A clinical multi-treatment scheme automatic comparison and analysis system and analysis method

Abstract

The application discloses a clinical multi-treatment scheme automatic comparison and analysis system and method, and belongs to the field of clinical data analysis. The system comprises a data input layer, an analysis engine layer and a structured output layer. Through batch construction of candidate treatment scheme pairs, automatic execution of propensity score matching parameter optimization, linkage completion of response rate and survival multi-endpoint analysis, and automatic triggering of report generation and global index clustering based on joint significance judgment rules, the technical problems in the prior art that multi-scheme comparison relies on manual repeated programming, matching analysis and multi-dimensional analysis are solved.

Description

Technical Field

[0001] This invention belongs to the field of clinical data analysis and drug evaluation technology, specifically relating to an automated comparative analysis system and method for multiple clinical treatment options. Background Technology

[0002] In clinical trials and real-world studies, assessing differences in efficacy, survival benefits, and safety among different treatment regimens is a crucial technical step supporting drug screening, clinical decision-making, and cohort studies. Current analytical methods primarily rely on statisticians developing scripts using tools such as SAS and R, or performing single analyses using software like SPSS, Stata, and JMP. While data management platforms offer relatively standardized data collection and quality control capabilities, they typically do not directly provide automated closed-loop analysis for batch comparisons of multiple treatment regimens.

[0003] The existing technology has at least the following problems: First, comparing multiple options relies on repeated manual coding, and the processes of data construction, group comparison, chart output, and report preparation are highly repetitive. Second, the optimization of the propensity score matching parameters is disconnected from subsequent baseline, survival, response rate and subgroup analysis, requiring repeated trial and error and manual connection. Third, the analysis results are scattered across multiple scripts, tables, graphs, and log files, making it difficult to form a unified structured result object and a searchable index; Fourth, for a large number of candidate treatment options, manual comparison of each option is time-consuming, carries a high risk of omissions, and lacks reusability and traceability of results.

[0004] Therefore, it is necessary to provide an automated comparison and analysis system and method for multiple clinical treatment options to improve the efficiency of batch comparison of multiple options, enhance the consistency of the analysis process, and improve the ability to summarize and report results. Summary of the Invention

[0005] The purpose of this invention is to provide an automated comparison and analysis system, method, electronic device, and storage medium for multiple clinical treatment options, in order to solve the technical problems in the prior art where multiple treatment options comparison relies on manual repetitive programming, the propensity score matching and multidimensional analysis processes are disconnected, and the analysis results are difficult to summarize and reuse.

[0006] An automated comparative analysis system for multiple clinical treatment options includes a data input layer, an analysis engine layer, and a structured output layer that work sequentially and collaboratively. The data input layer is used to receive clinical case report data, treatment plan combination data and field mapping rules, perform primary key association, field standardization mapping and data cleaning on multi-source data, and output a standardized analysis data table. The analysis engine layer is used to construct candidate treatment plan pairs in batches based on the standardized analysis data table, and to perform data preprocessing, automatic optimization of propensity score matching, baseline analysis, response rate analysis, survival analysis, subgroup analysis and joint significance determination on each candidate treatment plan pair in sequence, and output structured result objects. The structured output layer is used to receive the structured result object, perform graph standardization, report generation, index summarization and cluster analysis, and output a standardized report file and a global index file.

[0007] Preferably, the analysis engine layer and the structured output layer employ a dual-channel data transfer mechanism that combines real-time transfer of memory objects with persistent transfer of file-based results.

[0008] Preferably, the data input layer includes: The CRF data input submodule is used to read demographic tables, physical examination tables, follow-up record tables, treatment record tables, and efficacy-related tables, and to establish cross-table associations based on the patient's primary key; The protocol combination input submodule is used to read the protocol combination file, divide different analysis populations according to the worksheet, and define a set of candidate treatment protocol pairs for each row of records; The standardization mapping and cleaning submodule is used to map source fields to internal standard fields and perform missing value handling, valid value validation, grouping validity check, sample size threshold check, and time logic validation.

[0009] Preferably, the analysis engine layer includes: The batch scheduling module is used to build the dataset to be analyzed item by item according to the preset population and candidate treatment plan and trigger the analysis task. The data preprocessing module is used to remove missing samples from treatment groups, force treatment variables to be binary, and remove missing samples from covariates. The propensity score matching automatic optimization module is used to perform batch propensity score matching on multiple combinations of caliper values ​​and matching ratios and automatically filter the matching results. The baseline analysis module is used to output the baseline characteristic comparison results between treatment groups; The reaction rate analysis module is used to output reaction rate comparison results; The survival analysis module is used to output the results of survival endpoint comparisons; The subgroup analysis module is used to output the effect comparison results under different stratified variables; The significance determination module is used to output an overall significance indicator based on the reaction rate analysis results and the survival analysis results.

[0010] Preferably, the propensity score matching automatic optimization module performs the following operations: Construct a parameter combination mesh with multiple caliper values ​​and multiple matching ratios; Perform propensity score matching for each parameter combination separately; Calculate the overall standardized difference, the maximum standardized difference, and the percentage of threshold satisfaction for each matching result; Prioritize retaining matching results that meet strict balance criteria; When no matching results satisfy the strict balance criterion are found, the backsliding screening criteria are used to sort and select the best matching results. When no matching results meet the rollback filtering criteria, the original dataset is used as the rollback analysis set.

[0011] Preferably, the significance determination module adopts a joint significance determination rule: When the response rate analysis result is significant, or the result of at least one survival endpoint analysis is significant, the current candidate treatment regimen is determined to be significant for the overall population. When the current candidate treatment is determined to be significant overall, a full report is generated and an index is output. When the current candidate treatment is not determined to be significant overall, only result storage and index appending are performed.

[0012] Preferably, the structured output layer includes: The graphics standardization submodule is used to output survival curves, forest plots, or other statistical charts as standard charts of uniform size and arrange them according to preset layout rules. The report generation submodule is used to dynamically generate intermediate report fragments based on chapter templates and analysis results, and then compile them into a final report document; The global indexing and clustering submodule is used to summarize the significance results, sample information, statistics and matching information of multiple candidate treatment pairs, generate a global index file, and perform clustering analysis based on patient coverage patterns.

[0013] Preferably, the analysis engine layer further includes an optional safety analysis submodule, used to perform hierarchical statistics and significance tests on adverse events in the two treatment groups, and output a structured safety statistics table and text description; the safety analysis submodule can run independently or be connected to the batch scheduling module, and used together with the response rate analysis results and survival analysis results for the generation of a comprehensive report.

[0014] The present invention also provides an automated comparative analysis method for multiple clinical treatment options based on the system, comprising the following steps: Step S1: Receive clinical case report data and protocol combination files, associate multi-source data based on patient primary keys, perform field standardization mapping and data cleaning, and output standardized analysis data tables; Step S2: Analyze the candidate treatment plan pairs item by item according to the worksheets and row records in the scheme combination file to construct the dataset to be compared; Step S3: Perform data preprocessing, propensity score matching automatic optimization, baseline analysis, response rate analysis, survival analysis and subgroup analysis on the verified dataset to be compared in sequence to generate structured result objects; Step S4: Perform joint significance determination based on the reaction rate analysis results and survival analysis results, and output the overall significance indicator; Step S5: Store the structured result object as an intermediate result file, trigger graphical standardization and report generation for candidate treatment pairs that meet the overall significance condition, and index, summarize and cluster the results of multiple candidate treatment pairs.

[0015] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the above-described method for automatic comparison and analysis of multiple clinical treatment options.

[0016] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the above-described method for automatic comparison and analysis of multiple clinical treatment options is implemented.

[0017] The advantages and effects of this application are as follows: 1. The clinical multi-treatment option automatic comparison and analysis system designed in this application can automatically parse multiple populations and multiple treatment option pairs in the option combination file by batch building candidate treatment option pairs and batch scheduling analysis tasks. It can build the dataset to be analyzed item by item and trigger pipeline analysis. It eliminates the need for repeated manual coding and manual data splitting, which greatly reduces the human cost of repeated team building and repeated analysis, and at the same time reduces errors caused by human operation.

[0018] 2. The clinical multi-treatment option automatic comparison and analysis system designed in this application automatically constructs a parameter combination grid of caliper value and matching ratio through multi-parameter propensity score matching automatic optimization technology. It performs batch matching and calculates three balance indicators: overall standardized difference, maximum standardized difference, and threshold satisfaction ratio. Then, it selects the optimal matching results according to the hierarchical rule of "strict standard priority - backtracking standard sorting - original dataset as a fallback". This replaces the process of manual trial and error parameter tuning, reduces the workload of manual parameter tuning by more than 90%, and improves the scientificity and objectivity of the matching result selection.

[0019] 3. The clinical multi-treatment regimen automatic comparison and analysis system designed in this application eliminates the manual connection points between different analysis links by incorporating data preprocessing, propensity score matching, baseline analysis, response rate analysis, survival analysis, subgroup analysis and joint significance determination into an end-to-end unified automated process. This ensures that all candidate treatment regimens are analyzed using completely consistent methods and judgment criteria, fundamentally solving the problem of inconsistent analysis standards in multi-treatment regimen comparison.

[0020] 4. The clinical multi-treatment regimen automatic comparison and analysis system designed in this application automatically identifies clinically significant treatment regimen pairs by means of joint significance determination based on response rate and survival multiple endpoints and differential output. It generates a complete report only for regimen pairs determined to be overall significant, and only stores the core results for regimen pairs that are not significant, which greatly reduces meaningless redundant output and saves computing and storage resources.

[0021] 5. The clinical multi-treatment option automatic comparison and analysis system designed in this application integrates scattered scripts, tables and graphs into a unified structured result through unified structured result objects, standardized graphical report generation and global index clustering based on patient coverage patterns. It automatically generates output files with standardized formats and helps researchers quickly discover similar treatment options through cluster analysis, which significantly enhances the traceability, reusability and retrieval efficiency of the results.

[0022] 6. The clinical multi-treatment option automatic comparison and analysis system designed in this application uses a three-layer modular architecture and optional functional module plug-and-play design to support the independent operation of the security analysis sub-module or its integration into the main process. It is also compatible with various deployment methods such as local terminals, servers, and containerized clusters, and has good engineering scalability.

[0023] By employing a multi-parameter propensity score matching automatic optimization mechanism, the workload of manual parameter tuning is reduced, and the efficiency of matching result screening is improved. Baseline analysis, response rate analysis, survival analysis, subgroup analysis, and significance determination are incorporated into a unified process to improve the consistency of analysis criteria. At the same time, this application automatically generates graphs, reports, and global indexes through a structured output mechanism, enhancing the reusability, traceability, and retrieval efficiency of results. It can be extended to implement methods such as safety analysis, containerized deployment, and service-oriented scheduling as needed, possessing good engineering scalability. This enables batch automatic comparison of multiple treatment regimen combinations, reducing the cost of repeated team building and analysis.

[0024] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0025] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0027] Figure 1 This is a schematic diagram illustrating the CRF form standardization conversion principle of this invention. Figure 2 This is a system module architecture diagram of the present invention; Figure 3 This is the main flowchart of the present invention; Figure 4 This is a flowchart illustrating the batch processing logic for propensity score matching in this invention. Figure 5 Generate a pipeline diagram for the report of this invention; Figure 6 This is a schematic diagram of the global index clustering of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0029] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0030] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0031] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0032] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0033] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0034] Example 1: This example mainly introduces the system deployment and overall architecture.

[0035] Please see Figure 2 The system of the present invention includes a data input layer, an analysis engine layer, and a structured output layer. Figure 2 The system demonstrates the modular composition and interconnections of the automated clinical multi-treatment option comparison and analysis system of the present invention. The system consists of a data input layer, an analysis engine layer, and a structured output layer that work together in sequence. Each module performs data flow and linkage through standardized data structures and structured result objects.

[0036] CRF / Excel data input: Used to receive clinical data sources, including demographic, physical examination, follow-up, treatment records, efficacy / relapse / survival, etc. Excel or CRF exported data.

[0037] Medication regimen combination input: used to receive definitions of candidate treatment regimen pairs, such as dividing the population by worksheet, defining "Plan A vs Plan B" by row, and the corresponding patient set.

[0038] Standardization Mapping and Cleaning Module: Performs unified processing on the inputs from the "CRF / Excel Data Input" and "Drug Prescription Combination Input" modules, including at least primary key association, field standardization mapping, missing value and validity handling, and outputs a standardized analysis data structure that can be directly used for subsequent analysis.

[0039] Batch scheduling module: used to build the dataset to be analyzed item by item according to the preset population and candidate schemes and trigger the task, reflecting the system's batch processing capability.

[0040] Data preprocessing module: Used to perform basic validation and preprocessing on each dataset to be analyzed, including handling missing groupings, binary classification constraints, sample size / event number threshold checks, and handling missing covariates.

[0041] The PSM automatic optimization module is used to perform automatic optimization of propensity score matching when conditions are met. It typically forms a candidate matching set by traversing multiple caliper values ​​and matching ratio combinations, and selects the optimal matching dataset based on balance indicators; at the same time, the original dataset is retained for comparison.

[0042] Baseline analysis module: Used to output the baseline characteristics comparison results between treatment groups and identify significant imbalance variables to interpret the reliability and matching effect of subsequent results.

[0043] Reaction rate analysis module: Used to output intergroup comparison results and statistical test results for reaction rate-related endpoints.

[0044] Survival analysis module: used to output the results of intergroup comparisons of survival endpoints (such as survival curve difference tests, risk estimates, etc.).

[0045] Subgroup analysis module: Used to output subgroup effect comparison and interaction test results according to preset stratification variables.

[0046] Optional security analysis module: Used to output security-related comparison results such as adverse events; this module is optional and can run independently or be integrated into the main process.

[0047] Structured Result Object / Intermediate Result Storage Module: This module receives the outputs from various modules, forms a unified structured result object, and persists it (e.g., serializes it into an intermediate result file) to support reuse, traceability, and batch aggregation.

[0048] Graphics Standardization Module: This module outputs survival curves, forest plots, and other graphics according to standardized sizes, fonts, and layouts. It also allows for layout and format conversion to meet reporting and archiving requirements.

[0049] Report generation module: Used to automatically generate and output the structured results in C1 as a report document based on the chapter template and document style template.

[0050] Global Index and Clustering Module: This module is used to summarize the key statistical fields and significance status of multiple candidate protocol pairs to form a global index, and to perform clustering analysis based on information such as patient coverage patterns to support protocol selection, result deduplication, and topic summarization.

[0051] The system is deployed on a local analysis terminal or research server and executes in batch mode. Modules collaborate through a shared file system and a unified structured result object. In other embodiments, modules can be encapsulated as independent services, deployed in a containerized computing environment or server cluster, and communicate via shared storage, task queues, or program interfaces. The system employs a dual-channel data transfer mechanism combining real-time transfer of in-memory objects and persistent transfer of file-based results: in-memory objects are serialized in JSON format with a transmission latency of less than 10ms; file-based results store structured data in CSV format and image files in PNG format. All files are organized according to a directory structure of "analysis batch - scheme pair number" to ensure traceability.

[0052] Please refer to Figure 3 , Figure 3 This demonstrates the main workflow of the present invention for automatically comparing and analyzing multiple candidate treatment pairs in a batch processing scenario. The system uses "candidate treatment pairs" as the smallest processing unit to perform data construction, condition validation, core analysis, result storage, and report triggering in a loop. After all treatment pairs have been processed, significant treatments are summarized, represented in a matrix, and clustered, and finally a global index file is output.

[0053] Read protocol combination file: Used to read input files containing definitions of candidate treatment protocol pairs. These files can be divided into different analysis populations or cohorts by worksheet, serving as the source of batch comparison tasks.

[0054] Parse candidate treatment pair: Used to parse "Plan A vs Plan B" item by item according to the worksheet and row records, extract the names of the two plans and their corresponding patient identifier sets, and construct a binary grouped candidate comparison unit.

[0055] Associated standardized clinical fields: used to associate candidate comparison units with standardized clinical fields based on the patient's primary key to form a dataset to be analyzed; the standardized clinical fields include at least baseline covariates, response rate-related fields, and survival status and survival time fields.

[0056] Whether the analysis conditions are met: This is used to perform admission checks on the current dataset to be analyzed, including at least group validity checks (two groups must exist and be identifiable), binary classification constraints, sample size threshold or event number threshold checks, and key field availability checks. If the analysis conditions are not met, the system skips the current scheme pair and proceeds to the next scheme pair for processing.

[0057] Execute the core analysis process: This process is used to perform automated analysis on the dataset to be analyzed that has passed the admission verification. The process includes at least data preprocessing, automatic optimization of propensity score matching (if applicable), baseline analysis, response rate analysis, survival analysis, and subgroup analysis.

[0058] Generate structured result objects: These are used to encapsulate the output of the core analysis process into result objects with a unified structure. The result objects include statistics, graphical objects or graphical paths, text descriptions, and parameter information to support reuse, traceability, and subsequent output.

[0059] Append running index: Used to incrementally write key fields and significance status information of the current scheme pair into the running index so as to continuously accumulate global summary information during batch processing.

[0060] Overall significance: This is used to perform joint significance determination based on the results of response rate analysis and survival analysis, and output an overall significance indicator; when the overall significance is significant, it enters the report generation branch; when the overall significance is not significant, it does not trigger the generation of a complete report but still retains the structured results and index records.

[0061] Report generation and standardized plots: This feature is used to automatically generate reports and plots for overall significant scenarios, including generating chapter documents based on templates, outputting and standardizing statistical plots such as survival curves / forest plots, and compiling chapter content and plots into a final report output.

[0062] Loop through the next candidate pair: This function returns to the candidate pair parsing and analysis process, continuing to process the next candidate pair in batch order until all candidate pairs have been processed.

[0063] Filter all significant schemes: After batch processing, select the set of candidate scheme pairs that are significant overall from the cumulative index or intermediate results, and use them as input for global summary and cluster analysis.

[0064] Constructing a patient-scheme binary matrix: This is used to construct a binary matrix based on the set of patients covered by each significant scheme pair. The matrix elements are used to indicate whether a scheme pair covers a certain patient, thereby providing a structured representation of the "patient coverage pattern" of significant schemes.

[0065] Perform Jaccard distance calculation and hierarchical clustering: This is used to calculate the Jaccard binary distance or equivalent set distance index based on the binary matrix, and perform hierarchical clustering to group salient schemes according to the similarity of coverage patterns, and output the cluster structure or cluster labels.

[0066] Output global index file: This is used to merge the cluster labels and statistical fields of each significant scheme pair to form a global index file, which serves as a structured entry point for scheme selection, result deduplication, and topic summarization.

[0067] Example 2: Based on Example 1, this example mainly introduces data access and standardization.

[0068] Please see Figure 1 This embodiment mainly details the working principle of the data input layer. It explains how the system integrates multi-source CRF / Excel input tables (demographic, physical examination, follow-up, treatment records, efficacy, relapse, survival, etc.) with protocol combination inputs, converting them into an "internal standard analysis data table" that can be directly consumed by the subsequent analysis engine.

[0069] CRF / Excel input table: Represents raw or semi-structured clinical data tables from different sources, containing at least a patient primary key field (such as SUBJID) and related clinical fields.

[0070] Primary key association and field standardization: This refers to the process of unifying primary keys and mapping fields across multiple tables. Using SUBJID as the unique identifier for each patient, multiple source tables are associated and merged. Source fields are mapped to internal standard fields (e.g., AGE→Age, SEX→Sex, ETHNIC→Ethnic, ECOGPS→Ecog). Efficacy classification or BOR information is converted into internal efficacy fields (e.g., fields / indicators required to derive ORR). Survival fields (OS / OS_time) can be derived from follow-up and treatment start dates or imported from external outcome tables. PFS / PFS_time can be imported from relapse-related tables.

[0071] Protocol combination input: This represents the candidate comparison definition input for "Protocol A vs Protocol B". Typically, each line records a candidate protocol pair and contains two sets of patient IDs (separated by semicolons).

[0072] Grouping Parsing: This means parsing the combined input schemes into the system's internal grouping field Group, forming a binary grouping comparison data structure (Group=A / B).

[0073] Standardized quality control: This refers to the consistency and validity verification of the standardized candidate dataset, mainly including: missing value handling (removal or imputation rules); group validity (two groups must exist and the groups must be identifiable); binary classification constraint (forcing only two treatment groups to be retained for comparison); sample size thresholds (minimum number of cases per group, event number thresholds, etc.); and survival time logic verification (time must be non-negative, event date must not be earlier than the treatment start date, etc.).

[0074] Internal Standard Analysis Data Table: The output is a unified "patient-level analysis data table" with standardized field formats and naming conventions. It serves as the sole entry data structure for subsequent batch analyses (including ID, Group, baseline covariates, ORR / pCR, survival status and time, etc.).

[0075] The system receives clinical data from multiple sources and associates them based on patient primary keys.

[0076] Input data includes demographic forms, physical examination forms, follow-up record forms, treatment record forms, and forms related to efficacy or recurrence.

[0077] In the demographic table, fields such as age, sex, and ethnicity are mapped to the internal standard fields Age, Sex, and Ethnic; the physical fitness score field in the physical examination table is mapped to Ecog; and the survival status, date of death, and date of last follow-up in the follow-up record are combined with the treatment start date to derive the survival status and survival time fields.

[0078] The system also reads a treatment combination file. This file is divided into worksheets for different analysis populations, with each row defining a pair of candidate treatment plans. Each candidate treatment plan pair includes at least a plan name and a corresponding set of patient identifiers. The system then splits the patient identifier set to construct a patient-by-patient comparison dataset.

[0079] During the standardization mapping and cleaning process, the system executes the following rules: First, the patient's primary key is kept in string format throughout the entire process; Second, skip the format description lines in the source table according to preset rules; Third, when there are multiple visit records for the same patient, they are merged according to the preset visit rules; Fourth, remove missing samples from the treatment group and force the treatment variable to be binary; Fifth, remove samples with missing covariates and fill in predefined labels for missing character values; Sixth, verify the reaction rate range, time range, and date sequence logic; Seventh, when the sample size or number of events does not meet the preset conditions, the subsequent processing of the current candidate treatment pair is terminated.

[0080] Example 3: Based on Examples 1-2, this example mainly introduces the construction and scheduling process of batch tasks.

[0081] The system parses the treatment plan combination file item by item according to the worksheet and row records, constructing multiple candidate treatment plan pairs. Each candidate treatment plan pair is converted into a binary dataset to be analyzed, carrying the corresponding population identifier, treatment plan identifier, and patient set information. The batch scheduling module triggers subsequent analyses in a preset order, forming a batch processing pipeline.

[0082] Example 4, please refer to Figure 4 This embodiment details the working logic of the tendency score matching automatic optimization module. Figure 4 This demonstrates the batch processing and decision-making flow of the Propensity Score Matching (PSM) automatic optimization module of this invention. Instead of performing a single match with a single parameter, the system constructs a parameter combination grid for multiple clamping values ​​and matching ratios. Matching is performed on each parameter combination separately, and a balance index is calculated. Then, following a hierarchical strategy of "strict screening—backward screening—original data backward," the system automatically selects the optimal matching dataset, while retaining the original dataset for comparative analysis and verification.

[0083] Input raw two sets of comparison data: This is used to receive the dataset to be compared corresponding to the current candidate treatment plan. The dataset contains at least treatment grouping variables and preset covariates, and may also contain fields required for subsequent response rate and survival endpoint analysis.

[0084] Generate a grid of caliper value × matching ratio parameters: This is used to preset multiple caliper values ​​and multiple matching ratios, and construct a parameter combination grid as a set of candidate configurations for batch matching.

[0085] Propensity score matching is performed on a parameter combination basis: This is used to perform propensity score matching for each set of configurations in the parameter combination grid, generating several candidate matching datasets to cover the possibilities of different matching strengths and matching scales.

[0086] Calculate balance metrics: Used to calculate balance evaluation metrics for each candidate matching dataset. The metrics include at least the overall standardized difference, the maximum standardized difference, and the threshold satisfaction rate, which are used to quantify the degree of balance between the two sets of baseline covariates after matching.

[0087] Does a candidate matching set exist that meets strict criteria? This is used to determine whether there are matching results in the candidate matching dataset that meet strict balance requirements. Strict criteria are used to prioritize matching quality and avoid low-quality matching from affecting the reliability of subsequent conclusions.

[0088] Screening of a strictly qualified candidate set: This is used to select a set of matching results that meet strict criteria from all candidate matching results, and use it as a priority candidate set to proceed to the next step of optimization.

[0089] Sort by balance index for optimal matching: This method sorts the qualified candidate dataset according to the balance index and selects the matching dataset with the best overall balance as the final optimal matching dataset.

[0090] Does a candidate matching set exist that meets the rollback criteria? This is used to determine whether there are matching results that meet the rollback criteria when there are no results after strict screening. The rollback criteria are used to improve the matching success rate and process continuity while ensuring a certain balance.

[0091] Filtering and sorting qualified candidate sets for backoff: This is used to filter the set of matching results that meet the backoff criteria from the candidate matching results, and select the optimal matching dataset after sorting by the balance index.

[0092] Using the original dataset as the fallback analysis set: When no matching results are available in both strict filtering and fallback filtering, the original dataset is used as the fallback analysis set to enter the subsequent analysis process, thereby avoiding process interruption due to PSM failure.

[0093] Simultaneously output the optimal matching dataset and the original dataset: This is used to output the final selected optimal matching dataset and the original dataset at the same time and deliver them to the downstream analysis module for use; the optimal matching dataset is used to improve the comparability between groups, while the original dataset is used for result comparison, sensitivity analysis, and traceability retention.

[0094] For each dataset to be analyzed, the system presets multiple caliper values ​​and multiple matching ratios, forming a parameter combination grid. For each parameter combination, propensity score matching is performed, and the overall standardized difference, maximum standardized difference, and threshold satisfaction ratio are calculated. The strict balance criterion is: the maximum standardized difference is less than 0.1 and the threshold satisfaction ratio is greater than 90%; the backoff screening criterion is: sorted by overall standardized difference from smallest to largest, and the matching result with the smallest overall standardized difference is selected.

[0095] The system prioritizes retaining matches that meet strict balance criteria; if none exist, it uses a fallback selection criterion to rank and select the best matches from the candidate results; if no usable results are found, the original dataset is used as the fallback analysis set. Finally, the system retains both the original dataset and the optimized matching dataset for subsequent baseline analysis, survival analysis, response rate analysis, and subgroup analysis.

[0096] Example 5: Based on the above examples, this example mainly introduces core analysis and joint significance determination.

[0097] For the dataset to be analyzed that has passed the validation, the system performs the following steps in sequence: data preprocessing; automatic optimization of propensity score matching; baseline analysis; response rate analysis; survival analysis; and subgroup analysis.

[0098] The baseline analysis module outputs baseline differences between treatment groups and identifies significant imbalance variables. The response rate analysis module outputs inter-group comparisons of objective response rate or other efficacy endpoints. The survival analysis module outputs comparisons of survival curves and risk estimates. The subgroup analysis module outputs effect differences and interaction tests under different stratification conditions.

[0099] The system further performs a joint significance determination: when the response rate analysis result is significant, or at least one survival endpoint analysis result is significant, the current candidate treatment pair is determined to be overall significant. This determination result is used to drive subsequent report generation and index output. When the p-value of the response rate analysis result is less than 0.05, or the p-value of at least one survival endpoint analysis result is less than 0.05, the current candidate treatment pair is determined to be overall significant.

[0100] Example 6: Based on the above examples, this example mainly introduces security analysis.

[0101] The system includes a safety analysis submodule. This submodule performs hierarchical statistical analysis and significance testing on adverse events in the two treatment groups, and outputs a structured safety statistics table and text description. The safety analysis can run independently or be integrated into the main analysis workflow to form a comprehensive evaluation result together with response rate analysis and survival analysis. Adverse events are stratified according to systemic organ classification (SOC) and preferred terminology (PT), and the severity of adverse events is graded and coded according to CTCAE standards. The chi-square test is used to compare the differences in the incidence of adverse events between the two groups, and Fisher's exact test is used to compare the differences in the incidence of adverse events in small sample sizes. The Wilcoxon rank-sum test or Cochran-Armitage trend test is used to analyze the differences between groups in severity level distribution. Serious adverse events (SAEs) and adverse events leading to drug discontinuation or death are summarized and descriptively statistically analyzed separately.

[0102] Example 7, please refer to Figure 5 Based on the above embodiments, this embodiment mainly introduces structured output and report generation. Figure 5 This demonstrates the pipelined processing of "report generation" in the structured output layer of this invention. The system takes structured result objects or intermediate result files as input, and processes and standardizes statistical conclusions, tabular information, and graphical outputs through templates, ultimately generating a report document in a unified format via the report assembly module. Image format conversion is an optional post-processing step to meet different delivery requirements such as submission and archiving.

[0103] Structured result object / intermediate result file: used to provide all the input information required for report generation. The information includes at least the statistics output by each analysis module, significance determination results, tabular data, graphical objects or graphical paths, and necessary parameters and descriptive text. In one embodiment, the results are persistently stored in the form of an intermediate result file to support reuse and traceability.

[0104] Chapter templates: These are used to pre-define the chapter structure and content framework of the report. The chapter templates include at least the baseline, response rate, PFS, OS, subgroup, and methodology chapter templates to ensure that different schemes maintain a consistent report structure and format.

[0105] Template Variable Replacement Module: This module is used to inject statistical conclusions, numerical fields, table content, and graph paths from the structured results object as variables into the chapter template, and replace the placeholders in the template to generate chapter content that is consistent with the analysis results of the current scheme.

[0106] Generate chapter document fragments: This is used to generate chapter-level document fragments after template variable replacement is completed. These document fragments serve as input to the report assembly module and are used for subsequent merging of multiple chapters and unified formatting.

[0107] Statistical graph generation module: used to generate statistical graphs based on the graph parameters or graph objects recorded in the structured result object. The statistical graphs include at least survival curves, forest plots, etc., to visually display differences between groups and subgroup effects.

[0108] The plot standardization and layout module is used to standardize the output of generated statistical plots, including at least uniform size, font and layout, and to arrange multiple plots according to preset layout rules to meet the requirements of report layout specifications and output consistency.

[0109] Optional image format conversion module: This module is used to further convert standardized images into image files of preset resolution and format when needed, such as standard resolution images for journal submission or archiving; this module is optional and does not affect the main report generation process.

[0110] Document style templates: These provide a unified style specification for report documents, including heading levels, paragraph formatting, table styles, and image styles, to ensure that different solutions maintain consistency in the visual style and format requirements of the output reports.

[0111] The report compilation module is used to compile chapter document fragments, standardized figures, and document style templates, complete the combination of methodology chapters and results chapters, chapter sorting and unified layout, and output a single complete report document.

[0112] Final report document output: Used to output the final report deliverable to the user. The report document includes at least baseline, response rate, survival, subgroup and other analytical conclusions and corresponding charts, and has a reusable and traceable output structure.

[0113] The system persists the structured results objects from the core analysis output as intermediate result files. For candidate treatment pairs that are significant overall, the structured output layer performs the following operations: First, generate survival curves, forest maps, and other statistical graphs of uniform size; Second, the drawings are assembled according to the preset layout rules; Third, based on the chapter template, inject statistical results, table content and graph paths to generate chapter report fragments; Fourth, the methodology and results sections are compiled to generate the final report document.

[0114] In one embodiment, the image may be further converted into a standard resolution image file to meet the requirements of a paper, application, or archive.

[0115] The report generation submodule employs a templated chapter generation and dynamic compilation mechanism for the overall report, and includes at least the following: Call the corresponding chapter template according to different analysis types; The statistical conclusions, table information, and graphic paths are injected into the chapter template to generate the chapter document; The methodological content, results section content, and document style information are compiled and output into a single report document. The chapter template includes at least a methodological chapter, a baseline analysis chapter, a response rate analysis chapter, a survival analysis chapter, and a subgroup analysis chapter. The methodological chapter further includes a description of the treatment plan, definitions of measurement indicators, propensity score matching methods, and statistical analysis methods. The baseline analysis chapter further includes pre-match baseline comparisons and post-match baseline comparisons. The survival analysis chapter further includes progression-free survival analysis and overall survival analysis. The subgroup analysis chapter further includes subgroup analysis of survival endpoints and subgroup analysis of response rates. Optionally, the chapter template also includes a safety analysis chapter for adverse event grading statistics and intergroup comparisons. The report generation submodule injects statistical conclusions, table information, and graphic paths into the corresponding chapter template to generate the chapter document, and then automatically compiles it into a complete report document including a cover, table of contents, main text, and appendices.

[0116] Example 8, see Figure 6 Based on the above embodiments, this embodiment mainly introduces the implementation method of global indexing and clustering. Figure 6The global indexing and clustering process of this invention is demonstrated. After batch processing, the system summarizes all candidate solution pairs, first screening the candidate solution pairs that are generally significant, then constructing a binary matrix based on the patient coverage pattern and calculating the set distance, and then performing hierarchical clustering and optimization and verification of cluster partitioning, finally generating a global index file with cluster labels for solution screening, result deduplication, and topic summarization.

[0117] Screening candidate solution pairs that are significant overall: This is used to screen a set of candidate solution pairs that meet the overall significance judgment rules from the cumulative index or structured result objects, forming the target solution set for global cluster analysis.

[0118] Read the patient set corresponding to each significant protocol: This is used to read the set of patient identifiers included or covered by each overall significant candidate protocol pair to reflect the actual population coverage of the protocol pair.

[0119] Construct a patient-proposal binary matrix: This is used to structurally represent the "proposal pair-patient coverage relationship" as a binary matrix, where the matrix rows correspond to the overall significant candidate proposal pairs, the matrix columns correspond to the patient identifiers, and the matrix elements are used to indicate whether a proposal pair covers a certain patient, thereby mapping the coverage patterns of different proposal pairs to a computable matrix structure.

[0120] Calculate the Jaccard binary distance or equivalent set distance index: used to calculate the similarity or distance between any two scheme pairs based on the binary matrix. The Jaccard binary distance is used to characterize the degree of overlap between the patient sets of the two scheme pairs; in other embodiments, a distance index equivalent to the set overlap can also be used.

[0121] Perform hierarchical clustering (e.g., Ward.D2): This is used to perform hierarchical clustering based on the distance matrix mentioned above, to obtain the clustering structure or dendrogram between the protocol pairs, thereby merging protocol pairs with similar patient coverage patterns into the same cluster.

[0122] Cluster partitioning optimization and validation (profile coefficient search / permutation test): This is used to optimize and validate the segmentation method of hierarchical clustering results and determine the better cluster partitioning. The profile coefficient is used to evaluate the overall consistency of different cluster partitioning, and the permutation test is used to evaluate the statistical significance of differences between clusters, thereby improving the stability and interpretability of clustering results.

[0123] Generate cluster labels: These are used to assign cluster labels (e.g., res1, res2, etc.) to each overall significant candidate scheme pair based on the final determined clustering, so that they can be grouped, retrieved, and summarized in the index.

[0124] Merging cluster labels and statistical fields: This is used to merge cluster labels and statistical fields of each candidate solution pair to form a structured index record. The statistical fields include at least the solution identifier, the population to which it belongs, whether matching was used, sample size, number of events, and statistical and significance information of response rate / survival endpoint.

[0125] Output global index file: Used to output a global index file containing cluster labels and statistical fields, which serves as the system's structured master table entry point to support rapid filtering of significant schemes, merging of duplicate results, and thematic analysis summaries.

[0126] During batch processing, the system cumulatively records the significance status, sample information, matching information, and statistics of each candidate treatment pair. After batch processing is complete, it filters out the overall significant candidate treatment pairs from the cumulative index and constructs a patient-treatment binary matrix based on the covered patient set.

[0127] The system calculates the Jaccard binary distance or equivalent set distance index based on the binary matrix and performs hierarchical clustering. It then merges the cluster labels with the statistical fields of each candidate treatment pair and outputs a global index file. This global index file can be used for clue screening, result deduplication, thematic summarization, and subsequent manual review.

[0128] Example 9: Based on the above examples, this example mainly introduces application examples.

[0129] In one specific application embodiment, the system performs batch comparisons of immune checkpoint inhibitor combined with platinum-based chemotherapy regimens versus platinum-based chemotherapy alone in a lung cancer cohort. The system first reads the medication combination files, constructs comparison datasets for the combination regimen group and the chemotherapy-only group, and associates fields such as age, gender, performance status score, ethnicity, objective response rate, overall survival, and progression-free survival. After automatic optimization through propensity score matching, the baselines of the two groups are balanced.

[0130] Response rate analysis showed that the objective response rate was 46.4% in the combination therapy group and 27.3% in the chemotherapy-only group, with a statistically significant difference; the disease control rates were 80.4% and 56.4%, respectively. Survival analysis showed no significant differences in progression-free survival and overall survival between the two groups. Subgroup analyses were performed based on age, sex, and performance status score; interaction tests revealed no significant interaction.

[0131] Based on this, the system determines the scheme pair as overall significant and triggers the generation of a complete report, outputting a report document that includes a baseline statistics table, response rate table, survival curve, subgroup forest plot, and methodology section, while also including the scheme pair in the global index.

[0132] The above description is merely a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter alterations to these embodiments within the spirit and principles of the present invention, achieved through conventional substitutions or by achieving the same function without departing from the principles and spirit of the present invention, fall within the scope of protection of the present invention.

Claims

1. An automated comparative analysis system for multiple clinical treatment options, characterized in that, It includes a data input layer, an analysis engine layer, and a structured output layer that work in sequence. The data input layer is used to receive clinical case report data, treatment plan combination data and field mapping rules, perform primary key association, field standardization mapping and data cleaning on multi-source data, and output a standardized analysis data table. The analysis engine layer is used to construct candidate treatment plan pairs in batches based on the standardized analysis data table, and to perform data preprocessing, automatic optimization of propensity score matching, baseline analysis, response rate analysis, survival analysis, subgroup analysis and joint significance determination on each candidate treatment plan pair in sequence, and output structured result objects. The structured output layer is used to receive the structured result object, perform graph standardization, report generation, index summarization and cluster analysis, and output a standardized report file and a global index file; The data input layer includes: The CRF data input submodule is used to read demographic tables, physical examination tables, follow-up record tables, treatment record tables, and efficacy-related tables, and to establish cross-table associations based on the patient's primary key; The protocol combination input submodule is used to read the protocol combination file, divide different analysis populations according to the worksheet, and define a set of candidate treatment protocol pairs for each row of records; The standardization mapping and cleaning submodule is used to map source fields to internal standard fields and perform missing value handling, valid value validation, grouping validity check, sample size threshold check, and time logic validation. The analysis engine layer includes: The batch scheduling module is used to build the dataset to be analyzed item by item according to the preset population and candidate treatment plan and trigger the analysis task. The data preprocessing module is used to remove missing samples from treatment groups, force treatment variables to be binary, and remove missing samples from covariates. The propensity score matching automatic optimization module is used to perform batch propensity score matching on multiple combinations of caliper values ​​and matching ratios and automatically filter the matching results. The baseline analysis module is used to output the baseline characteristic comparison results between treatment groups; The reaction rate analysis module is used to output reaction rate comparison results; The survival analysis module is used to output the results of survival endpoint comparisons; The subgroup analysis module is used to output the effect comparison results under different stratified variables; The significance determination module is used to output an overall significance indicator based on the response rate analysis results and the survival analysis results; The propensity score matching automatic optimization module performs the following operations: Construct a parameter combination mesh with multiple caliper values ​​and multiple matching ratios; Perform propensity score matching for each parameter combination separately; Calculate the overall standardized difference, the maximum standardized difference, and the percentage of threshold satisfaction for each matching result; Prioritize retaining matching results that meet strict balance criteria; When no matching results satisfy the strict balance criterion are found, the backsliding screening criteria are used to sort and select the best matching results. When no matching results meet the rollback filtering criteria, the original dataset is used as the rollback analysis set; The significance determination module adopts a joint significance determination rule: When the response rate analysis result is significant, or the result of at least one survival endpoint analysis is significant, the current candidate treatment regimen is determined to be significant for the overall population. When the current candidate treatment is determined to be significant overall, a full report is generated and an index is output. When the current candidate treatment is not determined to be overall significant, only result storage and index appending are performed; The structured output layer includes: The graphics standardization submodule is used to output survival curves and forest maps as standard graphics of uniform size and to arrange them according to preset layout rules. The report generation submodule is used to dynamically generate intermediate report fragments based on chapter templates and analysis results, and then compile them into a final report document; The global indexing and clustering submodule is used to summarize the significance results, sample information, statistics and matching information of multiple candidate treatment pairs, generate a global index file, and perform clustering analysis based on patient coverage patterns. The analysis engine layer also includes an optional safety analysis submodule, which is used to perform hierarchical statistics and significance tests on adverse events in the two treatment groups, and output a structured safety statistics table and text description; the safety analysis submodule can run independently or be connected to the batch scheduling module, and is used together with the response rate analysis results and survival analysis results to generate a comprehensive report; The strict balance standard is: the maximum standardized difference is less than 0.1 and the threshold satisfaction rate is greater than 90%; the backoff screening standard is: sorted by overall standardized difference from smallest to largest, and the matching result with the smallest overall standardized difference is selected. The system employs a dual-channel data transfer mechanism that combines real-time transfer of in-memory objects with persistent transfer of file-based results between its various layers: in-memory objects are serialized in JSON format with a transmission latency of less than 10ms; file-based results store structured data in CSV format and image files in PNG format, with all files organized in a directory structure of "analysis batch - scheme pair number" to ensure traceability.

2. A method for automated comparative analysis of multiple clinical treatment options, characterized in that, The automated comparison and analysis system for multiple clinical treatment options described in claim 1 includes the following steps: Step S1: Receive clinical case report data and protocol combination files, associate multi-source data based on patient primary keys, perform field standardization mapping and data cleaning, and output standardized analysis data tables; Step S2: Analyze the candidate treatment plan pairs item by item according to the worksheets and row records in the scheme combination file to construct the dataset to be compared; Step S3: Perform data preprocessing, propensity score matching automatic optimization, baseline analysis, response rate analysis, survival analysis and subgroup analysis on the verified dataset to be compared in sequence to generate structured result objects; Step S4: Perform joint significance determination based on the reaction rate analysis results and survival analysis results, and output the overall significance indicator; Step S5: Store the structured result object as an intermediate result file, trigger graphical standardization and report generation for candidate treatment pairs that meet the overall significance condition, and index, summarize and cluster the results of multiple candidate treatment pairs.

3. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the automatic comparison and analysis method for multiple clinical treatment options as described in claim 2.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the automatic comparison and analysis method for multiple clinical treatment options as described in claim 2.

Images