JP6918353B2 - Death Predictor and Death Prediction Program - Google Patents

Description

The present invention relates to a death prediction device and a death prediction program.

In recent years, in the medical field, research and development have been carried out not only to utilize biometric information measured from a certain patient for the treatment and diagnosis of that patient, but also to utilize it for the treatment and diagnosis of other patients. ing.
The following Patent Document 1 and Patent Document 2 are exemplified as this kind of technology.

In Patent Document 1, mainly for cancer patients, the relationship between the blood analysis result and the elapsed time until the occurrence of an event (severity, death, etc.) is accumulated from the biological information on a large number of patients in the past. A predictor that predicts the timing of similar events in a subject patient using the regression analysis is disclosed. The accuracy (scale) of predicting the time of event occurrence in the present invention is preferably 12 months to 10 days, preferably 6 months to 1 day, when predicting the onset of a specific disease in the analysis target or the death of the analysis target. Monthly is preferred.

Patent Document 2 uses parameters such as the standard deviation of the patient's heart rate (heart rate variability), systolic blood pressure, and medical history, and analyzes these by an artificial intelligence-based approach to predict the survival of the patient. The method is described. Similar to the prediction device according to Patent Document 1, this prediction method also accumulates biological information measured from a large number of patients in the past and predicts the viability of the subject based on the data. Here, the viability is whether the patient dies or survives, which can be rephrased as the mortality rate.

Japanese Unexamined Patent Publication No. 2017-021727 Japanese Patent Application Laid-Open No. 2013-524865

Even if the time of death is predicted based on the disclosure contents of Patent Document 1, it is not practical because the predictable accuracy (scale) is insufficient, and its use in the medical field is limited.
Patent Document 2 urges medical personnel to preferentially intervene in treatment for patients with a high predicted mortality rate, and does not predict the time of death.

The present invention has been made in view of the above problems, and provides a death prediction device and a death prediction program for predicting the death time of a subject with high accuracy based on various information about the patient who is the subject. ..

According to the present invention, the death time of the subject is determined by using the acquisition means for acquiring a plurality of types of patient information about the subject at a frequency of less than a minute order and the patient information of the subject acquired by the acquisition means. A death prediction device including a death prediction means for predicting on the order of time or less is provided.

According to the present invention, the death time of the subject is determined by using the acquisition means for acquiring a plurality of types of patient information about the subject at a frequency of minutes or less and the patient information of the subject acquired by the acquisition means. A mortality prediction means for predicting less than an hourly order and a mortality prediction program for causing a computer to execute are provided.

According to the above invention, since the death time is predicted by processing using patient information acquired more frequently (less than minutes order) as compared with the prior art, higher accuracy (less than time order) is realized for the prediction. be able to.

According to the present invention, there is provided a death prediction device and a death prediction program that predict the death time of a subject with high accuracy based on various information about the patient who is the subject.

It is a system block diagram for realizing this invention. It is a figure which shows a specific example of the display screen of a computer terminal. It is a figure which shows the transition of the shock index about 19 patients who died in the intensive care unit. It is a figure which shows the time course of the modified early warning score derived for three patients who performed a treatment intervention in an intensive care unit. It is the figure which plotted the systolic blood pressure and the shock index from the decrease of blood pressure to the heart rate becomes 0 (zero), and performed the linear regression analysis and the non-linear regression analysis. It is a figure which visualized the classification method of the data group which is the object of regression analysis. It is a figure which shows the contrast between the prediction formula of a shock index and the measured value of a shock index. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database. It is a figure showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In all the drawings, the same components are designated by the same reference numerals, and the description thereof will be omitted as appropriate.

<About the death prediction device according to the present invention>
First, the death prediction device according to the present invention will be described with reference to FIGS. 1 and 2.
FIG. 1 is a system configuration diagram for realizing the present invention. FIG. 2 is a diagram showing a specific example of the display screen of the computer terminal 10.

The death prediction device according to the present invention is realized by, for example, a computer terminal 10 in which dedicated application software (death prediction program according to the present invention) is installed.
The computer terminal 10 can realize each function described later by executing the process related to the application software, and includes the hardware resources necessary for the realization. Here, the hardware resources are specifically the CPU and memory built in the computer terminal 10, the input device that accepts the user's operation input, and the screen necessary for the user's operation and realization of each function. An output device or the like that outputs voice or the like can be exemplified.

The computer terminal 10 can realize at least the following functions.
The first function is to acquire a plurality of types of patient information about a subject at a frequency of less than a minute order (hereinafter, referred to as an acquisition means).
The second function is to predict the death time of the subject on the order of time or less by using the patient information of the subject acquired by the acquisition means (hereinafter, referred to as the death prediction means).
Here, the term “minute order or less” means that the time interval for acquiring patient information by the computer terminal 10 does not exceed 1 hour (60 minutes), and more preferably it does not exceed 10 minutes. .. In the present embodiment, the time interval related to the acquisition of patient information by the computer terminal 10 is, in principle, one minute interval. Further, here, the time order or less means that the time interval for acquiring patient information by the computer terminal 10 does not exceed one day (24 hours), and more preferably the range does not exceed half a day (12 hours). It means that.
It should be noted that the computer terminal 10 does not need to acquire all of the plurality of types of patient information used for predicting the death time at a frequency of minutes or less, and a part of the acquired information may be repeatedly used.

The patient information used by the computer terminal 10 for predicting the death time of the subject includes information related to the patient's personal attributes (for example, the patient's name, gender, age, disease name, patient identification number, etc.) and the patient. Biological information (eg, body temperature, heart rate, respiratory rate, oxygen saturation, blood pressure, etc.) may be included. Here, the blood pressure may correspond to any of systolic blood pressure, diastolic blood pressure, and average blood pressure.

The acquisition destination of the patient information by the computer terminal 10 is, for example, the measuring instrument 20 and the database 40.
The measuring instrument 20 is a device that measures the biological information (vital signs) of the patient 30 who is the subject over time. Although the measuring instrument 20 is shown in FIG. 1 as if it were a single device, it may be an aggregate of a plurality of devices.
The database 40 stores a plurality of types of patient information acquired over time for a plurality of patients. The database 40 may be a dedicated database provided for the purpose of introducing the present invention, or may be a general database provided for a system (for example, an electronic medical record system) introduced separately from the present invention. It may be a database.
In FIG. 1, the computer terminal 10 is shown as being directly connected to the measuring instrument 20 and the database 40, but data is acquired from the measuring instrument 20 and the database 40 via a computer network (not shown). You may. Further, the computer terminal 10 may acquire the biological information measured by the measuring instrument 20 from the database 40 after temporarily storing it in the database 40.

At least a part of the patient information acquired by the computer terminal 10 is displayed on the display screen of the computer terminal 10. FIG. 2 illustrates a specific example of the display screen.
As shown in FIG. 2, the display screen of the computer terminal 10 can be roughly divided into four parts. In FIG. 2, these four parts are shown surrounded by broken lines. Note that these broken lines are not actually displayed.

The patient attribute display unit DR1 located at the uppermost stage is an area for displaying information related to the personal attributes of the subject (patient 30). In the present embodiment, the patient identification number, name, age at admission, gender, department transferred, admission classification number or surgery classification number, length of stay, and diagnosed disease name are displayed on the patient attribute display unit DR1. Will be done.
Although a plurality of diagnosed disease names can be displayed, only one type is displayed in FIG. 2.

The index transition display unit DR2 located on the right side of the middle row is an area for displaying the temporal change of the modified early warning score and the shock index. In the display of the index transition display unit DR2 in the present embodiment, the vertical axis is the modified early warning score or the shock index, and the horizontal axis is the time. In addition, the transition of the modified early warning score and the shock index is shown by a solid line, the transition of the value with + 1σ on each is shown by the alternate long and short dash line, and the transition of the value with -1σ on each is indicated by the alternate long and short dash line. Where σ is the standard deviation of the modified early warning score or shock index.
By observing how the modified early warning score and shock index change over time, it is possible to predict whether the patient's condition will worsen or improve. Furthermore, the patient's condition can be predicted with even higher accuracy by analyzing the variation of the modified early warning score and the shock index with reference to the values with ± 1σ added to each.
An upper limit is set on the horizontal axis of the index transition display unit DR2. For example, in FIG. 2, the upper limit is 20 minutes, and the modified early warning score and the change over time of the shock index are displayed within that range. The time zone displayed on the index transition display unit DR2 is displayed on the time zone display unit DR4, which will be described later.

The model display unit DR3 located on the left side of the middle row is an area for displaying a static scoring model. Here, the static scoring model is a patient information or a plurality of index values derived based on the patient information (for example, a modified early warning score and a shock index) are associated with each other on the vertical axis and the horizontal axis, respectively. It is a two-dimensional model to be displayed.
The static scoring model in this embodiment is divided into three regions: a state stable zone NZ, a caution zone WZ, and a terminal zone TZ. The state-stabilizing zone NZ is the region indicating that the condition of the patient 30 is the most stable, and the terminal zone TZ is the region indicating that the condition of the patient 30 is the most dangerous condition.
In the present embodiment, for convenience of explanation, these three areas are shown in an identifiable manner, but the computer terminal 10 does not necessarily have to display each area in an identifiable manner.

The time zone display unit DR4 located at the bottom is an area for displaying which part of the total time the time zone displayed on the index transition display unit DR2 at that time corresponds to. More specifically, it corresponds to the time zone in which the selected area SR (shaded portion in the time zone display unit DR4) of the total time is displayed on the index transition display unit DR2.

By displaying the display screen of the computer terminal 10 as described above, it is possible to "visualize" the general condition (condition) of the patient 30.
Hereinafter, each feature of the present invention will be described in detail.

<About the prediction function of death time>
Among the past inpatients accumulated in the database 40, 19 patients who have confirmed death in the intensive care unit are selected, and a prediction function of the death time is constructed based on the patient information related to these patients. It was investigated.
FIG. 3 shows the transition of the shock index for 19 patients who died in the intensive care unit, with the shock index on the vertical axis and the time (minutes) until death on the horizontal axis. The data group shown in FIG. 3 is a collection of data (12038 points in total) from when the systolic blood pressure falls below 80 mmHg until the heart rate becomes 0 (zero) and mapped.
Looking at FIG. 3, since there is no correlation between the value of the shock index and the time to death, it seems difficult to predict death using only the shock index.

In the medical field, when a doctor judges the timing when a patient is about to die, blood pressure, heart rate, respiratory rate, etc. are used as judgment criteria, and the age and gender of the patient are also taken into consideration. In order to enable such predictions in clinical practice, we derived two prediction functions focusing on the patient's age, gender, blood pressure, heart rate, and respiratory rate.
The two prediction functions are common in that they are models derived by multiple regression analysis, and differ in that the original data used in the analysis are different.

The first predictive function was derived by analyzing the data of systolic blood pressure of 80 mmHg or less among the vital signs of many patients who died in the intensive care unit. Specifically, the first prediction function was derived by the following procedure.
In the patient information regarding the selected patients, the timing at which the systolic blood pressure was 80 mmHg or less for 10 minutes or more and no increase in blood pressure was observed thereafter was defined as the data extraction start time.
The time when the heart rate became 0 was set as the death time, the death time was used as a reference (0 minutes), and the time was allocated in minutes so as to go back to the actual time at each timing before that. Therefore, each patient information to be analyzed can be paraphrased as patient information at the time of "n minutes until death (n is an integer)".
Of the patient information extracted as described above, those whose systolic blood pressure values are clearly disturbed compared to other time points are considered to be blood pressure values affected by blood gas analysis, etc., and are to be analyzed. Excluded from.
Using the remaining patient information, _{we performed a multiple regression analysis with the time to death (T 1} ) as the objective of the continuous variable, and performed age (Age), gender (Male, Female), systolic blood pressure (SBP), and heart rate. The following predictive functions were derived with variables (HR) and respiratory rate (RR). Since systolic blood pressure, diastolic blood pressure, and mean blood pressure are correlated with each other, only systolic blood pressure is used as a variable in this embodiment, but diastolic blood pressure and mean blood pressure can be used instead. Is.

As described above, the first predictive function is that for multiple patients who have died in the past, the systolic blood pressure and heart rate for each unit time during the period from when the systolic blood pressure falls below a predetermined value to the time of death. It can be rephrased as a function obtained by statistically analyzing the number and respiratory rate, taking into account the age and gender of each patient.
Then, the computer terminal 10 uses the systolic blood pressure, heart rate, and respiratory rate acquired at the time when the systolic blood pressure of the patient 30 who is the subject continues to be equal to or less than a predetermined value for a certain period of time or longer, and further, the age and sex of the patient 30. Is substituted into the first prediction function, the death time of the patient 30 can be predicted on the order of time or less.

Next, the second predictive function was derived by analyzing the vital signs of a large number of patients who died in the intensive care unit from the time of death to 120 minutes before. Specifically, the second prediction function was derived by the following procedure.
First, 19 patients similar to the above were selected.
In the patient information about the selected patient, the time when the heart rate becomes 0 is set as the death time, the death time is used as the reference (0 minutes), and the time is divided into minutes so as to go back to the actual time at each timing before that. Allotted in. Then, patient information from the time of death to 120 minutes before was extracted.
Using the extracted patient information, _{multiple regression analysis was performed with the time to death (T 2} ) as the objective of continuous variables, and age (Age), gender (Male, Female), systolic blood pressure (SBP), and heart rate were performed. The following predictive functions were derived with variables (HR) and respiratory rate (RR). The point that systolic blood pressure can be replaced with diastolic blood pressure or mean blood pressure is the same as the first prediction function.

As mentioned above, the second predictive function is for systolic blood pressure, heart rate and respiratory rate per unit time in the period from a predetermined time before the death time to the death time for multiple patients who have died in the past. Can be rephrased as a function obtained by statistically analyzing each patient's age and gender.
Then, the computer terminal 10 determines the systolic blood pressure, heart rate and respiratory rate acquired at the time when the patient 30 who is the subject is estimated to be a predetermined time before the death time, and further the age and sex of the patient 30. By substituting into the second prediction function, the death time of the patient 30 can be predicted on the order of time or less. Here, the predetermined time is preferably less than or equal to the accuracy of predicting the death time by the computer terminal 10, that is, less than or equal to the time order, and more preferably about several hours.
The method of estimating a predetermined time before the death time will be described later.

Death prediction using the two prediction functions described above is common in that it uses systolic blood pressure, heart rate, and respiratory rate at a point in time just before death, and also takes into account the age and gender of the patient. There is. In other words, the computer terminal 10 (death prediction means) sets a certain time point before the subject's death as a reference time by using the acquired patient information of the subject, and predicts the patient information of the subject acquired at the set reference time. By giving it to the function, the time of death of the subject can be predicted. Here, "determining as the reference time" is a process of determining one of the timings at which the computer terminal 10 (acquisition means) acquires the patient information as the reference time, and the determination is automatically performed by computer processing. It may be a mode determined manually by the user, or a mode in which computer processing and manual processing by the user are combined (for example, a plurality of selection candidates are presented to the user by computer processing, and then the user is used. May be an optional mode).
In this way, the computer terminal 10 predicts the death time after waiting for the arrival of the reference time by using a prediction function corresponding to the patient information at a certain time point (reference time) just before death, so that the accuracy is extremely high. The time of death can be predicted by (less than the time order).

<Method of estimating a predetermined time before death using the modified early warning score>
Subsequently, assuming that the computer terminal 10 uses a second prediction function ( _{a prediction function for deriving T 2} ), as a specific example of a method of estimating a predetermined time before the death time, a modified early warning score Will be described.

Here, a method for deriving the modified early warning score will be described. The table below is used to derive the modified early warning score.

The computer terminal 10 compares the respiratory rate (RR), oxygen saturation (SpO2), body temperature (BT), systolic blood pressure (ABPs), and heart rate (HR) acquired for each unit time with the above table. , Find the score for each item. Then, the total value of the obtained scores becomes the modified early warning score. Therefore, for example, when the patient information of the patient 30 at a certain time point is 18 breaths, 95% oxygen saturation, 38.5 degrees body temperature, 100 mmHg systolic blood pressure, and 100 heart rates, the modified type. The early warning score is 5 (= 0 + 1 + 1 + 2 + 1).
Since it is derived as described above, the modified early warning score is derived as an integer of 0 (zero) or more.

FIG. 4 is a diagram showing the time course of the modified early warning score derived for three patients who received treatment intervention in the intensive care unit. Each plot illustrated in FIG. 4 shows the measured values of the modified early warning score at that time. The solid line shown in FIG. 4 shows the average value of the modified early warning scores from that time point to 20 minutes before. In FIG. 4, the time when the treatment intervention is performed is set to "100 minutes", 50 minutes before the treatment intervention is set to "50 minutes", and 100 minutes before the treatment intervention is set to "0 minutes". ..

As shown in FIG. 4, the average value of the modified early warning score tends to increase with time toward the time of treatment intervention. Based on the tendency of such a modified early warning score to change, it is possible to automatically estimate that the patient has become severe (the time of death is near) by computer processing.
More specifically, the computer terminal 10 is a patient of a subject who has acquired a modified early warning score (first index value) derived based on heart rate, systolic blood pressure, respiratory rate, oxygen saturation and body temperature. Derived every unit time using information. Then, in the computer terminal 10, the slope index obtained by dividing the difference between the highest value and the lowest value of the derived modified early warning scores by the difference between the time point corresponding to the highest value and the time point corresponding to the lowest value sets the threshold value. It is possible to presume that the time point exceeded is reached before the predetermined time of death, and to set the time point as the above-mentioned reference time.

The above-mentioned estimation method based on the angle (slope index) between the highest point and the lowest point of the modified early warning score is a specific example. In addition to this, an estimation method using changes (fluctuations) in the plot pattern of the modified early warning score and variations from the standard deviation of the modified early warning score (that is, with respect to the index transition display unit DR2 shown in FIG. 2). An analysis-based estimation method) is also useful in the practice of the present invention.

<Method of estimating a predetermined time before death using regression analysis>
Subsequently, as a specific example of a method of estimating a predetermined time (for example, 2 hours) before the death time on the premise that the computer terminal 10 uses a second prediction function ( _{a prediction function for deriving T 2).} A case where regression analysis is used will be described.
This method focuses on systolic blood pressure and heart rate, and uses a shock index and systolic blood pressure. Normally, systolic blood pressure and heart rate are linked while the autonomic nerves are maintained. However, at the time of death in a terminally ill patient, this autoregulatory ability seems to be disrupted, causing a major disturbance in the relationship between shock index and systolic blood pressure.

FIG. 5 is a plot of systolic blood pressure and shock index from a decrease in blood pressure (systolic blood pressure is below 80 mmHg) to a heart rate of 0 (zero), and linear regression analysis and non-linear regression analysis are performed. Is.
There are variations when all the data are used, and the relationship is not clear (see FIG. 5 (a)). Therefore, the data were classified into 2 hours before death (see FIG. 5 (b)) and 2 hours before blood pressure decrease (see FIG. 5 (c)). It should be noted that the nonlinear regression analysis was used in the analysis on the assumption that one exponential decay is applicable.
The point that deviates from the non-linear regression analysis calculated by the plot up to 2 hours before death is that the interlocking of systolic blood pressure and heart rate has stopped due to autonomic imbalance, and it is considered that death is near. ..

From the above viewpoint, the following non-linear regression analysis was performed to construct a method for estimating 2 hours before the death time.

First, among the patient information on 19 patients who confirmed death in the intensive care unit, which was analyzed when constructing the above-mentioned prediction function of death time, all data after the systolic blood pressure dropped to 80 mmHg or less. Data of 12050 points was extracted.
Then, from the extracted 12050 points of data, inappropriate data (12 points) such as artifacts were excluded, and 12038 points were selected as the analysis target. Among these, 9915 points exist in the data group from the decrease in blood pressure to 2 hours before death (Volatile data set), and the data group from 2 hours before death to death (Terminal data set) is. There are 2111 points.
A set for constructing a prediction formula (Development Set; hereinafter referred to as a construction set) at random from each of the data group from the decrease in blood pressure to 2 hours before death and the data group from 2 hours before death to death, and prediction. It was divided into a set for evaluating the validity of the formula (Validation Set; hereinafter referred to as an evaluation set). As a result, the construction set for the data group from the decrease in blood pressure to 2 hours before death was 7982 points, and the evaluation set for the data group was 1983 points. The construction set for the data group from 2 hours before death to death was 1689 points, and the evaluation set for the data group was 422 points.
FIG. 6 shows a diagram that visualizes the classification method of the data group to be the target of the regression analysis shown above.

なお、上式における決定係数（R-squared）は0.6996となった。ここで決定係数とは、独立変数である収縮期血圧（SBP）によって説明される応答変数であるショックインデックス（SI）の変化に比例する量を表す指標である。 Of the 7982 point data group included in the construction set related to the data group from the decrease in blood pressure to 2 hours before death, a nonlinear regression analysis was performed using the 6293 point data group, and the following prediction formula was derived.

The coefficient of determination (R-squared) in the above equation was 0.6996. Here, the coefficient of determination is an index representing an amount proportional to a change in the shock index (SI), which is a response variable explained by the systolic blood pressure (SBP), which is an independent variable.

FIG. 7 is a diagram showing a comparison between the above-mentioned shock index prediction formula and the measured value of the shock index.
FIG. 7A plots 1689 points of the data included in the construction set related to the data group from the decrease in blood pressure to 2 hours before death, excluding the 6293 points used for the nonlinear regression analysis, and then plots the above. A curve representing the prediction formula is inserted. As shown in the figure, there are few plots in which the measured values of the shock index from the decrease in blood pressure to 2 hours before death deviate significantly from the prediction formula.
FIG. 7B is a plot of 1689 points, which is a construction set related to a data group from 2 hours before death to death, and then a curve representing the above prediction formula is inserted. As shown in the figure, the measured values of the shock index from 2 hours before death to death are conspicuously deviated from the prediction formula.

By multiplying the predicted shock index obtained by using the above-mentioned shock index prediction formula by the measured systolic blood pressure, it can be treated as the predicted heart rate. In other words, for multiple patients who have died in the past, the shock index and systole for each unit time during the period from when the systolic blood pressure falls below a predetermined value to 2 hours before the death time (predetermined time). By multiplying the systolic blood pressure prediction formula of the shock index obtained by nonlinear regression analysis of the blood pressure with the systolic blood pressure as a variable, a function (second function) for obtaining the predicted heart rate can be derived. ..

Using the data group corresponding to the construction set, the difference between the predicted heart rate obtained by the above function and the actually measured heart rate of the patient 30 was obtained. The results are shown in the table below, with the differences based on the construction set for the data group from 2 hours before death to 2 hours before death and the differences based on the construction set for the data group from 2 hours before death to 2 hours before death. show.

As shown in the table above, the difference based on the construction set related to the data set (Volatile data set) from the decrease in blood pressure to 2 hours before death is approximately 0 (zero) (median: 1.90, 95% confidence). Section: 1.1 to 2.9) tended to be.
On the other hand, the difference based on the construction set related to the data set from 2 hours before death to death (median: -9.8, 95% confidence interval:) is relatively different from 0 (zero). There was a tendency for it to be -10.9 to -8.5).

Furthermore, a nonparametric test is performed to compare and examine the difference based on the construction set for the data group from 2 hours before death to 2 hours before death and the difference based on the construction set for the data group from 2 hours before death to 2 hours before death. A log rank test was performed using the ROC (Receiver Operating Characteristic) curve. In these tests, a significant difference was found between the two (p <0.0001).

The table below shows the sensitivity and specificity obtained based on the difference between the predicted heart rate and the measured heart rate.
The numerical values described in the right column of the sensitivity column and the specificity column are confidence intervals of sensitivity or specificity at a confidence level of 95%. The numerical values listed in the rightmost column of each line are the likelihood ratios of sensitivity and specificity.

Looking at the case where the threshold is -10.05 (heart rate difference <-10.05) in the above table, when the heart rate difference <-10.05, the construction set for the data group from 2 hours before death to death 49.85% of the difference based on is included, and if the heart rate difference> -10.05, 75.84% of the difference based on the construction set for the data group from the decrease in blood pressure to 2 hours before death is included.
Furthermore, looking at the case where the threshold value is -9.950 (heart rate difference <-9.950) in the above table, when the heart rate difference <-9.950, it relates to the data group from 2 hours before death to death. 49.91% of the difference based on the construction set is included, and if the heart rate difference> -9.950, 75.31% of the difference based on the construction set for the data group from the decrease in blood pressure to 2 hours before death is included.
Judging from this analysis result, the present inventor has determined that it is appropriate to set the threshold value to less than -10.
This threshold is a specific example, and can be changed depending on the timing of estimation (how many hours before death is used as the reference time) and can be changed by changing the range of patient information to be analyzed. The threshold in the practice of the invention is not limited to this.

When the threshold value is less than -10, 409 points are present in the construction set (1689 points) related to the data group from the decrease in blood pressure to 2 hours before death plotted in FIG. 7 (a), and the ratio thereof. Is 24.2%.
In this case, of the construction set (1689 points) related to the data group from 2 hours before death to death plotted in FIG. 7 (b), 842 points were applicable, and the ratio was 49.9%. Is.

Subsequently, in order to evaluate the validity of the threshold value set using the construction set, the positive predictive value and the negative predictive value are obtained using the evaluation set.
When the threshold value is less than -10, a heart rate difference <-10 is treated as a positive finding. Here, 400 points of the evaluation set (1982 points) related to the data group from the decrease in blood pressure to 2 hours before death correspond to positive findings, and the evaluation set (421 points) related to the data group from 2 hours before death to death Of these, 222 points correspond to positive findings. Therefore, of these, the former is false positive data and the latter is true positive data.
When the threshold value is less than -10, a heart rate difference ≥ -10 is treated as a negative finding. Here, 1582 points of the evaluation set (1982 points) related to the data group from the decrease in blood pressure to 2 hours before death correspond to negative findings, and the evaluation set (421 points) related to the data group from 2 hours before death to death Of these, 199 points correspond to negative findings. Therefore, of these, the former is true negative data and the latter is false negative data.

Based on the above arrangement, the positive predictive value = 222 ÷ (400 + 222) = 35.7%. The negative predictive value = 1582 ÷ (1582 + 199) = 88.8%.
In this case, the sensitivity obtained based on the difference between the predicted heart rate and the measured heart rate was 52.7%, and the specificity was 79.8%.
Based on this result, it is considered that the method of estimating 2 hours before the death time based on the nonlinear regression analysis examined in the present embodiment has sufficiently high judgment accuracy.

The above estimation method is based on nonlinear regression analysis using patient information on multiple patients who have died in the past, but as another estimation method, linear regression analysis or nonlinear regression analysis using patient information on the subject itself is performed. It is also conceivable to perform this and treat the time point that deviates from the tendency of the data as being on the verge of death (predetermined time before the time of death).
8 to 13 are diagrams showing the relationship between the shock index and systolic blood pressure based on the patient information about some patients accumulated in the database 40.
More specifically, FIGS. 8-10 plot the shock index and systolic blood pressure based on patient information from hypotension to 2 hours before death, and each plot is subjected to linear or non-linear regression analysis. It represents a regression equation that expresses the relationship between the two.
Further, FIGS. 11 to 13 plot the shock index and systolic blood pressure based on patient information from 2 hours before death to the time of death, and use the same regression equation as the regression equation shown in FIGS. 8 to 10. It is the one that was inserted.
It is common to all the drawings that the horizontal axis represents systolic blood pressure and the vertical axis represents shock index. Then, "No. X (X is an arbitrary integer)" shown in each figure is a number for identifying each patient.

As is clear from comparing these drawings, the variation of the regression equation (predicted value) and plot (measured value) in the period from the decrease in blood pressure to 2 hours before death, and the period from 2 hours before death to death Comparing the regression equation and the variation of the plot, the latter variation is large for all patients.
In addition, the regression equation is different for all patients, and the degree of variation between the regression equation and the plot is also different for each patient. That is, the balance between systolic blood pressure and shock index varies from patient to patient, and it is considered possible to improve the prediction accuracy by creating a regression equation of the subject himself and obtaining the predicted heart rate.
In other words, at a certain point in time from when the subject's systolic blood pressure falls below a predetermined value to before death, the shock index and systolic blood pressure for each unit time based on the subject's patient information acquired before that point in time. Is derived by multiplying the systolic blood pressure prediction formula of the shock index obtained by linear regression analysis or non-linear regression analysis with the systolic blood pressure as a variable to obtain the predicted value of the heart rate (second function). be able to. Then, the function can predict the heart rate of the subject with higher accuracy than the uniform prediction formula derived based on the regression analysis using the patient information of another person, and the time of death is predetermined. It can be applied to computer processing that estimates the time before.

In the above, various methods for estimating a predetermined time before the death time by using regression analysis have been described. By applying any of these estimation methods, it is possible to estimate that it is 2 hours before the death time (the death time is just before the death time) by computer processing. Then, the estimated timing can be set as the reference time, and the death time of the subject can be predicted with high accuracy by using the patient information at the reference time.
In other words, the computer terminal 10 gives the acquired systolic blood pressure of the subject and derives the predicted heart rate for each unit time. Then, the computer terminal 10 estimates that the time when the difference between the derived predicted heart rate and the acquired heart rate of the subject falls below the threshold value has reached 2 hours before the time of death, and the time point Can be set as the above reference time.

<Method of estimating a predetermined time before death using a static scoring model>
Subsequently, assuming that the computer terminal 10 uses a second prediction function ( _{a prediction function for deriving T 2} ), a static scoring model is used as a specific example of a method of estimating a predetermined time before the death time. Will be described.
Here, the static scoring model is a two-dimensional model illustrated on the model display unit DR3 of FIG. The computer terminal 10 has a function for generating a static scoring model (hereinafter referred to as a model generation means) and a function for plotting actual measurement values on the static scoring model (hereinafter referred to as a scoring means). And have.

The model generation means is patient information acquired within a predetermined period including at least before the time when each patient becomes severe from a database 40 accumulating a plurality of types of patient information acquired over time for a plurality of patients. The first and second variables (for example, modified early warning score and shock index) are derived from the extracted patient information, and the derived first and second variables are on the vertical and horizontal axes, respectively. It is possible to generate a static scoring model corresponding to.
The scoring means acquires a plurality of types of patient information regarding the patient 30 as a subject for each unit time, derives the first variable and the second variable from the acquired patient information, and plots them sequentially on a static scoring model. go.
The static scoring model according to the present invention is "static" because the region division (state stabilization zone NZ, caution zone WZ, terminal zone TZ) does not change after being generated by the model generation means. It is called. Therefore, the division of the area related to the static scoring model can be variably changed depending on the conditions used by the model generation means, and the display mode of the static scoring model displayed on the display screen of the computer terminal 10 can also be changed. Can be changed.

The computer terminal 10 sets at least a part of the time when the plot is performed in the terminal zone TZ (specific area) which is a part of the two-dimensional area in the static scoring model a predetermined time before the death time (for example, 2 hours). It is presumed that the previous) has been reached, and the relevant time point may be set as the above-mentioned reference time. However, even if the plot is made in the terminal zone TZ, it is not necessary to immediately estimate that the plot has arrived before the predetermined time of death. This is because even if the condition of the patient 30 is temporarily changed and a plot is made in the terminal zone TZ, if it recovers immediately, it is considered unlikely that the patient's condition has become severe. Therefore, when plotting in the terminal zone TZ for a certain period of time (in a certain number of times in a row), it may be estimated that the item has arrived before a predetermined time of death.

In order to properly set the threshold of the terminal zone TZ, variables (for example, modified early warning score or shock index) obtained based on the data group from the decrease in blood pressure to 2 hours before death and 2 hours before death The threshold value of the variable can be obtained by comparing the sensitivity and specificity derived using the variable based on the data group up to death.
Since such a method of obtaining the threshold value applies the method of obtaining the threshold value of the difference in heart rate described above, detailed description here will be omitted.

Since the estimation method using the static scoring model as described above is based on two variables, the accuracy is improved as compared with the estimation method based on a single variable (for example, only the modified early warning score). As a result, the accuracy of predicting the time of death can be improved.

<About a modified example of the present invention>
Although the present invention has been described above in accordance with the embodiments, the present invention is not limited to the above-described embodiments, and includes various modifications, improvements, and the like as long as the object of the present invention is achieved. ..

Although the computer terminal 10 has been described as having a display screen in the above embodiment, the death prediction device or the computer that executes the death prediction program according to the present invention does not necessarily have to be provided with a display means (display or the like).

In the above embodiment, a case where a certain time point before the death of the subject is set as a reference time and the death time is predicted based on the measured value of the patient information at the reference time has been described. Not limited.
For example, if it can be determined that the subject is on the verge of death when the subject is brought to the intensive care unit, the time of death may be predicted based on the measured value of the patient information at that time.

This embodiment includes the following technical ideas.
(1) By using the acquisition means for acquiring a plurality of types of patient information about a subject at a frequency of minutes or less and the patient information of the subject acquired by the acquisition means, the death time of the subject is set to the time order or less. A death prediction device including a death prediction means for predicting with.
(2) The death prediction means sets a certain time point before the death of the subject as a reference time by using the patient information of the subject acquired by the acquisition means, and is acquired by the acquisition means at the determined reference time. The death prediction device according to (1), which predicts the death time of the subject by giving the patient information of the subject to the prediction function.
(3) The mortality prediction device according to (2), wherein the prediction function is a function having age, gender, systolic blood pressure, heart rate and respiratory rate as variables.
(4) The prediction function is either the period from the systolic blood pressure below the predetermined value to the death time or the period from the predetermined time before the death time to the death time for a plurality of patients who have died in the past. The mortality predictor according to (3), which is a function obtained by statistically analyzing systolic blood pressure, heart rate, and respiratory rate for each unit time in one of them, taking into consideration the age and sex of each patient.
(5) The mortality prediction means uses the patient information of the subject acquired by the acquisition means to obtain a first index value derived based on the heart rate, systolic blood pressure, respiratory rate, oxygen saturation and body temperature. The slope index is the threshold value obtained by dividing the difference between the highest value and the lowest value of the first index values derived for each unit time by the difference between the time point corresponding to the highest value and the time point corresponding to the lowest value. The mortality prediction device according to any one of (2) to (4), wherein the time point exceeded is set as the reference time.
(6) The mortality prediction means gives the systolic blood pressure of the subject acquired by the acquisition means to a second function with the systolic blood pressure as a variable, and derives and derives the predicted heart rate for each unit time. The death according to any one of (2) to (4), wherein the time point at which the difference between the predicted heart rate and the heart rate of the subject acquired by the acquisition means falls below the threshold time is defined as the reference time. Predictor.
(7) The second function is a shock index and systolic period for each unit time in the period from when the systolic blood pressure falls below a predetermined value to a predetermined time before the death time for a plurality of patients who have died in the past. The mortality predictor according to (6), which is a function for obtaining a predicted value of heart rate by multiplying a systolic blood pressure prediction formula obtained by nonlinear regression analysis of blood pressure with a systolic blood pressure as a variable. ..
(8) The second function is applied to the patient information of the subject acquired by the acquisition means before the time point at a certain point in time from when the systolic blood pressure of the subject falls below a predetermined value to before death. Based on the shock index for each unit time and the systolic blood pressure obtained by linear regression analysis or non-linear regression analysis, the shock index prediction formula with the systolic blood pressure as a variable is multiplied by the systolic blood pressure to predict the heart rate. The death prediction device according to (6), which is a function for obtaining.
(9) From a database accumulating a plurality of types of patient information acquired over time for a plurality of patients, patient information acquired within a predetermined period including at least before the time when each patient becomes severe is extracted. The first variable and the second variable are derived from the extracted patient information, and the derived static score is a two-dimensional model in which the first variable and the second variable correspond to the vertical axis and the horizontal axis, respectively. A model generation means for generating a ring model and a plurality of types of patient information regarding a patient who is a subject are acquired for each unit time, and the first variable and the second variable are derived from the acquired patient information to obtain the static score. A scoring means that sequentially plots on the ring model, the mortality predicting means, was performed on a specific region in which the plotting by the scoring means is a part of a two-dimensional region in the static scoring model. The mortality prediction device according to any one of (2) to (4), wherein at least a part of the time points is set as the reference time.
(10) By using the acquisition means for acquiring a plurality of types of patient information about the subject at a frequency of minutes or less and the patient information of the subject acquired by the acquisition means, the death time of the subject is set to the time order or less. A death prediction program that allows a computer to execute a death prediction means for predicting in.

10 Computer terminal 20 Measuring instrument 30 Patient 40 Database

Claims (10)

An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
By using the patient information of the subject acquired by the acquisition means, the death prediction means for predicting the death time of the subject on the order of time or less is provided .
The death prediction means
By using the patient information of the subject acquired by the acquisition means, a certain time point before the death of the subject is set as a reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
The prediction function is used for a plurality of patients who have died in the past in either the period from the systolic blood pressure below a predetermined value to the time of death or the period from a predetermined time before the time of death to the time of death. It is a function obtained by statistically analyzing systolic blood pressure, heart rate, and respiratory rate for each unit time, taking into account the age and gender of each patient.
A death prediction device characterized by the fact that. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
By using the patient information of the subject acquired by the acquisition means, the death prediction means for predicting the death time of the subject on the order of time or less is provided .
The death prediction means
The first index value derived based on the heart rate, systolic blood pressure, respiratory rate, oxygen saturation and body temperature is derived for each unit time using the patient information of the subject acquired by the acquisition means.
The reference time is the time when the slope index obtained by dividing the difference between the highest value and the lowest value of the derived first index values by the difference between the time point corresponding to the highest value and the time point corresponding to the lowest value exceeds the threshold value. Determine,
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction device characterized by the fact that. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
By using the patient information of the subject acquired by the acquisition means, the death prediction means for predicting the death time of the subject on the order of time or less is provided .
The death prediction means
The predicted heart rate is derived every unit time by giving the systolic blood pressure of the subject acquired by the acquisition means to the second function with the systolic blood pressure as a variable.
The time when the difference between the derived predicted heart rate and the subject's heart rate acquired by the acquisition means falls below the threshold value is set as the reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction device characterized by the fact that. The second function is a non-linear shock index and systolic blood pressure for each unit time in the period from when the systolic blood pressure falls below a predetermined value to a predetermined time before the death time for a plurality of patients who have died in the past. The mortality prediction device according to claim 3 , which is a function for obtaining a predicted value of heart rate by multiplying a prediction formula of a shock index using systolic blood pressure as a variable obtained by regression analysis by systolic blood pressure. The second function is a unit time based on the subject's patient information acquired by the acquisition means before the time point at a certain point in time from when the systolic blood pressure of the subject falls below a predetermined value to before death. A function that multiplies the systolic blood pressure by the systolic blood pressure to obtain the predicted value of the heart rate. The death prediction device according to claim 3. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
By using the patient information of the subject acquired by the acquisition means, the death prediction means for predicting the death time of the subject on the order of time or less is provided .
In addition
From a database that stores multiple types of patient information acquired over time for a plurality of patients, patient information acquired within a predetermined period including at least before the time when each patient becomes severe is extracted and extracted. A static scoring model, which is a two-dimensional model in which the first variable and the second variable are derived from the obtained patient information and the derived first and second variables correspond to the vertical axis and the horizontal axis, respectively, is obtained. Model generation means to generate and
Scoring in which a plurality of types of patient information regarding a patient as a subject are acquired for each unit time, the first variable and the second variable are derived from the acquired patient information, and the static scoring model is sequentially plotted. With means,
The death prediction means
At least a part of the time when the plot by the scoring means is performed on a specific area which is a part of the two-dimensional area in the static scoring model is set as a reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction device characterized by the fact that. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
A death prediction program for causing a computer to execute a death prediction means for predicting the death time of the subject on the order of time or less by using the patient information of the subject acquired by the acquisition means .
The death prediction means
By using the patient information of the subject acquired by the acquisition means, a certain time point before the death of the subject is set as a reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
The prediction function is used for a plurality of patients who have died in the past in either the period from the systolic blood pressure below a predetermined value to the time of death or the period from a predetermined time before the time of death to the time of death. It is a function obtained by statistically analyzing systolic blood pressure, heart rate, and respiratory rate for each unit time, taking into account the age and gender of each patient.
A death prediction program characterized by this. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
A death prediction program for causing a computer to execute a death prediction means for predicting the death time of the subject on the order of time or less by using the patient information of the subject acquired by the acquisition means .
The death prediction means
The first index value derived based on the heart rate, systolic blood pressure, respiratory rate, oxygen saturation and body temperature is derived for each unit time using the patient information of the subject acquired by the acquisition means.
The reference time is the time when the slope index obtained by dividing the difference between the highest value and the lowest value of the derived first index values by the difference between the time point corresponding to the highest value and the time point corresponding to the lowest value exceeds the threshold value. Determine,
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction program characterized by this. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
A death prediction program for causing a computer to execute a death prediction means for predicting the death time of the subject on the order of time or less by using the patient information of the subject acquired by the acquisition means .
The death prediction means
The predicted heart rate is derived every unit time by giving the systolic blood pressure of the subject acquired by the acquisition means to the second function with the systolic blood pressure as a variable.
The time when the difference between the derived predicted heart rate and the subject's heart rate acquired by the acquisition means falls below the threshold value is set as the reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction program characterized by this. An acquisition method for acquiring multiple types of patient information about a subject at a frequency of less than a minute order,
A death prediction means for predicting the death time of the subject on the order of time or less by using the patient information of the subject acquired by the acquisition means.
From a database that stores multiple types of patient information acquired over time for a plurality of patients, patient information acquired within a predetermined period including at least before the time when each patient becomes severe is extracted and extracted. A static scoring model, which is a two-dimensional model in which the first variable and the second variable are derived from the obtained patient information and the derived first and second variables correspond to the vertical axis and the horizontal axis, respectively, is obtained. Model generation means to generate and
Scoring in which a plurality of types of patient information regarding a patient as a subject are acquired for each unit time, the first variable and the second variable are derived from the acquired patient information, and the static scoring model is sequentially plotted. Means and
Is a death prediction program that lets a computer run
The death prediction means
At least a part of the time when the plot by the scoring means is performed on a specific area which is a part of the two-dimensional area in the static scoring model is set as a reference time.
By giving the patient information of the subject acquired by the acquisition means to the prediction function at the determined reference time, the death time of the subject is predicted.
A death prediction program characterized by this.

Images