A simulation model approach to analysis of the business case for eliminating health care disparities

David R Nerenz; Yung-wen Liu; Keoki L Williams; Kaan Tunceli; Huiwen Zeng

doi:10.1186/1471-2288-11-31

Research article
Open Access
Open Peer Review

A simulation model approach to analysis of the business case for eliminating health care disparities

David R Nerenz¹Email author,
Yung-wen Liu²,
Keoki L Williams¹,
Kaan Tunceli¹ and
Huiwen Zeng³

BMC Medical Research Methodology201111:31

https://doi.org/10.1186/1471-2288-11-31

Received: 19 August 2010
Accepted: 19 March 2011
Published: 19 March 2011

Open Peer Review reports

Abstract

Background

Purchasers can play an important role in eliminating racial and ethnic disparities in health care. A need exists to develop a compelling "business case" from the employer perspective to put, and keep, the issue of racial/ethnic disparities in health care on the quality improvement agenda for health plans and providers.

Methods

To illustrate a method for calculating an employer business case for disparity reduction and to compare the business case in two clinical areas, we conducted analyses of the direct (medical care costs paid by employers) and indirect (absenteeism, productivity) effects of eliminating known racial/ethnic disparities in mammography screening and appropriate medication use for patients with asthma. We used Markov simulation models to estimate the consequences, for defined populations of African-American employees or health plan members, of a 10% increase in HEDIS mammography rates or a 10% increase in appropriate medication use among either adults or children/adolescents with asthma.

Results

The savings per employed African-American woman aged 50-65 associated with a 10% increase in HEDIS mammography rate, from direct medical expenses and indirect costs (absenteeism, productivity) combined, was $50. The findings for asthma were more favorable from an employer point of view at approximately $1,660 per person if raising medication adherence rates in African-American employees or dependents by 10%.

Conclusions

For the employer business case, both clinical scenarios modeled showed positive results. There is a greater potential financial gain related to eliminating a disparity in asthma medications than there is for eliminating a disparity in mammography rates.

Keywords

Asthma
Mammography Screening
Emergency Room Visit
Business Case
Manage Care Plan

Background

Disparities in health among racial and ethnic groups in the US are well documented [1]. Although important exceptions can be found, a general pattern of poorer health (e.g., life expectancy, self-reported health status, incidence of disease) exists among members of racial or ethnic minority groups. One contributing factor to disparities in health is disparities in quality of health care; a similarly extensive literature documents disparities among racial and ethnic groups in terms of access to care and quality of care received [2, 3]. Disparities in quality refer to differences in levels of quality (based on defined measurable indicators) that have no clinical justification for members of different racial/ethnic groups [4].

Some disparities in health and health care reflect the combined effects of poverty and lack of health insurance [5]. Evidence of disparities in quality of care exists among employed and insured individuals, even among individuals in a single type of insurance or a single health plan [6–8]. Demonstration projects supported by the Commonwealth Fund and the Health Resources and Services Administration (HRSA) have shown that data on enrollees' race/ethnicity can be linked to health plan quality of care data to generate reports that identify disparities in quality of care at the individual health plan level [9–11].

Purchasers can play an important role in eliminating disparities by requiring health plans to analyze quality of care data separately by race/ethnicity and provide regular reports; and by using a variety of mechanisms inherent in the purchaser-supplier relationship (e.g., incentive payment systems) to focus attention on the disparity problem and change clinical practice [12, 13]. Thus a need exists to develop a compelling "business case" from the employer perspective to put, and keep, the issue of racial/ethnic disparities in health care on the radar screen [14]. A "business case" from the employer perspective inevitably will be somewhat different from one created from the health plan or provider perspective. Reductions in direct health care expenditures that follow from improved quality of care may or may not be relevant, for example, depending on whether the employer or a health plan is the at-risk entity in any specific time period.

In principle, the several components to a "business case for disparity reduction" from the employer perspective include, but are not necessarily limited to:

additional direct medical expenses attributable to disparities in quality of care;
costs of absenteeism, or lack of productivity while at work (sometimes called "presenteeism") related to poor health status that follows from health care disparities;
costs of absenteeism or diminished productivity related to receiving medical care for preventable conditions or complications;
costs of absenteeism or diminished productivity related to employees' obligations as caregivers for children or other family members (e.g., mother or father has to take day off from work when child with asthma has an exacerbation);
costs of training or hiring new staff when employees are disabled, retired, or die;
costs of specific interventions or general system changes designed to reduce or eliminate disparities;
potential additional costs of medical care as a result of eliminating disparities (e.g., additional costs of antidepressant medications among minority patients who were previously under-treated).

The elements of a business case for disparity reduction are similar to, perhaps even identical to, the elements of a business case for quality improvement. In both domains, the activities in question include data collection and analysis, identification of processes or outcomes in which quality is less than an ideal or target level, interventions of various types to improve quality, and additional cycles of data collection and analysis to determine whether quality has improved. Most of this work lies outside the usual routine of medical care for which payment is provided, so additional costs are incurred for which some sort of business case is required.

We illustrate both a method for calculating an employer business case for disparity reduction and a comparison of the business case in two clinical areas. We conducted analyses of the direct (medical care costs paid by employers) and indirect (absenteeism, productivity) effects of eliminating known racial/ethnic disparities in mammography screening and appropriate medication use for patients with asthma. We chose these two clinical domains because they represent areas in which disparities in specific quality of care measures have been documented and because a body of published data exist linking disparities in quality of care to medical care costs and data on absenteeism and productivity. We aimed to determine whether a positive business case exists for disparity reduction in either clinical case scenario in an effort to emphasize the importance for employers' ongoing participation in disparity reduction efforts.

Clinical Settings for Employer Business Case Modeling

Breast Cancer

Breast cancer is the second leading cause of cancer death among women. Each year, approximately 200,000 women in the US are diagnosed with breast cancer, and 40,000 will die of the disease [15]. Treatment can involve a difficult regimen of surgery, radiation therapy, chemotherapy, and reconstructive surgery for those women undergoing mastectomy. Even women who experience five-year disease-free survival have reduced quality of life during relatively extensive periods of treatment [16].

The probability of cure or extended remission of breast cancer depends on the stage of disease at diagnosis [17]. Mammography is part of a recommended set of breast cancer screening procedures (along with clinical breast exam and breast self-examination) whose main objective is diagnosis at the earliest possible stage of disease. Some variability exists among national guidelines for mammography frequency, but most guidelines recommend annual mammography for women over 50 [18]. The main benefit of increased mammography rates is in shifting the distribution of stage at diagnosis, and therefore the subsequent survival probabilities, among those women diagnosed with breast cancer [19]. The National Committee for Quality Assurance (NCQA) Health Employer Data and Information Set (HEDIS) for managed care plans includes a measure of mammography use, which is defined as the percentage of women between 50 and 65 who have been continuously enrolled in the plan for two years and have had at least one mammogram in the past two years [20].

There is some controversy about the extent to which there are meaningful racial/ethnic disparities in mammography rates. Recent national surveys have suggested that black, Hispanic, and white women all receive mammograms at approximately the same rate [21]. However, analyses of Medicare and other claims data sets typically show a disparity of 7-15 percentage points in mammography rates, with black and Hispanic women less likely to receive mammograms than non-Hispanic white women [22, 23]. These disparities have been found within some individual managed care plans, suggesting that at least some number of plans will find disparities in mammography rates when analyzing their own HEDIS data [24].

Eliminating a disparity in mammography rates would involve raising rates in black or Hispanic women to the level currently found for white women. For a typical health plan with relatively high current mammography screening rates, that might involve increasing the mammography rate for minority women from 70% to 80%. The question we addressed in this simulation was: What would be the benefits to an employer in terms of attendance and productivity at work, and medical care costs, for an employed population of African-American women if a typical disparity of 10% in mammography rates were eliminated?

Mandelblatt et al [25] used a simulation modeling approach to estimate the costs and benefits of improving breast cancer outcomes among African-American women through either improved screening or improved treatments after diagnosis. Their results showed a relatively modest benefit for improved screening in a hypothetical cohort of 40-year-old African-American women. We used a similar modeling approach, but the analyses were focused on the 50- to 65-year-old age group that is the defined denominator population for the HEDIS breast cancer screening measures, and the annual mammography patterns being modeled were selected to match those typically found among members of managed care plans. We used published data specific to African-American women on mammography rates, distribution of stage at diagnosis, survival by stage, and days worked to estimate the benefits of eliminating disparities in mammography rates between African-American and white women.

Asthma

Asthma is a common chronic disease affecting 10-20 million Americans in 2003 [26, 27]. (Variation in estimates depends on whether the specific measure is ever having had asthma, having had an asthma attack in the past 12 months, or report of currently having asthma.) Asthma is more prevalent in African-Americans and Hispanics than in non-Hispanic whites, so the burden of morbidity and mortality related to asthma occurs disproportionately in those groups [28]. Deaths from asthma are rare [29], but morbidity in the form of acute exacerbations and related days lost from work or school is common [30]. Data from national surveys and other studies suggest that members of minority groups experience more asthma-related lost work and school days than whites, even among those with comparable asthma severity levels [31, 32].

National guidelines since 1991 recommend use of inhaled preventive medications, including inhaled corticosteroids (ICS), as the preferred method of long-term management for patients with persistent asthma [33]. (Patients with only mild, intermittent asthma may be appropriately managed with acute, "rescue" medications as needed.) Based on these guidelines, NCQA has included a measure of Appropriate Medications for Asthma in its HEDIS measures since 2000 [34]. The measure is defined as the proportion of health plan members with persistent asthma who have been prescribed at least one of the approved preventive medications in the past year. The most recent average rate reported by the NCQA for commercial managed care plans publicly reporting their HEDIS data was approximately 70% in 2008, and the 90^th percentile rate for such plans was approximately 80% [35].

Data from a number of studies suggest that African-American children and adults are less likely to receive prescriptions for preventive medications (and use them) than their white counterparts [36, 37]. Since the medications are effective in all racial/ethnic groups and the national guidelines do not make different recommendations for different groups, disparities in use of recommended medications are presumed to be one of the key underlying reasons for disparities in morbidity and mortality [38, 39]. Improving rates of appropriate medication use in minority patients should result in a reduction in disparities in measures like lost work or school days, emergency room (ER) visit rates, hospital admission rates, or deaths [40].

Our goal in this project was to model the effects of improvements in appropriate medication rates for children, adolescents, and adults, on lost school or work days and on direct medical care costs [41, 42]. To link data on disparities in children's use of asthma medications to an employer business case, we made the assumption that an employed adult would have to take one day off work for every day in which a child had an asthma exacerbation requiring a physician office visit, an ER visit, or a hospital admission [43]. The model specifically focused on the effects of reducing or eliminating disparities in the HEDIS Use of Appropriate Medication for People with Asthma measure [44]. The modeling exercise asked the question of how many lost school or work days would be averted by increasing medication use rates in minority populations to those observed in non-Hispanic white populations. These benefits were estimated over time periods ranging from one to five years.

Methods

In calculating a business case for reduction of disparities in HEDIS measures, we used Markov simulation models to estimate the consequences, for defined populations of African-American employees or health plan members, of a 10% increase in HEDIS mammography rates or a 10% increase in appropriate medication use among either adults or children/adolescents with asthma. These improvements would correspond to elimination of disparities of the magnitude often observed for those two measures. Although we also report findings in terms of quality-adjusted life years (QALYs) gained as a measure of benefit to patients and society of reducing disparities, the primary endpoints in modeling the employer perspective were direct medical care expenses and cost of days missed from work. Our main analytic goal was to determine whether there was a positive business case for disparity reduction in either area, and if so, to determine whether the case was stronger in one clinical area than another.

Mammography Simulation Model

The basic approach used in this study was a Markov simulation model with annual cycles [45]. The model was set up to simulate the key events related to the diagnosis and treatment of breast cancer in a fixed cohort of African-American women aged 50-65, given different rates of mammography. All women in the simulation cohort started without cancer, and they either did or did not receive mammography in a given year. In each year, women may or may not have developed cancer; if cancer was diagnosed, the probability of advanced stage disease at diagnosis depended on whether mammography had been performed that year as part of an annual, biennial, or more infrequent schedule of mammography.

Women who did develop cancer in one year started the next year either in remission, with progressive disease, or dead, with probabilities depending on the stage at diagnosis during the year of diagnosis. Women who did not develop cancer in a given year started the next year without cancer and, again, either did or did not have mammograms. The model can be run for any desired number of years; for this study, the time period used was five years.

To assign values to the model, there were two health endpoints being modeled and two employment-related endpoints. In the mortality or survival analysis, the "dead" state was valued at 0 and all other states are valued at 1. For analysis of QALYs, each state was assigned a utility value between 0 and 1, based on values in published literature and a set of assumptions about how much of each year is spent in each state. (For example, a woman who dies of cancer in a given year does not die on the first day of the year, so she accumulates some QALY values during the part of the year in which she is alive.)

For analysis of employment endpoints, the probabilities of working and the average work loss days for those who are employed were estimated, on the basis of published literature, for each state in the model. Employment probabilities were estimated in an employment equation which included all the relevant factors affecting employment status - for example, breast cancer status, exogenous variables (age, race, education, etc.), and health insurance. Probit estimates were the most commonly used method to derive the probability of employment. The number of days lost from work due to undergoing treatment was also estimated. Then, monetary values of reductions in employment and increases in absenteeism for employers (e.g., cost to employer of replacing an employee missing a day of work due to illness) were calculated based on published data (see Table 1 for all model input parameters and sources).

Table 1

Input Parameters for Mammography Simulation Model

	Base Model Parameter	Range Used in Sensitivity Analysis	Source(s)
State Transition Probabilities
Annual Incidence of Cancer	.003		[64–67]
Probability of Annual, Biennial, and Sporadic Mammogram Patterns	HEDIS 70% - Annual = 0.197 - Biennial = 0.395 - Sporadic = 0.270 - None = 0.138		[59–63], authors' calculations
	HEDIS 80% - Annual = 0.242 - Biennial = 0.483 - Sporadic = 0.182 - None = 0.093
Increase in Probability of Mammography Following False Positive	.10 increase relative to probability in absence of false positive		[50]
Probability of Cancer Found, Given Screening Pattern	- Annual = .003 - Biennial = .003 - Sporadic = .005 - None = .008		[68, 69]
Distribution of Stage at Diagnosis with (W/O) Mammogram	Annual - In Situ = 0.13 - Stage I = 0.41 - Stage II = 0.37 - Stage III = 0.07 - Stage IV = 0.02		[70–76]
	Biennial - In Situ = 0.13 (0.04) - Stage I = 0.41 (0.32) - Stage II = 0.37 (0.43) - Stage III = 0.07 (0.13) - Stage IV = 0.02 (0.08)	In Situ = 0.04 - 0.21 Stage I = 0.32 - 0.57 Stage II = 0.16 - 0.43 Stage III = 0.04 - 0.13 Stage IV = 0.02 - 0.08
	Sporadic - In Situ = 0.17 (0.04) - Stage I = 0.29 (0.32) - Stage II = 0.40 (0.43) - Stage III = 0.09 (0.13) - Stage IV = 0.05 (0.08)
	None - In Situ = (0.04) - Stage I = (0.32) - Stage II = (0.43) - Stage III = (0.13) - Stage IV = (0.08)
Probability of Remission in Year of Diagnosis, Given Stage at Diagnosis	Stage I = 97.5% Stage II = 94.0% Stage III = 81.0% Stage IV = 66.0%		[73, 77]
Probability of Death from Other Causes	.01		[78, 79]
Probability of Continued Remission vs. Recurrent Disease, Given Stage at Diagnosis and Initial Remission	Stage I = 0.975 Stage II = 0.94 Stage III = 0.81 Stage IV = 0.66		[73, 77]
Probability of Death, Given Recurrent or Progressive Disease	0.79		[73, 77]
State Values
Utility of Well, Non-Cancer States	1.0		[80–86]
Utility of Newly Diagnosed Cancer, by Stage at Diagnosis	Stage I = 0.90 Stage II = 0.85 Stage III = 0.81 Stage IV = 0.81	Stage I = 0.80 - 0.98 Stage II = 0.80 - 0.98 Stage III = 0.80 - 0.98 Stage IV = 0.80 - 0.98	[80–86]
Utility of Undergoing Treatment, by Stage at Diagnosis	Stage I = -0.10 Stage II = -0.20 Stage III = -0.25 Stage IV = -0.25		[80–86]
Utility of Progressive Disease	0.40	0.30 - 0.69	[80–86]
Probability of Employment - no Cancer	1.00		Arbitrary - simulation assumes women employed
Probability of Employment in Year of Cancer Diagnosis, by Treatment Stage	In Situ = 0.86 Stage I = 0.70 Stage II = 0.60 Stage III = 0.50 Stage IV = 0.40		[88–94]
Probability of Employment in Year of Cancer Diagnosis, by Remission Stage	Stage I = 0.94 Stage II = 0.88 Stage III = 0.81 Stage IV = 0.75		[88–94]
Probability of Employment - Progressive Disease	0.56		[88–94]
Other Values
Costs of Initial Treatment, by Stage at Diagnosis	Stage I = $ 22,488.00 Stage II = $ 27,213.00 Stage III = $ 29,220.00 Stage IV = $ 31,476.00		[48]
Costs of Treatment - Recurrent or Progressive Disease	$ 33,000.00		[48]
Cost of Mammogram	$ 81.86		Federal Register, 2007
Cost of Followup after False Postive Test	$533		[106]
Hourly Wage for Time Missed from Work	$ 21.31		Department of Labor, May 2007 [46]

In general, we used estimates of absenteeism among women remaining employed, rates of women leaving work entirely, and medical care costs to employers for the six months following a new breast cancer diagnosis. All three are costs to employers that could be affected by increasing mammography rates and consequent shifts to earlier stage diagnoses. Based on the annual average wage of $21.31 per hour [46], the cost of lost productivity to the employer is the hours lost multiplied by hourly wage, assuming 8 hours worked per day and that absenteeism is the only cause of lost productivity.

When employees decide or must leave the workforce because of the severe illness, a turnover cost is incurred. The turnover cost to the employer comprises the expenses of recruiting and training new employees. Turnover costs generally include time and monetary costs to select, recruit and train a replacement, and lost productivity. A study based on the US Department of Labor estimated the costs to replace an employee to average 33 percent of the new hire's salary [47]. Assuming 2080 hours per year, using the estimated average hourly wage, the annual turnover cost is $14,627.

Medical costs of employed patients are also a burden for employers. In the model, costs for breast cancer care paid by private insurers (as estimated from available published literature) were presumed to represent costs to employers, since this would be literally true for self-insured employers, and true in a less direct sense for other employers as their insurance premiums would reflect recent costs of employees' cancer care. Initial and terminal phases of breast cancer treatments are the most costly [48].

Whenever possible, probabilities of events or utility values were obtained from studies of African-American women. The structure of the model is illustrated in Figures 1 and 2. The model was run for five annual cycles. All women started alive without cancer diagnosis in one of the four groups based on pattern of mammography, since the HEDIS denominator population definition excludes women with cancer diagnoses. Figure 1 shows the starting states for the model. At the beginning of the simulation, each woman in the simulation cohort was assigned to one of four groups based on her pattern of receipt of mammograms over multiple years - annual, biennial, sporadic, or none. The probabilities shown for each group reflect the probabilities of being in each group at the start of the simulation, for the 70% HEDIS rate scenario. Figure 1 also shows the set of states following each of the four "mammography pattern" groups.

Figure 1
**Mammography pattern groups and starting states of simulation model**.

Figure 2
**Breast cancer stages and treatments in simulation models**.

Figure 2 shows the initial options for women who received a mammogram in a given year under any of the three patterns (annual, biennial, sporadic) that would include the receipt of a mammogram during a five-year simulation period. (The same options were included in the model for women not receiving mammograms, but are not shown here.) The key feature of the model is the stage of cancer at diagnosis, which determines the subsequent probabilities of remission and survival. The triangle symbols at the right side of the figure signify that women in those branches return to the specific named states to start the next annual cycle of the model, or die and no longer continue in the simulation cycle.

The input parameters for the model are summarized in Table 1. The key parameters were: probability of annual, biennial, and less frequent mammography; the incidence of breast cancer in the population of employed African American women aged 50-65; distribution of stage at diagnosis given a particular pattern of mammography; probability of remission or progressive disease following initial treatment given stage at diagnosis; probability of continued remission vs. recurrence given remission after initial treatment; probability of death given recurrence or progressive disease; annual probability of death from other causes in African-American women aged 50-65; utilities assigned to various disease, treatment, or disease-free states; probability of employment given cancer diagnosed at specific stages; and medical care costs associated with cancer diagnosed at specific stages.

Sensitivity Analysis

Sensitivity analyses were conducted to determine how the life expectancy, QALYs, employment rates, working hours missed plus turnover costs and medical care costs change when the distribution of stage at diagnosis in the simulation model moves from the most "pessimistic" one (i.e., least benefit due to mammography) to the most "optimistic" one (i.e., greatest effect of mammography on stage). In addition to the base model described, four more versions of the model with different combinations of distribution of cancer stage from Table 1 were defined and used for sensitivity analysis. Based on the different combinations, models are considered as moving from the most pessimistic to the most optimistic models. The definitions of all five models are summarized in the Appendix.

The base model did not include the costs or effects of false-positive mammograms, but one recent review suggested that over 10% of mammograms will be read as suspicious for cancer and then followed up with biopsy or other tests, even though cancers are not found in the vast majority of these [49]. Two additional versions of the model were created - one with an explicit cost added for false positive tests requiring additional followup, and another with those costs included and a slight increase in the subsequent probability of mammography after a false positive test [50].

Other input parameters were not included in sensitivity analyses either because they had one widely accepted value (e.g., annual cancer incidence rate) or because they could be set arbitrarily to fit any specific new application of the model (e.g., increase in mammography rate; hourly rate for employees). The relationship between mammography and stage distribution at diagnosis seemed to be the relationship with greatest variation in published values and greatest potential impact on model outputs.

Asthma Basic Model Structure

The basic modeling approach was a Markov model simulating the likelihood of various events like acute exacerbations, ER visits or hospital admissions, lost school or work days, or deaths as a function of taking or not taking appropriate medications. The model (illustrated in Figure 3) was structured to capture the range of possible events during a month. Hypothetical patients in the model began each month either taking or not taking appropriate medications, and then experienced events during the month with probabilities that depended on whether or not they were taking medications. At the end of each month, patients "cycle back" and began a new month either taking or not taking appropriate medications. The model was run for 12, 24, 36, 48, or 60 monthly cycles to simulate experience over time periods from one to five years.

Figure 3
**Initial states for Markov model of appropriate medications for asthma**.

The model is similar in structure to that developed by Paltiel et al [51], but does not include forced expiratory volume (FEV1) as a core concept, because we were unable to find adequate data for that approach specific to individual racial/ethnic minority groups. When possible, parameters from the Paltiel model were used in our model to try to maximize comparability across the two modeling approaches.

The structure of the model is illustrated in Figures 3, 4, and 5. Figure 3 shows the first set of branches, or starting states. Figure 4 shows the next set of state transition options, with associated monthly probabilities in the situation where a child is not using appropriate medications. The same transition options exist in the model for children using appropriate medications, but with different probabilities. Figure 5 illustrates the complete set of transition states, for a child not using appropriate medications and experiencing an exacerbation requiring an ER or urgent care visit during the month. Similar sets of branches follow from other "exacerbation" branches, but with different probabilities.

Figure 4
**Second set of state transition options, illustrated for the initial state of not using appropriate medications**.

Figure 5
**Additional state transition options, given the occurrence of an exacerbation requiring an ER or urgent care visit**.

The primary input parameters for the model are summarized in Table 2. They included: probability of use of appropriate controller medications in a month, probability of exacerbations in a month given use or non-use of controller medications, probability of ER or urgent care visit or hospitalization given exacerbation, number of days missed from work given exacerbations of varying severity, medical care costs associated with treatment of exacerbations of varying severity, costs of controller medication [52], and utilities associated with specific states of exacerbations and freedom from exacerbations. The models were run for both adults and children; for children, days missed from school was used instead of days missed from work, but in analyzing days missed from work to calculate a business case from the employer perspective, it was assumed that an employed parent would miss a day of work for every day a child with asthma was out of school because of an exacerbation.

Table 2

Input Parameters for Asthma Medication Model

	Base Model Parameter	Range Used in Sensitivity Analysis	Source(s)
State Transition Probabilities
Probability of Asthma Medication Use	80%	70% (90%)	[37, 95, 96, 99]
Probability of Exacerbation not requiring ER visit or Admission	With Medication = 0.27 W/O Medication = 0.53	Same as 80%	[32, 95, 98]
Probability of Exacerbation Requiring ER Visit, not Admission	With Medication = 0.0382 W/O Medication = 0.0756	Same as 80%	[95, 98, 102–104]
Probability of Exacerbation Requiring Admission	With Medication = 0.0035 W/O Medication = 0.0069	Same as 80%	[49, 101]
Probability of Medication use Following Exacerbation	With Medication - no ER visit or Admission = 0.96 - ER visit, no Admission = 0.99 - Admission = 0.99	With Medication - no ER visit or Admission = 0.97 (0.99) - ER visit, no Admission = 0.99 (0.99) - Admission = 0.99 (0.99)	Authors' values, constrained by requirement that overall medication use be constant over entire simulation
	W/O Medication - no ER visit or Admission = 0.25 - ER visit, no Admission = 0.30 - Admission = 0.30	W/O Medication - no ER visit or Admission = 0.15 (0.4) - ER visit, no Admission = 0.25 (0.45) - Admission = 0.25 (0.45)
Probability of Medication Use Following Month of No Exacerbation	With Medication = 0.96 W/O Medication = 0.01	With Medication = 0.95 (0.965) W/O Medication = 0.01	Authors' values, constrained by requirement that overall medication use be constant over entire simulation
Probability of Death from Other Causes	0.0001	Same as 80%
State Values
Utility of Month without Exacerbation	With Medication = 0.0748 W/O Medication = 0.0756		[49]
Utility of Exacerbation not requiring ER visit or Admission	With Medication = 0.0706 W/O Medication = 0.0712		[49]
Utility of Exacerbation requiring ER visit, no Admission	With Medication = 0.0698 W/O Medication = 0.0704		[49]
Utility of Exacerbation requiring Admission	With Medication = 0.0649 W/O Medication = 0.0654		[49]
Days Missed per month from Work or School from Exacerbation not requiring ER visit or Admission	3 days		[95, 98]
Days Missed from Work or School from Exacerbation requiring ER visit, no Admission	4 days		[95, 98]
Days Missed from Work or School from Exacerbation requiring Admission	7 days		[33, 98]
Cost of Doctor Office Visit for Exacerbation	$155.44		[105], authors' calculation
Cost of ER Visit	$1,080.00		[105], authors' calculation
Cost of Admission	$13,512.00		[105], authors' calculation
Cost of Monthly Asthma Medications	$112.00		[50]
Hourly Wage for Time Missed from Work	$21.31		Department of Labor, May 2007 [46]

Patients are not consistent from month to month or year to year in their use of medications. Although a rate of adherence may be stable over time for a large population, individuals can alternate between periods of use and non-use, and the Markov model structure allows for this factor to be included. We presumed that medication use would be generally stable from month to month, but that some individuals taking medication and experiencing no exacerbations in a given month would not use medications in the following month (in various versions of the model we used proportions of 5-10%). Even some individuals experiencing ER visits or hospitalizations may not use medications in the following month (perhaps believing them to be ineffective); we estimated this proportion as 1%, with 99% of those experiencing ER visits or hospitalizations continuing on medications.

Among those not using medications in a given month, we presumed that 1% of those not experiencing exacerbations would begin to use medications in the following month, that 15% of those experiencing an exacerbation would begin to use medications in the following month, and that 25% of those requiring an ER visit or hospitalization would begin using medications in the following month.

In the absence of published data on these transitions among minority group patients with asthma, some of the transition probabilities are somewhat arbitrary. They were adjusted to produce the effect of gradually rising use of medication for the group as a whole over time. This trend of gradual increase is consistent with national data on improvement in the HEDIS asthma medication measure over time [9].