lost a loyal friend and trusted advisor when former board member Bernard S. Bloom, PhD, died at his home in Society Hill, Pennsylvania. A gifted economist, Bernie's contribution to the healthcare field was immeasurable and left an indelible impression on students and peers alike. 

The University of Pennsylvania was Bernie's second home for nearly 3 decades - first as a student, earning his doctorate degree in health policy, planning, and economics in 1978, and then as a professor and researcher. As a professor, Bernie is credited with teaching more than 4000 undergraduate students in his now classic introductory course on the healthcare system. Bernie's dynamic presence in the classroom inspired a generation of students to pursue careers in the healthcare field.

Bernie was responsible for conducting key research in the field of health economics and healthcare reform. One of the most memorable studies he led at Penn evaluated the availability of prescription drugs over the Internet. His study found that while the Internet may expand patient access to prescription drugs, it often comes at an increased cost and with little physician oversight. His provocative research has been widely published in many professional and peer-reviewed journals, including, of course, 

Bernie made lasting contributions to the 

not only as an advisor and consultant, but as an author and peer reviewer. Bernie shared his talents as a chief consultant for the World Health Organization in Europe and chief of Health Services Research at the Philadelphia Veterans Administration. He has also held several important positions at Penn over the years, including senior fellow at the Leonard Davis Institute for Health Economics and a fellow at of the Institute on Aging. Bernie also held faculty appointments with the departments of Psychiatry and Dental Care Systems. Known for his great sense of style, Bernie may have been just as likely to criticize your choice of tie as your research methods. But his characteristic candor succeeded in pushing students and researchers toward excellence - a standard that he demanded of himself.

Articles

FROM THE EDITORS

In Memoriam

Local television news is the number 1 source of information for most Americans, and media health reporting has increased significantly during the past 10 years. 

To evaluate the health topics and reporting characteristics of health stories on local television news across the United States. 

Content analysis of full-length broadcasts of local television news from a representative sample of the top 50 US media markets (122 stations). 

Two trained coders evaluated all health stories for topics and reporting characteristics. Any discrepancies were resolved by a third independent coder. 

Among 2795 broadcasts reviewed, 1799 health stories were aired. Seventy-six percent of all stories were about medical conditions. The median story airtime was 33 seconds. Breast cancer and West Nile virus were the 2 most common topics reported on. Among 1371 stories about disease, few gave recommendations, cited specific data sources, or discussed prevalence. Egregious errors were identified that could harm viewers who relied on the information. 

Local television news devotes significant airtime to health stories, yet few newscasts provide useful information, and some stories with factually incorrect information and potentially dangerous advice were aired. Regularly reaching 165 million people, local television news has the power to provide health information to most Americans. It is crucial that television reporting of health news be improved and that reporting errors be eliminated. 

Improving the public's health depends on effectively communicating health information.

Health professionals have limited time in which to disseminate health information,

and such personal patient communication does not reach persons who do not visit or who lack access to healthcare providers. As the number of uninsured people continues to increase and as the time physicians have to spend with patients continues to decrease, providers need to consider how health information is being disseminated to the public. 

One important means of dissemination is the mass news media, which have been shown in several areas to reach almost every segment of the American public and to influence what Americans know and what they consider important.

Health news coverage influences health knowledge and behaviors.

Television news, in particular, has an important influence that may be underappreciated by the medical profession. For example, following Katie Couric's 2000 discussion of colon cancer screening on NBC's 

However, most Americans' main source of information is not national network news but local television news.

57% of Americans regularly watch local television news, outpacing all other media sources by a margin of almost 2 to 1. Americans rate television as their primary source of health information, and more than 76% report acting on health information obtained from the mass media.

have evaluated how health news is reported in the print media, little is known about what health topics reach most of the public through local television news. Understanding the health information disseminated to most of the public via local television news allows managed healthcare professionals to anticipate questions, concerns, and resource needs for their patients and communities. With local television news being Americans' single most important news source, the objective of this study was to evaluate in a systematic way the health topics commonly reported on and the characteristics of this reporting on local television news for a specific period. 

This study was a content analysis of local television news health information reported during October 2002. The data source was the top 50 media markets (122 stations) in the United States. 

There are archives of many newspapers going back more than 100 years, and there is an archive of national network news broadcasts at Vanderbilt University, Nashville, Tenn. Although there have been scattered case studies of local news broadcasts in particular media markets for short periods, there has been no systematic national collection or archiving of local television news broadcasts, to our knowledge. In 2002, the NewsLab at the University of Wisconsin-Madison coordinated the archiving of more than 20 000 local television news broadcasts from a representative sample of 122 stations in the nation's top 50 media markets (reaching 66% of the nation's population) during September and October. To our knowledge, these data represent the largest, most systematic, and most representative sample of local television news coverage collected to date. 

Originally intended to assess local television coverage of American politics, full broadcasts were archived 7 days per week for 7 weeks. NewsLab provided DVD recorders to volunteers in each of the 50 media markets. We used this representative national sample of broadcasts to understand better how health information is reported on local television news across America. For the month of October, we studied the late local news programs, usually at 10 pm or 11 pm. These were typically the top-rated news broadcasts for all 122 stations. When only a 30-minute broadcast was aired during this time, we archived the full broadcast. When the broadcast lasted 60 minutes, we recorded the highest-rated 30-minute segment (the first 30 minutes or the last 30 minutes of the broadcast). Of the 3205 late local television news broadcasts aired during October 2002, NewsLab archived 2795 (87%). The missing 13% were not collected because of malfunctions in DVD recorder timers, human errors, or technical issues. Among the missing broadcasts, there was no pattern in terms of media markets, times, or dates, and there was no indication of any sort of systematic bias introduced into the data set of health stories by their absence. 

was the process of identifying individual health stories within each 30-minute local television news broadcast. Media assistants received a 2-day training session with us (JMP, KMG, EFF) and viewed each broadcast to identify health stories. 

We used a broad definition of a health story. We included any story contained within a 

health segment (ie, a segregated news segment with a recurring health title) and any story that discussed a health topic or a health issue. We excluded stories that reported a death, homicide, suicide, injury, or bioterrorism event but did not discuss it from a medical or a public health perspective. For example, a story about a firearm homicide was not included unless it discussed issues such as the epidemiology of gun injury or prevention. We also excluded stories about medical malpractice. 

Because of possible variations in resources throughout the sample, station demographics were gathered to assess health topic and reporting characteristics with respect to station location and media market size. Each health story was specified by station information, date aired, length of airtime, and network affiliate. Another media assistant independently clipped a 5% random sample of full-length local news broadcasts to assess clipping reliability and demonstrated 95% agreement. 

Two media assistants independently coded each health story for all topics other than the identification of health topics and provided written descriptive summaries. Any discrepancies were resolved by a third independent coder, and final data were based on agreement of at least 2 coders. 

International Classification of Diseases, Ninth

Five independent health professionals with an MPH or MD degree also reviewed the original broadcasts and coded the health topics by classifying them into 1 of 17 categories (the complete data are available in an appendix from the author). Sixteen categories were based on the 

(ICD-9-CM) classification system, and 1 category was for health stories discussing health information unrelated to a medical condition. We combined 2 medical conditions (congenital and perinatal) into 1 category in 

. Before the study, all 5 independent health professionals coded 58 health stories to allow for interrater reliability testing. We tested the interrater agreement for multiple raters and for multiple rating categories according to the method by Fleiss.

< .001), representing good interrater agreement. Within the full data set, each story was independently coded by the chief coder and by at least 1 other coder. We determined interrater agreement for each pair of health professional coders, calculating &kappa; values for 2 raters and for more than 2 categories according to the method by Fleiss.

The &kappa; values for the 4 rater pairs were 0.76, 0.80, 0.68, and 0.77. This represents good-to-excellent agreement. We tested the hypothesis that the agreement was not better than chance and found 

< .001 in each case. For all health topic discrepancies among pairs, a third health professional independently recoded all health topic codes that were discrepant after the initial coding, and the final data were based on agreement between at least 2 health professional coders. 

After each story was coded into 1 of 17 categories, the descriptive summaries for each health story were used to subdivide these general medical conditions into specific health topics. For example, within the general category of cancer, the descriptive summary enabled us to note under specific health topics the type of cancer that was the main focus of the story. If a single story discussed multiple health topics, the health topic receiving the largest proportion of time was coded as the main focus. 

Finally, because the same health story was at times reported on many local news broadcasts, we evaluated the entire sample for specific health stories that were repeated. This was accomplished by grouping all stories with similar descriptive titles under the same general medical condition category. Once this was done, each repeated story was viewed a second time to determine whether the story was in fact reporting on the same health issue. The story was defined as repeated if identical video was aired, if the story referenced the same source, or if identical statistics were reported. An independent health professional evaluated a 30% random sample of stories, with 95% agreement for identifying specific health topics and repeated health stories. Repeated stories that aired in more than 10 media markets were termed 

We described the characteristics of health reporting on local television news. These included health story length, stories reported within a franchised health segment, whether medical research was referenced in the story, and whether the story discussed prevalence, interviewed a health professional, or provided recommendations. 

The health topics were ranked by general medical condition and by a more specific health topic categorization within each general category. In addition, stories that aired in at least 10 media markets were ranked to demonstrate the most pervasive health stories. 

To determine whether exposure to certain health topics differed systematically by location, we compared the health topics and the characteristics of local television news health reporting by media market size (by quartile) within 4 geographic areas. These included the East, Midwest, South, and West; the data are available in an appendix from the author. 

Among 2795 local news broadcasts analyzed, 1799 health stories were aired. Forty percent of broadcasts reported at least 1 health story, and those stories comprised 11% of airtime on local television news that was nonsports related and nonweather related. Twenty-six percent of health stories on local television news gave specific recommendations, and 12% of all health stories reported prevalences of disease. The median health story airtime for local television news was 33 seconds; 40% of stories were part of franchised health segments. Only 27% of all stories interviewed a health professional. 

Among the health stories collected, 1371 (76%) were about medical conditions and were coded into 1 of the 16 categories based on the ICD-9-CM. The top 2 categories for general medical conditions were infectious disease and cancer. This finding was consistent for the entire sample, for each quartile of media market size, and for each of the 4 geographic locations. The median story airtime was similar for all stations. However, local television stations in the East and those in larger media markets were more likely to interview health professionals. Stations in the South reported twice as many West Nile virus stories as other geographic areas in the United States. 

The 2 most common specific health topics reported on local television news were breast cancer (10% [172 stories]) and West Nile virus (9% [166 stories]). Ischemic heart disease was the subject of 3% (58 stories), influenza immunization was the subject of 3% (53 stories), and obesity or weight loss was the subject of 3% (51 stories). 

Table 1 summarizes the breakdown of the stories among the 16 general medical conditions, with each category including only a few specific health topics. The specific health topics listed represent 70% of the 1371 health stories that reported on medical conditions. The 428 health stories not classified into a general medical condition category discussed a range of health topics, the most common of which were smoking cessation (7% [31 stories]) and prescription drug coverage (5% [23 stories]). 

lists the top 5 pervasive health stories aired on local television news.

The most pervasive health story discussed the effectiveness of breast self-examinations.

This story was reported 40 times, covering 26 different media markets, exposing 45 million households to this message. The sources of information for all of the top 5 pervasive health stories were medical journals or the Food and Drug Administration. 

Among the 1799 health story broadcasts, 38% of stories referenced a medical or scientific study generically by using phrases such as "researchers have found" or "a new study shows," with only 6% of all stories citing a specific source. Three journals (

) accounted for 54% of all journals cited. 

This systematic study of health information reported on local television news throughout the United States provides a more generalizable estimate of the health messages aired on local television news than previous studies.

We analyzed 1799 health stories, which comprised 11% of the nonweather-related, nonsports-related airtime for local television news broadcasts that contained health information during the study period. 

What information was conveyed in these stories? The most common general medical condition categories found in this study were infectious disease, cancer, injury or poisoning, and circulatory disease. These categories appear to coincide with common causes of death and disability in the United States.

However, the public health relevance diminishes when one considers specific health topics discussed. For example, 79% of 316 stories about infectious disease were related to the 4 topics of West Nile virus, food contamination, smallpox, and treatment of warts using duct tape. Only 6 stories were about human immunodeficiency virus (HIV) and 4 of these discussed how lemon juice might be useful in preventing HIV transmission during sexual intercourse. 

Although the most common specific health topic, breast cancer, is a significant cause of morbidity and mortality, West Nile virus is not. Less than 1% of the population who contract West Nile virus will develop severe disease.

Nonetheless, West Nile virus was the second most prevalent specific health topic reported on, comprising 9% of all health stories aired and 52% of stories about infectious disease during the study period. 

The reasons for this finding are hardly obscure. West Nile virus was a new topic in October 2002, and news media commonly report on uncommon events. The way in which it was reported is worthy of attention. By the media's failure to place health stories in context, the public's perceived risk may inappropriately increase.

 Most stories about West Nile virus reported the numbers of infected people and deaths within a certain community but did not put this illness into perspective. Perhaps by comparing these numbers with other causes of illness or death or by comparing the risk of West Nile virus with other more common causes of viral illness, such as influenza, would have enhanced the public's perception of this public health problem. In addition, these stories did not discuss known prevention techniques. In fact, only 18% of West Nile virus stories provided recommendations about how to prevent mosquito bites or about what to do if bitten by a mosquito. Managed care health professionals need to realize the deficiencies of television news health reporting and be cognizant of this reporting because it may directly influence patient care. Preventive messages omitted by the news media should be addressed by healthcare professionals when possible. This type of response to the news media's reporting about West Nile virus could have reduced the fears of the public, limited the number of people who sought medical attention after being bitten by a mosquito, and decreased the fatalities caused by West Nile virus. There is no mistaking that West Nile virus was news and should have been reported to the public. However, in the reporting about West Nile virus, opportunities were missed to provide the public with the most useful information. 

Almost 76% of the stories about cancer were about sex-specific cancers such as breast cancer, cervical cancer, testicular cancer, and prostate cancer. Only 5 of 259 cancer stories discussed lung cancer, the leading cause of cancer death in the United States.

Skin cancer, the most common form of cancer, was reported on 5 times,

and only 1 story discussed colon cancer. More alarming, 2 health stories reported that exercise may cause cancer. 

Because October is National Breast Cancer Awareness Month, it is not surprising that breast cancer stories were the most commonly reported cancer topic. The most pervasive cancer story discussed a study

evaluating the effectiveness of teaching breast self-examinations. This story was reported 40 times and aired in more than half the media markets in our sample. Common opening monologues contained statements such as "breast self-examinations may be a waste of time." This type of oversimplified headline is common in the mass media and may be misleading.

Despite these headlines, 65% of these stories gave recommendations, and almost 90% of the recommendations encouraged women to continue performing breast self-examinations. Moreover, there were differences of opinion among health professionals interviewed. For example, a health professional on one local television channel on the East Coast stated that "this is the best done study on this subject," while another health professional at a station in a western state commented that "this is a really terrible study." Viewers in different media markets received different recommendations based on the same health story. Health professionals and health organizations need to take the lead in understanding how to convey potentially controversial information so that the best science reaches the public in a comprehensible manner. 

Thirty-eight percent of all health stories and almost all of the pervasive health stories in our sample referenced scientific studies. Which studies get aired? Wading through the vast amount of published medical research is difficult enough for health professionals, let alone journalists. Health professionals should assist local television news producers in determining what health information to cover. The public desires health information that is pertinent to them.

Healthy People 2010 can provide a framework to help decide which studies should be distributed through the news media. This framework could help medical journal editors scrutinize their press releases so that the health topics provided to the media advance the current public health priorities. In addition, local television news editors could use Healthy People 2010 as a guide in choosing wire service stories that may most benefit the public. 

Time is precious. Health stories reported on local television newscasts last approximately 30 seconds, hardly enough time to deal with any complicated topic. Health organizations commonly produce health stories and make them available to local television stations via B-roll press releases.

These segments are often longer than the time allotted for health reporting on local television news, and stations may import any part of the video into the health story. Only small portions of these segments ever make it onto the air, and few give source attribution because of the limited airtime. It is unclear whether this fragmented reporting affects the message delivered, the story credibility, or the public's comprehension of health information reported on local television news. 

Only 6 stories in our study discussed HIV, a significant health problem in the United States.

Four of these stories discussed an in vitro study about the effect of lemon juice on sperm and speculated about (or presented as fact) the use of lemon juice as a form of contraception and its potential effect on preventing HIV transmission. All 4 stories cited a general reference to the research, only 1 story explained how to use the lemon juice, and 1 story simply stated to "apply it before sex." Only 1 story mentioned that this finding represented the results of a laboratory study and had not been tested in humans. Moreover, one of the stations misinterpreted the study altogether and stated that lemon juice may be a substitute for "costly HIV" medications. Egregious mistakes like this may harm a credulous public; further systematic evaluation of the accuracy of health messages reported on local television news is needed. Many local television news organizations strive to achieve certain standards: most weather reporters are certified meteorologists, yet there is no requirement for credentialing health reporters.

Establishing a "best practices" model for health reporting like that proposed by the Association of Health Care Journalists may enhance the ability of local television newsgroups to convey accurate and useful information.

The main limitation of this study is its temporal specificity. Newsworthy health articles will always be picked up by the media at specific times and will (and should) skew news coverage. Events such as National Breast Cancer Awareness Month and outbreaks of West Nile virus obviously affected the findings of our study. News is always being made, and any study performed during any period will be affected by events of that particular time. However, this limitation should not affect how the reported health information was evaluated in our study. Another limitation is that the study involved only late local news; broadcasts aired during other parts of the day may differ. However, the late local news broadcasts are typically the highest rated newscasts, reaching the greatest number of viewers. 

Despite the influence that mass media, particularly local news, could have on health literacy and the state of public health, little is known about how health information is comprehended by viewers. Future research efforts are needed to better understand how the public comprehends and uses health information obtained from the mass media. Regularly reaching 165 million people, local television health news has the power to convey health information and to influence health literacy among all segments of our society. Measures need to be implemented to encourage better health reporting on local television broadcasts and to eliminate egregious errors in reporting health news to the public. 

We thank Amy Drongowski, MA, and Sonia V. Kamat, MS, for their countless hours spent viewing the video files, managing the data set, and coding. 

From the Departments of Emergency Medicine (JMP, MJG, SKN) and Internal Medicine, History, and Health Management and Policy (JDH) and The Robert Wood Johnson Clinical Scholars Program and the Department of Veterans Affairs (JMP, JDH), University of Michigan, Ann Arbor; and the Departments of Political Science and Communications, University of Wisconsin-Madison (KMG, EFF). 

This study was supported by The Robert Wood Johnson Clinical Scholars Program and the Department of Veterans Affairs (JMP). 

TRENDS FROM THE FIELD

Medical News for the Public to Use? What's on Local TV News

To document the rise in copayments for patients in disease management programs and to call attention to the inherent conflicts that exist between these 2 approaches to benefit design. 

Data from 2 large health plans were used to compare cost sharing in disease management programs with cost sharing outside of disease management programs. 

The copayments charged to participants in disease management programs usually do not differ substantially from those charged to other beneficiaries. 

Cost sharing and disease management result in conflicting approaches to benefit design. Increasing copayments may lead to underuse of recommended services, thereby decreasing the clinical effectiveness and increasing the overall costs of disease management programs. Policymakers and private purchasers should consider the use of targeted benefit designs when implementing disease management programs or redesigning cost-sharing provisions. Current information systems and health services research are sufficiently advanced to permit these benefit designs. 

Two of the most prominent trends in health benefit design have been the rise in cost sharing for patients at the point of service and the proliferation of disease management programs. Increasing copayments reflect a desire by purchasers to shift much of the increasing cost of care to the patient at the point of service. Increased cost sharing will potentially lower premiums and also, in theory, lead to more cost-effective choices by patients. This trend is exemplified by a range of high-deductible benefit packages such as consumer-driven health plans and health savings accounts, as well as by increased cost sharing for pharmaceuticals and physician visits. 

Concurrently, beneficiaries are increasingly being enrolled in disease management programs.

Most large health plans offer enrollees access to programs designed to improve the clinical outcomes of diabetes mellitus, asthma, and heart failure, with growing emergence of programs for other diseases such as coronary artery disease. These disease management programs often are designed by insurers (including managed care plans) or provided by specialized vendors. In anticipation of more widespread adoption, Medicare has initiated a set of disease management demonstration projects. Disease management programs are predicated primarily on 2 beliefs: that these targeted interventions can improve health (which is supported by the literature) and that this improved health can reduce aggregate expenditures (which has less support in the literature).

The purposes of this paper are (1) to document the rise in copayments for patients in disease management programs and (2) to call attention to the inherent conflicts that exist between these 2 approaches to benefit design. On the one hand, disease management programs devote considerable resources to improving patient self-management, often by intervening to enhance adherence to certain behavioral modifications and compliance with specific medications. On the other hand, rising copayments create financial barriers that tend to discourage the use of recommended services.

If increasing copayments lead to enough underuse of recommended services, they could decrease the clinical effectiveness and increase the overall costs of disease management programs. 

Although existing literature indicates that, in general, rising copayments reduce adherence

and, in some targeted cases, worse economic outcomes,

we are not aware of any research that examines copayment changes within disease management programs and whether they differ from changes within the overall insured population. 

We focused on copayments for prescription pharmaceuticals. Medications are an important component of therapy for the chronic diseases targeted by disease management and have been subject to particularly rapid increases in patient cost sharing. For example, the percentage of workers facing triple-tiered formularies rose from 27% to 63% between 2000 and 2003.

In turn, the average copayment for prescription drugs listed on the preferred-formulary level grew 46% and the copayment for nonpreferred branded drugs grew more than 70% over this period.

Assessing the level of copayments for enrollees in disease management programs is difficult because no standard source for these data exists. We collected data from 2 large health plans. Plan 1 provided us with longitudinal data from 2001-2003 for participants in its congestive heart failure (CHF) and asthma disease management programs, as well as for its overall population. Plan 2 provided us with 2003 data for participants in its diabetes mellitus disease management program and for other plan members (including mostly members without diabetes, but also including some diabetic members not enrolled in disease management). 

The plans were selected based on personal contacts and a willingness to participate. One was a large plan in the Midwest that provided combined data for members in its preferred-provider, point-of-service, and traditional insurance products (which tended to have similar pharmacy benefits). The second plan was a large health plan in the West. Combined, the plans serve more than 5 million members. In each of these plans, the copayments faced by the members were largely driven by the decisions of employers, creating substantial variation within the plans in the copayments faced by members. 

Both plans had large, well-established disease management programs and provided us with information about copayment rates for the beneficiaries enrolled in selected disease management programs. In both cases, we collected data separately for generic, preferred branded, and nonpreferred branded medications. 

For plan 1, the data were based on claims experience, where the copayment for each prescription was measured as the out-of-pocket charge to the beneficiary. The vast majority of beneficiaries had benefit designs that charged a fixed dollar amount per prescription. In the cases where beneficiaries were charged a percentage of the prescription costs, the copayment rate was based on the absolute dollar amount charged to the enrollee. In some cases, beneficiaries were charged a differential copayment if they used mail-order services (or they could receive more days supply if they used mail-order services). We did not distinguish between mail-order and retail copayments, so the copayments reflect the retail/mail-order mix. Because the trend towards mail order was modest and not differentiated by participation in disease management, we do not believe this approach generated serious bias in the findings. To the extent that use of mail-order pharmacies has increased over time or is greater among disease management participants, any bias that does exist will tend to show lower and slower growth in copayments for patients enrolled in disease management programs. 

For plan 2, copayment data were based on information from individual benefit packages. Again, the vast majority of members faced a flat dollar copayment per prescription (which could vary by formulary tier). A small share of the population faced a percentage coinsurance rate (80% or 100%). We assigned these enrollees to the group whose copayment was more than $10. The copayments were based on retail purchase. 

Copayments rose for prescription drugs–independent of formulary tier–for individuals in disease management programs. 

reports the trends in copayments from 2001 through 2003 for preferred branded medications for participants in the CHF and asthma disease management programs in plan 1. The proportion of prescriptions with a copayment of more than $10 rose steadily from about 25% to about 40%. 

reports similar data from plan 1 for generic medications. In this plan, there was an upward trend for copayments in the $5 to $10 range for disease management participants buying generic medications. 

Copayments for Disease Management Participants

There was no evidence to suggest that individuals enrolled in disease management programs had appreciably lower copayments than those who were not enrolled. In plan 1, trends for the overall population were very similar to those reported for participants in disease management programs. In any given year, the share of prescriptions within any copayment category for participants in disease management programs was within 3 percentage points of the analogous share for individuals not participating in such programs. 

Plan 2 provided 2003 data for individuals enrolled in the diabetes mellitus disease management program compared with individuals not enrolled in any disease management program. The distributions in pharmaceutical copayments were very similar (

). Individuals in the diabetes disease management program were slightly less likely to face $10 copayments and slightly more likely to face $5 or $7 copayments for preferred branded medications. Because of the large sample size, this difference was statistically significant. However, we do not consider this small difference (mean copayments differed by about 25 cents) to be meaningful. 

This study examined data from only 2 plans. The focus on a few plans is common in the literature; however, like other studies, we cannot be sure of the generalizability of our results. Several factors increase our confidence in the general conclusion that medication copayment rates in disease management programs are rising rapidly and are similar to those outside of disease management. First, the 2 health plans cover a large number of employers, and in many cases the copayment rates and benefit design decisions are made at the employer level; therefore, the sample, although still not representative of all plans, is effectively a sample of many employers. A more accurate interpretation of our sample is probably that, in the areas served by these plans, employers offer the same copayment rates to disease management participants that they offer to other plan members. 

Second, we contacted individuals in the disease management and insurance industries to assess whether they felt that copayments differed for participants in disease management programs. Overwhelmingly, the response was that copayments were the same for individuals within and outside of disease management. For example, a partner at a large actuarial firm very familiar with benefit design practices reported that the firm was not aware of any plans that have different copayment structures for disease management participants versus nonparticipants (D. Mirkin, MD, oral communication, July 2005). 

A few exceptions were uncovered in this process. The most notable was the experience of Pitney Bowes, which lowered copayments in 2001 for medications used to treat diabetes mellitus and asthma. Although we are not aware of a peer-reviewed evaluation of this experiment, Pitney Bowes reported in the press that this reduction in copayments reduced spending by 6% for diabetic patients and 15% for asthma patients.

Another example was undertaken by the city of Asheville, North Carolina, which paid diabetic patients for a monthly consultation with a pharmacist and waived copayments for medications, lab tests, and glucose meters. The Asheville program also reported cost savings, and reports suggest several employers may adopt similar programs.

Finally, high-deductible plans also typically provide first-dollar coverage for preventive services (eg, immunizations). Medications generally would not be included as a preventive service eligible for first-dollar coverage. However, some high-deductible health plans have added certain classes of medications commonly used by those with chronic diseases to the list of services available for first-dollar coverage. For example, in 2006, Aetna plans to exclude a range of medications such as beta blockers, antihyperlipidemics, and angiotensin-converting enzyme inhibitors from the deductibles faced by enrollees in their high-deductible health plan.

As of 2004, at least, Aetna's decision does not seem to be the norm. We are not aware of any such programs in products other than high-deductible plans. 

In the plans we studied, the drug copayments for beneficiaries in disease management programs are about the same, and are rising just as fast, as those for their other enrollees. If disease management programs are able to increase compliance despite the higher copayments, participants in disease management programs may face higher total out-of-pocket costs. More important, the rise in cost sharing at the point of service appears to have occurred without attention to the clinical value of the services in question. Although there may be merit in greater consumer cost sharing in some instances and disease management in other instances, it does not make economic sense to combine greater cost sharing with disease management. 

If patients in the groups targeted by disease management face greater cost sharing, their consumption will be farther from the efficient level, and more disease management resources will be needed to move them to the appropriate level of care. Hence, cost sharing and disease management result in conflicting approaches to benefit design, effectively working against each other. 

Optimal benefit design should align the incentives created by both the cost-sharing and the disease management features of the benefit package. Financial incentives could be viewed as another lever for disease management to use. 

From a societal perspective, when there are different ways to increase consumption of underutilized but valuable services, efforts should first be targeted to increase consumption in a way that uses the fewest societal resources. For this reason, economic models would suggest that copayment rates for individuals in the highrisk subgroups targeted by disease management should be set to zero. Reducing copayments is a socially efficient way to achieve any given level of utilization instead of maintaining higher copayments and using real resources to increase utilization. 

Consider the following hypothetical example. Assume disease management criteria suggest that 100 people "should" take a lipid-lowering agent (or statin), but only 40% do so when faced with a $20 copayment per prescription. Assume this statin costs $100 per prescription. In this scenario, the total costs at baseline are $4000, with the insurer paying $3200 and the patients paying $800. If these 100 people are enrolled in a disease management program that costs $5 per patient (including those who do not adhere) and this program increases use to 60%, the new total cost is $6500, which reflects $6000 spent on the statin (60 Ã— $100) plus $500 (100 Ã— $5) spent on the disease management program. Of that, the insurer pays $4800 for the statin plus the $500 for the disease management program, which equals $5300. The payer also will reap some savings as a result of fewer coronary events. The patients collectively pay $1200 for the medications (and also may reap some financial savings from avoided heart attacks). The extra cost is presumably worth it even if the downstream cost savings do not completely offset the extra drug spending, because of the health benefit. If the health benefit did not justify the cost, then this subpopulation should not have been targeted by disease management. 

Now imagine that reducing the copayment to $0 also increases use to 60%. In this scenario, total costs are $6000 (60 Ã— 100). The insurer pays the full $6000, which exceeds the $5300 paid in the disease management scenario, but the employees pay nothing. The employer could raise the portion of the premium paid by employees (eg, by $1000) to offset the extra employer payment for care, thus reducing the employer costs to $5000 (saving $300 compared with the $5300 needed by the disease management program with a higher copayment). The employees would save $200 (the $1000 extra in premiums is less than the $1200 drug cost paid at the time of purchase). Note that the savings associated with reduced downstream costs still accrue in this example. Even with a $0 copayment, there will still be some underconsumption; therefore, traditional disease management activities could provide value. 

Many payers are concerned that lowering copayments will increase employer spending. This concern may reflect distributional issues that arise because the increased premium associated with lower copayments (and disease management) would be spread across all employees, while the consumer contribution is concentrated among those with the targeted condition. Because of risk aversion, economic theory would suggest that, on aggregate, spreading the costs would enhance the welfare of the group, but some employees who are at low risk for falling into the groups that benefit from disease management may object. 

Existing theoretical research, recognizing the conflict between greater cost sharing and adherence to recommended therapies, has argued for the adoption of benefit-based copayment designs, which maintain low copayments in clinical situations where consumption of the service has substantial clinical value and failure to utilize the given service could result in substantial harm.

Perhaps this points to an opportunity to use existing disease management program infrastructure for targeted lowering of copayments for underutilized but valuable medications, hence aligning plan and consumer incentives for value and efficiency. 

We thank Wendy Banka for research assistance and David Share, MD, MPH, and Ben Rowe, MA, for their assistance in obtaining the data for this study. We also thank Karen Fitzner, PhD, director of Research and Program Development at the Disease Management Association of America, and David Mirkin, MD, principal, Milliman Inc, for their support. 

From the University of Michigan, Ann Arbor, Mich. 

This work was funded by an unrestricted education grant from Pfizer Corp. 

Address correspondence to: Michael E. Chernew, PhD, University of Michigan, 109 S. Observatory, M3118, Ann Arbor, MI 48109-2029. E-mail: ggbz@umich.edu. 

MANAGERIAL

Rising Out-of-pocket Costs in Disease Management Programs

To use disaggregated data about metropolitan statistical areas (MSAs) and clinical conditions to better describe the variation in cost increases and explore some of the hypothesized influences. 

The study uses the state inpatient databases from the Healthcare Cost and Utilization Project, containing all discharges from hospitals in 172 MSAs in 1998 and 2001. The discharge summary information was combined with standardized hospital accounting files, surveys of managed care plans, MSA demographics, and state data pertaining to caps on medical malpractice awards. 

The analysis used descriptive comparisons and multivariate regressions of admission rate and cost per case in 9 leading disease categories across the MSAs. The increase in hospital input prices and changes in severity of illness were controlled. 

Metropolitan statistical areas with higher HMO market penetration continued to show lower admission rates, no less so in 2001 than in 1998. A cap on malpractice awards appeared to restrain admissions, but the net effect on hospital cost per adult eroded for those states with the most experience with award caps. Higher admission rates and increase in cost were found in several disease categories. 

The nationwide trend in hospital inpatient costs since 1990 shows a pronounced acceleration in both admission rates and cost per case starting about 1997-1998. According to the 2002 edition of American Hospital Annual Survey Trends, the annual increase of total cost of hospital inpatient care had fallen to nearly 0 by 1996. Cost per admission continued to decline until 1998. However, costs accelerated each year thereafter. The Medicare Payment Advisory Commission reported similar results for standardized Medicare accounting costs in their reports to Congress every March. 

By 2001, inpatient costs were rising nearly 5% per year, against a backdrop of declining general inflation. The acceleration was noted by a number of analysts who had been expecting hospital inpatient care to continue to offset or cushion increases in other components of healthcare spending.

In interview studies, many influences on hospital cost have been suggested, notably an easing of restrictions on choice of treatments, physicians, and hospitals in managed care plans.

According to these authors, the weakened restrictions in managed care were not due to legislation or regulatory action but to competition among plans for insured persons seeking choice in healthcare services, and competition among employers in some industries to satisfy employees seeking choice of healthcare plans and services. 

A number of other influences that may underlie hospital inpatient cost trends have been suggested in recent studies: beneficial technological advances, increased hospital bargaining power with health plans, and defensive practices to avoid large malpractice liability. First, when a consumer is well insured, it may be difficult to deny services and costly new technologies, regardless of whether they are cost effective, even in managed care plans. It may not be profitable either for hospitals or health plans to limit some questionable uses of new technology.

Second, the number of community general hospitals continued a gradual decline via closures and mergers during the period, and the potential bargaining power of hospitals in a number of geographic areas may have increased along with the number of health plans and the declining use of exclusive networks. (See the article by Devers et al for interview evidence in a number of communities.

) Finally, denial of new (or existing) services with any potential benefit can result in malpractice lawsuits. There is evidence that tighter liability laws and limits on awards can reduce the costs caused by "defensive practice" in treatment for heart attacks.

The purpose of this study was to use disaggregated data for metropolitan statistical areas (MSAs) for 1998 and 2001 to further explore some of these possible contributors to the increase in hospital inpatient costs. The study was restricted to adults with nonmaternity admissions. At the MSA level, variation in managed care penetration and hospital market competition, as well as years of experience with malpractice award caps, can be measured. Some factors that were not clear determinants of the overall national acceleration of cost since the mid-1990s may have effects that vary by geographic region. For example, hospital salaries, wages, and other input prices, as measured in the Centers for Medicare and Medicaid Services (CMS) "market basket update" did not accelerate in the mid-1990s, nor did the severity of illness of hospital patients increase significantly in the leading diagnostic groups.

Nevertheless, these 2 factors and the age distribution may have been important in the growth of cost in some geographic areas but not in others. 

The study further disaggregates hospitalized patients in each MSA to focus on 9 leading groups of conditions that were found to have led the growth in inpatient cost since 1993. Starting with those conditions that accounted for more than 50% of the increase in hospital cost, maternity-related and rehabilitation cases were dropped, leaving 9 condition groups that accounted for more than 22% of all adult admissions and more than 40% of the growth of cost. We started with the top 11 categories contributing about half of the growth in inpatient cost from 1993-2001 (contribution to growth was measured by the initial share of total cost multiplied by the percent increase). Maternity-related cases were dropped due to demographic trends and cycles in births unrelated to any of factors under discussion in this article. Rehabilitation cases were dropped because of suspected changes in the reporting of cases in distinct, specialized units. An apparently large increase in the rate of admission for rehabilitation cases was offset by a decline in average cost per case. The 9 remaining categories contributed 42% of the total growth of costs. A literature search found that only some of these condition groups were subject to recent and documented technology changes believed to be beneficial to patients by improving outcomes. 

A number of studies on the demand and cost for hospital services have illuminated the major influences on the behavior of patients, physicians, and hospitals. Demand for hospital services from the consumer point of view, once an injury or illness has occurred or is suspected, is assumed to depend on expected benefit and out-ofpocket cost, which itself depends on the features of the patient's insurance coverage. Physicians may act as imperfect agents (ie, their incentives may not be completely aligned with the best interest of the patient). This type of model was developed theoretically in the widely known 1991 study by McGuire and Pauly.

Competition among hospitals could affect the combinations of service, service quality, and cost available to consumers and physicians in local hospitals. In recent reviews of research on competition among hospitals, it was shown that the details of definition of market areas and the exact measure of competition did not have a big or qualitative effect on inferences about the effects on hospital expenses or quality.

One problem that has not been overcome in most past research, or in this study, is to distinguish hospitals that are effectively under the same ownership in a single market area due to a relationship with a multihospital system or joint venture. 

A full model taking patient, physician, and hospital objectives into account can be quite complex and difficult to fit empirically (see, eg, a rigorous attempt with a number of limiting assumptions by Gaynor and Vogt

). Nevertheless, several key working hypotheses can be plausibly offered. (1) Because alternatives to inpatient care are usually cheaper, and managed care plans have a financial incentive to substitute cheaper services, areas with a higher market penetration of managed care plans are hypothesized to have lower admission rates and cost. (2) Admission rates and cost would likely be higher for patients with diseases for which there is published evidence of beneficial changes in technology. (3) In areas with a higher proportion of uninsured patients, willingness and ability to pay for inpatient care are reduced; therefore, lower admission rates and costs are expected. (4) A cap on awards for medical malpractice is hypothesized to reduce inpatient costs, but this reduction would likely vary with the number of years since enactment, as found in earlier research by Kessler and McClellan.

The effect of hospital market consolidation and increased bargaining power with health plans does not yield a straightforward hypothesis for hospital cost. The main prediction is that prices and potential profitability would rise. For not-for-profit hospitals, this potential gain might be "spent" on a variety of the organizations' objectives in ways that increase the observed cost. On the other hand, hospitals could potentially reduce cost due to greater volume per hospital, or reduced quality, without having to lower prices. Therefore, no directional hypothesis about cost can be offered without a lot of detailed assumptions about the ownership structure of local hospitals, motives of the hospitals, and consumer preferences. 

We assembled a database of hospitals and inpatient stays for adults in 172 metropolitan areas from 22 states for 1998 and 2001. A total of 1706 hospitals were included in the ultimate dataset. The discharge summaries were part of the state inpatient databases maintained by the Agency for Healthcare Research and Quality (AHRQ). This data source, along with terms of availability to analysts, is described at http://www.hcup-us.ahrq. gov/home.jsp. The Clinical Classification System (also defined at the AHRQ Web site) is based on principal diagnosis, does not depend on choice of treatment, and defines mutually exclusive condition groupings (see http://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp). The study was narrowed to 9 top clinical categories of patients that contributed more than 40% of the national increase in total hospital cost since 1993. The clinical categories are acute myocardial infarction; coronary atherosclerosis; complication of device, implant, or graft; spondylosis, intervertebral disc disorders; cardiac dysrhythmias; osteoarthritis; respiratory failure, insufficiency; congestive heart failure, nonhypertensive; and nonspecific chest pain. 

Individual discharge and hospital data were aggregated or averaged, as appropriate, to the level of MSA, clinical condition, and year. These were denoted respectively by the variables 

for each observation. For a single year, the hypothesized forms of 2 key relationships can be viewed as linear functional relationships: 

 = f (proportion of adults in an MSA more than 75 years old, proportion of uninsured adults, proportion of population enrolled in HMOs, hospital competition index, characteristics of hospitals in the MSA, experience with a cap on malpractice liability, dummy variable for each condition group). 

) = g (area wage index, severity index for patients, proportion of uninsured adults, proportion of population enrolled in HMOs, hospital competition index, characteristics of hospitals in the MSA, experience with a cap on malpractice liability, dummy variable for each condition group). 

Because average cost is a nonnegative variable, the logarithm of average cost was meant to make the error distribution better approximate a normal distribution. It was advantageous to work with a first-difference model. Calculating the difference in either dependent variable for 2001 versus 1998 tended to remove a number of slow-changing unmeasured variables for each combination of MSA and clinical condition. In addition, if 1 hypothesis was that the restrictiveness of managed care plans eased, then the coefficient of HMO market share in either equation would be expected to fall over time. This hypothesis could be tested by including the initial value of that variable as well as the change in a regression. Focusing on just 1 determinant and letting 1 and 0 represent time periods, then (Y1 -Y0) = aX0 + ÃŸ(X1 -X0). Gathering terms, (ÃŸ - a) represents the net effect of the initial value, while ÃŸ is the effect in the subsequent year. Clearly the change in effect over time is positive only if a is positive. In the case of HMO market share, a positive result for a would mean that the restrictive effect eased over time. 

For 2 other key variables, the Herfindahl index of competition among hospitals and experience with malpractice award caps, the initial value and the change were included in the first-difference estimation model. The Herfindahl index (the sum of squared market shares of adult admissions for all hospitals in a market area) is an imperfect measure of concentration. Some hospitals in a market area may be under the ownership of a single system. This can't be reliably determined with the data in this study. For exploratory purposes, we included both the Herfindahl index and the proportion of local hospitals in multihospital systems. However, this variable and a number of other explanatory variables that were considered and later dropped had a relatively small change over the period for any MSA, as shown in 

; hence, they could contribute little explanatory power (see, eg, the proportion of admissions in government-owned hospitals, the proportion of admissions in teaching hospitals, and other MSA characteristics such as availability of primary care physicians). 

To fit the multivariate models, ordinary least squares (OLS) estimation was used. It is possible that the variation in some of the determining variables was endogenous to the behavior of the outcome. In particular, HMO market penetration and initiation of malpractice award caps may be partly determined by business decisions and legislative action affected by prior levels of hospital cost. It would be preferable to use instrumental variables for HMO market penetration and the presence of malpractice award caps. Several variables can provide some useful independent association with these 2 characteristics of market areas. We will discuss the correlations in the Results section. However, these additional variables also may have had an effect on cost and changes in cost, in which case they would not be ideal instrumental variables. 

Using the hospital discharge records, total charges for each case were reduced to a cost estimate, based on publicly available all-payer accounting data reported to CMS. This method has been used in a number of published studies with hospital claim files. The cost/charge ratios for hospitals in the Healthcare Cost and Utilization Project databases are documented in more detail at http://www.hcup-us.ahrq.gov, along with citations of published studies covering the method for imputing missing data and additional background studies of potential bias and refinements by condition. Hospital-wide, all-payer, inpatient cost/charge ratios were calculated from the CMS accounting reports, combining the estimated costs from distinct departments. Background studies for a few states were done with departmental cost/charge ratios overlaid on charges specific to diagnosis-related group (DRG). Diagnosis-related groups with a high proportion of routine bed unit charges, as opposed to ancillary department charges, tended to have costs underestimated. Unfortunately, many states do not report the detailed charges on discharge summaries. The condition (diagnostic) groupings used in this study were less subject to systematic bias because procedures were not a determinant of the classification, as they are for DRGs. 

Severity levels (1 through 4) were determined by the established All Patient Refined Diagnosis-Related Group (APR-DRG) software package from 3M company.

 Severity levels were based on principal and secondary diagnoses, and patient age and sex. Previous studies found that cost per case was significantly higher for severity levels 3 and 4. The proportion of cases at severity level 3 or 4 was the severity indicator used in this study. The area wage index for hospitals, by year, came from public CMS files. The Herfindahl index already has been discussed. This measure requires that data be available for all hospitals in an area–a strength of the source data for this project. Local area demographics were from the Area Resource File (including variables such as percentage of working-age adults who were unemployed and the proportion of the population in various age and education groups). The proportion of the population enrolled in HMOs for 1998 was published in the Area Resource File, based on data from Interstudy, Inc. This information is no longer based simply on location of the health plan office, but reflects the distribution by county of members. Similar data for 2001 from Interstudy were acquired for this study. 

Data on whether a state has a cap on malpractice awards (typically, a limit on awards for nonpecuniary or punitive damages), and the year enacted, were collected by Kessler and McClellan

and were updated by Encinosa and Hellinger in a recent study.

More than 40% of the states had caps in 1998. For those with caps, the number of years of experience was highly skewed because a few states have had caps for more than 20 years. For the MSAs in our regression analysis, 65 never had caps. Of the 107 with caps, the median number of years of experience was between 10 and 11, and the mode was 3 years (28 MSAs), but nearly as many had 23 years of experience. It is likely that the impact, if any, would occur gradually over time, accelerating at first and then decelerating with time. Using dummy variables for the presence and age of caps would help with descriptive comparisons of cost by MSA, but would not allow for increasing experience between 1998 and 2001. Also, a linear function of years in force would not allow for extra years to have more impact in areas with less experience at the beginning. 

We assumed that the impact would be proportional to a simple S-shaped function, 1 -exp(-ÃŸY

), where Y is the number of years in force. This type of nonlinear model has been used in research on technology innovation and impact. Values of ÃŸ were tested from a wide range. For intuitive assessment, one end of the range implied that half the impact of malpractice caps occurred within 3 years. The other end of the range implied that half the impact did not occur until 15 years. After a number of exploratory runs, assessing the effect of experience and the mean squared error of estimation for an equation, we adopted the value of ÃŸ = 0.007, which implies half of the full impact by 10 years. This is not meant to be a definitive estimate because other studies would want to assess the best value in the context of different models. In the current study with this assumed functional form and parameter value, significant associations were found by 10 years with little improvement in statistical models beyond that. Lower values of ÃŸ would imply that more than half the impact of caps would occur beyond the years of experience of more than half the MSAs in the study. This did not appear to be a reasonable assumption. Future studies may choose to explore this issue more fully. The formula for experience was calculated in both 2001 and 1998 to create the change. One state with 2 MSAs dropped award caps in the middle of the period. These MSAs were assumed to gain an additional 1.5 years of experience, and the total experience was assumed to affect admissions and cost even 1.5 years after caps were dropped. 

To evaluate the association of cost with particular diagnostic categories, we made a targeted literature search for medical advances. Our search used MEDLINE and other databases available at http://www.pubmed.gov. We used keyword searches for frequent principal diagnoses and procedures within each of the 9 categories from 1995 to 2003. To narrow the results, secondary key words included terms such as "new technology," "advances," "benefit," and "cost." It was not necessary that a technology be newly invented, but that statistical evidence about new or improved technologies for diagnosis or treatment be published. In some cases, effectiveness was still questioned, but perceived benefits by practicing physicians were reported. Eventually, 70 abstracts and papers were examined in detail by the project team. 

We noted explicit evidence in 4 of the 9 categories that would likely lead to increasing admissions and cost per case. Highlight examples of such evidence are as follows: 

More aggressive treatment of infections that can occur after implanting cardiac devices is expensive but leads to fewer deaths.

Dual-chamber, rate-modulated pacing was demonstrated to be superior to single-chamber pacing, despite shorter life of the device and higher cost.

In the other categories, changes in utilization were noted and discussed in the literature, but these changes were not linked to evidence of beneficial advances. For example, in the category of congestive heart failure, evidence about increased use of intensive-care units, pacemakers, defibrillators, and mechanical ventilation was offered. An increase in referral to teaching hospitals was noted in some articles pertaining to cardiovascular procedures, but this increase was not explicitly linked to particular improvements in technology that might be available. 

Basic descriptive statistics are provided in Table 1. For the dependent variables and the severity of illness, the variables were calculated over all the combinations of MSA and clinical condition used later to fit regressions. Clearly, there was a relatively high variance in the change of these variables over time. Most of the other variation was calculated at the MSA level only. There was considerable initial variation in the Herfindahl index, the proportion of hospitals in systems, and other measures of the mix of hospitals in an MSA. However, the changes over time were relatively small. There was relatively high variation in the experience with malpractice award caps, and the average increase was about one third of the initial average. The latter result is fairly substantial and due to nearly half the hospitals with caps being in the early upswing of the S-shaped experience curve. 

provides a selected set of breakdowns of the initial 1998 values for admission rate and cost together with the changes between 1998 and 2001, controlling for diagnostic category. In the first 2 rows, the HMO market penetration is divided at the median value. Admission rates per 1000 adults were about 10% lower already in 1998 for the areas with above-median HMO penetration (2.98 vs 3.22). Remarkably, the difference widened over time. The admission rate fell by 1.3% where the HMO penetration was relatively high, but was essentially flat when HMO penetration was below the median. Cost per case also was lower in areas with above-median HMO penetration ($7259 vs $7736). The difference shrank somewhat by the end of the period. The bivariate correlations of HMO market penetration with some variables that could affect the locational decisions of plans and self-selection by enrollees found a number of significant positive associations in 1998 with the hospital area wage index, the number of hospitals, the proportion of area population less than 19 years old, and the proportion of adults with 16+ years of education. In a multivariate regression for the HMO penetration in 1998 fitted to these variables, all had 

values greater than 1.0, and the F test was significant, with an adjusted 

of .21. (A number of other descriptors of area demographics and supply of physicians were considered and not found to be significantly correlated with HMO penetration.) These variables were not used as instruments in the later analyses because of concerns that the instruments themselves may have an unacceptable degree of association with the change in admissions and cost. 

The second set of 2 rows provides breakdowns for a median split of observations by the Herfindahl index in 1998. With above-median concentration, the MSA admission rate per 1000 adults was initially nearly 9% higher than that in the areas with below-median concentration. This difference tended to widen over time, as shown in the second column. This is consistent with the survey reports of increased bargaining power by hospitals. If each admission became more profitable, hospitals in more concentrated markets would compete with one another for more admissions. However, it also is possible to imagine a scenario in which less concentration (greater competition) led hospitals to make their services more diverse and attractive, yielding more admissions in total. Cost per case was somewhat higher with abovemedian concentration, but the rate of growth was slightly lower. One might expect, based on the survey studies, that cost would be higher and rising more in the more concentrated markets. However, as we noted earlier, the theoretical link between concentration and prices or profits does not immediately imply a link between concentration and cost. For example, in more concentrated areas there may be less competitive pressure to offer every new technology available for a small group of patients; hence, costs might be lower. The net effect, descriptively, appears to be that cost is higher with concentration, but the difference narrows over time. 

Descriptive breakdowns in relation to the existence of malpractice award caps and the years since initiation are provided later, so only a few correlation results will be offered here. States may have enacted award caps before 1998 for a number of reasons, including high and rapidly rising healthcare costs, high malpractice premiums, or a high burden on state budgets of Medicaid spending. A number of potentially exogenous determinants of legislative action (ie, not the level or rate of increase in costs) were examined. In a regression for whether a state had an award cap in 1998, all of the following variables were significant, with an absolute 

of .32: the proportion of Democrats in the state legislature (1995, positive effect), the continuation of a voluntary Certification of Need program (1995, negative effect), and a ban on corporate contributions to election campaigns (1998, negative effect). These were interesting associations, but again these variables could have some relation to future costs and hence were not deemed convincing for use as instrumental variables. Other measures were considered: malpractice premium rates for surgeons (1995) and the proportion of the state budget allocated to Medicaid (1995). These 2 variables were negatively correlated with the existence and age of award caps, suggesting that these were outcomes of caps. 

provides results for both the change in admission rates and the change in cost per case. Consider first the admission rate. The results suggest that HMO market penetration continued to restrain admission rates after 1998. A negative effect in the initial year indicates that restrictive effects did not weaken over time. The percentage of unemployed adults, included as a proxy for the uninsured population, had a significantly positive effect, counter to expectation. This could perhaps be due to less use of ambulatory and preventive services, resulting in more preventable admissions. Longer experience with malpractice award caps as of 1998 was associated with slower growth of admission rates, but the added experience as of 2001 did not have a significant association. This is consistent with a diminished effect over time. The hospital concentration measures did not have significant independent effects. 

In 3 of the 4 condition groups where evidence of beneficial technology advances was found, the increases in admission rate were significant. However, 2 other condition groups also had significantly high increases in admission rates. These coefficients are all relative to the category of acute myocardial infarction, which was a little below the average in the change of admission rate over time. The unexpected, large positive difference for nonspecific chest pain stands out as unexpected. Also, the increase in admissions with a principal diagnosis of osteoarthritis, after controlling for advanced age, was not expected. 

The last two columns in Table 3 provide results for the change in log of adjusted cost per case. Because the severity scores do take age into account, no age variables were needed. The severity index had a large effect on cost, as expected, after controlling for the type of condition. In these models, neither the initial level nor the change in the HMO market penetration had a significant association with cost. The proportion of unemployed adults was associated with lower cost per case, as expected. The Herfindahl index and the proportion of system membership did not have significant associations with cost. The level of experience with award caps in 1998 was not significant, but increased experience appeared to restrain the growth of cost. Because of the S-shaped function used to calculate experience, areas with fewest years of experience had the most increase in experience. Among the specific disease groups, only 2 (spondylosis, intervertebral disc disorder, and cardiac dysrhythmias) were significantly different from the default category. Both of these differences were expected on the basis of evidence of beneficial technology change. 

In view of the timeliness of the question of the effect of malpractice award caps, an effort was made to transform the results from Table 3, together with initial data in MSAs with varying numbers of years of experience with caps, to promote discussion of how the cost per adult may be associated with the experience with award caps. The results are shown in 

In the starting year of 1998, MSAs in states with no award cap had an average cost per adult in each of the 9 categories of admissions of about $25. The technique known as "smearing" was used with the OLS regression of log(cost) in Table 3 to estimate values of cost per case in 2001.

The average cost per adult in the states without caps increased by about 30% in 3 years. The average cost per adult increased by a smaller 21% rate in states with 1 to 10 years of experience with caps. Reading down the columns for cost per adult, there was a substantial drop in cost per adult after 10 years of experience with caps. This was associated with lower admission rates more so than with lower costs per case. However, the final column shows that the percent increase in cost per adult was higher after more than 10 years of experience, a somewhat faster increase than that for states with no award caps. This suggests erosion of the impact of caps after many years. 

Some major factors associated with the recent increases in hospital inpatient cost, seen by comparing MSAs on 9 leading groups of conditions, are not as simple as some observers have suggested. Any easing of restrictions in managed care plans probably did not ease the restraining effect of managed care plans on hospital admission rates per adult. Areas with higher unemployment rates saw higher admission rates, which was partially offset by a lower level of increase in cost per case. Hospital market concentration within MSAs, albeit difficult to measure in this study, was suggested as important in survey studies but with ambiguous theoretical effects. The level or change in concentration was not found to have a significant association with the change in admissions and cost. Incentives to practice defensive medicine appear to affect the admission rates and the increase of costs for the leading condition groups studied, as shown in the effects of prior experience and increasing experience with malpractice caps. Defensive practice also may account for the surprisingly large increase in admissions for some conditions (eg, nonspecific chest pain), but this requires more investigation to separate defensive practice from other incentives (eg, financial incentives for the hospital to admit insured patients from the emergency department, financial incentives for physicians who bill for in-hospital services independently). However, it is not known whether those financial incentives increased over this period. 

Beneficial changes in medical technology are likely to be important determinants of increases in admission rates and cost per case. Significant increases in admission rates were seen in 3 of the 4 condition groups with documented medical advances. Significant increases in cost per case were seen in 2 of the 4 condition groups with such documented advances. 

The study has clear limitations. The results cover a third of the hospitals in the country, and do not include hospitals outside MSAs. The leading disease groups in this study accounted for only about 42% of the increase in hospital inpatient costs. Better measures of local income and insurance coverage at the MSA level might contribute to an explanation of cost changes. The number and size distribution of competing health plans, as well as consumer out-of-pocket share of covered expense, should be considered as underlying determinants of hospital cost. The literature on technology changes is vast, and the search methods used in this study were not exhaustive. Therefore, some costly improvements in treatment for other diseases in the leading categories may have been missed, especially if the improvements were adopted before publication of large definitive studies in the literature. 

The possible effects of experience with malpractice award caps gave results at and above the upper end of the range (5%-9%) found by Kessler and McClellan

for heart attacks. Interestingly, the savings in cost for areas with more than 10 years of experience with award caps appeared to be weakening–the growth rate of cost during this period in those areas was slightly greater than the growth rate in all other areas. The results here cannot be considered as definitive because of the difficulty in finding convincing instrumental variables to control for possible endogeneity of the enactment and continuation of award caps. The topic is worthy of more research because policies to reduce incentives for defensive practice might yield savings in hospital inpatient cost that could be put to more effective use within the healthcare system. Improving the safety and outcomes of services will likely require investments in information systems, training, and organizational reforms for better coordination of services.

However, there may be other or better ways to achieve savings in inpatient cost. These might include reducing the profitability to physicians and hospitals of costly technologies with weak supporting evidence, or varying the consumer out-of-pocket responsibility for costs, depending on evidence of effectiveness. 

We plan to repeat this type of disaggregated study of admission and cost increases, with the data even more disaggregated to the level of hospital and payer group. Another avenue of exploration is to study which procedures and which payer groups within several of the condition groups account for more or less of the growth of expense, and how such trends might be related to beneficial technology advances and a variety of market forces. 

We acknowledge the data organizations that participated in the Healthcare Cost & Utilization Project state inpatient databases for 1998-2001: Arizona Department of Health Services; California Office of Statewide Health Planning & Development; Colorado Health & Hospital Association; Connecticut–Chime, Inc; Florida Agency for Health Care Administration; Georgia Association of Hospitals & Health Systems; Hawaii Health Information Corporation; Illinois Health Care Cost Containment Council; Iowa Hospital Association; Kansas Hospital Association; Maryland Health Services Cost Review Commission; Massachusetts Division of Health Care Finance and Policy; Missouri Hospital Industry Data Institute; New Jersey Department of Health & Senior Services; New York State Department of Health; Oregon Association of Hospitals & Health Systems; Pennsylvania Health Care Cost Containment Council; South Carolina State Budget & Control Board; Tennessee Hospital Association; Utah Department of Health; Washington State Department of Health; and Wisconsin Department of Health & Family Services. We also thank the editor of the journal and the reviewers for useful criticism and suggestions.

From the Center for Delivery, Organization and Markets, Agency for Healthcare Research and Quality, Rockville, Md. 

This research was wholly supported by the Agency for Healthcare Research and Quality. 

POLICY

Renewed Growth in Hospital Inpatient Costs Since 1998: Variation Across Metropolitan Areas and Leading Clinical Conditions

To determine if the use of a more detailed quality measure affected the finding of gender disparities in lipid management.

: Study subjects included 2589 patients with diabetes mellitus in a managed care plan in 2000-2001. We compared the quality of lipid management in men and women using the following 3 measures: (1) the traditional screening measure (measurement of low-density lipoprotein cholesterol [LDL-C] level), (2) the LDL-C level (ie, among those with an LDL-C level measured, an LDL-C level < 130 mg/dL [< 3.37 mmol/L]), and (3) a more detailed appropriate management measure (an LDL-C level < 130 mg/dL or an LDL-C level 

130 mg/dL plus statin initiation or intensification). Multivariate models were adjusted for clustering within clinic and for patient age, insulin use, hypertension, ischemic heart disease, cerebrovascular disease, congestive heart failure, and number of visits. 

In unadjusted analyses, using the traditional screening measure, women were less likely to be screened than men (

< .001). Using the LDL-C measure, women were less likely to have an LDL-C level less than 130 mg/dL (

= .047). However, using the appropriate management measure, women were as likely to receive appropriate management as men once their lipid levels were measured (

Quality measures that only examine LDL-C screening or LDL-C levels may demonstrate that women receive poorer lipid management than men among patients with diabetes mellitus. However, this gender disparity does not persist with the use of a more detailed measure. 

March 2006

In Memoriam

Medical News for the Public to Use? What's on Local TV News

Rising Out-of-pocket Costs in Disease Management Programs

Renewed Growth in Hospital Inpatient Costs Since 1998: Variation Across Metropolitan Areas and Leading Clinical Conditions

Gender Disparities in Lipid Management: The Presence of Disparities Depends on the Quality Measure

Robustness of a Computer-assisted Diabetes Self-management Intervention Across Patient Characteristics, Healthcare Settings, and Intervention Staff