 To quantify the prevalence of dyspeptic and gastroesophageal symptoms and peptic ulcer disease (PUD) in the United States and to identify factors affecting their prevalence.

 Systematic search of MEDLINE and Web of Science through November 2010.

We identified studies of US patients and evaluated a general (not disease-specific) adult sample that reported the prevalence of 1 or more upper gastrointestinal (GI) outcomes of interest, including dyspeptic symptoms, gastroesophageal symptoms, dyspeptic and/or gastroesophageal symptoms, or PUD. Proportions of individuals in each study reporting each symptom were pooled to derive separate prevalence estimates. Qualitative synthesis of data depicting multivariate relationships between covariates and upper GI outcomes was undertaken.

 A total of 36 citations representing 24 studies were included: 9 studies reporting dyspeptic symptoms (n = 14,181), 14 reporting gastroesophageal symptoms (n = 58,701), 5 reporting dyspeptic and/or gastroesophageal symptoms (n = 103,175), and 7 reporting PUD prevalence (n = 269,299). The pooled prevalences of dyspeptic, gastroesophageal, and dyspeptic and/or gastroesophageal symptoms were 16.3% (95% confidence interval [CI] 9.1%-25.1%), 24.2% (95% CI 18.2%-30.5%), and 35.2% (95% CI 14.9%-58.9%). The pooled prevalence for studies asking for shorter-term PUD recall was 3.3% (95% CI 2.2%-4.6%), with lifetime PUD prevalence estimated at 13.8% (95% CI 10.7%-17.0%). The influence of covariates evaluated as part of multivariate analyses was often inconsistent.

 It appears that upper GI symptoms and disorders are common in US inhabitants. We identified patient- and study-level factors that should be considered when assessing upper GI symptom prevalence and conducting future research.

	Systematic review and meta-analysis indicated that upper gastrointestinal symptoms and disorders are common to inhabitants of the United States.

		 The pooled prevalences of dyspeptic, gastroesophageal, and dyspeptic and/or gastroesophageal symptoms were 16.3%, 24.2%, and 35.2%.

		 In studies asking for shorter-term peptic ulcer disease (PUD) recall, the pooled prevalence was 3.3%; in studies asking for lifetime PUD recall, the prevalence was estimated at 13.8%.

		 The influence of covariates on prevalence rates was often inconsistent.

	The phrase “upper gastrointestinal (GI) symptoms” is commonly used to describe a wide range of complaints including dyspeptic and gastroesophageal symptoms, as well as peptic ulcer disease (PUD).

 Such symptoms are a common cause of healthcare utilization, resulting in increased direct medical costs

 as well as costs to individuals and society due to lost work time and productivity (indirect costs), disrupted social life, and lowered quality of life (intangible costs).

	Currently available studies suggest a high burden of disease due to upper GI symptoms.

 However, due to a lack of standardization of upper GI symptom definitions and differences in methodology used between studies, available prevalence estimates have varied greatly.

 Furthermore, data suggest that prevalence of upper GI symptoms can vary markedly between countries.

 The decision for managed care organizations to pay for medications depends on the prevalence of the condition in a population. When different studies report different values, it becomes difficult to estimate true population prevalence. Estimates provided by a systematic review and meta-analysis can be helpful in such situations. Despite this fact, no previous systematic review and meta-analysis focusing on US prevalence rates of upper GI symptoms have been undertaken. Finally, it has been proposed that associations between sociodemographic and other covariates can affect upper GI symptom prevalence

; however, no prior attempt has been made to gather and synthesize such data.

	In order to gain a better understanding of the prevalence of dyspeptic and gastroesophageal symptoms and PUD in the US population, and to more completely identify factors that affect their prevalence, we sought to conduct a comprehensive and systematic review of the literature on this topic.

	A systematic literature search using the MEDLINE (1950 through November 2010) and Web of Science (1994 through November 2010) computerized databases was conducted. The search combined keywords and medical subject headings (MeSH) terms related to upper gastrointestinal, dyspeptic, and gastroesophageal symptoms and PUD, along with “prevalence.” The complete search strategy is available in 

. Additionally, the references of each pertinent article identified were reviewed to locate other relevant published works.

	To be included in this systematic review, studies had to (1) be conducted in the United States (as clear differences in GI symptom prevalence exist between countries); (2) be published in English; (3) evaluate a general (not disease-specific) adult sample selected without prior knowledge or suspicion of the presence or absence of GI symptoms or disorders; and (4) report the prevalence of at least 1 upper GI outcome of interest, including dyspeptic symptoms (eg, postprandial fullness, early satiety, localized epigastric pain, diffuse epigastric pain, belching, nocturnal/fasting pain, abdominal distention), gastroesophageal symptoms (eg, heartburn, regurgitation), dyspeptic and/or gastroesophageal symptoms (at least 1 related dyspeptic or gastroesophageal symptom), or PUD (eg, gastric or duodenal ulcer). Database studies evaluating only the rate of healthcare utilization for upper GI symptoms or disease were not included. Only fully published manuscripts were considered. In studies with more than 1 published report on the same study population, the most recent publication was selected for analysis to avoid double-counting participants, although previous publications were reviewed to supplement for missing data where applicable. When studies with nonidentical but overlapping populations were identified, the most recent publication was again selected for analysis.

	Upon determination of inclusion, studies were further assessed to see whether they reported on the relationship between any sociodemographic or other covariates and upper GI symptom prevalence. To be included in this second portion of the systematic review, studies had to have (1) conducted multivariate analysis and (2) reported the results of at least 1 relationship between a covariate and upper GI symptom prevalence.

	In all situations, 2 investigators (DMS and SMC) determined study eligibility independently, with disagreements resolved by discussion or by a third investigator (CIC).

	Two investigators (DMS and CIC) performed all data extraction. Data collected for each study included authors, year of publication, sample size, US geographic region in which the study was conducted, population studied including setting, use of random or convenience sample, age and sex, method of study data collection (eg, postal, Internet, selfcompleted or interviewer-administered questionnaire), response rate, duration or recall of symptoms, time frame of population sampling/study, criteria used to define upper GI symptoms (including whether a validated tool was used), upper GI symptom prevalence estimates, 

value for the univariate relationship between a covariate and upper GI symptom, and effect size and 

 value for the multivariate relationship between a covariate and upper GI symptom.

	The methodological quality of prevalence studies was assessed using the following attributes: (1) adequacy of sampling by assessing whether participants were sampled randomly or by convenience, (2) use of a validated questionnaire or tool to define upper GI disorders, and (3) sufficiently high response rate (

We did not attempt to give a summary validity rating to each individual study in this systematic review, but rather used each criterion in sensitivity analysis as described below.

	Numerators (n = number of patients reporting the symptom) and denominators (N = total number of patients evaluated) were extracted from each study in order to compute symptom prevalence with accompanying 95% confidence intervals (CIs) using the Wilson score method without continuity correction.

The proportions of individuals in each study reporting symptoms were combined using a random-effects model to derive separate pooled prevalence rates of dyspeptic, gastroesophageal, or dyspeptic and/or gastroesophageal symptoms for all studies.

 Due to the wide variance identified in requested disease recall periods for PUD (upon endoscopy, the prior 12 months, or lifetime prevalence), an analysis pooling all studies was not conducted.

	Between-study heterogeneity was assessed using the I2 statistic with a threshold of 50% used to define a statistically significant degree of heterogeneity. To assess for the potential for publication bias, we reviewed Egger’s weighted regression statistic 

 <.05 suggesting a higher likelihood of publication bias) and constructed funnel plots in which the standard error of each study was plotted against its prevalence (with an asymmetrical funnel plot suggesting publication bias; funnel plots available from authors upon request).

	We conducted various subgroup and sensitivity analyses to examine the effect on dyspeptic, gastroesophageal, and dyspeptic and/or gastroesophageal symptom prevalence of various study characteristics and validity assessment criteria such as sampling (random or convenience); required duration or symptom recall (for dyspepsia, upper GI symptoms, and gastroesophageal symptoms, shorter recall included studies with 

3-month disease recall; for PUD, shorter recall included studies with 

12-month disease recall); US geographic region (national or regional sample); questionnaire/tool to ask about symptoms (validated or not); response rate (

70% or <70%); and year of publication (2006 to present or prior to 2006). For PUD, studies with shorter disease recall durations were pooled separately from studies with longer disease recall durations, and these analyses served as this outcome’s base-case results. No other PUD analyses were conducted due to the limited number of data points. The Cochran Q statistic was used to determine whether a statistically significant difference existed between subgroups. We considered 

 <.05 statistically significant for all analyses. All analysis was conducted using Stats- Direct version 2.7.2 (StatsDirect Ltd, Cheshire, England).

	Qualitative synthesis of data depicting the relationship between various covariates and upper GI symptoms is reported using descriptive statistics.

	Our literature search revealed 3096 nonduplicate citations. Upon title and abstract review, 3328 citations were excluded, leaving 578 citations for full-text review. Upon full-text review, 542 were excluded (

). Of note, 14 studies conducted in the Olmsted County, Minnesota, population were excluded from this analysis as they drew random samples out of the same study population and therefore were likely to have overlapping patient populations. A total of 36 citations, representing 24 unique studies, met our inclusion criteria.

These included 9 studies reporting dyspeptic symptoms, 14 reporting gastroesophageal symptoms, 5 reporting dyspeptic and/or gastroesophageal symptoms, and 7 reporting PUD prevalence. These studies were all observational in nature and were published between 1991 and 2010. Of the 24 studies, 11 (45.8%) reported results of multivariate analysis with an upper GI symptom as the dependent variable (

	Key characteristics of included studies can be found in

. Study sample sizes ranged from 248 to 226,953 patients. The total sample size for dyspeptic, gastroesophageal, dyspeptic and/or gastroesophageal symptoms and PUD studies was 14,181, 11,790, 58,701, and 284,922, respectively. Of the 24 unique studies, 9 (37.5%) used a randomly drawn sample, while the remaining 15 used convenience samples. Eleven studies (45.8%) sampled a national population, while the remaining studies drew participants from a regional area (city, state, or county). Eleven studies (45.8%) used validated questionnaires to elicit prevalence data on upper GI symptoms, with response rates from those reporting ranging between 17% and 92% (median of 70%).

	Nine studies (n = 14,181) reported prevalence data (symptom recall ranging from 1 to 12 months) on dyspeptic symptoms (

 The mean age range for these studies was 42 to 62 years, and females made up 27% to 100% of the studies’ populations. The prevalence of dyspeptic symptoms varied between 2.9% and 34.4%, with a pooled prevalence of 16.3% (95% CI 9.1%- 25.1%) (

). Significant heterogeneity between studies was observed (I

 = 99%). Review of the funnel plot (not shown, available upon request from authors) and Egger’s weighted regression statistic P values suggested the possibility of publication bias in the l symptom analysis (

	Upon subgroup analysis, it appeared that the prevalence of dyspeptic symptoms was greater in studies with longer symptom recall periods (22.8% vs 5.8% for shorter periods, 

<.001), when participants were drawn from a regional sample (23.2% vs 5.8% for national sample, 

 <.001), and when validated questionnaires were used (20.7% vs 4.7% for nonvalidated questionnaires, 

	Four studies reported results of multivariate analysis with dyspeptic symptom prevalence as the dependent variable (

 A single study suggested increasing age was an independent negative predictor of dyspeptic symptoms.

Various psychiatric disorders (including anxiety, depression, and posttraumatic stress disorder) were consistently found to be associated with an increased prevalence of dyspeptic symptoms.

 Use of prescription acid-suppressing medication,

were also found to be positive predictors, although they were only evaluated by 1 study each. Data on sex and race were inconsistent: 1 of 2 studies (50%) indicated that females had a higher prevalence of dyspeptic symptoms,

 and among studies reporting data on race, Caucasians had a higher prevalence in 1 study,

and a third study reported lower prevalence in Caucasians.

 Alcohol use and education level were not shown to be associated with an increased or decreased prevalence of dyspeptic symptoms.

	Fourteen studies (n = 58,701) reported prevalence data (symptom recall ranging from 3 to 12 months) on gastroesophageal symptoms (

 The mean age range for these studies was 42 to 66 years, and females made up 14% to 100% of the studies’ populations. The prevalence of gastroesophageal symptoms varied between 10.0% and 58.3%, with a pooled prevalence of 24.2% (95% CI 18.3%-30.5%) (

 >99%). Review of the funnel plot and Egger’s weighted regression statistic 

 values suggested the possibility of publication bias in the gastroesophageal symptom analysis (

	Upon subgroup analysis, it appeared that the prevalence of gastroesophageal symptoms was greater in studies published during or after 2006 (31.4% vs 19.1% for studies published prior to 2006,

Articles

WEB EXCLUSIVE

Systematic review and meta-analysis suggest that upper gastrointestinal symptoms and disorders are common to inhabitants of the United States.

US Prevalence of Upper Gastrointestinal Symptoms: A Systematic Literature Review

To describe trends in retail clinic use among commercially insured patients and to identify which patient characteristics predict retail clinic use.

Retrospective cohort analysis of commercial insurance claims sampled from a population of 13.3 million patients in 22 markets in 2007 to 2009.

We identified 11 simple acute conditions that can be managed at a retail clinic and described trends in retail clinic utilization for these conditions. We used multiple logistic regressions to identify predictors of retail clinic versus another care site for these conditions and assessed whether those predictors changed over time.

Retail clinic use increased 10-fold from 2007 to 2009. By 2009, 6.9% of all visits for the 11 conditions were to a retail clinic. Proximity to a retail clinic was the strongest predictor of use. Patients living within 1 mile of a retail clinic were 7.5% more likely to use one than those living 10 to 20 miles away (P <.001). Women ( 0.9%, P<.001), young adults ( 1.6%, P <.001), patients without a chronic condition ( 0.9%, P <.001), and patients with high incomes ( 2.6%, P <.001) were more likely to use retail clinics. All these associations became stronger over time. There was no association between primary care physician availability and retail clinic use.

If these trends continue, health plans will see a dramatic increase in retail clinic utilization. While use is increasing on average, it is particularly increasing among young, healthy, and higher income patients living close to retail clinics.

	We examined trends in retail clinic use among a commercially insured population and examined the predictors of retail clinic use.

		 Retail clinic use increased 10-fold among those seeking care for acute conditions.

		Young, healthy, higher income enrollees who lived close to retail clinics were more likely to use them throughout our study period, indicating that retail clinics continue to appeal more to a select group of patients rather than to a more general population.

		 We found no association between retail clinic use and physician access.

	Retail clinics are a new model for delivering ambulatory care with a focus on patient convenience. Located in pharmacies, grocery stores, or “big box stores” such as Walmart or Target, retail clinics focus on the treatment of a limited set of simple acute conditions or preventive care.

 As the number of retail clinics has grown over the last 5 years, there has been interest among payers in who will visit a retail clinic and for what reason.

 Initial studies found that the first users of retail clinics were primarily young, healthy patients who do not have a primary care provider.

In this article we describe (1) the trends in retail clinic use in a large commercially insured population, (2) patient characteristics that predict use of a retail clinic versus another care site, (3) whether the demographic profile of retail clinic users has changed, and (4) whether retail clinic use is higher in communities with a shortage of primary care.

	We used 2007 to 2009 claims and enrollment data provided by Aetna for their 13.3 million enrollees in 22 markets in which there are retail clinics. Aetna covered retail clinic visits throughout the 3 years. The copayment varied based on the employer, but in general was similar to the copayment for a physician visit. Aetna provided data on all enrollees who had at least 1 visit to a retail clinic (n =367,448), as well as a random sample of enrollees from the same markets who did not visit a retail clinic (n = 1,010,910). In our analyses we weighted all enrollees by the inverse of their likelihood of selection for our sample.

	We identified outpatient visits for a set of 11 acute conditions most commonly seen at retail clinics. We identified outpatient visits using the following Current Procedural Terminology

 codes: 99201-99205 and 99211-99215. The 11 “retail clinic—sensitive” conditions (associated 3-digit 

International Classification of Diseases, 9th Revision, Clinical Modification

 diagnosis codes) were upper respiratory infections (460, 465), sinusitis (461, 473), bronchitis (490, 466), pharyngitis (462, 463, 034), otitis media (381, 382), otitis externa (380), conjunctivitis (372), urinary tract infections (599, 595), allergic rhinitis (477), influenza (487), and unspecified viral infection (079). Together they accounted for 88% of acute care visits to retail clinics. We did not evaluate utilization trends for preventive care such as immunizations. Though immunizations accounted for 40% of visits to retail clinics, they are inconsistently recorded in nonretail clinic settings and patients also frequently receive immunizations at sites where no claim is issued (eg, work sites).

	We limited our population to enrollees under 65 years of age who lived within 20 miles of a retail clinic. We excluded enrollees over age 65 because they were likely to have coinsurance with Medicare; therefore, all claims might not have been captured in our data. We computed the geodesic distance from the center of each enrollee’s zip code to the center of each retail clinic’s zip code and dropped enrollees who lived more than 20 miles from any retail clinic. Our goal was to examine utilization trends in the market areas for retail clinics, and the vast majority (97.6%) of the enrollees who visited a retail clinic lived fewer than 20 miles from a retail clinic.

	Our predictors of retail clinic use were sex, age, distance to retail clinic, health status, income level, and access to primary care physicians. We included distance to a retail clinic because in prior studies proximity to providers was an important driver of use.

 To control for health status, we divided enrollees into 3 groups: no chronic conditions, 1 chronic condition, and 2 or more chronic conditions. The chronic conditions identified were 27 pediatric and adult chronic conditions used in prior work on risk adjustment.

	We used 2000 ZIP Code Tabulation Area median household income from the US Census Bureau as a proxy for enrollee income divided into 3 groups: low income (<2 times the federal poverty level), medium income (>2 times the federal poverty level to <$59,000), and high income (

$59,000). We chose $59,000 as a cutoff because it is the 90th percentile of US ZIP Code Tabulation Area median household income.

	To control for the availability of alternatives to retail clinics, we included fixed effects for each of the 22 healthcare markets. We also controlled for whether an enrollee resided in a zip code in which the majority of the population lived in a federally designated primary care Health Professional Shortage Area.

	In our first analysis, we tracked monthly retail clinic utilization for retail clinic—sensitive conditions in the entire study population. In our second analysis, we modeled the choice to use a retail clinic versus other providers for a retail clinic–sensitive condition. Our predictor variables were those listed above. The population of enrollees consisted of those that had a visit to any care site for a retail clinic–sensitive condition in 2007 to 2009. We ran a logistic regression predicting the likelihood of a patient visiting a retail clinic versus other care sites. In our third analysis, we evaluated changes in our predictors over time. We hypothesized that as the number of retail clinic visitors increased, they would become less distinct from the general population. We estimated separate prediction models for each year with the predictor variables listed above. In order to test the significance of the change in the effect of each of our predictors between 2007 and 2009, we conducted an analysis combining the 2007 and 2009 observations that included interactions with 2009 for all of our predictors. Statistical significance of the interaction terms would indicate statistically significant changes in the predictors between 2007 and 2009.

	For all of our logistic models, we used the Surveylogistic procedure in SAS version 9.22 (SAS Institute Inc, Cary, North Carolina). We weighted the enrollees by the inverse of the likelihood of selection and clustered our standard errors by zip code. To estimate the marginal effect of each predictor on the likelihood of retail clinic use, we used the method of predictive margins, also called recycled prediction.

	Among all 13.3 million enrollees in our study population, retail clinic use increased between January 2007 and December 2009 from a monthly rate of 0.3 visits per 1000 enrollees to 2.7 visits per 1000. Among the 3.8 million enrollees who had at least 1 retail clinic—sensitive visit to any site in 2007 to 2009, visits to retail clinics increased from a monthly rate of 0.6 visits per 1000 enrollees in January 2007 to 6.5 visits per 1000 enrollees in December 2009 (

). There was a cleaseasonal pattern with spikes in the winter months and troughs in the summer months.

	Of all enrollees who had a retail clinic—sensitive condition visit between 2007 and 2009, 210,763 (5.5%) visited a retail clinic. That fraction was 1.5%, 4.5%, and 6.9% in 2007, 2008, and 2009, respectively.

	In our model of retail clinic use at any time in 2007 to 2009, the strongest predictors were distance, age, chronic illness, income, and sex (

). Enrollees who lived less than 1 mile from a retail clinic were 7.5% (95% confi dence interval [CI] 6.6%-8.4%) more likely to visit a retail clinic than those who lived 10 to 20 miles away. Adults aged 18 to 44 years were more likely to visit a retail clinic than other age groups, and enrollees with no chronic conditions were 0.9% (95% CI 0.8%-1.0%) more likely to visit retail clinics than enrollees with 2 or more chronic conditions. We found that enrollees in zip codes with higher median household incomes were 2.6% (95% CI 2.1%-3.1%) more likely to use a retail clinic than those from zip codes with low median incomes. Women were 0.9% (95% CI 0.84%-0.98%) more likely to use retail clinics. We found no difference in the likelihood to use retail clinics for those who lived in Health Professional Shortage Areas.

	The Figure shows the monthly rate of retail clinic use broken down by the key population characteristics identifi ed by our model: proximity, age, health status, sex, and income (panels 2-6). In each panel except panel 4 (health status) and panel 6 (income), the distance between the top line and the other lines in the panel generally increases, indicating that the effect of that characteristic was getting stronger over time. For example, the rate of visits per 1000 enrollees from January 2007 to December 2009 for those who live within 1 mile of a retail clinic (panel 2) increased from 1.2 visits per 1000 enrollees in January 2007 to 10.5 visits in December 2009, while for patients 10 to 20 miles away it increased from 0.1 to 2.0 per 1000 enrollees.

	In our models we tested for a change in predictors over time by estimating separate multivariable logistic models for each year. The marginal effects became stronger over time. For example, the marginal effect for enrollees who lived within 1 mile of a retail clinic increased from 2.5% in 2007 to 11.3% in 2009 (

). The difference was statistically significant. The marginal effects for age, health status, and sex all increased significantly from 2007 to 2009. However, while the marginal effects of these predictors increased, the associated odds ratios remained steady (

of the predictor did not change because retail clinic use was increasing across the population.

	In a large commercially insured population, we saw a striking increase in retail clinic use between 2007 and 2009. In January 2007, shortly after retail clinics entered these markets, few enrollees visited them (less than 1 visit per 1000 enrollees). By December 2009, retail clinic use reached an average of almost 6 visits per 1000 enrollees for those who sought care for acute conditions. It is clear that enrollees are “voting with their feet” and that retail clinics are meeting an unmet need for simple acute care and/or addressing a shortage of traditional healthcare providers.

	Our results do not provide a definitive explanation of why enrollees are choosing retail clinics over other sources of care, but they provide some clues. Consistent with prior research,16 our results suggest that convenience is the strongest predictor of retail clinic use. Enrollees who lived close to a retail clinic were much more likely to use one. Controlling for distance to a retail clinic, we also found that enrollees in high-income zip codes were more likely to use a retail clinic. This is possibly consistent with prior studies that have shown that higher income patients place a higher value on their time.

Our results are not entirely consistent on whether primary care relationships drive retail clinic use. We found that healthier, young adults were the most common users of retail clinics. This group is less likely to have a primary care provider and therefore might visit a retail clinic because they have no alternative. However, we found no association between primary care availability and use of retail clinics. This is contrary to arguments that retail clinics fulfill unmet demand in areas of primary care shortage.

	After rapid growth in the number of retail clinics in 2007, there has been no significant net change in the number of retail clinics since mid-2008.

 Despite concerns about the profitability and long-term viability of retail clinics,

 we believe it is notable that the utilization of retail clinics among at least the commercially insured is growing rapidly.

	If these trends continue, payers that offer coverage of retail clinic visits can expect to see continued and rapid growth in the use of these clinics. Care initiated at retail clinics is 30% to 40% cheaper than similar care at physician offices and approximately 80% cheaper than similar care at emergency departments.

 If the growth in retail clinic visits we observed represents substitution for other sources of care, then the increase in retail clinic use could lead to lower costs. However, if these visits represented new utilization (ie, patients seeking care when previously they would have stayed at home), then costs could increase. Future work should explore whether retail clinic visits represent new utilization or replacement of visits to physicians or emergency departments.

	The authors gratefully acknowledge the Robert Wood Johnson Foundation’s Changes in Health Care Financing and Organization Initiative for financial support of this work and Aetna for providing access to their claims data. Ms Reid’s contributions to this work occurred prior to her employment at the Centers for Medicare & Medicaid Services.

Retail clinic use increased dramatically over 3 years, especially among young, healthy, higher income enrollees living close to retail clinics.

Trends in Retail Clinic Use Among the Commercially Insured

 To determine whether there is an association between the quality of child preventive care received and the existence of 1 or more chronic conditions.

 A retrospective study of all New York State children and adolescents enrolled in Medicaid managed care in 2008.

Using a cohort identified through mandatory annual quality reporting, a clinical algorithm was applied to administrative data to assign children to 3 health status levels: healthy/acute, minor chronic, and significant chronic. We performed bivariate and logistic regression analyses to compare the quality of care received by these 3 groups on 10 child-relevant preventive care services.

 One-fourth of the children in our cohort were deemed to have either minor or significant chronic health conditions. Children with chronic conditions generally had a higher or equal probability of receiving recommended preventive care compared with healthy or acutely ill children, even after controlling for member characteristics. For those services where children with chronic conditions were significantly more likely to receive a preventive care service, the risk ratios ranged from 1.03 to 1.11 for minor chronic children and from 1.03 to 1.17 for significant chronic children.

The quality of preventive health care for children with chronic conditions in New York State Medicaid managed care is equivalent to or better than that for healthy or acutely ill children. Investigating quality concerns for subpopulations of members by combining existing standardized quality measures with administrative health status data is a useful tool for informing state quality-improvement initiatives.

	Recommended preventive care can reduce chronically ill children’s need for intensive or specialized care and reduce unmet needs. In New York State Medicaid managed care, and other care systems, utilization of preventive services has been less than ideal. This study investigated the comparability of rates for chronically ill and healthier children.

		 Existence of chronic conditions did not adversely impact the ability of children to access and utilize recommended preventive care through Medicaid managed care.

		 Generally, children with minor and significant chronic conditions received care at levels comparable or 3% to 17% higher than healthy and acutely ill children

	The Health Resources and Services Administration’s (HRSA’s) Maternal Child Health Bureau defines children with special healthcare needs (CSHCN) as “Those who have or are at increased risk for a chronic physical, developmental, behavioral, or emotional condition and who also require health and related services of a type or amount beyond that required by children generally.”

 In the United States, estimates range between 10% and 20% for children with chronic health issues.

 The estimated prevalence in New York State (NYS) generally matches these national estimates, with 22.3% of all NYS children reported as having at least 1 chronic health condition.

	In terms of their more complicated health needs, CSHCN represents an even larger proportion of healthcare utilization.

While these children often require condition-specific services to optimize their health or functioning, they also benefit from preventive services, such as those included in the Recommendations for Preventive Pediatric Health Care.

 General preventive care for CSHCN is important because such services may reduce the need for intensive or specialized care beyond the standard treatment of their condition(s).

Additionally, there is evidence that preventive care visits provide an opportunity for physicians to discuss and manage a child’s chronic condition in addition to discussing recommended preventive care topics with parents, resulting in fewer unmet needs reported for CSHCN.

	Enrollment in health insurance is critical to accessing comprehensive and quality healthcare services, including preventive care.

 Yet possession of health insurance alone does not guarantee improved access or quality. In NYS, Medicaid managed care (MMC) plans provide health insurance for one-third of the state’s approximately 4.5 million total children and for 82% of children in Medicaid generally. While we know that in 2008 the majority of these children received the recommended number of well-child visits from birth through 6 years, rates were much lower for other recommended preventive services, including annual dental and adolescent well-care visits, a trend that has been historically consistent.

 Whether these rates have been even lower for CSHCN has not been studied.

	While healthcare researchers have examined many aspects of children’s healthcare, information is limited on how preventive care quality for CSHCN directly compares with that for children without special needs. Published research comparing these groups has been limited to utilization of specific service types, such as well-child visits, or general service use patterns, such as the number of physician visits.

 None have compared CSHCN with children without special needs across a range of recommended preventive services using standardized quality measures. Such research is particularly relevant as national interest in developing and expanding quality measures for children and their subpopulations gains momentum in conjunction with reducing disparities.

	In this study, we examined the performance of NYS MMC plans using 10 existing, standardized child quality measures (representing a subset of all recommended child preventive care), stratifying the children by 3 levels of health status. Our goal was to determine whether CSHCN (defined as having minor or significant chronic conditions) received a range of recommended preventive care at a level comparable to that of children without chronic conditions in NYS MMC.

	In this analysis, we limited our definition of CSHCN to children having at least 1 chronic condition as defined by using Clinical Risk Grouping software. This is a proprietary 3M product which categorizes individuals into health status levels based on the type and severity of existing health conditions. Clinical Risk Group (CRG) levels are determined using a combination of diagnosis, pharmacy, and procedure information obtained from submitted healthcare claims and encounters. CRGs place individuals into 1 of 9 mutually exclusive levels, ranging from healthy to catastrophic, which were designed to act as a tool for case identification, program evaluation, and case management.

 This method of defining CSHCN is consistent with other attempts to identify high-needs children using a categorical approach, although it results in a somewhat different definition of CSHCN than HRSA’s definition, which includes those at increased risk for a chronic condition.

	Since their creation, CRGs have been tested and used in a variety of research and program applications.

In unpublished analyses from the Chronic Condition module of the NYS-sponsored 2005 Consumer Assessment of Healthcare Providers and Systems (CAHPS) survey of MMC children, we found a high level of concordance (86%) between the CRG and self-reported positive screen for chronic conditions (based on the CAHPS CSHCN screener), indicating that CRGs are consistent with self-reported conditions. Other unpublished analyses examining children enrolled in Medicaid in 2008 through the Temporary Assistance for Needy Families program found that services for children in CRG levels 3 or higher cost 2.5 times more than for those in CRG levels 1 or 2. Since increased cost is generally associated with an increased number and/or intensity of services, this finding suggests that the service use of children in more chronic CRG categories is consistent with HRSA’s service-based definition.

	Based on the types of health conditions included in each CRG level, we created a health status indicator composed of the following 3 mutually exclusive categories (

): 1) Healthy/acute, comprising healthy children and children with significant acute illness, defined by CRG levels 1 and 2; 2) Minor chronic, comprising children with minor chronic diseases, defined by CRG levels 3 and 4; and 3) Significant chronic, comprising children with extensive chronic conditions ranging from dominant/moderate chronic diseases through catastrophic conditions, defined by CRG levels 5 through 9. The most common health conditions included in each health status category for our cohort are listed in Table 1.

	The CRGs for our cohort were calculated using 2008 encounter and claims data, except in the rare instances where there were insufficient member data, resulting in 2007 data being used. Children with no calculated CRG in either 2007 or 2008 and those not using any healthcare services were excluded from the analysis. Table 1 compares the proportion of children assigned to each of the 9 CRG categories within this study’s final cohort and within the NYS MMC child/adolescent population.

	For this study, quality of care refers to the receipt of recommended preventive healthcare services. All NYS managed care plans are required to annually submit healthcare quality data and information to the NYS Department of Health (DOH) as part of its Quality Assurance Reporting Requirements (QARR).

 Health plans submit aggregate summaries of plan performance based on claims and medical records for measures defined by the National Committee for Quality Assurance’s Healthcare Effectiveness Data and Information Set (HEDIS) and NYS DOH.

MMC plans are also required to submit member-level information to NYS for a subset of the full measure set. For this latter requirement, plans submit information for each MMC enrollee qualifying for at least 1 measure in the specified set. The submission identifies whether each member was eligible for each measure and, if eligible, whether the relevant service was received. Data submitted for QARR are reviewed by independent auditors for accuracy and completeness.

	For this study we examined 10 child-relevant HEDIS preventive care measures submitted to NYS DOH as part of the QARR MMC member-level reporting requirements for the 2008 measurement year. For 8 of the 10 measures, plans were allowed to submit samples instead of full population data, with the sampling methodology dictated by HEDIS specifications. Five of the measures were collected biannually and related data were collected for the 2007 measurement year.

	Our general approach to examining preventive care quality for our cohort was to compare each preventive care measure’s performance rate across the 3 health status groups. The performance rate is the proportion of children receiving the recommended preventive care service, as specified by HEDIS. Since each quality measure has its own HEDIS-defined requirements regarding member eligibility for inclusion, the children in each measure differ slightly but are not necessarily mutually exclusive. For the sampled measures, population performance rates were generated by applying measure- and plan-specific weights and then stratifying the analysis by plan.

	The sociodemographic variables used for these analyses were obtained from Medicaid enrollment data and include: sex, race/ethnicity, age group, aid category, receipt of cash assistance, and region of residence (

). Since sociodemographic variables are believed to be potential confounders in predicting access and quality, we chose to control for any disproportionate influence they might have on performance outcomes.

Primary care service utilization may also be related to receipt of recommended preventive care services, so we controlled for this factor as well.

	We began the analysis by examining the bivariate performance rates for each measure across the 3 health status categories. This was done using analysis of variance and Scheffé’s multiple comparison methodologies for the population measures and a generalized least squares methodology for the sampled measures. We then obtained relative risk ratios (RRs) using generalized linear modeling to determine the probability of measure performance for the minor chronic and significant chronic categories in comparison with the healthy/acute group. Relative RRs are more appropriate than odds ratios for predicting variables with common outcomes.

 For each measure we generated 2 RRs from separate regression models: 1) an unadjusted model predicting performance based solely on health status; and 2) an adjusted model predicting performance based on health status and member characteristics. This study was approved to include minors by the NYS DOH Institutional Review Board.

	There were 606,301 unique members who were identified as being included in at least 1 of the 10 examined measures. Three-fourths of our cohort fell into the healthy/acute category, 7.0% fell into the minor chronic category, and 17.8% were categorized as significant chronic (Table 2).

	In examining the sociodemographic information of the full cohort, males and females were evenly represented, with a consistent distribution across the age groups (Table 2). The largest proportion of children was Hispanic (39.3%), followed by an even split between non-Hispanic black (22.7%) and non-Hispanic white (21.1%). The vast majority of children (94.5%) were eligible for Medicaid through the Temporary Assistance for Needy Families program. Nearly one-third of the cohort received cash assistance. More than three-fourths of the children lived in the New York City (NYC) region. Significant chronic children were more likely to be male, non-white, and younger compared with the healthy/acute and minor chronic children. The significant chronic children also had higher rates of receiving Supplemental Security Income and cash assistance and had a slightly higher rate of living outside of NYC. Minor chronic children had the highest proportion of non-Hispanic whites and adolescents compared with the other 2 health status groups. Lastly, primary care utilization was lowest for the healthy/acute group and increased for both the minor chronic and significant chronic groups.

	Children in the significant chronic group had significantly higher bivariate performance rates than those in the healthy/ acute group for 7 of the 10 preventive care measures (

), were statistically lower on 2 measures (annual dental visit and chlamydia screening), and were equivalent on 1 measure (weight assessment/counseling—body mass index [BMI]). The adjusted RRs matched the bivariate trends for all measures except weight assessment/counseling–BMI and annual dental visit (

). For these 2 measures, the adjusted RRs showed children in the significant chronic group had a higher probability of receiving a BMI percentile during a weight assessment counseling visit and an equal probability of having a dental visit compared with those in the healthy/acute group. For those measures with higher adjusted RRs for the significant chronic group, the probability of receiving a preventive care service ranged from 3% to 17% higher compared with the healthy/acute group (Table 4). Chlamydia screening was the only preventive measure where the significant chronic group had a lower adjusted probability of performance, with the probability of receiving this screening being 4% lower for the significant chronic children.

	Minor chronic children had significantly higher bivariate quality rates compared with the healthy/acute group on 4 of the 10 preventive care indicators and were statistically equivalent on the remaining 6 measures (Table 3). After controlling for member characteristics, the RR trends remained consistent with the bivariate trends for all but 2 measures (childhood immunization combo 3 and well-child visits for the first 15 months) (Table 4). For the 15-month well-child measure, the minor chronic group’s adjusted RR became significantly higher compared with the healthy/acute group. For immunization, the adjusted RR became equivalent to that of the healthy/acute group. For those measures with higher adjusted RRs for the minor chronic group, the probability of receiving a preventive care service ranged from 3% to 11% higher for the minor chronic group compared with the healthy/acute group.

	Our primary goal was to gain a better understanding of preventive care quality for children with chronic conditions. Our results show that chronic conditions did not negatively impact child preventive care in NYS MMC. Specifically, after adjusting for member characteristics, children in the significant chronic group generally had a higher probability of receiving preventive services compared with those in the healthy/acute group, while those in the minor chronic group were at least comparable to children in the healthy/ acute group on all measures. Since we controlled for primary care utilization, this finding cannot be solely attributed to CSHCN having inherently more interactions with the healthcare system.

	For the chlamydia screening measure, the significant chronic group had a lower adjusted probability of receiving services compared with the healthy/acute group. The reason for the lower screening rate is not clear. One factor may be that some young women in the significant chronic group received reproductive services or medications for conditions not related to sexual activity but which the measure’s eligibility criteria specify as a determinant of sexual activity. Another factor may be that physicians assess the applicability of this service differently for patients with disabilities. Regardless, the magnitude of the disparity was relatively small.

	Overall, one-fourth of our cohort fell into the minor chronic and significant chronic categories. This proportionfalls within the higher end of other reported CSHCN rates but is consistent with the 22.9% rate reported for NYS public health insurance in the 2007 National Survey of Children’s Health’s Child Health and System Performance Profile.

 Generally we would expect to see a higher prevalence of CSHCN within our Medicaid-based cohort due to the known relationship between lower socioeconomic status and health issues, as well as the inclusion of disability in the Medicaid eligibility criteria.

	While our finding that children with chronic conditions are receiving preventive care at rates similar to healthy children in NYS MMC is important, this study also has implications beyond NYS. As federal child quality initiatives are implemented and healthcare reform focuses on the role of measurement to improve care quality and reduce disparities, states and health plans must find methods for collecting timely, cost-effective, and reliable data. While some consider self-reported health status and medical record review the gold standards, these methods are rarely timely or cost-effective and are usually limited to a subset of the population.

 Likewise, aggregate results from available claims and encounter data provide limited information regarding the quality of healthcare for subpopulations. NYS’s approach to these issues has been to synchronize and augment the strengths of existing information. The requirement of member-level data for selected quality measures has been integrated into the general requirements for annual state reporting. These requirements, in conjunction with an algorithm-based health status synthesized from claim and encounter data, create the flexibility to perform in-depth reporting and quality improvement analyses efficiently and consistently over time.

	The use of these standardized methods to define the quality of care and health status does have limitations. First, this study is limited to children insured through an MMC plan, meaning that uninsured children, children with private insurance, and those not enrolled in MMC are not represented. However, NYS MMC covers 32% of children in NYS, a substantial proportion of the NYS child and adolescent population. Additionally, the QARR quality indicators do not measure many of the diverse services and needs more frequently required by CSHCN, a national concern which is being partially addressed through Children’s Health Insurance Program Reauthorization Act of 2009 (CHIPRA)-funded measure development. These generalized measures do, however, address the healthcare services recommended for all children, including CSHCN, although many quality benchmarks for subpopulations are not available. Finally, while performance rates for children with chronic conditions met or exceeded those of healthy children for most measures, overall performance rates were still below 60% for many measures. While the NYS rates are actually higher than those reported nationally,

they still indicate that quality improvement is needed statewide.

The existence of chronic conditions did not adversely impact the ability of children in Medicaid managed care to access and utilize recommended preventive care services.

Preventive Care for Chronically Ill Children in Medicaid Managed Care

 To examine relative medication adherence of patients filling 90-day supplies of maintenance medications using retail and mail order channels. It was hypothesized that adherence rates would not differ across the 2 channels.

 A cross-sectional retrospective analysis was conducted using de-identified pharmacy claims data from a large pharmacy benefit manager (PBM) database over a 2-year period (January 2008 to August 2010). Patients who were continuously eligible for at least 12 months during this time frame, with benefit plan designs that allowed filling of 90-day supplies either at retail or by mail order pharmacy, were selected.

 Adherence was measured by medication possession ratio (MPR) within a 1-year period. Propensity score matching was employed to minimize differences between the Retail-90 group and Mail Order-90 group.

Overall, patients filling 90-day prescriptions for 9 therapeutic groups (antiasthmatics and bronchodilators, antidepressants, antidiabetics, antihyperlipidemics, antihypertensives, beta blockers, calcium channel blockers, diuretics, and thyroid agents) at retail pharmacies demonstrated a propensity score—matched average MPR that was statistically higher than for patients filling prescriptions via mail order (77.0% vs 76.0%). There were no significant differences in MPR (post-matching) between 90-day retail and mail order channels for individual therapeutic groups, except for antidiabetics (80.2% vs 83.1%).

 On a propensity-matched basis, patients who fill maintenance prescriptions at retail have a slightly, but statistically significantly, higher MPR than patients who fill their prescriptions by mail.

	Given that patient adherence to medication therapy is necessary in the management of chronic diseases, it is important to investigate adherence levels for patients using different distribution channels.

		Our study demonstrated that fulfillment of 90-day prescriptions at retail pharmacies resulted in medication adherence that is comparable to that of mail order.

		Retail pharmacy has the added advantage of providing pharmacist-led care, including face-to-face interaction with patients for medication management, which could improve health and well-being of patients, ultimately leading to cost savings.

	Patient adherence to medication therapy is an important aspect in the management of chronic diseases. Studies show that medication nonadherence is associated with increased hospitalizations and mortality for patients with chronic diseases, such as diabetes and coronary artery disease.

Despite the adverse effects of poor medication adherence, it is estimated that on average, 50% of patients do not take their medication as prescribed.

	Medication nonadherence carries a significant financial burden as well, estimated to be $100 billion for hospitalizations alone,

 impacting patients, employers, and health plans. However, these excess costs can be prevented. A study by Sokol et al

 found that medication adherence rates of 80% and higher are associated with decreased total medical costs for diabetes, high cholesterol, and hypertension. Furthermore, Balkrishnan et al

 showed that for diabetes every 10% increase in medication adherence could help cut medical costs by 8.6%.

	Many more Americans are using prescription drugs,

 a trend that is likely to continue as the US population ages

 and as the prevalence of chronic conditions increases.

 As a result, health plan sponsors have been seeking ways to improve adherence and impact overall healthcare costs.

	Mail order pharmacy is one channel for dispensing 90-day supplies of maintenance medications that pharmacy benefit managers (PBMs) have been using to help control prescription costs and increase generic utilization.

 Mail order pharmacy provides patients with the convenience of having medication delivered to their homes, while incurring lower costs through a reduced copayment incentive.

Some studies indicate use of mail order pharmacy may also contribute toward greater medication adherence.

 While patients using mail order pharmacy generally have access to a toll-free number for medication questions, they lose the ability to have a face-to-face interaction with a pharmacist,

 particularly when filling by mail order is mandatory rather than optional.

	Retail pharmacies also offer fulfillment of 90-day supplies of medication.

 This was implemented in an effort not only to compete with mail order pharmacy, but also to offer patients increased options for 90-day prescription fulfillment.

 Retail pharmacies, as a channel for fulfillment of maintenance prescriptions, offer several advantages over mail order pharmacy. First, patients seem to prefer filling their prescriptions at retail pharmacies compared with mail order, as demonstrated by higher utilization rates for retail pharmacies, regardless of incentive to use mail order.

 Second, this allows patients who prefer to fill their prescriptions at retail pharmacies the added convenience of doing so for all of their 30-day and 90-day medication needs. Third, retail pharmacies offer patients the opportunity to have face-to-face consultations with a pharmacist to address any concerns they may have about their medications. Finally, pharmacists are provided with an opportunity to proactively counsel patients on the importance of taking their medicines as prescribed, which has been demonstrated to improve adherence rates.

	The objective of this study was to compare medication adherence of patients who fill 90-day quantities of maintenance medications through retail versus mail order channels. We limited our analysis to prescriptions of 90-day quantities, rather than 30-day quantity fills, because maintenance-drug patients using this quantity of refills are likely to be stabilized with respect to medication and dosing regimen.

To our knowledge, this is the first study to compare medication adherence between the 2 delivery channels for 90-day prescriptions. We hypothesize that patients filling 90-day prescriptions at retail pharmacies will have adherence rates similar to those of patients employing mail order pharmacy.

	This retrospective cross-sectional analysis used de-identified demographic and pharmacy claims data obtained from alarge PBM database covering the 2-year period from January 1, 2008, to August 31, 2010. All patients who were continuously eligible for at least 12 months during this time and had a pharmacy benefit plan design that allowed for prescription fulfillment of 90-day supplies at either a retail pharmacy or by mail order with equal copays were considered for inclusion. Equivalence of benefits was determined by analyzing the distribution of patient cost sharing by client year, distribution channel, and drug type (generic, preferred, non-preferred), excluding claims where the patient payment amount was equal to the drug cost.

	The analysis was limited to those patients who had at least 1 pharmacy claim for a 90-day supply for any of the following therapeutic groups within the study period: antiasthmatics and bronchodilator agents, antidepressants, antidiabetics, antihyperlipidemics, antihypertensives, beta blockers, calcium channel blockers, diuretics, and thyroid agents. All drugs within each therapeutic group were included, regardless of dosage form. Therapeutic group was identified via the first 2 digits of the generic product identifier (GPI) associated with each pharmacy claim (respective GPI-2s: 44, 58, 27, 39, 36, 33, 34, 37, and 28). We limited our analysis to patients who were not new to therapy (identified by the presence of a prescription fill in the previous 6 months). Patients filling both 30-day and 90-day supplies for the same therapeutic group and those filling only 30-day supplies were excluded from this analysis. If a patient filled a prescription for more than 1 therapeutic group, they were included in each of the respective groups.

	Patients were grouped based on the pharmacy channel used to obtain their 90-day supply of medication. Patients who exclusively used a retail pharmacy for a single therapeutic group were placed into the "Retail-90" group. Patients who exclusively used mail order services for a single therapeutic group were placed into the “Mail Order-90” group. Patients who used both mail order services and retail pharmacies for a single therapeutic group were excluded (n = 21,948) in order to eliminate any potential variation caused by using both channels. Patients filling prescriptions for 90-day supplies at retail for 1 therapeutic group, but filling 90-day prescriptions through mail order for a different therapeutic group, would be included in both the Retail-90 and the Mail Order-90 group for each respective therapeutic group. Construction of the study groups is shown in the 

	We used the medication possession ratio (MPR) to calculate medication adherence for a 1-year period separately for each of the 9 therapeutic groups. MPR was calculated as the number of actual adherent days divided by 365, where actual adherent days is the total days’ supply of medication a patient had available in a 365-day period starting with the index date of the initial prescription filled for each therapeutic group. If a patient had any prescription refills beyond 365 days following the first fill, those prescriptions were excluded from the analysis. If a patient had a refill toward the end of the 365-day period, the number of days’ supply of that refill exceeding 365 days from the first fill was subtracted from the number of actual adherent days so that the maximum MPR could not exceed 100%. If a patient filled a prescription for more than 1 type of medication within a particular therapeutic group, then the days’ supplies of each medication were added together and divided by 365. In such cases, patients need not be adherent to every medication, as a single MPR was computed for all drugs within that particular class. Overall MPR represents the average MPR at the patient level. When a patient used medications from more than 1 therapeutic group, the group-specific MPRs were averaged to obtain a single MPR measure for that patient.

Patients' adherence to maintenance medications at retail pharmacy is slightly higher than those at mail order, presenting opportunities for pharmacists to provide quality care.

November 2011

US Prevalence of Upper Gastrointestinal Symptoms: A Systematic Literature Review

Trends in Retail Clinic Use Among the Commercially Insured

Preventive Care for Chronically Ill Children in Medicaid Managed Care

Medication Adherence for 90-Day Quantities of Medication Dispensed Through Retail and Mail Order Pharmacies

Potentially Preventable Hospitalizations Among Older Adults With Diabetes

Health Information Exchange Among US Hospitals

Persistence With Biologic Therapies in the Medicare Coverage Gap

Compliance and Persistence With Concomitant Statin and Oral Antihyperglycemic Therapy

Ethnic Differences in the Development of Albuminuria: The DISTANCE Study

Physician-Specific Variation in Medication Adherence Among Diabetes Patients