"SYSTEM FOR PREDICTING THE COMSUMPTION OF HEALTH CARE RESOURCES" FIELD OF THE INVENTION This invention concerns a method and data processing system for predicting the risk of a patient with chronic illness consuming excess health care resources. More specifically, the invention concerns a method and data processing system for: (1) defining the characteristics of chronically ill patients that are predictive of consuming excess health care resources; (2) developing and refining a risk scoring algorithm to predict the risk that a patient with a chronic illness will consume excess health care resources; and (3) applying the risk scoring algorithm to particular patients with chronic illnesses for various purposes which could include identifying patients for intervention techniques, improving the delivery of health care services and/or determining/adjusting insurance premiums.

BACKGROUND The medical community has developed a proven set of strategies for helping the lives of most patients with chronic illnesses. There are numerous strategies identified for the major disease categories, such as heart disease, lung disease, respiratory illnesses, cancer, etc. In the management of asthma, the National Heart, Lung, and Blood Institute, has published the National Asthma Education and PreVention Program.

Management strategies identified include, among others, allergy avoidance and the use of inhaled corticosteroid. Similarly, in the management of coronary artery disease following myocardial infarction, standards have been set to improve patient care and disease management including such strategies as reduction of the lifestyle behaviors that are known to increase the risk of complications and disease progression and possibly induce disease regression.

Nonetheless, the global health care system in the United States has been extremely ineffective at implementing these strategies. By way of example, asthma, which is considered by most experts to epitomize a chronic illness that can be managed

entirely outside of the hospital, nonetheless consumes over 2 billion dollars per year in hospital costs, at the same time generating indirect costs, such as work absenteeism, that exceed 2.5 billion dollars per year.

One of the significant problem areas in treating patients with chronic illness of any sort is that many chronically ill patients fail to utilize or stay with a proven cost-effective treatment regimen for their condition. As a result, these patients will consume a larger than expected amount of health care resources in the form of emergency room treatments, unscheduled doctor and health practitioner appointments and hospital/treatment facility stays. The consumption of resources by these patients is considered disproportionate when compared to other patients who have the same chronic conditions but yet are treated for these conditions without the necessity of such expensive interventions as emergency room treatments or hospital/treatment facility stays.

For purposes of this invention, a"chronic illness"is defined broadly to include any continuing disease process with no more than minor changes in symptoms from day to day if the disease process is treated properly. Some examples of chronic illnesses are diabetes, arthritis, asthma, chronic obstructive pulmonary disease (COPD) and coronary artery disease.

We have found that there are many reasons why certain patients consume excess health care resources, many of which are not directly tied to the patient's medical condition. A particular patient may not have adequate transportation to make scheduled medical appointments. Another patient may have complicating, psychological conditions, such as depression, that may result in the patient not taking his or her prescribed medication. Still other patients may be subject to problematic environmental factors (such as a chronic asthma patient being exposed to certain allergens in the home), or any combination of the foregoing factors. In short, we have found that many patient factors impact the consumption of health resources, including medical, environmental, socioeconomic, psychological and logistical factors.

A principal difficulty today with chronic illness is predicting which types of patients with chronic illness are likely to consume a disproportionate share of health

care resources. Significant benefits can be achieved by identifying which patients with chronic illness are at risk of incurring potentially avoidable resource utilization (such as emergency room visits and hospital/treatment facility stays). If health providers were armed with such predictive information, energy and resources could then be directed towards the patients who were at risk o consuming excess resources and appropriate steps could be taken to reduce the risk that these high risk patients will consume said resources.

Most of the predictive rules that have been developed in the medical area have been based on linear regression analysis. The problem with these types of models is that their usefulness diminishes as the complexity of the problem (including the number of factors involved) increases. For these reasons, no linear regression models exist for predicting the consumption of health care resources. Non-linear analysis using tools such as neural networks are well known and have been used successfully to develop predictive models outside the medical area. In addition, these tools have also been used in some medical contexts such as disease diagnosis. No one, however, has used these tools to develop predictive models for the consumption of excess health care resources.

Thus, there is presently no objective method or information system available for predicting the consumption of health care resources by those with chronic illnesses.

Indeed, there is a strong sentiment from many that health care issues are simply too complex for present information technology. J. D. Kleinke, a leading medical economist was quoted in the Wall Street Journal as stating that the practice of medicine is an art, which"is too complex to be digitized: it elements of chaos theory that the typical information technology would be hard pressed to... incorporate into product design. "L. Landro, Information Technology Could Revolutionize the Practice of Medicine. But Not Anytime Soon, Wall Street Journal (June 25, 2001). Consequently, without any objective information system or method, health professionals are left to make subjective consumption estimates based on their limited anecdotal experience with the patient in question and with prior patients. And health providers who have attempted to make predictions based on some anecdotal data have failed to focus on all the psychological, socioeconomic, logistical and environmental factors that are critical

to predicting which patients with chronic illnesses are likely to consume excess health care resources.

OBJECTS AND SUMMARY OF THE INVENTION In view of the foregoing, an object of this invention is to provide a method and data processing system for predicting the consumption of health care resources for patients with chronic illnesses of any type.

A further object of this invention is to provide a method and data processing system for determining the patient characteristics that are predictive as to whether a particular patient will consume excess health care resources.

A further object of this invention is to provide a method and data processing, system for developing a risk scoring algorithm for computing a chronically ill patient's risk of consuming excess health care resources based upon quantitative and non- quantitative assessment of the patient's predictive characteristics.

A further object of this invention is to provide a method and data processing system for applying a risk scoring algorithm to the patient's predictive characteristics in order to objectively determine a patient's risk of consuming excess health care resources.

A further object of this invention is to provide a method and data processing system for developing a risk scoring algorithm that can be used to determine and or adjust insurance premiums or otherwise estimate the costs that particular patients are likely to incur.

A further object of this invention is to provide a method and data processing system by which the risk-scoring algorithm is continually updated as additional data on chronically ill patients is collected.

A further object of this invention is to provide a method and data processing system for which patients with a high risk of consuming excess health care resources can be identified for specific interventions that can reduce the risk of excess consumption of healthcare resources and improve the patient's quality of life.

Briefly, the above and other objects of this invention are realized in a method and data processing system that involves the development of specific patient criteria that are predictive of the risk that a chronically ill patient will consume excess health care resources, that provides for the collection of chronically ill patient data that is used to predict the risk of excess health care resource consumption, that uses pattern recognition software to create a risk scoring algorithm for computing the risk that a particular patient will consume excess health care resources, that applies the risk scoring algorithm to the predictive patient criteria to generate a score that provides objective indication of the risk that a chronically ill patient will consume excess health care resources, that continues to refine the risk scoring algorithm as additional data on chronically ill patients is processed allowing specific focused application of healthcare resources to patient illness, that uses the risk scoring algorithm for purposes including developing intervention programs for specific patients who are a high risk of consuming excess health care resources, calculating and adjusting insurance premiums, and estimating health care costs.

The foregoing features of the present invention are more readily understood in the context of a specific illustrative example thereof described in detail herein below in conjunction with the following figures.

BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a block diagram of an illustrative network environment in which the present invention may operate.

Figure 2 is a block diagram of a process for creating the risk-scoring algorithm.

Figure 3 is a flow chart of the process by which the risk-scoring algorithm can be applied to particular patients with chronic illness.

Figure 4 is a flow chart showing the analysis of intervention methods for a particular patient.

Figure 5 is a flow chart of the processes by which the risk-scoring algorithm is continually updated as additional data on chronically ill patients is processed.

DETAILED DESCRIPTION OF THE INVENTION The present invention now will be described more fully hereinafter with reference to the accompanying drawings in which an illustrative example of the invention is shown. This invention may, however, be embodied in many different forms and should not be limited to the example set forth herein.

The present invention is described below with reference to flowchart illustrations of methods and data processing systems according to the illustrative example. It will be understood that each block of the flow chart illustrations, and combinations of blocks in the flow chart illustrations, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart block or blocks.

These computer program instructions may also be loaded onto a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart block or blocks.

The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

Referring now to Figure 1, a network is depicted in which the illustrative example of the invention may operate. The illustrative network includes a central server (10) that is accessible to one or more local servers (40) via a computer network, which is depicted here as the Internet (30) but could be any type of network. The

central server (10) includes a central database (20), which is the repository for data maintained by the system, including the data on chronically ill patients that is utilized by the system. The central database (20) can be implemented using standard database program such as Microsoft SQL Server application program.

The central server (10) also includes the application programs necessary to utilize the invention such as the pattern recognition software used by the system to create and update the risk scoring algorithm. Although the invention is not limited to particular type of pattern recognition software, the pattern recognition software used in this illustrative example is a neural network. The neural network software utilized is preferably BrainMaker v. 3.7 for Windows 95 and NT by California Scientific Software, which has a widespread and proven use in medicine; business management; publicly traded stocks, commodities and futures; science; and manufacturing.

The local servers (40) serve to provide the access by which remote users can communicate with the central server (10). Although only two are shown in Figure 1, there is no limit to the number of local servers that could be connected. In addition, while a remote user's access point is described here as being through a local server (40), the invention can be implemented through any type of access device capable of communicating with the central server (10). The local server (40) can provide various functions, including providing a means for a remote user to collect data on chronically ill patients and transmit the data to the central server. In addition, through a local server (40), a remote user can communicate requests to the central server (10) such as requesting that a risk score be generated for a particular chronically ill patient.

Referring now to Figure 2, a flow chart is shown of the process for developing a risk scoring algorithm. Once the risk scoring algorithm is developed, a predictive score for the consumption of health care resources can be assigned based on a patient's particular characteristics.

The risk scoring algorithm process begins with"Developing Initial Predictive Characteristics" (100). A set of initial characteristics need to be developed, which are believed to be predictors of the consumption of health care resources. The range of proposed characteristics is not limited to specific patient medical data. The experience

of the inventors here has shown that non-medical factors play a significant role in predicting the consumption of health care resources. The invention can make use of any patient characteristics that are believed to be predictive of health care resource consumption.

In the particular illustrative example of the invention described here, patient characteristics fall into six categories: socioeconomic, interventional, psychological, logistical and environmental. Socioeconomic factors are those relating to education, income and wealth. Interventional factors are those which can be brought to bear on the illness/or increased risk score in question and refer to the types of non- pharmacological treatments such as rehabilitation programs, behavior modifications, patient education, new diagnostic evaluations, environmental changes, etc. Medical factors refer to the process of diagnosis, prognosis, and application of pharmacological, surgical or nutritional modalities. Psychological factors relate to the patient's mental and emotional state such as moods or depression, and the behaviors present which either promote or impede wellness or recovery. Logistical factors refer to those that relate to patients ability to receive treatment such as miles between home and the nearest medical facility, and the availability of transportation, as well as the ability to access other resources recommended to maximize health. Environmental factors are those present in the school, home and/or occupational setting such as allergens, other pulmonary irritants, and extrinsic factors increasing, risk of illness.

Referring again to Figure 2, the next function after the initial predictive characteristics are developed is"Collecting Patient Characteristic Data" (110). A data gathering method such as a questionnaire is used to elicit from chronically ill patients the responses to the proposed predictive patient characteristics. The questionnaire should preferably elicit the socioeconomic, interventional, medical, pyschological and logistical factors noted previously that are believed to be predictive of the consumption of excess health care resources.

The preferred implementation of this is to create a Web page for the questionnaire on the central server (10) so that the questionnaire can be accessed from any remote location with access to the Internet (30). Once the data called for by the questions is elicited from the patient and entered and any required review is performed,

the data can be transmitted directly to the central database (20). Security features (high encryption) will need to be implemented to maintain patient confidentiality and prevent unauthorized use of the Web site and avoid corruption of the data. Although the questionnaire can be implemented electronically by means other than a Web site, the standardization of remote access through a Web site provides certain advantages over other methods.

The questionnaire can, of course, be completed on paper and then input into the local server (40) for transmission to the central server (10) or loaded directly from paper to the central server (10). However, given the number of criteria being collected on the questionnaire, the labor costs would be extremely high and transcribing the responses introduces significant risk of error.

The other data collection function that needs to be done before a risk scoring algorithm can be developed is"Collecting Patient Outcome Data" (120). Outcome data needs to be gathered for each of the patients whose characteristic data has been entered in the central database. The patient outcomes that are collected are those showing the consumption of excess resources--emergency room visits, hospital stays, unplanned hospital and doctor visits, and excessive use of medication.

The system can be set up to track any outcome and is not necessarily limited to the four listed above. However, these are the outcomes used in this particular illustrative example. The patient outcome data can be loaded via a Web site on the central server (10) in the same manner that patient characteristic data is entered. An initial set of both patient characteristic and outcome data needs to be collected before an accurate risk-scoring algorithm can be obtained.

Referring again to Figure 2, the"Develop Proposed Risk Scoring Algorithm" (130) is triggered once a predefined amount of patient characteristic and outcome data has been collected. The function, executed through software on the Central Server (40) retrieves the patient characteristic and outcome data from the central database (20) and transmits the data to the neural network, which will perform pattern recognition techniques on the data. This function can be set to execute at whatever frequency is desired on either a manual or automatic basis. As mentioned previously, the pattern

recognition/neural network software used in this illustrative embodiment is Brain Maker v. 3.7 for Windows 95 and NT.

The output of the"Develop Proposed Risk Scoring Algorithm" (130) function is a risk scoring algorithm which may or may not show to a reasonable accuracy the extent to which particular patient characteristic data can accurately predict the medical outcomes being tracked that show an excess consumption of health care resources.

Accuracy requires a certain amount of historical data. There is no specific amount of data that must be collected before a predictive algorithm to a reasonable accuracy can be generated. The amount of data that needs to be collected will depend on the type of pattern recognition software being used, the type of results collected, and the desired accuracy of the predictions. Generally, from a statistical standpoint the more relevant data that is collected, the more accurate the algorithm will be at predicting excess consumption of health care resources.

Here, with the neural network software being used in this illustrative example and the type of data being collected, it is estimated that data on about 500-600 patients is needed to generate an algorithm than can accurately predict the relative risk that any single patient will consume excess health care resources. In this illustrative example, accuracy here means that the predicted risk shall have no more than a 5 percent statistical error. Said differently, accuracy of prediction increases with the number of patients tested, resulting in the neural network's capability of reliably predicting relative risk with 95 percent accuracy. This accuracy figure is for illustrative purposes only. Depending on the desired objective, a different accuracy criterion ma, be utilized.

After generating a Proposed Risk Scoring Algorithm, the neural network software will perform the"Accuracy?" (150) function. What this function does is determine whether sufficient data has been collected so that the software can generate a risk scoring algorithm that can predict risk of excess health care consumption to the desired accuracy. If the risk scoring algorithm cannot predict risk to the desired accuracy, the process returns to the data collection step (110) so that additional data can be collected until the desired accuracy is achieved.

After sufficient data is collected to achieve the desired accuracy, the result from the neural network is a risk scoring algorithm that will determine to the desired accuracy which characteristics are most predictive of a patient who is at risk of consuming a disproportionate share of health care resources.

Based on the relative predictability of each characteristic, a weighted score can then be developed for each patient. The score can be translated into any type of number scale. In this illustrative example, a score of 10 signifies the highest risk of consuming disproportionate health care resources, whereas a score of 1 is the least risk of consuming disproportionate health care resources.

Referring now to Figure 3, the process is shown for assessing risk factors once a risk algorithm has been developed that has a satisfactory accuracy. The purpose here is to identify chronically ill patients that are at risk of incurring excess health care resources. At the same time, the patient characteristic data continues to be collected in the same manner as the data used for developing the initial risk-scoring algorithm.

Continually collecting data has the benefit of continuing to refine and make more accurate the risk scoring algorithm being used.

The first step to applying the risk scoring algorithm is the"New Patient"" (200) function. If the patient is determined to be new, then the same"Collecting Patient Characteristic Data" (110) function from Figure 2 must be performed so the relevant characteristics can be gathered. The preferred implementation of this is to have the patient characteristic data questionnaire accessible through the Web site on the central server (20). Once the patient characteristic data is collected, the"Execute Risk Scoring Algorithm" (220) function can be performed to determine the risk score for the patient at issue. This function collects the patient characteristic data and generates a 1-10 score which reflects the extent to which the patient is likely to consume excess health care resources. For convenience, a local server (40) can be set so that the patient data can be entered and the risk-scoring algorithm executed directly from the local server (40).

If the patient is not new and data has previously been loaded for a particular patient, the process will first retrieve from the central database (20) the data previously

entered. The Retrieve Patient Data (230) function performs the function of retrieving previously stored data. Then before executing the algorithm, the Enter Changes to Patient Data (240) permits existing data to be modified prior to executing the algorithm These functions are preferably implemented through a Web page that is accessible to each local server (40).

Once the"Execute Risk Scoring Algorithm" (220) function is performed the "Transmit Risk Score" (250) function transmits the score to the desired output location, which will preferably be a Web page on a local server (40). As noted previously, the risk score generated by the risk scoring algorithm for a particular patient with a chronic illness represents the risk that the patient will consume excess health care resources.

The recipient of the score can use it for a number of functions. For example, if the score indicates a high risk of consuming excess health care resources, this information can then be used by the responsible medical professionals to design a treatment and intervention strategy to address the risk prior to the consumption of excess health care resources. A high or low score can be used by insurance providers or other entities to adjust premiums or expected costs of services.

Referring now to Figure 4, a flow chart is shown describing the analysis of intervention methods for a particular patient. The process begins with the decision box of"Is There An Intervention In Place?" (260) for the particular patient. If there is an intervention in place, the next step in the process is to evaluate"Should The Intervention Be Modified?" (262). In this step, a medical professional is looking at the results of the present intervention the patient is on to determine whether or not it appears to be working in reducing the patient's consumption of health care resources.

If a decision is made not to modify the intervention, this process then proceeds to "Execute Risk Scoring Algorithm" (220) to generate a new risk score for the patient.

This step assumes that whatever new data there is for the patient (e. g. higher blood pressure value, loss of weight, change of or loss of employment or other changed circumstances) is entered prior to a new risk score being generated.

If a decision is made to modify an intervention, the process proceeds to the "Enter New Intervention" (266) step. In this step, a medical professional evaluates the patient's history and then defines the new appropriate intervention and enters that new

intervention into the patient's data. Again, the"Execute Risk Scoring Algorithm" (220) step will be executed, which in this case will show what the patient's risk score is based on the new intervention method.

Going back on Figure 4 to the"Is There An Intervention In Place?" (260) step, if the answer is no, the next step is"Should There Be An Intervention?" (264). If a medical professional evaluating the patient's history determines that an intervention should be instituted, the process again proceeds to the"Enter New Intervention" (266) step. After the appropriate intervention is defined and entered, the process again proceeds to the"Execute Risk Scoring Algorithm" (220) where the patient's risk score will be generated taking into account that an intervention has been developed for the patient. If the answer to"Should There Be An Intervention?" (264) is no, then the process proceeds directly to"Execute a Risk Scoring Algorithm" (220).

The significance of generating a new risk score after each one of the above scenarios is determine whether any changes being made are actually impacting a patient's risk score. In this way, it is possible for example to put in a variety of possible interventions to determine which one may be most beneficial for a particular patient based on his risk score. Of course, this system is still just a tool so if a risk score is not lowered for a particular patient based on a proposed intervention it does not necessarily mean that such an intervention should not be tried.

There may, for example, be unique factors for a particular patient or not enough data on particular intervention yet for a reliable prediction.

Referring now to Figure 5, a flow chart is shown for the processes by which the neural network is utilized to improve the predictability of the risk scoring algorithm, to improve the knowledge base on what patient factors are relevant to calculating the risk of consuming excess health care resources and to improve the intervention methods used on patients who are at high risk of consuming excess health care resources.

The key to these improvement processes described in Figure 5 is that as new patient characteristics and outcome data is collected, the data continues to be processed by the neural network. The"Collect Additional Patient Data" (300) function involves continuing to collect patient characteristic data by the methods discussed previously.

The"Generate New Risk Scoring Algorithm" (310) function updates the algorithm as more patient criteria and outcome data is processed through the neural network.

Updating the risk-scoring algorithm based on additional patient data can be done at any type of standard time interval, such as once a week, depending upon how much data has been collected.

The knowledge base on patient criteria can be improved because as patient results continue to be processed, the Identify Non-Predictive Factors (320) function of the neural network will identify patient criteria that do not need to be collected because the criteria are not predictive of consuming excess health care resources.

In addition, the neural network also allows new criteria to be added and tested to determine whether the criteria are predictive. Once new criteria are identified, the "Collecting Patient Data for New Criteria" (330) function collects the data using same data gathering methods discussed previously. The"Predictive?" (340) function processes this data to determine if it is predictive of the consumption of excess health care resources. If yes, the"Update Risk Scoring Algorithm" (350) function updates the algorithm based on the new criteria. If after a sufficient number of trials it is determined that a particular criteria appears not be predictive, the"Excluded Criteria" (360) function eliminates the new criteria.

Finally, the neural network permits assessments to be done on how well certain intervention techniques have succeeded in addressing high risk patients. This analysis begins with"Collect Data on Intervention Methods" (370). Outcomes are identified to patients who have received additional intervention services because of their high-risk.

Through the"Evaluate Intervention Data" (380) function, the neural network determines the extent to which the intervention services have resulted in the high-risk patient consuming less health care resources when compared to a similar high risk patient that did not receive similar intervention services.

The pattern recognition software will identify if the intervention services have resulted in a reduction in health care resource consumption, and if there has been an improvement in health outcomes.

The foregoing is illustrative of the present invention and is not to be construed as limiting thereof. Those skilled in the art will readily appreciate that many modifications are possible without materially departing from the novel teaching and advantage of this invention. Accordingly, all such modifications are intended to be included within the scope of this invention as defined in the claims. The invention is defined by the following claims, with equivalents of the claims to be included therein.