Methods, systems, and media for sleep disorder population cognitive risk assessment based on plasma multi-metal function modules

Abstract

The present application relates to the field of in vitro biomarker detection and medical data processing, and discloses a method and system for sleep disorder population cognitive risk assessment based on plasma multi-metal function module and a medium. Peripheral venous blood of a subject is collected and centrifuged to obtain plasma supernatant; ICP-MS is used to quantitatively detect the concentrations of at least 19 metals including calcium and magnesium; the metal concentration data is processed and divided into macro-metal module, metal enzyme / redox module, antioxidant-thyroid axis module, neural active trace element module and environmental exposure metal module; a joint model in the module is constructed to output cognitive risk probability; bootstrap resampling is used to calculate out-of-bag OOB-AUC distribution and winner probability for internal validation and target module determination; finally, individual cognitive risk probability value and / or risk stratification result are output, and an interpretable visualization result can be optionally generated. The present application has the advantages of standardized detection process, high discrimination performance, sufficient internal and external validation, and ease of clinical deployment.

Description

Technical Field

[0001] This invention belongs to the field of in vitro biomarker detection and medical data processing, and in particular relates to a method, system and medium for cognitive risk assessment of sleep disorder populations based on plasma multimetal functional modules. Background Technology

[0002] People with sleep disorders are at higher risk of impairment in cognitive domains such as attention, memory, and executive function. Current clinical cognitive screening methods mostly rely on scales or computerized assessments, which are easily affected by time costs, compliance, and subjective factors; some imaging or neurophysiological examinations are expensive and complicated, which is not conducive to large-scale, standardized, and repeatable screening in outpatient or physical examination settings.

[0003] Therefore, there is a need for a cognitive risk assessment method that is based on in vitro samples, has a standardized detection process, quantifiable output results, and can be robustly validated and externally promoted. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies in cognitive risk assessment for people with sleep disorders, which rely on subjective scales or complex examinations, are difficult to deploy quickly, and lack a stable and scalable objective quantitative indicator system, this invention provides a cognitive risk assessment method, system, and storage medium based on plasma polymetallic detection, functional module modeling, and internal and external validation, so as to realize the output of cognitive risk probability and risk stratification for people with sleep disorders.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: As a first aspect, a method for assisting in the cognitive risk assessment of people with sleep disorders is provided, including a model building phase and a risk inference phase; The model building phase includes: (1) Sample acquisition: peripheral venous blood samples were collected from subjects with sleep disorders in the training sample, and plasma supernatant was obtained by centrifugation after anticoagulation treatment. The cognitive assessment results of the subjects were obtained as the modeling outcome. (2) Quantitative analysis of metals: The plasma supernatant is subjected to quantitative analysis of multiple metal elements to obtain multi-metal concentration data; the multi-metal concentration data includes at least calcium (Ca) and magnesium (Mg), preferably including 19 metal elements: Mg, Ca, Fe, Cu, Zn, Mo, I, Se, Li, Co, Ni, As, Pb, Bi, Sb, Ba, Sr, Tl and Hg; the quantitative analysis of metal elements is preferably performed using inductively coupled plasma mass spectrometry (ICP-MS). (3) Preprocessing and feature construction: The metal concentration data obtained in step (2) is logarithmically transformed to obtain ln(X), and standardized based on the mean μ and standard deviation σ in the training dataset to obtain standardized metal features zln(X), where X represents the concentration of any metal element; when the concentration value of any metal element is lower than its detection limit, LOD / 2, LOD / √2 or multiple interpolation can be used for data processing before the logarithmic transformation and standardization are performed. (4) Functional module construction: The standardized metal features are divided into multiple functional modules according to the preset biological functional domains, and module features are constructed respectively; the functional modules include a macro metal module, a metalloenzyme / redox module, an antioxidant-thyroid axis module, a neuroactive trace element module, and an environmentally exposed metal module; preferably, the macro metal module includes Ca and Mg; the metalloenzyme / redox module includes Fe, Cu and Zn; the antioxidant-thyroid axis module includes Se and I; the neuroactive trace element module includes Li and Bi; the environmentally exposed metal module includes Sb, Ba, Sr, Tl and Hg; (5) Intra-module joint modeling and discriminant assessment: For each functional module, the standardized metal features and covariates within the module are input into the multi-metal joint Logistic risk assessment model within the module to calculate the cognitive risk probability p, and to plot the ROC curve and calculate the AUC; the covariates include one or more of age, gender, body mass index BMI, education level, HAMD-17 score and HAMA score; (6) Internal validation and module selection: Bootstrap resampling validation is performed on the Logistic risk assessment model corresponding to each functional module. In each resampling, the selected samples are used as training samples to build the model, and the out-of-bag samples that are not selected are used as validation samples to calculate the out-of-bag OOB-AUC, thereby obtaining the OOB-AUC distribution of each functional module; the proportion of each functional module that achieves the highest OOB-AUC in multiple resampling is further statistically analyzed as the winner probability; one or more target modules are determined based on the AUC, OOB-AUC distribution and winner probability. The risk inference stage includes: for the individual to be assessed, detecting the concentration of metal elements contained in the target module, or detecting the concentration of multiple metal elements including the metal element in the target module; performing data processing using logarithmic transformation and standardized parameters consistent with the model building stage, and inputting the processed target module features into the corresponding risk assessment model to output the individual's perceived risk probability p and / or risk stratification results; When the target module is a macro-metal module, the risk inference stage detects at least Ca and Mg, and outputs the cognitive risk probability p and / or risk stratification results based on the standardized characteristics of Ca and Mg.

[0006] Optionally, the method further includes: Correlation matrix construction: Calculate the pairwise Spearman rank correlation coefficient ρ between multi-metal elements to form a correlation matrix, and perform redundancy control, feature screening and / or model interpretation output on highly correlated metal pairs according to the preset correlation coefficient threshold; LASSO screening: During the model building phase, LASSO regression is performed on the candidate metal feature set, and λ_min and / or λ_1se are determined through cross-validation to screen core metal features related to cognitive risk; the candidate metal feature set is the 19 metal element features mentioned above, or metal element features within the target module; RCS Nonlinear Analysis: Restricted cubic spline RCS analysis is performed on the core metal features in the target module to construct a nonlinear dose-response relationship between metal concentration and perceived risk; preferably, restricted cubic splines are used to construct nonlinear dose-response features for ln(Ca) and / or ln(Mg), with nodes located at 5%, 35%, 65%, and 95% quantiles; External validation: Validate the discriminative performance of the target module or its simplified feature combination in an external population dataset; preferably, use the total calcium level and magnesium deficiency score MDS in the NHANES 2007-2012 dataset for external validation, calculate the weighted ROC curve and weighted AUC, and construct a Ca×MDS four-level joint stratification based on the total calcium level and MDS, and output the corrected OR value and its confidence interval.

[0007] Further, in step (6), the number of bootstrap resampling times B is 500-2000 times, preferably 1000 times; each resampling calculates the out-of-bag OOB-AUC; the winner probability is the proportion of a certain functional module that obtains the highest OOB-AUC in all resampling.

[0008] As a second aspect, a system for assisting in the cognitive risk assessment of individuals with sleep disorders is provided, comprising: The data acquisition module is used to receive the multi-metal quantitative detection results of the subject's plasma sample and / or the target module's metal quantitative detection results; The preprocessing module is used to perform logarithmic transformation and standardization on the metal concentration data; The module construction module is used to divide standardized metal features into multiple functional modules according to preset biological functional domains and construct module features. The model training and evaluation module is used to establish a multi-metal joint Logistic risk assessment model within the module and generate ROC curves and AUC. The internal validation module is used to perform bootstrap resampling, calculate the OOB-AUC distribution and winner probability to determine the target module; The risk inference module is used to output the individual's perceived risk probability p and / or risk stratification results based on the target module. The results output module is used to output AUC, winner probability, nomogram, risk stratification results, and / or risk assessment report.

[0009] As a third aspect, a computer-readable storage medium is provided having a program stored thereon, which, when executed by a processor, implements the method described in the first aspect above.

[0010] Compared with the prior art, the present invention has the following beneficial effects: (1) Objective quantification and deployability: Based on the in vitro quantitative detection of plasma metal elements and standardized data processing, this invention outputs the cognitive risk probability p and / or risk stratification results, which facilitates auxiliary screening and risk assessment in laboratory, outpatient or physical examination scenarios; (2) Modular modeling and improved discrimination performance: This invention divides the multi-metal features into multiple functional modules according to biological functional domains and establishes joint models within each module. This not only allows for comparison of the discrimination ability of different metal functional modules for cognitive risk, but also improves the biological interpretability of the model construction process. (3) The development stage and the application stage are distinguished: In the model building stage, the present invention can detect 19 metal elements and perform multi-module comparison. After the target module is determined, the risk inference stage can detect only the metal elements contained in the target module, such as Ca and Mg, thereby reducing the detection cost and improving the convenience of clinical deployment. (4) Robustness verification: This invention obtains the OOB-AUC distribution of each functional module through bootstrap resampling, and evaluates the advantage stability of different modules in repeated sampling through winner probability, thereby improving the robustness of target module selection; (5) Interpretability and clinical communication: This invention can be further combined with OR and confidence intervals, correlation matrix, LASSO screening, RCS dose-response curve, nomogram or risk assessment report to improve the interpretability of model results and clinical communication value; (6) Scalability: This invention can be externally validated in external population datasets such as NHANES, and risk association results can be output in a hierarchical manner through Ca and MDS, providing a basis for cross-population promotion and simplified model deployment. Attached Figure Description

[0011] Figure 1 This image shows the plasma polymetallic content detection results, functional module differences, and intermetallic correlation structures of individuals with sleep disorders. Figure 1 A represents the concentration comparison of Mg and Ca in the macro-metal module between the sleep disorder without cognitive impairment (SD) group and the sleep disorder with cognitive impairment (SDCI) group, used to demonstrate the differential characteristics of the macro-metal module in sleep disorder populations with different cognitive states. Figure 1 B represents the concentration comparison of Fe, Cu, and Zn in the metalloenzyme / redox module between the SD and SDCI groups, used to show the differences in plasma metal elements related to metalloenzyme activity and redox homeostasis in cognitive risk groupings. Figure 1 C represents the concentration comparison of I and Se in the antioxidant-thyroid axis module between the SD and SDCI groups, used to show the differences in cognitive risk groupings among metals or metalloids related to thyroid function and antioxidant defense. Figure 1 D represents the concentration comparison of Li and Bi in the neuroactive trace element module between the SD and SDCI groups, used to show the differences in cognitive risk groupings of metal elements associated with neural regulation or trace element exposure. Figure 1 E represents the concentration comparison results of Sb, Ba, Sr, Tl, and Hg in the environmental exposure metals module between the SD group and the SDCI group, used to show the differences in the cognitive risk grouping of environmental exposure-related metal elements. Figure 1 F is the Spearman correlation matrix of 19 plasma metal elements. The horizontal and vertical axes represent each metal element, and the colors represent the direction and magnitude of the correlation coefficient. Red indicates a positive correlation, and blue indicates a negative correlation. The color bars next to the matrix represent the functional modules to which each metal element belongs.

[0012] Figure 2 The graph shows the combined logistic regression effect size and ROC / AUC comparison results for five preset functional modules; among them... Figure 2 A is a multi-metal joint logistic regression forest plot within the module, used to show the adjusted association strength between metal elements and cognitive impairment risk in each functional module. Figure 2 B represents the ROC curves and AUC comparison charts of the Logistic risk assessment model corresponding to the five functional modules, used to evaluate the ability of different functional modules to discriminate the risk of cognitive impairment in people with sleep disorders.

[0013] Figure 3 This is a diagram showing the internal verification results based on bootstrap resampling, where... Figure 3A shows the winner probability results for each functional module based on OOB-AUC. The winner probability represents the proportion of times a certain functional module achieves the highest OOB-AUC out of the total number of resampling attempts during multiple bootstrap resampling validations. Figure 3 B is a box plot of the OOB-AUC distribution for the five functional modules. Each box plot represents the AUC distribution of the corresponding functional module in multiple bootstrap resampling. The dashed line represents the reference line with no discriminative power when the AUC is 0.5.

[0014] Figure 4 The LASSO feature screening process and the Ca and Mg restricted cubic strip RCS dose-response relationship diagram are shown. Figure 4 A is the LASSO regression coefficient path diagram, with the horizontal axis representing log(λ) and the vertical axis representing the regression coefficients corresponding to each candidate metal feature. Figure 4 B is the LASSO cross-validation error curve, with the horizontal axis representing log(λ) and the vertical axis representing binomial deviation (BinomialDeviance). The dashed lines in the graph represent the positions of λ_min and / or λ_1se determined by cross-validation, used to select feature combinations that balance predictive performance and model simplicity. Figure 4 C represents the restrictive cubic spline RCS curve between Ca and cognitive risk, with the x-axis representing ln[Ca concentration (mg / L)] and the y-axis representing the corrected ln(OR). The curve in the figure represents the dose-response relationship between Ca level and cognitive risk, and the shaded area represents the 95% confidence interval. Figure 4 D is the restricted cubic spline RCS curve between Mg and cognitive risk, with the horizontal axis being ln[Mg concentration (mg / L)] and the vertical axis being the corrected ln(OR). The curve in the figure represents the dose-response relationship between Mg level and cognitive risk, and the shaded area represents the 95% confidence interval.

[0015] Figure 5 The figure shows the results of NHANES external validation and Ca×MDS joint stratified risk assessment. Figure 5 A is a flowchart of the sample selection process for the external validation population of NHANES2007–2012. Figure 5 B is the weighted ROC curve from the NHANES 2007–2012 external validation data. This curve is used to evaluate the external discriminative ability of simplified feature combinations such as total calcium level and magnesium deficiency score MDS for cognitive risk. Figure 5 C is a forest diagram showing the impact of the combined Ca and MDS stratification on the risk of cognitive impairment. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are used to illustrate the technical solution and implementation effects of the present invention, and are not intended to limit the scope of protection of the present invention. Any equivalent substitutions or conventional adjustments made to the sample source, combination of detection elements, data preprocessing methods, statistical modeling methods, validation datasets or risk stratification thresholds based on the concept and technical solution of the present invention should be included within the scope of protection of the present invention.

[0017] This invention provides a method for assisting in the cognitive risk assessment of individuals with sleep disorders based on a plasma multi-metal functional module. The method includes a model building stage and a risk inference stage. In the model building stage, peripheral venous blood is collected from subjects with sleep disorders and centrifuged to obtain plasma supernatant. Plasma metal concentration data are obtained using a multi-metal element quantitative detection method. The multi-metal elements preferably include 19 metal elements: Mg, Ca, Fe, Cu, Zn, Mo, I, Se, Li, Co, Ni, As, Pb, Bi, Sb, Ba, Sr, Tl, and Hg, with at least Ca and Mg included. Subsequently, the metal concentration data undergoes logarithmic transformation (ln) and standardization (zln) to obtain standardized metal features. These standardized metal features are then divided into five functional modules according to preset biological functional domains, and a multi-metal joint logistic regression model is constructed within each module to output the cognitive risk probability p. The discriminative ability of different functional modules for cognitive risk is compared using ROC curves and AUC. Bootstrap resampling is used to calculate the out-of-bag (OOB-AUC) distribution and winner probability, thereby determining one or more target modules.

[0018] The five functional modules are not a direct, fixed classification of the 19 metal elements, but rather a functional classification based on the retained metal element set obtained after univariate logistic regression screening. The retained metal element set consists of 19 candidate metal elements that meet the preset statistical screening criteria after univariate logistic regression analysis, preferably those with an FDR q < 0.05 after error detection rate correction using the Benjamini-Hochberg method. Then, according to preset biological functional domains, the metal elements in the retained metal element set are mapped to the macro-metal module, metalloenzyme / redox module, antioxidant-thyroid axis module, neuroactive trace element module, and environmental exposure metal module. Candidate metal elements not included in the retained metal element set are not used as input features for the multi-metal joint logistic risk assessment model within the corresponding module.

[0019] After identifying the target module, LASSO regression can be used to screen 19 candidate metal features or metal features within the target module. Based on the core metal features obtained from the screening, a restricted cubic spline RCS model is constructed to characterize the nonlinear dose-response relationship between metal concentration and cognitive risk, improving the model's interpretability. For the individual to be evaluated, the concentration of metal elements contained in the target module, or the concentration of multiple metal elements including those in the target module, can be detected. The individual's cognitive risk probability and / or risk stratification results are output according to the data processing rules and model parameters determined during the model construction phase. Furthermore, external validation can be performed on external population datasets, and the generalization performance of the target module or its simplified feature combinations in different populations can be evaluated through a joint stratification approach.

[0020] The accompanying drawings in this specification are used to illustrate representative results obtained from each embodiment of the present invention. Figure 1 Used to demonstrate the differences in multimetallic functional modules of plasma and the related structures between metals; Figure 2 Used to display the joint logistic regression effect size and ROC / AUC results of each functional module; Figure 3 Used to demonstrate the OOB-AUC distribution and winner probability obtained from internal Bootstrap validation; Figure 4 Used to demonstrate the LASSO screening process and the RCS dose-response relationship of Ca and Mg; Figure 5 This is used to demonstrate the results of external validation and Ca×MDS joint stratification. The results shown in the figures will be further explained in conjunction with specific steps in the corresponding embodiments below.

[0021] Example 1: Research Subjects, Sample Processing, and Module Division This embodiment includes subjects with sleep disorders who have completed sleep assessment, cognitive assessment, and plasma polymetallic acid testing. In one specific embodiment, a total of 133 subjects with sleep disorders were ultimately included, of which 68 were in the sleep disorder without cognitive impairment (SD) group and 65 were in the sleep disorder with cognitive impairment (SDCI) group.

[0022] Inclusion criteria for participants included: meeting the diagnostic or screening criteria for sleep disorders, or having a PSQI threshold; completing cognitive function assessments; completing peripheral venous blood collection and plasma polymetallic analysis; and having covariate data such as age, sex, body mass index (BMI), education level, HAMD-17 score, and HAMA score.

[0023] Exclusion criteria included: comorbid severe neurological diseases, acute exacerbation of severe mental disorders, malignant tumors, severe liver and kidney dysfunction, acute or chronic infections, autoimmune diseases, pregnancy or lactation, recent high-dose mineral supplementation, significant hemolysis or severe lipemia in the sample, and missing key variables.

[0024] In this embodiment, the PSQI is used to define sleep disorders or sleep quality status, and is not directly used to define cognitive impairment. Cognitive impairment is determined based on computerized cognitive assessments, MMSE, MoCA, or specific cognitive domain scores; the threshold for determining cognitive impairment can be adjusted according to age, education level, or the specifications of the corresponding scale / assessment system.

[0025] Peripheral venous blood samples were collected from subjects, anticoagulated with EDTA, and centrifuged at 3000g, 10min, and 15℃ to obtain plasma supernatant. The supernatant was aliquoted and stored at -80℃, with no more than two freeze-thaw cycles. The concentrations of Mg, Ca, Fe, Cu, Zn, Mo, I, Se, Li, Co, Ni, As, Pb, Bi, Sb, Ba, Sr, Tl, and Hg in the plasma were determined using inductively coupled plasma mass spectrometry (ICP-MS). External standard calibration combined with internal standard correction was used during the detection process. Since Bi was the analyte, it was not used as an internal standard; Sc, Ge, Rh, In, or Ir were preferred internal standards, with a concentration range of 0.5–20 μg / L, preferably 5–10 μg / L. For each batch of tests, calibration blanks, reagent blanks, program blanks, low / medium / high concentration quality control samples, pooled plasma quality control samples, and repeat samples were prepared. One quality control sample was inserted for every 10–20 actual samples tested to monitor the stability of the detection process. For data below the detection limit, LOD / 2, LOD / √2, or multiple interpolation can be used for processing, and consistent data processing rules should be applied during the model training and subsequent inference phases.

[0026] Nineteen candidate metals were divided into five functional modules based on predefined biological functional domains: macrometals (Mg and Ca); metalloenzymes / redox mechanisms (Fe, Cu, and Zn); the antioxidant-thyroid axis (Se and I); neuroactive trace elements (Li and Bi); and environmentally exposed metals (Sb, Ba, Sr, Tl, and Hg). This module division was used to transform the multi-metal detection results into a modular set of features with biological interpretability.

[0027] Figure 1 The results showed that there were modular differences in the metallicity spectrum among sleep disorder patients with different cognitive states, suggesting that plasma metallic element levels may be related to cognitive risk. At the same time, there were different degrees of correlation among the 19 metallic elements. The Spearman correlation matrix between the metals can be used to show the correlation structure within and between modules, and can serve as the basis for subsequent redundancy control, feature screening and reduction and model interpretation.

[0028] Example 2: Training Set, Modeling, and ROC / AUC Evaluation The training dataset in this embodiment is the internal modeling cohort of sleep disorder subjects finally included in Example 1, for which metal detection, cognitive assessment, and covariate processing were completed. The metal concentration X is logarithmically transformed to obtain ln(X), and then standardized according to the mean μ and standard deviation σ of ln(X) in the training dataset to obtain zln(X). The calculation formula is as follows: Where X represents the concentration of any metal element, and μ and σ represent the mean and standard deviation of the corresponding metal element ln(X) in the training dataset, respectively. In the subsequent risk inference stage for the individuals to be evaluated, μ and σ determined from the training dataset are still used for standardization to avoid data leakage and model bias.

[0029] For each functional module, the zln characteristics of the metal elements within that module, along with covariates, are input into a logistic regression model. The outcome variable is the presence or absence of cognitive impairment. The covariates include one or more of age, gender, BMI, education level, HAMD-17 score, and HAMA score. The model outputs the cognitive risk probability p, the adjusted OR value for the metal elements within the module, and the 95% confidence interval. Furthermore, ROC curves are plotted and AUC is calculated to evaluate the ability of each functional module to discriminate cognitive risk in individuals with sleep disorders.

[0030] Figure 2 The results show that the joint logistic model of the five functional modules differs in effect size distribution and ROC / AUC performance, indicating that metal combinations from different biological functional domains contribute differently to cognitive risk. Figure 2 Figure A lists the odds ratio (OR), 95% confidence interval, and p-value for each metal element. An OR value less than 1 indicates that elevated levels of that metal element are associated with a reduced risk of cognitive impairment, while an OR value greater than 1 indicates that elevated levels of that metal element are associated with an increased risk of cognitive impairment. The figure shows that elements such as Ca in the macrometal module, I and Se in the antioxidant-thyroid axis module, and Li in the neuroactive trace element module are significantly associated with cognitive risk. Figure 2 Panel B shows that the AUC for the macrometal module was 0.900, the AUC for the antioxidant-thyroid axis module was 0.873, the AUC for the metalloenzyme / redox module was 0.792, the AUC for the neuroactive trace element module was 0.791, and the AUC for the environmental exposure metal module was 0.792. These results suggest that different metal functional modules have different risk discrimination performances, with the macrometal module exhibiting higher discrimination ability and potentially serving as a candidate target module for subsequent internal validation and risk inference.

[0031] Example 3: Bootstrap Internal Validation and Target Module Optimization This embodiment uses bootstrap resampling to perform internal validation of each functional module model. Bootstrap resampling does not generate new input features, but is an internal validation method used to evaluate model stability. Each time, sampling with replacement is performed from the training dataset, and samples with the same number of original samples are drawn as the bootstrap training set; samples not drawn in this sampling process constitute out-of-bag samples, or OOB samples for short.

[0032] In each bootstrap training set, a Logistic risk assessment model corresponding to each of the five functional modules is fitted. Then, the predicted probability and OOB-AUC of each module model are calculated using the corresponding OOB samples. The above process is repeated 500–2000 times, preferably 1000 times, to obtain the OOB-AUC distribution of each functional module.

[0033] Winner probability is defined as the proportion of times a particular functional module's OOB-AUC is the highest among all functional modules in multiple bootstrap validations, out of the total number of resampling attempts. The target module is determined based on a combination of the original AUC, the OOB-AUC distribution, and the winner probability. Specifically, if a functional module has a high original AUC, a high and stable OOB-AUC distribution, and a high winner probability, it can be identified as the target module.

[0034] Figure 3 The results show that the OOB-AUC distributions of each functional module differ in the bootstrap repeated sampling, and winnerprobability reflects the stability of different modules in achieving optimal discrimination performance in repeated validation. Figure 3 Figure A shows that a total of 1000 valid bootstrap validations were completed. The winner probability was 87.4% for the macro-metal module, 12.2% for the antioxidant-thyroid axis module, and 0.4% for the neuroactive trace element module. This result indicates that the macro-metal module most often achieves the best discriminative performance in repeated sampling validation and has good model stability. Figure 3B shows that the OOB-AUC distributions differ among different functional modules. The macro-metal module exhibits a generally higher and more stable OOB-AUC distribution, further supporting its selection as a target module for subsequent risk inference and clinical deployment. These results can be used to support the selection of target modules. It should be noted that the 19 metal elements are primarily used for multi-module comparison, target module optimization, and proof of inventiveness during the model development phase. In the final risk inference or clinical deployment phase, one or more target modules can be selected, and only the metal elements required for the target module can be detected. For example, when the target module is a macro-metal module, risk inference can be based on Ca and Mg, thereby reducing detection costs and improving the convenience of clinical application.

[0035] Example 4: LASSO screening and RCS nonlinear analysis This embodiment further employs LASSO regression to screen core metal features related to cognitive risk. LASSO feature screening can use standardized features of 19 candidate metals as input to verify which metals contribute most to cognitive risk among all candidate metals; it can also use metal features within the target module as input to further simplify and interpret the target module. Cross-validation is used to determine λ_min and / or λ_1se, and core metal features are screened accordingly.

[0036] When the LASSO results show that Ca and / or Mg are the core features, this mutually supports the ROC / AUC results and bootstrap internal validation results of the aforementioned macro-metal module, indicating that Ca and Mg make a significant contribution to cognitive risk assessment in individuals with sleep disorders. Subsequently, for Ca and / or Mg in the target module, a restricted cubic spline RCS model is established using ln(Ca) and / or ln(Mg) as continuous variables. The nodes can be set to 5%, 35%, 65%, and 95% quantiles, and the overall correlation P-value, nonlinear P-value, and dose-response curve are output within the framework of a covariate-corrected logistic model.

[0037] Figure 4The results show that the LASSO screening process can be used to identify core metal variables from candidate metal features, and the RCS curves can further demonstrate the nonlinear dose-response relationship between Ca and / or Mg and cognitive risk. The results show that as the penalty parameter λ changes, the regression coefficients of different metal features gradually shrink to 0; metal features that still retain non-zero coefficients can be considered as candidate core features related to cognitive risk. Furthermore, there is a significant overall association between Ca and cognitive risk (P=0.0046); the nonlinearity test shows P=0.0855, suggesting that the association between Ca and cognitive risk may mainly be linear or approximately linear. There is also a significant overall association between Mg and cognitive risk (P=0.0002); the nonlinearity test shows P=0.5540, suggesting that the association between Mg and cognitive risk mainly is linear or approximately linear. These RCS results can be used to further explain the continuous dose-response relationship between Ca and Mg and cognitive risk in the macro-metal module, helping to improve the model's interpretability and providing a basis for subsequently constructing a simplified risk inference model.

[0038] Example 5: NHANES External Validation and MDS Definition This embodiment uses an external population dataset to perform external validation of the target module or its simplified feature combination. The external validation dataset can be selected from the NHANES 2007–2012 multi-period merged data. The sample of the external validation population was screened, initially including a total of 30,442 participants from NHANES 2007–2012. Those under 18 years of age, those lacking sleep or cognitive-related data, those lacking BMI assessment data, those lacking depression assessment data, those lacking anxiety assessment data, those lacking total calcium data, and those lacking alcohol consumption information were excluded in sequence. Finally, 2,422 participants were obtained for external validation analysis, of which 2,143 were in the sleep disorder without cognitive impairment group (SD), and 279 were in the sleep disorder with cognitive impairment group (SDCI).

[0039] Inclusion criteria were: subjects with data on sleep status variables, cognitive outcome variables, serum total calcium or total calcium-related variables, MDS composition variables, age, sex, BMI, education level, depression or anxiety-related variables, and sampling weights.

[0040] Exclusion criteria were: minors, pregnant women, individuals with missing key variables, samples with extreme outliers, and individuals with severe renal dysfunction, malignant tumors, or other serious diseases that needed to be excluded based on the research objectives.

[0041] In external validation, total calcium level and magnesium deficiency score (MDS) can be used as simplified validation features for the target module. MDS can be scored according to the following items: current use of diuretics (1 point); current use of proton pump inhibitors (PPIs) (1 point); eGFR of 60–90 mL / min / 1.73 mcg / min. 2Record 1 point, eGFR < 60 mL / min / 1.73m 2 2 points are awarded for excessive drinking; 1 point is awarded for excessive drinking, defined as more than 2 drinks per day for men and more than 1 drink per day for women. The scores are added together to obtain the MDS, and the higher the score, the greater the tendency for magnesium deficiency.

[0042] For external validation, a weighted logistic model and weighted ROC curve were established under complex sampling weights, and a weighted AUC was calculated to evaluate the discriminative ability of the combination of total calcium level and MDS on cognitive risk. Furthermore, a four-level joint exposure grouping could be constructed based on high / low total calcium and high / low MDS, outputting the corrected OR value and 95% confidence interval for each group relative to the reference group.

[0043] Figure 5 The results show that external population data can be used to validate the generalization performance of the target module or its simplified feature combination; the weighted ROC / AUC results are used to evaluate the model's discriminative ability on external data. Figure 5 B shows a weighted AUC of 0.8220, indicating that this simplified feature combination still has good cognitive risk discrimination performance in the external population; Figure 5 C constructs a four-level combined exposure group based on high / low total calcium level and high / low MDS, and displays the adjusted OR and 95% confidence interval of each group relative to the reference group. The combined Ca×MDS stratification results can be used to demonstrate the difference in cognitive risk under the combined effect of calcium level and magnesium deficiency tendency, thereby supporting the cross-population promotion and clinical application of the target module.

[0044] Accordingly, the present invention provides a system for the cognitive risk assessment method for individuals with sleep disorders, characterized in that it includes: The data acquisition module is used to receive the results of multimetal quantitative detection and covariate data of plasma samples; The preprocessing module is used to perform logarithmic transformation and standardization of metal concentration; the module construction module is used to divide the standardized metal features into multiple functional modules according to preset functional domains. The model training and evaluation module is used to build a multi-metal joint Logistic model within each functional module and generate ROC curves and AUC. An internal validation module is used to perform bootstrap resampling and calculate the OOB-AUC distribution and winner probability to determine the target module; The risk inference module is used to output an individual's perceived risk probability value p and / or risk stratification results based on the target module. The results output module is used to output AUC, winner probability, nomogram, and / or risk assessment report.

[0045] Accordingly, the present invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the methods described above. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0046] In summary, this invention utilizes in vitro quantitative detection and standardized processing of plasma multimetals, employs modular multimetal co-modeling and performs ROC / AUC comparisons, evaluates model robustness through bootstrap internal validation and selects target modules, further characterizes core metal features and their dose-response relationships through LASSO screening and RCS analysis, and can complete external validation and joint stratified assessment in external population data such as NHANES, achieving objective quantitative output and widespread deployment of cognitive risk assessment for individuals with sleep disorders.

[0047] While the present invention has been disclosed with reference to the above embodiments, it is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made to the present invention by those skilled in the art without departing from the concept and scope of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A method for cognitive risk assessment of sleep disorder populations based on plasma polymetallic functional modules, characterized in that, It includes a model building phase and a risk inference phase. The model building phase includes the following steps: (1) Peripheral venous blood was collected from subjects with sleep disorders in the training sample, and plasma supernatant was obtained by centrifugation after anticoagulation treatment; (2) The plasma supernatant was subjected to quantitative detection of candidate multimetal panels to obtain candidate metal element concentration data; the candidate multimetal panels include Mg, Ca, Fe, Cu, Zn, Mo, I, Se, Li, Co, Ni, As, Pb, Bi, Sb, Ba, Sr, Tl and Hg; (3) The candidate metal element concentration data are logarithmically transformed and standardized based on the training dataset to obtain standardized metal feature vectors; (4) Perform univariate logistic regression screening on each candidate metal element in the standardized metal feature vector, and determine the set of retained metal elements according to the preset statistical screening criteria; then map the metal elements in the set of retained metal elements to five functional modules according to the preset biological functional domains and construct module features. The five functional modules are macro metal module, metalloenzyme / redox module, antioxidant-thyroid axis module, neuroactive trace element module and environmental exposure metal module. (5) For each functional module, the standardized metal features and covariates in the functional module are input into the multi-metal joint logistic risk assessment model in the module to obtain the cognitive risk probability value p, and the receiver operating characteristic curve (ROC) and area under the curve (AUC) of the functional module model are calculated. (6) Based on the model selection criteria, the target module for cognitive risk assessment is determined among the five functional modules; the model selection criteria include AUC comparison, out-of-bag OOB-AUC distribution under bootstrap resampling, and winner probability; The risk inference stage includes: inferring the cognitive risk of the individual to be assessed based on the target module and its corresponding multi-metal joint logistic risk assessment model within the module.

2. The method according to claim 1, characterized in that, In step (2), the quantitative detection of metal elements is performed using inductively coupled plasma mass spectrometry (ICP-MS).

3. The method according to claim 1, characterized in that, The metal element composition of the five functional modules is as follows: the macro-metal module includes Mg and Ca; the metalloenzyme / redox module includes Fe, Cu and Zn; the antioxidant-thyroid axis module includes Se and I; the neuroactive trace element module includes Li and Bi; and the environmental exposure metal module includes Sb, Ba, Sr, Tl and Hg.

4. The method according to claim 1, characterized in that, In step (3), the formula for calculating the standardized metal feature vector is: Where zln(X) is the standardized metal feature vector, μ and σ are the mean and standard deviation of ln(X) for all subjects, and X is the concentration of any metal element.

5. The method according to claim 4, characterized in that, When the concentration of any metal element is below its detection limit, LOD / 2 or LOD / can be used. Alternatively, data can be processed using multiple interpolation before undergoing the logarithmic transformation and standardization.

6. The method according to claim 1, characterized in that, Also includes: Based on the multi-metal concentration data obtained in step (2), calculate the pairwise Spearman rank correlation coefficient ρ to form a correlation matrix, and perform redundancy control, feature screening and / or model interpretation output on highly correlated metal pairs according to the |ρ| threshold.

7. The method according to claim 1, characterized in that, In step (6), the number of bootstrap resampling times B is 500-2000 times, preferably 1000 times; the out-of-bag OOB-AUC is calculated for each resampling; the winner probability is the proportion of a certain functional module that obtains the highest OOB-AUC in all resampling.

8. The method according to claim 1, characterized in that, The target module in the risk inference phase is the macro-metal module.

9. A system for implementing the cognitive risk assessment method for sleep disorder populations according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to receive the results of multimetal quantification and covariate data of plasma samples; The preprocessing module is used to perform logarithmic transformation and standardization on the metal concentration; The module building module is used to divide standardized metal features into multiple functional modules according to preset functional domains; The model training and evaluation module is used to build a multi-metal joint Logistic model within each functional module and generate ROC curves and AUC. An internal validation module is used to perform bootstrap resampling and calculate the OOB-AUC distribution and winner probability to determine the target module; The risk inference module is used to output an individual's perceived risk probability value p and / or risk stratification results based on the target module. The results output module is used to output AUC, winner probability, nomogram, and / or risk assessment report.

10. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the method for cognitive risk assessment of sleep disorder populations based on plasma polymetallic functional modules as described in any one of claims 1-8.

Images