WO2021073255A1 - 基于时序聚类的用药提醒方法及相关设备 - Google Patents

基于时序聚类的用药提醒方法及相关设备 Download PDF

Info

Publication number
WO2021073255A1
WO2021073255A1 PCT/CN2020/111324 CN2020111324W WO2021073255A1 WO 2021073255 A1 WO2021073255 A1 WO 2021073255A1 CN 2020111324 W CN2020111324 W CN 2020111324W WO 2021073255 A1 WO2021073255 A1 WO 2021073255A1
Authority
WO
WIPO (PCT)
Prior art keywords
medication
medication compliance
data
compliance
time series
Prior art date
Application number
PCT/CN2020/111324
Other languages
English (en)
French (fr)
Inventor
徐卓扬
赵惟
孙行智
左磊
刘卓
胡岗
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021073255A1 publication Critical patent/WO2021073255A1/zh

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61JCONTAINERS SPECIALLY ADAPTED FOR MEDICAL OR PHARMACEUTICAL PURPOSES; DEVICES OR METHODS SPECIALLY ADAPTED FOR BRINGING PHARMACEUTICAL PRODUCTS INTO PARTICULAR PHYSICAL OR ADMINISTERING FORMS; DEVICES FOR ADMINISTERING FOOD OR MEDICINES ORALLY; BABY COMFORTERS; DEVICES FOR RECEIVING SPITTLE
    • A61J7/00Devices for administering medicines orally, e.g. spoons; Pill counting devices; Arrangements for time indication or reminder for taking medicine
    • A61J7/04Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers
    • A61J7/0409Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers with timers
    • A61J7/0418Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers with timers with electronic history memory
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61JCONTAINERS SPECIALLY ADAPTED FOR MEDICAL OR PHARMACEUTICAL PURPOSES; DEVICES OR METHODS SPECIALLY ADAPTED FOR BRINGING PHARMACEUTICAL PRODUCTS INTO PARTICULAR PHYSICAL OR ADMINISTERING FORMS; DEVICES FOR ADMINISTERING FOOD OR MEDICINES ORALLY; BABY COMFORTERS; DEVICES FOR RECEIVING SPITTLE
    • A61J7/00Devices for administering medicines orally, e.g. spoons; Pill counting devices; Arrangements for time indication or reminder for taking medicine
    • A61J7/04Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers
    • A61J7/0409Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers with timers
    • A61J7/0481Arrangements for time indication or reminder for taking medicine, e.g. programmed dispensers with timers working on a schedule basis

Definitions

  • This application relates to the field of artificial intelligence, and in particular to a medication reminder method and related equipment based on time series clustering.
  • the chronic disease management system is a chronic disease management network system developed and designed for general hospitals and specialized hospitals to improve the treatment effect of chronic diseases.
  • patients with chronic diseases patients often take medications many times, and their medications may not always follow the doctor's instructions.
  • the patient may sometimes take the medicine as prescribed by the doctor, and sometimes not take the medicine as prescribed by the doctor. Reminding patients to follow doctor's prescriptions is the key to safe medication.
  • Adding medication reminder module to the chronic disease management system helps to improve the effectiveness of chronic disease treatment. The inventor found that when the current chronic disease management system reminds patients about medication, most of them are reminded according to the medication cycle and at the medication change node, but the time of medication reminder is not necessarily the time that has the greatest impact on the patient's medication effect. point.
  • the purpose of this application is to address the deficiencies of the prior art and provide a medication reminder method and related equipment based on time series clustering.
  • the medication compliance mode is obtained and combined with patients Obtain the medication compliance classification model based on the basic information of the medication compliance classification model.
  • the medication reminder can be used to remind the patient of key secondary medications, which can reduce the cost of medication reminders and increase the patient’s attention to improve disease control. Effectiveness, reducing the risk of complications and adverse outcomes.
  • the technical solution of the present application provides a medication reminder method and related equipment based on time series clustering.
  • This application discloses a medication reminder method based on time series clustering, which includes the following steps:
  • each medication compliance mode includes at least two medication compliance modes, and the medication compliance mode includes a plurality of medication time points to be used in the medication time points Remind users about medication;
  • the application also discloses a medication reminder device based on time series clustering, the device comprising:
  • Acquisition module set to acquire medication compliance data, clinical data, basic information data, and baseline index data of sample users, and obtain medication compliance time series based on the medication compliance data;
  • Clustering module configured to cluster the medication compliance time series to obtain a medication compliance pattern set, and perform a logistic regression analysis on the medication compliance pattern set according to the basic information data and clinical data to obtain the The therapeutic effect category of each medication compliance mode in the medication compliance mode set, wherein the medication compliance mode set includes at least two medication compliance modes, and the medication compliance mode includes multiple medication time points for use in The user is reminded of medication at the medication time point;
  • Training module set to train the medication compliance pattern set according to the basic information data and baseline index data to obtain a medication compliance pattern classification model:
  • Reminder module set to obtain the basic information data and baseline data of the target user, and input the basic information data and baseline data of the target user into the medication compliance pattern classification model to obtain the medication compliance pattern of the target user, A medication reminder is given to the target user according to the medication compliance mode of the target user.
  • the application also discloses a computer device, the computer device includes a memory and a processor, the memory is stored with computer readable instructions, when the computer readable instructions are executed by one or more of the processors, One or more of the processors execute the steps of the medication reminder method based on time series clustering as described below:
  • each medication compliance mode includes at least two medication compliance modes, and the medication compliance mode includes a plurality of medication time points to be used in the medication time points Remind users about medication;
  • the application also discloses a storage medium that can be read and written by a processor, and the storage medium stores computer instructions.
  • the computer-readable instructions are executed by one or more processors, one or more Each processor executes the steps of the medication reminder method based on time series clustering as described below:
  • each medication compliance mode includes at least two medication compliance modes, and the medication compliance mode includes a plurality of medication time points to be used in the medication time points Remind users about medication;
  • this application obtains a medication compliance model by performing time series clustering on medication compliance data of a large number of patients, and combines the basic information of the patient to obtain a medication compliance classification model, according to the medication compliance classification model Reminders on medications to patients can focus on the key secondary medications of patients. While reducing the cost of medication reminders, it can increase the patient's attention, so as to improve the effectiveness of disease control and reduce the risk of complications and adverse outcomes.
  • FIG. 1 is a schematic flow chart of a medication reminder method based on time series clustering according to the first embodiment of this application;
  • FIG. 2 is a schematic flowchart of a medication reminder method based on time series clustering according to a second embodiment of this application;
  • FIG. 3 is a schematic flowchart of a medication reminder method based on time series clustering according to a third embodiment of this application;
  • FIG. 4 is a schematic flowchart of a medication reminder method based on time series clustering according to a fourth embodiment of this application;
  • FIG. 5 is a schematic flowchart of a medication reminder method based on time series clustering according to a fifth embodiment of this application;
  • FIG. 6 is a schematic flowchart of a medication reminder method based on time series clustering according to a sixth embodiment of this application;
  • FIG. 7 is a schematic flowchart of a medication reminder method based on time series clustering according to a seventh embodiment of this application.
  • FIG. 8 is a schematic structural diagram of a medication reminder device based on time series clustering according to an embodiment of the application.
  • FIG. 1 The flow of a medication reminder method based on time series clustering according to the first embodiment of the present application is shown in FIG. 1. This embodiment includes the following steps:
  • Step s101 Obtain medication compliance data, clinical data, basic information data, and baseline index data of sample users, and obtain medication compliance time series according to the medication compliance data;
  • the sample user refers to a patient, and can obtain a large number of patient data from the big data of the hospital's patient management system; when extracting the patient's medication compliance data, the time period can be specified first, for example, Number within 1 year; that is, obtain medication compliance data for each patient within 1 year after diagnosis.
  • the "medication compliance data” refers to data on whether the patient's medication is in compliance with the doctor's order, and the medication compliance data of a patient is multiple pieces of data in the form of (TimeInterval, IsObey).
  • TimeInterval represents the time from the patient's enrollment (diagnosed) to a certain medication, the unit is "days”;
  • IsObey represents whether the patient is taking the medication according to the doctor's order, compliance is 1, and non-compliance is 0.
  • a patient was confirmed 4 times of medication through follow-up, and learned that he followed the doctor's prescription on the 5th day after enrollment, followed the doctor's prescription on the 30th day, did not follow the doctor's prescription on the 60th day, and followed the doctor's prescription on the 100th day . Then the patient's medication compliance data are (5,1), (30,1), (60,0), (100,1).
  • the "clinical data” includes clinical index data and clinical result data.
  • the clinical index data refers to the current clinical index of the patient after treatment.
  • the "clinical indicators after n years of medication” are: the average value of HbA1c (glycated hemoglobin) from n years after diagnosis to n+1 years after diagnosis;
  • the clinical outcome data refers to the results of patients after treatment, For example, for patients with diabetic nephropathy, whether ESRD (end-stage renal disease) occurs between n years after diagnosis and n+1 years after diagnosis.
  • the "basic information data” includes data such as gender, age, height, weight, whether to smoke, whether to drink, eating habits, past medical history, complications, and so on.
  • the "baseline index data” includes the values of various test inspection indexes when the patient is enrolled in the group: glycosylated hemoglobin, urine protein, blood uric acid, cholesterol, triglycerides, blood creatinine, fasting blood glucose, blood pressure and other data.
  • the medication compliance data of each user can be linearly interpolated to obtain the medication compliance time series of all users.
  • Step s102 cluster the medication compliance time series to obtain a medication compliance pattern set, and perform a logistic regression analysis on the medication compliance pattern set based on the basic information data and clinical data to obtain the medication compliance
  • the curative effect category of each medication compliance mode in the mode set wherein the medication compliance mode set includes at least two medication compliance modes, and the medication compliance mode includes multiple medication time points for use in the medication Remind users about medication at the time point;
  • the medication compliance time series of the sample users can be clustered, and the clustering method can be k-means, that is, k-means clustering.
  • the idea of the k-means algorithm is to divide the sample set into k clusters according to the distance between the samples for a given sample set. Make the points in the clusters as close together as possible, and make the distance between the clusters as large as possible. If expressed by a data expression, assuming that the cluster is divided into (C1, C2,...Ck), the goal of the traditional k-means algorithm is to minimize the square error E.
  • the logistic logistic regression model can be used to perform data analysis to obtain the pros and cons of each medication compliance mode, for example, which drugs are used in the medication compliance mode set Compliance modes are relatively good therapeutic, which medication compliance modes are relatively poorly therapeutic, and which medication compliance modes have no therapeutic effect, wherein the medication compliance mode set includes at least two medication compliance Mode, the medication compliance mode includes multiple medication time points, and is used to remind the user of medication at the medication time points.
  • Step s103 training the medication compliance pattern set according to the basic information data and baseline index data to obtain a medication compliance pattern classification model
  • the medication compliance mode set can be trained through a decision tree.
  • the decision tree is a CART classification decision tree model.
  • the decision tree is a tree structure used for classification.
  • Each internal node represents a pair For a test (judgment) of a certain attribute (feature), each edge represents a test result, and the leaf node represents a certain class or the distribution of a class.
  • the decision-making process of the decision tree needs to start from the root node of the decision tree.
  • the data to be tested is compared with the characteristic nodes in the decision tree, and the next comparison branch is selected according to the comparison result until the leaf node is the final decision result.
  • the basic information data and baseline index data of all users can be obtained as the features for classification;
  • the classification goal is to determine which medication compliance mode the user belongs to in the cluster, and then search for the best segmentation according to the Gini index minimization criterion Index and segmentation value, split the training set into two subsets. This process is repeated recursively in the generated subset, that is, recursive segmentation.
  • the stop condition is reached or the class labels of a training subset are the same, the recursion stops, and finally the medication compliance pattern classification model is obtained.
  • Step s104 Obtain basic information data and baseline data of the target user, and input the basic information data and baseline data of the target user into the medication compliance pattern classification model to obtain the medication compliance pattern of the target user.
  • the medication compliance mode of the target user is described to remind the target user of medication.
  • any user can be classified into the medication compliance pattern.
  • the information data and baseline data are input into the medication compliance pattern classification model to obtain the medication compliance pattern of the current user to be classified, that is, which medication compliance mode in the medication compliance mode set belongs to the current user to be classified, the The medication compliance mode is a time series trajectory, and finally, a medication reminder of the user to be classified is performed according to the identified medication compliance mode, and the medication reminder includes the time period in which the user is reminded of medication.
  • the medication compliance model is obtained by clustering the medication compliance data of a large number of patients, and the medication compliance classification model is obtained in combination with the basic information of the patient, and the medication is administered to the patient according to the medication compliance classification model.
  • Reminders can be used to emphatically remind patients of key secondary medications, which can reduce the cost of medication reminders while increasing the patient's attention, so as to improve the effectiveness of disease control and reduce the risk of complications and adverse outcomes.
  • Figure 2 is a schematic flow chart of a medication reminder method based on time series clustering according to a second embodiment of the application.
  • step s101 medication compliance data, clinical data, basic information data and data of sample users are obtained.
  • the baseline index data which obtains the medication compliance time series according to the medication compliance data, includes:
  • a first time period and a second time period are preset.
  • the first time period is used to obtain medication compliance data in any continuous time period in the historical data
  • the second time period is used to obtain the historical data.
  • two time periods can be preset, a first time period and a second time period.
  • the first time period is used to obtain medication compliance data in any continuous time period in the historical data
  • the second time period Segments are used to obtain clinical data in any continuous period of time in historical data.
  • Step s202 Obtain basic information data and baseline index data of the sample user, medication compliance data in the first time period, and clinical data in the second time period;
  • the basic information data and baseline index data can be obtained when the user sees a doctor, so there is no need to set a time period.
  • the first time period can be obtained.
  • the medication compliance data of the sample user in a period of time similarly, after the second period of time is set, the clinical data of the sample user in the second period of time can be obtained.
  • Step s203 Perform linear interpolation on the medication compliance data of the sample users in the first time period to obtain the medication compliance time series of the sample users.
  • the medication compliance data of the sample user can be measured with the field TimeInterval as the x-axis, the field IsObey as the y-axis, and "days" as the discrimination
  • the rate is linearly interpolated.
  • the sample users can get a time series of medication compliance with a time interval of days as the unit. For example: a patient's medication compliance data: (5,1),(30,1),(60,0),(100,1).
  • the medication compliance time series after linear interpolation is: (5,1),(6,1),(7,1)...(30,1),(31,0.967),(32,0.933),...( 60,0), (61,0.025), (62,0.05), (63,0.075)...(99,0.975), (100,1).
  • the user's medication compliance data is linearly interpolated to obtain the medication compliance time series, which is beneficial to clustering the medication compliance data.
  • step s102 the medication compliance time series are clustered to obtain medication compliance A collection of patterns, including:
  • Step s301 taking the medication compliance time series of the sample user as a sample
  • the data of each user can be regarded as a sample.
  • the medication compliance time series data of each user can be regarded as a sample.
  • Step s302 according to the formula for the sample Perform clustering, where In clustering, the square error E is minimized by calculating the dynamic time planning distance, and a penalty item is added to the dynamic time planning distance to obtain the medication compliance mode set.
  • Ci is the cluster in the cluster
  • I the mean vector of the cluster Ci, which can also be the center of mass.
  • the penalty term a*ln(t+1) is added when matching at the time difference t to increase the distance, where 0 ⁇ a ⁇ 1 is a parameter that controls the degree of punishment.
  • the state transition equation is:
  • DTW[i,j]: cost+minimum(DTW[i-1,j]//insertion
  • DTW[i,j]: cost+minimum(DTW[i-1,j]+a*ln
  • +1) can be called "single-step displacement penalty".
  • the ln function is used because when
  • 0 ⁇ a ⁇ 1 is the input parameter, which is the coefficient used to adjust the penalty caused by the displacement.
  • the user samples can be clustered more accurately.
  • step s102 the medication compliance time series are clustered to obtain medication compliance A collection of patterns, including:
  • Step s401 when clustering the medication compliance time series, obtain multiple medication compliance model candidate sets
  • clustering can be performed with different k values to obtain multiple candidate sets of medication compliance patterns, where k represents the number of medication compliance patterns. Since each patient has only one medication compliance time series, the patient belongs to only one mode.
  • step s402 among the multiple candidate sets of medication compliance mode, obtain the final medication compliance mode set according to the DB-Index criterion.
  • the number of medication compliance modes is determined according to the DB-Index criteria, and the user medication compliance modes can be clustered more accurately.
  • step s102 is based on the basic information data and clinical data to the medication compliance mode Collect and perform logistic regression analysis to obtain the curative effect category of each medication compliance mode in the medication compliance mode set, including:
  • Step s501 perform statistics on the clinical data of the sample user, and obtain the incidence rate of bad results of the sample user;
  • the adverse clinical results of the sample users can be inquired, and the adverse clinical results include various diseases or clinical indicators.
  • the incidence of adverse clinical results of the user can be obtained, for example, the occurrence of ESRD Rate, HbA1c substandard rate.
  • step s502 logistic regression is performed on the medication compliance mode set according to the incidence of the adverse results, and the curative effect category of each medication compliance mode in the medication compliance mode set is obtained.
  • the mode with the lowest incidence of adverse clinical results can be set as the reference mode, and then the data of each user can be used as a sample, and "which mode does the patient belong to" as the analysis feature and become a dummy variable, and add it Confounding factor (age, gender, etc.), do logistic regression based on whether the patient has a bad clinical result as y, the confounding factor data can be obtained from the basic information data, and finally judge the bad according to the p-value value in each mode Whether the incidence of clinical outcome is significantly higher than that of the reference model; and to determine the pros and cons of each medication compliance model; for example, suppose the clustering result is diabetic nephropathy, and the user in model 4 has the lowest incidence of ESRD (20.23%) ), then use mode 4 as the reference mode.
  • the feature of "which mode the patient belongs to” can be converted into 3 features by performing dummy variable conversion and deleting the reference column. Taking these three characteristics, age and gender (confounding factors) as the characteristics used in training, and taking the individual patient's clinical outcome (whether ESRD occurred) as y, a logistic regression analysis was performed. According to the p-value of these three analysis characteristics (mode 1, mode 2, mode 3), it can be defined whether the incidence of these three modes is significantly higher than that of the reference mode ESRD, as shown in Table 1.
  • step s104 the target user is reminded of medication according to the medication compliance mode of the target user ,include:
  • Step s601 Identify the medication compliance pattern of the target user according to the curative effect category, the curative effect category including poor treatment effectiveness and good treatment effectiveness;
  • the target user compliance mode can be identified, that is, it can be identified whether the medication compliance mode of the target user is poor in treatment effectiveness or good in treatment effectiveness.
  • Step s602 When the curative effect category of the target user is poor treatment effectiveness, the target user is treated according to the formula Do a medication reminder, where n is the time frequency and IsObey(t) is the compliance degree at time t.
  • the patient can be reminded to take medication at more time points, that is, during the time period when the medication reminder should be performed , The patient can be reminded to take the medicine at the frequency of once every n weeks at the time point t, and the calculation formula of n is Among them, IsObey(t) is the medication compliance of the medication compliance time series trajectory at time point t. Because the mode of poor treatment effectiveness tends to have lower medication compliance, the frequency of medication reminders for patients in this mode tends to be lower. higher. Modes with poor treatment effectiveness often require more time periods for medication reminders, and more reminders are given to patients belonging to this mode.
  • the medication effect of users with poor therapeutic effectiveness can be effectively improved.
  • FIG. 7 is a schematic flow chart of a medication reminder method based on time series clustering according to a seventh embodiment of the application. As shown in the figure, in step s104, the target user is reminded of medication according to the medication compliance mode of the target user ,include:
  • Step s701 preset a first threshold and a second threshold, the first threshold is used to compare the slope in the medication compliance mode, and the second threshold is used to compare the compliance;
  • two thresholds can be preset, a first threshold and a second threshold.
  • the first threshold is used to determine the magnitude of the slope of the trajectory in the medication compliance mode
  • the second threshold is used to determine the degree of medication compliance. the size of.
  • Step s702 obtaining a first medication time period corresponding to less than the first threshold and a second medication time period corresponding to less than the second threshold in the medication compliance mode of the target user;
  • the magnitude of each slope in the medication compliance mode of the target user can be obtained, and then the magnitude of each slope of the slope can be compared with the first threshold, and the slope less than the first threshold can be determined. Record it, and find out the time period corresponding to the slope less than the first threshold, and the time period is the first medication time period; then it can be found in the medication compliance mode of the target user that is less than the The medication compliance degree of the second threshold value, and the time period corresponding to the medication compliance degree is found, and the time period is the second medication time period.
  • step s703 a medication reminder is given to the target user during the first medication time period and the second medication time period.
  • the user can be reminded of medication in the medication time period, for example, to find a time period with a slope of ⁇ 0, during which the patient as a whole tends to not follow the doctor’s prescription for medication.
  • the center line drops; find the time period when the compliance degree is less than 0.5.
  • the patient group is increasingly complying with the doctor's prescription, the overall compliance is still at a low level, and the possibility of individual patients not following the doctor's prescription is still relatively high. Big.
  • the medication efficiency of the patient can be effectively improved.
  • FIG. 8 The structure of a medication reminder device based on time series clustering according to an embodiment of the application is shown in FIG. 8, and includes:
  • the embodiment of the present application also discloses a computer device, the computer device includes a memory and a processor, and computer-readable instructions are stored in the memory.
  • the computer-readable instructions are executed by one or more of the processors , Enabling one or more of the processors to execute the steps in the medication reminding method in the foregoing embodiments.
  • the embodiment of the present application also discloses a storage medium, the storage medium can be read and written by a processor, and the memory stores computer readable instructions.
  • the computer readable instructions are executed by one or more processors, One or more processors execute the steps in the medication reminder method described in the foregoing embodiments.
  • the computer program can be stored in a computer readable storage medium. When executed, it may include the procedures of the above-mentioned method embodiments.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or other volatile solid-state storage devices.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种基于时序聚类的用药提醒方法及相关设备,方法包括:获取样本用户的用药依从性数据,对用药依从性数据进行聚类,获得用药依从性模式,根据目标用户的用药依从性模式对目标用户进行用药提醒。可以对患者的关键次用药进行着重提醒,在降低用药提醒成本的同时提高患者重视,以提升病情控制的有效性,降低发生并发症与不良结局的风险。

Description

基于时序聚类的用药提醒方法及相关设备
本申请要求于2019年10月18日提交中国专利局、申请号为201910991530.X、发明名称为“基于时序聚类的用药提醒方法及相关设备”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及人工智能领域,特别涉及一种基于时序聚类的用药提醒方法及相关设备。
背景技术
慢性疾病管理系统是一种为综合性医院及专科医院为改进慢性病治疗效果而开发设计的慢性疾病管理网络系统。对于慢性病患者,患者往往会多次用药,其用药不一定是一直依从医嘱的。患者可能有时会按照医嘱用药,有时不按照医嘱用药。提醒病人遵循医嘱用药是安全用药的关键,在慢病管理系统中加入用药提醒模块有助于提高慢病治疗的有效性。发明人发现,目前的慢病管理系统在对病人进行用药提醒时,多数是按用药周期和在换药节点进行提醒,但这样用药提醒的时间点并不一定是对患者用药效果影响最大的时间点。而且,频繁地对所有慢病患者进行周期性的用药提醒是冗长而昂贵的,而且没有着重点的周期性用药提醒很可能得不到患者重视。故在进行用药提醒时,需要对患者最需要用药提醒的时间段进行鉴定,对患者的关键次用药进行着重提醒,才能在降低用药提醒成本的同时提高患者重视,以提升病情控制的有效性,降低发生并发症与不良结局的风险。
发明内容
本申请的目的在于针对现有技术的不足,提供一种基于时序聚类的用药提醒方法及相关设备,通过对大量患者的用药依从性数据进行时序聚类,获得用药依从性模式,并结合患者的基本信息获得用药依从性分类模型,根据所述用药依从性分类模型对患者进行用药提醒,可以对患者的关键次用药进行着重提醒,在降低用药提醒成本的同时提高患者重视,以提升病情控制的有效性,降低发生并发症与不良结局的风险。
为达到上述目的,本申请的技术方案提供一种基于时序聚类的用药提醒方法及相关设备。
本申请公开了一种基于时序聚类的用药提醒方法,包括以下步骤:
获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
本申请还公开了一种基于时序聚类的用药提醒装置,所述装置包括:
获取模块:设置为获取样本用户的用药依从性数据、临床数据、基本信息数据及基 线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
聚类模块:设置为对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
训练模块:设置为根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型:
提醒模块:设置为获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
本申请还公开了一种计算机设备,所述计算机设备包括存储器和处理器,所述存储器中存储有计算机可读指令,所述计算机可读指令被一个或多个所述处理器执行时,使得一个或多个所述处理器执行如下所述的基于时序聚类的用药提醒方法的步骤:
获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
本申请还公开了一种存储介质,所述存储介质可被处理器读写,所述存储介质存储有计算机指令,所述计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行如下所述的基于时序聚类的用药提醒方法的步骤:
获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
本申请的有益效果是:本申请通过对大量患者的用药依从性数据进行时序聚类,获得用药依从性模式,并结合患者的基本信息获得用药依从性分类模型,根据所述用药依从性 分类模型对患者进行用药提醒,可以对患者的关键次用药进行着重提醒,在降低用药提醒成本的同时提高患者重视,以提升病情控制的有效性,降低发生并发症与不良结局的风险。
附图说明
图1为本申请第一个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图2为本申请第二个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图3为本申请第三个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图4为本申请第四个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图5为本申请第五个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图6为本申请第六个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图7为本申请第七个实施例的一种基于时序聚类的用药提醒方法的流程示意图;
图8为本申请实施例的一种基于时序聚类的用药提醒装置结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本技术领域技术人员可以理解,除非特意声明,这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是,本申请的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作、元件和/或组件,但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作、元件、组件和/或它们的组。
本申请第一个实施例的一种基于时序聚类的用药提醒方法流程如图1所示,本实施例包括以下步骤:
步骤s101,获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
具体的,所述样本用户指的是患者,即可在医院的患者管理系统的大数据中获取大量患者的数据;在对患者的用药依从性数据进行提取时,可首先指定时间段,例如,1年内的数;即获取每个患者确诊后1年内的用药依从性数据。
具体的,所述“用药依从性数据”指的是患者用药是否依从医嘱的数据,一个患者的用药依从性数据是多条以(TimeInterval,IsObey)形式存在的数据。其中,字段TimeInterval表示患者入组(确诊)后到某次用药经历的时间,单位为“天”;字段IsObey表示该次用药患者是否依从医嘱用药,依从为1,不依从为0。例如:对一个患者通过随访确认了4次用药情况,得知他在入组后的第5天依从医嘱用药,第30天依从医嘱用药,第60天未依从医嘱用药,第100天依从医嘱用药。则该患者的用药依从性数据为(5,1),(30,1),(60,0),(100,1)。
具体的,所述“临床数据”包括临床指标数据及临床结果数据,所述临床指标数据指的是患者经治疗后的当前临床指标,例如,对糖尿病肾病患者,用药治疗的目的是降糖与保肾,则“用药n年后的临床指标”为:确诊后n年到确诊后n+1年间HbA1c(糖化血红蛋白)的平均值;所述临床结果数据指的是患者经治疗后的结果,例如,对糖尿病肾病患者,确诊后n年到确诊后n+1年间是否发生ESRD(终末期肾病)。
具体的,所述“基本信息数据”包括:性别、年龄、身高、体重、是否抽烟、是否喝酒、饮食习惯、既往病史、并发症等数据。
具体的,所述“基线指标数据”包括患者入组时的各检验检查指标值:糖化血红蛋白、 尿蛋白、血尿酸、胆固醇、甘油三酯、血肌酐、空腹血糖、血压等数据。
具体的,当获取到样本用户的用药依从性数据后,可对每个用户的用药依从性数据进行线性插值,获得所有用户的用药依从性时间序列。
步骤s102,对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
具体的,当获取到样本用户的用药依从性时间序列之后,可对样本用户的用药依从性时间序列进行聚类,所述聚类的方式可采用k-means,即进行k-means聚类,获得样本用户的用药依从性的多种模式集合;其中,k-means算法的思想是,对于给定的样本集,按照样本之间的距离大小,将样本集划分为k个簇。让簇内的点尽量紧密的连在一起,而让簇间的距离尽量的大。如果用数据表达式表示,假设簇划分为(C1,C2,...Ck),则传统的k-means算法的目标是最小化平方误差E。
具体的,当通过聚类获取到用药依从性模式集合之后,可通过logistic逻辑回归模型进行数据分析,获得各用药依从性模式的优劣分级,例如,在所述用药依从性模式集合中哪些用药依从性模式是治疗性比较好的,哪些用药依从性模式是治疗性比较差的,哪些用药依从性模式是没有治疗效果的,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒。
步骤s103,根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
具体的,可通过决策树对所述用药依从性模式集合进行训练,所述决策树是一个CART分类决策树模型,决策树是运用于分类的一种树结构,其中的每个内部节点代表对某一属性(特征)的一次测试(判断),每条边代表一个测试结果,叶节点代表某个类或类的分布。决策树的决策过程需要从决策树的根节点开始,待测数据与决策树中的特征节点进行比较,并按照比较结果选择选择下一比较分支,直到叶子节点作为最终的决策结果。因此首先可获取所有用户的基本信息数据和基线指标数据作为用来分类的特征;分类目标是判断用户属于聚类聚出的哪个用药依从性模式,然后根据基尼指数最小化准则,搜索最佳分割指标和分割值,分裂训练集为两个子集。这个过程不断的在产生的子集里重复递归进行,即递归分割。当达到停止条件或一个训练子集的类标都相同时递归停止,最终获得用药依从性模式分类模型。
步骤s104,获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
具体的,当获得用药依从性模式分类模型之后,可对任意一个用户进行用药依从性模式的分类,首先获取当前待分类用户的基本信息数据及基线数据,然后将所述当前待分类用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获取当前待分类用户的用药依从性模式,即当前待分类用户是属于用药依从性模式集合中哪一个用药依从性模式,所述用药依从性模式为一条时间序列轨迹,最后根据识别出的用药依从性模式进行当前待分类用户的用药提醒,所述用药提醒包括在哪个时间段对用户进行用药提醒。
本实施例中,通过对大量患者的用药依从性数据进行时序聚类,获得用药依从性模式,并结合患者的基本信息获得用药依从性分类模型,根据所述用药依从性分类模型对患者进 行用药提醒,可以对患者的关键次用药进行着重提醒,在降低用药提醒成本的同时提高患者重视,以提升病情控制的有效性,降低发生并发症与不良结局的风险。
图2为本申请第二个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s101,获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列,包括:
步骤s201,预设第一时间段及第二时间段,所述第一时间段用于获取历史数据中任意连续时间段内的用药依从性数据,所述第二时间段用于获取历史数据中任意连续时间段内的临床数据;
具体的,可预先设定两个时间段,第一时间段和第二时间段,所述第一时间段用于获取历史数据中任意连续时间段内的用药依从性数据,所述第二时间段用于获取历史数据中任意连续时间段内的临床数据。
步骤s202,获取样本用户的基本信息数据、基线指标数据,所述第一时间段内的用药依从性数据及所述第二时间段内的临床数据;
具体的,所述基本信息数据和基线指标数据可在用户看诊的时候获取,因此无需设定时间段,对于所述用药依从性数据,当设定第一时间段以后,可获取所述第一时间段内的样本用户的用药依从性数据,同样的,当设定第二时间段之后,可获取所述第二时间段内的样本用户的临床数据。
步骤s203,对所述第一时间段内的样本用户的用药依从性数据进行线性插值,获得样本用户的用药依从性时间序列。
具体的,当获取到样本用户的用药依从性数据后,可对样本用户的(TimeInterval,IsObey)用药依从性数据,以字段TimeInterval为x轴,以字段IsObey为y轴,以“天”为分辨率进行线性插值。则样本用户都可以得到一个以天为单位时间间隔的用药依从性时间序列。例如:一个患者有用药依从性数据:(5,1),(30,1),(60,0),(100,1)。则线性插值后的用药依从性时间序列为:(5,1),(6,1),(7,1)…(30,1),(31,0.967),(32,0.933),…(60,0),(61,0.025),(62,0.05),(63,0.075)…(99,0.975),(100,1)。
本实施例中,通过对用户的用药依从性数据进行线性插值,获取用药依从性时间序列,有利于对用药依从性数据进行聚类。
图3为本申请第三个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s102,对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括:
步骤s301,将样本用户的用药依从性时间序列作为样本;
具体的,每个用户的数据可看做一个样本,当对用户的依从性数据进行线性插值后,可将每个用户的用药依从性时间序列数据看做一个样本。
步骤s302,对所述样本根据公式
Figure PCTCN2020111324-appb-000001
进行聚类,其中
Figure PCTCN2020111324-appb-000002
并在聚类时通过动态时间规划距离计算最小化平方误差E,在所述动态时间规划距离中加入惩罚项,获得用药依从性模式集合。
具体的,当获取到所有用户的用药依从性时间序列样本后,根据公式
Figure PCTCN2020111324-appb-000003
进行k-means聚类,其中,Ci是聚类中的簇,
Figure PCTCN2020111324-appb-000004
是簇Ci的均值向量,也可以 成为质心。
具体的,在对用药依从性时间序列进行k-means聚类时,由于所述用药依从性时间序列在距离度量中,两个形状相似的时间序列不一定在x轴上是完全对齐的,因此可用DTW(动态时间规划)距离代替欧氏距离,计算最小化平方误差E,公式为
Figure PCTCN2020111324-appb-000005
这样经过时间窗折叠或者压缩后的时间序列有可能是匹配的。
具体的,考虑到时间位移的影响,DTW中,在时间相差t处匹配时加入惩罚项a*ln(t+1)使距离增加,其中0<a<1,是控制惩罚程度的参数,在传统DTW算法中,状态转移方程为:
DTW[i,j]:=cost+minimum(DTW[i-1,j]//insertion
DTW[i,j-1]//deletion
DTW[i-1,j-1])//match
其中insertion、deletion对应于时间位移,此时只有一个序列上的点在向后推移,相当于是在延迟。而在match时,两个序列上的点同时向后推移,在位移时,由于s[i]=t[j],故cost=0。虽然算法能容忍位移,但不希望位移毫无代价,因为“match”会比位移带来更高的相似度。所以可将状态转移方程修改为:
DTW[i,j]:=cost+minimum(DTW[i-1,j]+a*ln|i-j|+1)//insertion
DTW[i,j-1]+a*ln|i-j|+1//deletion
DTW[i-1,j-1])//match
其中,a*ln(|i-j|+1)可称为“单步位移惩罚”,使用ln函数是因为,当|i-j|增加时,即位移越远时,希望单步位移惩罚增加得越多。但不希望位移带来的惩罚过大,甚至比cost单位值还大很多,所以用ln函数放慢单步位移惩罚增加的速度。其中,0<a<1,是输入参数,是用来调节位移带来的惩罚大小的系数。
本实施例中,通过动态时间规划距离对用药依从性时间序列进行聚类,可以更精确的对用户样本进行聚类。
图4为本申请第四个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s102,对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括:
步骤s401,当对所述用药依从性时间序列进行聚类时,获取多个用药依从性模式候选集合;
具体的,在进行k-means聚类时,可以尝试不同的k值进行聚类,获取多个用药依从性模式候选集合,所述k表示用药依从性的模式的数量。由于每个患者只有一条用药依从性时间序列,该患者只属于其中一个模式。
步骤s402,在所述多个用药依从性模式候选集合中,根据DB-Index准则获取最终用药依从性模式集合。
具体的,所述“DB-Index准则”指的是Davies–Bouldin index,是聚类算法的一个度量标准,它同时考虑了类间的距离和类内的分散度,在获取到不同的k的用药依从性模式候选集合后,可以通过使得DB-Index最小的k值以获得最优聚类结果,最终获得用药依从性模式类别数量k;例如,假设当前考虑的疾病是糖尿病肾病,如果用糖尿病肾病患者的用药依从性时间序列进行聚类,根据DB-Index准则确定k=4,聚类后画出4个簇的簇中心。
本实施例中,通过DB-Index准则确定用药依从性模式数量,可以更精确的对用户用药依从性模式进行分簇。
图5为本申请第五个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s102,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,包括:
步骤s501,对样本用户的临床数据进行统计,获得样本用户的不良结果发生率;
具体的,可对样本用户的不良临床结果进行查询,所述不良临床结果包括各种疾病或者临床指标,通过对所述临床数据的统计,可获得用户的不良临床结果发生率,例如,ESRD发生率,HbA1c不达标率。
步骤s502,根据所述不良结果发生率对所述用药依从性模式集合进行逻辑回归,获得所述用药依从性模式集合中各用药依从性模式的疗效类别。
具体的,首先可将不良临床结果发生率最低的模式为定为参考模式,然后将每个用户的数据作为一个样本,并把“患者属于哪个模式”作为分析特征并变为哑变量,并加入混杂因子(年龄、性别等),以患者是否发生不良临床结果为y,做逻辑回归,所述混杂因子数据可从基本信息数据中获得,最后根据每个模式中的p-value值判断其不良临床结果发生率是否相对于参考模式显著更高;并由此确定各用药依从性模式的优劣分级;例如:假设聚类结果是糖尿病肾病,且模式4中的用户ESRD发生率最低(20.23%),则把模式4作为参考模式。对个体患者来说,将“患者所属哪个模式”这个特征进行哑变量转换和参考列删除,可转换为3个特征。将这3个特征和年龄、性别(混杂因子)作为训练所用的特征,将个体患者的临床结局(是否发生ESRD)作为y,进行逻辑回归分析。根据这3个分析特征(模式1、模式2、模式3)的p-value可界定这3个模式相对于参考模式ESRD的发生率是否显著更高,如表1所示。若p-value<0.05,则该模式更显著发生ESRD,为治疗有效性差的模式;若p-value>=0.05,则该模式为治疗有效性好的模式;参考模式(模式4)是治疗有效性最好的模式。
表1
Figure PCTCN2020111324-appb-000006
本实施例中,通过对用户的用药依从性模式集合进行逻辑回归,可以有效获得各用药依从性模式的优劣分级。
图6为本申请第六个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s104,根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括:
步骤s601,根据所述疗效类别对所述目标用户的用药依从性模式进行识别,所述疗效类别包括治疗有效性差及治疗有效性好;
具体的,当确定目标用户的用药依从性模式之后,可对所述目标用户依从性模式进行识别,即识别出目标用户的用药依从性模式是治疗有效性差的还是治疗有效性好的。
步骤s602,当所述目标用户的疗效类别为治疗有效性差时,对目标用户根据公式
Figure PCTCN2020111324-appb-000007
进行用药提醒,其中,n为时间频率,IsObey(t)为时刻t的依从度。
具体的,当对目标用户的用药依从性模式识别之后,如果目标用户的用药依从性模式是治疗有效性差的,那么可以在更多的时间点提醒患者用药,即在应当进行用药提醒的时间段,可在时间点t处以每n周一次的频率提醒患者用药,n的计算公式为
Figure PCTCN2020111324-appb-000008
其中,IsObey(t)为用药依从性时间序列轨迹在时间点t的用药依从度,由于治疗有效性差的模式往往有更低的用药依从度,则对该模式的患者进行用药提醒的频率往往会更高。治疗有效性差的模式需进行用药提醒的时间段往往更多,对属于该模式患者进行提醒的次数也就更多。
本实施例中,通过对治疗有效性差的用药依从性模式进行识别并进行用药提醒,可以有效提高治疗性差的用户的用药效果。
图7为本申请第七个实施例的一种基于时序聚类的用药提醒方法流程示意图,如图所示,所述步骤s104,根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括:
步骤s701,预设第一阈值及第二阈值,所述第一阈值用于比对所述用药依从性模式中的斜率,所述第二阈值用于比对依从度;
具体的,可预先设置两个阈值,第一阈值及第二阈值,所述第一阈值用于判断所述用药依从性模式中的轨迹的斜率大小,所述第二阈值用于判断用药依从度的大小。
步骤s702,获取所述目标用户的用药依从性模式中小于所述第一阈值对应的第一用药时间段及小于所述第二阈值对应的第二用药时间段;
具体的,首先可获取所述目标用户的用药依从性模式中每段斜率的大小,然后将所述每段斜率的大小与所述第一阈值进行比较,并将小于所述第一阈值的斜率记录下来,并找出所述小于所述第一阈值的斜率对应的时间段,所述时间段为第一用药时间段;接着可在所述目标用户的用药依从性模式中找出小于所述第二阈值的用药依从度,并找出所述用药依从度对应的时间段,所述时间段为第二用药时间段。
步骤s703,在所述第一用药时间段及第二用药时间段上对目标用户进行用药提醒。
具体的,当获取到用药时间段之后,可在所述用药时间段上对用户进行用药提醒,例如,找出斜率<0的时间段,在这个时间段患者整体倾向于不依从医嘱用药而使中心线下降;找出依从度<0.5的时间段,这个时间段中,尽管患者群越来越依从医嘱用药,但整体的依从度仍然处于低水平,个体患者不依从医嘱用药的可能性还是较大。
本实施例中,通过对用药依从性模式的轨迹的识别,获取斜率及依从度对应的用药时间段,并在所述用药时间段进行用药提醒,可以有效提高患者的用药效率。
本申请实施例的一种基于时序聚类的用药提醒装置结构如图8所示,包括:
获取模块801、聚类模块802、训练模块803及提醒模块804;其中,获取模块801与聚类模块802相连,聚类模块802与训练模块803相连,训练模块803与提醒模块804相连;获取模块801设置为获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;聚类模块802设置为对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行 用药提醒;训练模块803设置为根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;提醒模块804设置为获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
本申请实施例还公开了一种计算机设备,所述计算机设备包括存储器和处理器,所述存储器中存储有计算机可读指令,所述计算机可读指令被一个或多个所述处理器执行时,使得一个或多个所述处理器执行上述各实施例中所述用药提醒方法中的步骤。
本申请实施例还公开了一种存储介质,所述存储介质可被处理器读写,所述存储器存储有计算机可读指令,所述计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行上述各实施例中所述用药提醒方法中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,该计算机程序可存储于一计算机可读取存储介质中,该程序在执行时,可包括如上述各方法的实施例的流程。其中,前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)等非易失性存储介质、或其他易失性固态存储器件。
以上所述实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的各个技术特征所有可能的组合都进行描述,然而,只要这些技术特征的组合不存在矛盾,都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对本申请专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本申请构思的前提下,还可以做出若干变形和改进,这些都属于本申请的保护范围。因此,本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种基于时序聚类的用药提醒方法,其中,包括以下步骤:
    获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
    对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
    根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
    获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
  2. 如权利要求1所述的基于时序聚类的用药提醒方法,其中,所述获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列,包括:
    预设第一时间段及第二时间段,所述第一时间段用于获取历史数据中任意连续时间段内的用药依从性数据,所述第二时间段用于获取历史数据中任意连续时间段内的临床数据;
    获取样本用户的基本信息数据、基线指标数据,所述第一时间段内的用药依从性数据及所述第二时间段内的临床数据;
    对所述第一时间段内的样本用户的用药依从性数据进行线性插值,获得样本用户的用药依从性时间序列;
    则所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,包括:
    根据所述基本信息数据及所述第二时间段内的临床数据对所述用药依从性模式集合进行逻辑回归分析。
  3. 如权利要求2所述的基于时序聚类的用药提醒方法,其中,所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括:
    将样本用户的用药依从性时间序列作为样本;
    对所述样本根据公式
    Figure PCTCN2020111324-appb-100001
    进行聚类,其中
    Figure PCTCN2020111324-appb-100002
    并在聚类时通过动态时间规划距离计算最小化平方误差E,在所述动态时间规划距离中加入惩罚项,获得用药依从性模式集合。
  4. 如权利要求2所述的基于时序聚类的用药提醒方法,其中,所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括:
    当对所述用药依从性时间序列进行聚类时,获取多个用药依从性模式候选集合;
    在所述多个用药依从性模式候选集合中,根据DB-Index准则获取最终用药依从性模式集合。
  5. 如权利要求3所述的基于时序聚类的用药提醒方法,其中,所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,包括:
    对样本用户的临床数据进行统计,获得样本用户的不良结果发生率;
    根据所述不良结果发生率对所述用药依从性模式集合进行逻辑回归,获得所述用药依从性模式集合中各用药依从性模式的疗效类别。
  6. 如权利要求5所述的基于时序聚类的用药提醒方法,其中,所述根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括:
    根据所述疗效类别对所述目标用户的用药依从性模式进行识别,所述疗效类别包括治疗有效性差及治疗有效性好;
    当所述目标用户的疗效类别为治疗有效性差时,对当前用户根据公式
    Figure PCTCN2020111324-appb-100003
    进行用药提醒,其中,n为时间频率,IsObey(t)为时刻t的依从度。
  7. 如权利要求6所述的基于时序聚类的用药提醒方法,其中,所述根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括:
    预设第一阈值及第二阈值,所述第一阈值用于比对所述用药依从性模式中的斜率,所述第二阈值用于比对依从度;
    获取所述目标用户的用药依从性模式中小于所述第一阈值对应的第一用药时间段及小于所述第二阈值对应的第二用药时间段;
    在所述第一用药时间段及第二用药时间段上对目标用户进行用药提醒。
  8. 一种基于时序聚类的用药提醒装置,其中,所述装置包括:
    获取模块:设置为获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
    聚类模块:设置为对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
    训练模块:设置为根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型:
    提醒模块:设置为获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
  9. 一种计算机设备,其中,所述计算机设备包括存储器和处理器,所述存储器中存储有计算机可读指令,所述计算机可读指令被一个或多个所述处理器执行时,使得一个或多个所述处理器执行如下所述的基于时序聚类的用药提醒方法的步骤:
    获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
    对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
    根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
    获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述 目标用户的用药依从性模式对目标用户进行用药提醒。
  10. 如权利要求9所述的基于时序聚类的用药提醒设备,其中,所述获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列,包括以下步骤:
    预设第一时间段及第二时间段,所述第一时间段用于获取历史数据中任意连续时间段内的用药依从性数据,所述第二时间段用于获取历史数据中任意连续时间段内的临床数据;
    获取样本用户的基本信息数据、基线指标数据,所述第一时间段内的用药依从性数据及所述第二时间段内的临床数据;
    对所述第一时间段内的样本用户的用药依从性数据进行线性插值,获得样本用户的用药依从性时间序列;
    则所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,包括:
    根据所述基本信息数据及所述第二时间段内的临床数据对所述用药依从性模式集合进行逻辑回归分析。
  11. 如权利要求10所述的基于时序聚类的用药提醒设备,其中,所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括以下步骤:
    将样本用户的用药依从性时间序列作为样本;
    对所述样本根据公式
    Figure PCTCN2020111324-appb-100004
    进行聚类,其中
    Figure PCTCN2020111324-appb-100005
    并在聚类时通过动态时间规划距离计算最小化平方误差E,在所述动态时间规划距离中加入惩罚项,获得用药依从性模式集合。
  12. 如权利要求10所述的基于时序聚类的用药提醒设备,其中,所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,包括以下步骤:
    当对所述用药依从性时间序列进行聚类时,获取多个用药依从性模式候选集合;
    在所述多个用药依从性模式候选集合中,根据DB-Index准则获取最终用药依从性模式集合。
  13. 如权利要求11所述的基于时序聚类的用药提醒设备,其中,所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,包括以下步骤:
    对样本用户的临床数据进行统计,获得样本用户的不良结果发生率;
    根据所述不良结果发生率对所述用药依从性模式集合进行逻辑回归,获得所述用药依从性模式集合中各用药依从性模式的疗效类别。
  14. 如权利要求13所述的基于时序聚类的用药提醒设备,其中,所述根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括以下步骤:
    根据所述疗效类别对所述目标用户的用药依从性模式进行识别,所述疗效类别包括治疗有效性差及治疗有效性好;
    当所述目标用户的疗效类别为治疗有效性差时,对当前用户根据公式
    Figure PCTCN2020111324-appb-100006
    进行用药提醒,其中,n为时间频率,IsObey(t)为时刻t的依从度。
  15. 如权利要求14所述的基于时序聚类的用药提醒设备,其中,所述根据所述目标用户的用药依从性模式对目标用户进行用药提醒,包括以下步骤:
    预设第一阈值及第二阈值,所述第一阈值用于比对所述用药依从性模式中的斜率,所述第二阈值用于比对依从度;
    获取所述目标用户的用药依从性模式中小于所述第一阈值对应的第一用药时间段及小于所述第二阈值对应的第二用药时间段;
    在所述第一用药时间段及第二用药时间段上对目标用户进行用药提醒。
  16. 一种存储介质,其中,所述存储介质可被处理器读写,所述存储介质存储有计算机指令,所述计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行如下所述的基于时序聚类的用药提醒方法的步骤:
    获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列;
    对所述用药依从性时间序列进行聚类,获得用药依从性模式集合,根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别,其中,所述用药依从性模式集合中包括至少两个用药依从性模式,所述用药依从性模式包含多个用药时间点,用于在所述用药时间点上对用户进行用药提醒;
    根据所述基本信息数据及基线指标数据对所述用药依从性模式集合进行训练,获得用药依从性模式分类模型;
    获取目标用户的基本信息数据及基线数据,并将所述目标用户的基本信息数据及基线数据输入至所述用药依从性模式分类模型中,获得目标用户的用药依从性模式,根据所述目标用户的用药依从性模式对目标用户进行用药提醒。
  17. 如权利要求16所述的存储介质,其中,所述基于时序聚类的用药提醒的指令被处理器执行所述获取样本用户的用药依从性数据、临床数据、基本信息数据及基线指标数据,根据所述用药依从性数据获得用药依从性时间序列的步骤时,包括以下步骤:
    预设第一时间段及第二时间段,所述第一时间段用于获取历史数据中任意连续时间段内的用药依从性数据,所述第二时间段用于获取历史数据中任意连续时间段内的临床数据;
    获取样本用户的基本信息数据、基线指标数据,所述第一时间段内的用药依从性数据及所述第二时间段内的临床数据;
    对所述第一时间段内的样本用户的用药依从性数据进行线性插值,获得样本用户的用药依从性时间序列;
    则所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,包括:
    根据所述基本信息数据及所述第二时间段内的临床数据对所述用药依从性模式集合进行逻辑回归分析。
  18. 如权利要求17所述的存储介质,其中,所述基于时序聚类的用药提醒的指令被处理器执行所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合的步骤时,包括以下步骤:
    将样本用户的用药依从性时间序列作为样本;
    对所述样本根据公式
    Figure PCTCN2020111324-appb-100007
    进行聚类,其中
    Figure PCTCN2020111324-appb-100008
    并在聚类时通过动态时间规划距离计算最小化平方误差E,在所述动态时间规划距离中加入惩罚项,获得用药依从性模式集合。
  19. 如权利要求17所述的存储介质,所述基于时序聚类的用药提醒的指令被处理器执行所述对所述用药依从性时间序列进行聚类,获得用药依从性模式集合的步骤时,包括以下步骤:
    当对所述用药依从性时间序列进行聚类时,获取多个用药依从性模式候选集合;
    在所述多个用药依从性模式候选集合中,根据DB-Index准则获取最终用药依从性模式集合。
  20. 如权利要求18所述的存储介质,其中,所述基于时序聚类的用药提醒的指令被处理器执行所述根据所述基本信息数据及临床数据对所述用药依从性模式集合进行逻辑回归分析,获得所述用药依从性模式集合中各用药依从性模式的疗效类别的步骤时,包括以下步骤:
    对样本用户的临床数据进行统计,获得样本用户的不良结果发生率;
    根据所述不良结果发生率对所述用药依从性模式集合进行逻辑回归,获得所述用药依从性模式集合中各用药依从性模式的疗效类别。
PCT/CN2020/111324 2019-10-18 2020-08-26 基于时序聚类的用药提醒方法及相关设备 WO2021073255A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910991530.X 2019-10-18
CN201910991530.XA CN110812241A (zh) 2019-10-18 2019-10-18 基于时序聚类的用药提醒方法及相关设备

Publications (1)

Publication Number Publication Date
WO2021073255A1 true WO2021073255A1 (zh) 2021-04-22

Family

ID=69549560

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/111324 WO2021073255A1 (zh) 2019-10-18 2020-08-26 基于时序聚类的用药提醒方法及相关设备

Country Status (2)

Country Link
CN (1) CN110812241A (zh)
WO (1) WO2021073255A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110812241A (zh) * 2019-10-18 2020-02-21 平安科技(深圳)有限公司 基于时序聚类的用药提醒方法及相关设备
CN111696676A (zh) * 2020-05-28 2020-09-22 思派健康产业投资有限公司 一种基于依从性的慢病患者筛选方法
CN112037932A (zh) * 2020-09-09 2020-12-04 平安科技(深圳)有限公司 患者用药行为干预方法及装置、服务器、存储介质
CN117476165B (zh) * 2023-12-26 2024-03-12 贵州维康子帆药业股份有限公司 一种中成药药物药材智能管理方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103718208A (zh) * 2011-06-10 2014-04-09 Ai医疗科技公司 用于监控药物依从性的方法以及装置
US20140188507A1 (en) * 2012-12-28 2014-07-03 Industrial Technology Research Institute Lifestyle progression models for use in preventative care
US20170124281A1 (en) * 2015-10-29 2017-05-04 Cerner Innovation, Inc. Systems and Methods for Analyzing Medication Adherence Patterns
CN107391941A (zh) * 2017-07-26 2017-11-24 上海科瓴医疗科技有限公司 一种提高用药依从性的方法和系统
CN108109700A (zh) * 2017-12-19 2018-06-01 中国科学院深圳先进技术研究院 一种慢性病药物疗效评价方法和装置
CN110812241A (zh) * 2019-10-18 2020-02-21 平安科技(深圳)有限公司 基于时序聚类的用药提醒方法及相关设备

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140172437A1 (en) * 2012-12-14 2014-06-19 International Business Machines Corporation Visualization for health education to facilitate planning for intervention, adaptation and adherence
CN104915560A (zh) * 2015-06-11 2015-09-16 万达信息股份有限公司 一种基于广义神经网络聚类的疾病病种诊疗方案预测方法
CN109830302B (zh) * 2019-01-28 2021-04-06 北京交通大学 用药模式挖掘方法、装置和电子设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103718208A (zh) * 2011-06-10 2014-04-09 Ai医疗科技公司 用于监控药物依从性的方法以及装置
US20140188507A1 (en) * 2012-12-28 2014-07-03 Industrial Technology Research Institute Lifestyle progression models for use in preventative care
US20170124281A1 (en) * 2015-10-29 2017-05-04 Cerner Innovation, Inc. Systems and Methods for Analyzing Medication Adherence Patterns
CN107391941A (zh) * 2017-07-26 2017-11-24 上海科瓴医疗科技有限公司 一种提高用药依从性的方法和系统
CN108109700A (zh) * 2017-12-19 2018-06-01 中国科学院深圳先进技术研究院 一种慢性病药物疗效评价方法和装置
CN110812241A (zh) * 2019-10-18 2020-02-21 平安科技(深圳)有限公司 基于时序聚类的用药提醒方法及相关设备

Also Published As

Publication number Publication date
CN110812241A (zh) 2020-02-21

Similar Documents

Publication Publication Date Title
WO2021073255A1 (zh) 基于时序聚类的用药提醒方法及相关设备
WO2020181805A1 (zh) 糖尿病的预测方法及装置、存储介质、计算机设备
CN110246577B (zh) 一种基于人工智能辅助妊娠期糖尿病遗传风险预测的方法
CN107103207B (zh) 基于病例多组学变异特征的精准医学知识搜索系统及实现方法
AU2021103976A4 (en) Asthma diagnosis system based on decision tree and improved SMOTE algorithm
CN101911078A (zh) 基于疾病概率向量检索类似患者病例
US20220323018A1 (en) Automatic prediction of blood infections
Afsaneh et al. Recent applications of machine learning and deep learning models in the prediction, diagnosis, and management of diabetes: a comprehensive review
CN109243567B (zh) 一种基于处方数据挖掘的药物推荐方法
CN104866713B (zh) 基于增量局部鉴别子空间嵌入的川崎病和发烧诊断系统
CN116910172B (zh) 基于人工智能的随访量表生成方法及系统
Chen et al. Heterogeneous postsurgical data analytics for predictive modeling of mortality risks in intensive care units
CN114023441A (zh) 基于可解释机器学习模型的严重aki早期风险评估模型、装置及其开发方法
Ho et al. Imputation-enhanced prediction of septic shock in ICU patients
Tamin et al. Implementation of C4. 5 algorithm to determine hospital readmission rate of diabetes patient
CN116246768B (zh) 一种基于人工智能的mri影像检查智能分析管理系统
Al-Mualemi et al. A deep learning-based sepsis estimation scheme
CN114388095A (zh) 脓毒症治疗策略优化方法、系统、计算机设备和存储介质
CN117174330A (zh) 一种基于机器学习的IgA肾病患者治疗方案推荐方法
CN113571180A (zh) 基于c肽分层及脏器功能的2型糖尿病人工智能诊疗管理系统
CN115050451A (zh) 败血症临床用药方案自动生成系统
Han et al. Risk prediction of diabetes and pre-diabetes based on physical examination data
Lin et al. Medical Concept Embedding with Variable Temporal Scopes for Patient Similarity.
CN115691788A (zh) 一种基于异构数据的双重注意力耦合网络糖尿病分类系统
WO2022271572A1 (en) System and method for determining a stool condition

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20877560

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20877560

Country of ref document: EP

Kind code of ref document: A1