(C) 2006, 2010 Brian Altonen
What is ‘Groups’ Health?
“Groups Health” is a title I assigned to this project back in Late Winter- early Spring of 2004/2005. At the time I was developing methods for defining populations important to several PIP and QIA studies I was engaged in as a Quality Analyst. With the completion of my primary task using a national pharmacy database (Moderate to Severe Asthma), I turned to a review of other applications of this database for other projects, namely Diabetes. This led me to realize that the potential applications of this type of information were much more than my institution had expected. This realization in turn led me to design a new program in which a number of reviews could be carried out with this dataset, and applied to the numerous PIP and QIA studies underway.
A year or two earlier, I had developed a method for monitoring and reporting on population size for the entire membership of the various health systems related to where I worked as a QA analyst. I wrote a series of programs to access the institutional enrollment datasets at PEROT systems, performing headcounts in monthly-decimel (ages 1-3) and yearly increments. My next task thus became the design of a method to combined the outcomes of these two methods in order to develop a full population health review methodology.
The goals here were to develop a way to compare two datasets of population pyramids, and to develop an equation which could be used to define where the most significant change in prevalence was taking place. For example, for the smoking cessation studies I engaged in, the question that had to be answered was ‘does a change in habit take place more effectively in the 20-24, 25-29,30-34 . . . or 65-69, 70-74,75+ age group?’ And once an age range was understood to be the most effective age range for expecting changes to occur, at what specific year did this take place.
The problem presented here in terms of methodology was that we normally have some understanding as to what 5-year age range a group of people is most apt to show significant changes develop. But knowing the 5 year age range is sometimes not always that helpful. It ends up that people behave in such a way that a number ending with 0 or 5 rarely depicts the best age, in one year increments, for something to happen. A person doesn’t suddenly take on smoking at 15 or 20 years of age, and he/she doesn’t quit at 35 or 40 years of age, to use these as examples, not actual age values. An individual is most apt to take on smoking at a single year age sometime between 12 and 17, with men peaking at one age for large numbers first trying a cigarette and women peaking at another. Therefore, a new formula had to be developed to define that one year age when suddenly the bulk of the people take on or cease and particular unhealthy practice. This formula was fairly simple to imagine but a little difficult to develop and test.
Whereas the demographics review had already been a standard part of nearly all of my reports in recent years, the use of script data at a population level was a little more difficult to grasp the true values for. So my first uses uses of this data were focused on the identification of high incidence or prevalence rate conditions and their peak age bands, and high cost members and their primary cost-related dollar bands. As an example of the former, asthma patient ages were evaluated and peak 5- and 2-year age bands in terms of incidence/prevalence rates were defined. Cost evaluations involved the search for how to establish the boundaries between high, medium and low ranges, and a variety of ranges in between. The GIS systems allow for natural breaks and breaks defined by natural counts and age range distributions. In erratically behaving population systems, in which the number of 39 year olds in significantly different that the number of 40 year olds, the flow of transition of change that takes place from one age increment to the next is not as clear as it should be. A rolling average is often used to try and balance out this back and forth change in outcomes between high and low values, but this too softens the changes that are actually seen in the resulting rolling average graph that is produced.
Some statistical evaluation formulas have the ability to magnify the degree of difference thereby clearly pointing out the place where a significant change occurs. In some statistical equations, it makes a difference whether 39 or 40 is a peak in age related findings. It was these sensitive equations that were used to develop a formula that could be used to test for where the significant change actually occured in 1-year age increments. By using a rolling average method (known as moving windows in terms of evaluating satellite imagery), combined with a fairly sensitive statistical testing method, one can determine where the most important or largest impact or change is being made. This changes the previous 25 to 29 report of most important change to 28 years of age as the most significant one year of change in healhty of unhealthy practice habits. Knowing the exact year when these influences can be made can mean a significant difference in cost-related behaviors such as reducing the cost of mailing intervention letters to people in the right age group when the age group targeted is changed from 25 to 29, to 27 to 29–this represents a 40% reduction in overall costs and activities needed to engage in that intervention process. One has in theory just save the letter writers and institution mailing these letters approximately 40% of its costs.
The analysis of cost in raw dollars values is often a way to evaluate population health related performance as well. If we assumed that we are working with individuals whose overall scripts cost the system anywhere from >$100,ooo per year to greater than >$75,000 per quarter depending upon the reasons for this review, and wished to know how to reduce the highest cost practices, another method of population pyramid evaluations may be applied. The next task with my dataset then was to design a method of identifying the 20 to 50 highest cost members, in order to determine how and why these costs accrued and hopefully reduce these costs to some extent. In theory, in the end this could save the system more than a million dollars per year (according to one review of this actual work I performed at 25% of the state costs level, $3+M was identified as in need of future reductions.)
This analysis identified the source for these high costs, and required immediate administrative interventions. The steps that followed this line of review in turn also resulted in the implementation of some programs developed with the goal of utilization of high cost sources for drugs. Since there monthly script costing of as much as 40 times the normal costs incurred by other members with a similar prescription drug use history, a change in how these highest costs drugs were provided for patients was changed.
By then, it was apparent to me that the ways in which prescription drugs could be categorized, and evaluated for utilization rates and total costs, could play an important role in how we could better understand prescription drug use by particular population types. Did men utilize drugs differently than women? Were men more apt to engage in late refills? Were women more apt to skip a month for certain medication types due to side effects and related symptomatology?
By using the right categorization methods made available through the database provider, better techniques for developing a list of individuals with specific levels of disease history were implemented. I developed a way to rank and index individuals and assign them to specific classes of risk based on medication history, some times using specific forms of prescription drug use as specific health or disease monitoring indicators. The applications of this use of the information at the time was just limited to the studies ongoing about average costs per script per member per dose per day of use, a measure used to define the PIP for the next 3-year Medicaid study. This PIP was one of several dozen HEDIS measures I also reviewed throughout the year, looking for ways to develop lists of patients for whom interventions had to be developed or planned.
7 stages of change; age ranges for various HEDIS studies; core data assemblage
The next stage in this work came when the intranet became a source for overall systems data. The IT staff at the place where I worked placed all of the data for the past year onto the web in order for other offices to be able to access it and use it to carry out special research programs of their choosing. I decided to try to merge this data using Access with the prescription drug summary tables I could develop, and so came the next of research for me to engage in–the development of a population health status program in which slightly more than 300 types of illnesses or conditions were identified that could be monitored regularly to determine whether or not activities changed, such as increases in teen age visit claims due to STDs, changes in asthma ED and UC visits due to unknown causes, increases in childhood Upper Respiratory related cases, and reductions in preventive visits engaged in by patients with multiple comorbidities. One could also use this method to monitor the frequency of refills for certain health necessity medications, such as insulin renewals and refills for home glucose monitoring devices, and one could use this database to determine whether or not people were engaging in reduced utilization rates for high cost medications due to possible cost related problems.
In the end, it was determined that a number of HEDIS measures had to serve as base level information for defining the health of the various subsets of a population. These well-tested methods could be used to initiate this method of reviewing population health, but due to the limits in how the entire health of a population could be assessed using such a method, other types of measures had to be developed. For this reason, the intranet system data used to monitor visit, labs and claims information was evaluated along with prescription, and allied health script and non-script details when available, to measure the health of the various age-, gender- and ethnicity-defined subgroups identifiable with this information. These personal features combined with disease class information enabled me to define groups, many based on HEDIS guidelines, that could be monitored for change over time. Thus the GROUPS Health concept came to be and the name I assigned to my database used to perform this method of reviewing overall population health.
The methodology used for this project can be depicted using a simple flowchart. It was important to me that this type of work be manageable on a quarterly basis, although in the end it was found to be more helpful if reviewed on a triannual or biannual basis due to time limits in the workplace. Examples of some of the past work can be provided, in particular how the results are presented and how they should be interpreted. One of the major issues when dealing with claims and inpatient/outpatient data is that these reflect activities related to a disease or medical issue, but in no way represent incidence or prevalence in any way, shape or form. For this reason, I could not compare this information with any other data sources out there used for public health epidemiological surveillance reasons. For this reason, these methodologies of review Groups Health remain their own method of evaluating health, in need of constant comparison with other well established methods already tested and underway for more than a decade.
This method of evaluating population health has it applications to other programs already in place. In particular, by focusing on script and claims data, two sets of information that are available year round for certain overseers, methods of using this information for the good of the agency as well as the good of the population at large could be realized. Such is the purpose of this review. In the past ten or more years I have had the chance to review population health from the clinical side, QA side, and now educational side based on programs designed to inform health care providers about the ongoing changes in pharmacotherapeutics. I have also worked on this aspect of public health as an epidemiologist and GIS disease ecologist and population health analyst. The Groups Health method was developed in order to make better use of the information that has been accumulating and not being used in an effective manner by those in possession of this data. In the past, such data has served mostly internal purposes related to productivity, income and cost savings measures. Currently, there is enough information available in these systems to use this information in a highly respectable fashion, applying it to projects designed to have an impact on population health as a whole, intervening at one of the highest levels of potential impact that currently exist in standard evaluations of population health related levels.
In the past, the highest level of change in population health is considered to be events that lead to changes in population and personal health related statistics. The population level is typically viewed as a mega-level fashion of reviewing long-term effects of change. The ability of large organizations and corporations to engage in the practice of monitoring overall health in terms of millions and tens of millions of people opens the door for further intervention related activities to be developed. At the corporate level, these intervention-related studies are capable of serving as another income source. For the population in general, this adds an additional method for smaller groups such as insurance companies to be provided further input into the health fo their population as a whole, using a method of analysis that they themselves could engage in, but lack the manpower and resources to engage in such activities.
Groups Health measures fit in at Level 7 in the above table. Through Level 7 interactions, the individuals involved in these activities have the abilities to influence insurance companies, health care provider groups and even patients (indirectly) in such a way as to improve overall population health. The impact this level of engagement has on the overall program could include improvements in population health, to such an extent that certain PIP and QIA indicators, with measurable intervention related practices, could undergo change as well. The improved health practices that institutional and corporate based recommendations could result in, in the end, would require more frequent updates and changes in HEDIS and HEDIS-like measurements used to assess overall population health. This is the reason why I find the approach to understanding population by way of measuring prescription drug activities provides another way to impact public health at such a high level.