Michael Schatz, MD

 To develop and validate a predictive model for first severe chronic obstructive pulmonary disease (COPD) exacerbation using health insurance claims data and to validate the risk measure of controller medication to total COPD treatment (controller and rescue) ratio (CTR).

 A predictive model was developed and validated in 2 managed care databases: Truven Health MarketScan database and Reliant Medical Group database. This secondary analysis assessed risk factors, including CTR, during the baseline period (Year 1) to predict risk of severe exacerbation in the at-risk period (Year 2).

 Patients with COPD who were 40 years or older and who had at least 1 COPD medication dispensed during the year following COPD diagnosis were included. Subjects with severe exacerbations in the baseline year were excluded. Risk factors in the baseline period were included as potential predictors in multivariate analysis. Performance was evaluated using C-statistics.

223,824 patients. The greatest risk factors for first severe exacerbation were advanced age, chronic oxygen therapy usage, COPD diagnosis type, dispensing of 4 or more canisters of rescue medication, and having 2 or more moderate exacerbations. A CTR of 0.3 or greater was associated with a 14% lower risk of severe exacerbation. The model performed well with C-statistics, ranging from 0.711 to 0.714.

 This claims-based risk model can predict the likelihood of first severe COPD exacerbation. The CTR could also potentially be used to target populations at greatest risk for severe exacerbations. This could be relevant for providers and payers in approaches to prevent severe exacerbations and reduce costs.

To prevent worsening of chronic obstructive pulmonary disease (COPD) and its associated costs, identifying patients at high risk of a first severe exacerbation is critical.

A risk model for first severe COPD exacerbation was developed and validated using health insurance claims.

Total COPD treatment (controller and rescue) ratio (CTR) can predict first severe COPD exacerbation using health insurance claims.

COPD patients with a CTR of 0.3 or greater are at reduced risk of having a severe COPD exacerbation in the subsequent year.

Moderate exacerbation history, advanced age, use of rescue medication, chronic oxygen therapy, and type of COPD were also risk factors for first severe COPD exacerbation.

Chronic obstructive pulmonary disease (COPD) is the third leading cause of death in the United States, with more than 147,000 patients dying because of it in 2014.

 COPD poses a substantial economic burden on patients, healthcare systems, and society. In the United States, $32.1 billion in medical costs were associated with COPD in 2010, with costs projected to rise to $49.0 billion by 2020.

A major portion of COPD-related healthcare costs are attributable to exacerbations (sustained worsening of COPD from stable state, beyond normal day-to-day variations).

 Exacerbations are acute in onset and warrant additional treatment.

 Increased frequency of exacerbations leads to more hospital and emergency department (ED) visits.

 In 2009, approximately 740,000 hospitalizations were related to COPD.

 Hospitalization costs represent 85% of direct medical costs associated with COPD and are largely associated with exacerbations.

 Based on a Healthcare Cost and Utilization Project analysis, Perera et al showed that severe exacerbations, defined as hospitalizations for acute COPD exacerbations, were associated with increased aggregate costs, rising from $2.96 billion in 2006 to $3.47 billion in 2010.

 Furthermore, in a retrospective analysis of the Thompson Reuters MarketScan administrative database, Yu et al reported that COPD-related costs for patients with severe exacerbations were $7014 per 90 days compared with $658 per 90 days for patients with no exacerbations, with 83% of the costs associated with inpatient expenses.

 In another retrospective study, Abudagga et al reported that the costs associated with severe exacerbations were $18,120 and increased with each additional prior exacerbation.

Given the high costs associated with COPD exacerbations, it is important to identify patients at risk of exacerbation to target them for early treatment and improved prevention. Prior history of exacerbations is among the most important risk factors for the development of subsequent exacerbations.

 Previous clinical studies have identified additional risk factors for COPD exacerbations, with several groups developing exacerbation risk models.

 However, risk assessment models for COPD exacerbations utilizing readily accessible claims data that are available to health plans and healthcare groups, such as national quality-of-care organizations, are not available.

Identifying patients at risk for their first severe exacerbation is critical to target early and effective treatment, potentially slowing progression and associated cost increases. The present study was a secondary analysis from a wider study in which a medication ratio using pharmacy claims and a risk model were developed and validated. The primary results of this study have been published.

 In this secondary analysis, we aimed to construct and validate a predictive model of first severe COPD exacerbation event using health insurance claims, a source of data readily available to health plans and most quality-of-care organizations. This secondary analysis aimed to further validate the ability of the controller medication to total COPD treatment (controller and rescue) ratio (CTR) to predict first severe COPD exacerbation. The CTR is similar to the current Healthcare Effectiveness Data and Information Set (HEDIS) measure in asthma, the asthma medication ratio, and could have the potential to be used in a similar capacity.

 There were 2 objectives of the present analysis: 1) construction and validation of a risk model for first severe COPD exacerbation using patient demographics, comorbid conditions, and COPD treatment claims data as covariates; and 2) validation of the CTR as a measure of severe COPD exacerbation risk. The CTR is the ratio of total controller medications (inhaled corticosteroids [ICS], long&#8209;acting β

-adrenergic agonists [LABA], ICS + LABA, long&#8209;acting muscarinic antagonists, phosphodiesterase-4 inhibitors, and methylxanthines) to total controller plus rescue medications (short-acting β-agonists or short-acting muscarinic antagonists [SAMA]).

This retrospective study (GlaxoSmithKline study #HO-11-732) used the Truven MarketScan Commercial Claims and Encounter Database and Medicare Supplemental and Coordination of Benefits database (MarketScan database, Truven Health Analytics; Ann Arbor, Michigan) from 2006 to 2011 to construct and validate a risk model of prediction of first severe COPD exacerbation and to test the CTR measure. The Reliant Medical Group (Reliant database, Reliant Medical Group; Worcester, Massachusetts) was used for cross-validation.

The study start date (index date) was the second encounter with a COPD diagnosis (defined in study population section) within a year. Risk factors were assessed during the baseline period, which spanned 1 year, starting at the index date. This was followed by a 1&#8209;year at-risk period during which the occurrence of severe COPD exacerbations, the primary endpoint, was assessed (

Patients with COPD who were 40 years and older were included in the study. The index date of COPD status was defined as the second outpatient and/or ED encounter with a COPD diagnosis (

International Classification of Diseases, Ninth Revision, Clinical Modification

] codes 491.x [chronic bronchitis], 492.x [emphysema], or 496 [chronic airway obstruction]) within 1 year. Patients were also required to have at least 1 dispensing for an inhaled COPD medication or oral theophylline during the year following the index date and to have had continuous insurance eligibility (no more than 45 days without coverage) for at least 2 years following the index date.

Patients were excluded if they had a severe COPD exacerbation, defined as a COPD-related hospitalization during the baseline period, or at least 1 claim with an 

 diagnosis code for cancer (except cancer types that are typically considered to have little effect on lung function [eg, breast, prostate, or skin cancer except melanoma]) or other non-COPD lung disease during the baseline period.

The MarketScan database was used for model development and validation. This database includes health insurance claims data from 2006 to 2011, representing approximately 18 million covered individuals, and contains information on enrollment history and claims for medical (provider and institutional) and pharmacy services. The Reliant database was used for external validation. It includes longitudinal member-linked medical claims (physician and facility), pharmacy claims, enrollment records, laboratory results, and electronic health record information.

The primary endpoint of this study was the occurrence of a severe COPD exacerbation during the at-risk period (Year 2). A severe COPD exacerbation was defined as a hospitalization with either a primary diagnosis of COPD (excluding obstructive chronic bronchitis without exacerbation; 

code 491.20) or a secondary diagnosis for COPD with a primary diagnosis of respiratory failure (

 codes 518.81, 518.82, or 518.84). The choice of a 1-year window for the baseline and at-risk periods for the base case scenario made it possible to account for the seasonality of COPD exacerbations. Yet, given that risk factors were evaluated in the year prior to the at-risk period, they may not accurately reflect risk factors during the totality of the at-risk period. For example, the CTR may reflect more reliably the COPD medications patients received at the beginning of the at-risk period than those received toward the end of the at-risk period. As such, sensitivity analyses were conducted that limited the at-risk period to 90 days or 180 days following the baseline year (Year 1).

Candidate risk factors were assessed during the baseline period and used as predictors for severe exacerbation risk in the following year. Candidate risk factors were gender, age, region, insurance plan, H1N1 flu season (October 2009 to May 2010), pulmonologist visit, county characteristics (altitude; number of pulmonologists per 100,000 inhabitants; number of hospitals per 100,000 inhabitants; proportion of households below the low-income margin; median household income; proportion of patients without health insurance; proportions of high school dropouts and college graduates; urban/suburban/rural localization; and proportion of white, black, Hispanic, and Asian patients, derived from the Area Resource File

), type of COPD diagnosis, exacerbation history (moderate only [ie, defined as outpatient treated or ED visit for COPD with a dispensing for an oral corticosteroid within 7 days]), COPD medications (based on at least 1 dispensing), concomitant medications, procedures (flu and pneumococcal vaccines, use of chronic oxygen therapy, nebulizer, and spirometry; all based on Current Procedural Terminology and Healthcare Common Procedure Coding System codes), and comorbidities based on 

 Based on univariate associations among risk factors and the probability of an exacerbation in the at-risk period, risk factors were excluded if 

 tests for discrete variables and Wilcoxon tests for continuous variables. Risk factors present in less than 0.5% of the sample were excluded.

The choice of factors for the risk model was based on bootstrap resampling in combination with stepwise variable selection with logistic regression modeling, using a random sample of the MarketScan sample (development sample).

 A total of 1000 bootstrap samples were drawn; each was a random sampling with replacement from the original development sample, and each bootstrap replication was done using a different bootstrap sample. Within each bootstrap replication, all candidate risk factors were included, then stepwise elimination of risk factors was performed using the Akaike information criterion to identify predictors of exacerbations in the final model for the replication. Risk factors were chosen if they were selected in: 1) 100% of the 1000 bootstrap replications for the modeling of severe exacerbation or 2) at least 90% of the bootstrap replications for severe exacerbation and in 100% of the bootstrap replications for moderate or any type of exacerbations. CTR, which was defined as the number of canisters dispensed (converted to 30-day equivalent) for controller medication divided by the number of canisters dispensed for controller plus number of canisters of rescue medication (ie, SAMA, short&#8209;acting β

-adrenergic agonists [SABA], SAMA + SABA), was included in the final risk model. In order to assess the CTR as a predictive measure of first severe exacerbation, various ratio thresholds from at least 0.1 to at least 0.9 were tested. A cutoff of 0.3 was found to be the cutoff associated with the largest statistically significant risk reduction for severe exacerbations and has been shown to be the only cutoff associated with a statistically significantly reduced risk of any, moderate, and severe exacerbations in the previously published primary objectives of this study.

 Therefore, this ratio was used in the risk model of first severe COPD exacerbations.

The risk model was developed using the development sample, validated using the other random sample of the MarketScan data (validation sample), and cross-validated using the Reliant Medical Group data (cross-validation sample). Performance of the risk model was evaluated using C-statistics.

 Additional model analyses were performed on selected subpopulations, including those without a history of asthma and differing controller medication usage.

All statistical analyses were performed using SAS version 9.2 or newer (SAS Institute, Inc; Cary, North Carolina).

A total of 223,824 patients without history of severe exacerbations were included in the study, with 111,904 used in the development sample. Of the entire MarketScan sample of patients without history of severe exacerbations, approximately 80% experienced no exacerbations in the at-risk period (

). Patients who experienced exacerbations during the at-risk period tended to use more COPD medications and were more likely to use chronic oxygen therapy during the baseline period than patients who did not experience an exacerbation during the at-risk period (

Risk Factor Modeling for Severe Exacerbation

A total of 30 risk factors were included in the model; all were statistically significantly associated with severe exacerbations. The 5 risk factors with the largest impact on the probability of developing a severe exacerbation in the at-risk year were advanced age, use of chronic oxygen therapy, use of 4 or more canisters of rescue medication, having at least 2 prior moderate exacerbations, and type of COPD diagnoses received during the baseline period (

Patients with a CTR of 0.3 or greater had a lower risk of developing severe exacerbations (odds ratio [OR], 0.86; 95% CI, 0.77-0.96) (Table 2). In a subpopulation analysis, among patients with 4 or more fills for controller medications, a CTR of 0.3 or greater was associated with an even lower risk of severe exacerbation (OR, 0.76; 95% CI, 0.64-0.91) (

The model performed relatively well in predicting severe exacerbation risk, with C-statistics in the MarketScan development and validation and Reliant cross-validation samples of 0.713, 0.714, and 0.711, respectively (Table 2).

Most risk factors for severe exacerbation identified as most relevant using the 1-year at-risk period (eg, advanced age, use of chronic oxygen therapy, and having at least 2 prior moderate exacerbations in the baseline year) were also identified in sensitivity analyses using either a 90- or 180-day at-risk period (

]). Based on C-statistics, the performance of these models was similar to that of the model based on a 1&#8209;year at-risk period.

This report presents the development and validation of a risk model and the testing of the CTR for prediction of first severe COPD exacerbation using health insurance claims information. The main risk factors of first severe COPD exacerbation identified included advanced age, prior moderate exacerbation history, use of rescue medication, use of chronic oxygen therapy, and type of COPD diagnosis. A CTR of 0.3 or greater also was determined to predict first severe COPD exacerbation. The main advantage of the risk model and the CTR measure is that they do not rely on medical records, which are often unavailable to quality-of-care organizations and payers. Unlike previously reported measures, the CTR can be used to specifically assess the risk of first severe exacerbation and may help to identify which patients should be targeted for more intense management so that hospitalizations and costs could be reduced.

This model performed well in predicting the risk of first severe exacerbation at a population level using the MarketScan development and validation samples and the Reliant validation sample. Given the differences between the national MarketScan population and the regional Reliant database, the relatively consistent performance of the risk model in the Reliant population, with a C-statistic value of 0.711, indicates that the model and the CTR could be generalizable to heterogeneous populations.

Prior history of exacerbations is known to be among the most important risk factors for developing subsequent exacerbations.

 In an analysis of patients enrolled in the Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE) study, history of exacerbation was shown to be the best predictor of exacerbation, regardless of Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage.

 Previous incidence of exacerbations was also found to be an independent predictor of an exacerbation within 24 months (OR, 5.07; 95% CI, 2.55-10.07) in a study of 243 patients in the Netherlands.

 Our study confirmed that prior moderate exacerbations were a risk factor for a subsequent first severe exacerbation; however, a higher CTR value was associated with a reduced risk of first severe exacerbation regardless of exacerbation history. These results suggest that identifying patients at risk for their first severe exacerbation would be critical for targeting early and effective treatment, preventing worsening of the disease, and consequently reducing costs associated with COPD exacerbations.

There is a lack of published literature on risk factors for first severe COPD exacerbations. Müllerova et al evaluated the risk factors for a subcohort of patients with COPD from the ECLIPSE study who had experienced exacerbations requiring hospitalization for the first time during the 3-year study follow-up period.

 Significant risk factors identified were health status, COPD severity (based on forced expiratory volume in 1 second [FEV

]), increasing age, existence of emphysema or asthma, use of home oxygen, and elevated fibrinogen levels.

 Because data on COPD severity or fibrinogen levels were not available in the MarketScan data set, their relevance to exacerbation risk was not considered in this study.

Other studies have focused on the risk of severe exacerbation in general and did not model the risk of first severe COPD exacerbation.

 Garcia-Aymerich et al considered various clinical, medication compliance, lifestyle, and sociodemographic risk factors for COPD-associated hospitalizations.

 They found that previous COPD-associated hospitalizations, COPD severity (FEV

), under-prescription of long-term chronic oxygen therapy, and smoking were the most relevant risk factors for subsequent COPD-associated hospitalizations.

 In a follow-up study of patients from a clinical trial of tiotropium, Niewoehner et al identified increasing age, COPD severity, prior unscheduled ED visits or hospitalizations for COPD, cardiovascular comorbidities, and use of oral corticosteroids as significant risk factors for COPD-associated hospitalization.

Although previously published risk models of COPD exacerbations have focused on exacerbation frequency and COPD-associated hospitalizations, rather than the risk of first severe exacerbation, their findings were consistent with those of the current study in that factors such as increasing age and exacerbation history were highly relevant.

 Moreover, the performance of the claims-based model compares favorably with these models. In an exacerbation risk model developed by Bertens et al, in which exacerbation history, COPD severity, smoking, and vascular disease history were identified as risk factors, a C-statistics value of 0.66 was reported.

 In a model for frequent exacerbations published by Miravitiles et al, in which chronic mucus hypersecretion, COPD severity, and increasing age were associated risk factors, C-statistics values of 0.601 to 0.655 were obtained.

Certain methodologic limitations in this study should be noted. First, the identification of exacerbation events was based on definitional algorithms using health insurance claims data that have not been validated. These may have less than 100% sensitivity and specificity and may have misclassification errors for the outcome variables. Second, health insurance claims data are subject to inaccuracies and mistakes. In particular, an 

 diagnosis code on a medical claim is not an attestation that the patient has the diagnosis, because the code may represent a rule-out diagnosis or may be recorded incorrectly. This limitation applies to both the exacerbation events and the comorbidities used as risk factors in the models. Additionally, these records do not confirm that patients took the dispensed medication. A third study limitation was that included patients might have experienced severe exacerbations prior to the index date of this study. Finally, the covariates for the risk equations were limited to health insurance claims data only, which lack disease severity information (eg, GOLD stage). Nevertheless, using a subset of the Reliant sample, the performance of the risk model, including GOLD stage, was assessed and showed only a marginal improvement in incremental predictive accuracy.

Articles

A health insurance claims-based risk assessment tool to predict patients’ first severe chronic obstructive pulmonary disease exacerbation has been developed and validated.

Claims-Based Risk Model for First Severe COPD Exacerbation

To investigate differences in demographics, physician specialty, and medication use between patients who achieve high versus low ratios of controller to total asthma medications.

We used a Health Insurance Portability and Accountability Act—compliant claims database to identify patients aged 5 to 56 years with persistent asthma during a premeasurement year and a measurement year. Based on values in the measurement year, the ratio of controller to total asthma medications ratio was defined using the following formula: (Units of Controllers) / (Units of Controllers + Relievers). Descriptive analysis and multivariate logistic regression models were used to examine patients with high and low ratios.

The final study group comprised 38,538 patients with persistent asthma; 28,496 (73.9%) had high ratios. Specialty of usual-care physician differed (P <.001), with more high-ratio patients than low-ratio patients having an allergist or pulmonologist. Patients who received combination inhaled corticosteroid—long-acting β-agonist therapy (odds ratio [OR], 2.4) or leukotriene receptor antagonist therapy (OR, 3.5) were more likely to be in the high-ratio group compared with those dispensed a single inhaled corticosteroid. High-group and low-group assignment could be calculated by partial-year data: assignment based on 1 quarter of data was concordant with assignment based on full-year ratio in 91% of cases (Pearson product moment correlation coefficient, 0.864; κ statistic, 0.761), and assignment based on 2 quarters of data was concordant with full-year results in 94% of cases (Pearson product moment correlation coefficient, 0.928; κ statistic, 0.843).

A high ratio of controller to total asthma medications is associated with greater controller adherence and with more controller fills. The ratio can be calculated using 1 or 2 quarters of pharmacy claims data, at a time when intervention may reduce asthma-related exacerbations. Interventions that may improve the ratio include changing from single inhaled corticosteroid therapy and from asthma specialist care.

The ratio of controller to total asthma medications may be a useful tool for improving quality of care.

Patients with greater adherence to controller therapy are more likely to have high ratios.

Higher ratios are seen in patients treated with medications other than single inhaled corticosteroids.

Ratios may be calculated using less than a full year of data.

The Healthcare Effectiveness Data and Information Set (HEDIS), developed by the National Committee for Quality Assurance,

 reports quality measures for various health conditions. The asthma measure assesses the proportion of patients who are prescribed a controller medication in a given year, and it has become a focus of quality improvement programs nationally.

 The current measure may not be ideal for directing improvement efforts because patients with “appropriate” care (those with >1 controller medication prescribed) may actually have an increased risk of requiring emergency hospital care.

A revision to the measure that would address this issue has been proposed.

 This proposed measure, the ratio of controller to total (controller plus rescue) asthma medications, has been shown to be a better indicator of the need for emergency hospital care than the HEDIS measure.

 A higher ratio also is significantly related to improved asthmarelated quality of life, better disease control, and reduced symptoms.

 In addition to validating its association with relevant outcomes, studies

 of the ratio have demonstrated its performance among various plans and regions, tested different inclusion criteria, and examined the ideal cutoff point for high versus low ratios. However, little information is available regarding which patient populations perform well on this measure or what individual patient characteristics indicate poor performance. The type and number of controller and reliever medications that comprise the ratio have not been studied extensively. Having a better understanding of this ratio measure would improve plans’ (and physicians’) ability to identify patients at risk of having poorly controlled asthma and might help plans, through targeted intervention, improve their compliance with the measure.

The primary objectives of this study were to investigate differences between patients who achieve high versus low ratios of controller to total asthma medications, to identify patient characteristics that are associated with high ratios, and to compare the type and number of controller and reliever medications used by patients with high versus low ratios.

We conducted a cohort study to examine the characteristics of patients with high and low ratios of controller to total asthma medications. We used the PharMetrics database, a Health Insurance Portability and Accountability Act—compliant administrative claims database (

). This database contains adjudicated pharmacy and medical claims submitted by providers, healthcare facilities, and pharmacies. The study was exempt from human subjects review.

We included patients who had persistent asthma as defined by the current HEDIS measure during a premeasurement year (October 1, 2005, to September 30, 2006) and a measurement year (October 1, 2006, to September 30, 2007) who were aged 5 to 56 years during the measurement year and who were continuously enrolled during those 2 years. Patients with emphysema or chronic obstructive pulmonary disease (COPD) (

International Classification of Diseases, Ninth Revision, Clinical Modification 

] codes 492, 518.2, 491.2, 493.2, 496, and 506.4) were excluded.

The main outcome measure was the ratio of controller to total asthma medications, as described by Schatz et al.

 Patients with a ratio of at least 0.5 were classified as high-ratio patients, and those with a ratio of less than 0.5 were classified as low-ratio patients. Based on values in the measurement year, the ratio was defined using the following formula: (Units of Controllers)/(Units of Controllers + Relievers). Controller medications included cromolyn sodium, leukotriene modifiers, nedocromil, inhaled corticosteroids (including combination inhaled corticosteroid—long-acting β-agonists), methylxanthines, and omalizumab. Reliever medications included inhaled short-acting β-agonists. Long-acting β-agonists as individual drugs or as part of combination therapy were not counted in the numerator or the denominator for the ratio calculation.

To count medication units for oral medications, 1 U was considered equivalent to 1 dispensing event (for a 30-day supply). If a patient received a 90-day supply, it was considered 3 U. For inhalers, 1 U was 1 canister. For injected medications, 1 U was 1 claim, but if a subsequent claim had a service date within 21 days, it was ignored (eg, claims on days 1 and 15 counted as 1 U).

 We used HEDIS definitions to count dispensing events.

 For injected medications (omalizumab), each claim was counted as a dispensing event. The number of canisters dispensed in each claim was determined by a ratio of quantity to package size. For example, a claim of triamcinolone acetonide inhalation aerosol, 75 mcg/actuation (20-g package size), with a quantity of 20 was interpreted as 1 canister dispensed. Any claim with a ratio of quantity to package size of less than 1 was counted as 1 canister. For claims with a ratio of quantity to package size of at least 1, we rounded the number to a whole number of canisters. For example, claims of aerosol solution, 75 mcg/actuation (20-g package size), with quantities of 35 and 25 were counted as 2 canisters (35/20 = 1.75 [rounded to 2]) and 1 canister (25/20 = 1.25 [rounded to 1]), respectively. If a claim was for more than 12 canisters, it was truncated to 12 canisters.

We evaluated patient characteristics in the premeasurement year and in the measurement year. Characteristics evaluated in the measurement year included demographics (age, sex, and geographic region of care), specialty of usual-care physician, and controller or reliever use. To determine the specialties of the patients’ usual physicians, we reviewed office visit claims for evidence of evaluation and management services (defined as those with 

codes for office or other outpatient services, office consultations, and preventive medicine services). We identified the specialty of the physician with whom each patient had the most evaluation and management visits during the measurement year and considered that the patient’s usual-care physician.

 We reported extent of controller and reliever use in the measurement year and calculated controller medication possession ratios (MPRs), defined as the sum of the “days of supply” divided by the total number of days in the measurement year (365 days). If the days of supply exceeded 365, the total number of covered days was truncated to 365 days.

We assessed asthma control in the premeasurement year with variables that reflect impairment and risk using 2 key concepts described by the National Asthma Education and Prevention Program.

 To assess impairment, we measured whether patients had 6 or more short-acting β-agonist canisters dispensed in the premeasurement year. Such use has been shown to be a marker of impairment in previous investigations.

 To assess risk, we measured whether patients had the following: (1) at least 2 oral corticosteroid dispensing events, (2) any asthma-related (with primary diagnosis of asthma [ICD-9-CM code 493]) hospitalization, or (3) any asthma-related emergency department (ED) visits.

 We assessed impairment and risk in the premeasurement year rather than in the measurement year to understand the effect of baseline control on the ratio.

To describe patient clinical characteristics and chronic disease burden, we used claims in the premeasurement year and in the measurement year to capture more completely any relevant diagnoses. We evaluated clinical characteristics using individual and grouped diagnoses. We used Clinical Classifications Software, developed by the Agency for Healthcare Research and Quality (

http://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp

), to identify patients with mental disorders and disorders of the gastrointestinal system, conditions that may affect medication adherence and asthma severity.

 Individual diagnoses included reflux (ICD-9-CM codes 530.11 and 530.81), sinusitis (codes 461.x and 473.x), rhinitis (codes 477.0, 477.8, 477.9, and 472.0), acute upper respiratory tract infection (code 465.x), and nasal polyposis (code 471.x). To account for the burden of chronic health conditions, we applied the method by Hwang et al

 to assess the number of chronic health conditions each patient had.

We conducted a descriptive analysis to compare demographic and clinical characteristics, specialty of usual-care physician, and medication use between the high-ratio and low-ratio groups. In addition to high-ratio and low-ratio groups, we further divided patients into 4 groups based on their actual ratios (<0.25, 0.25 to <0.5, 0.5 to <0.75, and >0.75) and compared their characteristics. We studied quartiles for a more precise evaluation. We conducted bivariate analyses comparing selected characteristics between the high-ratio and low-ratio groups. Characteristics that were significantly associated with the ratio were included in the multivariate models as independent variables. We compared ratios calculated using the first quarter, the first 2 quarters, and the first 3 quarters of data with full-year ratios using Pearson product moment correlation coefficient and κ statistic.

We used 2 multivariate logistic regression models to estimate the association between certain characteristics and a high ratio. The first model excluded patients with a zero ratio (no controller use). It also excluded patients treated with only theophylline or cromolyn sodium, as these drugs are not commonly used as single agents in current practice (<1% of patients in the study). The second model included patients with no controller use. We reported adjusted odds ratios (ORs) and their 95% confidence intervals (CIs). All tests were 2-sided with a significance level of .05. All data transformations and statistical analyses were performed using SAS version 9.1

We identified 150,903 patients with persistent asthma in the premeasurement year and 156,649 in the measurement year. After excluding those who did not meet the age criteria, those who were not continuously enrolled, those who had evidence of emphysema or COPD, and those who did not use controllers or relievers in the measurement year, 38,538 patients remained (

Of the final study cohort, 28,496 (73.9%) were classified as having high (>0.5) ratios. The mean age did not differ between the groups with high versus low ratios. Age group, sex, and geographic region of care differed (

 <.001), with a higher proportion of high-ratio patients aged 5 to 9 years (14.1% vs 8.7%), female (55.2% vs 50.5%), and from the South (18.3% vs 13.6%). Specialty of usual-care physician in the measurement year differed (

 <.001), with a higher proportion of high-ratio than low-ratio patients having an allergist (7.6% vs 3.9%) or a pulmonologist (2.9% vs 2.2%) (

The mean MPR in the measurement year was 0.603 for high-ratio patients and 0.203 for low-ratio patients (Table 1). High-ratio patients used a mean of 10.3 U of controllers; low-ratio patients used a mean of 3.1 U (

 <.001). Controller medication use differed between groups (

 <.001): 7.7% of high-ratio patients used 0 to 3 U (lowest quartile) compared with 67.9% of low-ratio patients, and 35.3% of high-ratio patients used at least 12 U (highest quartile) compared with 4.1% of low-ratio patients. There were significant differences in reliever use during the measurement year (

 <.001): the mean number of reliever units used was 2.4 in the high-ratio group compared with 10.1 in the low-ratio group. Almost 30% of high-ratio patients had no reliever fills in the measurement year (0% of low-ratio patients had no reliever fills, as by definition having no reliever use produces a ratio of 1.0). The highest quartile of reliever use included those with 7 or more canisters used. Just over 8% of high-ratio patients and 63.3% of low-ratio patients were in this group.

There was a statistically significant difference between the groups in asthma control measures during the premeasurement year. High-ratio patients had less impairment, with only 16.4% filling 6 or more short-acting β-agonist canisters compared with 63.3% of low-ratio patients (

). High-ratio patients had lower risk than low-ratio patients in each of the areas measured (

 <.001 for each category). High-ratio patients had more total chronic health conditions (4.0 vs 3.6, 

 <.001) and more of the individual comorbidities assessed, except for mental disorders, which were more common among low-ratio patients (

). When we further classified patients into 4 ratio groups (<0.25, 0.25 to <0.5, 0.5 to <0.75, and >0.75), the results for medication use, asthma control, and comorbidities were similar (data not shown).

High-group and low-group assignment could be calculated using partial-year data. High-ratio vs low-ratio group assignment based on 1 quarter of data was concordant with group assignment based on full-year ratio in 91% of cases. Adding a second quarter of data increased concordance to 94%. Pearson product moment correlation coefficient of partial-year to full-year ratio (using a continuous [0-1.0] variable for the ratio) was 0.864 for 1 quarter and 0.928 for 2 quarters. κ statistic, calculated using the categories of high ratio and low ratio, was 0.761 for 1 quarter and 0.843 for 2 quarters (

The 2 multivariate models produced similar results. The model that excluded patients without controller use allowed us to estimate the effect of specific patterns of medication use on the ratio. In multivariate analysis of controller users, older age and male sex significantly lowered the odds of being in the high-ratio group, with an OR of 0.54 (95% CI, 0.48-0.61) among patients aged 10 to 17 years compared with patients aged 5 to 9 years and an OR of 0.69 (95% CI, 0.61-0.77) among patients aged 18 to 56 years compared with patients aged 5 to 9 years. Geographic region of care was associated with ratio group, with an OR of 1.14 (95% CI, 1.01-1.30) for the Northeast compared with the West and an OR of 1.18 (95% CI, 1.03-1.34) for the South compared with the West. Compared with patients having primary care physicians (internists and family physicians), patients having allergists as usual-care physicians had an OR of 1.35 (95% CI, 1.17-1.55) for being in the high-ratio group, and patients having pulmonologists had an OR of 1.26 (95% CI, 1.04-1.52) (

There was a statistically significant association between the type of controller used in the measurement year and the ratio group. Patients whose only controller was a combination inhaled corticosteroid—long-acting β-agonist product had 2.36 (95% CI, 2.15-2.58) times the odds of being in the high-ratio group compared with patients who used only single inhaled corticosteroids (Table 5). Similarly, patients with a leukotriene receptor antagonist as their only controller had an OR of 3.49 (95% CI, 3.15-3.87) for being in the high-ratio group compared with patients who used a single inhaled corticosteroid controller. Users of multiple controllers had an OR of 7.45 (95% CI, 6.80-8.17) for being in the high-ratio group.

Asthma control was associated with outcomes. Patients who filled 6 or more short-acting β-agonist canisters in the premeasurement year, a measure of impairment, had significantly lower odds (OR, 0.11; 95% CI, 0.10-0.12) of having high ratios than patients who filled fewer than 6 canisters (Table 5). Having any asthma-related hospitalization, any asthma-related ED visit, or having filled at least 2 oral corticosteroid prescriptions in the premeasurement year were all significantly associated with lower odds of having a high ratio. Patients with a mental disorder had lower odds of being in the high-ratio group (OR, 0.75; 95% CI, 0.70-0.81), whereas patients with rhinitis had higher odds (OR, 1.22; 95% CI, 1.14-1.31). Each additional chronic health condition increased the odds of being in the high-ratio group (OR, 1.04; 95% CI, 1.02-1.05).

Limitations and problems with the current HEDIS asthma measure have led to consideration of some modifications. One potential modification is the use of a ratio of controller to total asthma medications as the primary quality measure. Using a ratio that is associated with outcomes may result in improvements in the quality of asthma care, and the measure identifies a subpopulation that may benefit from attempts at improvements in care. In a study

 using 3 separate databases, a high ratio was consistently associated with improved outcomes. The proportion of patients with a ratio of at least 0.5 varied from 70% to 75% across the 3 databases. High ratios are also associated with improved patient-reported outcomes,

 but little has been published about specific differences between high-ratio and low-ratio patients or about the stability of the measure over time.

In the present study, 73.9% of patients had high ratios, similar to other published studies

; the mean age and the slight female preponderance were also similar to prior work. In an unadjusted analysis, children younger than 10 years, female patients, and those with allergists and pulmonologists as their usual-care physician were more likely to have high ratios, and these results persisted after adjustment for baseline characteristics. Other variables that had a statistically significant association with the high-ratio group in the multivariate analyses included lack of markers of impairment or risk, use of controllers other than single inhaled corticosteroids, absence of mental disorders, presence of rhinitis or acute upper respiratory tract infection diagnosis, and more chronic health conditions.

In multivariate analyses, the strongest variable associated with ratio group was the type of controller therapy, with single inhaled corticosteroids being associated with lower ratios than combination inhaled corticosteroid—long-acting β-agonist, leukotriene receptor antagonist, and multiple controllers. Patients are more adherent with combination inhaled corticosteroid–long-acting β-agonist inhalers than with concurrent treatment using separate inhaled corticosteroid and long-acting β-agonist inhalers.

 In addition, combination therapy is generally more effective than similar doses of inhaled corticosteroid monotherapy in relieving symptoms and improving lung function,

 so patients might be more consistent in their use of these therapies. Patients receiving long-acting β-agonist therapy may use less short-acting β-agonist, which may increase the ratio for combination products compared with single inhaled corticosteroid. However, if the reduced use of short-acting β-agonist reflects decreased symptoms, this is not an artificial increase. Adherence to leukotriene receptor antagonist therapy is also greater than adherence to inhaled corticosteroid monotherapy.

 Patients treated with multiple controllers were many times more likely to have high ratios, possibly as a result of the calculation used to determine the ratio. To be categorized as using multiple controllers, a patient needed at least 2 controller fills (eg, to identify >2 types of controllers), whereas some users of single therapies had only 1 fill, and a greater number of controller fills results in a higher ratio.

Ratios also increase with increasing controller adherence. This relationship between greater adherence as measured by the MPR and higher ratios may explain some of the associations we observed. Higher ratios were seen in patients who had allergists or pulmonologists as their usual-care physician. Provider knowledge and communication skills influence adherence to treatment, and specialist providers may be better at communicating complex regimens to their patients with asthma.

 Women had higher ratios, and some evidence suggests they may have better adherence than men.

 The presence of a mental disorder was associated with a lower ratio, and some mental disorders, including depression, decrease adherence to therapy.

 Many other factors associated with poor adherence were not captured in our study, such as patients’ education, socioeconomic status, disease understanding, and belief in treatment efficacy.

Some variables may mediate their effect on ratio group through their association with asthma control. Increasing numbers of chronic health conditions, as well as rhinitis and upper respiratory tract infections, all may identify patients with greater risk. These factors may also lead to more office visits and greater opportunities for prescribing controller medications. We used utilization-based measures to adjust our results for disease severity, but there were no clinical measures of severity in our database. This was a limitation of our analysis.

The goal of quality measures and tools for targeted interventions is to improve patient outcomes. Quality measures are provider oriented and aim to improve patient outcomes by improving provider adherence to treatment guidelines (eg, controllers for asthma). This is done by using quality measures as incentives for approaches such as public reporting and pay for performance. In contrast, population management (tools for targeted interventions) is patient oriented and aims to identify high-risk patients who need targeted interventions to improve outcomes. The ratio, which may be calculated from partial-year data, can be used as a quality measure or tool for population management to improve asthma outcomes. A plan interested in intervening for patients with low ratios could use data from 2 quarters or even 1 quarter to identify patients at risk for having a low ratio (if the quantities dispensed are in the 30-day to 60-day range). Interventions such as education, outreach, or medication counseling put in place after a partial-year

calculation of the ratio might not only increase the ratio but also (given the demonstrated link between the ratio and outcomes) reduce emergency hospital care. We found that patients who used combination inhalers versus single inhaled corticosteroid controllers and leukotriene receptor antagonist versus single inhaled corticosteroid were more likely to have high ratios. Using this information, interventions to improve patients’ ratios could be developed. For example, patients with poor control who are receiving single inhaled corticosteroids could switch to combination inhalers, and patients with poor adherence to single inhaled corticosteroids could switch to leukotriene receptor antagonists.22 Patients who received specialty care were more likely to have high ratios; therefore, it may be beneficial to refer patients with evidence of prior impairment or risk to asthma specialists. Studies to evaluate the effectiveness of these interventions should be considered.

The ratio of controller to total asthma medications may be useful as a measure of the quality of asthma care. The measure can be easily calculated using existing administrative data, and patients with higher ratios have improved asthma outcomes. Higher ratios are seen among patients with greater adherence to therapy and among patients treated with medications other than single inhaled corticosteroids. Ratios are stable; interventions focused on patients with low ratios based on 1 or 2 quarters of data may reduce asthma-related exacerbations. The ratio of controller to total asthma medications may be useful as a quality measure and as a tool for targeted intervention to improve asthma outcomes.

From the Partnership for Health Analytic Research (MSB, EC); Beverly Hills, CA; AstraZeneca, LP (BG, DM), Wilmington, DE; and Kaiser Permanente Medical Center (MS), San Diego, CA.

This study was funded by AstraZeneca, LP.

Drs Broder and Chang report receiving payment from AstraZeneca, LP for their involvement in conducting the study and preparing the manuscript. Drs Gutierrez and Meddis are employees of AstraZeneca, LP, the company that provided funding for the study. Dr Schatz reports receiving investigator-initiated grants from Aerocrine, Genentech, GlaxoSmithKline, and Merck.

Concept and design (MSB, BG, EC, DM, MS); acquisition of data (BG); analysis and interpretation of data (MSB, BG, EC, MS); drafting of the manuscript (MSB, BG, DM); critical revision of the manuscript for important intellectual content (MSB, BG, DM, MS); statistical analysis (EC); and obtaining funding (DM).

David Meddis, PhD, AstraZeneca, LP, 1800 Concord Pike, Wilmington, DE 19850. E-mail: 

1. National Committee for Quality Assurance. HEDIS 2008: Healthcare Effectiveness Data and Information Set. Vol 2, Technical Specifications. Washington, DC: National Committee for Quality Assurance (NCQA);2007:126-128. 2. Gelfand EW, Colice GL, Fromer L, Bunn WB III, Davies TJ. Use of the Health Plan Employer Data and Information Set for measuring and improving the quality of asthma care. Ann Allergy Asthma Immunol. 2006;97(3):298-305. 3. Berger WE, Legorreta AP, Blaiss MS, et al. The utility of the Health Plan Employer Data and Information Set (HEDIS) asthma measure to predict asthma-related outcomes. Ann Allergy Asthma Immunol. 2004;93(6):538-545. 4. Fuhlbrigge A, Carey VJ, Adams RJ, et al. Evaluation of asthma prescription measures and health system performance based on emergency department utilization. Med Care. 2004;42(5):465-471. 5. Schatz M, Zeiger RS, Vollmer WM, et al. The controller-to-total asthma medication ratio is associated with patient-centered as well as utilization outcomes. Chest. 2006;130(1):43-50. 6. Schatz M, Nakahiro R, Crawford W, Mendoza G, Mosen D, Stibolt TB. Asthma quality-of-care markers using administrative data. Chest. 2005;128(4):1968-1973.7. Schatz M, Stempel D; American College of Allergy, Asthma and Immunology Task Force; American Academy of Allergy, Asthma and Immunology Task Force. Asthma quality-of-care measures using administrative data: relationships to subsequent exacerbations in multiple databases. Ann Allergy Asthma Immunol. 2008;101(3):235-239. 8. OÃƒÂ¢Ã¯Â¿Â½Ã¯Â¿Â½Malley AS, Pham HH, Schrag D, Wu B, Bach PB. Potentially avoidable hospitalizations for COPD and pneumonia: the role of physician and practice characteristics. Med Care. 2007;45(6):562-570. 9. National Heart, Lung, and Blood Institute, National Institutes of Health, US Department of Health and Human Services. Expert Panel Report 3 (EPR3): guidelines for the diagnosis and management of asthma. Revised August 28, 2007. http://www.nhlbi.nih.gov/guidelines/ asthma/asthgdln.htm. Accessed February 17, 2009. 10. Schatz M, Zeiger RS, Vollmer WM, et al. Validation of a ÃƒÅ½Ã‚Â²-agonist long-term asthma control scale derived from computerized pharmacy data. J Allergy Clin Immunol. 2006;117(5):995-1000. 11. Schatz M, Cook EF, Nakahiro R, Petitti D. Inhaled corticosteroids and allergy specialty care reduce emergency hospital use for asthma. J Allergy Clin Immunol. 2003;111(3):503-508. 12. Schatz M, Nakahiro R, Jones CH, Roth RM, Joshua A, Petitti D. Asthma population management: development and validation of a practical 3-level risk stratification scheme. Am J Manag Care. 2004;10(1):25-32. 13. Cowen ME, Dusseau DJ, Toth BG, Guisinger C, Zodet MW, Shyr Y. Casemix adjustment of managed care claims data using the clinical classification for health policy research method. Med Care. 1998;36(7):1108-1113. 14. Hwang W, Weller W, Ireys H, Anderson G. Out-of-pocket medical spending for care of chronic conditions. Health Aff (Millwood). 2001;20(6):267-278. 15. Marceau C, LemiÃƒÆ’Ã‚Â¨re C, Berbiche D, Perreault S, Blais L. Persistence, adherence, and effectiveness of combination therapy among adult patients with asthma. J Allergy Clin Immunol. 2006;118(3):574-581. 16. Greenstone IR, Ni Chroinin MN, Masse V, et al. Combination of inhaled long-acting ÃƒÅ½Ã‚Â²2-agonists and inhaled steroids versus higher dose of inhaled steroids in children and adults with persistent asthma. Cochrane Database Syst Rev. 2005;4:CD005533. 17. Jones C, Santanello NC, Boccuzzi SJ, Wogen J, Strub P, Nelsen LM. Adherence to prescribed treatment for asthma: evidence from pharmacy benefits data. J Asthma. 2003;40(1):93-101. 18. World Health Organization. Adherence to Long-Term Therapies: Evidence for Action. Geneva, Switzerland: World Health Organization; 2003. 19. DiMatteo MR. Variations in patientsÃƒÂ¢Ã¯Â¿Â½Ã¯Â¿Â½ adherence to medical recommendations: a quantitative review of 50 years of research. Med Care. 2004;42(3):200-209. 20. DiMatteo MR, Lepper HS, Croghan TW. Depression is a risk factor for noncompliance with medical treatment: meta-analysis of the effects of anxiety and depression on patient compliance. Arch Intern Med. 2000;160(14):2101-2107. 21. Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005;353(5):487-497. 22. Tan H, Sarawate C, Singer J, et al. Impact of asthma controller medications on clinical, economic, and patient-reported outcomes. Mayo Clin Proc. 2009;84(8):675-684.

The ratio of controller to total asthma medications can be calculated using 1 or 2 quarters of data; high versus low ratios differentiate patient characteristics.

Ratio of Controller to Total Asthma Medications: Determinants of the Measure

: To identify an optimal cut-point score on the Mini-Asthma Quality of Life Questionnaire (mini-AQLQ) to predict subsequent asthma exacerbations, and to determine the additional risk conferred by a prior history of acute episodes.

: Cross-sectional survey linked to administrative records.

: A total of 1006 HMO patients with active asthma completed surveys that included the mini-AQLQ and prior-year history of acute episodes. Surveys were linked to administrative data that captured asthma emergency department and hospital care (emergency hospital care) for the year after the survey. Optimal mini-AQLQ cut-point scores were determined by stepwise logistic regression analyses using subsequentyear asthma emergency hospital care as the outcome and various mini-AQLQ cut-points as the predictors. Predictive properties of the 2 risk factors (mini-AQLQ cut-points and prior acute episodes) were determined.

: A mini-AQLQ cut-point of 4.7 was most significantly associated with subsequent exacerbations in patients without a history of prior acute episodes. The presence of either a mini-AQLQ score <4.7 or a history of prior acute episodes provided high sensitivity (90.4%) and identified a group nearly 6 times more likely to require emergency hospital care than patients with neither risk factor. The presence of both risk factors provided high specificity (79.2%) and resulted in a risk ratio of 9.5 compared with the absence of both risk factors.

: Asthma-specific quality of life and a history of acute episodes can be used together to identify patients with clinically meaningful higher and lower risks of subsequent acute exacerbations.

Asthma-specific quality of life and prior history of exacerbations can be used to define future risks of asthma exacerbation. 

Either a score of <4.7 on the Mini-Asthma Quality of Life Questionnaire or a history of prior acute episodes provided high sensitivity (90.4%) and identified a group nearly 6 times more likely to require emergency hospital care than patients with neither risk factor. 

The presence of both risk factors provided high specificity (79.2%) and resulted in a risk ratio of 9.5 compared with the absence of both risk factors. 

Such information could be useful for determining baseline risks in clinical trials as well as for individual and population asthma management.

Asthma is a common medical problem that causes substantial morbidity, including hospitalizations and unscheduled physician and emergency department visits.

 Moreover, chronic asthma is associated with very large direct and indirect costs, with a large portion of the direct costs attributable to emergency hospital care.

Several studies have shown that asthma-specific quality of life is associated with the subsequent risk of emergency hospital care.

 For this information to be clinically useful, though, a cut-point score on the quality-of-life measure needs to be defined, so that patients can be classified meaningfully according to higher and lower risk for subsequent exacerbations. None of the prior studies have attempted to establish the optimal quality-of-life cut-point score that maximizes both sensitivity and sensitivity in predicting subsequent emergency hospital care.

In addition, to use quality of life clinically to predict subsequent emergency hospital care, its relationship to other clinically defined risk factors should be clarified. Prior acute episodes have been shown to be the strongest predictor of subsequent acute episodes.

 Three studies have shown that asthma quality of life is independent of prior acute episodes in predicting subsequent emergency hospital care,

 but these studies have not provided a clinically applicable algorithm that incorporates both prior utilization and asthma-specific quality of life. Moreover, 2 of the prior studies defined prior utilization based on computerized data, which may not be available in all clinical settings.

The purpose of this study was to use quality-of-life survey data as well as other information obtained from the patient to develop a practical algorithm for identifying patients at increased risk of subsequent emergency hospital care. This involved identifying a quality-of-life cutpoint score that optimally predicts subsequent emergency hospital care in patients with and without a history of unscheduled asthma visits in the past year and determining the additional risk conferred by a prior history of acute episodes.

Patients in this study have been previously described.

 Surveys were sent by the Kaiser Permanente Care Management Institute in August 2000 to a random sample of Kaiser Permanente Medical Care Program adult members aged 18-56 years from the Northern California (n = 3072) and Northwest (n = 543) regions who were diagnosed as having persistent asthma in 1999 based on Healthcare Effectiveness Data and Information Set (HEDIS) criteria.

 The HEDIS criteria comprise 1 or more of the following administrative database observations in a 12-month period: (1) 4 or more asthma medication dispensings, (2) 1 or more emergency department visits or hospitalizations with a primary diagnosis of asthma, or (3) 4 or more asthma outpatient visits with 2 or more asthma medications dispensings.

Completed surveys were returned from 2219 members (61%), of whom 1998 (90%) confirmed a diagnosis of physician- diagnosed asthma. The current study was restricted to the 1100 respondents who met the HEDIS criteria for persistent asthma in 2000 (as well as 1999) and thus had electronic utilization information for 2001 available. Of these patients, 1006 subjects had complete quality-of-life data (see Survey Information section) and are the subjects of this report. The study was approved by the Northern California Region and the Nortwest Region Kaiser Permanente Institutional Review Boards. Survey Information The survey included information regarding age, sex, race or ethnicity, educational attainment, household income, and smoking history. The survey also included the mini-Asthma Quality of Life Questionnaire (mini-AQLQ),

 which includes 15 questions in 4 domains (symptoms, activity, emotions, and environment). Scores range from 1-7, with higher scores indicating better quality of life. 

Prior utilization was assessed by the following questions:

1. &#8220;In the past 12 months, how many times did you get treatment for an acute asthma attack at a doctor&#8217;s office, urgent care facility or emergency department (ED)?&#8221; Responses range from &#8220;None&#8221; to &#8220;7 or more.&#8221;

2. &#8220;When was your most recent overnight hospitalization for asthma?&#8221; Responses included &#8220;I have never been hospitalized overnight for asthma,&#8221; &#8220;Within the last week,&#8221; &#8220;Within the last month,&#8221; &#8220;Within the last 6 months,&#8221; &#8220;Within the past year,&#8221; and &#8220;More than one year ago.&#8221;

Prior acute episodes were defined as an answer of one or more on question 1 above, an overnight hospitalization in the past year on question 2, or both.

Demographic and utilization characteristics of the population were evaluated by means of descriptive analyses. The optimal cut-point of the mini-AQLQ for predicting year 2001 emergency hospital care was determined by means of stepwise logistic regression analyses with emergency hospital care as the outcome and various mini-AQLQ cut-point levels as the predictors. Mini-AQLQ cut-points between 3.0 and 6.0 at 0.1 increments were offered as potential predictors. The analyses were done separately in patients with and without a history of acute episodes in the prior year.

Hypothesis testing for univariate analyses was by means of Wilcoxon (continuous variables) or chi-square (categorical variables) tests. Multivariable analyses were performed using logistic regression methodology in which the outcome was year 2001 emergency hospital care and the predictors were (1) mini-AQLQ score (optimal cut-point) and (2) prior acute episodes (as defined above, in the Survey Information section). To determine the clinical significance of the findings, predictive properties were evaluated. Sensitivity was defined as the proportion of subjects with subsequent utilization who were in the high-risk group. Specificity was defined as the proportion of participants without subsequent utilization who were not in the high-risk group. Positive predictive value was defined as the proportion of patients in the high-risk group with subsequent utilization. Negative predictive value was defined as the proportion of participants not in the high-risk group who did not experience subsequent utilization. Finally, a risk ratio was calculated as the risk of the outcome in patients with the risk factor or factors divided by the risk in patients without the risk factor or factors.

Nominal 2-tailed statistical significance for all analyses was set at 

 <.05. All analyses were performed using SAS statistical software (SAS version 8.2 for Windows, SAS Institute Inc, Cary, North Carolina).

A mini-AQLQ cut-point score of 4.7 was chosen in the stepwise logistic regression model for patients without prior acute episodes (odds ratio [OR] = 3.1; 95% confidence interval [CI] = 1.4, 6.9). An mini-AQLQ cut-point score of 4.3 was chosen in the stepwise logistic regression model for patients with prior acute episodes (OR = 3.5; 95% CI = 2.1, 5.9). In the total cohort, a low mini-AQLQ score (either cut-point) or a history of acute episodes in the prior year was associated with an approximate 20% risk of subsequent emergency hospital care compared with a 5% to 6% risk in patients without the predictor (

Although a cut-point of 4.3 was more significantly related to subsequent exacerbations in the logistic regression model than a cut-point of 4.7 in patients with prior episodes, the difference would appear to be clinically unimportant because the 2 cut-points produced equal ORs for subsequent exacerbations in these patients (Table 1). Because a cut-point of 4.7 was associated with higher ORs in the total cohort and in patients without prior acute episodes, this mini-AQLQ cut-point was chosen to define higher risk in subsequent analyses.

Prior acute episodes (OR = 3.8; 95% CI = 2.4, 6.1) and a mini-AQLQ score &#8804;4.7 (OR = 3.4; 95% CI = 2.1, 5.4) were independently related to emergency hospital care the following year in the multivariable logistic regression analysis. The C statistic for the area under the receiver operating characteristic curve was 0.69 for prior acute episodes alone, 0.68 for a mini-AQLQ score &#8804;4.7 alone, and 0.75 when both were included in the model. 

Because prior acute episodes and low mini-AQLQ score appeared to be independent predictors of subsequent emergency hospital utilization, the predictive properties of the presence of either or both of these risk factors were evaluated (

). The presence of either risk factor provided high sensitivity (90.4%) and identified a group nearly 6 times more likely to require emergency hospital care than patients with neither of these risk factors. The presence of both risk factors provided high specificity (79.2%), and patients with both risk factors had a risk ratio of 9.5 compared with patients who had neither risk factor (Table 2).

Emergency hospital care exacts substantial human and economic costs. Identification of patients at increased risk of requiring such care could facilitate targeted intervention to prevent such occurrences. The current study shows that information easily obtainable from the patient (asthma-specific quality of life and a history of episodes requiring unscheduled visits in the past year) can be used to identify patients at clinically meaningful increased risk of future acute episodes. Low quality of life (based on data-derived optimal cut-point scores) and prior acute episodes each independently increase the risk of requiring such care approximately 4-fold. Ninety percent of patients who subsequently required emergency hospital care had 1 of these risk factors. More than 25% of patients with both risk factors required subsequent asthma hospitalizations or ED visits, a nearly 10-fold increased risk compared with that of patients who had neither of these high-risk markers.

Prior studies identified low quality of life as a risk factor for subsequent asthma ED visits and hospitalizations,

Asthma-specific quality of life and a history of acute episodes can be used together to identify risks of subsequent acute exacerbations.

Michael Schatz, MD

Articles

Claims-Based Risk Model for First Severe COPD Exacerbation

Ratio of Controller to Total Asthma Medications: Determinants of the Measure

Asthma-specific Quality of Life and Subsequent Asthma Emergency Hospital Care

Validity of the Asthma Control Test™ Completed at Home

Development and Validation of a Medication Intensity Scale Derived From Computerized Pharmacy Data That Predicts Emergency Hospital Utilization for Persistent Asthma

How Well Do the HEDIS Asthma Inclusion Criteria Identify Persistent Asthma?