TW201926157A - Method for predicting the daily life function of disabled people capable of properly arranging caring resources and reducing waste of caring resources - Google Patents

Abstract

A method for predicting the daily life function of disabled people includes inputting the measurement results of a plurality of rehabilitation assessment indicator sets of a plurality of disabled persons and their corresponding daily life function states into a mechanical learning machine; selecting the variable values of several rehabilitation assessment indicator sets with the best classification performance through the mechanical learning machine; using a supervised mechanical learning algorithm to establish a daily life function state prediction model in cooperation with the corresponding daily life function states. Finally, inputting a test result of the plurality of rehabilitation assessment indicator sets of a new examinee into the daily life function state prediction model, and using the internal calculations of the daily life function state prediction model to obtain a prediction result of the daily life function state.

Description

A method for predicting the function of daily life of disabled person

The present invention provides a method for predicting the function of daily life of a disabled person, which is to predict the future functional state of daily life of the subject through the establishment of a functional state prediction model for daily life, and to appropriately allocate care resources to reduce unnecessary Waste of care resources.

A person who is physically and mentally disabled (referred to as a disabled person) is defined as the loss of part or all of the physical or mental function, so that his or her daily life needs assistance from others. The degree of disability is usually judged by the Daily Life Function Scale (ADLs) or the Instrumental Daily Living Function Scale (IADLs), which are usually classified into mild, moderate, and severe disability according to the condition. According to statistics, Taiwan had nearly 670,000 disabled people in 2011, of which about 410,000 were over 65 years old. By 2020, it will be estimated to be 860,000 disabled, and those who are over 65 will exceed 600,000. .

Today's society often lacks care during mild disability, and then quickly deteriorates into moderate and severe disability. Today's medical technology is enough to sustain the lives of the disabled, and the number of people with moderate or severe disability will be long-term. More and more, the burden on the state and society is getting heavier and heavier.

In order to overcome the above problems, the prior art uses a manual assessment of a single risk factor's rehabilitation assessment scale to evaluate the function of the disabled's daily life. However, the above method lacks systematic evaluation and does not use multiple laboratory test data. Unable to assess the distribution of the overall data and inefficiency, it is not easy to know the future functional status of the disabled, and then It affects its efficacy in clinical use, so there is still room for improvement in correctness, timeliness and reproducibility of interpretation results.

The present invention provides a method for predicting the function of daily life of a disabled person, mainly by applying a plurality of rehabilitation assessment index sets to the plurality of disabled persons, and inputting the measurement results and the daily functional status of each disabled person. In the mechanical learning machine, the number of variables of the rehabilitation evaluation index set with the best classification efficiency is selected through the mechanical learning machine, and the corresponding daily life function state is matched, and the supervised mechanical learning algorithm is used to establish the daily routine. The life function state prediction model, finally, the new subject is subjected to the measurement of a plurality of rehabilitation evaluation index sets, and then the test results are input into the daily life function state prediction model, and the daily functional state calculation can be performed. And get the predicted results of the functional status of daily life; the results obtained can remind the subject to follow up, achieve high efficiency to predict the future life status of the subject, and allocate care resources to reduce unnecessary waste of care resources For its main purpose.

The second object of the present invention is to solve the problem of big data interpretation by the daily function state prediction model, which includes diverse information, which enables medical personnel to learn the body of the disabled from more aspects. The situation and daily life functions, so the more evaluation indicators, the better the evaluation will be, but the big data is directly judged by the clinical medical staff, and may have its shortcomings in timeliness and correctness. Therefore, the present invention utilizes supervision. The mechanical learning algorithm is able to analyze the difference in the distribution of the disability assessment indicator sets from the existing data to the greatest extent, and find the classification basis from the overall data distribution appearance to predict the future case. The daily life state is mild, moderate or heavy; therefore, using the established daily life function state prediction model instead of human interpretation, it increases the overall interpretation efficiency. And the correct rate, and the daily life function state prediction model can also be copied to the user's terminal for use in various aspects.

Figure 1 is a schematic flow chart of the present invention

2 is a ROC curve of a toolive daily life function score prediction model of the present invention

Figure 3 is the performance of the toolive daily life function score prediction model of the present invention, taking the area under the ROC curve as an index

Figure 4 is a comparison table of the supervised mechanical learning algorithms and the single indicator predictive performance of the composite index with the area under the ROC curve as an index.

For the embodiments and principles of the present invention, please refer to the following description: First, referring to FIG. 1, the subject concept provided by the present invention is a method for predicting the daily function of a disabled person, and multiple statistical scores of the multiple disabled persons. Rehabilitation assessment scale and laboratory test values are used to establish a set of rehabilitation assessment indicators; after the multiple disabled persons are individually tested for multiple rehabilitation assessment indicators, the measurement results and their daily lives are corresponding to each disabled person. The function status of the life is input into the mechanical learning machine; the variable values of the rehabilitation evaluation index sets with the best classification performance are selected through the mechanical learning machine, and the corresponding daily functional state is matched with the supervised mechanical learning calculation The method is to establish a predictive model of daily life function state; finally, the new subject is subjected to the measurement of a plurality of rehabilitation evaluation index sets, and the test results are input into the daily life function state prediction model, and the daily life function can be performed. The operation of the state, and the prediction result of the function state of daily life, and then the subject can be reminded to follow the prediction result Action in the state.

The date of determination of the daily functional status and the evaluation date of the rehabilitation assessment scale are between 2 weeks and 1 year.

The rehabilitation assessment scale is the Modified Rankin Scale (MRS), the Barthel Index, the Functional Oral Intake Scale (FOIS), and the Mini Nutrition Assessment Scale. (Mini Nutrition Assessment, MNA), Healthy Quality of Life Measurement Questionnaire (Euro QoL-5D), Instrumental Daily Life Function Scale (IADL Scale), Berg Balance Scale (BBS), Walking Rate (Gait) Speed), Six minutes walking test (6MWT), Fugl-Meyer Assessment (FMA), Mini-mental state examination (MMSE), Activity Activity Scale (Motor Activity Log, MAL), Concise Chinese Aphasia Test (CCAT), or any combination of the above.

The functional status of daily life is assessed using the Barthel Scale, the Barrier Index, the Instrumental Daily Function Score (IADL Scale), or the Rayleigh Correction Scale Score (MRS).

The laboratory test values are full blood test (CBC), white blood cells differential counts, total protein, albumin, Leukocyte Esterase, C-reactive protein ( C-Reactive Protein), Procalcitonin, Erythrocyte Sedimentation Rate, Lactate, Lactate Dehydrogenase, Sugar, Na(Na), Potassium Ions (K), calcium (Ca), chloride (Cl), magnesium (Mg), ferrous (Fe ²⁺ ), iron (Fe ³⁺ ), urea nitrogen (Urea Nitrogen), creatinine ( Creatinine), Cystatin C, Bilirubin, Low Density Lipoprotein (LDL), High Density Lipoprotein (HDL), Triglyceride, Total cholesterol, blood sugar, microalbumin, glycosylated hemoglobin (HbA1C), homocysteine, lipoprotein A, Uric acid or the like random combination.

The supervised mechanical learning algorithm uses logical regression, K-neighbor method, support vector machine, neural network learning, decision tree, random forest, Bayesian decision method or any combination of the above.

Referring to FIG. 2 and FIG. 3 together with FIG. 1 , the specific implementation manner is as follows: This embodiment is based on the rehabilitation evaluation index set of the brain stroke patients in the north hospital area of Chang Gung Hospital, and the rehabilitation evaluation index is set. After inputting the corresponding daily function state, it is input into the mechanical learning machine, and the predictive model of daily life function state is established through the supervised mechanical learning machine algorithm.

Subject conditions (inclusion, exclusion conditions), number: The subjects were rehabilitation data of patients with cerebral apoplexy in the Changyuan Hospital North Campus. The medical records were retrospectively and no additional subjects were required.

DESIGN AND METHODS: The main clinical information on the functional status of daily life is the Modified Rankin Scale (MRS), the Barthel Index, and the Functional Oral Intake Scale (Dental). FOIS), Mini Nutrition Assessment (MNA), Healthy Quality of Life Measurement Questionnaire (Euro QoL-5D), Instrumental Daily Life Function Scale (IADL Scale), Berg Balance Scale (Berg Balance Scale) , BBS), step Gait speed, Six minutes walking test (6MWT), Fugl-Meyer Assessment (FMA), Mini-mental state examination (MMSE), Motor Activity Log (MAL), Concise Chinese Aphasia Test (CCAT) scale score. The 264 adults, at the same time, have evaluation data on the functional status of daily life at the time of discharge. Therefore, in the above data, the variable selection method is used to select the variable value of the daily function state with the best classification efficiency. Variable selection: After the preliminary data screening, this example uses univariate analysis, selects the appropriate univariate statistical method according to the variable characteristics (chi-square test and t-test), and selects the Rayleigh correction scale (MRS), Berg. The Balanced Balance Scale (BBS) and the Instrumental Daily Life Function Scale (IADL Scale) are the variables with the best classification performance. After selecting the best variables, input them into the mechanical learning machine using complex supervised mechanical learning algorithms such as Logistic Regression (LR), Decision Tree (DT), Random Forest (Random Forest, RF), K Nearest Neighbor (KNN), Support Vector Machines (SVM), Artificial Neuron Network, Bayesian Network, etc. Forecast model.

Data review period, implementation period of this embodiment: data backtracking from March 2014 to October 2016.

Evaluation and verification method of the results: This embodiment calculates the distribution of each rehabilitation evaluation index set, and trains the daily life functional state prediction model according to the variable and its numerical value, and internally verifies the predictive ability of each daily life function state prediction model. . The effectiveness of the daily life functional state prediction model will be verified by the receiver operating characteristic curve (ROC curve), and the area under the curve will be calculated simultaneously.

Efficacy: Figure 2 and Figure 3 show that the daily life function state prediction model is tested using different supervised mechanical learning algorithms. The area under the ROC curve of these supervised mechanical learning algorithms can be used as a performance evaluation to predict instrumentality. The performance of the IADL Scale. The supervised mechanical learning algorithms used include logical regression, decision trees, random forests, and K proximity. The area under the curve of logistic regression is 0.84, the decision tree is 0.75, the random forest is 0.86, and the K is 0.77. In general, the various mechanical learning methods are quite good, including logical regression and random forest. It is better.

Figure 4 is based on the average area under the ROC curve (AUC average) as an indicator, comparing the use of complex indicators (with different supervised mechanical learning algorithms) and a single indicator to predict the performance of daily functional status after discharge. The scale value of the subject at the time of admission is used as a baseline to predict the scale value after discharge. The supervised mechanical learning algorithm used: the logistic regression AUC average is 0.796 (AUC standard error is 0.015), the random forest AUC average is 0.792 (AUC standard error is 0.014), and the support vector machine AUC average is 0.774 ( The AUC standard error is 0.028). The single indicator used: the average AUC of the Barometer scale is 0.756 (AUC standard error is 0.029), the average AUC of the instrumental daily life function scale is 0.681 (AUC standard error is 0.035), and the Berg balance scale AUC The average value is 0.720 (AUC standard error is 0.032). From the above point of view, the standard error of using the complex index (supervised mechanical learning algorithm) is significantly smaller than the standard error of the single index, and the average AUC of each supervised mechanical learning algorithm is greater than the average of the AUC of the single index, which proves The addition of multiple rehabilitation assessment indicators in the rehabilitation assessment indicator set can not only increase the effectiveness of daily life functional status prediction, but also use the different supervised mechanical learning algorithms to conduct rehabilitation assessment index set data learning and use. Can be significantly Improve the predictive performance of daily functional status.

Therefore, the present embodiment analyzes a plurality of multiple rehabilitation evaluation indicators using the daily life function state prediction model, and can accurately predict the daily life function state.

It is to be noted that the above-described embodiments are merely illustrative of the principles of the invention and its advantages, and are not intended to limit the scope of the invention. Modifications and variations of the embodiments can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention should be as described in the appended claims.

Claims (7)

A method for predicting the function of daily life of a disabled person, comprising: A. establishing a plurality of rehabilitation assessment indicator sets, the rehabilitation evaluation indicator sets having a statistical assessment score and a laboratory test value; After the multiple disabled persons are respectively subjected to the test of the plurality of rehabilitation evaluation index sets, the measurement results of the rehabilitation evaluation index sets and the daily function states of the respective disabled persons are input into the mechanical learning machine; Selecting the variable value of the rehabilitation evaluation indicator set with the best classification efficiency by the mechanical learning machine, matching the corresponding daily life function state, and establishing the daily life function state prediction model through the supervised mechanical learning algorithm; D. The new subject is subjected to the measurement of a plurality of rehabilitation assessment indicator sets, and the results of the test are input into the daily life function state prediction model, and the daily functional state calculation can be performed, and the daily life function can be obtained. The predicted result of the state. For example, in the method for predicting the daily function of the disabled person according to the first aspect of the patent application, after obtaining the prediction of the functional state of the daily life, the subject may be reminded to take the action of the subsequent corresponding state according to the predicted result. A method for predicting the function of daily life of a disabled person according to claim 1, wherein the date of determination of the daily functional status and the evaluation date of the rehabilitation assessment scale are between 2 weeks and 1 year. A method for predicting the daily function of a disabled person as described in claim 1, wherein the rehabilitation assessment scale is a Modified Rankin Scale (MRS), a Barthel Index, Functional oral intake scale (Functional oral intake scale, FOIS), Mini Nutrition Assessment (MNA), Healthy Quality of Life Measurement Questionnaire (Euro QoL-5D), Instrumental Daily Life Function Scale (IADL Scale), Berg Balance Scale (Berg Balance Scale) , BBS), Gait speed, Six minutes walking test (6MWT), Fugl-Meyer Assessment (FMA), Mini-mental state examination , MMSE), Motor Activity Log (MAL), Concise Chinese Aphasia Test (CCAT), or any combination of the above. A method for predicting the function of daily life of a disabled person according to the first aspect of the patent application, wherein the functional status of daily life is a Barthel Index, a daily life function score (IADL) Scale) or Modified Rankin Scale (MRS) for evaluation. A method for predicting the function of daily life of a disabled person as described in claim 1, wherein the laboratory test values are a complete blood test (CBC), white blood cells differential counts, and total protein (Total Protein). ), Albumin, Leukocyte Esterase, C-Reactive Protein, Procalcitonin, Erythrocyte Sedimentation Rate, Lactic Acid, Lactic Acid Lactate Dehydrogenase, Sugar, Sodium (Na), Potassium (K), Calcium (Ca), Chloride (Cl), Magnesium (Mg), Ferrous (Fe ^{2 +} ), iron ion (Fe ³⁺ ), urea nitrogen (Urea Nitrogen), Creatinine, Cystatin C, Bilirubin, Low Density Lipoprotein (LDL) , High Density Lipoprotein (HDL), Triglyceride, Total cholesterol, Blood sugar, Microalbumin, Hemoglobin (HbA1C), Homocysteine Amino acid (Homocysteine), lipoprotein (L Ipoprotein A), Uric acid or any combination of the above. A method for predicting the function of daily life of a disabled person as described in claim 1, wherein the supervised mechanical learning algorithm uses logical regression, K-neighborhood, support vector machine, neural network learning, decision tree , random forest, Bayesian decision making, or any combination of the above.