This cross-sectional study was approved by the Research Ethics Committee of the First Affiliated Hospital, Anhui Medical University (Ethics approval number: PJ2024-06-63). All participants involved in this study provided written informed consent in accordance with the Declaration of Helsinki.

From June 2024 to February 2025, patients who underwent clinical evaluation, laboratory tests, and multimodal neuroimaging assessments at the Department of Nephropathy of The First Affiliated Hospital of Anhui Medical University were enrolled in this study. Furthermore, Fifty-two HCs matched for gender, age, and years of education were recruited from the local community, with three participants excluded due to cerebral infarction or claustrophobia. All patients fulfilled the following inclusion criteria: aged 18–80 years, undergoing continuous ambulatory peritoneal dialysis (CAPD) with standard lactate-based peritoneal dialysis solution (containing 15 or 25 g glucose, 5.38 g sodium chloride, calcium chloride, 0.051 g magnesium chloride, and 4.48 g sodium lactate per 1,000 mL), stable dialysis duration >3 months, and right-handedness. Exclusion criteria were as follows: (1) History of craniocerebral injury or other neurological disorders (e.g., traumatic brain injury, intracranial space-occupying lesions); (2) Prior history of impaired consciousness, affective disorders, or neuropsychological conditions (e.g., depression, anxiety disorders, schizophrenia); (3) Documented substance dependence or abuse (including alcohol, medications, or illicit drugs); (4) Patients with renal transplantation, acute infections, heart failure, or severe hepatic dysfunction; (5) Inability to complete neuropsychological assessments or cranial magnetic resonance imaging (MRI). All imaging data were independently evaluated by two experienced radiologists blinded to participants’ clinical and cognitive profiles. Participants exhibiting suboptimal image quality, severe white matter disease, cerebral infarction (excluding lacunar infarcts), or intracranial hemorrhage were excluded from subsequent analyses.

Clinical and laboratory data of patients were obtained using standardized forms. The following indicators were extracted: Demographic data, including age, gender, years of education, and dialysis duration; Physical examination variables, including systolic blood pressure (SBP), diastolic blood pressure (DBP), height, and body weight with dry abdomen; Health history, including the history of diabetes and peritonitis, which was collected from the electronic medical records of our hospital; and laboratory variables from serum, urine, and peritoneal dialysis fluid, including white blood cell count, hemoglobin, serum albumin, blood urea nitrogen (BUN), creatinine, uric acid, residual glomerular filtration rate (rGFR), C-reactive protein (CRP), serum iron metabolism indicators, alkaline phosphatase, total cholesterol, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), triglycerides, calcium-phosphorus product, serum bicarbonate concentration, intact parathyroid hormone (iPTH), etc. Standard laboratory methods were used to measure the dialysate/plasma creatinine ratio (D/P Cr), total creatinine clearance (CCr), and total Kt/V on an automated chemical analyzer. The specific details are shown in Table 1. Residual renal function, expressed as rGFR, was calculated using the following formula:rGFR = 1/2[urine Cr(μmol/L)/serum Cr(μmol/L) + urine urea(mmol/L)/serum urea(mmol/L)] × urine volume(mL)/1,440 (Nolph et al., 1993). These values were obtained through the peritoneal equilibrium test (PET). Patients should receive standard CAPD treatment before the PET, and the nocturnal peritoneal dialysis fluid should be retained in the abdominal cavity for 8–12 h. No blood biochemical tests were performed on the control subjects.

Neurological examinations were conducted for all participants by two senior investigators in a quiet environment one day prior to MRI examinations. Cognitive function was evaluated using the Beijing-Chinese version of the MoCA, a validated cognitive screening instrument, as the primary outcome measure (Lu et al., 2011). It assesses seven domains: visuospatial/executive function, naming, attention, abstraction, language, delayed memory, and orientation, with a total score ranging from 0 to 30 points (the higher the score, the better the function). To mitigate educational bias, an additional point was added to the final score for individuals with educational attainment <12 years. All participants of mild cognitive impairment (MCI) were diagnosed in adherence to the Petersen criteria, which encompass: (1) subjective memory complaints, (2) MoCA score <26 (adjusted for educational attainment if necessary), (3) normal function in daily life, and (4) absence of dementia (Marcos et al., 2016). In addition, A standardized neuropsychological battery was administered to evaluate emotional status, cognitive processing speed, attentional capacity, executive function, and language abilities: (1) Emotional profiling: self-rating anxiety scale (SAS) and self-rating depression scale (SDS) were utilized to quantify anxiety and depressive symptoms; (2) Processing speed: symbol digit modalities test (SDMT) assessed visual-motor information processing efficiency. The outcome measure is the number of correct symbol-digit pairings completed within a fixed time period. (3) Attention metrics: digit span test (forward and backward versions) measured auditory-verbal working memory and attentional control. (4) Executive function: trail making test (TMT) evaluated cognitive flexibility, while stroop color-word test quantified inhibitory control. The core outcome measure of the former is the total time (in seconds) required to complete the trail-making task, while the outcome indicator of the latter is the total time (in seconds) needed to finish all items under each condition. (5)Language competence: verbal fluency test (VFT)-Fruit Category was employed to assess semantic retrieval and lexical access.

MRI images of each subject were acquired on a 3.0 T MRI system (Siemens Healthcare, Erlangen, Germany) equipped with a 64 - channel head matrix coil. All participants were instructed to maintain strict supine positioning throughout scanning procedures, with foam padding and noise-reducing earplugs applied to minimize motion artifacts and acoustic interference. Quantitative susceptibility mapping was obtained using a three-dimensional multi-echo gradient echo sequence with the following parameters: echo time (TE) 6.3 ms, repetition time (TR) 35 ms, field of view (FOV) 220 × 192 mm², flip angle 20°, acquisition matrix 416 × 291, slice thickness 2.0 mm with 0.4 mm interslice gap, 72 contiguous slices, and total acquisition time 5.3 min. Structural imaging employed a T1-weighted 3D brain volume sequence (MPRAGE) with optimized parameters: TE 2.9 ms, TR 2300 ms, FOV 256 × 240 mm², flip angle 9°, isotropic matrix 256 × 256, 1.0 mm slice thickness with 0.5 mm gap, 208 slices, achieving full brain coverage in 5.1 min. Additional FLAIR imaging was acquired using the following protocol: TE 98 ms, TR 8000 ms, FOV 220 × 200 mm², flip angle 150°, matrix size 320 × 224, 5.0 mm slice thickness with 1.0 mm gap, 24 slices, completed within 98 s.

The QSM data were reconstructed and analyzed using the SuscEptibility mapping PIpeline toolfor phAse images (SEPIA) toolbox (Chan and Marques, 2021) in the MATLAB program (MathWorks, Natick, MA). First, non-brain tissues were removed from the raw phase and magnitude images using the Brain Extraction Tool (BET) with a fractional intensity threshold of 0.3. Phase unwrapping was subsequently performed using a Laplacian-based algorithm to resolve phase discontinuities. Local background magnetic field contributions were corrected via the Laplacian Boundary Value (LBV) method. The susceptibility distribution map was reconstructed using an inverse filter with a k-space threshold of 0.15 to suppress high-frequency noise artifacts. Spatial smoothing was applied using a Gaussian kernel with a full width at half maximum (FWHM) of 2 mm to enhance signal-to-noise ratio (SNR).

The subsequent processing steps were performed using SPM12 based on MATLAB 2017a. The reconstructed QSM images were spatially normalized to the Montreal Neurological Institute (MNI) standard space through a one-step registration approach. The specific workflow consisted of two phases: (1) The Normalise (Estimate) module in SPM12 was employed to perform non-linear registration between the magnitude map of the first echo sequence from individual subjects and the tissue probability map (TPM) in MNI space, during which the deformation matrix was generated and preserved; (2) The Normalize (Write) module was then applied to propagate the preserved deformation field onto the QSM images, achieving spatial normalization of QSM data into MNI space (Figure 1). Finally, all normalized QSM images were resliced to achieve isotropic spatial resolution of 2 × 2 × 2 mm³ through voxel interpolation.

The normalized susceptibility maps were aligned to the automated anatomical Labeling (AAL) parcellation template, and voxel-wise magnetic susceptibility values within each of the 116 brain regions were automatically extracted using code written in MATLAB R2017a. Regional susceptibility characteristics were quantified as the arithmetic mean of voxel values per ROI. To evaluate the relationship between magnetic susceptibility values and cognitive performance, we selected iron-rich subcortical nuclei as ROIs, guided by prior research findings and accounting for the anatomical limitations of the AAL template (Haacke et al., 2005; Moon et al., 2016). The ROIs in this study included the thalamus, caudate nucleus, globus pallidus, putamen, hippocampus, and amygdala.

All statistical analyses were performed using R (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org), EmpowerStats (version 3.0; X&Y Solutions Inc., Boston, MA; https://www.empowerstats.com), and SPSS software (version 25.0; IBM Corp., Armonk, NY, United States). Normality of data distribution was assessed using the Kolmogorov–Smirnov test. Continuous variables conforming to a normal distribution were expressed as mean ± standard deviation (SD), whereas skewed variables were reported as median with interquartile range (IQR). Categorical variables were summarized as frequencies and percentages.

For group comparisons, independent two-sample t-tests and Mann–Whitney U tests were applied to evaluate differences in normally and non-normally distributed continuous variables, respectively, between the PD and HC groups. Pearson’s chi-square test was utilized for categorical variables. For multiple comparisons, tests were corrected using the Bonferroni method (p < 0.05/116 = 0.0004).

Multivariable linear regression models were constructed to examine the association between MoCA scores and ferritin levels. Generalized additive models (GAM) with smooth curve fitting were employed to explore potential curve relationships between MoCA scores and ferritin. To investigate threshold effects in the MoCA-ferritin correlation, subgroup analyses were conducted using stratified multivariable regression models, incorporating interaction terms to assess variability across predefined subgroups (e.g., age, gender, or BMI). The correlation was analyzed using Pearson’s correlation coefficient for normally distributed data or Spearman’s rank correlation coefficient for non-normally distributed data between the mean QSM values of ROIs, cognitive scores, and laboratory data. However, to mitigate the elevated risk of type I errors arising from multiple comparisons, false discovery rate (FDR) correction methodology was implemented throughout the analytical process. The statistically significant p value was set at 0.05.

This study enrolled 52 patients undergoing peritoneal dialysis, categorized into peritoneal dialysis with mild cognitive impairment (PD-MCI) (n = 34) and peritoneal dialysis with normal cognition (PD-CN) (n = 18) groups based on cognitive function assessments. Demographic and clinical characteristics are summarized in Tables 1, 2. Significant intergroup differences were observed in demographic profiles: the PD-MCI group exhibited a higher median age (p = 0.023) and shorter educational attainment (median 7 years vs. 10 years, p < 0.001). Gender distribution (61.8% females in PD-MCI vs. 66.7% in PD-CN) and other baseline characteristics, including BMI, diabetes prevalence, and dialysis duration, showed no significant differences (all p > 0.05).

Laboratory analyses revealed marked disparities in iron metabolism parameters: serum ferritin levels were significantly elevated in the PD-MCI group (p < 0.001), with correspondingly higher transferrin saturation (p < 0.001). Hemodynamically, the PD-CN group demonstrated higher diastolic blood pressure (p = 0.039). Although the PD-MCI group showed a trend toward higher uric acid levels and calcium-phosphorus product values, these differences approached but did not reach statistical significance (p = 0.053 and p = 0.071, respectively). No significant intergroup variations were detected in the remaining laboratory and clinical parameters, including inflammatory markers (CRP), peritoneal transport characteristics, and nutritional indices (all p > 0.05).

As shown in Table 2, patients undergoing PD exhibited multidimensional neuropsychological impairments compared to HC. The two groups showed no statistically significant differences in age, gender distribution, or years of education (all p > 0.05). However, neuropsychological assessments revealed the following: (1) Emotional dimensions: The PD group scored significantly higher on the SAS and SDS than the HC group (both p < 0.001). (2) Global cognitive function: The PD group demonstrated significantly lower MoCA scores (p < 0.001). (3) Processing speed: The SDMT scores were markedly reduced in the PD group (p < 0.001). (4) Working memory: Both forward and backward Digit Span Test scores were significantly diminished in the PD group (both p < 0.001). (5). Executive function: The Stroop Test revealed significantly prolonged completion times for word naming, color naming, and interference tasks in the PD group. (6). Verbal fluency: The PD group showed significantly lower scores on the VFT. Notably, no significant differences were observed between the two groups in completion times for TMT-A or TMT-B. These findings suggest that PD patients experience broad cognitive impairment, with pronounced deficits in information processing speed, working memory, executive function, and emotional regulation, while visual search and task-switching abilities remain relatively preserved.

This study employed a two-stage regression analysis to investigate factors influencing MoCA scores in peritoneal dialysis patients. Univariate linear regression analysis revealed significant associations between MoCA scores and age (β = −0.348, 95% CI: −0.554 to −0.143, p = 0.001), years of education (β = 1.191, 95% CI: 0.485 to 1.896, p = 0.001), serum ferritin (β = −0.040, 95% CI: −0.054 to −0.025, p < 0.001), and transferrin saturation (β = −0.310, 95% CI: −0.567 to −0.053, p = 0.019) (Supplementary Table 1). The borderline significant variable, dialysis duration (p = 0.052), was included in the multivariate model for adjustment. Following multivariate linear regression analysis, age (β = −0.204, 95% CI: −0.400 to −0.007, p = 0.042) and serum ferritin (β = −0.031, 95% CI: −0.046 to −0.017, p < 0.001) retained independent statistical significance (Table 3). Specifically, a 100 μg/L increase in serum ferritin correlated with a 3.1-point decline in MoCA scores. However, the associations with education years (p = 0.614) and transferrin saturation (p = 0.512) were attenuated in the adjusted model. These findings indicate that advanced age and iron overload are independent risk factors for cognitive impairment in peritoneal dialysis patients.

The generalized additive models with smooth curve fits were employed to delineate the nonlinear relationship between SF and MoCA scores, revealing a distinct inflection point at 258.4 μg/L (Figure 2). Subsequent threshold effect analysis using a two-piecewise linear regression model (Table 4) demonstrated divergent associations: above this threshold (SF ≥ 258.4 μg/L), each 1 μg/L increase in SF corresponded to a 0.053-point decline in MoCA scores (95% CI, −0.069 to −0.037). Conversely, below this threshold, each 1 μg/L increase in SF is linked to a slight increase of approximately 0.015 points in MoCA scores (95% CI, −0.018 to 0.047). Serum ferritin and MoCA scores appear to exhibit an inverted U-shaped relationship. Likelihood ratio tests confirmed significant improvement in model fit for the piecewise model compared to the standard linear model (p < 0.001), indicating superior data representation. The area between the upper and lower dashed lines is represented as 95% CI. Each point shows the magnitude of the SF and is connected to form a continuous line. Analyses were adjusted for age, educational attainment, dialysis duration, and transferrin saturation.

Subgroup analyses were performed to assess the robustness and consistency of the SF-MoCA score relationship across potential effect modifiers, including age, gender, BMI, educational attainment, dialysis duration, diabetes, hypertension, and peritoneal transport status. Detailed results were reported in Supplementary Table 2. All interaction tests demonstrated non-significant effects (all P for interaction>0.05), indicating that the inverse association between SF and MoCA scores remained stable regardless of these stratification variables.

Following Bonferroni correction, significant intergroup differences in iron deposition were observed among PD and HC across specific brain regions (Table 5). The PD group exhibited significantly higher susceptibility values in the left amygdala, and right putamen compared to HC (p < 0.005/116 = 0.0004). No statistically significant differences were detected in other regions, including the hippocampus, caudate nucleus, globus pallidus, or thalamus (all p > 0.05/116 = 0.0004).

The MoCA total score was significantly correlated with the susceptibility values of the left amygdala (Figure 3A). Subsequently, the correlation analysis revealed distinct correlation patterns between magnetic susceptibility values in the left amygdala and multidimensional cognitive assessments (Figures 3B–F and Supplementary Table 3). Specifically, the left amygdala exhibited significant positive correlations with SAS (r = 0.297, p = 0.004, FDR corrected) and SDS (r = 0.295, p = 0.004, FDR corrected), while demonstrating negative correlations with DST-forwards, DST-backwards, and VFT (r = −0.371, p < 0.001; r = −0.226, p = 0.019, and r = −0.216, p = 0.022, FDR corrected, respectively). However, the right putamen demonstrated no significant correlations with cognition function measures (all p > 0.05).

Moreover, the susceptibility values in the left amygdala showed no significant correlation with hemoglobin, serum iron, ferritin, or transferrin saturation in peritoneal dialysis patients (all p > 0.05) (Supplementary Figures 1A–D).