Local progression stage gastric cancer neoadjuvant chemotherapy curative effect prediction model construction method and prediction system

Abstract

The embodiment of the invention discloses a local progression stage gastric cancer neoadjuvant chemotherapy curative effect prediction model construction method and a prediction system. According to the method, clinical factors related to the NCT curative effect of the LAGC patient are reported by utilizing system evaluation and meta analysis, an NCT curative effect prediction model of the LAGC patient based on general clinical characteristics is constructed, and further verification is carried out by utilizing a large sample LAGC patient queue in the real world, so that a new thought and clue are provided for prediction of the NCT curative effect and determination of a treatment scheme of the LAGC patient, and the method is suitable for popularization and application. And a basis is provided for individualized and precise treatment of LAGC patients.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of medical technology, and specifically to a method for constructing a model and a prediction system for predicting the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer. Background Art

[0002] The overall efficacy of neoadjuvant chemotherapy (NCT) combined with surgery in the treatment of locally advanced gastric cancer (LAGC) has been recognized. Compared with surgery alone combined with adjuvant chemotherapy, preoperative NCT can improve the survival rate of most LAGC patients, improve the clinical symptoms of patients' tumors, eliminate potential micrometastases, and prevent and reduce postoperative recurrence and metastasis. However, due to the heterogeneity of tumors, gastric cancer patients with similar clinical characteristics have different treatment effects. Therefore, accurately evaluating the efficacy of NCT for gastric cancer and accurately predicting gastric cancer patients who are sensitive to NCT are important parts of individualized treatment of LAGC.

[0003] Using the general clinical characteristics of the tumor, including the patient's personal situation, biochemical indicators, and clinical related information, to construct a prediction model based on clinical characteristics is a good way to predict the efficacy of neoadjuvant chemotherapy (NCT) for advanced gastric cancer and to evaluate the individualized regimen. At present, some researchers have explored the factors affecting the efficacy of NCT for advanced gastric cancer and developed relevant prediction models. However, most of the existing prediction models are developed based on post hoc analysis of randomized controlled trials. These prediction models based on general clinical characteristics to predict the efficacy of NCT for gastric cancer still have the following problems: 1) The model is only based on small cross-sectional studies, and there is a lack of systematic research. The influencing factors closely related to the efficacy of NCT are not completely clear; 2) There is a lack of information on the weight of risk factors; 3) The sample size of the reported studies is small, and the results are not credible; 4) The clinical characteristics used to construct the prediction model are not comprehensive; 5) The research is often limited to one of the NCT chemotherapy regimens, and the clinical application is poor. Summary of the invention

[0004] To this end, the embodiment of the present invention provides a method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer and a prediction system. The present invention systematically evaluates and meta-analyzes literature reports, obtains clinical prediction factors related to the NCT efficacy of LAGC patients, constructs an NCT efficacy prediction model for LAGC patients, and verifies it using a large sample LAGC patient cohort in the real world. The model constructed by this method only relies on easily accessible clinical parameters, improves the accuracy of NCT treatment decisions for advanced gastric cancer, provides new ideas and clues for the prediction of NCT efficacy and the determination of treatment plans for LAGC patients, and provides a basis for individualized and precise treatment of LAGC patients.

[0005] In order to achieve the above purpose, the embodiment of the present invention provides the following technical solutions:

[0006] According to a first aspect of an embodiment of the present invention, the present invention provides a method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer, the method comprising the following steps:

[0007] (1) Identify and screen the database to form a derivation cohort;

[0008] (2) Conduct a systematic review and meta-analysis of the derivation cohort to screen out clinical predictive factors associated with the efficacy of neoadjuvant chemotherapy for advanced gastric cancer;

[0009] (3) To evaluate the risk of clinical predictive factors and to construct a model for predicting the efficacy of neoadjuvant chemotherapy for advanced gastric cancer;

[0010] (4) Use a validation cohort to validate the constructed efficacy prediction model;

[0011] Among them, the clinical predictive factors related to the efficacy of neoadjuvant chemotherapy for advanced gastric cancer are: CEA, tumor location, Lauren classification, differentiation, infiltration depth, lymph nodes, clinical stage, Her-2, and Ki67.

[0012] Furthermore, the contents of identifying and screening the database include:

[0013] Articles were identified by database search, with the search terms for Chinese databases being: gastric cancer, gastric malignancies, neoadjuvant chemotherapy, preoperative chemotherapy, efficacy, and effect; and the search terms for English databases being: Cancer of the stomach, neoadjuvant chemotherapy, Neoadjuvant Treatment, and risk factor;

[0014] Duplicates were removed, and articles meeting the criteria were obtained and assessed using the Newcastle-Ottawa Scale.

[0015] Furthermore, the contents of the systematic review and meta-analysis include:

[0016] The following information was extracted from the derivation cohort and entered into the database: details of the study design, region and year of study publication, patient demographics (age and sex), number of enrolled patients and incidence, study outcomes, identified risk factors and their RRs with corresponding 95% CIs, and identified influencing factors for the number of included studies ≥3;

[0017] For influencing factors with ≥9 included studies, publication bias analysis was performed using visual funnel plots. If the funnel plot was inverted and symmetrical, it indicated that there was no publication bias; if the funnel plot was asymmetrical, it indicated that there was publication bias. For influencing factors with 3-8 included studies, the Egger linear regression test was performed to assess publication bias. When P < 0.10, it indicated that there was publication bias, and factors related to efficacy were determined.

[0018] The relative risks (RRs) and 95% confidence intervals (CIs) of efficacy-related factors in each study were extracted, and the combined statistical analysis was performed using Stata SE 15.1 software, and forest plots were drawn to present the results;

[0019] The heterogeneity of the studies was tested by Q test. When the result was P ≥ 0.10 and I 2 When the Q test result was P < 0.10 or I 2 When the heterogeneity was >50%, it indicated that the heterogeneity was statistically significant, and a random effects model was selected for meta-analysis, and sensitivity analysis was performed, and descriptive reports were made for studies with obvious clinical heterogeneity;

[0020] Finally, among the generated pooled RRs and corresponding 95% CIs, when the 95% CI did not contain 1 or P ≤ 0.05, it indicated that the pooled statistic was statistically significant and that the factor was different between the two groups.

[0021] Furthermore, the risk scores of clinical predictive factors were calculated, and the contents of the prediction model for the efficacy of neoadjuvant chemotherapy for advanced gastric cancer were constructed, including:

[0022] The combined RRs of all clinical predictive factors and their different stratifications were extracted. The β coefficient of each influencing factor was calculated based on the combined RRs and their corresponding 95% CI. The calculation formula was: β = ln (RR). The approximate number of β multiplied by 10 was taken and one decimal place was retained. The one decimal part was 0-0.2 and retained as 0, 0.3-0.7 was retained as 0.5, and 0.8-0.9 was retained as 1.0, and the score of each influencing factor was obtained.

[0023] All influencing factors in the prediction model were classified according to the meta-analysis and NCCN guidelines for gastric cancer, and a score was assigned to each category. The total score was obtained by adding the scores of all parameters in the prediction model.

[0024] Furthermore, the validation cohort includes validation cohort 1 and validation cohort 2, and the inclusion criteria are: real-world LAGC patients who underwent NCT and radical surgery, received at least 2 cycles of NCT before surgery and underwent radical gastrectomy after chemotherapy, excluded patients aged > 75 years, excluded patients who failed to complete neoadjuvant chemotherapy due to changes in their condition, excluded patients with residual gastric cancer, and excluded patients with other malignant tumors. The validation cohorts were divided into validation cohort 1 and validation cohort 2 based on the different onset times.

[0025] Furthermore, the validation of the constructed efficacy prediction model in validation cohorts 1 and 2 included: using the real-world validation cohorts 1 and 2 to evaluate the generalizability of the prediction model and calculate the actual effectiveness of NCT treatment for LAGC patients; assigning points to each predictive factor according to the prediction model and calculating the total score of each patient, and calculating the sensitivity, specificity and area under the curve (AUC) at different critical values; selecting the score corresponding to the maximum Youden index to determine the optimal critical value, and dividing the patients into two risk levels: medium-low risk group and high risk group; then drawing the ROC curves of the two groups of patients, respectively, and selecting the maximum Youden index of the two groups, and dividing the patients into four risk levels: relatively low, medium, high and extremely high risk.

[0026] According to a second aspect of an embodiment of the present invention, the present invention provides a system for predicting the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer, comprising:

[0027] Input module: Input the following clinical indicators of the patient: CEA, tumor location, Lauren classification, differentiation, invasion depth, lymph nodes, clinical stage, Her-2, Ki67;

[0028] Analysis module: a prediction model for the efficacy of neoadjuvant chemotherapy for advanced gastric cancer established by any of the methods described in the above items is provided;

[0029] Output module: Output the predicted results of the neoadjuvant chemotherapy efficacy for patients with advanced gastric cancer.

[0030] Furthermore, the input unit, the analysis unit and the output unit are connected in a wired and / or wireless manner.

[0031] The embodiments of the present invention have the following advantages:

[0032] We developed and validated a transparent "white box" model. By uniformly processing and analyzing the clinical characteristics, treatment regimens, efficacy evaluation and other data of patients in multiple related studies in the derivation cohort, we effectively reduced the possible bias and limitations of a single study and more accurately revealed the relationship between these factors and the efficacy of NCT. Meta-analysis can dig out clinical characteristics or biomarkers related to the efficacy of NCT that were ignored in a single study. This potential information is of great significance for improving the accuracy and practicality of the prediction model. The constructed model was further validated in two real-world patient external validation cohorts to determine the stability and reliability of the model. Its AUC was 0.760 and 0.786, respectively, which is better than the existing NCT efficacy prediction model for LAGC patients based on general clinical data, and can more accurately distinguish patients' response to NCT. At the same time, the model integrates more clinical characteristics, considers multiple NCT regimens, and is more adaptable to clinical needs. It greatly improves the limitations of the existing model, which includes a small number of patients and clinical characteristics, and usually only includes one NCT regimen. In addition, we quantified the performance evaluation indicators of the model. The model further divided patients into four risk levels by calculating the risk score and the corresponding Youden index, which can distinguish high-risk patients from other populations. In validation cohorts 1 and 2, the resistance rate of patients in the high-risk group was 4.05 times and 2.55 times that of patients in the low-risk group, respectively, and the resistance rate of the extremely high-risk group increased to 6.6 times and 5 times, respectively, confirming the effectiveness of the risk scoring system in predicting the resistance of gastric cancer patients to neoadjuvant chemotherapy. Our research results show that this method can be used as a powerful tool to predict the benefit of NCT treatment.

[0033] The model of the present invention incorporates 9 easily accessible clinical features based on Meta-analysis and real-world data. The model integrates a large amount of data, mines potential information, has strong generalization ability, can accurately predict efficacy and prognosis, effectively support clinical decision-making, is convenient for network application, and can be updated and improved later. At the same time, the prediction model constructed based on data from multiple studies can better meet the needs of different populations and different clinical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0035] Figure 1 The derivation queue provided by the present invention is incorporated into the flow chart;

[0036] Figure 2The verification queue provided by the present invention is included in the flow chart;

[0037] Figure 3 Among the 22 influencing factors provided by the present invention, 9 factors were related to the NCT efficacy of LAGC patients;

[0038] Figure 4 Receiver operating characteristic curve provided by the present invention, A: RECIST group in validation cohort one, B: TRG group in validation cohort one, C: RECIST group in validation cohort two, D: TRG group in validation cohort two;

[0039] Figure 5 A bar graph of drug resistance risk for the four risk groups provided by the present invention in validation cohorts one and two;

[0040] Figure 6 The Kaplan-Meier curves of LAGC patients in different risk score groups provided by the present invention;

[0041] Figure 7 The network deployment of the prediction model provided by the present invention. DETAILED DESCRIPTION

[0042] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. 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.

[0043] Example 1

[0044] 1. Data Source

[0045] 1.1 Derivation Queue

[0046] Articles were identified by searching electronic databases (from inception to August 2023) in English (Pubmed, Web ofscience, Embase, Cochrane) and Chinese (CNKI, Wanfang, VIP) databases. The search strategy included a combination of text and subject terms (MeSH). The search terms for Chinese databases were: "gastric cancer", "gastric malignancy", "neoadjuvant chemotherapy", "preoperative chemotherapy", "efficacy", "effect", etc.; the search terms for English databases were: "Cancer ofthe stomach", "neoadjuvant chemotherapy", "Neoadjuvant Treatment", "risk factor", etc. To verify potential relevant studies, references in the retrieved articles were manually reviewed. A flowchart of the study selection method is shown in the figure. Figure 1 As shown, 25 cohorts (including 2 prospective cohorts and 23 retrospective cohorts) were obtained as derivation cohorts, and all studies reported the risk ratios (RRs) or standardized mean differences (SMDs) of the influencing factors and the corresponding 95% CI, and were evaluated using the Newcastle-Ottawa scale. A total of 4014 gastric cancer patients from Europe (including Germany), Asia (including China), and America (including the United States) were included, with clinical stage cT1-2N+M0 or cT3-4bN±M0 locally advanced gastric cancer, of which 80% were Asians and 20% were whites.

[0047] 1.2 Verification Queue

[0048] This study collected LAGC patients who underwent NCT and radical gastrectomy in Liaoning Provincial Cancer Hospital between June 1, 2014 and July 31, 2021 as validation cohort 1, and collected LAGC patients who underwent NCT and radical gastrectomy in Liaoning Provincial Cancer Hospital between August 1, 2021 and January 31, 2022 as validation cohort 2. According to the 8th edition of the AJCC cTNM staging system for gastric cancer, the clinical stage was cT1-2N+M0 or cT3-4b any N and M0 for the first-time LAGC patients. Inclusion criteria were patients aged between 18 and 75 years; patients diagnosed with gastric epithelial malignant tumors by gastroscopy biopsy after consultation, with no specific pathological type required; patients who received at least 2 cycles of NCT before surgery and underwent radical gastrectomy after chemotherapy. Exclusion criteria were patients with concurrent primary active malignant tumors in other sites or other malignant tumors within 3 years before treatment and patients with residual gastric cancer. Finally, in validation cohort one, 12 patients aged >75 years, 27 patients who failed to complete NCT due to changes in their condition, 17 patients with gastric remnant cancer, 12 patients with other malignant tumors, and 20 patients with incomplete clinical data were excluded from 841 LAGC patients; in validation cohort two, 2 patients aged >75 years, 3 patients who failed to complete NCT due to changes in their condition, 1 patient with gastric remnant cancer, 5 patients with other malignant tumors, and 1 patient with incomplete clinical data were excluded from 112 LAGC patients. The remaining 753 patients with advanced gastric cancer were included in validation cohort one, and 100 patients with advanced gastric cancer were included in validation cohort two. Overview of the patient selection process for inclusion in the study. Figure 2 shown.

[0049] 1.3 Evaluation of NCT treatment response

[0050] The Mandard tumor regression grading system was used to determine the efficacy of NCT in the patient's postoperative specimens: grade 1, complete regression, no residual tumor cells; grade 2, very few cancer cells remaining; grade 3, local tumor regression, fibrosis reaction exceeds residual cancer cells; grade 4, cancer cells exceed fibrosis reaction; grade 5, almost no response. Mandard grades 1-3 were defined as the chemotherapy response group, and grades 4-5 were defined as the chemotherapy non-response group. Two senior pathologists evaluated the patient's pathological response to NCT treatment under double-blind conditions. If there was a disagreement, a third pathology expert arbitrated to reach a consensus.

[0051] The response to NCT in patients was evaluated using the Response Evaluation Criteria in Solid Tumors (RECIST v1.1): Complete Response (CR) means that all target lesions disappear, Partial Response (PR) means that the sum of the diameters of all target lesions decreases by ≥30%, Progressive Disease (PD) means that the sum of the diameters of all target lesions increases by ≥20%, and Stable Disease (SD) means that the reduction is insufficient to meet the conditions of PR, or the increase is insufficient to meet the conditions of PD. Patients with CR or PR were classified as clinical remission (effective group). Two senior radiologists evaluated the changes in CT images of patients before and after NCT treatment in a double-blind manner. If there was a disagreement, a third radiologist would arbitrate to reach a consensus.

[0052] In this study, TRG 1-3 and TRG 4-5 were considered to correspond to CR / PR and SD / PD in RECIST 1.1, respectively.

[0053] 2. Research Methods

[0054] 2.1 Systematic review and meta-analysis

[0055] The following information was extracted from the included studies and entered into the database: details of study design, region and year of study publication, patient demographics (age and sex), number of enrolled patients and incidence, study outcomes, identified risk factors and their RRs with corresponding 95% CIs.

[0056] For factors with ≥9 included studies, publication bias analysis was performed using visual funnel plots. If the funnel plot was inverted and symmetrical, there was no publication bias; if the funnel plot was asymmetrical, there was publication bias. For factors with 3-8 included studies, the Egger linear regression test was performed to assess publication bias, and when P < 0.10, publication bias was indicated. Publication bias assessment of included studies can improve the reliability and authenticity of the results of the systematic review.

[0057] In the meta-analysis of the 25 cohort studies in the derivation cohort, the relative risks (RRs) and 95% confidence intervals (CIs) of the efficacy-related factors in each study were first extracted. To reduce the impact caused by the differences in NCT efficacy evaluation criteria in different studies, the efficacy evaluation criteria for the same influencing factors were not distinguished, and the correspondence between TRG grading and RECIST grouping was performed according to the established definition, that is, TRG 1-3 and TRG 4-5 were considered to correspond to CR / PR and SD / PD in RECIST 1.1, respectively. If a study provided data on both efficacy evaluation criteria, tumor response grading (TRG) was preferred as the evaluation criterion. When a certain influencing factor was evaluated by TRG or RECIST alone in at least two studies, subgroup analysis of TRG or RECIST was performed for the influencing factor. Stata SE 15.1 software was used for combined statistical analysis, and forest plots were drawn to present the results.

[0058] The heterogeneity of the studies was tested by Q test. When the result was P ≥ 0.10 and I 2 When the Q test result was P < 0.10 or I 2 When the factor is >50%, it means that the heterogeneity is statistically significant. The random effects model is selected for meta-analysis and sensitivity analysis. Descriptive reports are made for studies with obvious clinical heterogeneity. Finally, in the generated combined RRs and corresponding 95% CI, when the 95% CI does not contain 1 or P≤0.05, it means that the combined statistic is statistically significant, and the factor is different between the chemotherapy effective group (TRG1-3 or CR / PR) and the chemotherapy ineffective group (TRG4-5 or SD / PD), indicating that this factor is a related factor for chemotherapy efficacy. When the included factors are divided into different levels of stratification in different studies (for example, some studies divide age into ≤55 years and >55 years, while some studies divide age into ≤60 years and >60 years), the studies with the same stratification will be merged, and the stratified analysis and subgroup analysis of the variable will be carried out according to the above steps based on the TRG and RECIST efficacy judgment criteria of the factor.

[0059] 2.2 Model development

[0060] The NCT efficacy prediction model was developed according to the method proposed by Sullivan et al. First, the clinical predictive factors included in the prediction model were selected from the above-mentioned systematic reviews and meta-analyses. Secondly, the combined RRs of all influencing factors and their different stratifications were extracted. The β coefficient of each influencing factor was calculated based on the combined RRs and their corresponding 95% CIs, and the calculation formula was: β = ln (RR). Then, the approximate number of β multiplied by 10 was taken and one decimal place was retained (the 1 decimal part was 0-0.2 and retained as 0, 0.3-0.7 and retained as 0.5, and 0.8-0.9 were retained as 1.0), and the scores of each influencing factor were obtained. Finally, all the influencing factors in the prediction model were classified according to the meta-analysis and the NCCN guidelines for gastric cancer, and a score was assigned to each category. The total score was obtained by adding the scores of all parameters in the prediction model.

[0061] 2.3 Model Validation

[0062] In order to verify the predictive ability of the prediction model, we used the real-world validation cohorts 1 and 2 to evaluate the generalizability of the prediction model and calculated the actual effective rate of NCT treatment for LAGC patients. After assigning points to each influencing factor according to the prediction model, the total score of each patient was calculated, and the sensitivity, specificity, and area under the curve (AUC) at different critical values ​​were calculated. The score corresponding to the maximum Youden index was selected to determine the optimal critical value, and the patients were divided into two risk levels: medium-low risk group and high-risk group. Then, the ROC curves of the medium-low risk group and high-risk group were drawn respectively, and the maximum Youden index of the two groups was selected to divide the patients into four risk levels: relatively low, medium, high, and very high risk. In addition, we also generated Kaplan-Meier curves for each risk group to preliminarily explore the relationship between different risk levels and survival outcomes. All analyses were performed using SPSS software 26.0 and R 4.2.0 software.

[0063] 2.4 Network deployment of the model

[0064] In order to provide a more convenient NCT resistance risk assessment tool for the clinic, we use the Shiny framework in R language on the server side to convert the data into a web page that is convenient for doctors and patients to use. First, we define the score mapping table to associate clinical indicator options with scores. Next, we use the 'fluidPage' function to create a concise and clear user interface, in which we enable the title panel to clearly convey the function of the calculator and set a selection input box in the sidebar to facilitate users to select different clinical indicator options. At the same time, the sidebar also sets two action buttons, "Calculate Risk Level" and "Clear Options with One Click", to facilitate doctors to operate and reselect. After that, we define the `reactiveValues` object `user_scores` to store the scores selected by the user. When the user selects different options, the calculator listens to the event through `observeEvent` to ensure that the value of `user_scores` is updated in a timely and accurate manner. Finally, we use the `calculate_score` function to obtain the corresponding score from the score mapping table according to the option selected by the user, and then calculate the total score, and determine the patient's risk level based on the total score.

[0065] 3. Description of baseline characteristics of the included cohort

[0066] 3.1 Baseline characteristics of the derivation cohort

[0067] The patients in the 25 study cohorts included in the derivation cohort were aged between 18 and 85 years old, and their baseline characteristics are shown in Table 1. According to the Newcastle-Ottawa Scale, the above 25 studies were all high-quality studies, with scores exceeding 8 points (see Table 2). 2190 patients in the derivation cohort were evaluated as having chemotherapy efficacy, with a chemotherapy efficacy rate of approximately 54.6%.

[0068] Table 1 Baseline characteristics of the 25 cohorts included in the systematic review and meta-analysis

[0069]

[0070]

[0071]

[0072] Table 2 Newcastle-Ottawa quality assessment scale for 25 cohort studies

[0073]

[0074] 3.2 Baseline characteristics of the validation cohort

[0075] Of the 753 patients with advanced disease included in validation cohort 1, 27.0% were female. The mean age ± SD at baseline was 59.05 ± 8.89 years. The proportion of smokers was 56.2%, and 33.5% of patients were long-term drinkers. 43.2% of patients had progressive weight loss at the onset of disease, and 58.8% of patients had symptoms that affected their normal life. 63.6% of patients received SOX regimen NCT, 9.0% of patients received XELOX regimen NCT, and 45.3% of patients received more than 2 cycles of NCT. The baseline characteristics of patients in validation cohort 1 are shown in Table 3. In validation cohort 1, 496 patients (65.9%) were evaluated as chemotherapy effective by TRG grading, and 461 patients (61.2%) were evaluated as chemotherapy effective by RECIST score.

[0076] Of the 100 patients with advanced disease included in the validation cohort II, 32.0% were female. The mean age ± SD at baseline was 64 ± 8.49 years. The proportion of smokers was 46%, and 27% of patients were long-term drinkers. 35% of patients had progressive weight loss at the onset of disease, and 20% of patients had symptoms that affected their normal life. 60% of patients received SOX regimen NCT, 11% of patients received XELOX regimen NCT, and 46% received more than 2 cycles of NCT. The baseline characteristics of patients in the validation cohort II are shown in Table 4. In the validation cohort II, 65 patients (65%) were evaluated as chemotherapy effective by TRG grading, and 66 patients (66%) were evaluated as chemotherapy effective by RECIST score. There was no significant difference in the intergroup comparison of the baseline characteristics of the validation cohorts I and II.

[0077] Table 3 Baseline characteristics of patients in validation cohort 1

[0078]

[0079]

[0080] Table 4 Baseline characteristics of patients in validation cohort 2

[0081]

[0082]

[0083] 4. Model derivation results

[0084] 4.1 Results of a systematic review and meta-analysis of the efficacy of neoadjuvant chemotherapy for gastric cancer based on 25 cohorts

[0085] Data of the 25 included cohorts were extracted and merged. Among the 62 influencing factors mentioned in the studies, 22 were included in more than two cohort studies. These 22 influencing factors included: gender, age, smoking, CEA, CA199, tumor site, gross classification, histological classification, Lauren classification, degree of differentiation, depth of invasion, lymph node metastasis, clinical stage, NCT regimen, NCT cycle, Her-2, EGFR, Ki67, P53, Her-2 and P53, TOPOII, and neural infiltration. We conducted stratified analysis based on the different levels of inclusion of the same influencing factor in different studies. The stratified factors included age (≤55 years / > 55 years and ≤60 years / > 60 years), tumor location (cardia fundus, gastric body, gastric antrum and whole stomach), degree of differentiation (well differentiated, moderately differentiated, poorly differentiated, undifferentiated and well-moderately differentiated), Lauren classification (intestinal type, diffuse type, mixed type and intestinal type + diffuse type), depth of invasion (T2, T2-T3, T3 and T4), lymph nodes (N0, N1, N2 and N3), clinical stage (stage II, stage I + II and stage III), chemotherapy regimen (SOX, XELOX and FOLFOX), and chemotherapy cycles (2 cycles, 3 cycles and 4 cycles). Bias analysis showed that none of the 22 influencing factors had publication bias and were determined to be factors related to efficacy.

[0086] Next, we performed subgroup analyses of factors that influenced the efficacy evaluation criteria of TRG or RECIST alone in at least two studies. In the TRG subgroup, the factors we analyzed included sex, tumor site, gross type, degree of differentiation, depth of invasion, and lymph node metastasis. In the RECIST subgroup, the factors we analyzed included sex, tumor site, gross type, and degree of differentiation.

[0087] The results showed that 9 of the 22 influencing factors were associated with the efficacy of NCT in LAGC patients. These 9 factors and their combined RRs were: CEA ≤ 5 μg / L (1.10), gastric cardia fundus tumor (1.77), intestinal type gastric cancer (1.28), well-differentiated (1.70), T2 (1.90), N0 (1.75), clinical stage I+II (1.42), Her-2 (+++) (4.16), Ki67 < 10% (2.29) ( Figure 3 ).

[0088] 4.2 Prediction model construction

[0089] In order to further construct a prediction model based on the factors related to NCT efficacy obtained by meta-analysis, we converted the RR value into a β coefficient and calculated the risk score for each influencing factor. Table 5 lists the number of studies, sample size, combined RRs, 95% CI, β coefficient, and risk score involved in the prediction model for NCT efficacy of gastric cancer.

[0090] Table 5 Number of studies, sample size, pooled RRs, 95% CI, β coefficient and risk score

[0091]

[0092] Based on this, we developed a prediction model for the efficacy of NCT in LAGC patients based on general clinical characteristics (Table 6). The nine parameters and their risk scores included in this prediction model are: CEA (≤5 μg / L-1 point, >5 μg / L 0 points); tumor site (cardia 5.5 points, gastric antrum -1.5 points, other sites 0 points); Lauren classification (intestinal type 2.5 points, diffuse type 1.5 points, mixed type 0 points); differentiation (well differentiated 5.5 points, moderately differentiated 3.5 points, poorly undifferentiated 0 points); depth of invasion (T2 6.5 points, T3 4 points, T4 -2.5 points); lymph nodes (N0 5.5 points, N1 3 points, N2 0 points, N3 -5 points); clinical stage (I-II stage 3.5 points, III-IV stage 0 points); Her-2 [Her-2 (+++) 14.5 points, Her-2 (-~++) 0 points]; Ki67 (<10% 0 points, ≥10% -2.5 points). This prediction model is recommended for LAGC patients aged 18-85 years, mainly for whites and Asians.

[0093] Table 6 Prediction model of NCT efficacy in LAGC patients

[0094]

[0095] 4.3 Model Validation

[0096] The receiver operating characteristic curve (ROC curve) of the NCT efficacy prediction model for LAGC patients in the validation cohort 1 is shown in Figure 4 A, Figure 4 As shown in B, the receiver operating characteristic curve (ROC curve) in the validation cohort 2 is as follows Figure 4 C, Figure 4 D. The results showed that the AUC for evaluating the outcome with the TRG score in validation cohort 1 was 0.760 (95% CI 0.725–0.795), and the AUC for evaluating the outcome with the TRG score in validation cohort 2 was 0.786 (95% CI 0.691–0.880). The validation model performance of the subgroup with TRG as the outcome was better.

[0097] To further evaluate the relationship between patient risk scores and NCT efficacy and prognosis, we used the above optimal cutoff values ​​to classify patients with a total score above 7.75 into a low- to medium-risk group in the validation cohort, and those with a total score below 7.75 into a high-risk group. Subsequently, we plotted ROC curves within these two groups and used the maximum Youden index to determine additional optimal cutoff points. This step helped us further subdivide patients into four risk levels: very high risk (<-1.5), high risk (-1.5 to 7.5), medium risk (8 to 15), and low risk (≥15). In two independent validation cohorts, the distribution of these risk groups was: low risk group (n=198, 18), medium risk group (n=218, 26), high risk group (n=185, 38), and very high risk group (n=152, 18). In both cohorts, risk level was significantly negatively correlated with the efficacy of NCT treatment (R1=0.86, R2=0.98, P<0.001) ( Figure 5 A. Figure 5 B), and the linear trend test showed that the higher the risk level, the greater the proportion of poor efficacy (P<0.05).

[0098] We used the Kaplan-Meier curve to conduct an in-depth analysis of the association between different risk levels and patient survival outcomes. We found that the survival differences between different risk levels were statistically significant (P<0.05). As the patient's risk level of drug resistance increased, the patient's prognosis gradually worsened (P<0.05) (e.g. Figure 6 ).

[0099] 5. Network deployment of prediction models

[0100] Based on the NCT efficacy prediction model, we developed a drug resistance risk calculator. The calculator can evaluate the risk by inputting the patient's clinical indicators and provide corresponding risk levels and recommendations. The purpose of this tool is to help doctors and patients understand the condition more comprehensively and provide accurate support for clinical decision-making ( Figure 7 ).

[0101] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer, characterized in that: The method comprises the following steps: (1) Identify and screen the database to form a derivation cohort; (2) Conduct a systematic review and meta-analysis of the derivation cohort to screen out clinical predictive factors associated with the efficacy of neoadjuvant chemotherapy for advanced gastric cancer; (3) To evaluate the risk of clinical predictive factors and to construct a model for predicting the efficacy of neoadjuvant chemotherapy for advanced gastric cancer; (4) Use a validation cohort to validate the constructed efficacy prediction model; Among them, the clinical predictive factors related to the efficacy of neoadjuvant chemotherapy for advanced gastric cancer are: CEA, tumor location, Lauren classification, differentiation, infiltration depth, lymph nodes, clinical stage, Her-2, and Ki67.

2. The method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 1, characterized in that: The database was identified and screened to include: Articles were identified by database search, with the search terms for Chinese databases being: gastric cancer, gastric malignancies, neoadjuvant chemotherapy, preoperative chemotherapy, efficacy, and effect; and the search terms for English databases being: Cancer of the stomach, neoadjuvant chemotherapy, Neoadjuvant Treatment, and risk factor; Duplicates were removed, and articles meeting the criteria were obtained and assessed using the Newcastle-Ottawa Scale.

3. The method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 1, characterized in that: The contents of the systematic review and meta-analysis include: The following information was extracted from the derivation cohort and entered into the database: details of the study design, region and year of study publication, patient demographics, number of enrolled patients and incidence, study outcomes, identified risk factors and their RRs with corresponding 95% CIs, and identified influencing factors for the number of included studies ≥3; For influencing factors with ≥9 included studies, publication bias analysis was performed using visual funnel plots. If the funnel plot was in an inverted symmetrical shape, it indicated that there was no publication bias; if the funnel plot was asymmetrical, it indicated that there was publication bias. For influencing factors with 3-8 included studies, the Egger linear regression test was performed to assess publication bias. When P < 0.10, it indicated that there was publication bias, and factors related to efficacy were determined. The relative risks (RRs) and 95% confidence intervals (CIs) of efficacy-related factors in each study were extracted, and the combined statistical analysis was performed using StataSE 15.1 software, and forest plots were drawn to present the results; The heterogeneity of the studies was tested by Q test. When the result was P ≥ 0.10 and I 2 When the Q test result was P < 0.10 or I 2 When the heterogeneity was >50%, it indicated that the heterogeneity was statistically significant, and a random effects model was selected for meta-analysis, and sensitivity analysis was performed, and descriptive reports were made for studies with obvious clinical heterogeneity; Finally, among the generated pooled RRs and corresponding 95% CIs, when the 95% CI did not contain 1 or P ≤ 0.05, it indicated that the pooled statistic was statistically significant and that the factor was different between the two groups.

4. The method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 1, characterized in that: The contents of risk scoring of clinical predictive factors and construction of a prediction model for the efficacy of neoadjuvant chemotherapy for advanced gastric cancer include: The combined RRs of all clinical predictive factors and their different stratifications were extracted. The β coefficient of each influencing factor was calculated based on the combined RRs and their corresponding 95% CI. The calculation formula was: β = ln (RR). The approximate number of β multiplied by 10 was taken and one decimal place was retained. The one decimal part was 0-0.2 and retained as 0, 0.3-0.7 was retained as 0.5, and 0.8-0.9 was retained as 1.0, and the score of each influencing factor was obtained. All influencing factors in the prediction model were classified according to the meta-analysis and NCCN guidelines for gastric cancer, and a score was assigned to each category. The total score was obtained by adding the scores of all parameters in the prediction model.

5. The method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 1, characterized in that: The validation cohort includes validation cohort one and validation cohort two. The inclusion criteria are: real-world LAGC patients who underwent NCT and radical surgery, who received at least 2 cycles of NCT before surgery and underwent radical gastrectomy after chemotherapy, excluding patients aged > 75 years, patients who were unable to complete neoadjuvant chemotherapy due to changes in their condition, patients with residual gastric cancer, and patients with other malignant tumors. The validation cohorts were divided into validation cohort one and validation cohort two based on the different onset times.

6. The method for constructing a prediction model for the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 1, characterized in that: The validation of the constructed efficacy prediction model in validation cohorts 1 and 2 included: The generalizability of the prediction model was evaluated using real-world validation cohorts one and two, and the actual effectiveness of NCT treatment for LAGC patients was calculated. Each predictive factor was scored according to the prediction model, and the total score of each patient was calculated. The sensitivity, specificity, and area under the curve (AUC) at different critical values ​​were calculated. The score corresponding to the maximum Youden index was selected to determine the optimal critical value, and the patients were divided into two risk levels: medium-low risk group and high risk group. Then, the ROC curves of the two groups of patients were drawn, and the maximum Youden index of the two groups was selected to divide the patients into four risk levels: relatively low, medium, high, and extremely high risk.

7. A system for predicting the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer, characterized in that: include: Input module: Input the following clinical indicators of the patient: CEA, tumor location, Lauren classification, differentiation, invasion depth, lymph nodes, clinical stage, Her-2, Ki67; Analysis module: provided with a prediction model for the efficacy of neoadjuvant chemotherapy for advanced gastric cancer established by the method as described in any one of claims 1 to 6; Output module: Output the predicted results of the neoadjuvant chemotherapy efficacy for patients with advanced gastric cancer.

8. The system for predicting the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer according to claim 7, characterized in that: The input unit, the analysis unit and the output unit are connected in a wired and / or wireless manner.

Images