Title:
Healthcare management system and method of predicting high utilizers of healthcare services
Kind Code:
A1


Abstract:
A healthcare management system and method identify individuals at risk of high near-term healthcare use. On step of the method includes collecting information from an individual for a predetermined set of predictive factors. Another step of the method includes assigning, based upon the information from the individual, a separate value to each predictive factor of the predetermined set of predictive factors. The method further includes the step of generating, based upon a predetermined predictive model and the separate values assigned to the predetermined set of predictive factors, a risk level of the individual utilizing healthcare services at a predetermined level within a prospective time span.



Inventors:
Meek, Julie A. (Greenwood, IN, US)
Lyon, Brenda L. (Indianapolis, IN, US)
Lynch, Wendy D. (Lakewood, CO, US)
Application Number:
10/426773
Publication Date:
10/16/2003
Filing Date:
04/30/2003
Assignee:
The Haelan Corporation
Primary Class:
International Classes:
G06Q10/10; G06Q50/22; G06Q50/24; G16H10/60; (IPC1-7): G06F17/60
View Patent Images:



Primary Examiner:
RAJ, RAJIV J
Attorney, Agent or Firm:
Intellectual Property Group (Indianapolis, IN, US)
Claims:

What is claimed is:



1. A method of managing healthcare services, comprising the steps of: collecting information describing an individual's perceived health relative to their subjective expectation of health; and generating, based upon a predictive model, an output suggestive of future healthcare service use by the individual.

2. The method of claim 1, wherein the output is a risk value indicative of the likelihood of the individual utilizing healthcare services at a predetermined level within a prospective time span.

3. The method of claim 1, further including the step of assigning, based upon the information, a separate value to each of a plurality of predetermined factors in a set of predictive factors.

4. The method of claim 1, wherein the information collected comprises at least one data point for each of a plurality of predetermined predictive factors.

5. The method of claim 4, further including the step of comparing one collected data point to a set value for the respective predictive factor to determine if the collected data point is indicative of either a high risk or a low risk of future healthcare service use, and assigning, based upon the comparing step, a value to the predetermined predictive factor.

6. The method of claim 5, wherein each of the predictive factors is inserted into the predictive model to generate the output.

7. A method of managing healthcare services, comprising the steps of: collecting information directly from an individual for a predetermined set of predictive factors; assigning, based upon the information from the individual, a separate value to each predictive factor of the set of predictive factors; and generating, based upon a predetermined predictive model and the separate values assigned to the predetermined set of predictive factors, a risk level of the individual utilizing healthcare service at a predetermined level.

8. A method of managing healthcare services, comprising the steps of: providing a survey including a plurality of questions; receiving response information from an individual for each of said plurality of questions of the survey; determining, from the information of each question, whether the response is indicative of one of a high or a low future use of healthcare services; assigning values to a plurality of predictive factors based upon the determining step; and generating, based upon the assigned values, an output suggestive of future use of healthcare services.

9. The method of claim 8, wherein the survey is presented by a computer.

10. The method of claim 9, further including the step of providing the individual with at least one intervention program in response to the generating step generating an output suggestive of a high risk of future use of healthcare services.

11. The method of claim 10, wherein the intervention program is provided substantially contemporaneously with the receiving of responses from the individual.

12. The method of claim 8, wherein the plurality of questions includes a question regarding the individual's perceived health.

13. The method of claim 8, wherein the output is a risk value indicative of the likelihood of the individual utilizing healthcare services at a predetermined level.

14. A method of managing healthcare service including the steps of: collecting information about an individual including a plurality of data points corresponding to a plurality of factors in a set of predictive factors; determining, for each data point, what the data point indicates about use of healthcare services by the individual; and generating, based on a predictive model, an output suggestive of future use of healthcare services by the individual.

15. The method of claim 14, further including the step of assigning a value to each factor in the set of predictive factors, each assignment being based upon the determining step for each of the corresponding data points, wherein the generating step is performed by inserting into the predictive model the assigned values for the set of predictive factors.

16. The method of claim 14, wherein the determining step categorizes the data points as either indicative of a high use of healthcare services or a low use of healthcare services.

17. The method of claim 14, wherein the determining step includes the step of assigning to each predictive factor one of two dichotomous values.

18. The method of claim 14, wherein the output is a risk value indicative of the likelihood of the individual utilizing healthcare services at a predetermined level.

19. A method of managing healthcare service including the steps of: analyzing healthcare information utilizing a predictive model represented by the equations z=β1X12X23X34X4 . . . βn-1Xn-1nX1X2+C, and Phighuse=ez/(1+ez), wherein β1, β2, β3, β4, . . . βn represent regression coefficients, X1, X2, X3, X4, . . . Xn-1 represent predictor variables, the predictor variables are dichotomous variables, and the values of the dichotomous predictor variables are determined based upon the healthcare information.

20. The method of claim 19, wherein individuals with a Phighuse value greater than a threshold value are identified as individuals with a high-risk of high near-term healthcare use.

21. The method of claim 19, wherein the analyzing step causes the generation of reports.

22. The method of claim 21, wherein the reports are one or more reports selected from the group including: a general health report, advice as to how the individual may improve their overall health, information about appropriate intervention programs, and follow-up coaching reports from care providers.

23. The method of claim 21, wherein the reports are one or more reports selected from the group including: a summary of the wellness of a managed care group, a list of individuals identified as high risk, and a report identifying intervention programs which address wellness needs of a managed care group.

Description:

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is a continuation of U.S. patent application Ser. No. 09/520,419, filed Mar. 8, 2000, which is expressly incorporated herein by reference.

FIELD OF THE INVENTION

[0002] The present invention relates generally to healthcare management systems, and more particularly to a healthcare management system and method for identifying individuals at risk for high near-term healthcare use.

BACKGROUND OF THE INVENTION

[0003] In commercial adult managed care populations, typically 10-20% of the population will incur 80-90% of the risk exposure for the risk-bearing payer. A major challenge in today's managed care environment is prospectively predicting which members are at risk for high utilization to enable appropriate targeting of proactive case management and intervention programs. These members could enjoy better health and incur fewer health-related expenses if they could be identified early and managed proactively.

[0004] Historically over the last 30 years, health assessment tools for adult populations (18-65 years of age) have been built upon clinical risk/chronic disease models to guide clinical decision making, to serve as outcome measures, or to automate aspects of care such as automated problem lists, laboratory results, and healthcare reminders. The emphasis has been on practical clinical issues, resulting in limited predictive power in terms of targeting those at risk for higher levels of healthcare-seeking behavior.

[0005] Increasingly, the ability of risk-bearing payers to approach population health management in a cost and quality-effective manner, will be dependent upon sound predictive models which guide care toward members who have a higher probability of becoming a high care user in the near-term. Adding momentum to the efforts, technological advances such as ASP and dynamic HTML Web tools, widespread connectivity to the Internet, and digital, variable-data print technologies have only recently become available to make population-based care management a real option versus a rhetorical question.

[0006] Theoretically and empirically, evidence suggests that near-term healthcare utilization may not be as strongly tied to traditional disease and clinical risk models as once thought. Indeed, 40-60% of office visits are often tied to non-disease based concerns. Also, from a practical perspective, risk/disease models often miss key prescriptive data which may guide the health counseling process (i.e., areas of readiness to change, stress levels, levels of perceived self-management, etc.) and key outcome measures (i.e., improvements in functional ability, levels of stress emotions, disease and medication compliance, etc.). In a practical sense, claims-based stratification methods are limited by the inherent inadequacies of the claims data itself, including the lack of claims history from new enrollees and the several month time lags often required to access and process the predictive information.

[0007] Therefore, a need exists for a healthcare management system and method that identify, based upon easily ascertained self-reported information, those members of a managed healthcare group who are at risk for high near-term healthcare use

SUMMARY OF THE INVENTION

[0008] The present invention addresses the above-identified need, as well as others, with a method of managing healthcare services. On step of the method includes collecting information from an individual for a predetermined set of predictive factors. Another step of the method includes assigning, based upon the information from the individual, a separate value to each predictive factor of the predetermined set of predictive factors. The method further includes the step of generating, based upon a predetermined predictive model and the separate values assigned to the predetermined set of predictive factors, a risk level of the individual utilizing healthcare services at a predetermined level within a prospective time span.

[0009] Pursuant to another embodiment of the present invention, there is provided a healthcare management system including a processor, and a memory operably coupled to the processor. The memory includes instructions that, when executed by the processor, cause the processor to assign, based upon information from an individual, a separate value to each predictive factor of a set of predictive factors. The instructions of the memory, when executed by the processor, further cause the processor to generate, based upon a predetermined predictive model and the separate values assigned to the predetermined set of predictive factors, a risk level of the individual utilizing healthcare services at a predetermined level within a prospective time span.

[0010] Pursuant to a further embodiment of the present invention, there is provided a computer readable medium for a healthcare management system. The computer readable medium including instructions that, when executed by the healthcare management system, cause the healthcare management system to assign, based upon information from an individual, a separate value to each predictive factor of a set of predictive factors. The instructions, when executed by the healthcare management system, further cause the healthcare management system to generate, based upon a predetermined predictive model and the separate values assigned to the predetermined set of predictive factors, a risk level of the individual utilizing healthcare services at a predetermined level within a prospective time span.

[0011] It is an object of the present invention to provide an improved healthcare management system and method.

[0012] It is also an object of the present invention to provide a new and useful healthcare management system and method.

[0013] It is another object of the present invention to provide a healthcare management system and method for identifying high risk individual users of healthcare services.

[0014] It is yet another object of the present invention to provide a computer readable medium for configuring a healthcare management system to identify individuals at risk of high near-term healthcare use.

[0015] It is yet another object of the present invention to provide a healthcare management system and method for identifying members of a managed care group who are at risk to utilize healthcare services at a predetermined usage rate within prospective time span.

[0016] The above and other objects, features, and advantages of the present invention will become apparent from the following description and the attached drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

[0017] FIG. 1 is a simplified block diagram of an exemplary healthcare management system which incorporates various features of the present invention;

[0018] FIG. 2 illustrates an exemplary operating environment in which the healthcare management system of FIG. 1 may be utilized; and

[0019] FIG. 3 is a flowchart depicting the general operation of the healthcare management system of FIG. 1.

DESCRIPTION OF EXEMPLARY EMBODIMENTS

[0020] While the invention is susceptible to various modifications and alternative forms, exemplary embodiments thereof have been shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that there is no intent to limit the invention to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.

[0021] Referring now to FIG. 1, a simplified block diagram of an exemplary healthcare management system (“HMS”) 100 is shown which incorporates various features of the present invention. As will be explained in more detail, the exemplary HMS 100 is generally operable to identify members (i.e. individuals), based upon perceived health information, who are at risk of high near-term healthcare use. Besides identifying members of the managed care group who are at risk for high near-term healthcare use, the HMS 100 enables proactive case management of the identified members. For example, the HMS 100 enables healthcare managers identify individuals who are at risk of incurring significant healthcare expenses in the near-term so that intervention programs may be targeted to these identified individuals in order to help these individuals to enjoy better health and to incur fewer health-related expenses.

[0022] To this end, the exemplary HMS 100 includes a processor 102, memory 104, a system bus 106, controllers 112, 114, 116, 118, network interface 120, and devices 122, 124, 126, 128. The hardware architecture of the exemplary HMS 100 is generally illustrative of general purpose computing systems such as personal computers, workstations, and servers currently manufactured by Gateway, Dell, Macintosh, Sun Microsystems, and IBM. While the hardware architecture of the exemplary HMS 100 is generally illustrative of general purpose computing systems, the HMS 100 may also be implemented as a special purpose computing device that is generally operable to perform a limited set of tasks associated with healthcare management.

[0023] Returning to the description of the exemplary HMS 100, the processor 102 includes one or more processing units such as the central processing units of Intel, AMD, IBM, National Semiconductor, and Transmeta. The processor 102 is generally operable to execute software and/or firmware routines stored in the memory 104. As a result of executing the software and/or firmware routines of the memory 104, the processor 102 controls the general operation of the HMS 100 and the devices 122, 124, 126, 128 via the respective controllers 112, 114, 116, 118. Moreover, the processor 102 as a result of executing software and/or firmware routines of the memory 104 is generally operable to implement a predictive model that identifies, based upon perceived health, members of a managed care group who are at risk of high near-term healthcare use.

[0024] The memory 104 of the exemplary HMS 100 is operable to store data and code used by the processor 102. To this end, the memory 104, in an exemplary embodiment, includes standard random access memory for storing the data and software code needed by the processor 102. However, the memory 104 may alternatively include other volatile memory types such as DRAM, SDRAM, and SRAM for storing data and software code and/or non-volatile memory such as ROMs, PROMs, EEPROMs, and flash memory for storing data and firmware code.

[0025] The system bus 106 is generally operable to interconnect the processor 102, the memory 104, and the controllers 112, 114, 116, 118. To this end, the system bus 106 in the exemplary embodiment includes an address bus and data bus which enable the various components of the exemplary HMS 100 to communicate with one another.

[0026] The removable media drive 122 is generally operable to read information from and write information to the removable storage medium 150, and the removable storage media controller 112 is generally operable to provide the processor 102 with an interface to the removable media drive 122. The removable media drive 122 may include various computer readable and/or writeable storage media devices such as floppy disk drives, DVD drives, tape drives, DAT drives, and ZIP drives in order to access corresponding types of removable storage media. The removable media drive 122 may also include non-volatile memory interfaces for accessing non-volatile memory devices such as PROMS, EEPROMS, and Flash Memory.

[0027] The mass storage device 124 is generally operable to store data and/or software code of the exemplary HMS 100, and the mass storage controller 114 is generally operable to provide the processor 102 with an interface to the data and/or software code stored by the mass storage device 118. To this end, the mass storage device 118 may include various computer readable and/or writeable media devices such as hard disk drives, floppy disk drives, CD-ROM drives, DVD-RAM drives, RAID devices, and/or Disk-On Chip devices to name a few. The HMS 100 may alternatively be implemented without a mass storage device 122 and mass storage controller 112. For example, the exemplary HMS 100 may be implemented with data and firmware code stored in a non-volatile manner in the memory 104.

[0028] The video display device 126 is operable to provide a visual display to a user of the exemplary HMS 100, and the video controller 116 is operable to provide the processor 102 with an interface to the video display device 126. To this end, the video display device 126 may include CRT displays, LCD displays, and/or standard television displays. The HMS 100 may alternatively be implemented without a video display device 126, and video controller 116. For example, the HMS 100 may be implemented to output results to an attached printer (not shown), the removable storage media 150, and/or another computing device attached to the HMS 100 via the network interface 120, a USB port (not shown), parallel port (not shown), or FireWire interface (not shown).

[0029] The input device 128 is operable to provide a user of the exemplary HMS 100 with an interface to control the operation of the exemplary FMS 100 and/or to provide the exemplary HMS 100 with information. Moreover, the I/O interface controller 118 is operable to provide the processor 102 with an interface to the input device 128. To this end, the input device may include a mouse, keyboard, touch pad, push buttons, touch screen, and/or digital scanner.

[0030] The network interface 120 is generally operable to couple the exemplary HMS 100 to a network such as the Internet, an intranet, and/or an extranet. The network interface 120 generally enables the exemplary HMS 100 to send and receive information to and from other computing devices coupled to the network. To this end, the network interface 120 may be implemented with a network interface card such as an Ethernet card or Token Ring card. The network interface 120 may also be implemented with an analog modem for use over POTS telephone lines such as a 28.8K or 56K modem. The network interface 120 may also be implemented with a digital modem such as a cable modem for use over a cable distribution network, an ISDN modem for use over an ISDN telephone line, or a DSL modem for use over a POTS telephone lines that support DSL.

[0031] An exemplary operating environment is illustrated in FIG. 2. As illustrated, the HMS 100 is operably coupled to client computing devices 2021, 2022 . . . 202X via a network 200. The network 200 may include multiple public or private LANs and/or WANs that are operably coupled to one another via routers, switches, hubs, gateways, proxies, and/or firewalls. In an exemplary embodiment, the network interface 120 of the HMS 100 is operably coupled to a LAN that provides a gateway to the Internet. As a result, the exemplary HMS 100 is accessible to client computing devices 2021, 2022 . . . 202X that are either coupled to the LAN or the Internet, thus providing the services of the HMS 100 to people located locally and throughout the world.

[0032] In an exemplary embodiment, the HMS 100 communicates with client computing devices 2021, 2022 . . . 202X in accordance with the Hyper Text Transport Protocol (HTTP) and other Internet Protocols (IP). Accordingly, the services of the HMS 100 are generally available to any client computing device that supports the appropriate IP protocols. Most personal computers such as those manufactured by Dell and MacIntosh include web browsing software that provide an interface to services delivered via the IP protocols. However, besides personal computers, many other devices are being developed that include web browsing capabilities. For example, set-top boxes such as Microsoft's WebTV and certain digital mobile phones include web browsing capabilities and could be used to access the services of the HMS 100.

[0033] As stated above, the HMS 100 identifies individuals who are at risk of high near-term healthcare use based upon perceived health information and other information. Perceived health is defined as a person's composite evaluation of how he or she is feeling and doing. People experience health as either some level of wellness or illness. If a person is experiencing unpleasant or uncomfortable physical sensations and/or emotions in addition to not functioning up to their perceived capability level, the person perceives some level of illness. Some degree of wellness, on the other hand, is perceived when the person experiences pleasant/comfortable physical sensations and emotions combined with an acceptable level of self-evaluated functioning.

[0034] The difference between the perceived health and disease/risk models is that a person can experience wellness in the presence or absence of disease. For example, a person who has diabetes can experience wellness in that from the person's perspective the person is feeling and doing well. Conversely, a person can experience illness in the absence of discernable disease in that from the person's perspective the person is not feeling nor doing well. From a practical viewpoint, people seek care when their sense of feeling and doing is below their perceived threshold of tolerance.

[0035] Evidence also shows a correlation between (i) an individual's beliefs and preferences regarding the healthcare system and (ii) particular types of healthcare-seeking behavior. In particular, individuals who place high levels of faith in the healthcare system or desire the input of health professionals to make health decisions tend to use the healthcare system more often.

[0036] As a result of the above observed correlations, the exemplary HMS 100 attempts to identify individuals at risk of high near-term healthcare use based upon information about an individual's perceived health, beliefs and preferences regarding the healthcare system, past levels of healthcare use, and demographics. To this end, the HMS 100 in step 310 of FIG. 3 obtains such information from an individual via a health perception questionnaire. An exemplary questionnaire is provided in Appendix A and includes questions that elicit information for a predetermined set of predictive factors which include past healthcare use factors, demographic factors, perceived health factors, disease factors, healthcare compliance factors, healthcare belief factors, and healthcare preference factors.

[0037] The exemplary HMS 100 is operable to display the health perception questionnaire on the video display 126 and receives responses to the questions of the questionnaire via the input device 128. Moreover, the exemplary HMS 100 is further operable to transmit the questionnaire to client computing devices 2021, 2022 . . . 202X via the network 200, and receives responses to the questions of the questionnaire from client computing devices 2021, 2022 . . . 202X via the network 200. The HMS 100 may also include various other mechanisms for receiving responses to the health perception questionnaire. For example, the HMS 100 may include a digital scanner that is operable to read from a printed questionnaire an individual's responses to the questions of the questionnaire. The HMS 100 may also obtain responses to the questionnaire from archived information stored on a removable medium 150, the mass storage device 124, or a database accessible via the network 200. Further yet, them HMS 100 may obtain responses to the questionnaire via telephone (e.g. an interactive voice response telephone system).

[0038] After obtaining information from an individual, the exemplary HMS 100 analyzes the information based upon a predictive model developed from similar information collected from other individuals and their prospective use of healthcare services. As will be explained in more detail below, the exemplary HMS 100 analyzes the information based upon a logistic regression predictive model represented by the following equations:

z=β1X12X23X34X4 . . . β38X3839X1X2+C,

Phighuse=ez/(1+ez)

[0039] where β1, β2, β3, β4 . . . β39 represent the regression coefficient shown in Table 2 of Appendix B; X1, X2, X3, X4 . . . X38 represent the predictor variables shown in the questionnaire of Appendix A and Table 2 of Appendix B; and C represents the regression constant shown in Table 2 of Appendix B. Those skilled in the art should appreciate that predictor variables X1 and X2 are interaction variables. Moreover, other analysis tools such as neural networks, genetic algorithms, and fuzzy logic could also be used to develop the predictive model.

[0040] In order to utilize the above logistic regression predictive model, the exemplary HMS 100 in step 330 processes an individual's responses to the health perception questionnaire in order to obtain suitable values for the predictor variables X1, X2, X3, X4 . . . X38. More specifically, the predictor variables X1, X2, X3, X4 . . . X38 of the logistic regression are converted to dichotomous variables (i.e. variables that can have one of only two values such as 0 or 1). Accordingly, the exemplary HMS 100 assigns an appropriate dichotomous value to each of the predictor variables X1, X2, X3, X4 . . . X38 based upon the responses to the health perception questionnaire. In the exemplary embodiment, the HMS 100 generally assigns a first dichotomous value of “0” to each variable X1, X2, X3, X4 . . . X38 that the responses suggest are not indicative of the individual being at risk of high near-term healthcare use, and assigns a second dichotomous value of “1” to each variable X1, X2, X3, X4 . . . X38 that the responses suggest are indicative of the individual being at risk of high near-term healthcare use. Table 2 of Appendix B illustrates the possible responses to the questionnaire for each predictive variable and corresponding dichotomous value assigned to the predictive variable X1, X2, X3, X4 . . . X38. Moreover, for each variable each variable X1, X2, X3, X4 . . . X38 corresponding to an answered or an inappropriately answered question, the exemplary HMS 100 assigns the variable a value of “0” in order to be conservative in identifying individuals as high near-term users of healthcare services.

[0041] After assigning a dichotomous value to each of the predictor variables X1, X2, X3, X4 . . . X38, the HMS 100 in step 340 calculates, based upon the above logistic regression equations, a value indicative of whether the individual is at risk of high near-term healthcare use. Essentially, the exemplary HMS 100 in step 340 obtains for the individual a probability value, ranging between 0 and 1, that is indicative of the probability that the individual is a high-risk for high near-term healthcare use.

[0042] The HMS 100 then in step 350 determines based upon the probability value obtained for the individual in step 340 whether the individual should be identified as a high-risk or a low-risk person. In particular, the exemplary HMS 100 makes such a determination by comparing the obtained probability value to a predetermined threshold value. The threshold value is preselected to strike a balance between sensitivity and specificity of the measurement. Sensitivity is defined as the percent of people who become high care users and are so identified prospectively by the predictive formula. Specificity is defined as the percent of people who do not become high care users and are so identified by the formula. From a practical viewpoint, the threshold value may also be selected dependent upon the sponsoring plan's objectives, intervention capacity, financial parameters, and expectations. The exemplary HMS 100 utilizes a threshold value of 0.47 and identifies any individual with a probability value greater than the threshold value as a high-risk of high near-term healthcare use. Furthermore, besides identifying an individual as a high-risk if their associated probability value Phighuser exceeds the threshold value, the exemplary HMS 100 also marks an individual as a high risk if the individual fails to answer or incorrectly answers more than a predetermined number of questions.

[0043] The HMS 100 in step 360 outputs the results of the analysis. In particular, the exemplary HMS 100 is operable to display output results on video display 126 or transmit the output results to a client computing device 2021, 2022 . . . 202X for display. In the exemplary embodiment, the HMS 100 is operable to generate several types of reports and restrict the generation of certain reports to authorized personnel. For example, a member of the managed care group (i.e. an individual responding to the health perception questionnaire) may cause the HMS 100 to generate reports which (i) provide a general health report, (ii) advise as to how the individual may improve their overall health, (iii) provide information about appropriate intervention programs, and (iv) provide follow-up coaching reports from care providers. On the other hand, personnel of a managed care group may cause the HMS 100 to later generate similar reports about the above identified member and additional reports that (i) summarize the wellness of the managed care group, (ii) list individuals identified as high risk, and (iii) identify intervention programs which address wellness needs of the managed care group.

[0044] Development and Validation of the Exemplary Predictive Model

[0045] The predictive model of the exemplary HMS 100 was developed from an initial member population of 20,000 members ages 18-65 with claims data showing more than 6 primary care visits within the last year. A health perception questionnaire containing 70 question/126 response was mailed to the 20,000 members. Subsequently, for the 5,285 members who responded, claims data were obtained from the health plan for the next six months yielding 4,271 members who responded to the health perception questionnaire and were also continuously enrolled during the entire study time period. Those members who completed an intervention program targeted to high utilizers (N=61) were removed from subsequent analyses, leaving 4,210 for the final sample. The sample included 3 female respondents for every male respondent. To assure that the predictive model was developed for a gender-representative population, cases were weighted (3:1) to approximate the normal gender distribution of 51.2% female to 48.8% male. Both the original and weighted sample descriptives are listed in Table 1 of Appendix B.

[0046] As indicated above, logistic regression was selected for developing the predictive model; however, other analysis tools such as neural networks, genetic algorithms, and fuzzy logic could also be used to develop the predictive model. Logistic regression is especially suited for highly skewed data such as the six-month claims encounter data. For this reason, members confirmed as outliers in terms of six-month total dollars or encounters were left in the data set, because the conversion to a dichotomous variable mitigates the outlier effect.

[0047] To prepare all questionnaire information for logistic analysis, each survey item of the questionnaire was converted to a dichotomous variable with a value of “0” indicating low-risk and a value of “1” indicating high-risk responses for each item. In cases where item responses could not be conceptually categorized as high-risk versus low-risk, analyses were run comparing responses with subsequent care use to determine the final coding. All missing survey item data (<10% on each variable) was coded as “0” to be conservative.

[0048] “Total encounters” over the subsequent six-month period was chosen as the dependent variable. An encounter was defined as a unique person with a unique provider on a unique date. Thus for example, if a person went to their physician, was subsequently sent to a lab and finally to pick up a prescription, the total encounters would equal three. With the impact of DRG's on hospital stays and the increasing movement of services to the outpatient and office settings, this definition of encounters was felt to more accurately represent total healthcare use with accompanying total dollars.

[0049] The decision to dichotomize the dependent variable, essentially deciding where to mark the cut-off for a high (1) versus low (0) care user, was a more difficult challenge. A practical rule was used that compares the percent of the population with the percent of premium dollar used. What seems to occur with regularity across adult commercial populations is that the top 2% of care users utilize 40% of the premium, the next 18% of high care users utilize the next 40% of the premium; the next 65% of the members will use the remaining 20% of the premium and 15% of the population will have no care use within the given year. The total member population for the health plan at that time was 36,354 people.

[0050] The sample population was ranked from highest to lowest based upon total claims dollars, and each segment of the above identified segments were then weighted until the population reached 36,354. Subsequently, the resulting data set was checked to assure that the percentage of the population coincided with the percent of premium dollar used. Using this method, 3.6% of the extrapolated population used the first 40% of the premium and 22.3% used the next 40% of the premium. By observing the mean, median and range of encounters for this combined 25.9% group versus all others in the population, the dependent variable cut-off was determined to be 6 or greater encounters in a six month period as the high-risk group coded as “1”.

[0051] After the dependent variable and all of the potential predictor variables were coded with dichotomous values, logistic regression was applied using all perceived health, risk, disease, compliance, belief/preference and demographic variables. A split-half technique was used where the predictive model is developed on a randomly selected first half of the data set (N=2,067) and then tested for its stability on the second half of the population (N=2,143). Using standard logistic regression on the first split-half of the weighted sample, 50 variables were found to have significance levels at 0.25 or less including 3 interaction variables in the initial analysis. Subsequent to this initial analysis, a forward stepwise regression was run using these 50 variables on the first split-half, with 39 variables entering the equation. The results of the forward stepwise logistic regression of the probability of becoming a high encounter member using the 39 final predictor variables on the first split-half are shown in Table 2 of Appendix B. The obtained logistic model correctly predicted 68.08% of the observed high care use group and 61.93% of the observed low care use group.

[0052] Next, the predictive model was converted to a formula, which could be used to test the first and second split-halves. First, the formula calculates each person's probability (Phighuse) of becoming a high care user by inserting his or her responses to the 39-predictor variables into the logistic formula:

z=β1X12X23X34X4 . . . β38X3839X1X2+C,

Phighuse=ez/(1+ez)

[0053] where β1, β2, β3, β4 . . . β39 represent the regression coefficient shown in Table 2 of Appendix B; X1, X2, X3, X4 . . . X38 represent the predictor variables shown in the questionnaire of Appendix A and Table 2 of Appendix B; and C represents the regression constant shown in Table 2 of Appendix B.

[0054] Next, the member is predicted to be high-risk or low-risk according to whether his or her resulting probability value Phighuse exceeds a threshold value that is determined a priori. The threshold value was selected to strike the optimal balance between the sensitivity and specificity of the measure. The probability values for the entire weighted respondent population ranged from 0.06 to 0.99 with a mean of 0.53, median of 0.52 and a standard deviation of 0.18. Various threshold values were evaluated for their efficacy using these parameters resulting in a selection of 0.47 as the final probability threshold value. This resulted in 59.9% of the entire weighted respondent population classified as high-risk, which when extrapolated to the entire population equaled approximately 6.9% of the total member population. This confirms, as stated previously, that the study sample had a disproportionate number of past and subsequent high care utilizers.

[0055] To test the predictive validity of the predictive model and the HMS 100, the formula was tested on both halves of the population with the Chi Square test using predicted (0,1) vs. observed (0,1) care use groups. In the first split-half (see Table 3 of Appendix B), the formula resulted in 72.5% (sensitivity) of the high-risk group predicted correctly and 52.7% (specificity) of the low-risk group. Positive predictive value was 63.5% and negative predictive value was 62.9%. In the second split-half, the formula resulted in 64.2% (sensitivity) of the high-risk group predicted correctly and 46.7% (specificity) of the low-risk group. Positive predictive value was 57.3% and negative predictive value was 53.9%.

[0056] For the total weighted population, the predicted high-risk members have significantly higher encounters and dollars than the low-risk group as detailed in Table 4. High-risk group average encounters exceeded the low-risk group by nearly 3 encounters as well as an average of $301 more in average care use dollars. Reduction in these dollars by even a net 25% through proactive intervention would have netted this health plan conservatively a half million dollars annually.

[0057] Moreover, the selection of 0.47 for the probability threshold value resulted in approximately 7% of the study population potentially needing prospective health advocacy. Accordingly, the health perception questionnaire and the HMS 100 are effective in identifying individuals at high risk for future high near-term care use without placing an exorbitant burden upon proactive case managers.

[0058] To test the health perception model of the exemplary HMS 100 against more traditional stratification methods, a logistic model was developed which used merely traditional predictor variables. The predictor variables used were previous hospitalization in the last six months, age, and presence of a chronic disease. At a probability threshold value of 0.56, sensitivity was 60.1% and specificity was 51.1%. Results showed that compared to the traditional method of identifying high-risk members, the health perception model of the exemplary HMS 100 added 12.4% to the sensitivity and 1.6% to the specificity in the first split-half, thus contributing considerable predictive power. In addition, because the exemplary HMS 100 uses self-reported information versus claims data, the time lag from assessment to intervention is shortened from several months to less than 24 hours from the initiation of data collection. Care providers also are not subsequently required to complete an additional health assessment, because the exemplary HMS 100 provides prescriptive as well as predictive information to guide health-counseling efforts.

[0059] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description is to be considered as exemplary and not restrictive in character, it being understood that only exemplary embodiments have been shown and described and that all changes and modifications that come within the spirit of the invention are desired to be protected. embedded image embedded image embedded image embedded image embedded image embedded image embedded image embedded image embedded image embedded image