Renal cell carcinoma prognosis lifetime prediction method based on data set internal and external verification
Through the method based on internal and external verification of the data set, key clinical factors related to the prognostic survival of patients with renal cell carcinoma were screened out, and the prediction model was constructed and optimized, which solved the problem of low prediction accuracy of existing models and improved the accuracy of prediction models. and practicality.
Patent Information
- Application Number
- CN202411542818.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-21
AI Technical Summary
The existing renal cell carcinoma prognostic survival prediction model has low prediction accuracy due to excessive selection of clinical factors, and cannot fully consider the specificity and complexity of the disease.
Using an internal and external verification method based on data sets, final clinical factors with higher correlation with prognostic survival in patients with renal cell carcinoma were gradually screened out, and the initial model was constructed and iteratively optimized through internal and external verification to reduce the impact of redundant factors.
By gradually screening key clinical factors, the impact of redundant information on the predicted results is reduced, the accuracy and practicality of the prediction model are improved, and more powerful support is provided for patients' personalized treatment.
Smart Images
Figure CN119495426A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of clinical medical diagnostics, and in particular to a method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set. Background Art
[0002] Renal cell carcinoma is the most common renal malignancy in adults, accounting for more than 85% of all renal tumors. In recent years, the incidence of renal cell carcinoma has been on the rise worldwide, partly due to the popularity of imaging examinations and advances in early detection technology. Despite improvements in early detection and treatment methods, the high invasiveness and heterogeneity of renal cell carcinoma make it a highly dangerous disease. The prognosis of RCC patients is variable, with a 5-year survival rate ranging from more than 90% in early localized cases to less than 10% in advanced or metastatic cases. Therefore, accurately predicting the prognosis and survival of patients with renal cell carcinoma is crucial to improving the survival rate of patients with renal cell carcinoma.
[0003] In recent years, with the advancement of science and technology, the field of clinical research on breast cancer has ushered in a deep integration of machine learning and statistical methods. Researchers are committed to building efficient patient risk assessment prediction models to cope with the complexity and diversity of breast cancer. In this process, they innovatively used advanced strategies such as filtering and variance selection. The core of filtering, as a feature selection technique, is to quantify the potential value of each clinical factor in the prediction model by carefully scoring it. Subsequently, based on strict threshold settings or considerations based on the number of clinical factors, this method can accurately eliminate those clinical factors with low scores and limited contribution to the model, thereby achieving preliminary screening and optimization of features. On the other hand, the variance selection rule starts from the perspective of data variability and evaluates its potential impact on the model's predictive ability by calculating the variance of each clinical factor. On this basis, this method can automatically eliminate those clinical factors with small variance, that is, small data fluctuations, and may not provide enough information for the model, further improving the simplicity and accuracy of the model.
[0004] However, it is worth noting that although the above methods have improved the efficiency and performance of predictive model construction to a certain extent, they still have certain limitations. Specifically, these methods often fail to fully consider the specificity and complexity of the disease itself when constructing the model, but take all clinical factors into consideration equally. This non-differentiated processing method not only leads to the increase of redundant information in the model, but also may mask the most critical clinical features for specific disease types or stages. Therefore, in order to overcome this problem and further improve the accuracy and practicality of the prediction model, we urgently need to explore a smarter and more accurate feature selection scheme. This scheme should be able to fully combine the pathophysiological characteristics of breast cancer with the actual clinical needs, and through in-depth exploration and analysis of the intrinsic connections and differences between various clinical factors, accurately screen out the key features that have the greatest impact on the prediction results. Only in this way can we build a more efficient and reliable breast cancer risk assessment prediction model to provide more powerful support for personalized treatment and precision medicine for patients. Summary of the invention
[0005] The purpose of the present invention is to provide a method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set, so as to improve the problem of low accuracy of the existing prediction model due to excessive selection of clinical factors.
[0006] The present invention provides a method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set, comprising the following steps:
[0007] Obtaining the prognosis survival period and multiple related clinical factors and final clinical factors of patients with renal cell carcinoma, constructing a sample data set after associating the prognosis survival period with the final clinical factors, and dividing the sample data set into a training sample set and a validation sample set;
[0008] An initial model is constructed, and after the initial model is trained based on the training sample set, internal verification is performed based on the training sample set. If the internal verification passes, a prediction model with good internal verification is obtained. Otherwise, the number of training sample sets is increased, and the training sample sets are continuously iterated for training;
[0009] The prediction model that has passed the internal verification is externally verified based on the verification sample set. If the external verification is passed, the prediction model with good external verification is obtained. Otherwise, the number of training sample sets is increased, and the training sample sets are continuously iterated for training and internal verification is performed again.
[0010] The final clinical factors of renal cell carcinoma patients who passed the test were obtained and the final prediction results were obtained based on the externally validated prediction model.
[0011] The beneficial effect of the method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set provided by the present invention is that the method gradually screens out the final clinical factors that are more highly correlated with the prognosis and survival period of renal cell carcinoma patients, thereby reducing the influence of redundant factors on the accuracy of the prediction results.
[0012] Optionally, the relevant clinical factors include: age, body mass index, smoking, other diseases, family history, complications, occupational exposure, tumor site, CEA, VHL gene mutation, tumor size, lymph node metastasis, distant tumor metastasis, tissue type, tumor grade, immunotherapy, radiotherapy and radiotherapy cycle, targeted therapy, and traditional Chinese medicine treatment. Optionally, the relevant clinical factors include: age, body mass index, smoking, other diseases, family history, complications, occupational exposure, tumor site, CEA, VHL gene mutation, tumor size, lymph node metastasis, distant tumor metastasis, tissue type, tumor grade, immunotherapy, radiotherapy and radiotherapy cycle, targeted therapy, and traditional Chinese medicine treatment.
[0013] Optionally, the final clinical factors are: comorbidities, tumor size, lymph node metastasis, distant tumor metastasis, tumor grade, tissue type, immunotherapy, and targeted therapy.
[0014] Optionally, the process of obtaining the final prediction result includes:
[0015] Input each final clinical factor into a preset prediction model for weighted summation to obtain a total score value, which is the sum of comorbidity score value, tumor size score value, lymph node metastasis score value, tumor distant metastasis score value, tumor grade score value, tissue type score value, immunotherapy score value, and targeted therapy score value;
[0016] The total score value is mapped to obtain a final prediction result, which is the survival period of patients with renal cell carcinoma.
[0017] Optionally, the process of obtaining the final clinical factors comprises:
[0018] The probability P value of each preset clinical factor is obtained, and all preset clinical factors with P values less than the preset threshold are eliminated, and the remaining preset clinical factors are used as the final clinical factors.
[0019] Optionally, the preset threshold is 10%.
[0020] Optionally, when internal verification is performed based on the training sample set, it includes:
[0021] Based on the prediction model, the final clinical factors in the training sample set are internally verified to obtain the prediction results of the renal cell carcinoma patients to be internally verified, and after comparing the prediction results of the renal cell carcinoma patients to be internally verified with the corresponding true results in the test sample set, the similarity score, calibration curve, and decision analysis curve between the prediction results of the breast cancer patients to be internally verified and the true results are obtained; when the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the internal verification sample set fit each other, and the net benefit of the internal verification sample set is greater than the excellent net benefit threshold, the internal verification is passed, otherwise the number of initial sample sets is increased and iteration continues.
[0022] Optionally, external validation based on the validation sample set includes:
[0023] Based on the prediction model that has been internally verified, the final clinical factors in the verification sample set are externally verified to obtain the prediction results of the renal cell carcinoma patients to be externally verified, and after comparing the prediction results of the renal cell carcinoma patients to be externally verified with the corresponding real results in the verification sample set, the similarity score, calibration curve, and decision analysis curve between the prediction results of the renal cell carcinoma patients to be externally verified and the real results are obtained; when the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the training sample set and the verification sample set fit each other, and the net benefits of the training sample set and the verification sample set are greater than the excellent net benefit threshold, the external verification is passed, otherwise the number of training sample sets is increased and iteration continues. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a basic step flow chart of a method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set provided by the present invention;
[0025] Figure 2 The present invention provides a nomogram for predicting the prognosis and survival period of renal cell carcinoma based on the validation of the data set inside and outside the data set;
[0026] Figure 3 It is a regression verification diagram provided by the present invention;
[0027] Figure 4 is a consistency index curve diagram of the training sample set and the verification sample set provided by the present invention;
[0028] Figure 5 is a calibration curve diagram of the training sample set and the verification sample set provided by the present invention;
[0029] Figure 6 It is a decision analysis curve diagram of the training sample set and the verification sample set provided by the present invention. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.
[0031] The present invention provides a method for predicting the prognosis and survival of renal cell carcinoma based on internal and external validation of a data set. Figure 1 , the method comprises the following steps:
[0032] S1. Obtaining the prognosis survival period and multiple related clinical factors and final clinical factors of patients with renal cell carcinoma, correlating the prognosis survival period with the final clinical factors and then constructing a sample data set, and dividing the sample data set into a training sample set and a validation sample set;
[0033] S2, constructing an initial model, and after training the initial model based on the training sample set, performing internal verification based on the training sample set, and obtaining a prediction model with good internal verification if the internal verification passes, otherwise increasing the number of training sample sets, and continuing to iterate the training sample sets for training;
[0034] S3, the prediction model that has passed the internal verification is externally verified based on the verification sample set. If the external verification is passed, the prediction model with good external verification is obtained. Otherwise, the number of training sample sets is increased, and the training sample sets are continuously iterated for training and the internal verification is re-performed;
[0035] S4. Obtain the final clinical factors of renal cell carcinoma patients who have passed the test and obtain the final prediction results based on the externally validated prediction model.
[0036] In some embodiments, when executing step S1, the process of dividing the sample data set into a training sample set and a validation sample set includes:
[0037] According to the inclusion and exclusion criteria, a total of 129 patients diagnosed with renal cell carcinoma in Jingdezhen Second People's Hospital from January 2016 to December 2023 were selected for inclusion in the study as the sample data set;
[0038] The sample data set is divided into a training sample set and a validation sample set according to the random number table method. There are 86 cases in the training sample set and 43 cases in the validation sample set.
[0039] Specifically, the inclusion criteria must be met: (1) Patients diagnosed with primary renal cell carcinoma by histopathology and cytopathology. (2) All medical records of the patients are complete. (3) The patient's staging information and grading information are complete and available. (4) Postoperative adjuvant regular radiotherapy.
[0040] Furthermore, the exclusion criteria must be met: (1) patients with missing baseline data, unclear clinical stage and grade. (2) patients lacking some clinical or laboratory diagnostic information such as lymph node status, surgical and chemoradiotherapy treatment measures, CEA, etc. (3) patients with tumors of other organs and tissues. (4) patients with a history of other surgeries. (5) patients with complications such as other temperamental serious lesions such as heart, liver, kidney, etc. and patients with incomplete clinical data. (6) patients who did not receive surgical treatment. (7) patients with no follow-up records. (8) patients who died of other causes or whose cause of death is unknown.
[0041] Specifically, the relevant clinical factors in step S1 refer to factors that may affect the prognosis and survival of patients with renal cell carcinoma clinically, including:
[0042] Age, body mass index, smoking, other diseases, family history, complications, occupational exposure, tumor site, CEA, VHL gene mutation, tumor size, lymph node metastasis, distant tumor metastasis, tissue type, tumor grade, immunotherapy, radiotherapy and radiotherapy cycle, targeted therapy, and traditional Chinese medicine treatment are shown in Table 1 .
[0043] Table 1 Quantification scheme of relevant clinical factors
[0044]
[0045] Furthermore, the initial clinical factors were subjected to a proportional risk test and the P value of each initial screening clinical factor was calculated. The clinical factor VHL mutation with a P value of < 0.05 was eliminated, and the remaining 8 factors were the preset clinical factors, which were: comorbidities, tumor size, lymph node metastasis, distant tumor metastasis, tumor grade, tissue type, immunotherapy, and targeted therapy.
[0046] See also Figure 3, select logλ=-3, after regression graph verification, the confirmed clinical factors are still: comorbidity, tumor size, lymph node metastasis, distant tumor metastasis, tumor grade, tissue type, immunotherapy, targeted therapy. After regression verification, the final clinical factors are the same as the preset clinical factors. Therefore, the present invention uses comorbidity, tumor size, lymph node metastasis, distant tumor metastasis, tumor grade, tissue type, immunotherapy, and targeted therapy as the final clinical factors. The quantification scheme of the final clinical factors is shown in Table 2.
[0047] Table 2 Quantification scheme of final clinical factors
[0048]
[0049] In practice, the probability P value of each preset clinical factor is obtained, all preset clinical factors with P values less than a preset threshold are eliminated, and the remaining preset clinical factors are used as final clinical factors, and the preset threshold is 10%.
[0050] In some embodiments, the process of performing step S2 to perform internal verification based on the training sample set includes:
[0051] Based on the prediction model, the final clinical factors in the training sample set are internally verified to obtain the prediction results of the renal cell carcinoma patients to be internally verified, and after comparing the prediction results of the renal cell carcinoma patients to be internally verified with the corresponding true results in the test sample set, a similarity score, a calibration curve, and a decision analysis curve are obtained between the prediction results of the breast cancer patients to be internally verified and the true results;
[0052] When the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the internal validation sample set fit each other, and the net benefit of the internal validation sample set is greater than the excellent net benefit threshold, the internal validation passes; otherwise, the number of initial sample sets is increased and iterations continue.
[0053] In some embodiments, the process of performing step S3 to perform external verification based on the verification sample set includes the following steps:
[0054] Based on the prediction model that has been internally verified, the final clinical factors in the verification sample set are externally verified to obtain the prediction results of the renal cell carcinoma patients to be externally verified, and after comparing the prediction results of the renal cell carcinoma patients to be externally verified with the corresponding true results in the verification sample set, a similarity score, a calibration curve, and a decision analysis curve are obtained between the prediction results of the renal cell carcinoma patients to be externally verified and the true results;
[0055] When the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the training sample set and the validation sample set fit each other, and the net benefit of the training sample set and the validation sample set is greater than the excellent net benefit threshold, the external validation passes; otherwise, the number of training sample sets is increased and iterations continue.
[0056] Specifically, the consistency index is used to estimate the similarity between the actual result score and the predicted result score ranking. Its theoretical value is between 0 and 1. When the consistency index is > 0.5, it means that the prediction model with good internal verification has better performance. The calculated consistency index curve of the verification sample set is as follows: Figure 4 As shown in CD, the results show that the consistency index of the validation sample set is greater than 0.5, which indicates that the prediction model with good external validation fits the validation sample set well.
[0057] Furthermore, the calibration curve of the validation sample set is as follows Figure 5 As shown in EH, the results showed that the prediction model with good external validation showed good calibration performance and was able to accurately predict the actual survival rates at one, two, three, and five years.
[0058] Furthermore, Figure 6 The decision analysis curve of the validation sample set is shown, and it can be seen that the personal preset threshold of patients with renal cell carcinoma is 40% (that is, the probability of 5-year survival of patients with renal cell carcinoma is 40%, then patients with renal cell carcinoma will choose advice), and the net benefit when using the nomogram to make a decision is 0.105, which shows that further treatment and advice have additional net benefits than all treatment options or no treatment options. By considering the patient's specific situation and predicted results, doctors can develop more precise and targeted treatment plans, thereby improving treatment outcomes and reducing unnecessary treatment costs.
[0059] Furthermore, it can be seen from the similarity curve, calibration curve, and decision result analysis curve that the external verification is passed and the prediction results can be obtained based on the prediction model that has been internally verified.
[0060] In some embodiments, the process of obtaining the final prediction result in step S4 includes:
[0061] S4-1. Input each final clinical factor into a preset prediction model for weighted summation to obtain a total score value, which is the sum of the comorbidity score value, tumor size score value, lymph node metastasis score value, tumor distant metastasis score value, tumor grade score value, tissue type score value, immunotherapy score value, and targeted therapy score value;
[0062] S4-2. Map the total score value to obtain a final prediction result, where the final prediction result is the survival period of the renal cell carcinoma patient.
[0063] In fact, when executing step S4-2, the total score value is reflected on the nomogram, and the total score value is mapped to the nomogram according to the nomogram to obtain the survival period of the renal cell carcinoma patient.
[0064] For details, see Figure 2 , map each final clinical factor to the top score of the nomogram, obtain the corresponding score of each final clinical factor, accumulate the scores corresponding to each final clinical factor and map them to the total score, and map the total score downward to obtain the one-year, three-year, and five-year survival rates of patients with renal cell carcinoma.
[0065] Further, the corresponding Chinese and English explanations of the drawings are as follows:
[0066] Points: The score corresponding to each final clinical factor
[0067] Comorbidity
[0068] T: tumor size
[0069] N: Lymph node metastasis
[0070] M: distant metastasis of tumor
[0071] Grade: Tumor grade
[0072] Tissue: tissue type
[0073] Immunotherapy
[0074] Target: Targeted therapy
[0075] Total Points: Total score
[0076] 12-Month Survival: One-year survival rate
[0077] 36-Month Survival: Three-year survival rate
[0078] 60-Month Survival: Five-year survival rate
[0079] Partial Likelihood Deviance: Partial Likelihood Deviance
[0080] Coefficients: Coefficients
[0081] Concordance index: consistency index
[0082] variables: clinical factor variables
[0083] Time(month): Time(month)
[0084] Fraction Surviving (12 / 36 / 60 / 84) Months: The probability of actual survival (12 months / 36 months / 60 months / 84 months)
[0085] Predicted(12 / 36 / 60 / 84)Months Survival: Predicted probability of survival (12 months / 36 months / 60 months / 84 months)
[0086] Net benefit: Net benefit
[0087] Threshold probability: Threshold probability
[0088] Treat All: Receive treatment
[0089] Treat None: Do not receive treatment
[0090] Smoothed net benefit: Smoothed net benefit
[0091] Validation: Validation sample set
[0092] Training: training sample set
[0093] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.
Claims
1. A method for predicting the prognosis and survival of renal cell carcinoma based on internal and external validation of a data set, characterized in that: include: Obtaining the prognosis survival period and multiple related clinical factors and final clinical factors of patients with renal cell carcinoma, constructing a sample data set after associating the prognosis survival period with the final clinical factors, and dividing the sample data set into a training sample set and a validation sample set; An initial model is constructed, and after the initial model is trained based on the training sample set, internal verification is performed based on the training sample set. If the internal verification passes, a prediction model with good internal verification is obtained. Otherwise, the number of training sample sets is increased, and the training sample sets are continuously iterated for training; The prediction model that has passed the internal verification is externally verified based on the verification sample set. If the external verification is passed, the prediction model with good external verification is obtained. Otherwise, the number of training sample sets is increased, and the training sample sets are continuously iterated for training and internal verification is performed again. The final clinical factors of renal cell carcinoma patients who passed the test were obtained and the final prediction results were obtained based on the externally validated prediction model.
2. A method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: The relevant clinical factors include: age, body mass index, smoking, comorbidities with other diseases, family history, complications, occupational exposure, tumor site, CEA, VHL gene mutation, tumor size, lymph node metastasis, distant tumor metastasis, tissue type, tumor grade, immunotherapy, radiotherapy and radiotherapy cycle, targeted therapy, and traditional Chinese medicine treatment.
3. The method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: The final clinical factors are: comorbidities, tumor size, lymph node metastasis, distant tumor metastasis, tumor grade, tissue type, immunotherapy, and targeted therapy.
4. The method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: The process of obtaining the final prediction result includes: Input each final clinical factor into a preset prediction model for weighted summation to obtain a total score value, which is the sum of comorbidity score value, tumor size score value, lymph node metastasis score value, tumor distant metastasis score value, tumor grade score value, tissue type score value, immunotherapy score value, and targeted therapy score value; The total score value is mapped to obtain a final prediction result, which is the survival period of patients with renal cell carcinoma.
5. The method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: The process of obtaining the final clinical factors includes: The probability P value of each preset clinical factor is obtained, and all preset clinical factors with P values less than the preset threshold are eliminated, and the remaining preset clinical factors are used as the final clinical factors.
6. A method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 5, characterized in that: The preset threshold is 10%.
7. The method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: When internal verification is performed based on the training sample set, it includes: Based on the prediction model, the final clinical factors in the training sample set are internally verified to obtain the prediction results of the renal cell carcinoma patients to be internally verified, and after comparing the prediction results of the renal cell carcinoma patients to be internally verified with the corresponding true results in the test sample set, the similarity score, calibration curve, and decision analysis curve between the prediction results of the breast cancer patients to be internally verified and the true results are obtained; when the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the internal verification sample set fit each other, and the net benefit of the internal verification sample set is greater than the excellent net benefit threshold, the internal verification is passed, otherwise the number of initial sample sets is increased and iteration continues.
8. The method for predicting the prognosis and survival period of renal cell carcinoma based on internal and external validation of a data set according to claim 1, characterized in that: When external validation is performed based on a validation sample set, it includes: Based on the prediction model that has been internally verified, the final clinical factors in the verification sample set are externally verified to obtain the prediction results of the renal cell carcinoma patients to be externally verified, and after comparing the prediction results of the renal cell carcinoma patients to be externally verified with the corresponding real results in the verification sample set, the similarity score, calibration curve, and decision analysis curve between the prediction results of the renal cell carcinoma patients to be externally verified and the real results are obtained; when the similarity score is greater than the excellent score threshold, the calibration curves of the predicted disease probability and the actual disease probability in the training sample set and the verification sample set fit each other, and the net benefits of the training sample set and the verification sample set are greater than the excellent net benefit threshold, the external verification is passed, otherwise the number of training sample sets is increased and iteration continues.