A prediction model and web-based prediction score calculator for the comorbidity of psoriasis and diabetes

Abstract

The present invention relates to a prediction model for the comorbidity of psoriasis and diabetes, wherein the prediction model for the comorbidity of psoriasis and diabetes is constructed by the following methods: S1 data collection and preprocessing; S11 collecting data of psoriasis patients; S12 supplementing and preprocessing missing data; S2 variable selection: S21 using the least absolute shrinkage and selection operator regression method to screen key variables; S22 selecting important predictive variables S3 model construction: S31 using a multifactor logistic regression model for modeling; S32 optimizing model parameters by 10-fold cross validation during model construction; S4 model evaluation: S41 using the area under the receiver operating characteristic curve to evaluate the predictive performance of the model; S42 using the Hosmer-Lemeshow test to evaluate the calibration of the model; S43 verifying the net effect of the model by decision curve analysis and clinical impact curve. The present invention also provides a prediction webpage version prediction score calculator. The prediction results and analysis reports provided by the system of the present invention provide strong support for clinical decision-making and patient management.

Description

Technical Field

[0001] The present invention relates to the technical field of medical informatics, and in particular to a prediction model for the comorbidity of psoriasis and diabetes and a web-based prediction score calculator. Background Art

[0002] Psoriasis is a chronic, inflammatory skin disease. In recent years, studies have found that patients with psoriasis have a significantly increased risk of developing diabetes. In existing clinical research and practice, there is a lack of an effective tool that can predict the risk of diabetes in patients with psoriasis at an early stage. This leads to difficulties in early identification and intervention of high-risk patients with diabetes comorbidity in clinical practice. Therefore, it is of great clinical significance to develop a prediction model based on big data and machine learning to assess the risk of diabetes comorbidity in patients with psoriasis.

[0003] Chinese patent document CN114783607A discloses a method for constructing a surgical transfusion risk prediction model and a network calculator thereof, which uses a univariate logistic regression model to calculate the correlation between each clinical indicator and the transfusion risk, screens out clinical indicators related to the transfusion risk for multivariate logistic regression calculation, and screens out independent predictive factors to construct a transfusion prediction model; and generates a network calculator based on the formula of the weight coefficient score of each predictive factor in the prediction model; collects key information data sets of surgical patients, inputs the key information data sets into the network calculator, and obtains the predicted transfusion probability. The transfusion prediction model constructed by the present invention can predict the probability of perioperative transfusion of surgical patients before surgery, and a network calculator is generated based on the model, which can be applied to medical electronic medical records to provide automated electronic decision-making for preoperative planning and blood management.

[0004] However, there are no reports on a prediction model and web-based prediction score calculator for the comorbidity of psoriasis and diabetes. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies in the prior art and provide a prediction model and a web-based prediction score calculator for the comorbidity of psoriasis and diabetes. By collecting and analyzing a variety of clinical data of patients, early prediction of the risk of comorbidity of diabetes in psoriasis patients can be achieved, thereby providing decision support for clinicians, optimizing patient management, and reducing the incidence of comorbidity of diabetes.

[0006] In a first aspect, the present invention provides a prediction model for the comorbidity of psoriasis and diabetes, wherein the prediction model for the comorbidity of psoriasis and diabetes is constructed by the following method:

[0007] S1 Data collection and preprocessing:

[0008] S11 collects data on patients with psoriasis;

[0009] S12 supplements and preprocesses missing data;

[0010] S2 variable selection:

[0011] S21 uses the least absolute shrinkage and selection operator regression methods to screen key variables;

[0012] S22 Select important predictor variables

[0013] S3 model building:

[0014] S31 was modeled using a multivariate logistic regression model;

[0015] During the model building process, S32 optimized the model parameters through 10-fold cross validation;

[0016] S4 Model Evaluation:

[0017] S41 used the area under the receiver operating characteristic curve to evaluate the predictive performance of the model;

[0018] S42 used the Hosmer-Lemeshow test to assess the calibration of the model;

[0019] S43 The net effectiveness of the model was verified by decision curve analysis and clinical impact curve.

[0020] As a preferred example, step S12 of supplementing and preprocessing the missing data includes the following steps:

[0021] S121 Identification and supplementation of missing data, S122 Outlier processing, S123 Variable standardization, S124.

[0022] Data consistency check, S125 data partitioning and S126 data security and desensitization.

[0023] As another preferred example, step S21 uses the least absolute shrinkage and selection operator regression method to screen key variables, specifically including: age, BMI, marital status, history of drug allergy, history of tumor, history of joint swelling and pain, history of finger or toe joint swelling, nail lesions, nail pitting, PASI score and psoriasis course.

[0024] In a second aspect, the present invention provides a web-based prediction score calculator for predicting the comorbidity of psoriasis and diabetes, which is constructed by the following method:

[0025] S1 Data collection and preprocessing:

[0026] S11 collects data on patients with psoriasis;

[0027] S12 supplements and preprocesses missing data;

[0028] S2 variable selection:

[0029] S21 uses the least absolute shrinkage and selection operator regression methods to screen key variables;

[0030] S22 Select important predictor variables

[0031] S3 model building:

[0032] S31 was modeled using a multivariate logistic regression model;

[0033] During the model building process, S32 optimized the model parameters through 10-fold cross validation;

[0034] S4 Model Evaluation:

[0035] S41 used the area under the receiver operating characteristic curve to evaluate the predictive performance of the model;

[0036] S42 used the Hosmer-Lemeshow test to assess the calibration of the model;

[0037] S43 validates the net effectiveness of the model through decision curve analysis and clinical impact curve;

[0038] S5: A web-based calculator where users can input the patient’s clinical data and the system will calculate the diabetes risk score based on the trained model.

[0039] As a preferred example, the system outputs the classification of diabetes risk according to the results of model calculation, which is divided into low risk, medium risk and high risk.

[0040] As another preferred example, S12 supplements and preprocesses the missing data, including the following steps:

[0041] S121 Identification and supplementation of missing data, S122 Outlier processing, S123 Variable standardization, S124.

[0042] Data consistency check, S125 data partitioning and S126 data security and desensitization.

[0043] As another preferred example, step S21 uses the least absolute shrinkage and selection operator regression method to screen key variables, including: age, BMI, marital status, history of drug allergy, history of tumor, history of joint swelling and pain, history of finger or toe joint swelling, nail lesions, nail pits, PASI score and psoriasis course

[0044] The advantages of the present invention are:

[0045] 1. Improve prediction accuracy: Based on big data and advanced machine learning algorithms, it can accurately predict the risk of diabetes comorbidity in psoriasis patients and identify high-risk patients.

[0046] 2. Early intervention: By identifying high-risk patients early, clinicians can take timely intervention measures to reduce the incidence of diabetes and improve the quality of life of patients.

[0047] 3. Optimized management: The prediction results and analysis reports provided by the system provide strong support for clinical decision-making and patient management, and improve the quality of medical services. Description of the drawings:

[0048] Figure 1 A prediction model and web-based prediction score calculator flow chart for the comorbidity of psoriasis and diabetes. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0050] Example 1

[0051] Please see Figure 1 A prediction model and web-based prediction score calculator flow chart for the comorbidity of psoriasis and diabetes.

[0052] 1. Data collection and preprocessing:

[0053] Data on psoriasis patients were collected from 49 hospitals across the country, including gender, age, BMI, medical history and other health-related factors.

[0054] The data comes from a nationwide multi-center collaboration under the China Psoriasis Diagnosis and Treatment Center platform, including 49 qualification-reviewed hospitals across the country, and contains clinical data of 17,124 psoriasis patients collected from June 24, 2020 to June 30, 2021.

[0055] The data collected covers a number of health-related factors including:

[0056] (1) Demographic information

[0057] Including gender, age, body mass index (BMI), marital status, education level, etc.

[0058] (2) Lifestyle and medical history

[0059] Including smoking history, drinking history, drug allergy history, tumor history, etc.

[0060] (3) Psychological and emotional state

[0061] The patient's psychological status was recorded, including the degree of depression (none, mild, severe, extremely severe).

[0062] (4) Involvement of the musculoskeletal system

[0063] This includes a history of painful swollen joints, heel pain, swollen finger or toe joints, and any joint involvement.

[0064] (5) Skin involvement

[0065] It includes the involvement of the scalp, palms and soles of feet, external genitalia, as well as manifestations such as nail pitting and nail lesions.

[0066] (6) Other system involvement

[0067] Possible other systemic manifestations, such as nail involvement, nail pitting, etc., were recorded.

[0068] (7) Psoriasis condition

[0069] These included the course of psoriasis, family history, disease progression (such as gradual worsening, stable, remission), severity of pruritus (none, mild, severe, very severe), and the Psoriasis Area and Severity Index (PASI) and the proportion of body surface area (BSA) affected.

[0070] Data were collected from all participating institutions through a standardized electronic data collection form (eCRF). To ensure the accuracy of the data, participating institutions were required to review the data before uploading it, and the research team regularly performed data quality control, including consistency checks and verification of missing data.

[0071] Missing data were supplemented and preprocessed to ensure the completeness and accuracy of the data.

[0072] Data preprocessing

[0073] To maximize the completeness and accuracy of the data, we performed the following preprocessing steps on the collected data:

[0074] (1) Identification and supplementation of missing data

[0075] · Analyze the proportion and pattern of missing values ​​to identify potential Missing at Random (MAR).

[0076] Multiple imputation is used to handle randomly missing data. Specific methods include:

[0077] Impute missing values ​​based on multiple relevant variables (e.g., age, BMI, PASI score, etc.).

[0078] Generate multiple imputed data sets to reduce bias caused by missing values.

[0079] (2) Outlier processing

[0080] Analyze outliers for continuous variables using boxplots and scatterplots.

[0081] ·Review and correct the data based on possible input errors, combined with patient medical history records and hospital feedback.

[0082] (3) Variable standardization

[0083] Standardize or normalize continuous variables (such as BMI, age, PASI score) to ensure that the variable scale is consistent.

[0084] Encode categorical variables (such as marital status, smoking history, etc.) as dummy variables to suit modeling needs.

[0085] (4) Data consistency check

[0086] Check whether there are any inconsistencies in the patient's multiple records and ensure the biological plausibility of key variables (such as PASI score, body surface area involvement ratio, etc.).

[0087] (5) Data division

[0088] 5.1. Training set (model development):

[0089] Data source: collected from the China Psoriasis Diagnostic and Treatment Center platform.

[0090] Time range: June 24, 2020 to June 30, 2021.

[0091] Size: Contains clinical data of 17,124 psoriasis patients from 49 hospitals.

[0092] Purpose: Used for model development, including variable screening and construction of logistic regression models.

[0093] 5.2. Validation set (model validation):

[0094] The data set division and specific description of the validation set were not explicitly mentioned in the paper, but the model performance evaluation adopted internal validation (1,000 bootstrap resampling), and the performance and clinical application value of the model were verified by ROC curve, Hosmer-Lemeshow test, DCA, CIC, etc.

[0095] (6) Data security and desensitization

[0096] During data collection and processing, all data will be stripped of patient identity information to ensure privacy protection.

[0097] 2. Variable selection:

[0098] Use the least absolute shrinkage and selection operator (LASSO) regression method to screen key variables. LASSO regression reduces the weight of unimportant variables by introducing regularization terms, thereby achieving variable selection and model simplification.

[0099] Detailed process of variable screening using LASSO regression

[0100] The least absolute shrinkage and selection operator (LASSO) regression method is used to screen key variables. LASSO regression effectively reduces the weight of unimportant variables by introducing regularization terms, thereby achieving variable selection and model simplification. The specific process is as follows:

[0101] (1) The basic principle of LASSO regression:

[0102] LASSO regression is a linear regression method that makes the regression coefficients sparse by adding L_1 regularization terms to the objective function. The objective function formula is:

[0103]

[0104] in:

[0105] yi: target variable (e.g., diagnosis of T2DM).

[0106] xij: The j-th predictor variable for the ith sample.

[0107] βj: regression coefficient, which indicates the influence of the variable on the target.

[0108] λ: Regularization parameter, which controls the compression strength of variable weights.

[0109] (2) Compression and selection of variable weights:

[0110] As the regularization parameter λ increases, the regression coefficient is gradually compressed to zero.

[0111] Unimportant variables are completely removed (i.e., βj = 0), thereby reducing the impact of redundant variables and retaining key variables with strong predictive power for the target event.

[0112] (3) Cross-validation optimization λ:

[0113] A 10-fold cross validation method was used to evaluate the model performance under different λ parameters.

[0114] During cross validation, the optimal lambda value (i.e. the point at which the model has the best predictive performance on the validation dataset) is sought.

[0115] (4) The final selected variables

[0116] LASSO regression screened out predictor variables with non-zero coefficients and entered the subsequent multivariate logistic regression model.

[0117] This study finally screened out 15 variables as candidate variables, including age, BMI, PASI score, history of joint swelling, nail lesions, etc., which were further reduced to 11 independent predictors by multivariate logistic regression.

[0118] Through the LASSO regression method, the variable selection process not only ensures the simplicity and interpretability of the model, but also effectively avoids the problem of multicollinearity, laying a solid foundation for subsequent modeling.

[0119] Finally, select several important predictor variables, which have high predictive value in the model.

[0120] Through LASSO regression screening and multivariate logistic regression analysis, 11 variables with significant predictive value for type 2 diabetes (T2DM) in patients with psoriasis were finally identified. The selection of these variables was entirely based on the results of statistical analysis, including:

[0121] 1. Age

[0122] 2. BMI (Body Mass Index)

[0123] 3. Marital status

[0124] 4. History of drug allergy

[0125] 5. History of tumor

[0126] 6. History of joint swelling and pain

[0127] 7. History of swollen finger or toe joints

[0128] 8. Nail lesions

[0129] 9. Nail concave

[0130] 10.PASI score (Psoriasis Area and Severity Index)

[0131] 11. Psoriasis course

[0132] The above variables are non-zero coefficient variables screened by LASSO regression method, and their statistical significance is further verified by multi-factor logistic regression model. The final selection of these variables is completely data-driven, reflecting their important role in the model prediction performance.

[0133] 3. Model construction:

[0134] The multivariate logistic regression model is suitable for dealing with binary classification problems (such as whether psoriasis patients have diabetes) and predicting the probability of the target event through multiple independent variables. The model is constructed based on the Akaike Information Criterion (AIC) to optimize the balance between the complexity and goodness of fit of the model.

[0135] (1) Model construction method based on AIC

[0136] Variable screening and model optimization:

[0137] Through the stepwise regression method, the AIC criterion is used as the model selection criterion to maximize the explanatory power of the model while reducing redundant variables.

[0138] AIC criterion formula:

[0139] AIC=-2·log(Likelihood)+2·k

[0140] Among them, Likelihood represents the likelihood function of the model, and k is the number of model parameters.

[0141] (2) Specific steps:

[0142] First, the initial variable set is screened through LASSO regression to reduce redundant variables.

[0143] Then, variables were gradually added or removed from the logistic regression model, and the model with the lowest AIC value was selected as the final model.

[0144] (3) Results:

[0145] The optimal model included multiple significant predictors (e.g., age, BMI, PASI score, etc.) and was validated by further performance validation (e.g., AUC, Hosmer-Lemeshow

[0146] Test) to prove its reliability.

[0147] The final logistic regression model formula is as follows:

[0148]

[0149] in:

[0150] P: probability of target event (diabetes occurrence);

[0151] β0: intercept, representing the baseline probability;

[0152] β1,β2,…,βn: regression coefficients of independent variables;

[0153] X1, X2, …, Xn: multiple predictor variables (such as age, BMI, PASI score, etc.).

[0154] The prediction probability formula is:

[0155]

[0156] During the model building process, 10-fold cross validation was used to optimize the model parameters to improve the stability and generalization ability of the model.

[0157] 10-fold cross validation to optimize model parameters

[0158] (1) Purpose: To improve the stability and generalization ability of the model and avoid overfitting problems.

[0159] (2) Process:

[0160] The dataset is randomly divided into 10 equal parts, 9 of which are used for training each time, and the remaining 1 is used for validation;

[0161] Rotate 10 times to ensure that every piece of data is verified;

[0162] Select the optimal model parameters (such as regression coefficients and regularization parameters) by comparing cross-validation performance indicators (such as accuracy and AUC values).

[0163] (3) Effect: Cross-validation can effectively improve the robustness of the model, enabling it to maintain good prediction performance in different sample distributions.

[0164] 4. Model evaluation:

[0165] The area under the receiver operating characteristic curve (AUC) was used to evaluate the q of the model. The higher the AUC value, the stronger the discrimination ability of the model.

[0166] Area under the receiver operating characteristic curve (AUC)

[0167] Definition: AUC (Area Under the Curve) measures the model's ability to distinguish between positive and negative samples, with a value range of 0.5 to 1.0.

[0168] Evaluation method: Draw the ROC curve and evaluate the discriminative ability of the model by calculating the area under the curve.

[0169] Interpretation: The higher the AUC value, the better the model is at distinguishing high-risk from low-risk patients; AUC>0.75 indicates that the model has high discriminative power.

[0170] The Hosmer-Lemeshow test was used to assess the calibration of the model and ensure the reliability of the prediction results.

[0171] Hosmer-Lemeshow test

[0172] (1) Definition

[0173] Used to assess the calibration ability of the model, that is, the consistency between the predicted probability and the actual incidence.

[0174] (2) Methods

[0175] Divide the predicted probability into multiple groups (e.g., 10 groups) and compare the actual incidence rate within each group with the predicted probability.

[0176] A p-value > 0.05 indicates that the model is well calibrated and there is no significant difference between the predicted and actual values.

[0177] (3) Two key lines in the calibration curve

[0178] Ideal line: A 45° diagonal line, indicating that the model predictions are completely consistent.

[0179] Model fitting line: Indicates the actual calibration of the model. The closer it is to the ideal line, the better the calibration performance.

[0180] (4) Internal verification

[0181] The model performance was verified by 1,000 bootstrap resamplings to ensure the stability of the calibration ability in different sample distributions.

[0182] · Decision curve analysis (DCA) and clinical impact curve (CIC) were used to verify the net benefit and clinical utility of the model.

[0183] Decision Curve Analysis (DCA) and Clinical Impact Curve (CIC)

[0184] Decision Curve Analysis (DCA):

[0185] The clinical application value of the model was evaluated by the net benefit (NetBenefit) under different probability thresholds.

[0186] Net benefit formula:

[0187] Net Benefit=(True Positive Rate-False Positive Rate×Weight)

[0188] The weight is determined by the threshold probability, reflecting the clinical tolerance for false positives and true positives.

[0189] High net benefits indicate that the model has higher practicality in different clinical decision-making scenarios.

[0190] Clinical Impact Curve (CIC):

[0191] Show the relationship between the number of high-risk patients predicted by the model and the actual number of high-risk patients, verifying the predictive stability and clinical applicability of the model.

[0192] The CIC curve shows that at certain high thresholds, the predicted value matches the actual value to a high degree, which supports the clinical promotion of the model.

[0193] 5. System Implementation:

[0194] Develop a web-based calculator where users can enter their patients’ clinical data and the system will calculate a diabetes risk score based on the trained model.

[0195] System Implementation

[0196] (1) Risk calculator

[0197] Provide a web-based calculator where users input the patient's clinical data (such as age, BMI, PASI score, etc.) and the system calculates in real time the probability of psoriasis patients having diabetes.

[0198] (2) Visual interface

[0199] The input box is used to fill in the patient's key variable data.

[0200] Display the probability of diabetes in people with psoriasis in clear text or simple visualization.

[0201] (3) Technical Implementation

[0202] The system supports online access and automatically calls the trained model for calculation after inputting data.

[0203] All data is encrypted, stored and processed to ensure patient privacy.

[0204] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and supplements without departing from the method of the present invention. These improvements and supplements should also be regarded as the scope of protection of the present invention.

Claims

1. A prediction model for the comorbidity of psoriasis and diabetes, characterized in that: The prediction model for the comorbidity of psoriasis and diabetes was constructed by the following method: S1 Data collection and preprocessing: S11 collects data on patients with psoriasis; S12 supplements and preprocesses missing data; S2 variable selection: S21 uses the least absolute shrinkage and selection operator regression methods to screen key variables; S22 Select important predictor variables S3 model building: S31 was modeled using a multivariate logistic regression model; During the model building process, S32 optimized the model parameters through 10-fold cross validation; S4 Model Evaluation: S41 used the area under the receiver operating characteristic curve to evaluate the predictive performance of the model; S42 used the Hosmer-Lemeshow test to assess the calibration of the model; S43 The net effectiveness of the model was verified by decision curve analysis and clinical impact curve.

2. The prediction model for the comorbidity of psoriasis and diabetes according to claim 1, characterized in that: S12 supplements and preprocesses missing data, including the following steps: S121 Identification and supplementation of missing data, S122 Outlier processing, S123 Variable standardization, S124. Data consistency check, S125 data partitioning and S126 data security and desensitization.

3. The prediction model for the comorbidity of psoriasis and diabetes according to claim 1, characterized in that: Step S21 uses the least absolute shrinkage and selection operator regression method to screen key variables Specifically including: age, BMI, marital status, history of drug allergy, history of tumor, history of joint swelling and pain, history of swollen finger or toe joints, nail lesions, nail pitting, PASI score and psoriasis course.

4. A web-based prediction score calculator for predicting the comorbidity of psoriasis and diabetes, characterized by: Constructed by: S1 Data collection and preprocessing: S11 collects data on patients with psoriasis; S12 supplements and preprocesses missing data; S2 variable selection: S21 uses the least absolute shrinkage and selection operator regression methods to screen key variables; S22 Select important predictor variables S3 model building: S31 was modeled using a multivariate logistic regression model; During the model building process, S32 optimized the model parameters through 10-fold cross validation; S4 Model Evaluation: S41 used the area under the receiver operating characteristic curve to evaluate the predictive performance of the model; S42 used the Hosmer-Lemeshow test to assess the calibration of the model; S43 validates the net effectiveness of the model through decision curve analysis and clinical impact curve; S5: A web-based calculator where users can input the patient’s clinical data and the system will calculate the diabetes risk score based on the trained model.

5. The web-based prediction score calculator for predicting the comorbidity of psoriasis and diabetes according to claim 4, characterized in that: The system outputs the diabetes risk classification based on the results of model calculation, which is divided into low risk, medium risk and high risk.

6. The web-based prediction score calculator for predicting the comorbidity of psoriasis and diabetes according to claim 4, characterized in that: S12 supplements and preprocesses missing data, including the following steps: S121 Identification and supplementation of missing data, S122 Outlier processing, S123 Variable standardization, S124. Data consistency check, S125 Data partitioning and S126 Data security and desensitization.

7. The web-based prediction score calculator for predicting the comorbidity of psoriasis and diabetes according to claim 4, characterized in that: Step S21 uses the least absolute shrinkage and selection operator regression method to screen key variables, including: age, BMI, marital status, history of drug allergy, history of tumor, history of joint swelling and pain, history of finger or toe joint swelling, nail lesions, nail pitting, PASI score and psoriasis course.

Images