CN109360657B - Time period reasoning method for selecting samples of hospital infection data - Google Patents

Abstract

The invention discloses a time period reasoning method for sample selection of nosocomial infection data. In the process of data sampling, the recorded diagnosis date is used as the benchmark, and the real infection date or samples of the last few days of the real infection date are selected, wherein Samples are drawn within the time unit length inferred from the “before time period” and the “post time period” in the past, and the average value is obtained for evaluation. Its beneficial effects are: adopting the method of time period reasoning to solve the problem of sample selection, effectively obtaining the infected samples during the period of time in the infection state, and the method is generalizable, and can be tried for time series data. The basic conditions for the use of the device are similar to basic experience.

Description

A time-segment reasoning method for sample selection of nosocomial infection data

technical field

The invention relates to a nosocomial infection data mining technology, in particular to a time period reasoning method applied to the selection of nosocomial infection data samples in the nosocomial infection big data analysis and modeling process.

Background technique

In the field of nosocomial infection, a large number of economic losses and casualties are caused by nosocomial infection every year. The analysis and modeling of nosocomial infection data is a difficult problem in medical data analysis and modeling. The quality of nosocomial infection data is poor, the sample construction is difficult, and the data There is no good precedent for the analysis and modeling of hospital infection. However, with the gradual attention of hospital infection incidents, a monitoring and early warning model has been constructed to carry out real-time monitoring and early warning of hospital infection cases, thereby helping clinicians to conduct timely monitoring and warning. Intervention and treatment have become a very valuable issue. In recent years, major hospitals have begun to establish their own hospital infection monitoring information systems. However, these monitoring and early warning systems are of mixed quality and their effects are not satisfactory. Most of these problems are caused by the difficulty of big data analysis and modeling of hospital infection. Successful cases are used as guidance and reference, and each case solves a small part of the problem. It is difficult to comprehensively explain and analyze the difficulties of hospital infection modeling. Solutions to data modeling have also been proposed in the existing literature, but there are various problems.

For example, the literature (Lin Yusong, Wang Peipei, Liu Wei, et al. Research on the preprocessing method of medical examination data [J]. Computer Application Research, 2017, 34(4): 1089-1092.) proposed a data cleaning method, through linear Function data transformation and other methods eliminate data duplication and other problems. However, in hospital infection data, the problem of data missing is more common, and the data missing of different attributes is different. For example, body temperature basically does not appear a lot of missing, while Lab tests may be missing for a week in a row, and a “one size fits all” approach to all data is unreasonable.

For another example, the literature (Kotsiantis S B, Kanellopoulos D, Pintelas P E. Datapreprocessing for supervised leaning[J]. International Journal of Computer Science, 2006, 1(2): 111-117.) described missing data and wrong data in the field of machine learning. Some common processing methods have been introduced. For missing values, methods such as the use of mean and special values can be used. However, from the purpose of modeling, these methods are not very suitable, because the ultimate purpose of modeling is to The most important thing is to provide early warning or real-time monitoring for hospital-infected patients. The most important thing is to provide early warning basis for the final warning patient. These basis are generally to show the real value of the patient instead of the processed value, so that it is convenient for doctors to carry out Reasonable diagnosis, so directly modifying the value or using a special value is not suitable for this case.

Another example, literature (Li Hong, Liang Peifeng, Pan Dongfeng, et al. Application of autoregressive moving average mixed model in the prediction of hospital infection incidence [J]. Chinese Journal of Hospital Infection, 2013, 23(11): 2693.) A time series model is proposed to monitor the development trend of nosocomial infection, with the purpose of early warning and reducing the risk of nosocomial infection. However, this early warning model has two obvious shortcomings. One is that the model indirectly monitors the incidence of nosocomial infection, which is generally a retrospective study after the event. Intervention and treatment of infection. Second, the model is a formula-type calculation model, which is not interpretable and difficult to analyze. The data used in the model is based on the Ningxia People's Hospital, without extensive testing in other hospitals. Whether it can be generalized remains to be tested.

In the process of big data analysis and modeling of nosocomial infection, the difficulties encountered mainly include the following:

(1) The problem of missing nosocomial infection data. Nosocomial infection data has the characteristics of timeliness, which determines that the time range of the patient's detection data must be taken into account when using the data, and the problem of missing nosocomial infection data increases the difficulty of big data analysis and modeling of nosocomial infection;

(2) The problem of dividing positive and negative samples of nosocomial infection data. Nosocomial infection data samples are mainly divided into two categories, one is infected samples and the other is non-infected samples. How to divide these two types of samples into positive and negative examples is a more important issue. However, the actual problem is more complicated. Non-infected samples are relatively easy to obtain. It is only necessary to randomly select a few days of data from patients who have not suffered nosocomial infection as non-infected samples. The selection of infected samples is difficult. Most of the patients with nosocomial infection have been hospitalized for a long time, and they may only be in the real infection state for a period of time. Other times are normal. It is more difficult to obtain the data of the infection state in this period. In nosocomial infections, patients who have been diagnosed or reported as nosocomial infections generally have an "infection date" diagnosed by a doctor, which is referred to as the "diagnosis date" here, which is used to determine that the patient was infected on that day. Naturally, the "diagnosis date" is taken as the infection sample. However, the actual investigation found that this date is an inferred date by the doctor, and most of them are not accurate. The actual date of infection of the patient may be before or here. After the date, the date is not very strict, and the literature (Zhang Xiaowei, Meng Lihui, Zheng Jia, et al. Discussion on different statistical methods for the underreporting rate of nosocomial infection [J]. Chinese Journal of Hospital Infectious Diseases, 2006, 1.). Similar issues are discussed.

Therefore, in the process of big data analysis and modeling of nosocomial infection, the existing technology needs to be improved in view of the above-mentioned defects.

SUMMARY OF THE INVENTION

In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is to provide a time period reasoning method for sample selection of nosocomial infection data, so as to solve the problem of nosocomial infection data sample selection in the process of nosocomial infection big data analysis and modeling. question.

Before proceeding to the elaboration of the content of the invention, the terms appearing in the document need to be explained and defined.

Valid time range: Refers to the general term for the time range of the patient/patient detection data. For example, data such as body temperature, stool frequency, heart rate and respiratory rate have high timeliness, and there will be differences basically every day, then the data use 24 hours can be used as the range, and data beyond 24 hours are no longer considered for use. For example, microbiological examinations and laboratory examinations have low timeliness, and data within three to five days can be considered valid, then 72 or 120 hours is the range, and this range is collectively referred to as the "valid time range". The "effective time range" is generally determined based on experience or data in the reference literature, and can also be established according to the actual modeling purpose. The standard refers to the action time of some features in "Nosocomial Infection Diagnostic Standards (Trial)".

Diagnosis date: In nosocomial infection, patients who have been diagnosed or reported as nosocomial infection generally have an "infection date" diagnosed by a doctor, which is referred to as the "diagnosis date" here.

Date of infection: The date when the patient actually became infected is the date of infection.

Previous time period: Select infection samples with the diagnosis date as the base date, and the length of the time unit of forward reasoning is the previous time period.

Post-time period: The infection sample is selected based on the diagnosis date, and the length of the time unit of subsequent inference is the post-time period.

In order to solve the above problems, the present invention provides a time period reasoning method for selecting a nosocomial infection data sample, comprising the following steps:

Step 1, determine the characteristics of the nosocomial infection data, and classify the characteristics according to the "valid time range", the characteristic set is denoted as F, and k represents the kth characteristic in the set F;

Step 2, the set composed of all patients is denoted as S, the patient m is obtained in the set S, and the positive and negative sample set N is generated for the patient m;

Step 3, after the positive and negative sample set N is generated in step 2, the set of nosocomial infection patients is denoted as C, and the set of the diagnosis date of the infected patient is denoted as Cd;

Step 4, randomly extract n patients from the set C, and obtain the diagnosis dates corresponding to the n patients;

Step 5: Diagnose the n patients in Step 4, and obtain the arrays A_pre and A_end composed of the data of the "pre-time period" and "post-time period" of the n patients;

Step 6, the two arrays in step 5 are summed and then averaged to obtain two averages avg_pre=sum(A_pre)/n, avg_end=sum(A_end)/n; these two averages are used as a set Two parameters for time-segment inference for all patients in C, approximate estimates of the "pre-time" and "post-time" for all patients in set C;

Step 7, update data to generate sample set D and carry out modeling test according to sample set D;

Step 8, continuously fine-tune avg_pre and avg_end according to the test results to obtain the final required value.

It should be noted that the positive sample in step 2 is the sample of patient m with nosocomial infection, and the negative sample is the sample of patient m without nosocomial infection.

Further, if m is a patient in the positive sample, then m is recorded as the mth patient in S; if m is a patient in the negative sample, then m is a randomly selected patient.

Further, the method for updating data in step 7 adopts an incremental updating method, including the following steps:

Step 7a, sort the positive and negative sample set N in step 2 in ascending order according to the sequence of time from front to back, so as to ensure that in the process of incremental update, the time is arranged from front to back, so as to ensure that when updating. always new value overwrites old value;

Step 7b, store the sample i with the earliest time in the sample set N into the sample set D, and store it in the set T according to the characteristics of the hospital infection data determined in step 1, respectively record Tk_v and Tk_date, indicating the set N The value of the k-th feature in the set T corresponding to the sample i in , and the date of the value;

Step 7c: Perform missing value judgment on the second and subsequent samples i in the sample set N, update the missing values, and retain the non-missing values; if the value Tk_v of the feature Tk of the sample i is a missing value, then Find the value Tk_v and Tk_date of the feature Tk in the sample set D in reverse order. If the value in the sample set D is not empty, and the difference between Tk_date and Tk_date in i does not exceed the "valid time range", then take the value out Update the Tk_v of sample i to replace the missing value. The reverse order traversal is required here to ensure that the traversed samples in the set D are always closest to the current sample in time, and the following is the same; if the value in the sample set D is not If it is empty, but it exceeds the "valid time range", the traversal is introduced to keep the missing state of the k-th feature of sample i; if the value in the sample set D is empty, continue to traverse the next value.

Step 7d, store the updated or retained samples into the sample set D, read subsequent samples in the order of step 5 and save the sample data;

In step 7e, when i=N is obtained by repeating steps 7c and 7d, the reading is completed, and the construction of the sample set D is completed.

The present invention also provides an analysis and modeling method for solving the sampling of nosocomial infection data samples by means of time-segment reasoning, comprising the following steps:

Step A1, determine the characteristics of the nosocomial infection data, and classify the characteristics according to the "valid time range";

Step A2, determine the patients who generate positive and negative samples, wherein the positive samples are patient samples with nosocomial infection, and the negative samples are patient samples without nosocomial infection;

Step A3, adopting the method of time period reasoning to divide the positive and negative samples, and the specific implementation method is as described in the foregoing steps 1-8;

Step A4, the method of "incremental update" is used to generate the sample set, and the specific implementation method is as described in the aforementioned steps 7a-7e;

Step A5, analyze and model the final sample set.

The present invention also provides an analysis and modeling system for solving the sampling of nosocomial infection data samples by means of time-segment reasoning, which at least includes a database in which all patient sets S and case data of patients in the set S are stored; a sample generation a module that generates a sample set according to the sample generation conditions, for example, an infected patient set and a non-infected patient set are generated according to the infection situation of the patient; a sample division module that divides the sample set generated by the aforementioned sample generation module into a sample set required for analysis and modeling; and A data update module, the data update module implements the update of missing data values through the aforementioned steps 1-7.

The present invention also provides an implementation method of an analysis and modeling system for solving the sampling of nosocomial infection data samples by means of time-segment reasoning, comprising the following steps:

Step B1, according to the information in the database, sort out and clarify the required patient data items in the nosocomial infection data and design a corresponding XML storage structure;

Step B2, the sample generation module organizes the patient's data into a sample format of the required data according to the set sampling period as the sample and the data item as the feature, and generates the required sample set;

In the above step B2, the data of nosocomial infection are organized into samples, each of these samples is the data of a patient in the set sampling period, and the features in the samples are increased according to the aforementioned incremental update method. The quantitative update will eventually generate a sample set consisting of samples of several patients in the set sampling period.

In step B3, the sample division module divides the sample set according to the final classification label, and generates a final sample set that is distinguished from the infected sample and the non-infected sample;

Step B4, the divided sample set is incrementally updated by the data update module;

Step B5, after the update of the sample set is completed, build a model according to the general method of modeling.

Further, in step B1, the file is stored in the form of XML, which contains the basic information of the patient, such as case number, gender, age, date of infection, etc., and contains the basic information of the patient's admission, such as admission diagnosis, admission department, The admission date, etc., includes the information of the sampling period set by the patient during the hospitalization, such as body temperature, doctor's order, laboratory examination, microbiological examination, imaging examination, and disease course records; in addition to the storage function of the patient's information, this storage solution The most important thing is to facilitate the organization and application of data. Each item in XML can be taken out separately and used in combination with other items, and each item has an accurate time. It can also be organized in time series, depending on the way of use. to the needs of developers.

The present invention also provides a computer-readable medium for addressing nosocomial infection data sample sampling and nosocomial infection data analysis and modeling over a computer network, comprising a set of instructions that, when executed, result in at least one A computer implementation solves the problem of sample sampling of nosocomial infection data in the process of nosocomial infection data analysis and modeling as well as post-sampling data analysis and modeling.

By implementing the above-mentioned method provided by the present invention to solve the sampling of nosocomial infection data samples, the method has the following technical effects:

(1) The problem of difficulty in dividing samples due to inaccurate infection dates is solved by using time-segment reasoning. In the past, there were many cases in which nosocomial infection samples were divided by patients, and the collection of infection data required a lot of manual review. Distinguishing the two types of samples solves the problems of difficult sample selection and difficult sample division, and effectively obtains infected samples during the period of infection.

(2) The problem of missing data or real-time data utilization is solved by the incremental update method. In the previous methods for dealing with missing data and real-time data of nosocomial infections, most of them are to evaluate the missing values of the data, and delete the samples with more missing data and no longer use them. This is not very reasonable, because although the missing values are more However, if the few values in it are real-time data, it has reference value. This method can basically solve the problem of missing most data by using incremental update.

(3) The method of classifying different features according to the "valid time range" is proposed to solve the problem of different time validity of different features.

(4) The way of XML is used for storage, which solves the problem that the hospital data is complicated and difficult to use. In the previous methods of processing hospital infection data, most of them still processed and analyzed data derived from databases and related programs, and did not design a more general data structure for data storage and processing alone. In addition to the advantages of convenient storage and processing, this method can also manage data on a patient-by-patient basis. All the specific information of each patient is summarized into one file, which is not only conducive to data management, but also convenient for R&D workers. A retrospective study of the data greatly facilitates the application of the data.

(5) The basic process and several difficulties of "nosocomial infection big data analysis and modeling" are clearly described, and the basic ideas for the analysis and modeling of nosocomial infection data are clarified.

Description of drawings

The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings, so as to fully understand the purpose, characteristics and effects of the present invention.

Fig. 1 is a flowchart of nosocomial infection data analysis and modeling in the embodiment of the present invention;

2 is a flowchart of time segment reasoning in an embodiment of the present invention;

Fig. 3 is the incremental update flow chart in the embodiment of the present invention;

Fig. 4 is the flow chart of the realization method of analysis modeling system in the embodiment of the present invention;

FIG. 5 is a partial feature classification table in "Nosocomial Infection Diagnostic Standards (Trial)" in an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The "valid time range" mentioned in the following embodiments refers to the general term for the time range of the time limit for patient/patient detection data. There will be differences every day, so the data use can be used as a range of 24 hours, and the data that exceeds 24 hours will not be considered for use. For example, microbiological examinations and laboratory examinations have low timeliness, and data within three to five days can be used. If it is considered to be valid, then 72 or 120 hours can be used as the range, which is collectively referred to as the "valid time range". The "effective time range" is generally determined based on experience or data in the reference literature, and can also be established according to the actual modeling purpose. The standard refers to the action time of some features in "Nosocomial Infection Diagnostic Standards (Trial)".

"Date of diagnosis" refers to the "date of infection" diagnosed by a doctor in patients who have been diagnosed or reported as a hospital infection, which is referred to as the "date of diagnosis" here.

"Date of infection" refers to the date when the patient actually became infected as the date of infection.

"Previous time period" refers to the selection of infection samples based on the diagnosis date, and the length of the time unit of forward reasoning is the previous time period.

"Post time period" refers to the selection of infection samples based on the date of diagnosis, and the length of the time unit for subsequent inference is the post time period.

As shown in Figure 1, the data analysis and modeling process of hospital infection includes the following steps:

Step A1: Determine the characteristics of the nosocomial infection data, such as body temperature, pulse and C-reactive protein, etc., and form a characteristic set of nosocomial infection, which is denoted as F, and k represents the k-th feature in the set F; Perform classification to generate a set T, where Tk represents the category to which the k-th feature belongs;

Among them, the purpose of the "effective time range" is that the time length of the impact of different features on the human body is different, which is generally determined based on experience or data in the reference literature, or can be established according to your actual modeling purpose. The standard recommends referring to The action time of some features in the "Nosocomial Infection Diagnostic Standards (Trial)" is shown in FIG. 5 , and some classifications are given in this embodiment, which can be used for reference.

Among them, the determination of the feature set mainly relies on some features summarized in "Nosocomial Infection Diagnostic Standards (Trial)" and some features obtained from papers or doctors. This part of the work is mainly completed in the stage of demand investigation and analysis.

Step A2: Determine the patients who generate positive and negative samples. This step is to divide the data by patient unit, wherein the positive samples are the samples of patients with nosocomial infection, and the negative samples are the samples of patients without nosocomial infection; first, it is necessary to Obtaining patients with nosocomial infection is easier to obtain, because patients with nosocomial infection have been diagnosed by the hospital or have been reported, and you can directly get the "diagnosis date" corresponding to these patients and these patients, and then , non-hospital infection patients can take those patients in the hospital who have not been diagnosed with nosocomial infection. Since there are many patients in this group, a combination of stratified sampling and random sampling is adopted. Stratification is carried out, and then random sampling is used to select some patients in each stratum, and the number of patients finally selected generally does not exceed 10 times the number of patients with nosocomial infections;

It should be noted that this step is used to determine which patients have nosocomial infections and which patients are non-hospital infections. These are not samples for modeling, because one patient is not suitable as a sample, and each patient is in hospital during the period. A certain period of time is in the state of infection, while other times are normal. Only the period of time in the state of infection can be used as an infection sample, that is, the sample has a time-series nature.

In step A3, time-segment reasoning is used to divide positive and negative samples; after nosocomial infection patients and non-hospital infection patients are determined, positive and negative samples can be generated in time series. In this case, samples are mainly generated in units of days, so each day of the patient's stay in the hospital can be used as a sample. However, it is not necessary to generate samples for the data of each day of the patient's stay in the hospital. For patients with non-hospital infections, The data of certain days during the patient's stay in the hospital can be extracted by random sampling. For patients with hospital infection, the method of "time period reasoning" can be applied to extract the data of the corresponding time period. When dividing positive and negative samples, it is necessary to try several times to find reasonable values for “period” and “post time period”, and it is generally recommended that the two time periods should not exceed 5 days; the process of using time inference is shown in Figure 2, including:

Step A3a, the collection of hospital infection patients is denoted as C, and the collection of its diagnosis date is denoted as Cd;

Step A3b, randomly extract n patients from the set C, and obtain the diagnosis dates corresponding to these n patients;

Step A3c, further diagnose the n patients according to the "Nosocomial Infection Diagnostic Criteria (Trial)", and obtain the arrays A_pre and A_end composed of the "pre-time period" and "post-time period" of the n patients, and analyze the n patients. The two arrays of patients are summed and averaged respectively, and the average values of the two groups of values are obtained as avg_pre=sum(A_pre)/n and avg_end=sum(A_end)/n. These two averages can be used as the time of all patients C. Two parameters of segment inference;

Step A3d, adopting an incremental update method to generate a sample set and perform a modeling test;

Step A3e, continuously fine-tune avg_pre and avg_end according to the test results, such as +1 or -1 at the same time, to optimize the set and finally obtain a value with better effect.

Step A4, after the positive and negative samples are divided, use the "incremental update" method to generate a sample set; this step is the same as the previous step A3d, and here it is necessary to follow the "valid time range" to which the data characteristics of step 1 belong. To incrementally update different features, it should be noted that the positive samples obtained by applying time-segment inference for hospital-acquired patients are continuous in time, so this method can solve most of the missing data problems. However, non-hospital infection patients Due to the use of random sampling, it is difficult to ensure that the time is continuous. The "incremental update" here may not solve the problem of missing data. For this situation, it needs to be handled according to the actual situation. If there are too many missing values, it can be considered. When selecting samples from patients with non-hospital infection, it is sufficient to randomly select consecutive days; the method of "incremental update" is used to process the missing values of the samples, as shown in Figure 3. The specific steps include:

In step A4a, the set of all patients in the aforementioned step A3 is denoted as S, and m represents the mth patient in S;

Step A4a, traverse the set S, obtain a hospital infection patient m in S, perform "time period reasoning" on m to generate a set N of positive and negative samples, and sort N in ascending order of the date of the day. During the incremental update, the time is arranged from small to large, so as to ensure that the new value always covers the old value during the update. If the patient m is a non-infected patient, the random sampling method is used to generate the sample set N;

Step A4b, start to traverse the sample set N, the first sample i is the sample with the smallest time, directly store it in the sample set D, and classify the characteristics of the sample i into the set T, record Tk_v and Tk_date, indicating the first sample i. the value of the k features and the date of that value;

Step A4c, start to traverse the second and all subsequent samples i, and judge the value Tk_v of each feature Tk in i. If the value is a missing value, go to step 5, otherwise the value is retained and does not do any processing ;

Step A4d, if the value Tk_v of the feature Tk of the sample i is a missing value, find the values Tk_v and Tk_date of the feature Tk in the sample set D in reverse order, if the value in D is not empty and the difference between Tk_date and Tk_date in i If the value does not exceed the "valid time range", the value is taken out and updated to Tk_v of sample i to replace the missing value. If the value in D is not empty but exceeds the "valid time range", the traversal is introduced to keep the kth feature of sample i The missing state of , if the value in D is also empty, continue to traverse the next value. Reverse order traversal is required here to ensure that the traversed samples in set D are always closest to the current sample in time;

Step A4e, after the update or reservation is completed, this sample is stored in the sample set D and the next sample is read, i.e. i=i+1;

In step A4f, it is judged whether i=N is established. If so, the traversal is completed, and the construction of the sample set D is completed. If it is not established, continue to the next step.

Step A5: Perform analysis, modeling, testing and optimization on the final sample set; this step is the same as the aforementioned step A3e; after the sample set is generated, the main difficulties of the nosocomial infection data are basically solved. At this time, the follow-up work can basically be completed according to the basic process of data analysis and machine learning. However, it should be noted that the choice of machine learning algorithm is not arbitrary, and the early warning results of the hospital infection monitoring and early warning model generally need to be interpretable. Therefore, the algorithm must choose an explanatory algorithm, such as decision tree, random forest and logistic regression, etc., while deep learning, support vector machine and other algorithms are not recommended; the process of modeling and testing is shown in the figure 3, this part still uses the traditional algorithm and steps, the steps are as follows:

Step A5a, model the sample set D, and suggest selecting algorithms such as decision tree, random forest, and logistic regression, which are interpretable, and record the sensitivity and specificity indicators of the algorithm on the test set;

Step A5b, after recording the sensitivity and specificity indicators, fine-tune avg_pre and avg_end, perform modeling and testing again, and record the two indicators;

Step A5c, perform multiple modeling tests, and find two indicators with the best effect. At this time, avg_pre and avg_end are basically the best values;

After the final model is built, it can be integrated online. This part will vary greatly according to different systems, but the model is basically universal.

The present invention also provides an analysis and modeling system based on an incremental update method to solve the lack of nosocomial infection data, which at least includes a database that stores all patient sets S and case data of patients in the set S; a sample generation module , generating a sample set according to the sample generation conditions, for example, generating an infected patient set and a non-infected patient set according to the infection situation of the patient; a sample dividing module, dividing the sample set generated by the aforementioned sample generating module into a sample set required for analysis and modeling; and a A data update module, the data update module implements the update of missing data values through the aforementioned steps A4a-steps A4f.

An implementation method of an analysis and modeling system based on an incremental update method to solve the missing data of nosocomial infection, as shown in Figure 4, includes the following steps:

Step B1, according to the information in the database, sort out and clarify the required patient data items in the nosocomial infection data and design a corresponding XML storage structure;

Step B2, the sample generation module organizes the patient's data into a sample format of the required data according to the set sampling period as the sample and the data item as the feature, and generates the required sample set;

In the above step B2, the data of nosocomial infection are organized into samples, each of these samples is the data of a patient in the set sampling period, and the features in the samples are increased according to the aforementioned incremental update method. The quantitative update will eventually generate a sample set consisting of samples of several patients in the set sampling period.

In step B3, the sample division module divides the sample set according to the final classification label, and generates a final sample set that is distinguished from the infected sample and the non-infected sample;

Step B4, the divided sample set is incrementally updated by the data update module;

Step B5, after the update of the sample set is completed, build a model according to the general method of modeling.

Further, in step B1, the file is stored in the form of XML, which contains the basic information of the patient, such as case number, gender, age, date of infection, etc., and contains the basic information of the patient's admission, such as admission diagnosis, admission department, The admission date, etc., includes the information of the sampling period set by the patient during the hospitalization, such as body temperature, doctor's order, laboratory examination, microbiological examination, imaging examination, and disease course records; in addition to the storage function of the patient's information, this storage solution has the function of storing the patient information. The most important thing is to facilitate the organization and application of data. Each item in XML can be taken out separately and used in combination with other items, and each item has an accurate time. It can also be organized in time series, depending on the way of use. to the needs of developers.

A computer-readable medium for selecting sample collections and nosocomial infection data analysis and modeling over a computer network, comprising a set of instructions that, when executed, cause at least one computer to execute a solution to the nosocomial infection data Analyze the problem of sample set selection in the modeling process and the data analysis and modeling after sample set selection.

The preferred embodiments of the present invention have been described above in detail. It should be understood that those skilled in the art can make many modifications and changes according to the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art through logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the present invention shall fall within the protection scope determined by the claims.

Claims (10)

1. A time period reasoning method for selecting a sample of hospital infection data is characterized by comprising the following steps:

step 1, determining the characteristics of hospital infection data, classifying the characteristics according to an effective time range, wherein a characteristic set is marked as F, k represents the kth characteristic in the set F, and the effective time range is a general name of the aging time range of patient/patient detection data;

step 2, recording a set composed of all patients as S, obtaining a patient m in the set S, and generating a positive and negative sample set N for the patient m;

step 3, after the positive and negative sample set N is generated in the step 2, recording a hospital infected patient set as C, and recording a set formed by diagnosis dates of infected patients as Cd;

step 4, randomly extracting n patients from the set C, and obtaining diagnosis dates corresponding to the n patients;

step 5, diagnosing the n patients in the step 4, and obtaining arrays A _ pre and A _ end formed by data of 'previous time period' and 'next time period' of the n patients, wherein the 'previous time period' is the time unit length obtained by selecting infection samples by taking the diagnosis date as the reference date and deducing in the previous time, and the 'next time period' is the time unit length obtained by selecting infection samples by taking the diagnosis date as the reference date and deducing in the next time;

step 6, summing the two arrays in the step 5, and then averaging to obtain two average values avg _ pre ═ sum (a _ pre)/n and avg _ end ═ sum (a _ end)/n; these two averages serve as two parameters for time period inference for all patients in set C, approximating the "previous time period" and "subsequent time period" for all patients in set C;

step 7, updating the data to generate a sample set D and carrying out modeling test according to the sample set D;

step 8, continuously fine-adjusting the avg _ pre and the avg _ end according to the test result to obtain a final required value;

wherein, the positive sample in step 2 is the m-sample of the patient with the hospital infection, and the negative sample is the m-sample of the patient without the hospital infection.

2. A time period inference method according to claim 1, wherein if m is a patient in the positive sample, then m is scored as the mth patient in S; if m is the patient in the counterexample sample, then m is the randomly drawn patient.

3. The time period inference method of claim 1, wherein the method of updating data in step 7 is an incremental update method, comprising the steps of:

step 7a, sequencing the positive and negative sample sets N in the step 2 in an ascending order according to a sequence of time from front to back so as to ensure that the time is arranged from front to back in the incremental updating process, thereby ensuring that a new value always covers an old value during updating;

step 7b, storing the sample i with the earliest time in the sample set N into a sample set D, correspondingly storing the sample i into a set T according to the characteristics of the hospital infection data determined in the step 1, and respectively recording Tk _ v and Tk _ date which represent the value of the kth characteristic in the set T corresponding to the sample i in the set N and the date of the value;

step 7c, carrying out missing value judgment on the second and all the subsequent samples i in the sample set N, updating the missing values, and reserving the un-missing values;

step 7D, storing the updated or reserved samples into the sample set D, reading subsequent samples according to the sequence of the step 5 and storing sample data;

and 7e, when the step 7c and the step 7D are repeated to obtain that i is equal to N, the reading is completed, and the construction of the sample set D is completed.

4. A time period inference method according to claim 3, wherein, in step 7c, if the value Tk _ v of the signature Tk of the sample i is a missing value, the values Tk _ v and Tk _ date of the signature Tk are found in reverse order in the sample set D, and if the value in the sample set D is not empty and the difference between Tk _ date and Tk _ date in i does not exceed the "valid time range", the value is updated to Tk _ v of the sample i instead of the missing value.

5. A time period inference method according to claim 3, characterised in that in step 7c, if the value Tk _ v of the feature Tk of the sample i is a missing value, the values Tk _ v and Tk _ date of the feature Tk are found in reverse order in the sample set D, and if the value in the sample set D is not empty but exceeds the "valid time range", the missing state of the kth feature of the sample i is deduced by traversal.

6. A time period inference method according to claim 3, characterised in that in step 7c, if the value Tk _ v of the signature Tk of the sample i is a missing value, the values Tk _ v and Tk _ date of the signature Tk are found in reverse order in the sample set D, and if the value in the sample set D is empty, the next value is continued to be traversed.

7. An analytical modelling approach to address hospital infection data sample selection by the time period inference method of any of claims 1-6, comprising the steps of:

step A1, determining the characteristics of hospital infection data, and classifying the characteristics according to an effective time range;

step A2, determining patients generating positive and negative samples, wherein the positive sample is a patient sample with nosocomial infection, and the negative sample is a patient sample without nosocomial infection;

step A3, dividing positive and negative examples samples by adopting a time period reasoning mode, wherein the specific implementation mode is as described in the step 1 to the step 8;

step A4, generating a sample set by using an incremental update method, wherein the specific implementation manner is as described in steps 7a-7 e;

step a5, analytically modeling the final sample set.

8. An analytical modelling system for addressing hospital infection data sample selection by the time period inference method of any one of claims 1-6, comprising at least a database in which the case data of all patients in set S and in set S are stored; the sample generation module generates a sample set according to the sample generation condition; the sample dividing module is used for dividing the sample set generated by the sample generating module into a sample set required by analysis and modeling; and the data updating module realizes the updating of the missing data value through the steps 1 to 8.

9. An implementation method of an analytical modeling system for resolving hospital infection data sample selection by the time period inference method of claim 8, comprising the steps of:

step B1, according to the information of the database, the patient data items needed in the hospital infection data are sorted and defined and a corresponding XML storage structure is designed;

b2, the sample generating module arranges the patient data into the sample format of the needed data according to the set sampling period and the data item as the characteristic, and generates the needed sample set;

in the step B2, the data of nosocomial infection is arranged into samples, each of which is the data of a patient in a set sampling period, and the incremental updating method according to claim 3 is used to incrementally update the features in the samples, so as to finally generate a sample set consisting of a plurality of samples of patients in the set sampling period;

step B3, the sample dividing module divides the sample set according to the finally classified labels to generate a sample set after the final infection sample and the non-infection sample are distinguished;

step B4, the divided sample set is updated incrementally through a data updating module;

and step B5, after the sample set is updated, establishing a model according to a general modeling method.

10. A computer readable medium for selecting a sample set and hospital infection data analysis and modeling over a computer network, comprising a set of instructions which, when executed, cause at least one computer to perform the steps of solving the problem of sample selection during the hospital infection data analysis and modeling process and analyzing and modeling the data after sample selection according to any one of claims 1-6.

Images