Apparatus and method for categorizing health care utilization5486999Abstract An apparatus and method for categorizing health care utilization provides an efficient aid in identifying patients who are seeking inappropriate care. The invention involves a computer system having a neural network responsive to several input variables to categorize the utilization characteristics of the patient. The input variables define selected characteristics of a patient. In one embodiment, a screening process identifies patients who are at high risk to an immediate threat to their health and eliminates those least likely to be seeking inappropriate care. Claims What is claimed is: Description Microfiche Appendix: Appendix B is a microfiche appendix of six microfiche and 298 frames.
TABLE 1
__________________________________________________________________________
INPUT VECTOR VARIABLES
Variable
Number
Variable Name Range Format.sup.1
__________________________________________________________________________
1 Anxiety 0-5 d
2 Depression 0-5 d
3 Somatization 0-5 d
4 Total ADS 0-15 dd
5 Age 18-100+ ddd
6 Sex M = 0, F = 1
b
7 Gross Charges Total 0-99999.sup.2
ddddd
8 Total Count 0-9999 dddd
9 Gross Charges Out-Patient
0-99999.sup.2
ddddd
10 Out-Patient Count 0-9999 dddd
11 Gross Charges DOV 0-99999.sup.2
ddddd
12 DOV Count 0-9999 dddd
13 Gross Charges Non-DOV 0-99999.sup.2
ddddd
14 Non-DOV Count 0-9999 dddd
15 Number of different PCP's seen
0-99 dd
16 Number of different specialists seen
0-99 dd
17 Number of different Out-Patient Diagnoses
0-999 ddd
18 Number of in-patient or out-patient diagnostic ranges
0-19 dd
19 Patient Driven Definite yes = 1; no = 0
b
20 Illness Driven yes = 1; no = 0
b
21 Patient Driven Possible yes = 1; no = 0
b
__________________________________________________________________________
.sup.1 "d" denotes one decimal digit (0-9); "b" denotes one binary digit
(0-1)
.sup.2 hundreds of dollars, rounded to the nearest $100.00
Table 1 also provides information on the organization of each patient input vector data file. As shown in Table 1, each input variable is assigned by a variable number. Each input variable also has a range and a data format. The format column describes the representative format for the variable in the input vector data files. For example, the input variable 1, "Anxiety," may have a value ranging from -5 (see "Range" column), whereby "0" indicates no anxiety and "5" indicates extreme anxiety. This value is represented in an input vector data file by a single decimal digit ranging from 0-5. This is illustrated in Table 1 by the single "d" reference in the "Format" column. In the present embodiment, twenty-one variables have been selected for the input vectors for the neural network 20 for categorization purposes. In the present embodiment, each of the input vector variables is scaled to a value ranging from "-1" to "1." Alternatively, the input vector variables are scaled to values ranging from "0" to "1" or "-1" to "0." For instance, in the embodiment where the scaling is for the range of -1 to 1, and the range of raw values for a variable is 1-5, the raw value "1" will scale to -1 and the raw value "5" will scale to 1. This type of scaling for neural networks is well understood in the art. These twenty-one input vector variables, each of which represents a particular characteristic of a patient, are: Anxiety ("ANX"), Depression ("DEP"), Somatization ("SOM"), Combined Values of Anxiety, Depression and Somatization ("TOTAL ADS"), Age, Sex, Total Gross Charges, Total number of line item charges ("Total Count"), the Total Out-Patient Gross Charges, the total number of Out-Patient Visits ("Out-Patient Count"), the Total Gross Charges attributable to Doctor Office Visits ("Gross Charges DOV"), the number of Doctor Office Visits ("DOV Count"), the Total Gross charges attributable to Non-Doctor Office Visits ("Gross Charges Non-DOV ), the number of non-Doctor Office Visits ("Non-DOV Count"), the number of different Primary Care Physicians ("PCP") seen, the number of different specialists seen, the number of different Out-Patient Diagnoses, the number of inpatient or out-patient diagnostic ranges, whether definite patient driven care diagnoses are present ("Patient Driven Definite"), whether illness driven care diagnoses are present ("Illness Driven"), and whether possible patient driven care diagnoses are present ("Patient Driven Possible"). The values of the first four input vector variables (Anxiety, Depression, Somatization, and total ADS) are obtained from various answers provided by the patient to the Lifestyle Questionnaire. In the Questionnaire, five questions relating to the patient's anxiety, depression and somatization are listed. These questions are designed to provide some information about the psychological make-up of the patient. With reference to the Lifestyle Questionnaire in Appendix A, five questions (questions 21-25) in the Lifestyle Questionnaire are designed to determine the anxiety level of a patient. Each affirmative response to questions 21 through 25 indicates that the patient experiences a certain level of anxiety. A score of "1" is assigned to each affirmative response and a score of "0" is assigned to each negative response. A sum of the values for questions 21-25 becomes the value for the input variable number 1, the Anxiety variable. Thus, as explained above, the value of the input variable for Anxiety may range from 0-5, whereby a score of 0 indicates that the patient experiences no or very little anxiety and a score of 5 indicates that a patient experiences substantial anxiety. As shown in the Lifestyle Questionnaire in Appendix A, questions regarding depression and somatization are listed to identify the existence of these psychological ailments in a patient. Questions 26-30 of the Lifestyle Questionnaire in Appendix A relate to the Depression input variable depicted in Table 1. Questions 31-35 relate to the Somatization input variable from Table 1. As with the Anxiety variable, because there are five questions which relate to depression and to somatization, the Depression input variable and the Somatization input variable each has a range from 0-5, as depicted in Table 1. The score of the Anxiety, Depression, and Somatization input variables are summed to generate the Total ADS variable listed in Table 1. Thus, the Total ADS variable has a range of 0-15. The values of the variable numbers 5 and 6, Age and Sex, are also obtained from information listed in the Lifestyle Questionnaire or can be obtained from patient medical records. The Age input variable, variable number 5, ranges from 18 to 100+. This value is represented in the data file by three decimal digits. The Sex input vector variable, variable number 6, is defined as having a value of 0 for male and 1 for female. This variable is represented with a single binary digit (0 or 1) in the patient input vector data file (represented by the single "b" in the Format column). The values of variable numbers 7-18 are advantageously obtained from the patient's medical records. The "Gross Charges" variable, variable number 7, refers to the gross dollar amount of charges that the patient has incurred through his or her use of health care. The value for this variable is recorded in hundreds of dollars and ranges from 0-99999. As illustrated in Table 1, this value is represented with five decimal digits in the data file for each patient. The "Total Count" variable, variable number 8, refers to the total number of line item charges that the patient has incurred. This value has a defined range from 0-9999 and is represented with four decimal digits in each data file. The "Gross Charges Out-Patient" variable, variable number 9, represent the Total Out-Patient Gross Charges incurred. The value of this variable is recorded in each patient data file in hundreds of dollars ranging from 0-99999 in a field in the data file having five decimal digits. The "Out-Patient Count" variable, variable number 10, represents the total number of out-patient visits of the patient. In the present embodiment, out-patient visits refers to the combined number of doctor office visits and non-doctor office visits (any medical encounter without visiting the attending physician, such as lab work). The value of this variable is listed in the data file with a from 0-9999 in a field of having four deemed digits. The "Gross Charges DOV" variable, variable number 11, represents the total dollar charges attributable to doctor office visits by the patient. The value of this variable is listed in the patient data file in hundreds of dollars, ranging from 0-99999 in a field in the data file of five decimal digits. The "DOV Count" variable, variable number 12, refers to the number of times a patient has visited a doctor at his or her office. The value of this variable is listed as a number in the patient data file ranging from 0-9999, in a field having four decimal digits. The "Gross Charges Non-DOV" variable, variable number 13, represents the total charges attributable to non-doctor office visits. In the present embodiment, a non-doctor office visit is any medical encounter which did not involve the attendance of a physician, including visits to laboratories for medical testing. The value of this variable is listed in the data file in hundreds of dollars, ranging from 0-99999, in a field having five decimal digits. The "Non-DOV Count" variable, variable number 14, represents the number of non-doctor office visits by a patient. The value of this variable is listed in the patient data file, ranging from 0-9999, in a field having four decimal digits. The "number of PCP's seen" variable, variable number 15, represents the number of different primary care physicians seen by a particular patient. The value of this variable is listed in the data file, ranging from 0-99, in a field having two decimal digits. The "number of different specialists seen" variable, variable number 16, represents the number of different specialists seen by a patient. The value of this variable is listed in the data file, ranging from 0-99, in a field having two digits. The "number of different Out-Patient Diagnoses" variable, variable number 17, refers to the number of outpatient diagnoses made by physicians seen by the patient. The value of this variable is listed in the data file, ranging from 0-999, in a filed having three digits. The "number in-patient or out-patient diagnostic ranges" variable, variable number 18, represents the number of predetermined ranges into which in-patient or out-patient diagnoses for a patient are classified. As well understood in the art, there are several predefined diagnostic ranges for different organ systems in the human body (e.g., Infectious and Parasitic Disease, Neoplasms Endocrine, nutritional and metabolic diseases, etc.). The diagnostic ranges of the present embodiment are listed in Appendix C. The diagnoses that fall into each range are recorded in each patient's medical records. The patient receives a count of "1" for each different range in which the patient has diagnoses. The total number of ranges into which diagnoses for a patient are categorized is the value for variable number 18. The value of this variable is listed in the data file, ranging from 0-19, in a field having two decimal digits. The "Patient Driven Definite" variable, variable number 19, represents whether any definite patient driven diagnoses are present. An exemplary list of definite patient driven diagnoses is provided in Appendix B. The value of this variable is represented in a patient data file with a binary digit, where yes=1 and no=0. The "Illness Driven" variable, variable number 20, represents whether any illness driven diagnoses are present. An exemplary list of illness driven diagnoses is provided in Appendix B. The value of this variable is represented in a patient data file with a binary digit, where yes=1 and no=0. The "Patient Driven Possible" variable, variable number 21, represents whether any possible patient driven diagnoses are present. An exemplary list of possible patient driven diagnoses is provided in Appendix B. The value of this variable is represented in a patient data file with a binary digit, where yes=1 and no=0.
TABLE 2
______________________________________
Variable Variable
Number Name Range Format.sup.1
______________________________________
1 category 1 Yes = 1, No=0 b
2 category 2 Yes = 1, No=0 b
3 category 3 Yes = 1, No=0 b
4 category 4 Yes = 1, No=0 b
______________________________________
As briefly explained above, in order to train a neural network a correct output or output vector is provided for each input vector in a training set. TABLE 2 is a format chart of patient output vectors for the neural network 20. The output vector basically describes the category of care for the patient. In the present embodiment, the output vector comprises the values corresponding to the four categories of health care utilization described above. Category 1 represents Patient Driven Care, Category 2 represents a combination of Patient Driven Care and Illness Driven Care, Category 3 represents Illness Driven Care and Category 4 represents Possible Patient Driven Care. It will be understood that additional categories are also envisioned, and the particular categories selected do not limit the scope of the present invention. As illustrated in TABLE 2, each category of care is assigned a value of 0 or 1, where 0 indicates that a particular category of care is inapplicable and 1 indicates that the patient is classified in that category of care. Preferably, for any given input vector, only one category in the output vector has a value of 1. In other words, only one category of the four categories listed applies for any given patient.
TABLE 3
__________________________________________________________________________
Column Number
00000000011111111112222222222333333333344444444445555555555666666666677777
777778888888
12345678901234567890123456789012345678901234567890123456789012345678901234
567890123456
5 2 4 22 41 1 1129 473 378 468 21 26 355 422 4 3 23 10 0 1 1 0 1 0 0
Input Var Number Out. Var Number
1 2 3 44 555 6 77777 8888 99999 1111 11111 1111 11111 1111 11 11 111 11 1
2 2 1 2 3 4
0000 11111 2222 33333 4444 55 66 777 88 9 0 1
__________________________________________________________________________
As mentioned above, information for each patient is represented in a patient data file which is configured to be read by the computer 12. For training purposes, the data file for each patient in the training set includes the expected output vector. For patients to be analyzed after training, the output vectors are unknown, and provided by the neural network 20. An exemplary patient data file format is illustrated in TABLE 3. In TABLE 3, the Column Number, Input Var Number and Output Var Number labels, and corresponding reference indexes are provided for ease of description. The column numbers in TABLE 3 are read from top to bottom. Therefore, the top two rows of numbers make up the column numbers. The third row is the variable value. The fourth and fifth rows of numbers, read vertically, identify the variable number. For example, the value of input variable number 1 (identified by "1" in the fourth row of numbers in the first column), listed in column 01 (identified by the "0" in row one and the "1" in row two in the first column) is 5 (provided in the third row of numbers, column one). Because variable number 1 is Anxiety, as described above, the 5 indicates that the patient is experiencing significant anxiety. The next column, column 02 serves as a field separator (i.e., it provides a separator between the data fields for input variables 1 and 2). In the present embodiment, each data field in the patient file is separated by a separator as depicted in TABLE 3. Other field separators, such as commas, are well understood in the art. The third column, column 03 contains the value of input variable number 2, which represents Depression. The value of this variable is 2, which indicates that the patient is suffering from some depression. In one embodiment, in an actual patient data file, the patient data (i.e., the actual variable value depicted in the third row of numbers in TABLE 3) is recorded without the column number or variable number indexes. These indexes may prove helpful to human operators, but the computer 12 uses the actual variables as input. Additionally, the "Column Number", "Input Var Number" and "Out. Var Number" headings are removed. The remainder of the input variables are represented in columns 05-77. Columns 79-85 depicted in TABLE 3 contain the values for the output variables. For example, the value of output variable number 1, listed in column 79 is 0. This indicates that this category of care, Category 1, is inapplicable to the patient. The value of output variable number 2, listed in column 81, is 1. This indicates that the patient is identified as a utilizer of this category of care, Category 2 (i.e., the patient is seeking a combination of patient driven care and illness driven care). The remaining output variable columns (Columns 83 and 85) contain a "0" in the example depicted in TABLE 3. As explained above, in order to train the neural network 20, expected output vectors corresponding to input vectors are provided. Accordingly, training the neural network 20 requires that a judgment be made with respect to categorizing the health care utilization of the training set of patients. Advantageously, the training set is substantially representative of the patient population. Typically, the more vectors (each vector representing one patient) provided in the training set, and the more accurate the associated expected output vectors for each input vector, the more reliable the training of the neural network. FIG. 2 illustrates the general factors used in the present embodiment to ascertain an expected classification of particular patients into one of four categories of care for purposes of training the neural network 20. In other words, these factors were selected to generate the original output vectors for the training set. It should be noted that the parameters used to generate the training set may or may not correlate to the function ultimately provided by the neural network 20 based upon the 21 input variables illustrated in TABLE 1. In the present embodiment, Category 1 represents Patient Driven Care, Category 2 represents a combination of Patient Driven Care and Illness Driven Care, Category 3 represents Illness Driven Care and Category 4 represents Possible Patient Driven Care. For purposes of generating expected output vectors for the training set, a particular patient's utilization of health care is classified as Patient Driven Care, Category 1 utilization, if the patient displays all the following characteristics: (1) the patient receives high scores on the Lifestyle Questionnaire. In the present embodiment, a patient's questionnaire scores are considered high if the patient receives: (a) a score equal to or greater than 3 for any of the ANX, DEP or SOM input vector variables; or (b) a Total ADS Score equal to or greater than 6. (2) no illness driven diagnosis is present; (3) The patient may have had a high number of different diagnoses categories. In the present embodiment, a patient is considered as having a high number of different diagnostic categories if the patient: (a) has diagnoses in six of more diagnostic ranges and is younger than 60 years old; or (b) has diagnoses in 8 or more diagnostic ranges, and is 60 years old or older. (4) The patient may be diagnosed as having a definite patient driven care disorder or two or more possible patient driven care disorders. A listing of illness driven diagnoses and definite and possible patient driven diagnoses is provided in Appendix B. As depicted in FIG. 2, a patient's utilization is classified mixed patient driven care and illness driven care, Category 2 utilization, if the patient has the following characteristics: (1) the patient scores highly on the questionnaire, as described above; (2) an illness driven diagnosis is present; (3) the patient may be diagnosed in a high number of diagnostic ranges; and (4) The patient may be diagnosed as having a definite patient driven care disorder or two or more possible patient driven care disorders. A patient's utilization is classified as illness driven care, Category 3 utilization, if the patient exhibits the following characteristics: (1) the patient has low scores in the Lifestyle Questionnaire. In the present embodiment, a patient scores low on the questionnaire if (a) he receives a score of less than 3 on each of the ANX, DEP or SOM input vector variables; and (b) his Total ADS Score is less than 6. (2) an illness driven diagnosis is present; (3) the patient may be diagnosed in a high number of diagnostic ranges; and (4) no definite patient driven diagnosis is present, and less than two possible patient driven care diagnoses are present. A patient's utilization is classified as Possible Patient Driven Care, category 4 utilization, if the patient exhibits the following characteristics: (1) patient scores low in the questionnaire, as discussed above for category 3; (2) no illness driven diagnosis is present; (3) the patient has been diagnosed in a high number of diagnostic ranges; and (4) The patient may be diagnosed as having a definite patient driven care disorder or two or more possible patient driven care disorders. In training the neural network, many presentations of the training set and weight adjustments are required for the network to converge (become stable). During training, the neural network 20 is periodically tested to determine how well it generalizes from specific examples it has learned. To test the network, the trained net is presented with novel patient vectors, and the accuracy of the responses of the net is recorded. Training and testing proceed, until the neural network achieves the desired degree of accuracy on new input vectors or until accuracy no longer improves. A generalized training process 30 for the neural network 20 of the present invention is depicted in the flow diagram of FIG. 3. The training process 30 begins, as represented in a start block 32. First, input and output vectors are scaled to the -1 to 1 range discussed above, and any other preprocessing of the training data is completed, as represented in an action block 33. The next step is to initialize the neural network 20, as represented in an action block 34. In this step, variables used in the neural network are initialized. In particular, the weights used for the neural network inputs and between layers are selected as random values, and the learning rate, network architecture, activation function, and other parameters are also selected. Input vectors and their corresponding target output vectors are then presented to the neural network in random order, as represented in an action block 38. The input vectors and target output vectors are preferably read from a file created either manually by an operator via the keyboard 14 or by providing commands to the processing system 10 to retrieve the appropriate patient data files from the training set on the storage media 16. In the present embodiment, the data in the patient data files is preferably stored in a conventional ASCII format. Each data file in the training set consists of two vectors of numbers: an input vector and a target output vector. Each vector consists of a series of numbers separated by one or more blank spaces, as explained above with reference to TABLE 3. The input vector represents the patient data which is presented to the network, and the target output vector represents the target response of the network (i.e., the output which should be produced by the network in response to this input vector). Based on the input vector, and the weights initially assigned, the neural network calculates the output response, as represented in an action block 40. Next, the neural network 20 performs a comparison between the calculated output response with the target output to provide an error signal, as represented in an action block 42; and the weights from each input are adjusted to partially compensate for the error signal, as represented in an action block 44. The amount of compensation in the adjustment of the weights is dependent upon the selected learning rate for the neural network 20. If the learning rate is too fast, the compensation is too great for each error signal; and the neural network 20 loses what the neural network "learned" from each vector when the subsequent vector is processed. Accordingly, a learning rate between 0 and 1 is selected in the present embodiment to provide less than complete adjustment for each error signal, as well understood in the art. Next, a determination is made whether this iteration is the last iteration before testing, as represented in a decision step 46. In other words, several iterations with the training set are generally completed before testing. If the current iteration is the last iteration, the training process 30 proceeds to testing, as represented in the action block 48. If the iterative process is not complete, the neural network 20 repeats the steps represented in the action blocks 38-46, until all vectors in the training set have been presented to the neural network 20 a number of times. When the iterative process depicted by steps 38-46 has completed (decision block 46), the neural network 20 is tested, as represented in the test network action block 48. In this step, the neural network 20 receives new input vectors, not part of the training set. The output response of the neural network 20 is then calculated and compared to a target response. If the neural network 20 produces accurate results, the operator may determine that training is adequate, as represented in a decision block 50, and the training process is complete, as represented in an end block 52. If the result is unacceptable and further improvement is desired, the neural network 20 repeats steps 38-46 for a new training set, or for the original training set, or a combination of some or all of the original training set with additional new input vectors. A variety of network models may be used in the neural network 20. Typical models include the back-propagation network, the radial basis function network and the learning vector quantization network, as known in the art. For each neural network type, the appropriate training processes is utilized, as well understood in the art. After the neural network 20 is trained, it may be used to process new information (new input vectors). In one embodiment of the present invention, prior to utilizing the trained neural network 20 to analyze health care utilization of patients, patient information may be preprocessed to identify patients who are at "high risk." In the present embodiment, high risk patients are those patients considered to be suffering from an immediate threat to their health. The flow diagram of FIGS. 4A and 4B illustrate one possible embodiment of a prescreening process 60 in accordance with the present invention. Beginning at a start block 62, the process proceeds to a read patient data action block 64. Patient data is accepted via keyboard input and/or via data files. As represented in a decision block 66, a determination is made whether the patient has "high" Lifestyle Questionnaire scores, a represented in a decision block 66. In the present embodiment, a patient is considered to have a high Lifestyle Questionnaire score if a "CAGE" score is greater than or equal to 2, a MEBS score is greater than or equal to 8, or a score is equal to or greater than 3 on any one of the ANX, DEP or SOM input variables or a total ADS score is equal to or greater than 6. The CAGE score is obtained from the Lifestyle Questionnaire, and relates generally to psychological problems due to alcohol addiction. With reference to the Lifestyle Questionnaire in Appendix A, four questions in the Questionnaire relate to the CAGE score. Specifically, Questions 16 through 19 inquire whether (1) the patient has ever felt that he should Cut Back on his drinking, (2) if the patient has ever been Annoyed by others' criticisms of his drinking, (3) if the patient has ever felt Guilty about his drinking habits and (4) if the patient has ever consumed alcohol as soon as he wakes up so as to steady his nerves or to overcome a hangover. A score of "1" is assigned to each affirmative response and a score of "0" is assigned to each negative response. CAGE stands for "Cut back, Annoyed, Guilty, Eye Opener," which are attributes related to alcoholism. The total from the four questions provides the patient's total CAGE score. If the CAGE score is equal to or greater than 2, the patient is identified as having a high Lifestyle Questionnaire score. The MEBS score is also obtained from the Lifestyle Questionnaire. As seen in the Lifestyle Questionnaire, ten questions (questions 5 through 14) relate to MEBS. These questions relate to a patients caffeine usage. If the patient answers yes to eight or more of questions 5 through 14, the patient is identified as having a high Lifestyle Questionnaire score. The patient scores for ANX, DEP and SOM are also obtained from the Lifestyle Questionnaire. As seen in the Lifestyle Questionnaire, questions 21-25 relate to anxiety (ANX), questions 26-30 relate to depression (DEP) and questions 31-35 related to somatization (SOM). As explained above, if a patient has three or more yes answers in any of the ANX, DEP or SOM categories of questions, or has a combined total of six or more for all three categories, the patient is identified as having a high Lifestyle Questionnaire score. Any patient identified as having a high Lifestyle Questionnaire score is identified as being at high risk, as represented in an action block 68. This information is then forwarded to a case manager, as represented in an action block 70. The case manager may automatically receive this information via a computer network, or the operator for the pre-processing system may provide this information to the case manager. It should be understood that data for all patients could be pre-processed before the case manager is notified. For instance, each patient that is to be forwarded to a case manager could be identified in the pre-processing; and a list of all such patients, with corresponding recommendations, forwarded to the case manager. However, if a patient's Lifestyle Questionnaire score is not high (decision block 66), or the high risk identification is complete (action block 68), those patients who appear to have a narcotic dependency problem are identified, as represented in a decision block 76. If a patient has been provided with two or more narcotic prescriptions in two or more consecutive quarters by two or more doctors at two or more pharmacies, the patient is identified as being at high risk. as represented in the action block 68. This information is then forwarded to the case manager, as illustrated in the action block 70. It should be understood that the Lifestyle Questionnaire data and pharmacy (narcotic) data from all patients could be pre-processed before the case manager is notified. If the patient's records do not suggest a narcotic dependence (decision block 74), a determination is made whether more patients are available for pre-screening, as represented in a decision block 78. If more patients are available for screening, the steps represented in block 64-78 are repeated. If no further patient files are available, patients are ranked according to the total cost of claims paid for the patient group under consideration, as represented in an action block 80 (FIG. 4B, via continuation point A). Next, patients with the least frequent utilization are eliminated as unlikely candidates for inappropriate utilization. In the present embodiment, the top 10% (dollar amount) for total claims paid for the group under consideration are kept for consideration. The bottom 90% (by total dollars of claims paid) are removed from consideration. In one embodiment, only the top 10% of utilizers are processed by the neural network 20, as represented in an action block 84. Alternatively, all patients are passed to the neural network 20 for processing. Pre-screening for high risk may be advantageous even when all patients are analyzed with the neural network 20, in that the case manager receives a quick indication of the likelihood of problems. The flow diagram of FIG. 5 illustrates a generalized operational process 90 for the trained neural network 20. Beginning at a start block 92, control proceeds to an action block 94. As represented in the action block 94, the patient input vector is presented to the trained neural network 20. The neural network 20 calculates the output vector for the selected patient, as represented in an action block 96. Next, a determination is made whether all patients, from a group of patients to be analyzed, have been analyzed by the neural network 20, as represented in a decision block 98. The neural network repeats the processing until all patients have been analyzed. The process completes in an end block 100. The output vector for each patient is advantageously stored in the patient's data file, or in a separate file categorizing patients by their category of health care utilization. Those patients who have been classified in Category 1, Category 2 or Category 4 are forwarded to case managers for further examination in an attempt to identify inappropriate usage, and to treat the underlying problems for the patients. It will be understood, that for each patient continually utilizing health care in an inappropriate way, many health care dollars will be spent in treatments which will not assist in the underlying problem. Thus, by characterizing patients as having the potential of inappropriate utilization, the case managers can assist in obtaining proper diagnoses, and eliminating the unnecessary expenses of treating ailments which merely stem from the underlying problem. Although the present invention has been described in terms of certain preferred embodiments, other embodiments can be readily devised by one skilled in the art in view of the foregoing. For instance, processing methods other than neural networks may be appropriate given the factors suggesting utilization characteristics depicted in FIG. 2. Additionally, further factors may also be considered in making the above determinations. Accordingly, the scope of the present invention is defined by reference to the appended claims.
APPENDIX C
______________________________________
OUTPATIENT
ICD 9 CODES & CLASSIFICATION
______________________________________
001-139 Infectious and Parasitic Disease
140--230
Neoplasms
240-279.9
Endocrine, nutritional and metabolic diseases
280-289 Immunity diseases; diseases of the blood and blood
forming organs
290-319 Mental disorders
320-389 Diseases of the nervous system and sense organ
390-459 Diseases of the circulatory system
460-519 Diseases of the respiratory system
520-579 Diseases of the digestive system
580-629 Diseases of the genitourinary system
630-676 Complications of pregnancy, childbirth and the
puerperium
680-709 Skin and subcutaneous tissue
710-739 Diseases of the musculoskeletal system &
connective tissue
740-759 Congenital anomalies
760-779 Certain conditions originating in the perinatal
period
780-799 Symptoms, signs and ill defined conditions
800-999 Injury and poisoning
E Codes
V Codes
______________________________________
|
Same subclass Same class Consider this |
||||||||||