Type 2 diabetes patient personalized follow-up visit strategy generation and optimization method

Abstract

The invention discloses a real world data-based personalized follow-up visit strategy generation and optimization method for patients with type 2 diabetes mellitus, which comprises the following steps of: calculating a patient state transition probability by utilizing a Markov decision process model through integrating demographic characteristics, a baseline health state, a laboratory examination result and life style data; and further, an individual follow-up visit frequency and a follow-up visit mode are dynamically generated. When a patient's clinical event or health status changes, the method may adjust the follow-up plan in real time. Compared with an existing standardized follow-up visit method, the method has remarkable advantages in the aspects of reducing the occurrence rate of complications, prolonging the health life and reducing the medical cost, is particularly suitable for basic medical institutions and resource-limited environments, and has universality and popularization value.

Description

Technical Field

[0001] This invention belongs to the field of chronic disease management and digital healthcare, specifically relating to a method for generating and optimizing a personalized follow-up strategy for patients with type 2 diabetes, and more specifically to a method for generating and optimizing a personalized follow-up strategy for patients with type 2 diabetes based on real-world data. Background Technology

[0002] Type 2 diabetes is a globally prevalent chronic metabolic disease, with its prevalence continuously rising due to population aging, obesity, and lifestyle changes. Diabetes patients often require long-term management to reduce the risk of complications, improve quality of life, and decrease healthcare costs. Follow-up management is a crucial aspect of diabetes control; through regular follow-ups, physicians can understand the patient's disease progression and treatment adherence, thereby adjusting medication and lifestyle interventions accordingly.

[0003] Currently, both international and domestic guidelines recommend that patients with diabetes undergo regular follow-up. For example, the American Diabetes Association (ADA) and the European Association for Diabetes (EASD) recommend increasing the frequency of follow-up for newly diagnosed patients or those with poor glycemic control, while China's primary public health guidelines stipulate that patients with type 2 diabetes should undergo four evenly spaced face-to-face follow-up visits per year.

[0004] However, the existing standardized follow-up model has the following shortcomings: (1) Lack of consideration for individual differences. Patients vary significantly in age, underlying diseases, lifestyle habits, and risk of complications, and fixed-frequency follow-up is difficult to meet the needs of different groups; (2) Uneven allocation of medical resources. In primary healthcare institutions, excessive follow-up will lead to waste of resources, while insufficient follow-up may delay the detection of the disease; (3) Lack of data-driven decision-making tools. At present, the adjustment of follow-up frequency and method mainly relies on clinical experience, lacking a systematic and dynamic scientific method, making it difficult to achieve precise management.

[0005] Therefore, how to optimize the frequency and method of follow-up while ensuring clinical efficacy, and achieve individualized diabetes follow-up management, has become an urgent technical problem to be solved. Summary of the Invention

[0006] Purpose of the Invention: Addressing the problems of existing diabetes follow-up strategies, such as lack of individualization, low resource utilization efficiency, and a lack of scientific decision-making tools, the purpose of this invention is to provide a method for generating and optimizing personalized follow-up strategies for type 2 diabetes patients based on real-world data. This method can dynamically determine the follow-up frequency and method based on the patient's multi-dimensional health information and make real-time adjustments as the condition changes, thereby reducing medical costs while ensuring clinical effectiveness and improving the precision of follow-up management.

[0007] Technical Solution: To achieve the above-mentioned objectives, the present invention provides a method for generating and optimizing personalized follow-up strategies for patients with type 2 diabetes, comprising the following steps: (1.1) Data collection and preprocessing: The first follow-up data and electronic health record information of patients were collected as input data, including demographic characteristics, baseline health status, laboratory test indicators, lifestyle data and past complications; missing data were repaired using machine learning imputation methods to obtain a complete input dataset as the development dataset; (1.2) State modeling: Based on the input data, a finite state Markov decision process model is constructed to calculate the transition probability between different health states and to establish a state-action-reward function framework; (1.3) Generation of personalized follow-up strategies: Using the Markov decision process model, the personalized follow-up frequency and follow-up method for patients are dynamically generated within a given time range; (1.4) Dynamic adjustment and optimization: When the patient’s health status or risk of complications changes, the Markov decision process model is updated in real time based on the latest input data, and the existing follow-up strategy is dynamically adjusted to optimize clinical efficacy and economic cost. (1.5) Results output and application: The generated personalized follow-up strategy will be applied to the patient's chronic disease management system to assist medical institutions and medical staff in formulating and adjusting follow-up plans.

[0008] Furthermore, the data preprocessing in step (1.1) includes: (2.1) The random forest model is used to impute missing data, which can handle both numerical and categorical variables at the same time, and can recover missing values ​​through nonlinear mapping without assuming the data distribution. (2.2) Variables with missing rates exceeding a preset threshold are removed to ensure the stability and reliability of the input data; (2.3) Standardize the first follow-up data from different sources, including unifying the timestamp format, unifying the unit of the detection index, removing duplicate records, and correcting outliers.

[0009] Furthermore, the construction of the Markov decision process model in step (1.2) includes: (3.1) Based on the first follow-up data, the health status is divided into a finite set of statuses, which includes a state without complications, a state with a single complication, a state with multiple complications, and a state of death. (3.2) By performing statistical analysis on the first follow-up data, the transition probabilities between finite states are estimated, and a state transition matrix is ​​constructed; (3.3) Define follow-up behavior as the action space, and the follow-up behavior includes two categories: follow-up frequency and follow-up method; (3.4) Set a reward function to comprehensively measure the patient's health outcome and medical cost, wherein the health outcome is represented by the number of days the patient survives without complications, and the medical cost is represented by the follow-up and treatment expenses per unit time. (3.5) Solve for the optimal follow-up policy within a given time range using a policy optimization algorithm to maximize the expected value of the reward function.

[0010] Furthermore, the generation and dynamic adjustment of personalized follow-up strategies in steps (1.3) and (1.4) include: (4.1) After inputting the patient's initial health status and individual characteristics, a Markov decision process model is used to generate individualized follow-up frequency and follow-up method. The follow-up frequency includes monthly, quarterly and semi-annual follow-up, and the follow-up method includes face-to-face follow-up, remote follow-up and telephone follow-up. (4.2) When a patient develops complications or changes in health status during the follow-up process, the transfer probability is recalculated based on the updated second follow-up data, and the follow-up strategy is updated in real time. (4.3) The dynamic optimization goal of the personalized follow-up strategy is to minimize the overall cost of follow-up and treatment while ensuring clinical efficacy; (4.4) The updated personalized follow-up strategy will be output to the chronic disease management system in real time for clinicians and patients to refer to and implement.

[0011] Furthermore, the application scenarios of the method include: (5.1) Used as a follow-up decision support tool in community-level primary healthcare institutions to support medical staff in carrying out personalized management under limited resources; (5.2) Application in telemedicine or mobile health management platforms, using electronic health record systems to input first follow-up data in real time and update follow-up strategies; (5.3) Promote its use in regional or national public health chronic disease management systems to develop individualized follow-up policies to reduce the incidence of complications and overall healthcare costs.

[0012] Furthermore, step (1.2) includes: (6.1) The input parameters of the Markov decision process model are estimated using machine learning methods. The input parameters include state transition probabilities, health outcome indicators and cost-utility parameters. (6.2) The machine learning methods mentioned include random forest, gradient boosting decision tree or other nonlinear regression algorithms, used to mine complex relationships between variables from the first follow-up data; (6.3) When clinical data is insufficient, supplement the model parameters by combining external evidence-based medicine evidence and expert knowledge to ensure the accuracy and robustness of the model; (6.4) Validate the generalization performance of the Markov decision process model by cross-validation and external independent datasets.

[0013] Furthermore, the output of step (1.5) uses a cost-benefit analysis method, including the following steps: (7.1) Calculate the patient’s health outcome indicators, including complication-free survival days or quality-adjusted life days, based on the simulation results of the personalized follow-up strategy; (7.2) Calculate the total medical costs during the follow-up and related treatment process, including outpatient follow-up expenses, drug costs and treatment costs for complications; (7.3) Calculate the incremental cost-effectiveness ratio (ICER) between personalized follow-up strategies and standardized follow-up strategies, using the following formula:

[0014] in, and These represent the average costs of personalized follow-up strategies and standardized follow-up strategies, respectively. and These represent the health outcome indicators corresponding to personalized follow-up strategies and standardized follow-up strategies, respectively. (7.4) When the ICER is lower than the preset willingness to pay threshold, it is determined that the personalized follow-up strategy has significant cost-effectiveness advantages.

[0015] Furthermore, it also includes external verification steps: (8.1) Select real-world third follow-up data other than the first follow-up data for external validation. The third follow-up data comes from type 2 diabetes patients managed by a medical institution in another region. (8.2) Keeping the input data and parameter settings unchanged, input the third follow-up data into the personalized follow-up strategy to generate an individualized follow-up strategy and output the corresponding health outcome and cost result; (8.3) When the external validation results are consistent with the results of the development dataset in terms of improvement in clinical outcomes and reduction in costs, the method is deemed to have universality and generalizability.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects: (1) A method for generating and optimizing personalized follow-up strategies for type 2 diabetes based on real-world follow-up data is proposed, which can take into account individual differences on the basis of group patterns and achieve dynamic decision-making; (2) By constructing a Markov decision process model, health outcomes (quality-adjusted life days, QALDs) and medical costs are incorporated into the reward function to achieve dual optimization of clinical efficacy and economic efficiency; (3) The incremental cost-effectiveness ratio (ICER) of personalized follow-up strategy and standardized follow-up strategy was calculated by using cost-benefit analysis. The results showed that personalized follow-up has significant cost-benefit advantages. (4) External validation was conducted on community population data in another region. The validation results were consistent with the development dataset, proving that the method has robustness and universality. (5) This invention can be realized by relying on routine follow-up and electronic health records. It is applicable to community primary medical institutions, telemedicine platforms and regional public health systems, and has good prospects for promotion and social benefits. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the method for generating and optimizing personalized follow-up strategies for type 2 diabetes patients based on real-world data, as described in this invention. Figure 2 This is a schematic diagram of the state transitions in the Markov decision process model of the present invention, including a state without complications, a state with a single complication, a state with multiple complications, and a state of death. Figure 3 This is a schematic diagram comparing the personalized follow-up strategy generated by this invention with the existing standardized follow-up strategy; Figure 4 This diagram illustrates the validation results of the present invention on external community population data, comparing the differences in health outcomes and medical costs between personalized follow-up strategies and standardized follow-up strategies. Detailed Implementation

[0018] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] Example 1: Data Collection and Preprocessing (1) Data source The dataset used in this invention is derived from first follow-up data and electronic health record information from multiple community health service centers in Nanjing. The data includes: Demographic characteristics: such as age and sex; Baseline health status: such as duration of diabetes, body mass index (BMI). Laboratory test indicators: such as glycated hemoglobin (HbA1c), fasting blood glucose, and blood pressure; Lifestyle data: such as smoking and drinking habits, exercise habits; Previous complications: such as nephropathy, retinopathy, peripheral neuropathy, and lower extremity arterial disease / diabetic foot.

[0020] (2) Missing data processing Missing data is imputed using machine learning methods, with a random forest model being the preferred choice. This method can handle both numerical and categorical variables simultaneously, predicting missing values ​​through nonlinear mapping without requiring assumptions about the data distribution.

[0021] Variables with a missing rate exceeding 10% were removed to ensure the stability and reliability of the input data.

[0022] (3) Standardization process To ensure data consistency, first follow-up data from multiple sources were standardized, including: The follow-up time should be standardized using a standard timestamp format; Standardized testing units for laboratory indicators; Remove duplicate records and retain only valid values; Correct outliers.

[0023] (4) Sample data Table 1 provides examples of input data for some patients at their first follow-up visit: Table 1. Example of Patient Input Data

[0024] Note: HbA1c represents glycated hemoglobin level, and BMI represents body mass index.

[0025] (5) Flowchart description like Figure 1 As shown, the overall process of the method of the present invention includes multiple stages: First, patients with type 2 diabetes were divided into a standardized follow-up group and a non-standardized follow-up group according to the existing management model; The standardized follow-up group received face-to-face follow-ups at a fixed frequency (e.g., once every three months). Non-standardized follow-up varies dynamically based on individual circumstances; All patients’ health data (such as health records, laboratory test results, diagnostic records, lifestyle, etc.) were incorporated into the Markov decision model; Based on this model, a personalized follow-up management plan can be generated; Ultimately, cost-effectiveness analysis and field-controlled verification were used to evaluate the superiority of the personalized follow-up protocol in terms of clinical outcomes and economic costs.

[0026] Figure 1It fully demonstrates a closed-loop system from data collection to decision generation and effect evaluation, reflecting a personalized follow-up strategy optimization framework based on real-world data.

[0027] Example 2: State Modeling (1) State division Based on the patients' first follow-up data, this invention divides the health status of type 2 diabetes patients into a finite set of states, specifically including: S0: No complications; S11–S14: Single complication status, respectively: S11: Diabetic nephropathy; S12: Retinopathy; S13: Peripheral neuropathy; S14: Lower extremity arterial disease / diabetic foot; S21–S26: Any combination of two complications; S31–S34: Any combination of three complications; S4: The simultaneous presence of four complications; S5: Death state (absorption state, no further transfer after entering).

[0028] The patient is in only one of the above states at any given time.

[0029] like Figure 2 The diagram shown is a state transition diagram of the Markov decision process model established in this invention, clearly illustrating the possible evolutionary paths of a patient between various health states. The arrows in the diagram indicate the direction and probability of state transitions. State S0 can transition to various complication states or remain unchanged, and may eventually enter S5 (death). The transitions between each state are influenced by the patient's follow-up actions; for example, the frequency and method significantly affect the numerical value of the transition probability.

[0030] This figure also illustrates the dynamic process of complex complication combinations (such as S31–S34, S4) gradually converging to the endpoint S5, which is an important basis for modeling the evolution of chronic disease course within the patient's life cycle.

[0031] (2) State transition probability In this invention, the evolution of a patient's health status is modeled using a Markov Decision Process (MDP) model. The key definitions in this model are as follows: : indicates that the patient was on the The health status at each time point takes values ​​within the aforementioned finite set of states. : indicates that the patient was on the Follow-up behaviors (actions) received at each time point, including follow-up frequency and follow-up methods; : indicates that the patient was on the Health status at a specific point in time; : indicates the state Take action below Then transition to state The conditional probability.

[0032] By longitudinally tracking and statistically analyzing the first follow-up data, the transition probabilities between each state are estimated, and a transition matrix is ​​formed. This provides the foundation for strategy generation.

[0033] (3) Action space Follow-up behavior is defined as an action space, which includes: Follow-up frequency: monthly, quarterly, semi-annually; Follow-up methods: face-to-face follow-up, remote follow-up, and telephone follow-up.

[0034] Different actions can affect the probability of state transition and related medical costs.

[0035] (4) Reward function The reward function takes into account both healthy outcomes and medical costs. Health utility: expressed in quality-adjusted life days (QALDs). Utility value for patients without complications. The value was 0.87, indicating peripheral neuropathy. The value was 0.77, indicating retinal disease. The value was 0.83, indicating kidney disease. The value was 0.75, indicating lower extremity arterial disease / diabetic foot. The value is 0.69. If the patient is in a combined state, the value is calculated by weighting the utility value of the individual complication.

[0036] Studies have shown that for diabetic patients, HbA1c levels increase by 0.25% for every 35.2 km increase in travel distance for outpatient follow-up. A 1% increase in HbA1c leads to a 0.21-point decrease in quality of life. Furthermore, participation in follow-up improves quality of life by 0.097 points. However, the likelihood of participation decreases by 7.6% for every 24 km increase in travel distance. Therefore, the impact of travel barriers on the health utility of diabetic patients, expressed as the health utility of diabetic patients, is as follows:

[0037] patient i exist t The health utility value at that time is expressed as , means as follows:

[0038] : Quality-adjusted life days (QALDs) utility value in the absence of complications; : Represents the health utility value for patients with different complications; : indicates the first The patients in time Does the patient have the first Complications; : indicates the first The patients in time Is the patient in a state without complications? : indicates the first Did the patient [do something] within the time frame? Do you accept outpatient follow-up? :patient One-way distance to the follow-up location (unit: km); The loss of health benefits due to distance.

[0039] Medical costs include the cost of basic diabetes treatment, treatment of complications, and follow-up care. Costs for those without complications are age-related.

[0040] Treatment of complications The costs are as follows: neuropathy =5500 yuan / year, retinal disease =4500 yuan / year, kidney disease =12,000 yuan / year, lower extremity arterial disease / diabetic foot = The annual cost is 8000 yuan. Regarding follow-up costs, according to the requirements of China's public health service standards, diabetic patients are entitled to four free outpatient follow-up visits per year. Any additional outpatient follow-up visits are charged at 12 yuan per visit, including a 2 yuan registration fee and a 10 yuan blood glucose test fee. Community follow-up costs are calculated based on home care service fees, at 40 yuan per visit. Telephone follow-up costs are determined by the nurse's hourly wage and average follow-up duration, at 13.8 yuan per visit. The total cost of follow-up examinations is recorded as follows: , means as follows:

[0041] Representing patients i exist tWhether to choose outpatient follow-up. Representing patients i exist t Whether to choose telephone follow-up, Representing patients i exist t Whether to choose community follow-up.

[0042] patient i In time t The total cost is expressed as , indicates the following

[0043] Representing patients i exist t The state of being suffering from complications.

[0044] (5) Strategy optimization Within a given timeframe, the optimal follow-up strategy is determined by employing strategy optimization algorithms (such as strategy iteration and value iteration), comprehensively considering the health utility and cost parameters described in Table 2. The optimization objective is: Minimize the total cost of medical care and follow-up while ensuring the best possible patient health outcomes; The incremental cost-effectiveness ratio (ICER) of personalized follow-up versus standardized follow-up is calculated using cost-benefit analysis. When the ICER is lower than the willingness-to-pay threshold, the personalized follow-up strategy is considered optimal.

[0045] Example 3: Generation and Dynamic Adjustment of Personalized Follow-up Strategies (1) Strategy generation After inputting the patient's initial health status and individual characteristics, a personalized follow-up strategy is generated using the Markov decision process model described in Example 2. The strategy output includes the follow-up frequency and follow-up method. Follow-up frequency is recommended based on personalization. Follow-up methods include face-to-face follow-up, remote follow-up, and telephone follow-up.

[0046] For example, in Figure 3 In Mr. Wang's example: For example patients without complications, the optimal strategy is to conduct 5 follow-ups within the next year, and the follow-up methods should be community follow-up, outpatient follow-up, telephone follow-up, community follow-up, and telephone follow-up respectively. The recommended frequency and method of follow-up will vary depending on the patient's comorbidity status. These strategies vary depending on different health conditions, reflecting personalized and differentiated management.

[0047] Table 2. Follow-up management plan for example patients

[0048] (2) Dynamic adjustment A patient's health status may change over time. This invention optimizes existing follow-up protocols through a dynamic adjustment mechanism: When a patient develops a new complication, the status is automatically updated to the corresponding single or multiple complication status, and a new personalized follow-up strategy is generated in real time. For patients with stable conditions, if no increase in risk is found in multiple consecutive follow-ups, the follow-up interval can be extended to reduce resource consumption. For high-risk patients (such as those with persistently elevated HbA1c or rapidly deteriorating renal function), the follow-up interval should be shortened and face-to-face follow-up should be given priority.

[0049] (3) Result output The generated personalized follow-up strategies are output through the chronic disease management system for clinicians and patients to refer to and implement. The output includes: Recommended follow-up frequency (e.g., "time schedule for each follow-up visit within the next year"); Recommended follow-up methods (e.g., "Recommended follow-up methods for each follow-up visit within the next year"); Strategy adjustment prompts (e.g., "Complication progression has been detected, and the follow-up plan has been updated to monthly outpatient follow-up").

[0050] Example 4: Cost-Benefit Analysis (1) Calculation of health outcomes Based on the personalized follow-up strategy generated in Example 3, patients' quality-adjusted life days (QALDs) were simulated during the long-term follow-up period. As shown in Table 3, personalized follow-up generally improved patients' health outcomes.

[0051] For example, the mean QALDs for patients without complications under standardized follow-up was 244.7 days / year, while under personalized follow-up it increased to 258.0 days / year; for patients with complications, personalized follow-up significantly reduced health loss, extending it by an average of 2.2-17.4 days / year.

[0052] Table 3. Effects of different follow-up management protocols for example patients

[0053] Looking at the community as a whole, the average QALDs under standardized follow-up was 229.5 days / year, while under personalized follow-up it increased to 260.9 days / year.

[0054] (2) Cost calculation The total medical costs within the same period include: Basic treatment costs; Cost of treatment for complications; Follow-up costs (outpatient visit: 12 yuan / visit; community follow-up: 40 yuan / visit; telephone follow-up: 13.8 yuan / visit). Simulation results show that personalized follow-up for patients at different initial stages of comorbidities consistently reduced long-term total costs (Table 4).

[0055] Table 4. Costs and cost-effectiveness ratios of different follow-up management plans for example patients.

[0056] Similar results were observed at the community-wide level, with the average cost being 22,486.19 yuan under the current standardized follow-up strategy. Under the proposed personalized follow-up strategy, the average cost was 14,986.14 yuan.

[0057] (3) ICER calculation Compare the differences in health outcomes and costs between personalized and standardized follow-up, and calculate the incremental cost-effectiveness ratio (ICER):

[0058] in, and These represent the average costs of personalized follow-up and standardized follow-up, respectively. and These represent the corresponding health outcome indicators for both.

[0059] In the simulation results: Taking sample patients with different comorbidities as examples, the incremental cost-utility ratio (ICER) of personalized follow-up ranged from -174.20 yuan / QALD to -1900.70 yuan / QALD (reduced costs and improved efficacy), indicating that personalized follow-up is an absolutely superior strategy. Similar results were observed at the population level, with the ICER of personalized follow-up being -238.86 yuan / QALD.

[0060] (4) Determination of willingness to pay threshold Sensitivity analyses conducted at different willingness-to-pay thresholds (e.g., 1–3 times GDP per capita) showed that personalized follow-up demonstrated cost-effectiveness advantages. This indicates that the method is not only clinically superior but also economically feasible and has value for large-scale implementation.

[0061] Example 5: External Validation and Promotion Application Scenarios (1) Sources of external validation data To verify the universality of the method of this invention, a third follow-up dataset, independent of the development dataset, was selected, consisting of type 2 diabetes patients managed by a community healthcare institution in another region. This data differed from the development dataset in terms of age distribution, baseline health status, and complication spectrum.

[0062] (2) Verification method Keeping the input variables and parameter settings of the Markov decision process model unchanged, the external follow-up data is input into the personalized follow-up strategy generation and optimization method described in Examples 1 to 4, the personalized follow-up strategy is output, and the health outcome and cost result are calculated.

[0063] (3) Verification results The results show that the personalized follow-up strategy performs consistently with the development population in the external community: under the current standardized follow-up strategy, the average QALDs are 242.6 days, and the total cost is RMB 19,154.98. When the suggested personalized follow-up strategy is adopted, the average QALDs are 271.1 days, the total cost is RMB 14,277.8, and the incremental cost-utility ratio (ICER) is -RMB 171.13 / QALD, far below the willingness-to-pay threshold. These results demonstrate that the method of this invention has good robustness and cross-regional promotion value.

[0064] like Figure 4 The figure shown is a year-to-year cumulative comparison of the personalized follow-up strategy of this invention and the current standardized follow-up strategy in terms of health outcomes (QALDs) and medical costs: Figure 4 (a) shows the growth trend of cumulative quality-adjusted life days (QALDs) over 12 months between personalized and standardized follow-up strategies: In the early stages of follow-up (months 1–7), the growth of QALDs for both strategies was roughly the same, with the standardized follow-up having a slight advantage. Starting from month 8, the personalized follow-up group (circle ○) surpassed the standardized follow-up group (circle △), and the gap continued to widen over time. By month 12, the cumulative QALDs of the personalized follow-up was significantly higher than those of the standardized strategy, indicating that its health benefits have a time-cumulative and long-term advantage.

[0065] Figure 4 (b) in the figure shows the corresponding cumulative medical cost trend, indicating that personalized follow-up performs better in terms of cost control: the costs of the two groups are similar in the early stage, but from the 5th month onwards, the cumulative cost of standardized follow-up grows faster; by the 12th month, the cost of the standardized follow-up group is close to 19,000 yuan, while the cost of the personalized follow-up group is controlled within 15,000 yuan.

[0066] This figure visually verifies that the personalized follow-up method based on Markov models and real data proposed in this invention can significantly reduce medical expenses while ensuring clinical efficacy, and is sustainable and scalable.

[0067] (4) Promotion of application scenarios The method of this invention is not only applicable to community-level medical institutions, but can also be applied to the following scenarios: Telemedicine and mobile health platform: Real-time input of follow-up data and updating of personalized strategies through electronic health records; Regional public health management system: Promote personalized follow-up policies at the regional or national level and optimize the allocation of medical resources; Resource-constrained environments: Utilize low-cost follow-up methods (such as telephone or remote follow-up) to achieve effective coverage of chronic disease management.

Claims

1. A method for generating and optimizing personalized follow-up strategy for type 2 diabetes patients, characterized in that, Comprising the following steps: (1.1) Data collection and preprocessing: Collecting the first follow-up data and electronic health record information of patients as input data, including demographic characteristics, baseline health status, laboratory examination indicators, lifestyle data and previous complications; Missing data is repaired using machine learning imputation methods to obtain a complete input dataset as a development dataset; (1.2) State modeling: Based on the input data, a finite state Markov decision process model is constructed to calculate the transition probabilities between different health states and establish a state-action-reward function framework; (1.3) Generation of personalized follow-up strategy: Using the Markov decision process model, the personalized follow-up frequency and mode of the patient are dynamically generated within a given time range; (1.4) Dynamic adjustment and optimization: When the patient's health status or complication risk changes, the Markov decision process model is updated in real time based on the latest input data, and the existing follow-up strategy is dynamically adjusted to optimize the clinical effect and economic cost; (1.5) Result output and application: The generated personalized follow-up strategy is applied to the patient's chronic disease management system to assist medical institutions and medical staff in developing and adjusting follow-up plans.

2. The personalized follow-up strategy generation and optimization method of claim 1, wherein, The data preprocessing in step (1.1) includes: (2.1) Impute missing data using a random forest model, which can handle both numerical and categorical variables, and recover missing values through nonlinear mapping without assuming data distribution; (2.2) Variables with missing rates exceeding a pre-set threshold are excluded to ensure the stability and reliability of the input data; (2.3) Standardize the first follow-up data from different sources, including unifying timestamp format, unifying detection indicator units, removing duplicate records, and correcting outliers.

3. The personalized follow-up strategy generation and optimization method of claim 1, wherein, The Markov decision process model construction in step (1.2) includes: (3.1) Divide the health status into a finite state set based on the first follow-up data, including no complication state, single complication state, multiple complication state and death state; (3.2) Estimate the transition probabilities between finite states by statistical analysis of the first follow-up data and construct a state transition matrix; (3.3) Define follow-up behavior as action space, including follow-up frequency and follow-up mode; (3.4) Set up a reward function to comprehensively measure the patient's health outcome and medical cost, with the patient's complication-free survival days as the health outcome and the follow-up and treatment costs per unit time as the medical cost; (3.5) Solve the optimal follow-up strategy within a given time range by a strategy optimization algorithm to maximize the expected value of the reward function. 4.The personalized follow-up strategy generation and optimization method of claim 1, wherein, The personalized follow-up strategy generation and dynamic adjustment in steps (1.3) and (1.4) include: (4.1) After inputting the initial health status and individual characteristics of the patient, the Markov decision process model is used to generate individualized follow-up frequency and follow-up mode, including monthly, quarterly and semi-annual follow-up, and face-to-face, remote and telephone follow-up; (4.2) When the patient has complications or changes in health status during follow-up, the transition probability is recalculated based on the updated second follow-up data, and the follow-up strategy is updated in real time; (4.3) The dynamic optimization goal of the personalized follow-up strategy is to minimize the comprehensive cost of follow-up and treatment under the premise of ensuring clinical efficacy; (4.4) The updated personalized follow-up strategy is output to the chronic disease management system in real time for reference and execution by clinicians and patients.

5. The personalized follow-up strategy generation and optimization method of claim 1, wherein, The application scenarios of the method include: (5.1) As a follow-up decision support tool in community primary medical institutions, it supports medical staff to conduct personalized management under limited resources; (5.2) Applied in telemedicine or mobile health management platform, the first follow-up data is input in real time through electronic health record system and the follow-up strategy is updated; (5.3) Popularized in regional or national public health chronic disease management system, used to develop individualized follow-up policy to reduce the incidence of complications and overall medical costs. 6.The personalized follow-up strategy generation and optimization method of claim 1, wherein, The step (1.2) includes: (6.1) Use machine learning methods to estimate the input parameters of the Markov decision process model, including state transition probability, health outcome indicators and cost-effectiveness parameters; (6.2) The machine learning method includes random forest, gradient boosting decision tree or other nonlinear regression algorithm, which is used to mine the complex relationship between variables from the first follow-up data; (6.3) When the clinical data is insufficient, combine external evidence-based medical evidence and expert knowledge to supplement the model parameters to ensure the accuracy and robustness of the model; (6.4) The generalization performance of the Markov decision process model is verified by cross-validation and external independent data set.

7. The personalized follow-up strategy generation and optimization method of claim 1, wherein, The output of the results of step (1.5) uses cost-effectiveness analysis method, including the following steps: (7.1) Calculate the health outcome indicators of the patient based on the simulation results of the personalized follow-up strategy, including the number of complication-free survival days or quality-adjusted life days; (7.2) Calculate the total medical cost during follow-up and related treatment, including outpatient follow-up expenses, drug costs and complication treatment costs; (7.3) Calculate the incremental cost-effectiveness ratio ICER between the personalized follow-up strategy and the standardized follow-up strategy, the formula is: wherein, and respectively represent the average cost of the personalized follow-up strategy and the standardized follow-up strategy, and respectively represent the health outcome indicators corresponding to the personalized follow-up strategy and the standardized follow-up strategy. (7.4) When the ICER is lower than the preset willingness to pay threshold, it is determined that the personalized follow-up strategy has significant cost-effectiveness advantage.

8. The personalized follow-up strategy generation and optimization method of claim 1, wherein, It also includes an external verification step: (8.1) Select real-world third follow-up data other than the first follow-up data for external verification, which comes from type 2 diabetes patients managed by medical institutions in another region; (8.2) Keep the input data and parameter settings unchanged, input the third follow-up data into the personalized follow-up strategy, generate individualized follow-up strategy and output the corresponding health outcomes and cost results; (8.3) When the external verification results and the development data set results are consistent in terms of clinical outcome improvement and cost reduction, it is determined that the method has universality and generalizability.

Images