Method for predicting lifetime of non-tumor peaceful patient based on clinical data

Abstract

The invention relates to a method and a model for predicting the lifetime of a non-tumor peaceful patient based on clinical data. The method comprises the following steps of: collecting and sorting previous hospitalization clinical data of a non-tumor peaceful treatment death patient, and carrying out standardized pretreatment; carrying out attribute selection and reduction on the high-dimensional data from two perspectives of correlation and redundancy by using a maximum information coefficient (MIC); the method comprises the following steps: respectively carrying out attribute weighting by applying a Pearson correlation coefficient and a Spearman rank correlation coefficient; a weighted naive Bayes classification model is used for modeling; according to the method, clinical factors related to the lifetime of the non-tumor Anning treatment patient can be objectively analyzed and predicted, and guidance is provided for Anning clinical admission standard formulation and early and later treatment plans. The model can be realized in the form of a doctor workstation and the like, and the medical work efficiency and accuracy can be greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method and model for predicting the survival period of non-tumor palliative care patients based on clinical big data. Background Art

[0002] In clinical non-oncology palliative care, efficient and accurate assessment of patient survival plays a crucial role in establishing patient admission criteria (currently, no specific criteria exist for clinical admission to non-oncology palliative care) and in pre- and post-treatment planning. This requires two key criteria: immediate and accurate assessment and the ability to retrospectively identify relevant factors or develop assessment criteria. Historically, clinical survival prediction has been categorized into two main scenarios: non-acute survival assessment when chronic diseases and complications reach a certain level of severity; and survival assessment when acute organ failure develops on this basis. The former includes the FRAIL scale, the Clinical Frailty Scale (CFS), and the Charlson Comorbidity Index, while the latter includes the Acute Physiology and Chronic Health Evaluation (APACHE II, III, and IV), the Sepsis-Associated Organ Failure Assessment (SOFA), the Simplified Acute Physiology Score (SAPS II), and the Early Warning Score (MEWS). Other approaches include using single or multiple independent mortality-related factors to discriminate based on the failure of a specific organ or system; and systematic prediction based on trained algorithms such as long-short-term memory neural networks. The above evaluation methods have problems such as limited considerations, complicated steps, and black box nature, which result in deficiencies in objectivity, accuracy, and clinical practicality.

[0003] Clinical big data refers to all types of data that are directly or indirectly related to clinical practice and can be collected clinically. Since most non-cancer palliative care patients are in states of dementia, unconsciousness, etc. and are unable to correctly express their symptoms, research data is mainly based on objective data in electronic medical records.

[0004] Clinically, there is significant uncertainty in assessing the survival of non-oncology palliative care patients. On the one hand, death is inevitable due to multiple organ failure in the elderly, often caused by factors such as aspiration and infection, while the timing of these events is contingent. On the other hand, even in the elderly and frail, the body's powerful, individualized self-compensatory and repair capabilities play a key role. Clinical observations have shown that survival is related to multiple factors, including direct factors such as diet, biochemical indicators, and treatment measures, and indirect factors such as pensions and family relationships. Some seemingly unimportant factors often play a key role. For example, a patient receiving enteral nutrition support through a nasogastric tube often survives far longer than expected. Elderly patients with femoral neck fractures who have three or more children often choose conservative treatment. Statistics show that the mortality rate within one year due to complications of bed rest reaches 50%. Furthermore, the risk of death increases significantly around the winter solstice and Qingming Festival. Therefore, multi-feature assessments based on clinical big data guarantee accuracy and facilitate the development of admission criteria based on clinical retrospective analysis of relevant factors.

[0005] Survival-related factors are complex, high-dimensional data. Due to the advanced age, disability, dementia, and multi-organ damage of patients, who often experience multi-organ failure in the terminal stage, these factors exhibit the following characteristics: First, complex linear and nonlinear relationships exist between survival and these factors; second, multicollinear correlations of varying strength exist between these factors; third, the importance, or weight, of each factor in assessing survival varies, and actual data for each factor are often incomplete. The vast majority of these patients experience repeated hospitalizations, with frequency increasing significantly near the end of their lives. Hospital databases contain a large amount of objectively extractable, constantly changing data over time. These data often possess three characteristics: sensitivity, e.g., the proportion of patients who die when the data exhibits an abnormality or change; specificity, e.g., the proportion of patients who die when the data exhibit an abnormality or change; and spatiotemporal specificity. The constant changes in these factors over the course of the disease indicate varying degrees of disease progression. The first two reflect the correlation between the cross-sectional temporal relationship of each factor and survival at a specific moment in time, while the latter reflects the correlation between the longitudinal spatiotemporal changes of each factor and survival.

[0006] Based on the above data characteristics, algorithm design needs to consider the following: 1. In the medical field, interpretability is more important than predictive accuracy; 2. The method should be simple, with low-cost and high-efficiency computation; 3. It should simultaneously address linear and nonlinear issues of related factors, data dimensionality reduction, redundancy removal, multicollinearity, weighting issues, and the handling of skewed distributions that are unsuitable for statistical analysis; 4. Data incompleteness is a common issue. Therefore, a comprehensive evaluation method based on fuzzy mathematics theory is considered, as its robustness is suitable for handling these issues.

[0007] With the widespread adoption of information technology in hospitals, efficient data integration is possible, and the establishment of a continuously dynamically optimized system can significantly improve the efficiency of medical staff. For example, after a patient is admitted to the hospital and undergoes an examination and medical history, the doctor's workstation intelligently displays survival and short-term mortality risks, analyzing relevant risk factors, and its accuracy continues to improve. Summary of the Invention

[0008] In response to the above technical problems and needs, the purpose of the present invention is to provide a method and model for predicting the survival of non-tumor palliative care patients based on clinical big data, which is of great significance to clinical non-tumor palliative care work.

[0009] To achieve the above-mentioned purpose, the present invention provides a technical solution comprising the following steps: Step 1, establishing an evaluation index system, extracting one or more hospitalization data within one year of deceased patients who meet the non-cancer palliative care admission conditions from the hospital medical record system; Step 2, performing standardized preprocessing on the extracted data; Step 3, feature screening; Step 4, weighting relevant factors; Step 5, constructing a weighted naive Bayes classification model; and outputting the category that maximizes the posterior probability as the classification result.

[0010] Preferably, a method for predicting the survival of non-cancer palliative care patients based on clinical data is provided, wherein the evaluation index system is established by obtaining one or more hospitalization data within one year of deceased patients who meet the non-cancer palliative care admission requirements from the hospital electronic medical record system, specifically including: basic information: age, gender, height, weight; vital signs: body temperature, heart rate, blood pressure, respiratory rate; blood and body fluid tests: routine blood, urine, and stool tests, coagulation function; biochemical tests: liver and kidney function, electrolytes, protein, blood lipids, cardiac markers, blood gas; nuclear medicine tests: thyroid function, adrenal function; functional tests: CT, chest X-ray, B-ultrasound, MRI; medical history: diagnosis, chief complaint, main medications; nursing: nasogastric feeding and other nursing plans; comprehensive geriatric and other assessments: nutritional status, swallowing function, Padua, Barthel, mental faculties, sleepiness, delirium CAM, dementia CDR scores, etc.

[0011] Preferably, a method for predicting survival of non-cancer palliative care patients based on clinical data, wherein the feature screening uses the maximum information coefficient (MIC) to perform feature redundancy deletion and simplification;

[0012] MIC calculation formula:

[0013] Where MIC(X; Y) is the MIC value between variables; B is the number of grids; I(X; Y) is the mutual information value, that is,

[0014] Where I(X;Y) is the mutual information between variables X and Y, and P(x,y) is the joint probability density between the variables.

[0015] Preferably, a method for predicting the survival of non-cancer palliative care patients based on clinical data, wherein the weighted related factors are weighted by using the survival period as the dependent variable and other factors as independent variables to perform linear regression analysis on all selected data using SPSS26.0 statistical software;

[0016] ① Select factors with significance less than 0.05 for Pearson correlation analysis. The correlation coefficient calculation formula is:

[0017] P xy is the correlation coefficient between survival period X and each factor Y; Represents an m-dimensional vector x j and y j The mean of

[0018] ② Factors with a significance greater than 0.05 were analyzed using Spearman rank correlation analysis; the Spearman rank correlation calculation formula is: where γ XY is the rank correlation coefficient between survival X and each factor Y, p i and q i The variables x i ,y i rank, n is the number of samples in the data;

[0019] ③P xy and γ XY Converted into weights of each factor

[0020] or W ij is the weight of the i or j factor.

[0021] Preferably, a method for predicting the survival of non-tumor palliative care patients based on clinical data, wherein the weighted naive Bayes classification model is constructed, and the posterior probability h nb (x) Calculation formula:

[0022] Prior probability P(C i ) Its maximum likelihood estimate can be calculated based on the existing data samples

[0023] Among them, |D| represents the total number of samples, |D ci | indicates that the category value is C i The number of samples;

[0024] Conditional probability P(A j |C i ) If 0 occurs, Laplace calibration is introduced and the corrected conditional probability is estimated as:

[0025] Among them, |D ci,aj | indicates that the category is c i And attribute A j The value is a j The total number of samples, N j Represents attribute A j The number of possible values;

[0026] The output makes the posterior probability h nb (x) The category (survival period) that obtains the maximum value is taken as the classification result.

[0027] BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 It is a principle block diagram of the present invention; DETAILED DESCRIPTION

[0028] The following is combined with Figure 1 The present invention is described; it should be understood that it is only used to illustrate the present invention rather than to limit the scope of protection of the present invention; those skilled in the art can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

[0029] The present invention provides a method for predicting the survival of non-tumor palliative care patients based on clinical data, such as Figure 1 The method comprises the following steps:

[0030] Step S1: Establish an evaluation index system, specifically including:

[0031] 1. Case selection:

[0032] Death records were collected after the hospital upgraded its five-level medical record management system in 2017. The screening criteria were as follows: ① patients with a clear tumor diagnosis, undiagnosed obvious organ mass, and significantly elevated levels of two or more cancer markers were excluded; ② patients were over 60 years old; ③ patients met the admission criteria for non-cancer palliative care before death, i.e., patients with multiple organ damage or failure and contraindications to active treatment measures such as surgery, hemodialysis, interventional therapy, and pacemakers; and ④ patients were included in the data of one or more hospitalizations within 1 year from the time of surgery, interventional therapy, hemodialysis, pacemaker, or no active treatment for 2 months after the date of death. Patients whose treatment plan was changed during the same hospitalization due to significant changes in their condition were recorded as new records. This created a sample database indexed by the number of days from death (survival) 0, 1, 2, 3…n (n ≤ 360 days) (using a 3-day step size of 0, 3, 6,…n when necessary).

[0033] 2. Data collection:

[0034] Including basic information: age, gender, height, weight; vital signs: body temperature, heart rate, blood pressure, respiratory rate; blood and body fluid examination: routine blood, urine and stool tests, coagulation function; biochemical examination: liver and kidney function, electrolytes, protein, blood lipids, cardiac markers, blood gas; nuclear medicine examination: thyroid function, adrenal function; functional examination: CT, chest X-ray, B-ultrasound, MRI; medical history: diagnosis, chief complaint, main medication; nursing: nasogastric feeding and other nursing plans; comprehensive geriatric and other assessments: nutritional status, swallowing function, Padua, Barthel, mental intelligence, drowsiness, delirium CAM, dementia CDR score, etc.

[0035] Step S2: Data standardization preprocessing:

[0036] The collected data is classified into count data, measurement data, and text data. Feature extraction is performed on text data. For example, diagnoses are classified based on the ICD-10 classification system. Chief complaints are classified based on symptoms in the 8th edition of the Diagnostics Standards. Primary therapeutic drugs are classified according to the General Pharmacopoeia. The specific classification methods mentioned above are based on the opinions of clinical experts and, when necessary, are determined using Delphi or the Analytic Hierarchy Process. Categorical feature data are converted to binary form using One-Hot encoding, and count data are normalized using the Z-score to meet subsequent modeling requirements.

[0037] Step S3: Feature screening

[0038] The maximum information coefficient (MIC) is used to measure the degree of association between two variables (linear or nonlinear relationship). The MIC calculation formula is: Where MIC(X; Y) is the MIC value between variables, which ranges from 0 to 1; B is the number of grids, which is generally 0.6 times the total amount of data; I(X; Y) is the mutual information value, that is,

[0039] Where I(X;Y) is the mutual information between variables X and Y, and P(x,y) is the joint probability density between the variables;

[0040] MIC is used to remove redundant features and simplify them;

[0041] Step S4: weighting of relevant factors

[0042] SPSS 26.0 statistical software was used to perform linear regression analysis on all selected data, with survival as the dependent variable and other factors as independent variables. Variable assignment method: continuous variables retained their original values; categorical variables were represented by "yes" and "no"; and ordered variables were represented by ascending values such as 0, 1, 2, etc., to indicate the corresponding level of clinical characteristics.

[0043] 1. Select factors with significance less than 0.05 for Pearson correlation analysis. The correlation coefficient calculation formula is:

[0044] P xy is the correlation coefficient between survival period X and each factor Y; Represents an m-dimensional vector x j and y j The mean of

[0045] 2. Factors with significance greater than 0.05 were analyzed using Spearman rank correlation analysis;

[0046] Spearman rank correlation calculation formula:

[0047] where γ XY is the rank correlation coefficient between survival X and each factor Y, p i and q i The variables x i ,y i Ranking;

[0048] 3. P xy and γ XY Converted into weights of each factor

[0049] or W ij is the weight of the i or j factor;

[0050] Step S5: Construct a weighted naive Bayes classification model (WNBC):

[0051] Prior probability P(C i ) Its maximum likelihood estimate can be calculated based on the existing data samples Among them, |D| represents the total number of samples, |D ci | indicates that the category value is C i The number of samples;

[0052] Conditional probability P(A j |C i ) If 0 occurs, Laplace calibration is introduced and the corrected conditional probability is estimated as:

[0053] Among them, |D ci,aj | indicates that the category is c i And attribute A j The value is a j The total number of samples, N j Represents attribute A jThe number of possible values; output posterior probability h nb (x) The category (survival period) that obtains the maximum value is taken as the classification result.

Claims

1. A method for predicting the survival of non-cancer palliative care patients based on clinical data, characterized in that: include: Step 1: Establish an evaluation indicator system and extract data from the hospital medical record system on one or more hospitalizations within one year for deceased patients who meet the non-cancer palliative care admission criteria; Step 2: Standardize and preprocess the extracted data; Step 3: Feature screening; Step 4: Weighting of relevant factors; Step 5: Construct a weighted naive Bayes classification model; output the category that maximizes the posterior probability as the classification result.

2. A method for predicting survival of non-tumor palliative care patients based on clinical data as claimed in claim 1, characterized in that The evaluation index system is established by extracting one or more hospitalization data within one year from the hospital medical record system of deceased patients who meet the non-cancer palliative care admission criteria, including basic information: age, gender, height, and weight; Vital signs: temperature, heart rate, blood pressure, respiratory rate; blood and body fluid tests: routine blood, urine, and stool tests, coagulation function; biochemical tests: liver and kidney function, electrolytes, protein, blood lipids, cardiac markers, blood gas; nuclear medicine tests: thyroid function, adrenal function; functional tests: CT, chest X-ray, B-ultrasound, MRI; Medical history: diagnosis, chief complaint, main medications; nursing: nasogastric feeding and other nursing plans; comprehensive geriatric assessment and other evaluation methods: nutritional status, swallowing function, Padua, Barthel, mental faculties, sleepiness, delirium CAM, dementia CDR scores, etc.

3. A method for predicting the survival of non-tumor palliative care patients based on clinical data as claimed in claim 1, characterized in that : The feature screening uses the maximum information coefficient (MIC) to perform feature redundancy deletion and simplification; MIC calculation formula: Where MIC(X; Y) is the MIC value between variables; B is the number of grids; I(X; Y) is the mutual information value, that is, Where I(X;Y) is the mutual information between variables X and Y, and P(x,y) is the joint probability density between the variables.

4. A method for predicting survival of non-tumor palliative care patients based on clinical data as claimed in claim 1, characterized in that The weighted related factors were analyzed by linear regression using SPSS 26.0 statistical software with survival as the dependent variable and other factors as the independent variables. ① Select factors with significance less than 0.05 for Pearson correlation analysis. The correlation coefficient calculation formula is: P xy is the correlation coefficient between survival period X and each factor Y; Represents an m-dimensional vector x j and y j The mean of ② Factors with a significance greater than 0.05 were analyzed using Spearman rank correlation analysis; the Spearman rank correlation calculation formula is: where γ XY is the rank correlation coefficient between survival X and each factor Y, p i and q i The variables x i ,y i rank, n is the number of samples in the data; ③P xy and γ XY Converted into weights of each factor or W ij is the weight of the i or j factor.

5. A method for predicting the survival of non-tumor palliative care patients based on clinical data as claimed in claim 1, characterized in that : The weighted naive Bayes classification model is constructed, and the posterior probability h nb (x) Calculation formula: The output makes the posterior probability h nb (x) The category (survival period) that obtains the maximum value is taken as the classification result.

Images