Using natural language processing to find adverse events

Abstract

A method is provided for automatically classifying a patient's clinical description using natural language processing by a computer processor. The clinical description relates to the use of a ventricular assist device to treat the patient. The method includes receiving at least one clinical description comprising text. The method then further includes determining the location of a target word within the text. Furthermore, the method includes determining the presence of at least one negative word within an active region comprising a predetermined number of words, including the target word, appearing immediately before and after the target word within the text. The method then includes determining the presence of at least one body part word within the active region. Finally, the method includes determining that if the active region contains a negative word or a body part word, the clinical description will be ignored.

Description

[0001] Cross-reference of related applications This application claims the benefit of priority from 35 USC §119(e) of U.S. Provisional Application No. 62 / 784,192, filed December 21, 2018, the contents of which are incorporated herein by reference in their entirety. Background Technology

[0002] Cardiovascular conditions can reduce a patient's quality of life. Various treatment options have been developed for treating these cardiac conditions, ranging from medications to mechanical devices and transplants. Ventricular assist devices (VADs), such as heart pump systems and catheter systems, are commonly used in treating the heart to provide hemodynamic support and promote recovery. Some heart pump systems are percutaneously inserted into the heart and can operate in parallel with the patient's own heart to supplement cardiac output. One such heart pump system is Impella, manufactured by Abiomed, Inc., Danvers, Massachusetts. ® This is a series of devices. Some of these medical procedures may cause adverse events in patients during treatment. This may be due to improper use of the device or suboptimal device configuration.

[0003] After a patient has been treated using a medical device such as a VAD, a detailed description of the treatment provided to the patient, along with any clinical indications during such treatment, is recorded. Such records have traditionally been made manually using clinical sketching or by entering data into a computer. Alternatively, details of the treatment are provided to a device that converts speech to text (e.g., a recorder or an over-the-lap microphone running speech recognition software) and stored as a text file. Such files are typically recorded in a patient data repository and made available to other clinicians who need access to the patient's medical records.

[0004] Clinicians must typically determine whether a patient has experienced adverse events during treatment before deciding on any further treatment. Adverse events include, for example, bleeding, hemolysis, and ischemia that may have occurred while the patient was being treated with VADs (e.g., due to the use of Impella). ® (Pump and new guidewire). This type of determination involves reading and examining clinical records to manually determine whether any such adverse events have occurred during treatment. Manual examination involves several degrees of freedom. For example, certain parts of the text of the clinical description may be omitted or difficult to decipher, or the interpretation of the clinical description may vary from person to person. Furthermore, in order to obtain clinical indicators of the treatment plan (e.g., success rate), a repository of clinical descriptions needs to be analyzed to obtain representative indicators. Manual examination of a large number of clinical descriptions will be time-consuming and, due to the aforementioned degrees of freedom, may involve several inaccuracies.

[0005] Attempts to automate the analysis of clinical descriptions involve the use of natural language preprocessing and machine learning, such as bagging and random forests, logistic regression, and regression trees. These algorithms are complex and recursive, and consume significant processor resources on computer systems, especially when they are not prone to convergence. Machine learning using such algorithms typically requires substantial amounts of training data before a machine model can be relied upon. Therefore, for smaller data repositories, insufficient training data will result in unstable machine learning models whose outputs will be unreliable when analyzing clinical descriptions. Summary of the Invention

[0006] The method and system described herein use natural language processing and keyword search performed by a processor of a computing device to determine whether a clinical description relates to treatment involving an adverse event. The method begins by receiving at least one clinical description comprising text. The processor then determines the location of a target word within the text. The processor then proceeds to determine the presence of at least one negative word within an active region comprising a predetermined number of words, including the target word, appearing immediately before and after it within the text. Next, the processor determines the presence of at least one body part word within the active region. The method then includes determining that if the active region contains a negative word or a body part word, the clinical description will be ignored.

[0007] By searching for keywords in the text of clinical descriptions, no training of machine learning algorithms (which involves training data) is required, thereby freeing up system resources of the computing device. The keyword search nature of the method and system disclosed herein does not monopolize the processor of the computing device performing the analysis of the clinical descriptions.

[0008] In some embodiments, the method further includes processing the text to generate word identifiers, identifying and grouping word identifiers including variant forms of the words; and performing a keyword search on the text using the grouped word identifiers. In other embodiments, the method includes tagging the clinical description if the active region does not contain: negative words and body part words. In some embodiments, the method includes writing the tagging into a title in the clinical description. In some embodiments, the predetermined number of words in the active region is at least three words. In other embodiments, the predetermined number of words in the active region is three. In some embodiments, at least one negative word includes any of the following: 'no', 'not', 'nor', 'non', 'without', 'never', and 'false'. In some embodiments, the clinical description is obtained from an Acute Myocardial Infarction Cardiogenic Shock (AMICS) repository.

[0009] In another embodiment, a system is provided for automatically classifying clinical descriptions of patients. The system includes at least one ventricular assist device (VAD) for treating patients. The system also includes a controller that communicates with the VAD and is configured to generate at least one clinical description of the treatment for a patient using the VAD. Furthermore, the system includes a data repository for storing the clinical descriptions of treatments. The system also includes a processor that communicates with the data repository and is configured to perform the method according to any of the foregoing embodiments. In some embodiments, the system disables the use of the VAD if the number of clinical descriptions containing adverse events exceeds a predetermined threshold.

[0010] In yet another embodiment, a system is provided for automatically classifying clinical descriptions of patients, each clinical description being associated with the use of a ventricular assist device in the patient. The system includes a processor configured to perform the method according to any of the foregoing embodiments.

[0011] In a further embodiment, a computer program including computer-executable instructions is provided, which, when executed by a computing device including a processor, cause the computing device to perform the method according to any of the foregoing embodiments. Attached Figure Description

[0012] The above and other objects and advantages will become apparent when considered in conjunction with the accompanying drawings, wherein the same reference numerals always refer to the same parts, and wherein: Figure 1 An illustrative system for finding adverse events in clinical descriptions according to an embodiment of the present invention is shown; Figure 2 An illustrative flowchart of a keyword search method using natural language processing is shown; Figure 3 An illustrative flowchart of a method for identifying adverse events in a clinical description according to embodiments of the present disclosure is shown; Figure 4 It shows the use of Figure 1 The method optimizes the length of the active region around the target word in relation to the number of false positives. Figure 5A and Figure 5B It shows Figure 1 The use of methods in clinical descriptions containing negative words; and Figure 6A and Figure 6B It shows Figure 1 The method is used in clinical descriptions that include terms related to body parts. Detailed Implementation

[0013] To provide a general understanding of the methods and systems described herein, certain illustrative embodiments will be described. Although the embodiments and features described herein are specifically described as combining the use of natural language processing to automatically detect adverse events in clinical descriptions involving the use of ventricular assist devices (VADs), it will be understood that all components and other features outlined below can be combined with each other in any suitable manner and can be adapted and applied to other types of medical treatments with their associated clinical descriptions.

[0014] The systems and methods described herein use Natural Language Processing (NLP) to automatically detect the occurrence of adverse events in clinical descriptions. NLP is used to perform a keyword search within the active region of target words contained in the clinical description. Once a keyword is found, the processor considers the clinical description to be related to (or not, as appropriate, to) the treatment in which an adverse event has occurred. In some embodiments of this disclosure, tags are written into the headings of the clinical description text file. By searching for keywords in the clinical description text, no training of machine learning algorithms (involving training data) is required, thereby freeing up system resources of the computing device. The keyword search nature of the methods and systems of this disclosure does not monopolize the processor of the computing device performing the analysis of the clinical description.

[0015] Figure 1 A block diagram of a system 100 for automatically detecting the occurrence of adverse events in clinical description 110 is shown. System 100 includes a computing device 120, such as a laptop computer, that communicates with a patient data repository 130. For simplicity, Figure 1 Only the processor 125 of the computing device 120 is shown in this disclosure. However, it will be understood that the computing device 120 also includes other components typically associated with computing devices, such as volatile memory (e.g., random access memory RAM), non-volatile memory (e.g., read-only memory ROM), a display, and a connection bus that enables communication between these components, all of which are included in this disclosure.

[0016] The computing device 120 includes a processor 125 capable of executing machine-readable instructions to perform operations on text data using natural language processing. The computing device 120 communicates with a patient data repository 130 that includes patient data obtained from various medical institutions. According to certain embodiments of this disclosure, the patient data repository 130 may include an Acute Myocardial Infarction-Cardiac Shock (AMICS) database compiled and maintained by a CRM such as Salesforce.com, Inc. The AMICS database 130 stores data on treatments from high-risk percutaneous coronary intervention (PCI) patients and patients with cardiogenic shock. The AMICS database 130 may also store VAD 140-specific data that can be used for treatment in a VAD database 135. The VAD database 135 may include operating parameters for each device.

[0017] Patient data includes clinical descriptions 110 stored in the AMICS database 130 following treatment of patients with cardiogenic shock. Such treatments include the use of medical devices to alleviate the patient's condition, such as, for example, a VAD 140. VADs provide ventricular support to patients with cardiogenic shock and may include, but are not limited to, Impella. ® Pumps, extracorporeal membrane oxygenation (ECMO) pumps, balloon pumps, and Swan-Ganz catheters. Impella ® The pump may include Impella 2.5. ® Pump, Impella 5.0 ® Pumps, Impella CP ® Pumps and Impella LD ® All of these pumps are manufactured by Abiomed, Inc., Danvers, MA.

[0018] VAD 140 is connected to controller 150, which enables physician 160 to operate VAD 140 when treating patient 170. Such operation may include navigating the VAD within patient 170 and adjusting the operating parameters of VAD 140 to suit the patient's condition. Operating parameters include, but are not limited to, purge volume, flow rate, and pump speed. According to certain embodiments of this disclosure, controller 150 may include an automated Impella system manufactured by Abiomed, Inc. of Danvers, Massachusetts. ® Controller (AIC).

[0019] Each VAD 140 may include at least one sensor that collects data from the patient 170 when the VAD is used to treat a patient. The patient data is transmitted as a signal to the controller 150. Such data may include, but is not limited to, mean arterial pressure (MAP), left ventricular pressure (LVP), left ventricular end-diastolic pressure (LVEDP), pulmonary artery wedge pressure (PAWP), pulmonary capillary wedge pressure (PCWP), and pulmonary artery occlusion pressure (PAOP). The controller 150 transmits the patient data to an AMICS database 130, which stores data for post-treatment analysis. The AMICS database 130 may also be provided with additional data from the physician 160 (e.g., notes from the treated patient) that can be stored along with the patient data.

[0020] Data from patients and physicians may be stored in repository 130 as clinical descriptions 110. In some embodiments of this disclosure, clinical descriptions 110 may be stored in AMICS database 130 as at least one text file with a *.txt extension. Clinical descriptions 110 may include text in any language (e.g., English) and / or shorthand (e.g., clinical shorthand). An illustrative clinical description is shown in Table 1. The text file may also include header information containing identifying data, such as, for example, the patient's name and the name of the medical institution, patient demographics, date, and time (not shown in Table 1). It will be understood that the above are exemplary embodiments of clinical descriptions, and the term 'clinical description' encompasses any group of machine-readable characters containing information relating to medical procedures performed on a patient, such as cardiovascular treatment using a VAD. Table 1. Descriptive clinical descriptions in *.txt format.

[0021] Clinical description 110 is evaluated by computing device 120 to classify various events that occur during the corresponding treatment. For example, events may include the occurrence of adverse events, malfunction of the treatment device, and success of the treatment. Clinical description 110 can be selected based on a set of specified criteria, such as, for example, geographic region, time period, diagnosis type, patient age, and type of treatment device used (e.g., by Impella). ® (VAD consisting of CP pumps). For example, such criteria can be predetermined or entered by a clinician operating a laptop computer 120 via a graphical user interface (GUI). Each selected clinical description 110 is then analyzed by the processor 125 of the computing device 120 using a software-implemented natural language processing (NLP) algorithm. Examples of NLP software include, but are not limited to, Apache OpenNPL, Mallet, ELIZA, and cTAKES.

[0022] Natural Language Processing (NLP) algorithms determine whether keywords are present in each selected clinical description 110. Keywords may include target words acting on or related to at least one word in the clinical description 110. Keywords can be used to categorize the selection of clinical descriptions 110. According to embodiments of this disclosure, target words may be used to describe adverse events (e.g., bleeding, hemolysis, or ischemia) that have occurred during cardiac treatment in a patient. Examples of target words may include “bleed,” “clot,” and “heart,” and examples of keywords may include “not,” “non,” and “no.” Such keywords and target words may be pre-determined and stored in the memory of computing device 120 for specific types of analysis. Alternatively, keywords and target words may be input by a clinician operating computing device 120 via a GUI of NLP software. The NLP then analyzes the presence of keywords in an active area surrounding the target words to determine whether a specific event occurred during the corresponding treatment. For example, the NLP may identify the occurrence of an adverse event and label the clinical description 110 as containing an adverse event 122 or not containing an adverse event 124.

[0023] Figure 2 A flowchart illustrating a natural language processing method 200 according to an embodiment of the present disclosure is shown. Figure 2 Method 200 in the middle is Figure 1 The processor 125 of the computing device 120 executes the procedure. The method begins in step 210, where the processor 125 of the computing device 120 obtains a selection of clinical descriptions 110 from the AMICS database 130. As previously described, the selection of clinical descriptions 110 can be based on a set of specified criteria, such as, for example, geographic region, time period, diagnosis type, patient age, and type of treatment device used (e.g., VAD). In step 220, the NPL algorithm segments the text of each clinical description 110 into segments or identifiers, referred to as text tokenization / symbolization. Depending on the NPL algorithm used, certain characters in the text (such as punctuation characters) may be ignored. Each identifier serves as a semantic unit for processing the text associated with the selected clinical description.

[0024] After the text is tokenized, method 200 then proceeds to step 230, also known as lemmatization, where similar tokens are grouped together based on their variant forms so they can be analyzed as individual items. Essentially, lemmatization (or stemming) links tokens with the same base form (root word) and groups them together so that the tokens can be processed in a similar manner. For example, in English, the verb "to walk" might appear as "walk," "walked," "walks," and "walking." Here, the base form is "walk," which can be looked up in a dictionary. The output of lemmatization step 230 is a bag-of-words (BOW) containing groups of tokens, each group having an associated base form.

[0025] Once a Word Layout (BOW) is formed for the selected clinical description 110, a keyword search can be performed (step 240). The NPL performs lemmatization on the target word and identifies its root word. Next, the NPL scans the BOW to determine if the root word of the target word appears in the BOW. If such determination is positive, i.e., if the BOW contains a root word that matches the root word of the target word, then the selected clinical description 110 is considered to contain the target word. Conversely, if the determination is negative, i.e., if the BOW does not contain the root word of the target word, then the selected clinical description 110 is considered not to contain the target word.

[0026] Figure 3 A flowchart of a method 300 for automatically classifying clinical description 110 according to an embodiment of the present disclosure is shown. Figure 3 Method 300 in the middle is Figure 1 The processor 125 of the computing device 120 executes the procedure. Similar to method 200, method 300 begins at step 310, where the processor 125 of the computing device 120 obtains a selection of a clinical description 110 from the AMICS database 130. As previously described, the selection of the clinical description 110 can be based on a set of specified criteria, such as, for example, geographic region, time period, diagnosis type, patient age, and type of treatment device used (e.g., VAD). These criteria can be specified by the clinician via the GUI of the computing device 120.

[0027] In step 320, processor 125 uses an NPL algorithm running thereon to determine the location of a target word in each selected clinical description 110. Once the location of the target word is identified, method 300 further uses NPL procedure 200 to identify an active region associated with the target word. The active region includes a predetermined number of words within the text of the selected clinical description 110 that immediately precedes and follows the target word. The active region also includes the target word. The predetermined number of words may be stored within computing device 120 or may be provided as input from a clinician via a GUI. The predetermined number of words defines the size of the active region (i.e., the granularity of method 300) and will be referred to hereinafter as the granularity size.

[0028] Then, method 300 continues to analyze the active regions in each selected clinical description 110. Here, processor 125 uses NPL method 200 to search for keywords in the active regions of each selected clinical description 110. As described above, keywords act on or are related to target words in each clinical description 110. According to embodiments of this disclosure, keywords may include negative words or body part words. Negative words may include, but are not limited to, 'no,' 'not,' 'neither,' 'not,' 'never,' and 'false. The presence of a negative word in the active region of a target word reverses or invalidates the ordinary meaning of the target word. For example, if the clinical description is "...dry groin area with no signs of bleeding...", the presence of the negative word 'none' invalidates the meaning of the target word 'bleeding' appearing in the groin area. Therefore, when the negative word 'no' is detected in the active region, processor 125 marks this clinical description as unrelated to the adverse event, which is groin bleeding.

[0029] In a similar manner, body part terms can include any body part, such as, for example, 'legs', 'arms', 'abdomen', and 'groin'. The presence of a body part term in the active region of a target term invalidates the ordinary meaning of the target term. Unlike negative terms, according to embodiments of this disclosure, the presence of a body part term indicates that no adverse event (e.g., bleeding) has occurred in the heart. For example, if the clinical description is “...the patient is very ill and they feel that her abdomen is bleeding…”, the presence of the body part term 'abdomen' invalidates the meaning of the target term 'bleeding' because it is indeed unrelated to the heart. According to embodiments of this disclosure, it will be assumed that any adverse event in the clinical description of the body part term that does not act on the target term occurs in the patient's heart. Therefore, when the body part term 'abdomen' is detected in the active region, processor 125 marks this clinical description as unrelated to the adverse event of cardiac bleeding. It will be understood that the NPL keyword search can be further customized to match words in the active region with specific body part terms (e.g., 'heart') and to mark the clinical description based on the match.

[0030] Return to reference Figure 3 Once the location of the target word in the text of clinical description 110 is determined, method 300 proceeds to step 330, where it is further determined whether a negative word exists in the active region. If a negative word exists in the active region ('yes' in step 330), the clinical description is marked as not containing an adverse event. In some embodiments of this disclosure, the mark may be written into the title of each clinical description's text file (e.g., in ASCII characters), and the marked clinical description 110 may be written back to the AMICS database via computing device 120.

[0031] If no negative word is present in the active region ('No' in step 330), method 300 proceeds to step 340, where it is further determined whether the body part word exists in the active region. If the body part word exists ('Yes' in step 340), the clinical description is marked as not containing an adverse event, as in step 350. If no body part word exists ('No' in step 340), the clinical description is marked as containing an adverse event, as in step 360. In some embodiments of this disclosure, the markings for both 'Yes' and 'No' cases in step 340 can be stored in the title of a text file for each clinical description, and the marked clinical description 110 can be written back to the AMICS database by the computing device 120.

[0032] This disclosed strategic keyword search scans the text within each clinical description 110 to locate target words and any specified keywords within the active area of ​​the target word. Once keywords are identified, the clinical description is tagged and analysis moves to the next text file. Compared to NPL using machine learning, this does not burden the processor and thus frees up the processing power of the computing device.

[0033] Figure 4 An optimized diagram 400 of the NPL algorithm based on the size of the active region for detecting negative words relative to the target word, according to an embodiment of this disclosure, is shown. In effect, this optimizes the granularity of method 300. As previously stated, the number of false positives (FP) refers to the number of times processor 125 fails to detect the presence of a negative word or body part word in the active region when the negative word or body part word is actually present, while true positives (TP) refer to the negative word or body part word being present and detected. In fact, according to clinical description 110, FP is a missed detection because processor 125 misses the detection of a bleeding event in the patient's heart by incorrectly identifying negative words or body part words in the active region of the clinical description. Similarly, the number of false negatives (FN) refers to the number of times processor 125 incorrectly detects the presence of a negative word or body part word when it is not actually present, while true negatives (TN) refer to the absence or non-detection of a negative word or body part word.

[0034] exist Figure 4 In this context, the size of the active region is considered relative to the number of words immediately preceding and following the target word. Optimization is performed relative to the number of false positives. Curve 410 in optimization chart 400 illustrates the occurrence of false positives, showing the minimum change in the active region with a granularity of three or more words. Furthermore, Table 2 shows exemplary values ​​for TP, FP, FN, and TN as the granularity of the active region changes. The values ​​in Table 2 reinforce... Figure 4 The trend shown indicates that the number of missed detections (i.e., FPs) decreases as the size of the active region increases. The rate of reduction in FPs stabilizes after a critical granularity in the active region. According to embodiments of this specification, the critical size of the active region is considered to be three words. Activity Area predict TP FP FN TN 0 73 39 34 1 346 1 66 39 27 1 353 2 63 39 24 1 356 3 61 39 22 1 358 4 60 39 21 1 359 5 59 39 20 1 360 Table 2. Optimization of FP with active area size.

[0035] Figure 5A and Figure 5B An example is shown of using the NPL algorithm to automatically detect the presence of negative words in clinical descriptions 500 and 550 based on the aforementioned system and method. Figure 5A In this context, target word 510 is 'bleeding', and the granularity of active region 520 is three. Keyword 530 is the negative word 'no'. When the negative word 530 appears within the active region 520 of target word 510, the clinical description 500 is marked as not containing adverse events. Similarly, in Figure 5B In this context, target word 560 is 'bleeding', and the granularity of active region 570 is three. Keyword 580 is the negative word 'no'. When the negative word 580 appears within the active region 570 of target word 560, the clinical description 550 is marked as not containing an adverse event. In both examples above, according to... Figure 2 The NPL processing performed by the method shown is used to identify target words, negation words, and words in the active region.

[0036] Figure 6A and Figure 6B An example is shown of automatically detecting the presence of body part terms in clinical descriptions 600 and 650 using the NPL algorithm according to the system and method described above. In Figure 6, the target term 610 is 'bleeding', and the granularity of the active region 620 is three. The keyword 630 is the body part term 'abdomen'. When the body part term 630 appears within the active region 620 of the target term 610, the clinical description 600 is marked as not containing an adverse event. As previously stated, it is assumed that no adverse event in the clinical description of the target term occurs in the heart of the patient without the body part term acting on it. Therefore, when the body part term 'abdomen' is detected in the active region 620, the processor 125 marks the clinical description 600 as unrelated to the adverse event of cardiac bleeding. Similarly, in Figure 6B In this example, target term 660 is 'bleeding', and the granularity of active region 670 is three. Keyword 680 is the body part term 'groin'. When body part term 680 appears within active region 670 of target term 660, clinical description 650 is marked as not containing adverse events. In both examples above, according to... Figure 2 The NPL processing of method 200 shown is used to identify target words, body part words, and words in active areas.

[0037] Table 3 presents a confusion matrix, providing example figures to illustrate the effectiveness of the methods and systems of this disclosure. Table 3 compares the results of performing a full keyword search using NPL for each word in the clinical description with a strategic negative word and / or body part keyword search using NPL to search only negative words and / or body part keywords in the active region of the target word according to embodiments of this disclosure. Metrics associated with the confusion matrix include precision, recall, and accuracy. Precision is determined using the following formulas: TP / (TP+FP); recall is determined using the following formula: TP / (TP+FN); and accuracy is determined using the following formula: (TP+TN) / (TP+FP+FN+TN). For the illustrative confusion matrix in Table 3, the precision, recall, and accuracy for the full keyword search are 53.4%, 97.5%, and 91.7%, respectively, while the same metrics for the strategic negative and / or body part keyword search are 74.0%, 92.5%, and 96.2%, respectively. These comparative metrics demonstrate that strategic negation and / or body part keyword search identifies keywords with greater precision and accuracy. These figures indicate that the strategic negation and / or body part keyword search of this disclosure outperforms traditional NPL techniques while minimizing system resource usage for performing this type of natural language processing.

[0038] As previously described, after tagging each selected clinical description 110, the tag can be stored in the header of each clinical description text file, and the text file can be written back to the AMICS database 130 by the computing device 120. For this purpose, clinicians can be able to filter tagged clinical descriptions from the AMICS database based on certain criteria (e.g., VAD type, patient age, medical institution name) to obtain the percentage of clinical descriptions containing the target term. For example, a clinician operating the computing device 120 can be able to obtain from the AMICS database the Impella used on male patients aged 50-55 years in Boston, Massachusetts. ®2.5 Data related to heart pumps in male patients who experienced cardiac bleeding during cardiac surgery. If such data is lower than the statistics shown for eligible patients, it may indicate various problems. These problems may include, for example, an incorrect cardiac surgical procedure performed on the patient, or a malfunction in the VAD used that requires correction. In cases of incorrect cardiac surgical procedure, such data may initiate further training at the medical facility. In cases of suspected malfunction in the VAD, such data can be used for quality control during the manufacture of such devices. Furthermore, if the number of labeled clinical descriptions exceeds a predetermined threshold, data obtained from labeled clinical descriptions containing adverse events can be used to trigger a lockout mechanism to warn physicians against the use of the VAD. Therefore, automatically labeling clinical descriptions, as described above, will enable clinicians to receive feedback that can improve the treatment provided to patients. Table 3. Comparison of confusion matrices.

[0039] The foregoing is merely an illustration of the principles of this disclosure, and the device can be practiced in ways other than those described. These embodiments are presented for illustrative purposes and not for limitation. It should be understood that while the methods disclosed herein are shown for use in automated ventricular assist systems, they can be applied to systems that will be used in other automated medical systems.

[0040] Variations and modifications will occur to those skilled in the art upon reading this disclosure. The disclosed features can be implemented in any combination and sub-combination (including multiple dependent combinations and sub-combinations) with one or more other features described herein. The various features described or illustrated above (including any components thereof) can be combined or integrated into other systems. Furthermore, certain features may be omitted or not implemented.

[0041] Examples of alterations, substitutions, and modifications are those that can be determined by a person skilled in the art and can be made without departing from the scope of the information disclosed herein. All prior art referenced herein is incorporated in its entirety by reference and forms part of this application.

Claims

1. A method for automatically classifying a patient's clinical description, the method comprising: One or more processors receive at least one clinical description file from a data repository comprising multiple clinical description files, wherein each clinical description file includes text relating to the use of a ventricular assist device (VAD) for the corresponding patient; The one or more processors are used to determine the position of the target word within the text of the at least one clinical description document; The one or more processors determine whether at least one of a plurality of predetermined negative words is located within an active region, wherein the active region includes the target word and a predetermined number of words appearing in the text immediately before and after the target word; The one or more processors determine whether at least one of a plurality of predetermined body part terms is located within the activity area; and In response to determining that the active area does not contain at least one of the plurality of predetermined negative words or at least one of the plurality of predetermined body part words, a marker is written into the title of the at least one clinical description document by the one or more processors, wherein the marker indicates that the at least one clinical description document contains an adverse event.

2. The method according to claim 1, wherein determining the position of the target word comprises: Process the text of the at least one clinical description file to generate word identifiers; Identify and group word identifiers, including inflections of words; and Perform a keyword search on the text using grouped word identifiers.

3. The method according to claim 1, wherein the target word is "bleeding", "clot" or "heart".

4. The method according to claim 1, wherein the plurality of predetermined negative words are selected from "not", "not", "not either", "non", "never", or "false".

5. The method of claim 1, wherein the plurality of predetermined body parts are selected from "leg", "arm", "abdomen" or "groin".

6. The method according to claim 1, further comprising: In response to determining that the number of clinical description files containing adverse events exceeds a predetermined threshold, at least one VAD is disabled using the one or more processors.

7. The method according to claim 1, further comprising: In response to determining that the number of clinical description files containing adverse events exceeds a predetermined threshold, further training is initiated at the medical facility using one or more processors.

8. The method according to claim 1, further comprising: In response to determining that the number of clinical description files containing adverse events exceeds a predetermined threshold, a manufacturing quality control process is initiated using one or more processors.

9. A system for automatically classifying a patient's clinical description, the system comprising: The data repository includes multiple clinical description files, each containing text related to the use of a ventricular assist device (VAD) for the corresponding patient; and One or more processors that communicate with the data repository and are configured to: Receive at least one clinical description file; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined negative words is located within an active region, wherein the active region includes the target word and a predetermined number of words appearing in the text immediately before and after the target word; Determine whether at least one of a plurality of predetermined body part terms is located within the activity area; and In response to determining that the active area does not contain at least one of the plurality of predetermined negative words or at least one of the plurality of predetermined body part words, a marker is written into the header of the at least one clinical description document, wherein the marker indicates that the at least one clinical description document contains an adverse event.

10. The system of claim 9, further comprising: At least one VAD used to treat the patient; and The controller that communicates with the VAD and the data repository. The controller is configured to transfer patient data from the at least one VAD to the data repository, and The data repository is configured to store clinical description files, which include patient data from the at least one VAD.

11. A non-transitory computer-readable storage medium having instructions stored thereon, the instructions causing the one or more processors, when executed by one or more processors, to: Receive at least one clinical description file from a data repository that includes multiple clinical description files, wherein each clinical description file includes text relating to the use of a ventricular assist device (VAD) for the corresponding patient; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined negative words is located within an active region, wherein the active region includes the target word and a predetermined number of words appearing in the text immediately before and after the target word; Determine whether at least one of a plurality of predetermined body part terms is located within the activity area; and In response to determining that the active area does not contain at least one of the plurality of predetermined negative words or at least one of the plurality of predetermined body part words, a marker is written into the title of the at least one clinical description document, wherein the marker indicates that the at least one clinical description document contains an adverse event.

12. A method for automatically classifying a patient's clinical description, the method comprising: One or more processors receive at least one clinical description file from a data repository comprising multiple clinical description files, wherein each clinical description file includes text relating to the use of a medical device on the corresponding patient. The one or more processors are used to determine the position of the target word within the text of the at least one clinical description document; The one or more processors determine whether at least one of a plurality of predetermined body part words is located within an active region, wherein the active region includes the target word and a predetermined number of words appearing in the text immediately before and after the target word; and In response to determining that the active area does not contain at least one of the plurality of predetermined body part terms, the one or more processors write the tag into the title of the at least one clinical description document.

13. The method of claim 12, further comprising: The one or more processors determine whether at least one of a plurality of predetermined negative words is located within the activity area. In response to (a) determining that the active area does not contain at least one of the plurality of predetermined negative words and (b) determining that the active area does not contain at least one of the plurality of predetermined body part words, the marker is written into the title of the clinical description document.

14. The method of claim 12, wherein the medical device is a ventricular assist device (VAD).

15. The method of claim 12, wherein the marker indicates that the at least one clinical description file contains an adverse event.

16. A system for automatically classifying a patient's clinical description, the system comprising one or more processors configured to: Receive at least one clinical description file from a data repository, the at least one clinical description file comprising text relating to the use of a medical device on a patient; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined body part terms is located within an active region, wherein the active region includes the target term and a predetermined number of words appearing in the text immediately before and after the target term; and In response to determining that the active area does not contain at least one of the plurality of predetermined body part terms, a marker is written into the title of the at least one clinical description document.

17. The system of claim 16, further comprising the medical device.

18. The system of claim 16, further comprising the data repository, wherein the data repository comprises a plurality of clinical description files, and wherein each clinical description file comprises text relating to the use of a medical device on a corresponding patient.

19. A non-transitory computer-readable storage medium having instructions stored thereon, the instructions causing the one or more processors, when executed, to: Receive at least one clinical description file from a data repository, the at least one clinical description file comprising text relating to the use of a medical device on a patient; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined body part terms is located within an active region, wherein the active region includes the target term and a predetermined number of words appearing in the text immediately before and after the target term; and In response to determining that the active area does not contain at least one of the plurality of predetermined body part terms, a marker is written into the title of the at least one clinical description document.

20. A method for automatically classifying a patient's clinical description, the method comprising: One or more processors receive at least one clinical description file from a data repository comprising multiple clinical description files, wherein each clinical description file includes text relating to the use of a medical device on the corresponding patient. The one or more processors are used to determine the position of the target word within the text of the at least one clinical description document; The one or more processors determine whether at least one of a plurality of predetermined body part words is located within an active region, wherein the active region includes the target word and a predetermined number of words appearing in the text immediately before and after the target word; and In response to determining that the active area contains at least one of the plurality of predetermined body part terms, a marker is written into the header of the at least one clinical description file by the one or more processors, wherein the marker indicates that the medical device has been successfully used to treat the corresponding patient.

21. A system for automatically classifying a patient's clinical description, the system comprising one or more processors configured to: Receive at least one clinical description file from a data repository, the at least one clinical description file comprising text relating to the use of a medical device on a corresponding pair of patients; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined body part terms is located within an active region, wherein the active region includes the target term and a predetermined number of words appearing in the text immediately before and after the target term; and In response to determining that the active area contains at least one of the plurality of predetermined body part terms, a marker is written into the title of the clinical description document, wherein the marker indicates that the medical device has been successfully used to treat the corresponding patient.

22. A non-transitory computer-readable storage medium having instructions stored thereon, the instructions causing the one or more processors, when executed, to: Receive at least one clinical description file from a data repository, the at least one clinical description file comprising text relating to the use of a medical device on a corresponding pair of patients; Determine the location of the target word within the text of at least one clinical description document; Determine whether at least one of a plurality of predetermined body part terms is located within an active region, wherein the active region includes the target term and a predetermined number of words appearing in the text immediately before and after the target term; and In response to determining that the active area contains at least one of the plurality of predetermined body part terms, a marker is written into the title of the clinical description document, wherein the marker indicates that the medical device has been successfully used to treat the corresponding patient.

Images