Webpage calculator construction and evaluation method for anticipated sadness occurrence probability of family caregiver of cancer patient

Abstract

The invention relates to the technical field of psychological prediction, and particularly discloses a webpage calculator construction and evaluation method for anticipated sadness occurrence probability of a family caregiver of a cancer patient, and the method comprises the steps: collecting the related data of the family caregiver and the cancer patient, determining the diagnosis standard of anticipated sadness through potential profile analysis, and evaluating the anticipated sadness occurrence probability of the family caregiver. And screening key predictive variables by using LASSO-logistic regression, constructing a predictive model, and developing an online webpage calculator based on a shiny technology. According to the method, the expected sadness risk of the family caregiver can be rapidly evaluated, the method has good distinction degree, calibration degree and clinical practicability, and a scientific basis is provided for medical staff to identify high-risk family caregivers and provide targeted support.

Description

Technical Field

[0001] The present invention relates to the technical field of psychological prediction, and particularly to a method for constructing and evaluating a web calculator for predicting the probability of anticipatory grief among family caregivers of cancer patients. Background Art

[0002] Cancer, as a global health problem, not only has a serious impact on patients, but also brings significant psychological and physiological burdens to family caregivers. Anticipatory grief is a common psychological reaction among family caregivers, manifested as emotions such as sadness, anger, guilt, or anxiety, which may seriously affect their mental health and caregiving ability when severe. Currently, the assessment of anticipatory grief mainly relies on tools such as the Chinese version of the Anticipatory Grief Scale, but these methods are time-consuming, complex to operate, lack clear diagnostic criteria and quantitative prediction capabilities, and are difficult to meet the needs of quickly identifying high-risk family caregivers in clinical practice. Therefore, it is of great significance to develop an efficient and accurate tool for predicting the risk of anticipatory grief. Summary of the Invention

[0003] To solve the above problems, the purpose of the present invention is to provide a method for constructing and evaluating a web calculator for predicting the probability of anticipatory grief among family caregivers of cancer patients, which can simply and accurately predict the probability of anticipatory grief among family caregivers of cancer patients.

[0004] The present invention provides a method for constructing and evaluating a web calculator for predicting the probability of anticipatory grief among family caregivers of cancer patients, and the method includes: S101: Collect data of family caregivers of cancer patients; S102: Preprocess the data; S103: Use latent profile analysis to identify risk categories of anticipatory grief among family caregivers and determine diagnostic criteria; S104: Use LASSO-logistic regression to screen key predictive variables and construct a prediction model for anticipatory grief; S105: Develop an online web calculator based on the prediction model to support users to input data in real time and obtain the probability of anticipatory grief; S106: Evaluate the prediction performance of the web calculator.

[0005] Among them, step S101 includes: collecting general demographic characteristics of family caregivers and cancer patients and the Chinese version of the Anticipatory Grief Scale; the scores of family caregivers measured.

[0006] Among them, step S102 includes: identifying and processing missing values in the data; deleting predictive variables with missing values exceeding a preset percentage, and for predictive variables with missing values less than the preset percentage, using multiple imputation methods to fill in the missing data.

[0007] Among them, step S103 includes: classifying the scores of the seven dimensions of the Chinese version of the Anticipatory Grief Scale using latent profile analysis, identifying the anticipatory grief risk categories based on indicators such as AIC, BIC, aBIC, entropy value, BLRT, and LMR tests, and determining the diagnostic criteria.

[0008] Among them, step S104 includes: using the LASSO regression method to shrink the coefficients and screening out the latent variables that make important contributions to the prediction results. A prediction model is constructed through multiple logistic regression: P = e x / (1 + e x ) x = -0.5325 - 0.9488×education level + 1.1497×monthly income + 0.8417×monthly income + 0.4051×monthly income - 1.8115×physical condition - 1.6797×physical condition + 1.4429×care duration + 0.6012×care duration + 0.9058×cancer type + 0.7472×patient's job - 1.0816×diagnosis time - 0.9802×diagnosis time. The reasons why the four parameters of monthly income item, physical condition, care duration, and diagnosis time appear multiple times in the model are that these variables are statistically significant and are therefore retained in the equation; if there is no statistical significance, they will not be replicated and enter the equation.

[0009] P is the predicted probability, and e is the base of the natural logarithm; for education level, those with a university degree (including junior college) or above are assigned 1, otherwise 0; for monthly income, those with 2000 - 4000 yuan are assigned 1, those with 4000 - 6000 yuan are assigned 2, those with > 6000 are assigned 3, otherwise 0; for physical condition, those in good condition are assigned 1, those in very good condition are assigned 2, otherwise 0; for care duration, those with 7 - 12 months are assigned 1, those with > 1 year are assigned 2, otherwise 0; for cancer type, those with lung cancer are assigned 1, otherwise 0; for the patient's job, those without a job are assigned 1, otherwise 0; for diagnosis time, those with 6 - 12 months are assigned 1, those with > 12 months are assigned 2, otherwise 0.

[0010] Among them, step S105 includes: extracting the regression coefficients of each predictive variable in the multiple Logistics regression model, and calculating the change amount of the predictive variable by multiplying the regression coefficient by the range of the variable in the data; selecting the predictive variable with the largest change amount as the benchmark variable and assigning it 100 points; then, calculating the scores of other predictive variables, and the formula is: score = 100×(change amount of other variables / change amount of the benchmark variable); adding up the scores of all predictive variables to get the total score, and calculating the final predicted probability through the conversion formula.

[0011] Among them, step S106 includes: drawing a receiver operating characteristic curve (ROC) and calculating the area under the curve (AUC) to evaluate the prediction ability of the web calculator; using the Bootstrap method to perform repeated sampling with replacement for a preset number of times, drawing an internal calibration curve and performing a goodness-of-fit test; drawing a decision curve analysis (DCA) curve, and evaluating the actual utility of the model in clinical practice by analyzing the net benefit at different probability thresholds.

[0012] The present invention develops an efficient, accurate web calculator based on machine learning algorithms and computer programming techniques, which can accurately identify the anticipatory grief of family caregivers of cancer patients as early as possible. The web calculator can visualize complex model information. Medical staff can input the basic information of the caregiver into the model to obtain the probability of anticipatory grief of the family caregiver of the cancer patient, providing a simple, intuitive, and rapid method for medical staff to diagnose the anticipatory grief of family caregivers of cancer patients. The specific advantages are as follows: (1) High efficiency: Users can quickly input variables and obtain prediction results in real time, significantly saving the evaluation time; (2) Accuracy: The model has good discrimination and calibration, ensuring the reliability of the prediction results; (3) Practicality: It helps medical staff to identify high-risk family caregivers in a timely manner, provide personalized interventions, and improve the quality of care. Description of the Drawings

[0013] Figure 1 is a flowchart of a method for constructing and evaluating a web calculator for the probability of anticipatory grief of family caregivers of cancer patients provided by an embodiment of the present invention; Figure 2 is a schematic diagram of determining the risk categories of anticipatory grief of family caregivers by latent profile analysis provided by an embodiment of the present invention; Figure 3 is a schematic diagram of screening variables by LASSO regression analysis provided by an embodiment of the present invention; Figure 4 is a schematic diagram of constructing a risk probability web calculator using a multiple logistic regression model provided by an embodiment of the present invention; Figure 5 is a schematic diagram of an ROC curve provided by an embodiment of the present invention; Figure 6 is a schematic diagram of a self-sampling (Bootstrap) internal calibration curve provided by an embodiment of the present invention; Figure 7 is a schematic diagram of a decision curve analysis (DCA) curve provided by an embodiment of the present invention. Detailed Embodiments

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0015] Figure 1 It is a flowchart of a method for constructing and evaluating a web calculator for the probability of anticipatory grief among family caregivers of cancer patients provided by an embodiment of the present invention. The method includes the following steps: S101. Collect data of family caregivers of cancer patients In this embodiment, the collected data is from family caregivers (hereinafter referred to as FCs) of lung cancer and breast cancer patients who are hospitalized and receiving treatment in a certain hospital. The specific data includes two parts: 1. General demographic characteristics of family caregivers and cancer patients: gender, age, educational level, monthly income, physical condition, care duration, relationship with the patient, etc. of FCs; gender, age, cancer type, work situation, diagnosis time, and TNM stage of cancer patients.

[0016] 2. Anticipatory grief score data: Measure the anticipatory grief level of FCs through the Chinese version of the Anticipatory Grief Scale (AGS). The Anticipatory Grief Scale (AGS) was developed by American scholars Theute et al. in 1991 and was initially used to evaluate the anticipatory grief of caregivers of dementia patients. Later, it was also used for caregivers in hospice care, cancer patients, and critically ill patients. The scale has 27 items, covering 7 dimensions, namely grief, sense of loss, anger, irritability, guilt, anxiety, and ability to complete tasks; there are a total of 27 items, and the score range is from 27 to 135 points. The higher the score, the more severe the anticipatory grief. In this embodiment, the internal consistency Cronbach's α coefficient of the AGS scale is 0.95.

[0017] In this embodiment, a total of 670 cases of FCs data are collected, and finally 642 cases of valid data are screened out, and the data validity rate is 95.8%.

[0018] S102. Preprocess the data Preprocess the collected FCs data to ensure the integrity and consistency of the data. The specific steps include: 1. Missing value identification and handling: Conduct missing value analysis on the data, and eliminate samples or variables with a missing value rate exceeding 20%. For variables with a missing value rate less than 20%, use the multiple imputation method to fill in the missing data. In this embodiment, the data missing rate is 1.09%, and the filling is completed through multiple imputation.

[0019] 2. Data standardization and distribution analysis: Conduct skewness and kurtosis analysis on continuous variables (such as age, score, etc.) to verify whether the data conforms to the normal distribution. In this embodiment, the skewness is 0.596 and the kurtosis is 0.099, meeting the requirements of the normal distribution. Therefore, the measurement data is described by the mean ± standard deviation.

[0020] S103. Use latent profile analysis to identify the risk categories of anticipatory grief among family caregivers and determine the diagnostic criteria In this step, based on the scores of the 7 dimensions of the AGS scale, latent profile analysis is performed on 642 FCs to identify the risk categories of anticipatory grief. The specific process is as follows: 1. Model fitting: Start from a 1-class model and gradually increase the number of classes to 5 classes, and calculate the fitting parameters of each model, including AIC, BIC, aBIC, entropy value, LMR, and BLRT.

[0021] 2. Selection of the best model: Considering the fitting indicators and the practical significance of classification comprehensively, select the 3-class model as the best model (AIC = 10174.121, BIC = 10308.058, aBIC = 10212.810, Entropy = 0.906, LMR P < 0.001, BLRT P < 0.001). This model classifies FCs into three categories: low anticipatory grief type (54.5%), medium anticipatory grief type (30.5%), and high anticipatory grief type (15.0%).

[0022] 3. Determination of diagnostic criteria: Define the low anticipatory grief type as "non-case", and the medium and high anticipatory grief types as "cases". Through ROC curve analysis, the best cut-off value of the AGS scale is determined to be 70 points (sensitivity 1.000, specificity 0.994, Youden index 0.994), which is used as the diagnostic criteria for anticipatory grief.

[0023] Figure 2 It is a schematic diagram for determining the risk categories by latent profile analysis, showing the distribution characteristics of the standardized means of the three types of FCs in each dimension.

[0024] S104. Use LASSO-logistic regression to screen key predictive variables and construct an anticipatory grief prediction model In this step, the LASSO-logistic regression method is used to screen key predictive variables and construct a prediction model. The specific steps are as follows: 1. Variable screening: Initially, 49 independent variables were included, including the educational level, monthly income, physical condition, care duration of FCs, and the cancer type, work situation, diagnosis time, etc. of the patients. LASSO regression was used to compress the variable coefficients and exclude redundant variables. Finally, 7 key predictive variables were screened out: the educational level, monthly income, physical condition, care duration of FCs, the cancer type, work situation, and diagnosis time of the patients.

[0025] Figure 3 Figure 6 shows the schematic diagram of variable screening for LASSO regression, presenting the variable selection path and cross-validation results.

[0026] 2. Model construction: The 7 variables screened out were incorporated into a multiple Logistic regression model to calculate the regression coefficients and odds ratios (OR) of each variable. The model formula is as follows: P = e x / (1 + e x ) x = -0.5325 - 0.9488×Educational level + 1.1497×Monthly income + 0.8417×Monthly income + 0.4051×Monthly income - 1.8115×Physical condition - 1.6797×Physical condition + 1.4429×Care duration + 0.6012×Care duration + 0.9058×Cancer type + 0.7472×Patient's work - 1.0816×Diagnosis time - 0.9802×Diagnosis time; P is the predicted probability, and e is the base of the natural logarithm; for the educational level, those with a university degree (including junior college) or above are assigned 1, otherwise 0; for the monthly income, those with 2000 - 4000 yuan are assigned 1, 4000 - 6000 yuan are assigned 2, >6000 are assigned 3, otherwise 0; for the physical condition, those in good condition are assigned 1, those in very good condition are assigned 2, otherwise 0; for the care duration, those with 7 - 12 months are assigned 1, >1 year are assigned 2, otherwise 0; for the cancer type, those with lung cancer are assigned 1, otherwise 0; for the patient's work, those without a job are assigned 1, otherwise 0; for the diagnosis time, those with 6 - 12 months are assigned 1, >12 months are assigned 2, otherwise 0.

[0027] S105. Develop an online web calculator based on the prediction model Based on the above multiple Logistic regression model, an online web calculator was developed using Shiny technology. The specific construction process is as follows: 1. Score calculation: Extract the regression coefficients of each predictor variable in the model, calculate the change in each variable by multiplying the regression coefficient by the range of the variable; select the variable with the largest change (for example, the care duration is 7 - 12 months, and the change is 1.4429) as the baseline variable and assign it 100 points; the score calculation formula for other variables is: Score = 100×(change in this variable / change in baseline variable); add up the scores of all variables to get the total score (range 0 - 400 points).

[0028] 2. Probability conversion: Calculate the probability of anticipatory grief occurrence through the conversion relationship between the total score and the predicted probability.

[0029] Figure 4 The figure shows the schematic diagram of constructing a web calculator using a multiple Logistic regression model, demonstrating the score calculation and probability conversion processes.

[0030] S106. Performance evaluation of the web calculator To verify the prediction performance of the web calculator, it is evaluated from three aspects: discrimination, calibration, and clinical utility: 1. Discrimination: Plot the ROC curve and calculate the AUC value. The AUC of the internal validation set is 0.671 (sensitivity 0.670, specificity 0.608), indicating that the model has a certain discriminatory ability. The schematic diagram of the ROC curve is as Figure 5 shown.

[0031] 2. Calibration: Use the Bootstrap technique for 1000 times of repeated sampling with replacement and plot the internal calibration curve. The P value of the Hosmer - Lemeshow test for the training set is 0.095, and for the internal validation set is 0.801, both indicating good model fitting.

[0032] Figure 6 The figure shows the schematic diagram of the internal calibration curve, showing a high consistency between the predicted value and the actual value.

[0033] 3. Clinical utility: Plot the DCA curve to analyze the net benefit at different probability thresholds. The net benefit of the training set is significantly higher than the baseline within the threshold range of 18% - 80%, and that of the internal validation set is within the range of 34% - 62%, indicating that the model has clinical utility value. The schematic diagram of the DCA curve is as Figure 7 shown.

[0034] Taking the FCs of a lung cancer patient as an example, the information is as follows: The education level is high school, the monthly income is 4000 - 6000 yuan, the physical condition is average, the care duration is 7 - 12 months, the patient has retired, and the diagnosis time is more than 12 months.

[0035] After inputting into the web calculator, the total score is 266 points and the predicted probability is 80%, indicating that the risk of anticipatory grief occurring in these FCs is relatively high and timely intervention is required.

[0036] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A web calculator for the expected probability of sadness in family caregivers of cancer patients and a method for evaluating the probability of sadness in family caregivers, characterized in that: include: S101: Collect data from family caregivers of cancer patients; S102: preprocessing the data; S103: Use latent profile analysis to identify risk categories for anticipatory grief in family caregivers and determine diagnostic criteria; S104: Use LASSO-logistic regression to screen key predictor variables and construct an anticipatory sadness prediction model; S105: Develop an online web calculator based on the prediction model to support users to input data in real time and obtain the expected probability of sadness; S106: Web calculator prediction performance evaluation.

2. The method according to claim 1, characterized in that Step S101 includes: collecting general demographic characteristics of family caregivers and cancer patients and family caregiver scores measured by the Chinese version of the Anticipatory Grief Scale.

3. The method according to claim 1, characterized in that Step S102 includes: identifying and processing missing values ​​in the data; deleting predictor variables containing missing values ​​exceeding a preset percentage, and for predictor variables containing missing values ​​less than a preset percentage, using a multiple interpolation method to fill in the missing data.

4. The method according to claim 1, characterized in that: Step S103 includes: using latent profile analysis to classify the scores of the seven dimensions of sadness, sense of loss, anger, irritability, guilt, anxiety, and ability to complete tasks in the Chinese version of the Anticipatory Sadness Scale, identifying the anticipatory sadness risk category based on AIC, BIC, aBIC, entropy, BLRT and LMR test indicators, and determining the diagnostic criteria.

5. The method according to claim 1, characterized in that Step S104 includes: using the LASSO regression method to shrink the coefficients, screening out potential variables that have important contributions to the prediction results, and constructing a prediction model through multivariate logistic regression: P=e x / (1+e x ) x=-0.5325-0.9488×education level+1.1497×monthly income+0.8417×monthly income+0.4051×monthly income-1.8115×physical condition-1.6797×physical condition+1.4429×care time+0.6012×care time+0.9058×cancer type+0.7472×patient's work-1.0816×diagnosis time-0.9802×diagnosis time; P is the predicted probability, e is the base of the natural logarithm; educational level: college degree and above is assigned a value of 1, otherwise it is assigned a value of 0; monthly income: 2000-4000 yuan is assigned a value of 1, 4000-6000 yuan is assigned a value of 2, and >6000 yuan is assigned a value of 3, otherwise it is assigned a value of 0; physical condition: good is assigned a value of 1, very good is assigned a value of 2, otherwise it is assigned a value of 0; length of care: 7-12 months is assigned a value of 1, >1 year is assigned a value of 2, otherwise it is assigned a value of 0; cancer type: lung cancer is assigned a value of 1, otherwise it is assigned a value of 0; patient employment: no job is assigned a value of 1, otherwise it is assigned a value of 0; time of diagnosis: 6-12 months is assigned a value of 1, >12 months is assigned a value of 2, otherwise it is assigned a value of 0.

6. The method according to claim 1, characterized in that Step S105 includes: extracting the regression coefficient of each predictor variable in the multivariate Logistics regression model, and calculating the change of the predictor variable by multiplying the regression coefficient with the range of the variable in the data; selecting the predictor variable with the largest change as the baseline variable and assigning it 100 points; then, calculating the scores of other predictor variables, the formula is: score = 100 × (change of other variables / change of baseline variable); adding the scores of all predictor variables to obtain the total score, and calculating the final prediction probability through the conversion formula.

7. The method according to claim 1, characterized in that Step S106 includes: drawing a receiver operating characteristic curve (ROC) and calculating the area under the curve (AUC) to evaluate the predictive ability of the web calculator; using the Bootstrap method to perform a preset number of repeated samplings with replacement, drawing an internal calibration curve and performing a goodness of fit test; drawing a clinical decision curve (DCA) to evaluate the actual utility of the model in clinical practice by analyzing the net benefits under different probability thresholds.

Images