild-to-moderate dementia was defined as a clinical dementia rating (CDR) of 0.5, 1, or 2 [21]. Patients had either AD with varying degrees of white matter changes or SVaD. All patients diagnosed with clinically probable AD fulfilled the criteria for probable AD proposed by the NINCDS ADRDA (National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association)[22]. Patients diagnosed with SVaD had to have symptoms fulfilling both the clinical criteria for vascular dementia as described in the Diagnostic and Statistical Manual of Mental disorders, 4th edition (DSM-IV) and the imaging criteria of the CREDOS study. These criteria included periventricular capping or banding ≥10 mm in length and deep WMLs ≥25 mm in length.

Patients falling between the full diagnostic criteria for AD and SVaD were regarded as having AD, with or without WMLs. Therefore, rather than dichotomizing these patients into either AD or SVaD categories, we classified all patients according to the ischemic scale the CREDOS defined (Table 1). We therefore excluded patients with degenerative etiologies other than AD, such as idiopathic Parkinson's disease, diffuse Lewy body disease, corticobasal degeneration, and progressive supranuclear palsy. We also excluded patients with structural lesions such as territorial infarction, intracranial hemorrhage, brain tumor, and hydrocephalus. Those with current or past neurological or psychiatric illnesses such as schizophrenia, epilepsy, encephalitis, and severe head trauma were ruled out.

To investigate the relationship between WMLs and LIs, we selected from the MRI300 study patients with SVaD and those with AD containing moderate degree of WMLs. We considered the latter as having AD with cerebrovascular disease (CVD), one of the subtypes of vascular cognitive impairment (VCI) [23, 24]. We designated these two groups of patients as having with vascular cognitive impairment with subcortical vascular lesion (VCI-S). Of the 289 patients enrolled into the MRI300 study, 128 (44.2%) were chosen, including 106 with AD with CVD and 22 with SVaD.

The study was approved by the institutional review board of all participating centers. Written informed consent was obtained from patients and their caregivers.

All subjects were assessed in 2005 and 2006 by MRI, following a standard protocol within 6 months of the clinical investigation. Cerebral MRI included axial T1-, T2-weighted images, gradient echo sequences, and fluid-attenuated inversion recovery (FLAIR) images (1.5T Vision Philips, Siemens, General Electric). The entire brain was imaged in the axial plane with 1.6-10 mm slice thicknesses.

Three neurologists trained in rating ischemic white matter changes, and blinded to clinical and functional data, rated the white matter changes on the T2 axial and/or FLAIR images. The CREDOS rating scales were developed by the CREDOS study steering committee, with modifications from Fazekas' [25] and Scheltens's [26] scales. WMLs were evaluated in periventricular WML (P rating) and deep subcortical WML (D rating) areas separately and the D and P ratings were combined to provide a final ischemia degree. Deep subcortical WMLs were divided into D1 (deep subcortical WML<10 mm), D2 (10≤deep subcortical WML<25 mm), and D3 (≥25 mm) based on the longest diameter of lesions. Periventricular WMLs were classified into P1 (capping and banding <5 mm), P2 (between P1 and P3), and P3 (capping or banding≥10 mm) based on the size of cap and band, which were perpendicular and horizontal to the ventricle, respectively. The results were combined to provide a ischemia rating as mild (D1P1, D1P2), moderate (neither minimal nor severe group), or severe (D3P3)(Table 1). As previously mentioned, D3P3 was used as the imaging criteria for SVaD diagnosis. The inter-rater reliabilities for assessment of periventricular, deep, or total white matter changes were excellent (κ=0.726-0.905).

The number of lacunae was rated visually by a single rater (HY Kim, MD) blinded to clinical information and diagnosis. LIs were defined as areas of focal hyperintensity on T2-weighted images, hypointensity on T1-weighted images and hypointensity with perilesional halo on FLAIR of size 3 to 20 mm and located in the subcortical white matter, basal ganglia and thalamus. Non-lacunar territorial infarcts were excluded from the counting of LIs. Ambiguous lesions were reviewed through consultation with a neuroradiologist (RH Juh, PhD), blinded to patient history and clinical information.

The ANALYZE software package (Mayo Clinic, USA), version 9.0, was used to process Images, to estimate WML volume and to rate LIs. Due to the different slice thicknesses of the FLAIR images, we utilized ANALYZE software to resample the original FLAIR images with slice thicknesses 4 mm, followed by the use of the region-of-interest (ROI) function within the package, allowing the rater to create automated traces along with WML on individual images. When the largest diameter of a WML was adjacent to the ventricular lining (e.g. adjacent to the occipital and frontal horns, or the lateral ventricles), that WML was considered periventricular; otherwise, it was considered deep subcortical WML. To measure its volume, the ROI of each slice was multiplied by the slice thickness (4 mm) and all ROI volumes from each images was summed to calculate the total volume, in cubic milliliters. All summation processes were performed automatically by the ANALYZE program.

General cognitive function was measured using in Korean version of Mini-Mental State Examination (K-MMSE) and instrumental ADL (IADL) tasks were measured using the Seoul-Instrumental Activities of Daily Living (S-IADL) [27]. We evaluated the S-IADL from the perspective of current performance.

Since WML volume data were positively skewed, we assessed the relationship between the natural log of WML volume and LIs. Univariate comparisons between demographic data and periventricular and deep subcortical WML volume and the number of LIs were performed using univariate Poisson log-linear models. Periventricular and deep subcortical WML volumes were entered separately, along with adjusted variables, into the multivariate Poisson log-linear models. Adjusted variables were selected for multivariate analysis if their p values were below 0.2 in the univariate log-liner model. Spearman's correlation analysis was used to assess the relationship among K-MMSE, S-IADL and WML volume. Statistical significance was set at p<0.05. All statistical analysis was performed using SPSS for Windows version 18.0 (Windows, Chicago, IL, USA) or SAS, version 9.1 (SAS Institute, www.sas.com).