A method for stratifying and early warning of relapse risk of acute leukemia

Abstract

The present application relates to the technical field of risk prediction based on medical record data, and particularly relates to an acute leukemia relapse risk stratification early warning method. Based on the similar situation of relapse influencing factor values between historical medical record samples and the similar situation of first relapse intervals, the similarity of relapse influencing factor values of the historical medical record samples and the current medical record sample is adjusted to screen similar samples and preliminarily quantify the relapse risk value of the current medical record sample. Further, the relapse risk change curve is fitted by combining the first relapse interval difference, the relapse influencing factor difference and the adverse habit characteristic value difference between the historical medical record samples, and the influence law of the dynamic change of the adverse habit on the relapse risk is quantified. Finally, the adverse habit characteristic value difference of the current medical record sample from discharge to the current time is mapped to the curve to obtain the relapse risk change amount, and the relapse risk value is fused to form a dynamically adjusted risk assessment index, thereby improving the accuracy of the relapse risk early warning of the current medical record sample.

Description

Technical Field

[0001] This invention relates to the field of risk prediction technology based on medical record data, and specifically to a method for risk stratification and early warning of acute leukemia relapse. Background Technology

[0002] Acute leukemia is a malignant clonal disease of the blood system, usually divided into myeloid and lymphoid lineages, and each lineage is further divided into different subtypes. Although techniques such as chemotherapy, targeted therapy, and hematopoietic stem cell transplantation have significantly improved the initial remission rate of patients, and some subtypes can achieve complete remission and cure, relapse is still the main cause of treatment failure and death. Therefore, early warning and assessment of relapse risk in the later stage is still very important. Its main purpose is to take effective treatment measures in the early stage of relapse, avoid missing the best intervention opportunity, and effectively prolong the patient's survival expectation and quality of life.

[0003] When predicting the relapse risk of leukemia patients, current technologies usually consider the values ​​of relapse influencing factors reflected by the patient's physical condition, such as age and examination indicators. However, in reality, unhealthy lifestyle habits are also a key factor in the relapse risk of acute leukemia patients, and it is a dynamic influencing factor in the later stages. If the impact of unhealthy habits on the relapse risk is ignored, the final risk assessment will not accurately reflect the patient's actual condition. Summary of the Invention

[0004] To address the fact that unhealthy lifestyle habits are a key factor in the relapse risk of acute leukemia patients, and that this is a dynamic influencing factor in the later stages, ignoring the impact of unhealthy habits on relapse risk would lead to a technical problem where the final risk assessment cannot accurately reflect the patient's actual condition. The purpose of this invention is to provide a stratified early warning method for acute leukemia relapse risk, and the specific technical solution adopted is as follows:

[0005] Obtain historical medical record samples, including the interval between the first relapses, the characteristic values ​​of unhealthy habits at the time of the first relapse, and the values ​​of various factors influencing relapse; obtain current medical record samples, including the characteristic values ​​of unhealthy habits at the time of discharge, the characteristic values ​​of unhealthy habits at the current time, and the values ​​of various factors influencing relapse;

[0006] By utilizing the similarity of relapse influencing factor values ​​and the similarity of first relapse intervals among historical medical record samples, the similarity characteristics of relapse influencing factor values ​​between historical and current medical record samples are adjusted to screen out similar samples for the current medical record sample; the relapse risk value of the current medical record sample at the current time is determined based on the first relapse interval of similar samples.

[0007] Based on the analysis of the differences in the interval between the first relapse, the differences in the values ​​of relapse influencing factors, and the differences in the values ​​of unhealthy habits among historical medical record samples, the change in the risk of relapse of the current medical record sample at the current time is determined according to the differences in the values ​​of unhealthy habits between the current medical record sample at the time of discharge and the current time.

[0008] By integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, a stratified early warning system can be implemented.

[0009] Furthermore, the method for obtaining similar samples includes:

[0010] For each recurrence influencing factor, historical medical records with the same recurrence influencing factor value are grouped into one category, thus obtaining a cluster.

[0011] For each relapse influencing factor, the similarity of the first relapse interval of historical medical record samples in the cluster was analyzed, and the importance index of each relapse influencing factor was determined by combining it with the quantitative characteristics of the cluster.

[0012] For each relapse influencing factor, the absolute value of the difference between the relapse influencing factor values ​​of each historical medical record sample and the current medical record sample is negatively correlated and mapped to the value, which is used as the medical record similarity value between each historical medical record sample and the current medical record sample.

[0013] The similarity values ​​between historical and current medical record samples were adjusted using the importance index of relapse influencing factors, thereby obtaining the similarity feature value between each historical and current medical record sample.

[0014] Among all historical medical record samples, those with similarity feature values ​​greater than a preset similarity threshold are considered as similar samples to the current medical record sample.

[0015] Furthermore, the method for obtaining the importance indicators includes:

[0016] In each cluster of each relapse influencing factor, the absolute value of the difference between the first relapse intervals of any two historical medical record samples is negatively correlated and mapped to the value, which is used as the relapse similarity factor between the two historical medical record samples.

[0017] The normalized value of the ratio of the mean recurrence similarity factor among all historical medical record samples in all clusters of each recurrence influencing factor to the number of clusters is used as the importance indicator for each recurrence influencing factor.

[0018] Furthermore, the method for obtaining the similarity feature values ​​includes:

[0019] Under all factors influencing recurrence, the similarity values ​​between each historical medical record sample and the current medical record sample are weighted and fused using the importance index of the factors influencing recurrence, thereby obtaining the similarity feature value between each historical medical record sample and the current medical record sample.

[0020] Furthermore, the method for obtaining the recurrence risk value includes:

[0021] Calculate the time interval between the current time and the discharge time of the current medical record sample, and use it as the test interval;

[0022] The test interval is compared with the first recurrence interval of each similar sample. If the test interval is greater than or equal to the first recurrence interval, the recurrence risk factor of the test interval is set to a preset value; if the test interval is less than the first recurrence interval, the ratio of the test interval to the first recurrence interval is used as the recurrence risk factor of the test interval.

[0023] The mean of the recurrence risk factors between the interval to be tested and all similar samples is used as the recurrence risk value of the current medical record sample at the current time.

[0024] Furthermore, the method for obtaining the change in relapse risk includes:

[0025] Based on the differences in relapse influencing factor values ​​among all historical medical record samples, medical record similarity parameters are determined among historical medical record samples to divide medical record sample groups.

[0026] Based on the differences in the first relapse interval, medical record similarity parameters, and differences in the characteristic values ​​of bad habits among historical medical record samples in the medical record sample group, the curve of relapse risk change was determined.

[0027] The difference between the unhealthy habit characteristic values ​​of the current medical record sample at the current time and at the time of discharge is taken as the test value, and the value of the vertical axis corresponding to the test value in the recurrence risk change curve is taken as the change in recurrence risk of the current medical record sample at the current time.

[0028] Furthermore, the method for obtaining the medical record sample group includes:

[0029] In all historical medical record samples, the absolute value of the difference between the values ​​of the recurrence influencing factors under each recurrence influencing factor for any two historical medical record samples is taken as the deviation factor. The sum of the deviation factors of these two historical medical record samples under all recurrence influencing factors is negatively correlated and mapped to the value of the medical record similarity parameter between the two historical medical record samples.

[0030] Each historical medical record sample is grouped together with the historical medical record sample corresponding to the maximum value among all similarity parameters of all medical records.

[0031] Furthermore, the method for obtaining the recurrence risk change curve includes:

[0032] In each medical record sample group, one historical medical record sample is used as the first sample, and the other historical medical record sample is used as the second sample.

[0033] The difference in the unhealthy habit feature values ​​between the first sample and the second sample is taken as the unhealthy habit change; the product of the difference in the first relapse interval between the first sample and the second sample and the medical record similarity parameter is numerically adjusted using the hyperbolic tangent function to obtain the relapse risk change value.

[0034] A two-dimensional coordinate system is constructed by using the change in unhealthy habits as the horizontal axis and the change in relapse risk as the vertical axis. This allows all data points corresponding to the medical record sample groups to be mapped into the two-dimensional coordinate system. Curve fitting is then performed on all data points to obtain the relapse risk change curve.

[0035] Furthermore, the stratified early warning system, which integrates the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, includes:

[0036] By integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, the recurrence risk index of the current medical record sample at the current moment is determined.

[0037] When the recurrence risk index is greater than or equal to the preset level 3 risk threshold, a level 3 warning is triggered. When the recurrence risk index is less than the preset level 3 risk threshold but greater than or equal to the preset level 2 risk threshold, a level 2 warning is triggered. When the recurrence risk index is less than the preset level 2 risk threshold, a level 1 risk warning is triggered. The severity of the level 1, level 2, and level 3 risk warnings increases sequentially.

[0038] Furthermore, the method for obtaining the recurrence risk indicator includes:

[0039] The normalized sum of the recurrence risk value and the change in recurrence risk of the current medical record sample at the current moment is used as the recurrence risk indicator of the current medical record sample at the current moment.

[0040] The present invention has the following beneficial effects:

[0041] First, historical and current medical record samples are obtained. Common methods assess future relapse risk by analyzing relapse-influencing factors (such as age and examination indicators) in the medical record samples. In general medical practice, the relapse risk of current patients is often assessed based on the relapse history of previous patients. This can be summarized as medical record samples with similar conditions having similar recovery abilities, and therefore, similar relapse risks. Therefore, by utilizing the similarity of relapse-influencing factors and the first relapse interval between historical medical record samples, the similarity characteristics of historical and current medical record samples are adjusted to screen for similar samples of the current medical record. Based on the first relapse interval of these similar samples, a preliminary estimate of the potential relapse risk value of the current medical record sample is then made. Furthermore, since the patient's unhealthy habits are also a key influencing factor on the relapse risk of acute leukemia patients, and this characteristic is a late-stage dynamic factor that cannot be directly reflected in the aforementioned relapse risk value, it needs to be analyzed separately. Based on a comprehensive analysis of the differences in three characteristics among historical medical record samples—unhealthy habit characteristics, first relapse interval, and relapse influencing factors—this study clarifies the specific change in relapse risk due to unhealthy habits in the current medical record sample, considering the differences in unhealthy habit characteristics between the current sample and the present moment. Finally, this change in relapse risk is integrated with the relapse risk value to determine a more accurate risk assessment characteristic parameter that accurately reflects the patient's actual condition. When used for stratified early warning of relapse risk in the current medical record sample at the current moment, this parameter can effectively improve the accuracy of early warning. Attached Figure Description

[0042] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a method for risk stratification and early warning of acute leukemia relapse provided in one embodiment of the present invention;

[0044] Figure 2 A flowchart illustrating a method for obtaining similar samples according to an embodiment of the present invention;

[0045] Figure 3 This is a flowchart illustrating a method for obtaining changes in recurrence risk, as provided in one embodiment of the present invention. Detailed Implementation

[0046] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for risk stratification and early warning of acute leukemia relapse according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0048] The following description, in conjunction with the accompanying drawings, details the specific scheme of the acute leukemia relapse risk stratification and early warning method provided by this invention.

[0049] Please see Figure 1 The diagram illustrates a flowchart of a method for risk stratification and early warning of acute leukemia relapse according to an embodiment of the present invention. The method includes the following steps:

[0050] Step S1: Obtain historical medical record samples, including the first relapse interval, the characteristic values ​​of bad habits at the time of the first relapse, and the values ​​of various relapse influencing factors; obtain current medical record samples, including the characteristic values ​​of bad habits at the time of discharge, the characteristic values ​​of bad habits at the current time, and the values ​​of various relapse influencing factors.

[0051] Even after acute leukemia patients are discharged from the hospital in complete remission, it is still necessary to monitor and warn of the risk of relapse, so as to ensure that more effective treatment measures can be taken in the early stage of relapse and avoid missing the best intervention opportunity.

[0052] First, it is necessary to collect historical medical record samples of patients with acute leukemia. These samples can be obtained from the hospital's electronic medical record system. The historical medical record samples should include discharge records and treatment records of acute leukemia patients. Specifically, this embodiment of the invention needs to extract the first relapse interval (the time interval between the first discharge and the first relapse), the values ​​of various relapse influencing factors at the time of the first relapse, and the values ​​of unhealthy habits at the time of the first relapse. Acute leukemia, as a malignant hematological disease, is generally considered to have a relapse risk related to factors such as patient age, gender, examination indicators, and disease subtype. Therefore, this embodiment sets the relapse influencing factors as age, gender, disease subtype, whether a donor transplant has been performed, and blood parameters. Furthermore, for text-based relapse influencing factors, [the following is omitted as it is not relevant to the main point]. After converting the data into vectors using the bag-of-words model, the modulus is obtained as the relapse influencing factor value. For numerical relapse influencing factors, the values ​​are normalized and then used as the relapse influencing factor values. The method for obtaining the unhealthy habit feature values ​​can be as follows: the unhealthy lifestyle habits of acute leukemia patients are recorded periodically through questionnaires and scored. Unhealthy lifestyle habits can include: smoking index, drinking index, staying up late index, stress index, etc. In the historical medical record sample, the mean of the index of each unhealthy lifestyle habit in all questionnaires between the first discharge time and the first relapse time is used as the value of each unhealthy lifestyle habit in the historical medical record sample at the first relapse time. Then, the values ​​corresponding to all unhealthy lifestyle habits are mapped to one-dimensional values ​​(L1 norm can be used) to serve as unhealthy habit feature values.

[0053] It is also necessary to obtain the current medical record sample, which is the medical record sample to be used for relapse risk prediction. This is also based on the hospital's electronic medical record system. The current medical record sample includes discharge records and treatment records of patients with acute leukemia. Since the current medical record sample is in a state of unpredictable relapse risk, the following are extracted: the unhealthy habit feature value at the time of discharge (the acquisition method is the same as the historical medical record sample, and the time period is from the time of admission to the time of discharge), the unhealthy habit feature value at the current time (the acquisition method is the same as the historical medical record sample, and the time period is from the time of discharge to the current time), and the values ​​of various relapse influencing factors at the current time (the acquisition method is the same as the historical medical record sample).

[0054] At this point, the necessary data can be extracted from the hospital's electronic medical record system; the bag-of-words model is a well-known technique, and the specific process will not be elaborated here.

[0055] In this embodiment of the invention, the collection and acquisition of patient medical record data are authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0056] Step S2: Utilize the similarity of recurrence influencing factor values ​​and the similarity of first recurrence intervals among historical medical record samples to adjust the similarity features of recurrence influencing factor values ​​between historical and current medical record samples, and screen out similar samples for the current medical record sample; determine the recurrence risk value of the current medical record sample at the current moment based on the first recurrence interval of similar samples.

[0057] Acute leukemia, a malignant hematological disease, can achieve complete remission in some patients with certain subtypes after treatment such as stem cell or bone marrow transplantation. However, a high risk of relapse remains, which is generally considered to be related to factors influencing relapse. In general medical practice, the relapse risk of a current medical record can be assessed by comparing it to historical relapse data. This is because, for almost all diseases, patients with similar physiological conditions are generally considered to have similar recovery abilities, thus the relapse risk of the current record can be predicted using similar historical records. However, different relapse-influencing factors contribute differently to the final similarity calculation. For example, acute leukemia patients over 60 years of age typically have a higher relapse rate and worse prognosis than those under 60 years of age with the same condition. In this case, age contributes more significantly to the similarity of relapse risk in the calculation. Therefore, for acute leukemia, it is necessary to determine which relapse-influencing factors are the key factors before identifying historical records similar to the current record. Therefore, by utilizing the similarity of relapse influencing factor values ​​and the similarity of first relapse intervals between historical medical record samples, the similarity characteristics of relapse influencing factor values ​​between historical and current medical record samples can be adjusted to screen out similar samples of the current medical record sample.

[0058] Preferably, in one embodiment of the present invention, the method for obtaining similar samples includes:

[0059] Please see Figure 2 The diagram illustrates a method flowchart for obtaining similar samples according to an embodiment of the present invention, which includes the following steps:

[0060] Step S201: Under each recurrence influencing factor, classify the historical medical record samples based on the recurrence influencing factor value to obtain clusters.

[0061] There are many factors that influence the relapse of acute leukemia, such as age and examination indicators. Therefore, we adopted a controlled variable approach to analyze the importance of each relapse influencing factor separately. First, under each relapse influencing factor, historical medical records with the same relapse influencing factor value were grouped into one category to obtain clusters (each cluster should include at least 2 historical medical records; if a cluster contains only one historical medical record, it is discarded and will not be analyzed in subsequent processes). Through clustering, we can observe the relapse situation of historical medical records with different values ​​under the same relapse influencing factor.

[0062] Step S202: Under each relapse influencing factor, analyze the similarity of the first relapse interval of historical medical record samples in the cluster, and combine it with the quantitative characteristics of the cluster to determine the importance index of each relapse influencing factor.

[0063] The first relapse interval is an important indicator for assessing relapse characteristics. Therefore, in each cluster of each relapse influencing factor, the absolute value of the difference between the first relapse intervals of any two historical medical records is calculated. The smaller the absolute value of this difference, the higher the relapse similarity between the two historical medical records. Therefore, this absolute value of the difference is negatively correlated to correct the logical relationship and serves as a relapse similarity factor between the two historical medical records. The larger the relapse similarity factor, the higher the similarity between the two historical medical records under that relapse influencing factor, and thus the greater the importance of that relapse factor in the relapse. This negative correlation mapping can be performed using the formula... ,in, Let x represent an exponential function with the natural constant e as the base, and let x represent the independent variable.

[0064] At this point, there is a recurrence similarity factor between every two historical medical record samples in each cluster.

[0065] The fewer the number of clusters formed under each recurrence influencing factor, the higher the numerical similarity of historical medical records under that factor, and thus the greater the importance of that factor to recurrence. Therefore, the normalized value of the ratio of the mean of the recurrence similarity factor among all historical medical records in all clusters for each recurrence influencing factor to the number of clusters is used as the importance index for each factor. Based on the aforementioned analytical logic, a higher importance index indicates that the timing and characteristics of recurrence in historical medical records are more similar under that factor, and thus the factor is more important. Normalization is a technique well-known to those skilled in the art; the normalization function can be linear normalization or standard normalization, etc., and the specific normalization method is not limited here.

[0066] Step S203: For each relapse influencing factor, analyze the differences in the relapse influencing factor values ​​between historical medical record samples and current medical record samples, and determine the medical record similarity value.

[0067] For each relapse influencing factor, the absolute value of the difference between the relapse influencing factor values ​​for each historical medical record sample and the current medical record sample is calculated. The smaller the absolute value of the difference, the higher the similarity between the data performance of the historical and current medical record samples. Therefore, the value after negative correlation mapping is used as the medical record similarity value between each historical and current medical record sample. The larger the medical record similarity value, the greater the likelihood that the historical medical record sample will serve as a similar sample for subsequently assessing the relapse risk of the current medical record sample. This negative correlation mapping can be performed using the formula... ,in, Let x represent an exponential function with the natural constant e as the base, and let x represent the independent variable.

[0068] Step S204: Adjust the medical record similarity value between historical medical record samples and current medical record samples using the importance index of recurrence influencing factors, thereby obtaining the similarity feature value between each historical medical record sample and the current medical record sample.

[0069] The greater the importance index of a recurrence influencing factor, the greater its contribution or weight should be in assessing the similarity between historical and current medical record samples. Therefore, under all recurrence influencing factors, the importance index of each recurrence influencing factor is used to weight and fuse the medical record similarity values ​​between each historical and current medical record sample. Specifically, the importance index of each recurrence influencing factor is used as a weight, and it is multiplied by the medical record similarity value between each historical and current medical record sample under each recurrence influencing factor to obtain a weighted similarity value. Finally, the sum of the weighted similarity values ​​of each historical and current medical record sample under all recurrence influencing factors is normalized and used as the similarity feature value between each historical and current medical record sample. The larger the similarity feature value, the higher the data similarity between the historical and current medical record samples. Normalization is a technique well-known to those skilled in the art, and the choice of normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0070] Step S205: Based on similarity feature values, filter out similar samples of the current medical record sample from all historical medical record samples.

[0071] Among all historical medical record samples, those with similarity feature values ​​greater than a preset similarity threshold are considered as similar samples to the current medical record sample.

[0072] It should be noted that the preset similarity threshold is 0.6, and the specific value can be adjusted according to the implementation scenario, which is not limited here.

[0073] After filtering out similar samples of the current medical record, the recurrence risk value of the current medical record at the current moment can be quantified based on the first recurrence interval of the similar samples.

[0074] Preferably, in one embodiment of the present invention, the method for obtaining the recurrence risk value includes:

[0075] First, the time interval between the current moment and the discharge moment of the current medical record sample is calculated as the interval to be tested. In this embodiment of the present invention, in the historical medical record sample, before the time of recurrence is reached, it is approximately assumed that the risk of recurrence is linearly related to time. That is, in the historical medical record sample, the closer the time is to the first recurrence moment, the higher the risk of recurrence. When the time reaches the first recurrence moment, the risk of recurrence is the highest.

[0076] Therefore, the interval to be tested is compared with the first recurrence interval of each similar sample. If the interval to be tested is greater than or equal to the first recurrence interval, the recurrence risk factor of the interval to be tested is set to a preset value (the preset value is 1 in this embodiment of the invention); if the interval to be tested is less than the first recurrence interval, the ratio of the interval to be tested to the first recurrence interval is used as the recurrence risk factor of the interval to be tested.

[0077] Finally, the mean of the recurrence risk factors between the test interval and all similar samples is taken as the recurrence risk value of the current medical record sample at the current moment. The higher the recurrence risk value, the higher the recurrence risk of the current medical record sample at the current moment.

[0078] To illustrate the above logic, consider the following example: If the time frame is in months, the test interval is 3 months, and there are two similar samples in the current medical record sample. The first recurrence interval of similar sample a is 2 months. Since the test interval is longer than the first recurrence interval of similar sample a, the recurrence risk factor for the current medical record sample should be a preset value of 1 when predicting its recurrence characteristics based on the recurrence of similar sample a. The first recurrence interval of similar sample b is 6 months. Since the test interval is shorter than the first recurrence interval of similar sample b, the recurrence risk factor for the current medical record sample should be 3 / 6. The higher the value, the higher the recurrence risk. Finally, the average of the recurrence risk factors between the current medical record sample and the two similar samples is taken as the recurrence risk value of the current medical record sample at the current moment.

[0079] Step S3: Based on the analysis of the differences in the first relapse interval, the differences in the values ​​of relapse influencing factors, and the differences in the values ​​of unhealthy habits among historical medical record samples, determine the change in relapse risk of the current medical record sample at the current time based on the differences in the values ​​of unhealthy habits between the current medical record sample at the time of discharge and the current time.

[0080] Step S2 above is an initial relapse risk assessment based on historical medical record samples. In reality, unhealthy lifestyle habits are also a key influencing factor on the relapse risk of acute leukemia patients. Unhealthy habits after discharge are often difficult to directly represent in the initial relapse risk because they are a dynamic influencing factor in the later stages, and the greater the accumulation of unhealthy habits, the higher the impact on relapse risk. Therefore, by comparing a large number of users, we can analyze the differences in unhealthy habit characteristic values, the differences in the first relapse interval, and the differences in relapse influencing factor values ​​among historical medical record samples, thereby constructing a model of the impact of unhealthy habits on relapse risk. Then, based on this, according to the differences in unhealthy habit characteristic values ​​between the current medical record sample at the time of discharge and the current time, we can determine the change in relapse risk of the current medical record sample at the current time.

[0081] Preferably, in one embodiment of the present invention, the method for obtaining the change in relapse risk includes:

[0082] Please see Figure 3 The diagram illustrates a method flowchart for obtaining the change in relapse risk according to an embodiment of the present invention, the method comprising the following steps:

[0083] Step S301: Based on the differences in recurrence influencing factor values ​​among all historical medical record samples, determine the medical record similarity parameters among historical medical record samples to divide the medical record sample groups.

[0084] For any two patients, the greater the similarity in their physical conditions, the greater the difference in the interval between their first relapses, which can be attributed to the patients' unhealthy habits to some extent.

[0085] Therefore, we can first analyze the similarity characteristics between historical medical record samples to group them: Among all historical medical record samples, the absolute value of the difference between the recurrence influencing factor values ​​of any two historical medical record samples under each recurrence influencing factor is taken as a deviation factor. The smaller the deviation factor, the greater the similarity between the two historical medical record samples under that recurrence influencing factor. We calculate the sum of the deviation factors of the two historical medical record samples under all recurrence influencing factors, and then apply a negative correlation mapping to this sum to correct the logical relationship, obtaining the medical record similarity parameters of the two historical medical record samples. Based on the aforementioned analysis, the larger the medical record similarity parameter, the higher the similarity of the patients' physical conditions represented by the two historical medical record samples. The negative correlation mapping here can be achieved using the formula... ,in, Let x represent an exponential function with the natural constant e as the base, and let x represent the independent variable.

[0086] At this point, each historical medical record sample has a medical record similarity parameter with each of the remaining historical medical record samples. Therefore, for any historical medical record sample, it is combined with the historical medical record samples corresponding to the largest medical record similarity parameter in all the remaining historical medical record samples to form a medical record sample group.

[0087] At this point, the pairing of medical record sample groups can be completed. The two historical medical record samples in each medical record sample group are highly consistent in terms of relapse influencing factors, which means that the patient status they represent is relatively consistent.

[0088] Step S302: Determine the recurrence risk change curve based on the differences in the first recurrence interval between historical medical record samples in the medical record sample group, the differences in medical record similarity parameters, and the differences in the characteristic values ​​of bad habits.

[0089] In each medical record sample group, one historical medical record sample is designated as the first sample, and the other as the second sample. The difference in the unhealthy habit characteristic values ​​between the first and second samples is taken as the change in unhealthy habits. The first relapse interval between the two historical medical record samples can approximate the relapse risk (the larger the first relapse interval, the smaller the relapse risk). Therefore, the difference in the first relapse interval can approximate the change in relapse risk. When the medical record similarity parameter between the two medical record samples is larger, it indicates that the change in the first relapse interval is more due to the change in unhealthy habit factors. In other words, the medical record similarity parameter can be used as the confidence level of the change in the first relapse interval caused by unhealthy habits. Therefore, the product of the difference in the first relapse interval between the first and second samples and the medical record similarity parameter is calculated, and the product is numerically adjusted using the hyperbolic tangent function (th function) to obtain the relapse risk change value corresponding to each medical record sample group. At this point, if the change in unhealthy habits between the first and second samples is positive, it indicates that the unhealthy habits of the second sample have improved compared to the first sample. If the change in risk between the first and second samples is negative and the smaller the value, it indicates that the relapse risk of the second sample is smaller compared to the first sample.

[0090] At this point, each medical record sample group corresponds to a set of data (change in unhealthy habits and change in relapse risk). A two-dimensional coordinate system is constructed with the change in unhealthy habits as the horizontal axis and the change in relapse risk as the vertical axis, thus mapping the data points corresponding to all medical record sample groups to the two-dimensional coordinate system. Based on the least squares method, curve fitting is performed on all data points to obtain the relapse risk change curve. The relapse risk change curve characterizes the change in the unhealthy habit characteristic value that affects the relapse risk.

[0091] It should be noted that the least squares method is a well-known technique, and the specific process will not be elaborated here.

[0092] Step S303: Map the difference in the characteristic values ​​of bad habits of the current medical record sample at the time of discharge and at the current time to the curve of recurrence risk change, thereby determining the amount of change in recurrence risk of the current medical record sample at the current time.

[0093] The difference between the unhealthy habit characteristic values ​​of the current medical record sample at the current time and at the time of discharge is taken as the test value. The test value reflects the change in unhealthy habits at the current time compared to the time of discharge. Then, the value of the vertical axis corresponding to the test value in the recurrence risk change curve is taken as the change in recurrence risk of the current medical record sample at the current time. Based on the analysis in step S302, it can be seen that the smaller the negative value of the change in recurrence risk, the greater the degree of reduction in recurrence risk at the current time compared to the time of discharge.

[0094] Step S4: Integrate the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment to perform stratified early warning.

[0095] In step S2, the initial relapse risk of the current medical record sample at the current moment can be quantified to obtain the relapse risk value. In step S3, by analyzing the influence of bad habit characteristics, the change in the relapse risk at the current moment compared to the discharge moment can be obtained. Therefore, the two can be integrated to obtain a more comprehensive indicator for stratified early warning of the relapse risk of the current medical record sample at the current moment.

[0096] Preferably, in one embodiment of the present invention, the stratified early warning is achieved by integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, including:

[0097] By combining the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, a recurrence risk indicator for the current medical record sample at the current moment is determined: a larger recurrence risk value indicates a higher risk of recurrence; a negative and smaller change in recurrence risk indicates a greater reduction in the risk of recurrence at the current moment compared to the time of discharge. Therefore, the normalized sum of the recurrence risk value and the change in recurrence risk of the current medical record sample at the current moment is used as the recurrence risk indicator for the current medical record sample at the current moment. The normalization method used here is... function.

[0098] When the recurrence risk index is greater than or equal to the preset level 3 risk threshold, a level 3 warning is triggered. When the recurrence risk index is less than the preset level 3 risk threshold but greater than or equal to the preset level 2 risk threshold, a level 2 warning is triggered. When the recurrence risk index is less than the preset level 2 risk threshold, a level 1 risk warning is triggered. The severity of the level 1, level 2, and level 3 risk warnings increases sequentially.

[0099] It should be noted that in this embodiment of the present invention, the preset third-level risk threshold is 0.6, the preset second-level risk threshold is 0.3, and the preset first-level risk threshold is 0.1. The specific values ​​can be adjusted according to the implementation scenario and are not limited here.

[0100] In summary, we first obtained historical and current medical record samples. Common methods assess future relapse risk by evaluating relapse-influencing factors (such as age and examination indicators) in these samples. In general medical practice, the relapse risk of current patients is often assessed based on the relapse history of previous patients. This can be summarized as medical record samples with similar conditions exhibiting similar recovery abilities, and therefore, similar relapse risks. Thus, by utilizing the similarity of relapse-influencing factors and the first relapse interval between historical and current medical record samples, we adjusted the similarity features between them to identify similar samples for the current medical record. Based on the first relapse interval of these similar samples, we can preliminarily estimate the potential relapse risk value of the current medical record. Furthermore, since patients' unhealthy habits are also a key influencing factor on the relapse risk of acute leukemia patients, and this characteristic is a dynamic factor that cannot be directly reflected in the aforementioned relapse risk values, it needs to be analyzed separately. Based on a comprehensive analysis of the differences in three characteristics among historical medical record samples—unhealthy habit characteristics, first relapse interval, and relapse influencing factors—this study clarifies the specific change in relapse risk due to unhealthy habits in the current medical record sample, considering the differences in unhealthy habit characteristics between the current sample and the present moment. Finally, this change in relapse risk is integrated with the relapse risk value to determine a more accurate risk assessment characteristic parameter that accurately reflects the patient's actual condition. When used for stratified early warning of relapse risk in the current medical record sample at the current moment, this parameter can effectively improve the accuracy of early warning.

[0101] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0102] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for risk stratification and early warning of acute leukemia relapse, characterized in that, The method includes: Obtain historical medical record samples, including the interval between the first relapses, the characteristic values ​​of unhealthy habits at the time of the first relapse, and the values ​​of various factors influencing relapse; obtain current medical record samples, including the characteristic values ​​of unhealthy habits at the time of discharge, the characteristic values ​​of unhealthy habits at the current time, and the values ​​of various factors influencing relapse; By utilizing the similarity of relapse influencing factor values ​​and the similarity of first relapse intervals among historical medical record samples, the similarity characteristics of relapse influencing factor values ​​between historical and current medical record samples are adjusted to screen out similar samples for the current medical record sample; the relapse risk value of the current medical record sample at the current time is determined based on the first relapse interval of similar samples. Based on the analysis of differences in the first relapse interval, relapse influencing factor values, and unhealthy habit characteristic values ​​among historical medical record samples, the change in relapse risk for the current medical record sample at the current time is determined according to the differences in unhealthy habit characteristic values ​​between the current medical record sample and the current time. This includes: determining medical record similarity parameters among all historical medical record samples based on the differences in relapse influencing factor values ​​to divide the medical record sample groups; within each medical record sample group, selecting one historical medical record sample as the first sample and the other as the second sample; and comparing the unhealthy habit characteristic values ​​between the first and second samples. The difference between the first and second samples is used as the change in unhealthy habits. The product of the difference in the first relapse interval between the first and second samples and the medical record similarity parameter is adjusted using the hyperbolic tangent function to obtain the change in relapse risk value. A two-dimensional coordinate system is constructed with the change in unhealthy habits as the horizontal axis and the change in relapse risk value as the vertical axis, thereby mapping the data points corresponding to all medical record sample groups to the two-dimensional coordinate system. Curve fitting is performed on all data points to obtain the relapse risk change curve. The difference between the unhealthy habit feature values ​​of the current medical record sample at the current time and the discharge time is used as the test value. The value of the vertical axis corresponding to the test value in the relapse risk change curve is used as the change in relapse risk of the current medical record sample at the current time. By integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, a stratified early warning system can be implemented.

2. The method for risk stratification and early warning of acute leukemia relapse according to claim 1, characterized in that, The method for obtaining similar samples includes: For each recurrence influencing factor, historical medical records with the same recurrence influencing factor value are grouped into one category, thus obtaining a cluster. For each relapse influencing factor, the similarity of the first relapse interval of historical medical record samples in the cluster was analyzed, and the importance index of each relapse influencing factor was determined by combining it with the quantitative characteristics of the cluster. For each relapse influencing factor, the absolute value of the difference between the relapse influencing factor values ​​of each historical medical record sample and the current medical record sample is negatively correlated and mapped to the value, which is used as the medical record similarity value between each historical medical record sample and the current medical record sample. The similarity values ​​between historical and current medical record samples were adjusted using the importance index of relapse influencing factors, thereby obtaining the similarity feature value between each historical and current medical record sample. Among all historical medical record samples, those with similarity feature values ​​greater than a preset similarity threshold are considered as similar samples to the current medical record sample.

3. The method for risk stratification and early warning of acute leukemia relapse according to claim 2, characterized in that, The methods for obtaining the importance indicators include: In each cluster of each relapse influencing factor, the absolute value of the difference between the first relapse intervals of any two historical medical record samples is negatively correlated and mapped to the value, which is used as the relapse similarity factor between the two historical medical record samples. The normalized value of the ratio of the mean recurrence similarity factor among all historical medical record samples in all clusters of each recurrence influencing factor to the number of clusters is used as the importance indicator for each recurrence influencing factor.

4. The method for risk stratification and early warning of acute leukemia relapse according to claim 2, characterized in that, The method for obtaining the similarity feature values ​​includes: Under all factors influencing recurrence, the similarity values ​​between each historical medical record sample and the current medical record sample are weighted and fused using the importance index of the factors influencing recurrence, thereby obtaining the similarity feature value between each historical medical record sample and the current medical record sample.

5. The method for risk stratification and early warning of acute leukemia relapse according to claim 1, characterized in that, The method for obtaining the recurrence risk value includes: Calculate the time interval between the current time and the discharge time of the current medical record sample, and use it as the test interval; The test interval is compared with the first recurrence interval of each similar sample. If the test interval is greater than or equal to the first recurrence interval, the recurrence risk factor of the test interval is set to a preset value; if the test interval is less than the first recurrence interval, the ratio of the test interval to the first recurrence interval is used as the recurrence risk factor of the test interval. The mean of the recurrence risk factors between the interval to be tested and all similar samples is used as the recurrence risk value of the current medical record sample at the current time.

6. The method for risk stratification and early warning of acute leukemia relapse according to claim 1, characterized in that, The method for obtaining the medical record sample group includes: In all historical medical record samples, the absolute value of the difference between the values ​​of the recurrence influencing factors under each recurrence influencing factor for any two historical medical record samples is taken as the deviation factor. The sum of the deviation factors of these two historical medical record samples under all recurrence influencing factors is negatively correlated and mapped to the value of the medical record similarity parameter between the two historical medical record samples. Each historical medical record sample is grouped together with the historical medical record sample corresponding to the maximum value among all similarity parameters of all medical records.

7. The method for risk stratification and early warning of acute leukemia relapse according to claim 1, characterized in that, The method of integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment to perform stratified early warning includes: By integrating the change in recurrence risk and the recurrence risk value of the current medical record sample at the current moment, the recurrence risk index of the current medical record sample at the current moment is determined. When the recurrence risk index is greater than or equal to the preset level 3 risk threshold, a level 3 warning is triggered. When the recurrence risk index is less than the preset level 3 risk threshold but greater than or equal to the preset level 2 risk threshold, a level 2 warning is triggered. When the recurrence risk index is less than the preset level 2 risk threshold, a level 1 risk warning is triggered. The severity of the level 1, level 2, and level 3 risk warnings increases sequentially.

8. The method for risk stratification and early warning of acute leukemia relapse according to claim 7, characterized in that, The methods for obtaining the recurrence risk indicators include: The normalized sum of the recurrence risk value and the change in recurrence risk of the current medical record sample at the current moment is used as the recurrence risk indicator of the current medical record sample at the current moment.

Images