Portions of this review (c) 2006 Brian Altonen  

Developing a Population Health Monitoring Tool based on Prescription Data


There were two projects ongoing from 2004 to 2006.  The first dealt with making better use of internal institutuional data pertaining to inpatient and out patient vistis, labs, and other internal sources of medical information such as claims, survey results, hot line utilization, disease registeries and the like.  The end product of this was an access program that could automatically import several large datasets and prepare the information for a standard set of epidemiological queries, such as number of teens visiting a clinic for an STD or number of asthma related ED visits.  The second project was focused on the naturional pharmacy database.  The purpose of this work was to learn how to make the best use of this database, by applying it to two PIPs and QIAs set up for internal quality improvement activities: the care management studies of asthma and diabetes.

Both of these projects took up a lot of my spare time.  Through the use of the institution’s pdf-form medical records database, I was able to manage the annual HEDIS and PIP/QIA projects and develop a few of my own.  At one point I decided to make heavy use of the pharmacological data to determine what other measures could be developed out there.  This goal was established after I was asked to review pregnancy cases in order to develop a study on pregnancy-related diabetes onset.  Due to low case counts this project was cancelled following two series of pre-evaluations. 

In 2005 I began the process of evaluating a series of large institutional datasets stored on the intranet (the National NCPDP Pharmacy Database, for details see www.ncpdp.org).  The purpose of this was to be able to review the complete records for all office visits, lab tests, inpatient and outpatient activities, and prescription drug utilization histories, with the goal of assessing three populations for various HEDIS, NCQA, QIA and PIP-related measures.  With the exception of the complete demographic list (ca. 500K rows in length), all of these databases consisted of 3.5 million or more rows (event entries), with the largest database consisting of prescription data and containing more than 20 millions rows of data.  The second largest dataset (ca. 5-10M rows) was the documentation of lab visits and other non-clinical or office events (xrays, screening activities, diagnostic visits, etc.). 

The task at hand was to link these four major datasets together to form a  single database using Access.  Patient IDs were used to link this information together.   These datasets could then be merged with a dataset developed for the external pharmacy use dataset, with an number of entries of about 3 million to 6 million scripts per quarter, averaging about 3 to 7 scripts per patient on the average if high cost individuals are excluded.

One of the major issues with linking up institutional datasets is the periodic upgrades they go through.  Most of the original datasets used to produce the institutional dataset are updated monthly, and on different days of the month.  This means that the structure and content of the initial datasets used to form the database bears different identifiers each month, requiring that links be re-established each time the system is opened for a periodic reassessment of population health.  Aside from this issue, very little limited the ability of these datasets to be linked within the Access system for the sole purpose of developing a program meant to identify population health related features.

A quick review of the isolated datasets pertaining to office and lab/diagnostic visits revealed that there were certain practices that could be monitored in frequency relative to the population as a whole as well as its parts.  For example, one of the most common reasons for children 12 to 18 years of age to have lab visit data, aside from standard physiochemical measures like blood screenings, urinalyses,  and the like, pertained to STDs.  This suggested that a measure of STD visits could be used to quantify activities by this particular age group, thus forming the first epidemiological measure for this study.  Another lab/diagnostic visit found to be helpful in identifying high risk children 12-18 years of age was the information on asthma-related testing, in particular pulmonary function testing.  A matching document indicating an evaluation and education visit related to asthma history was used to verify an important history of childhood asthma cases; each case had to be subsequently reviewed for elimination of other forms of asthma common to children in this age range such as exercise-induced onset.

The screening of children’s age groups consisted of a number of measures designed to document population health for children 0-2 years of age, 3-5 years of age, 6-11 years of age, and 12-18 years of age.  These measures included a review of office visits, screenings, diagnostic tests, a number of very specific inpatient and outpatient activities, and prescription drug use top define a number of risk factors that had to be considered at the population level when reviewing childhood health and disease.  The various types of health related measures that could be obtained using this method included the discovery of medical records documenting evidence for important early stage physical and mental disability or “handicapped” states, important psychological histories, evidence for pre-smoking, smoking and drug use history, and various forms of chronic disease history ranging from genetic conditions such as Type I diabetes, to environmentally-linked factors such as asthma and severe astham/COPD-related problems during the early years, to behavioral problems such as type 2 diabetes and several childhood-linked psychological states of profiles based on regular and specialized care office visits.

The next population at risk evaluated for developing a method for profiling a single group was the 18 to 24 years old.  For several studies focused on this age group, age ranges extended slightly in either direction based upon the topic under review.  Chlamydia evaluations for example included patients as young at 15 years of age.  Prenatal-Postpartum studies, used to measure the healthy behaviors and activities of both the mother and child, focused on women as young as 14 (or 12) years of age and as old as 25.   One of the major purposes of this review of this particular age group was to document preventive health changes that take place due to change in age group status.  In particular, one of the most common problems noted with all Medicaid and Medicare systems has been the drop in health care and preventive care provisions once the age of 18 is reached.  For asthmatics this results in a significant reduction in drop in frequencies for pulmonary function testing, regular educational and allergy profiling visits, and the participation in local seasonal flu shot programs.  For children with epilepsy or some other chronic neurological and employment-related condition, this led to a reduction on quality of care given in many cases due to lack of active participation on behalf of the patient, with lack of follow-up on this problem by health care providers in general.  For URI, repeated visits became a measurable feature. 

Since this age range (esp. for 4 – 16 yo) is one of the most important periods in life regarding occupation choice, development and success, measurable features such as missed health care visits, incompleted immunizations, failure to manage physiological problems such as BMI and exercise-nutrition habits, can be considered potentially valuable methods that can be employed in order to improve long-term lifestyle conditions and practices.  Failure to maintain a decent health profile during this phase of life can in turn be linked to the higher likelihood for failures to complete schooling, to obtain work, and ultimately to prevent long-term complications brought on by poor quality of life features and the related high costs for care.

The third age range used to evaluate health conditions was typically 24 to 44 years of age.  The list of major health related problems assessed for evaluating the health of this population was extensive, and included most medical conditions that potentially become high cost conditions during the later years of life.  These health states that were evaluated included many included in the listing of HEDIS measures for this age range and several others not evaluated by HEDIS or NCQA methods that were, due to this study, identified as potentially valuable health indicators for this particular age range.  Evidence for preventive or reactive care for these types of diseases included inpatient and outpatient data linked to ICD-identified visits for COPD/Asthma, NIDDM, CHF, CVD, HTN, epilepsy, MS, a variety of chronic autoimmune syndromes, GERD, fibromyalgia, LBP, PVD, pre-CRF, etc.  For each of the chronic disease states, a standard NCQA approach is taken to evaluate success at a preventive or self-prescribed health management level.  As an example. to evaluate epilepsy history using this model of records review, one can assess consistency of script history and data, the number of ED or UC visits related to seizure-related complications, the number of preventive office visits engaged in on a per year basis, the numbers of screenings an individual goes through per year for such a condition (blood screenings, 1+EEG/year, counseling visits when necessary, etc.).  Likewise for evaluating NIDDM patients, utilization of various script information, office visit, self-measured glycemic levels, quarterly HBA1c, etc, can all be used to assess the state of health of this particular patient type.  Other measures focused upon for this age range are the standard HEDIS/NCQA measures–Cervical Cancer Screening and Breast Cancer Screening and several that are not a part of most NCQA activities such as Prostate Cancer visits.

It is expected that during this age range, several diseases may demonstrate a tendency to occur in the same individual.  In such cases, combined disease profiling methods can be used to measure and quantify the highest risk patients such as DM-HTN-HL cases, Epilepsy-Drug Abuse cases, Asthma-Allergy-COPD cases, and CVD with secondary chronic disease management of any shape or form.  A number of preventive screening or clinical assessment procedures carried out for patients in this age range may also be included in the clinical setting, and so recorded as part of a normal preventive care visit.  These may include such measures as QOL surveys, mental health evaluations for identified high risk/+ history cases, etc. etc.

The final two age groups evaluated using the given datasets involved 45 to 64 year olds and 65+.  The first group is the highest cost group and typically requires the most preventive and non-preventive visits, the most prescriptions, and the high inpatient and outpatient surgical and specialist visit demands.  Unlike the 24 to 44 yo age group, for which numerous preventive activities can be engaged in, this population poses the highest financial risk to the health acre system in general.  Some of the highest cost surgical procedures occur in these individual, and some of the longest periods of unemployment for health purposes can ensue when a medical problem begin to take center stage.  People already in these two age ranges have to undergo both an aggressive form of preventive and an equally aggressive form of reactive, palliative or QOL-related series of health care services.   Costs attributed to amounts of care and medical services for the second (65+) group can be at least double those of high-cost members of the younger group, with requirements for new high costs services such as 100 day LTC stays and such.

The preventive measures to be engaged in by the 45-64 yo group include a continuation of preventive visit activities such as HBA1c assessments, breast and cervical cancer screening activities (unless no longer eligible of in need due to medical history), PFT/COPD evaluations, etc..  For individuals still employed, these activities are most important in order to maintain significant levels of productivity and the related QOL in the workplace.    This is also the period in which Colorectal screening must be employed (preferably between 50 and 55 yo, or 52 and 57 yo depending upon the methods of measurement utilized for this review); in cases where Colorectal screening is not performed or avoided, other replacement methods can be used as well, such as Flexible Sigmoid, Barium Contrast procedure, or as a last measure, an annual FOBT.   Other evaluations involving this age group have to include continued evaluations of any chronic disease related states the individual has, such as monitoring for a worsening seizure or hyperglycemic events history.  Preventive screenings have to be engaged in such as lab tests designed to predict or delay onset of renal failure, onset of insulin dependency, or habituation of use of chronic pain relief prescriptions.

The most important activities to be monitored for 65+ members pertain to age-specific events such as pre-prostate and early prostate-cancer states, early to mid colorectal cancer states, COPD/URI high risk states, insulin dependency, QOL problems onset related to prolonged neurological conditions or disorders, glaucoma onset, the increased possibility of late age related diabetes onset, and the possibility of age related changes in behavioral and mental health conditions (Alzheimer’s, etc.).

Each of the above mentioned or inferred medical states, based on their ICD ID, can be evaluated for amounts of activity engaged in at a per member/patient level, and per PCP/specialist level.  These values do not necessarily indicate whether or not disease itself is occurring or preventive actions are being taken.  Instead, these values at the population (per patient/per service provider level) can be used to monitor the activity taking place within a given system.  In some cases, we might expect to see increases in certain screening practices ensue due to increased publicity of this practice or increase intervention activities.  Other times, increases may instead be due to general health related changes.  Typically, slowly progressing increases in reporting frequency may be attributed to increases in public awareness and education activities.  Sometimes these increases are seen in spike, and can mimic actual increases in incidence of a condition.  For the most part, infectious conditions are monitored and spikes can be recognized using this method of evaluating the entire dataset.  Other times, standardized rates can be developed for particular programs and program types, with the goal of increasing such things as engagement on a particular screening event (i.e. colorectal) or increasing your likelihood of engaging in a timely follow-up to a positive breast cancer screening outcome. 

In many cases, it is not the single event such as a claim- or hospital records identified visit that is not an indicator of any importance.  It is the combination of events that really tells us an entire story of what is taking place.  Whereas the use of a particular drug per person, such as an asthma remedy, is often measured at a per patient level, this measure alone is not an indicator of asthma-related or COPD-related risk.  Instead, it is the combination of events such as combined use of non-steroidal asthma and allergy medications versus steroidal and non-steroidal and allergy medications that defines the difference between high risk and low risk groups, and their numbers of patients.  Once the inpatient and outpatient data is included in this analysis, we can get a better count of people fully engaged in preventive activities, versus those just partially or barely engaged in such practices and behaviors.  The risk value used to assess a population is not so much a product of the drug information and isolated case history data as it is the results of an evaluation of the combination of preventive and reactive events that took place.

The DH database was tested for evaluating a little more than 300 disease states or conditions in association with pre-defined age groups, using the above methodologies and relying heavily upon NCQA/HEDIS information sources.  In essence it was found that specific ICDs and scripts could be used to evaluate specific indicator-related events that take place in a person’s life.  The numbers of these events may be used much like any incidence or prevalence value, with the exception that one must remember that these value define visits activity in many cases, and are not indicative of any actual diagnostic conclusions.  Nevertheless, this method can be used to assess a large populations health status and help define where certain preventive activities may have to be increase in frequency or improved in order to improve long-term health and cost-related outcomes.  When actual diagnostic and service related summaries can be made with this data, we can also assess whether or not the predictive methods developed based on case and visits data can be employed to meet program or population related needs and produce a measurable success.  This type of outcome can then be related to overall costs incurred preventively as well as the economic result of a series of health-related consequences brought on by a particular condition.

the following methodology has been established for relying primarily upon prescription information to develop a method of monitoring population health in a large health care system.  This method is used to evaluate internal, Perot systems, HEDIS and non-HEDIS related health information.  The following pages  define the methodologies applied to developing this type of population health monitoring tool.