US20020012899A1

US20020012899A1 - Computer-automated implementation of user-definable decision rules for medical diagnostic or screening interpretations

Info

Publication number: US20020012899A1
Application number: US09/044,487
Authority: US
Inventors: Terese Finitzo; Kenneth D. Pool
Original assignee: Individual
Current assignee: Optimization Zorn Corp
Priority date: 1997-03-20
Filing date: 1998-03-19
Publication date: 2002-01-31
Anticipated expiration: 2018-03-19
Also published as: US7228315B2; US6394811B2; US20020081558A1

Abstract

Disclosed is a method and a technique for allowing a nonprofessional to utilize professional-tailored software to carry out an analysis without the professional being present. In the preferred embodiment, a decision tree is provided having multiple levels of analysis, where each level includes plural rules for optional use. Each rule is selectable for use by the professional, and if selected, the rule has provisions for allowing the professional to assign a numerical value for comparison with the client data. In the preferred embodiment, the professional can enter into the levels of the decision tree how many of the selected rules must pass or fail before the level is considered as a pass or fail condition. Also, provisions are made in the software to provide a “refer” output, where the test results are indeterminate and the patient must be tested again while yet present at the test facility.

A nonprofessional can attend to all the patent testing to obtain medical data from the patient—in the form of digital data. The nonprofessional can then apply the digital data test results of the patient to the software decision tree. Since the rules of the software decision tree were tailored by the professional, the output of the software represents a professional analysis that was obtained by a nonprofessional.

Description

This application claims the benefit of prior pending application Ser. No. 60/041,420 filed Mar. 20, 1997, the entire disclosure of which is incorporated herein by reference.[0001]

The purpose of EVA is to leverage the expertise of the professional, be it a physician, audiologist, therapist or scientist to use and leverage their expertise to provide an interpretation for a technician level person to complete particular tasks. For example, EVA is used to provide an automatic interpretation of an auditory brain stem response to evaluate the neurologic status of a patient. EVA can provide an interpretation for an otoacoustic emission to let a technician know whether a particular test conducted on an infant or an adult was a good quality test and what the result was. At this point EVA can be used certainly for specific interpretations of auditory function, but can also be used, for example, in an interpretation of sonograms, of electrocardiograms, of any kind of diagnostic assessment or screening assessment that requires a professional to interpret the test and provide the result. The beauty of EVA it that is allows the individual professional to utilize their own criteria as to what is considered a pass or a fail, or what is a test in which they would want additional information. It is the professional, the physician again, the professional's rules that are applied in EVA. If, for example, there were two different laboratories with two different physicians who had two different views of what was concerning in a sonogram on an infant, each of those physicians could utilize EVA to put in his own rules so that his own rules would be the ones that applied in the case of their interpretation for a patient. In other words, it is the ever vigilant backup to reduce errors, to be right there—using the physician's expertise at the time the test is done without needing the professional to be present. And again, if two different physicians in two different laboratories in two different places in the country had different interpretive criteria they can take the technology and apply their own rules so that the laboratory presents their view. This is what EVA embodies.

The input to EVA can be a number of different measures, sonograms, waveforms, it could be waveform morphology interpretation. It could be computerized or electronically generated. It could be specific numbers that are taken from a computer or from a computer generated test. EVA would read the data file, read the absolute values in the data files and extract those values and then look for features that would produce the rules. If one is monitoring brain function, EVA would pull out the amplitude criteria of the waveform that was generated from the brain; EVA would pull out frequency characteristics. EVA would extract information from a wave emission, picked up from an amplifier in the ear canal and it would evaluate features from that emission. Physiologic data from different systems could be assessed. EVA would take the representation and then pull out the key pieces that the professional chose to call important. And that is the beauty of it. It is not an interpretation, but it is the expert's view of the test procedure, to leverage his expertise, to allow him or her to do more.

In FIG. 1, the first focus is to establish the features that will be used in defining the rules for evaluating a particular test, diagnostic procedure, or screening measure. In [0004] Block 1, the technology specific electronic representation is the first feature that needs to be considered. This might be the specific embodiment of the IL088, transient otoacoustic emission technology by (OAE) Otodynamics London. The electronic representation might be waveform analysis or might be a waveform drawn from a particular company's EEG equipment. It might take the Grason Stadler GSI60 distortion product otoacoustic emission waveform manufactured by Grason Stadler, and apply the same kinds of rules to that. It could take Biologics distortion product equipment manufactured by Bio-logic in Mundelein, Ill., and apply a professional's criteria to that electrodiagnostic equipment as well. EVA might take an EKG and pull out data from there. The data is taken from the technology (Block 1) and incorporated into a PC based computer running under Windows. EVA is written in Borland's C++ programming language at this time.
For example, when an OAE test is completed, a probe is placed in an infant's ear and a stimulus is presented to the baby's ear (a sound, a click) and the amplifier, also attached in the probe assembly, picks up the emission from the cochlea of the infant ear, and feeds that back into the computer into the technology specific application where features are displayed. Among the features are the stimulus level, the noise level, the number of quiet and noisy stimuli presented, the background noise level, the amplitude of the various components in the emission, the time of the testing, the frequency of the various components, etc. [0005]
These data then reside in the technology specific program completed that has just been run. Based on [0006] Block 2 of FIG. 1 entitled Establishing Features, there is an extraction or classification of specific data elements as features. This is the process. The data elements are now features selected by the professional and based on his/her judgment. A data element might be the signal to noise ratio of the otoacoustic emission at 1600 hz. Another data element might be the number of quiet sweeps that was presented. All these are elements that could be used in an analysis of the particular test. A data element is a potential feature. There are data elements that might not be used as features by a particular professional because he or she does not consider them to be critical to analysis. Another professional might use that particular data element. The various features that the professional determines are then named to develop rules for rule development. This is shown in Block 3 and it is the rule development of EVA that is based on the professional's decisions.
The output of EVA, after the professional has made his decisions on rule development and they have been implemented via the C++ programming language, the output is a decision, a decision about the test that was completed on a particular patient. That decision could be that the test was “technically inadequate” and needed to be redone. It could be that the test was a “pass”, it could be that the test was a “fail”, and that additional testing was needed. It might be that in a particular embodiment, if you looked at cardiac function, it might be that no additional testing was needed based on the results of the decision tree rules that were implemented by EVA. A professional might decide to use a morphologic analysis of a waveform to determine if a response was present in that waveform. This might produce the decision that a patient had abnormal brain stem function. That could be an additional implementation of a rule by EVA. [0007]
In FIG. 2, under User Rule Selection, is a series of Blocks. The first block is [0008] Block 4 to determine classification levels. The embodiment of Block 4 in FIG. 2 is shown in FIG. 5. Set interpretive assistant parameters for the analysis decision tree. It is the analysis decision tree in which the various classification levels are determined. The classification levels are a technical fail level, a pass level, and another technical fail level, and the issues are whether the criteria are met, whether the criteria are sufficient for a pass, necessary for a pass or contributing to a pass. The issue with a tech fail is whether the technical criteria for “refer” are met or not. Note in FIG. 6 the technical minimums for analysis. Note in FIG. 7 there is the sufficient criteria for pass, and that again is the specific embodiment for the ILO Otodynamics Transient Evoked Otoacoustic Emission. In FIG. 8 is the necessary criteria for pass, and in FIG. 9 there is the contributing criteria for pass, necessary but contributing. In FIG. 10 there is shown the technical minimums for a refer result. Refer in this case to send the patient for additional evaluation. Notice the right hand corner of FIG. 5. The directions to the professional are shown telling him or her how to program, really how to sub-program EVA and make the decision tree work for that professional. Again, it is the professional's judgments that are being implemented, not generic decisions. It is the professional who is doing the testing, who is responsible for the testing and who is responsible for the interpretation. The rationale behind the EVA is to again leverage the expertise of the professional to allow that person to conduct more testing in a high quality and cost efficient manner and to improve patient care through automation.
In [0009] Block 4 Determine Classification Levels—that has now been done. The professional would begin at the first level (Block 5). The first level is the first evaluative criteria that the professional has determined will be important in his decision making criteria. This is very higherarchical interpretive assistant. The first level in the specific embodiment is the technical minimum requirement for a test to be analyzed to evaluate them.
The technical minimums for analysis is shown in FIG. 6. Note that the professional makes judgments and the judgments are related to various criteria that are pulled out and extracted from the particular technology in question. It could be the IL088, it could be the GSI60 Distortion Product, it could be an auditory brain stem response by Bio-logic, it could be Nicolet auditory brainstem response equipment. It could be EMG equipment produced by Nicolet or Bio-logic. In this case the implementation is the IL088. The first decision that the professional needs to make is: what is the minimum number of quiet sweeps that have to have been collected for a test to be analyzed further. The professional can make that decision and increase or decrease the value. In this case the minimum number of quiet sweeps was set at 60 as the standard. If this were a auditory brain stem response, the minimum number of quiet sweeps might be set at 1000 or 2000 sweeps. The technical minimums for analysis allows the professional to assess other features: the maximum number of quiet sweeps, a percentage of quiet sweeps, minimum and maximum, the lowest stimulus intensity that would be allowed as well as the highest stimulus intensity that would be allowed. Note that the second selected technical minimum for analysis is that the maximum peak stimulus cannot exceed 85 dB. Only two values in this level were selected: a minimum number of quiet sweeps and a maximum peak stimulus. The user (or the professional) involved had many other choices to make. This then is [0010] Block 6 in FIG. 2 “Get User Rule Values” for this level. The rule values would be the minimum number of quiet sweeps of 60 and the maximum peak stimulus of 85 dB.
The next Block is [0011] Block 7 in FIG. 2: is there another level to examine? If no, save these rules and go on. If yes, get the user rule values for this level. Referring to the next level in the embodiment. In the interpretive assistant parameters in FIG. 5, the next level is to detail the sufficient criteria necessary for a test to be a pass for the decision of the professional to make a test, a pass. This is seen in FIG. 7. It sets the sufficient criteria for pass. The rules that the professional would make in this case: what is the whole wave response in dB. What is the whole wave correlation? What is the net response at 800 hz, at 1600 hz, at 2400 hz, at 3200 hz and at 4000 hz? Basically, the professional would define his necessary results for the test to be a pass. Those then would be the second level that would need to be evaluated by EVA. (FIG. 7)
The question in User Rule Selection: is there another level? If no, EVA saves the rules so far, and then makes the decision based on what has been done. If there is another level, go back and get the rules for this level. In this case, there is a necessary criteria for a pass. What is sufficient, what is necessary? The necessary criteria that this professional has made are in FIG. 8. There are seven necessary criteria that have been delineated and two that are absolutely required. That is, the whole wave correlation between, in this case, the IL088 two waveforms that are being collected, must be fifty percent or better. Note also that the signal to noise ratio, ([0012] net response 4000 hz), the signal to noise ratio has to be at least 6 dB. That is a requirement for the test to be considered a pass. The question again must be asked: is there another level? If not, then EVA saves these user rules; if yes, EVA goes back down to the next level.
The next level is “contributing criteria” for the test to be considered a pass. Again, in the specific embodiment of the IL088 Quick Screen Transient Otoacoustic Emission test (quick screen testing mode FIG. 9). There are basically another seven criteria that are contributing to a pass. In this case, the professional involved has selected three that contribute. Note that the net response (again, this is signal to noise ratio) at 1600 hz has to be at least 3 dB, the net response at 2400 hz has to be at least 3 dB, and the net response at 3200 hz has to be at least 6 dB. Note also that in terms of contributing criteria, two of those three are required for this to be contributing criteria. Is there another level? If no, save the rules, if yes, go back and look again. In this case, there is one more level to look in FIG. 10. [0013]
FIG. 10 is the technical minimum criteria for a professional to decide the test is a refer or a fail. Refer again means to refer a patient on for additional assessment. Note that there are eight criteria that are being examined, or that the user or professional can decide on—the minimum number of quiet sweeps, the maximum number of quiet sweeps, etc. Note that in this case the professional has made two decisions, that the minimum stimulus intensity (the trough stimulus) has got to be no lower than 70 dB, and that the minimum trough percent stimulus stability has to be 60 percent. Either of these two criteria are enough to classify this test as a technical fail. In other words, if the overall stimulus is less than 70 dB, the test is not a fail or a refer, it becomes a technical fail, or a technically inadequate test. Similarly, if the stimulus stability (and that refers specifically to whether the probe in the baby's ear is in there firmly, is not at least 60, then the test result is interpreted as a technically inadequate test (or a tech fail) rather than a true fail. What that EVA offers the screener, or the tester, or the technician at the bedside is immediate feedback, that “do it over”, that it is not a technically acceptable test. This “do it over—it is not technically acceptable” will actually improve patient care because the technician has a chance to correct an error before the patient has left, before money has been spent, and a redo of something that was done poorly the first time is necessary. This specific piece of information allows the screener to self correct at a point when it can still impact the quality of care for the patient. [0014]
Is there another level? If not, save these rules ([0015] Block 8 of FIG. 2). The user has made his selections on his rules, on the decision criteria. The user can be sitting in his office, setting up EVA to implement his decisions. Once the user is done with this step, it is these decisions that will be implemented in the nursery, the laboratory, to wherever the test is going to be performed by whoever is going to perform the test. When the technician begins to do a real test on a real patient, these are the rules that will be implemented, and that the technician will automatically follow, and EVA will provide the technician with immediate feedback as to what the professional wanted. Even if a test is done at 2:00 in the morning when the professional is not present, his or her criteria will be implemented consistently and repeatedly. Screener fatigue will not enter, errors won't happen because of inadvertent oversight. The professional's rules are applied day in and day out.
Following the actual testing, whether diagnostic test or screening, the results are displayed for the screener to know what the result was. If the results are a pass and a refer, a report is immediately generated and this report contains, again, the words of the professional. In FIGS. 13, 14 and [0016] 15 the professional chooses the language. This is another one of the levels that the professional will set up: report generation. Automatically, that report will be printed, printed at 2:00 in the morning immediately following the test. The test results will be automatically extracted from the technology specific electronic representation and that information will be put into the report. So once again the screener cannot make an error and print or insert an incorrect letter into a patient record. EVA goes in, and looks at the test results, makes the decision in real time, takes that result and places it in the letter that will be generated promptly. One more level of potential error in medical care is eliminated by the automatic decision and application of the decision and generation of reports.
FIG. 3 is the application of the rules. [0017] Block 9 in FIG. 3 is “data submitted”. This refers to the extraction of data from the electronic technology or electronic representation. Data is submitted (Block 10) and the user rules are retrieved for each level (Block 11). For each rule in a level (Block 12) the question is asked, does the data satisfy the rule (Block 13)? Review FIG. 6 and note that there are technical minimums for analysis. The minimum number of quiet sweeps 60. The maximum peak stimulus is less than 85 dB. At this level then, there are two rules in the level. The data is examined by Eva to see if it satisfies each of these two rules. “Does the data satisfy the rule” is the question posed in Block 13. In FIG. 3 (Block 14), EVA “asks” is another rule required? Following a test is completion, the technician retrieves the EVA test result to determine the next steps for a particular patient.
In this example, there are other levels. What are the sufficient criteria for a pass is shown in FIG. 7. What are the necessary criteria for a pass in shown in FIG. 8. We note that the whole wave correlation has to be 50% in FIG. 8. Note that at 4,000 Hz, the signal of the emission has to exceed the noise level of the emission by 6 dB. That is another required rule. If there was no, not another rule required, the next question is: another level required? Whole wave correlation has to be 50%. That's one rule. Is another rule required, yes. The net response has to be a 6 dB signal to noise ratio. Within a level, there are those two rules in this case. [0018]
The next question is, is there another level required? In this case, there are the contributing criteria for a pass. The answer is yes, another level is required. This is [0019] Block 15. There are contributing criteria for pass and there are three rules in this level. There are three rules that the professional has set for contributing criteria (FIG. 9). Only two of those criteria or two of those rules have to be implemented or have to be required for the contributing criteria to be met. If a net response or a signal to noise ratio at 1600 Hz of 3 dB is met, the 3 dB criteria at 2400 Hz need not be met if the 3200 Hz, 6 dB signal to noise criteria is met. Note in FIG. 9, number of criteria required is 2.
Another level required? (Block [0020] 15) This is FIG. 10 and there are two tech minimums marked or decided upon by the professional in this case. One is the minimum trough stimulus. The other is the percent stimulus stability. Either of those is sufficient to classify the result as the tech fail. Note that it says the minimum of above criterion required to classify as a tech fail is one. The professional establishing the rules could have changed that number to two if he/she has felt it was correct/needed.
At this point at [0021] Block 16 in FIG. 3 is the decision. All levels and all rules in each level have been assessed. No more rules are required and no more levels are available to evaluate. EVA returns the classification of the test results based on the professional's interpretive criteria. The classification is the result. From that result, the report is generated to the referring physician or referring clinician. Classification is pass, fail or refer, or a technically inadequate test. Classification might be refer for additional diagnostic cardiac testing. Classification might be do another test. Classification can be basically whatever the professional decides is the next step based on the rules that he/she is generating. FIG. 12 is a handprinted copy of a results report. FIGS. 13, 14 and 15 are example reports. FIG. 13 is a report to a physician showing that the baby in this case born on Mar. 14, 1997 did not pass the test in the right ear or the left ear, that is, the baby “referred” in both ears. The words in the letter are the words that the professional determines to be acceptable to him. They can be modified by another professional. FIG. 14 is a letter to a physical and the results of the letter are a pass in both the right ear and the left ear on this infant. FIG. 15 shows a test result of pass in the right ear and refer in the left ear. Again, the words are generated by the clinician of record. Note there is no letter for a technically inadequate test although there are words that can be generated. In general, technically inadequate results are required to be done again in this example.
The specific embodiment in FIG. 4 is of the three elements. The first element is that the values saved by the IL088 Otoacoustic emissions equipment manufactured by Otoacoustic Dynamics are read from the ILO generated data file. Those elements are read and then they are named. Specific values are named for each level and for each rule in a level. For the second element, the user selects the values, the ranges and the features. Finally, the data are retrieved from the ILO data file and the macro, the software or Eva retrieves the user rules and applies them to data for classification. Note then that the code is the implementation of [0022] element 3. The code can be referred to as FIG. 11 and it is really the specific embodiment of the third element which is that the macro retrieves the user rules and applies them to classification.
It is through this that patient care is improved because feedback is provided to the technician immediately. The technician can learn immediately from his/her errors and alter the outcome for the patient before they have left and eliminate an inadequate costly test session. [0023]
In summary, this software system works for the caregiver at the bedside by reducing errors. It leverages the costly professional's time and energy. It facilitates prompt and accurate patient care through rule governed decision analysis. When widely implemented it will reduce health costs. [0024]

Claims

What is claimed is:

1. A method for evaluating medical information, comprising the steps of:

processing medical information to select various features therefrom, as specified by a doctor;

developing rules for the evaluation of the medical information based on the features extracted therefrom;

developing classification rules and criteria as input by the doctor for a determination of a pass, fail and refer of a decision based on the selected features;

presenting displays for the doctor so that various criteria is displayed so that the doctor can make judgments thereon and provide inputs as to said judgments; and

extracting test results and generating a report.