A method for constructing a prediction model for depression comorbid chronic pain
By acquiring and fusing biomarker data of inflammatory pathways and tryptophan metabolism pathways, a predictive model for chronic pain comorbid with depression was constructed, which solves the problem of limited diagnostic efficacy in existing technologies and achieves efficient and accurate disease prediction and diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF ZHENGZHOU UNIV
- Filing Date
- 2026-04-15
- Publication Date
- 2026-05-29
Smart Images
Figure CN122117426A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of bioinformatics and clinical diagnosis, and specifically relates to a method for constructing a predictive model for chronic pain comorbid with depression. Background Technology
[0002] Major Depressive Disorder (MDD) and chronic pain often co-occur, significantly increasing the disease burden on patients and posing a major challenge to treatment. Currently, the diagnosis of chronic pain comorbid with depression relies primarily on patient subjective reports and physician clinical experience, lacking objective, quantitative biological indicators to support the diagnosis. Therefore, there is an urgent clinical need to develop a method that can objectively and accurately identify or predict whether patients with depression will have comorbid chronic pain, enabling early patient classification and precise intervention.
[0003] While existing research has explored the roles of inflammatory responses (such as IL-6 and IL-12) and tryptophan metabolic pathway disorders (such as the kynurenine pathway) in depression and pain, it has largely focused on biomarkers of single biological pathways, resulting in limited diagnostic or predictive efficacy. More importantly, there is a lack of a systematic methodological framework that can effectively integrate biomarker information from different pathways and is specifically designed for constructing high-performance DCP prediction models.
[0004] Existing technologies have failed to provide a clear, reproducible process for addressing this specific clinical problem. Summary of the Invention
[0005] To address the above shortcomings, this invention provides a method for constructing a predictive model for chronic pain comorbid with depression, comprising the following steps:
[0006] S1. Obtain multi-dimensional datasets: Obtain the first biomarker dataset and the second biomarker dataset for the patient group with depression comorbid with chronic pain (DCP), the patient group with depression alone (DNP) as a control, and the healthy control group (HC).
[0007] The first biomarker dataset, derived from inflammatory pathways, contains concentration data of at least one pro-inflammatory factor;
[0008] The second biomarker dataset, derived from the kynurenine metabolic pathway of tryptophan, includes ratio data of at least one metabolite.
[0009] S2. Feature selection: The first and second biomarker datasets were standardized, and statistical methods were used to analyze the expression differences of each biomarker among patients with depression comorbid chronic pain (DCP), patients with depression alone (DNP), and healthy controls (HC). Biomarkers with statistically significant differences were selected as candidate features.
[0010] S3. Model Construction: Using the candidate features selected in step S2 as independent variables, a prediction model is constructed using machine learning algorithms.
[0011] S4. Model Evaluation: The predictive performance of the constructed model is evaluated using methods such as receiver operating characteristic (ROC) analysis, internal cross-validation, or external independent dataset validation. This includes, but is not limited to, calculating the area under the curve (AUC), sensitivity, specificity, and other metrics to confirm the effectiveness and stability of the model.
[0012] Furthermore, the pro-inflammatory factors in the first biomarker dataset include interleukin-6 (IL-6) and interleukin-12 (IL-12).
[0013] Furthermore, the metabolite ratio in the second biomarker dataset is the ratio of kynurenic acid to 3-hydroxykynurenine (KYNA / 3-HK).
[0014] Furthermore, the machine learning algorithm used in step S3 is the Binary Logistic Regression algorithm.
[0015] Furthermore, the validation method used in step S4 is receiver operating characteristic (ROC) curve analysis or Bootstrap internal validation to assess the stability of the model.
[0016] Compared with the prior art, the present invention has the following advantages:
[0017] 1. Based on objective biomarker data, the constructed model can accurately distinguish whether depression is accompanied by chronic pain;
[0018] 2. The method of specifically combining biomarkers of inflammatory pathways and kynurenine metabolic pathways and constructing models can more comprehensively reflect the complex pathophysiological mechanisms of diseases.
[0019] 3. By using a complete and systematic technical process from data acquisition, feature selection, model building to performance verification, it has good repeatability;
[0020] 4. The model constructed based on the method of this invention can be developed into a clinical auxiliary diagnostic tool, providing an objective basis for formulating individualized treatment plans. Attached Figure Description
[0021] Figure 1 This is the ROC curve of the joint prediction model in this invention.
[0022] Figure 2 This is the ROC curve of the joint prediction model in this invention after Bootstrap correction. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] Example
[0025] This embodiment provides a method for constructing a predictive model for chronic pain comorbid with depression. In this embodiment, a total of 119 patients with depression were recruited and, after clinical diagnosis and evaluation, were divided into a depression comorbid with chronic pain group (DCP group, n=59), a simple depression group (DNP group, n=60), and a healthy control group (HC group, n=60).
[0026] Inclusion criteria: (1) meeting the DSM-5 diagnostic criteria for depression; (2) aged 13-65 years; (3) patients in the DCP group also met the diagnostic criteria for chronic pain (pain lasting more than 3 months or intermittent pain for 6 months).
[0027] Exclusion criteria: (1) suffering from other serious mental illnesses or neurological diseases; (2) having serious physical illnesses, autoimmune diseases or acute infections; (3) having recently used anti-inflammatory drugs or immunosuppressants.
[0028] Five ml of blood was collected from the median cubital vein of each subject in the morning on an empty stomach using EDTA anticoagulant tubes. Within two hours of collection, the blood was centrifuged at 3000 rpm for 15 minutes at 4°C. The supernatant (plasma) was carefully aspirated, aliquoted, and frozen at -80°C for later use until subsequent testing to ensure sample quality and the reliability of test results.
[0029] Then proceed with the following steps:
[0030] S1-1, Detection of inflammatory pathway markers
[0031] Commercially available enzyme-linked immunosorbent assay (ELISA) kits for human interleukin-6 (IL-6) and interleukin-12 (IL-12) were used for detection. Three groups of plasma samples and standards were added to pre-coated ELISA plates, incubated, washed, and then biotinylated antibody and horseradish peroxidase-labeled streptavidin were added. After incubation and washing again, the substrate TMB was added for color development, and finally the stop solution was added. The absorbance (OD value) was measured at 450 nm, and the concentrations (pg / ml) of IL-6 and IL-12 in the samples were calculated according to the standard curve.
[0032] S1-2, Detection of biomarkers in the kynurenine metabolic pathway
[0033] Absolute quantification of kynurenic acid (KYNA) and 3-hydroxykynurenic acid (3-HK) in plasma was performed using high-performance liquid chromatography-tandem mass spectrometry (LC-MS / MS). A 50 μL plasma sample was taken, an internal standard was added, and after protein precipitation, the supernatant was injected. Analysis was performed using an Agilent 1290 Infinity II LC system and a Sciex QTRAP 6500+ mass spectrometer. Using established standard working curves, the concentrations of KYNA and 3-HK in the sample were accurately quantified (nmol / L). Based on the detection results, the key ratio KYNA / 3-HK was calculated.
[0034] S2, Data Analysis and Feature Selection
[0035] Data analysis was performed using IBM SPSS Statistics 27.0 and R 4.5.1 statistical software, and graph visualization was performed using Origin 2024 and R 4.5.1. First, a normality test was conducted. Data conforming to a normal distribution were analyzed using ANOVA for differences among the three groups, while data not conforming to a normal distribution were analyzed using the Kruskal-Wallis test. Factors showing differences among the three groups were compared pairwise post-hoc, and the Bonferroni method was used to correct for p-values. A p-value < 0.05 was defined as statistically significant.
[0036] The analysis results are shown in Table 1:
[0037] Table 1
[0038] biomarkers DCP group (n=59) DNP group (n=60) Group HC (n=60) H / F p-value IL-6 (pg / ml) 86.856（117.578） 62.820±33.494 42.577±20.455 47.536 <0.001 IL-12 (pg / ml) 5.428（2.183） 4.614±1.767 4.868（1.910） 14.573 <0.001 KYNA / 3-HK (ratio) 0.384(0.235) 0.565(0.395) 0.691(0.557) 44.694 <0.001
[0039] In the table above, H is the Kruskal-Wallis test statistic, and F is the one-way ANOVA statistic. It should be noted that normally distributed data are expressed as mean ± standard deviation; non-normal data are expressed as median (interquartile range, i.e., the value of P75-P25).
[0040] Compared with the DNP group, the DCP group had a lower plasma IL-6 concentration ratio, which was statistically significant (P<0.001).
[0041] Compared with the DNP group, the DCP group had a lower plasma IL-12 concentration ratio, which was statistically significant (P<0.001).
[0042] Compared with the DNP group, the DCP group had a lower KYNA / 3-HK ratio in plasma, which was statistically significant (P<0.001).
[0043] Based on the statistically significant differences mentioned above, IL-6, IL-12, and KYNA / 3-HK were ultimately selected as candidate features for constructing the predictive model, which laid a solid biological foundation for the model's effectiveness.
[0044] S3, Model Building
[0045] S3-1. A joint prediction model is constructed using the binary logistic regression algorithm. The group classification of the research subjects' DCP and DNP groups is used as the dependent variable, and age and gender are used as control variables. The continuous variable values of the three candidate features (IL-6, IL-12, KYNA / 3-HK) selected in the previous step are used as independent variables and standardized (while controlling for age and gender). The backward likelihood ratio (backward LR) method is used to screen variables, with the elimination criterion being P > 0.05. All independent variables are then included in the model for fitting.
[0046] After model training, the final logistic regression equation (joint prediction model) is as follows:
[0047] Logit(P)=β0+1.346[IL-6]+0.710[IL-12]-1.294[KYNA / 3-HK]
[0048] Where P is the probability of predicting an individual as a DCP patient, and β0 is a model constant.
[0049] S3-2. Setting up a comparison model: Inflammation prediction model
[0050] For comparison, the same logistic regression algorithm was used, with only the inflammatory pathway markers IL-6 and IL-12 as independent variables, to construct an inflammation prediction model;
[0051] S3-3. Setting up a comparison model: Tryptophan metabolism prediction model
[0052] For comparison, the same logistic regression algorithm was used, with only the kynurenine metabolic pathway marker KYNA / 3-HK as the independent variable, to construct a tryptophan metabolism prediction model;
[0053] S4-1. Evaluate the internal performance of the model.
[0054] like Figure 1 As shown, to evaluate the predictive efficacy of the constructed joint prediction model, the receiver operating characteristic (ROC) curve of the model was plotted, and the results are shown in Table 2.
[0055] Table 2
[0056] AUC Confidence interval Sensitivity Specificity Yoden Index Inflammation model 0.755 0.668-0.841 0.638 0.746 0.384 Tryptophan model 0.739 0.650-0.829 0.661 0.696 0.357 Joint model 0.871 0.807-0.936 0.759 0.855 0.613
[0057] The ROC analysis results of this joint prediction model are as follows: the area under the curve (AUC) is 0.871, its 95% confidence interval (CI) is 0.807-0.936, and the P value is <0.001.
[0058] Diagnostic efficacy: Based on the principle of maximizing the Youden index, the optimal diagnostic cutoff point for the model was determined. At this cutoff point, the model's diagnostic sensitivity was 75.9%, and its specificity was 85.5%.
[0059] The AUC value is much greater than 0.75, indicating that the model has good or even excellent discrimination ability.
[0060] The inflammation prediction model was calculated to have an AUC of 0.755 (95% CI: 0.668-0.841).
[0061] The AUC of the tryptophan metabolism prediction model was calculated to be 0.739 (95% CI: 0.650-0.829).
[0062] Therefore, the diagnostic efficacy of the combined predictive model (AUC=0.871) is significantly better than that of the model using only inflammatory factors (AUC=0.755) and the model using only tryptophan metabolism ratio (AUC=0.739). This result strongly demonstrates that combining biomarkers from two different dimensions—inflammatory pathways and kynurenine metabolism pathways—can produce a synergistic effect, and its predictive ability is not simply a superposition of information from individual pathways.
[0063] S4-2. Verify the stability of the model.
[0064] To further test the stability and reliability of the model and avoid overfitting, a Bootstrap internal validation method was repeated 1000 times. Figure 2 As shown in Table 3, the receiver operating characteristic (ROC) curves after Bootstrap correction for the joint prediction model are as follows:
[0065] Table 3
[0066] AUC Confidence interval Sensitivity Specificity Yoden Index Inflammation model 0.701 0.481-0.897 0.598 0.739 0.336 Tryptophan model 0.692 0.483-0.873 0.647 0.667 0.314 Joint model 0.823 0.653-0.976 0.722 0.811 0.533
[0067] Validation results: After 1000 resampling validations, the corrected AUC value was calculated to be 0.823 (95% CI: 0.653-0.976).
[0068] After 1000 Bootstrap resampling corrections, the corrected AUC value was 0.823. The corrected AUC value was only slightly lower than the original value, indicating that the joint prediction model constructed in this invention is robust, has a low risk of overfitting, and has good stability and generalization potential in clinical applications.
[0069] Thus, from rigorous sample collection and processing to accurate biomarker detection, scientific statistical screening and model building, and finally through comprehensive performance evaluation and validation, a predictive model capable of efficiently distinguishing between DCP and DNP patients was successfully constructed.
[0070] It should be noted that the appendix Figure 1-2 In the diagram, green represents the inflammation prediction model, orange represents the tryptophan metabolism prediction model, and purple represents the combined prediction model.
[0071] It should be noted that the structure described in this invention can be implemented in many different forms and is not limited to the embodiments described. Any equivalent transformations made by those skilled in the art based on the description and drawings of this invention, or direct or indirect applications in other related technical fields, such as the loading and unloading of other items, are included within the protection scope of this invention.
Claims
1. A method for constructing a predictive model for chronic pain comorbid with depression, characterized in that, Includes the following steps: S1. Obtain multi-dimensional datasets: For the patient group with chronic pain comorbid with depression, the patient group with depression alone, and the healthy control group, obtain the first biomarker dataset from the inflammatory pathway and the second biomarker dataset from the tryptophan-kynurenine metabolic pathway, respectively. S2. Feature selection: Statistical methods were used to screen for biomarkers that showed significant differences among patients with depression and chronic pain, patients with depression alone, and healthy controls as candidate features. S3. Model Construction: Using the candidate features selected in step S2 as independent variables, a prediction model is constructed using machine learning algorithms. S4. Model Evaluation: Evaluate the predictive performance of the constructed model using at least one validation method.
2. The method for constructing a predictive model for chronic pain comorbid with depression as described in claim 1, characterized in that: The first biomarker dataset contains concentration data for interleukin-6 and interleukin-12.
3. The method for constructing a predictive model for chronic pain comorbid with depression as described in claim 1, characterized in that: The second biomarker dataset contains data on the ratio of kynurenic acid to 3-hydroxykynurenine.
4. The method for constructing a predictive model for chronic pain comorbid with depression as described in claim 1, characterized in that: The machine learning algorithm used in step S3 is the binary logistic regression algorithm.
5. The method for constructing a predictive model for chronic pain comorbid with depression as described in claim 1, characterized in that: The verification method used in step S4 is either receiver operating characteristic (ROC) curve analysis or Bootstrap internal verification.