This Article Abstract Reprint (PDF) Submit a response to this article Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services E-mail this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Archive Download to Citation Manager Reprints & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Dubois, R. W. Articles by Wade, S. Search for Related Content PubMed PubMed Citation Articles by Dubois, R. W. Articles by Wade, S. Related Collections Managed Care Pharmaceuticals Quality Of Care State/Local Issues - California

DataWatch

Geographic Variation In The Use Of Medications: Is Uniformity Good News Or Bad?

Abstract

 
Studies have repeatedly found much geographic variation in use of surgical and diagnostic procedures. This study of the variability of medication use for specific conditions in eleven California regions finds surprisingly few differences among regions. The difference between the highest- and lowest-use areas was far less than we anticipated and amounted to only 30–40 percent for many drugs. We explore five potential explanations for low geographic variability: financial incentives, impact of managed care, study design elements, characteristics of California, and pharmaceutical marketing and education efforts. To determine whether these findings represent good or bad news will require further study.

Methods

Top Methods Study Results Why The Low Variability? Good News Or Bad? NOTES

 
Our study used administrative claims data from three California health plans (Blue Cross of California, Blue Shield of California, and PacifiCare of California). Each plan submitted claims for a sample of 250,000–300,000 randomly selected members who had continuous medical and outpatient pharmacy coverage from 1 April 1998 through 30 September 1999. The data included pharmacy (NDC level), facility (UB92), and professional (HCFA 1500) claims/encounters for the sampled members. We selected members of large group plans to eliminate any biases inherent in benefit coverage and underwriting differences for small-group and individual coverage. To ensure complete capture of health care use for these patients, we excluded members with Medicare supplemental coverage.

The aggregated California health plan population included 638,188 patients (that is, members who had coverage and received services during the study period). The average age for this population was 46.1 years, with slightly more females than males (53 percent versus 47 percent). The distribution of health plan types for the aggregated population was as follows: health maintenance organization (HMO), 39 percent; preferred provider organization (PPO), 60 percent; and point-of-service plan (POS), 1 percent. The pharmacy benefit structure for the population was flat copayment, 5 percent; two-tier closed formulary, 23 percent; two-tier open formulary, 60 percent; three-tier open formulary, 3 percent; and unknown, 14 percent.

Analytic approach. The study population resided in California. To perform geographic comparisons, we identified eleven regions within the state. These regions represent aggregations of adjacent counties to form reasonable sample sizes and geographic distinctions. The size of the regions ranged from a low of 16,052 members to a high of 166,316 members (Exhibit 1).

We selected five disease or pharmacologic areas for study: depression (treated with antidepressants), asthma, congestive heart failure, rheumatoid/osteoarthritis, and respiratory tract infections. We chose these conditions on the basis of the frequency of their occurrence, the potential for large amounts to be spent on their treatment, links between them and high-cost conditions that could be avoided with treatment, and other economic and clinical factors.

Using two approaches, we examined geographic variation in medication use: (1) their use in patients with specific diagnoses, and (2) their use in the entire population. To perform the diagnosis-based analysis, we identified patients with the target medical conditions and patients using key medications based on International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis coding and medication use. Patient identification algorithms were consistent with existing definitions for external quality assessment tools, such as those used by the National Committee for Quality Assurance (NCQA). For each condition we identified selected medications that directly treat it (Exhibit 2). We also identified medications that treat complications and comorbidities of the target condition (for example, use of the anticoagulant coumadin for patients with congestive heart failure who are at risk for blood clots). We determined the rate of use of these medications, overall and by region. We defined use as any prescription for that medication during the study period.

For the population-based approach to analyzing geographic variation, we calculated population rates of use for the same medication (for example, use of angiotensin converting enzyme, or ACE, inhibitors per 100 members in each region regardless of diagnosis). These rates were age and sex standardized to the 1999 U.S. population.

Finally, we calculated population rates of use of selected diagnostic and therapeutic procedures. For these procedures, the rate of use was based upon the total population (for example, upper gastrointestinal endoscopies per 100 members), not upon specific diagnosis-based subpopulations (for example, endoscopies per 100 members diagnosed with peptic ulcer).

Measures of variation. A commonly used measure of variation is the standard deviation. This measure shows variability around the mean. However, if the mean differs greatly between groups, then a comparison of standard deviations would not be informative. We used the coefficient of variation, which expresses the standard deviation as a proportion of the mean and is thus independent of it. Coefficient of variation is defined as (standard deviation/mean) x 100.³ We also used a second measure of variation: the ratio of the highest-use region to the lowest-use region.

Study Results

Top Methods Study Results Why The Low Variability? Good News Or Bad? NOTES

 
We observed a surprisingly low amount of geographic variation in use of medications. In the diagnosis-based analysis we found the lowest amount of variation for the use of selective serotonin reuptake inhibitors (SSRIs) in patients on antidepressant therapy (Exhibit 3). We saw the highest amount of variation for the use of Cox-2 inhibitors for rheumatoid/osteoarthritis and for the use of "new" antidepressants that are not SSRIs.

Top Methods Study Results Why The Low Variability? Good News Or Bad? NOTES

Our findings on medication use are striking. These findings differ greatly from frequently observed variation in the use of surgical and diagnostic procedures. In one study of eleven procedures, only surgery for hip fracture had variability this low (ratio of highest to lowest was 1.9).⁴ On average, the authors observed a threefold higher variation (ratio of highest to lowest was 4.7), with a nearly eightfold variation in use of radical prostatectomy and more than a tenfold variation in use of carotid endarterectomy. Lucian Leape found similar variability among counties about the same size as our California regions for carotid endarterectomy (high-low ratio, 8.2), upper gastrointestinal tract endoscopy (high-low ratio, 3.9) and coronary angiography (high-low ratio, 12.1).⁵ Similarly high variability has also been reported for diagnostic procedures. On average, 10 percent of Medicare beneficiaries underwent echocardiography in 1995. Rates varied by state from 5 percent in Oregon to 15 percent in Michigan, for a threefold variation.⁶

Despite the ample body of literature on variation in use of procedures, few studies have examined the geographic variability in use of medications. One study of patients hospitalized for acute myocardial infarction examined the medications prescribed at discharge. That study focused on just one category of patients and did not determine use of the medications after discharge.⁷ At an October 2001 conference, the pharmacy benefit management company Express Scripts presented data based on a 1.2 million person sample from their population. Researchers observed a 1.55-fold variation in use of medications between the highest- and lowest-use states (11.9 versus 7.65 prescriptions per person, respectively).⁸

Why did so little variation in use of medications occur when so many studies have shown wide variation in the use of procedures? Discussions in the literature try to explain the causes of variation. Most typically, these include potential differences in illness prevalence (and thus, a fundamental variation in the actual need for surgery), mixed opinions of the effectiveness of the surgery, the discretionary nature of certain interventions, and differing patient preferences. However, our study requires an alternative discussion, to explore why we observed less geographic variation.

Methodological differences? First, did our results reflect the particular methodology employed in the study? Prior research examined variation in use based upon the entire population, not on a subset of patients with a particular diagnosis. Our study examined geographic variation using both a diagnosis approach (the use of a drug within a group of patients with a relevant diagnosis) as well as a population approach (the use of a drug within an entire population). As Exhibits 3 and 4 indicate, our results were consistent regardless of analytic framework. Examining variation within a group of patients with a particular diagnosis has the benefit of clinical precision. However, it could show low amounts of variability, since the patients have already been diagnosed. Alternatively, the population vantage point examines the use of a drug in an entire region. While more similar to prior study designs, it has the potential limitation that observed variability could result from differences in disease prevalence, in the likelihood of receiving a diagnosis, or in physicians’ propensity to treat with a particular drug regimen.

Selection of clinical area? Second, did the low amount of variability reflect a selection of clinical areas with minimal "uncertainty" about what works? For several of the drug categories (such as ACE inhibitors for congestive heart failure and inhaled corticosteroids for asthma), a firm body of scientific evidence and well-recognized practice guidelines support their use.⁹ However, for other categories in which we observed little variation, no definitive literature or guidelines exist (for example, the use of brand-name versus generic antibiotics for respiratory tract infections).

Lack of financial incentives? Third, did the lower amount of variability reflect the lack of financial incentives to either use or not use a class of medications? For procedures, financial incentives exist for physicians to perform the procedure, and financial disincentives exist for the patient because of the high cost of those interventions. For medications, in contrast, the cost is more modest, and the physician experiences few, if any, financial repercussions from prescribing decisions.

Influence of drug industry marketing? Fourth, does the relative uniformity of medication use reflect the efforts of the pharmaceutical industry to "educate" physicians and patients? We raise this as a potential explanation worth greater investigation. In 2000 the drug industry spent approximately $17 billion on marketing its drugs or "educating" the public on the efficacy of these agents. Unlike surgical procedures or diagnostic tests, which have few or far more modest marketing constituencies (and generally are not marketed directly to patients), drug firms’ marketing efforts may truly educate consumers and providers and lead to greater uniformity of practice. Further research is needed to correlate pharmaceutical marketing intensity with patterns of use, to determine the plausibility of this explanation.

Influence of managed care? Fifth, could our findings reflect the influence of managed care in reducing variation? Our study examined medication use in California, a mature managed care market. Disease management, prior authorization, networks of providers, utilization review, drug formularies, and multiple financial constraints may influence the extent and type of care that providers offer and patients receive. We could not explore this possibility from the literature, because almost all of the published studies examined variation in care among Medicare beneficiaries. This elderly population entered managed care in any number only recently, and the studies have primarily observed procedure use in an environment with relatively few controls.

To begin to explore this possibility, we examined variation in use of procedures in the same California managed care environment (Exhibit 5). We observed lower rates of variability among procedures than historically observed (high-low ratios of approximately 2–3 in our study versus three- to eightfold variability in the studies discussed above), which supports a managed care impact on variation. However, we still observed approximately twice the variability in procedure use compared with drug use (coefficient of variation for procedures and drugs, respectively, 27 percent versus 14 percent). Further side-by-side comparison of drug and procedure variability in managed care and non–managed care environments would help to elucidate the impact of managed care on variations in care.

Unique population characteristics? Sixth, do our findings reflect unique characteristics of California or of our population? Our study used three separate health plans, so our results reflect a composite picture of three of the largest California-based managed care organizations. Moreover, a subset of the Dartmouth Atlas of Health Care 1999 showed a sixfold variation in rates of back surgery in California hospital service areas, a sixfold variation in heart bypass surgery, and a threefold variation in use of partial mastectomies for breast cancer.¹⁰ We cannot exclude unique characteristics of California as a potential explanation for our findings. Studies in other geographic areas would add further support to our approach.

Good News Or Bad?

Top Methods Study Results Why The Low Variability? Good News Or Bad? NOTES

 
Reducing variability has always been an important goal of quality assurance. Do our findings of greater uniformity in use of drugs mean good news? We cannot be certain. When variability is great, similar patients may receive very different interventions. In that circumstance, reducing variation has the potential to improve quality. In contrast, our finding of minimal variability does not mean that patients received the "right amount" of care. Generally, the same proportion of patients received therapy across the state, but it is unclear whether the "right" patients were treated or whether the "right level" of therapy was given. To answer this more fundamental question, a set of appropriateness criteria must be applied to examine medication use. Unfortunately, when such criteria have been applied to surgical procedures, high and low rates of use did not correlate with appropriateness.¹¹ Perhaps when these criteria are applied to use of medications, the answers will be more straightforward. For many of the conditions we studied, no equivalently effective alternative exists—thus, higher use may equate to more appropriate use. For example, most patients with asthma need inhaled corticosteroids, and patients with congestive heart failure have improved outcomes with ACE inhibitors. Surgery or physical therapy does not represent medication replacements. Thus, high-use areas for these medications may well achieve better health outcomes than low-use areas do. Moreover, despite relatively little variability, for certain conditions such as congestive heart failure and asthma, even the small amount of variability that exists suggests underuse of highly effective medications and opportunities for improvement in care.

Editor's Notes

 
Robert Dubois is senior vice-president for business development at Zynx Health in Beverly Hills, California. Elaine Batchlor is vice-president of the California HealthCare Foundation in Oakland. Sally Wade is director of research at Zynx Health.

The authors thank Jeff Kamil, Nancy Stalker, and Cheryl Tanigawa for assisting them with obtaining data and for their contributions to the Advisory Board. They also greatly appreciate the assistance of Marianne Laouri in the analytic design, review of results, and preparation of the manuscript.

NOTES

Top Methods Study Results Why The Low Variability? Good News Or Bad? NOTES

 

1. J. Wennberg and A. Gittelsohn, "Small Area Variations In Health Care Delivery," Science 182, no. 117 (1973): 1102–1108.[Abstract/Free Full Text]
2. R.W. Dubois et al., "Explaining Drug Spending Trends: Does Perception Match Reality?" Health Affairs (Mar/Apr 2000): 231–239.
3. C.H. Hennekens, J.D. Buring, and S.L. Mayrent, Epidemiology in Medicine (Boston/Toronto: Little, Brown, and Company, 1987).
4. J.D. Birkmeyer et al., "Variation Profiles of Common Surgical Procedures," Surgery 124, no. 5 (1998): 917–923.[Medline]
5. L.L. Leape et al., "Does Inappropriate Use Explain Small-Area Variations in the Use of Health Care Services?" Journal of the American Medical Association 263, no. 5 (1990): 669–672.[Abstract]
6. F.L. Lucas, J.E. Wennberg, and D.J. Malenka, "Variation in the Use of Echocardiography," Effective Clinical Practice 2, no. 2 (1999): 71–75.
7. G.T. O’Connor et al., "Geographic Variation in the Treatment of Acute Myocardial Infarction: The Cooperative Cardiovascular Project," Journal of the American Medical Association 281, no. 7 (1999): 627–633.[Abstract/Free Full Text]
8. Express Scripts, "Outcomes Conference 2001,"<www.express-scripts.com/other/news_views/outcomes_conf.htm> (11 October 2001).
9. National Heart, Lung, and Blood Institute, Guidelinesfor the Diagnosis and Management of Asthma, Expert Panel Report 2, NIH Pub. no. 97–4051 (Bethesda, Md.: NHLBI, July 1997), 20; R. Garg and S. Yusuf, for the Collaborative Group on ACE Inhibitors Trials, "Overview of Randomized Trials of Angiotensin-Converting Enzyme Inhibitors on Mortality and Morbidity in Patients with Heart Failure," Journal of the American Medical Association 273, no. 18 (1995): 1450–1456[Abstract/Free Full Text]; and M. Gomberg-Maitland, D.A. Baran, and V. Fuster, "Treatment of Congestive Heart Failure: Guidelines for the Primary Care Physician and the Heart Failure Specialist," Archives of Internal Medicine 161, no. 3 (2001): 342–352.[Abstract/Free Full Text]
10. "Geography Is Destiny: California Variations in Medical Practice," as reported by Dartmouth Atlas of Health Care 1999 (Chicago: AHA Press, 1999).
11. Leape et al., "Does Inappropriate Use Explain Small-Area Variations?"

                      What's this?

This article has been cited by other articles:

				J. Gao and J. Campbell Trend and variation of prescription drug cost in the veterans health-care system Health Serv Manage Res, February 1, 2008; 21(1): 14 - 22. [Abstract] [Full Text] [PDF]

				D. S. Budnitz, D. A. Pollock, K. N. Weidenbach, A. B. Mendelsohn, T. J. Schroeder, and J. L. Annest National surveillance of emergency department visits for outpatient adverse drug events. JAMA, October 18, 2006; 296(15): 1858 - 1866. [Abstract] [Full Text] [PDF]

				A. W. Dick, J. D. Klein, L. P. Shone, J. Zwanziger, H. Yu, and Peter. G. Szilagyi The Evolution of the State Children's Health Insurance Program (SCHIP) in New York: Changing Program Features and Enrollee Characteristics Pediatrics, December 1, 2003; 112(6): e542 - 550. [Abstract] [Full Text] [PDF]