Medical evidence suggests that the current technologies used for chest radiography for screening purposes are not cost-effective and may be clinically harmful.

More surprisingly, a recent metaregression analysis published in the November issue of the 

found that despite formal recommendations, chest radiography screening is still prescribed by primary care physicians, and prescription trends seem to decrease only slowly over time. This is a very important issue considering the worrisome proportions that healthcare costs–and, specifically, imaging costs–are amassing.

For this reason, it would be important to be able to evaluate the impact of chest radiography screening on healthcare spending. Unfortunately, from the reported analysis by Mauri et al the impact of this practice on health expenditure cannot be estimated. More specifically, it would be useful if Mauri et al had reported the actual practice of chest radiography screening in the general population and not only the physicians' prescription habits. This is likely to represent a bias, because the rate of prescription of a certain test may be consistently different from its real practice among the general population.

Given that the same authors, in 2 previous studies, supported that in Greece the rate of prescription of the test by primary care physicians was 3 times higher than its real practice (77% vs 20%),

it is surprising that they did not report this information or comment on this issue, which constitutes a possibly unnecessary limitation in the work they present. 

1. Mauri D, Kamposioras K, Proiskos A, et al. 

Old habits die hard: chest radiography for screening purposes in primary care. 

The new era of medical imaging–progress and pitfalls. 

Cancer screening by primary care physicians: a comparison of rates obtained from physician self-report, patient survey, and chart audit. 

4. Kamposioras K, Casazza G, Mauri D, et al. 

Screening chest radiography: results from a Greek cross-sectional survey. 

5. Proiskos A, Loukidou E, Kamposioras K, et al. 

Screening chest radiography in primary care: an underestimated belief. 

Prescribed chest radiography for screening purposes (S-CRx) is not a cost-effective test–in any setting considered. Its practice may still be harmful because its positive predictive value is low, and further evaluation of false-positive findings might be associated with increased cost and risk from additional diagnostic/therapeutic interventions. To assess whether S-CRx recommendation by primary care physicians changes over time, a systematic review of literature was recently published. As healthcare costs continue to climb to a worrisome proportion, and more specifically the costs that imaging are amassing,

critics observed that test prescription may be consistently different from its real practice among the general population,

did not provide relevant data for the calculation of expenditure related to S-CRx implementation among the population. It was thus suggested that the lack of analyses of S-CRx practice patterns was a limitation of the study. We agree with this criticism. In December 2005, a systematic review of literature was conducted to retrieve relevant peer-reviewed studies evaluating the rate of S-CRx practice among the population. Studies investigating both patients' desire and actual practice were regarded as eligible. 

An electronic search resulted in 23 528 hits (PubMed/MEDLINE 23 184 hits, Thompson Scientific library 304 hits, and Cochrane library 40 hits). Potentially eligible studies were selected by abstract/title, and 203 full-paper manuscripts were thereafter retrieved. The Panhellenic Association for Continual Medical Research archive for the related literature was further perused. 

but only 3 of them were reporting the actual practice of S-CRx

; consequently, we did not proceed to statistical analysis. 

the actual S-CRx practice was 37% for smokers and 32% for nonsmokers, while patients' perception was that they underwent S-CRx at higher rates (45% and 39%, respectively). It is worth mentioning that physicians underestimated their performance of S-CRx; 10% for nonsmokers and 33% for smokers. 

the actual rate of S-CRx performed was double the expected rate based on physicians' recommendations. Patients' educational status seemed to be an independent correlation factor, because only 30% of patients with post-high school education desired a yearly S-CRx compared with 94% of patients without post-high school education (

40% of smokers or patients with occupational risk underwent SCRx in the past 2 years, while in the Mandel et al study,

the patient-years per test rate was only 9%. 

It is obvious that data from literature are not enough to help us estimate the actual practice of S-CRx. Indeed, to our knowledge, only 4 studies

had addressed this issue, still including our report published in 2006 (

Further survey studies are therefore needed to reveal the actual impact of this avoidable screening tool in health economy. 

3. Kamposioras K, Casazza G, Mauri D, et al. 

4. Proiskos A, Loukidou E, Kamposioras K, et al. 

5. Mauri D, Kamposioras K, Proiskos A, et al. 

6.Woo B, Woo B, Cook EF, Weisberg M, Goldman L. 

Screening procedures in the asymptomatic adult. Comparison of physicians' recommendations, patients' desires, published guidelines, and actual practice. 

Screening for cancer by residents in an internal medicine program. 

Screening guidelines in a family medicine program: a five-year experience. 

9. Zemencuk JK, Feightner JW, Hayward RA, Skarupski KA, Katz SJ. 

Patients' desires and expectations for medical care in primary care clinics. 

Articles

LETTERS TO THE EDITORS

Impact of Chest Radiography Screening on Healthcare Spending

To assess enrollment selection bias between a standard Medicare health maintenance organization (HMO) and a higher-priced social health maintenance organization (SHMO) offering full prescription drug and unique home-based and community-based benefits and to assess how adverse selection was handled through SHMO finances. 

Kaiser Permanente Northwest offered the dual-choice option in the greater Portland region from 1985 to 2002. Analysis focused on 3 "choice points" when options were clear and highlighted for beneficiaries. Data collected included age and sex, utilization 1 year before and after the choice points, health status data at enrollment (1999-2002 only), mortality, and cost and revenues. Data were extracted from health plan databases. 

Hospital, pharmacy, and nursing facility utilization for 1 year before and after the choice points are compared for HMO and SHMO choosers. Health and functional status data are compared from 1999 to 2002. Utilization and mortality data are controlled by age and sex. 

SHMO joiners evidenced adverse selection, while healthier members tended to stay in the HMO, with leaner benefits. Despite adverse selection, the health plan maintained margins in the SHMO, assisted by frailty-adjusted Medicare payments and member premiums. 

This high-low option strategy sought to offer the "right care at the right time" and may be a model for managed care organizations to serve aging and disabled beneficiaries under Medicare's new special needs plan option. 

For 25 years, social health maintenance organizations (SHMOs) have operated prototypes for Medicare special needs plans serving disproportionate shares of beneficiaries with severe or disabling conditions. This study looks at how one health plan used a SHMO to control costs. 

Kaiser Permanente offered its SHMO alongside its standard health maintenance organization (HMO), and sicker and more disabled beneficiaries chose the high-benefit, high-cost social HMO. 

Health plans are ill advised to offer special needs plans with benefits for frail beneficiaries (eg, personal care, homemakers, and adult day services) unless Medicare offers a frailty adjuster and allows these benefits to be paid for with savings. 

The Medicare Prescription Drug, Improvement and Modernization Act of 2003 allows Medicare to contract with a new type of managed care organization called a special needs plan (SNP). The SNPs exclusively or disproportionately serve 1 of the following 3 categories of special needs beneficiaries: institutional residents, Medicaid eligibles, and those with severe or disabling conditions.

 Besides this targeting provision, the legislation and regulations to date do not require SNPs to do much that is "special" for special needs beneficiaries. Nor is there much experience in the Medicare health maintenance organization (HMO) market about what types of benefits to offer these beneficiaries or how beneficiaries will choose SNPs versus other options, particularly in the third category focused on beneficiaries with chronic illnesses and disabilities. 

This article details the experience of Kaiser Permanente Northwest (KPNW) from 1985 to 2002, during which time it offered Medicare-aged beneficiaries a choice between a standard HMO and a social health maintenance organization (SHMO).

Operating under Medicare demonstration waivers, SHMOs can be seen as a precursor to SNPs for beneficiaries with chronic illnesses and disabilities. They offer complete prescription coverage plus benefits for "community care" (eg, personal care, homemaking, adult day services, personal emergency response systems, and short institutional respite care), as well as service coordinators to manage these latter benefits and to coordinate with medical care. 

The KPNW strategy for marketing the SHMO was similar to its approach to its earlier HMO demonstration,

including mailings, newspaper advertisements, and internal marketing to aged KPNW HMO members and to members turning age 65 years ("age-ins"). The SHMO members were also allowed to switch to the HMO. Marketing representatives explained how to compute personal out-of-pocket drug copayments under both plans, because this was often the deciding factor. 

Our hypothesis is that the 2 KPNW Medicare programs attracted different memberships: persons joining the SHMO would be sicker, while healthier HMO members would stay with the HMO's leaner benefits and lower monthly premiums. If Medicare capitations were sensitive to risk characteristics and if member premiums covered additional costs, the programs could operate in a cost-effective manner. 

We gathered and analyzed data on benefits, prices, marketing, enrollment levels, enrollee characteristics, and utilization. After discussing the historical record with KPNW staff, we decided to illustrate choice patterns by focusing analysis on specific "choice points" when options were clear and highlighted for beneficiaries. We call these beneficiaries the decision makers. 

The first choice points are geographic expansions when the SHMO expanded from Multnomah and Washington counties (greater Portland) to Clackamas County (Oregon) in 1990, and to Clark County (Washington) in 1995. The KPNW Medicare HMO members in the new counties, as well as age-ins, were offered the SHMO option. The second choice point is 1991, when the differential in member premiums between the HMO and the SHMO jumped from $26 to $59 because of the repeal of the Medicare Catastrophic Coverage Act. During this period, we analyze those who chose to convert from the SHMO to the HMO and those who converted to the SHMO despite premium differences. The third choice point is 1999 to 2002, when switching was allowed at any point during the year. Self-report health status data on new members to the 2 programs were available to compare selection bias. 

summarizes the benefit and price differences between the HMO and the SHMO during the study period. 

We linked the SHMO membership files with the KPNW membership files to form a consolidated file that merges small eligibility gaps (<90 days) when subscribers changed programs. These data include only Medicare-aged beneficiaries who were individual members of the HMO or the SHMO, as the SHMO option was not offered to groups because it was not available throughout the service area. We used the ZIP code for residential address at year end to determine the county. We selected persons continuously enrolled during the calendar year before the decision point and created a variable for the number of months enrolled in the year after the decision point. Date of birth and enrollment year were used to determine age. The membership system holds a termination date and a reason code for the end of every enrollment episode. 

Utilization data are from KPNW systems. Hospital and nursing facility days beyond the cutoffs were truncated to year end. The following 2 pharmacy utilization measures were developed: total days supplied and total number of prescriptions. Health status data for beneficiaries joining from 1999 to 2002 were collected using the KP standard mailed questionnaire (response rate, 80%-90%).

The following 2 probability risk indexes have been developed using these data: the risk of frailty (being dependent on another person for daily care) and the risk of advanced illness care (likelihood of surviving the next 36 months). Definitions are available in the online 

Analysis files were created from multiple annual administrative files using SAS statistical software (SAS Institute, Cary, NC) to form longitudinal person-based files. SPSS version 13.0 (SPSS Inc, Chicago, Ill) was used for 

tests, analysis of variance, and logistic regression analysis to test for group differences. Unadjusted means of the outcome variables are reported, and the odds for joining the SHMO are adjusted by age and by months of eligibility. 

At the end of 1985, the SHMO had 3087 members, and the HMO had almost 10 000 nongroup members. The HMO peaked at about 18 000 individual members in 1987 and fell to 16 000 in 2002. The SHMO membership peaked at 6081 in 1990 and then fell (after premiums jumped) to around 4000 for the rest of the study period. The proportion of SHMO members 85 years or older rose from 7.7% in 1985 to 23.6% in 2002. The percentage of HMO members 85 years or older held steady at 10% during the study period. 

Choice Point 1: 1990 and 1995 Expansions. 

In late 1989, HMO nongroup members and age-ins were informed that the SHMO would be offered in Clackamas County. Clackamas is a large, mostly rural county whose northwest corner reaches the Portland area. The analysis is based on 2094 persons who were offered the choice and had a full year of health plan data before and after the decision point. Of these decision makers, 354 (16.9%) joined the SHMO. Switchers did not differ in age from nonswitchers, but switchers had higher hospital and prescription use before and after switching (

). A multiple regression analysis controlling for age (the statistics are available online in 

) found that, for every additional 10 days of hospital use in the prior year, decision makers were 1.26 times more likely to choose the SHMO. Also, those who selected the SHMO were 1.57 times more likely to have died by 2002, after controlling for age. Expansion to Clark County in 1995 resulted in 11.5% of decision makers switching to the SHMO, with similar differentials in preutilization and postutilization after controlling for age (data not shown). 

Choice Point 2: 1991 Heightened Dual Choice. 

In 1991, the SHMO premium rose from $75 to $125, while the HMO premium rose only from $49 to $66. The KPNW marketing brochure presented options in a manner that was unique during the study period. Not only were HMO members in covered counties mailed information about how to switch to the SHMO, but members in the SHMO were also explicitly offered the option to switch to the cheaper and less comprehensive HMO. 

Almost 1000 family units asked for information to join or leave the SHMO. Having peaked at 6081 members in December 1990, SHMO membership fell to 5170 a year later. The proportion of age-ins choosing the SHMO decreased from 40% in 1985 to 17% in 1991. After this increase in the premium differential, SHMO membership dropped steadily for 10 years to a low of 3913 at the end of 2000. 

We compared preutilization and postutilization for the 3 decision groups (selected SHMO, dropped SHMO, and remained in SHMO), controlling for age. Results show that members leaving the SHMO in 1991 had fewer total hospital days, prescriptions filled, and prescription days supplied in 1990 than those who selected the SHMO or chose to remain in the SHMO (

). Those leaving the SHMO continued to have lower hospital and drug utilization than the other 2 groups in 1991 (

). For the group selecting the SHMO, the 2 prescription measures increased from the preperiod to the postperiod from a mean days' supply of 765 to 924. This could indicate declining health status or a constrained demand relieved by a richer prescription benefit. 

Choice Point 3: Internal Consolidation From 1999 to

This analysis used 4612 individuals joining the HMO or the SHMO in this 4-year period for whom we had self-report health data. The mean age for 1623 SHMO joiners from KP was 72.3 years, which was significantly younger than the mean age for 276 SHMO joiners from the community (74.8 years). The mean age for the HMO joiners was 73.2 years. There were 2 other significant differences between the SHMO joiner subgroups: The community group was more impaired than the KP joiners in instrumental activities of daily living movement tasks (meals, shopping, chores, and getting places) and in instrumental activities of daily living cognitive tasks (money management, taking medications, and using the telephone). The former difference disappeared after controlling for age, while the latter did not. Because only 1 difference remained after age adjustment, the 2 subgroups of SHMO joiners were merged for the selection bias analysis. 

The goal of the selection bias analysis was to take a snapshot of utilization and health status differences between those who joined the SHMO (n = 1899) and those who joined the HMO (n = 2713). Consistent with prior analytic findings, KPNW records show that the SHMO joiners in the year after joining used 100% more total hospital days, 91% more skilled nursing facility days, 64% more total days of prescription drug supply, and 57% more total drug records than the HMO joiners, adjusting for eligibility months and age (data not shown). The health status survey data show the SHMO joiners reporting double the rates of hospitalization, almost 5 times the rates of emergency department use, and triple the rates of daily use of 5 or more prescriptions (

). Odds ratios for frailty measures and for use of community care services and devices also show wide differentials. These utilization and functional status differences reflect underlying differences in major chronic disease: 51% of the SHMO joiners reported diabetes mellitus, heart trouble, or lung problems, compared with 32% of the HMO joiners (unadjusted for age). Moreover, logistic regression analysis, adjusting for age, demonstrated that the SHMO joiners were more than twice as likely to be at risk of frailty and advanced illness. 

How did the SHMO survive adverse selection? The key factor was a Medicare waiver for a frailty-adjusted capitation formula, which paid about double the standard Medicare HMO underwriting factors for members who resided in the community but who met nursing facility preadmission screening criteria. To maintain budget neutrality, the formula also paid a little less than the standard factors for members who did not meet these criteria.

This formula was phased out from 2004 to 2007 in favor of a frailty adjustment to Medicare's new diagnosis-adjusted payment.

Another payment difference is that the SHMO was paid 100% of the estimated county-based fee-for-service equivalents, while the HMO was paid 95%. 

As SHMO members' frailty rose beyond community averages, revenues rose above what the program would have received using the standard formula. For example, in 1992 the SHMO accounted for 13% of KPNW Medicare members and 14% of Medicare revenues. By 2002, the SHMO represented only 9% of membership but still accounted for 13% of Medicare revenues. The ratio of the SHMO to HMO Medicare payments per member per month (PMPM) went from 1.13 in 1992 to 1.48 in 2002. 

The higher Medicare payments were justified by higher acute care utilization. The ratio of SHMO to HMO hospital days per 1000 rose from 1.03 in 1986 to 1.40 in 1992 and to 1.66 in 2002. The ratios of skilled nursing facility days per 1000 and outpatient visits per 1000 were much steadier, rising from 1.28 in 1986 to 1.64 in 1992 and to 1.51 in 2002 for skilled nursing facilities and from 1.30 in 1986 to 1.24 in 1992 and to 1.23 in 2002 for outpatient visits. In addition, costs for community care and resource coordination grew from $21 PMPM in 1986 to a high of $98 PMPM in 2001, amounting to about 15% of Medicare revenues in all study years. During the later years, more than 30% of the members lived at home but met nursing facility preadmission screening criteria, and about 25% received community care.

The rising costs in both programs, relative to Medicare revenues, led to increases in beneficiary premiums, which have been much higher in the SHMO than in the HMO. The differential has been more than 2:1 since 1997, with the absolute differential about $100 per month since 2000 (Table 1). In addition, community care benefit copayment revenue rose from about $5 PMPM in the early 1990s to more than $10 PMPM by 2001. 

Finally, the effect of the high-low choice on overall Medicare revenues should be considered. Although the standard diagnosis-based Medicare payment system and the frailty-adjusted version were actuarially sound standing alone, this is threatened by our finding that frailer elders tended to pick the high-option side of a high-low choice and that healthier elders tended to choose the HMO. Clearly, the 2 payment formulas produced a more favorable revenue situation for KPNW than would have occurred with only 1 formula, aside from the "extra" 5% in the SHMO payment. 

Since the inception of Medicare managed care, policy makers and providers have struggled to find the combination 

Selection Bias Between 2 Medicare Capitated Benefit Programs

 of appropriate payment systems and care models to serve increasingly frail beneficiaries. Kaiser Permanente Northwest used subtle, but potentially important, changes in financing, marketing, care delivery, and organization to create a new choice of benefits for Medicare elders–a standard HMO supplement alongside a more expensive SHMO with full prescription drug and community care benefits. Despite adverse selection into the SHMO, KPNW successfully adjusted the member premiums and copayments to maintain margins, assisted by frailty-adjusted Medicare payments. Rising cost and benefit differentials further drove adverse selection, but the SHMO remained viable and compelling, attracting persons with poorer health seeking more services. This high-low strategy sought to offer the "right care at the right time" to beneficiaries, and many were ready to pay when the time came. 

Because the SHMO waivers end in December 2007, it is not clear which elements of this approach will be available to KPNW or other health plans in the future. One option is for the SHMOs at KPNW, Elderplan, and SCAN Health Plan (the initial sites) to become "disproportionate share" SNPs, because 25% of members in each of the plans meet nursing facility level-of-care requirements.

Elderplan took this option in 2006. However, the new regulations make no provision for frailty-adjusted payment for SNPs not operating under demonstration waivers. Given KPNW's experience with adverse selection, Medicare Advantage plans would be ill advised to offer such a choice to beneficiaries without similar payment protections. The Centers for Medicare & Medicaid Services is phasing out the current SHMOs' frailty adjuster from 2008 to 2010. 

One policy option for Medicare would be to give the frailty adjuster only to SNPs that offer benefits designed for special needs beneficiaries (eg, community care, comprehensive prescription coverage, and more). Kaiser Permanente Northwest met these criteria in 2006, offering comprehensive drug coverage through the "doughnut hole" in the SHMO and an increase in the value of its community care benefit from $12 000 to $14 400 per year. The beneficiary premium jumped to $215 a month versus $76 for the HMO, but the SHMO continued its modest growth. The fact that significant numbers of beneficiaries continue to pay this high monthly premium shows the value they place on the benefits. 

Unfortunately, even if the Centers for Medicare & Medicaid Services were to extend the frailty adjuster to protect against adverse selection, their policy for 2006 further discouraged the option of offering community care benefits by prohibiting Medicare Advantage plans from using savings to pay for services that are not "health care related," including "homemaker services...to assist people to meet personal, family, or domestic needs."

Under this policy, it will be difficult if not impossible for SNPs to follow the SHMO model and offer benefits to meet nonmedical special needs. 

We are indebted to Kevin Lutz, MFA, for his assistance with the manuscript. 

From Schneider Institute for Health Policy, Heller School for Social Policy and Management, Brandeis University, Waltham, Mass (WL); and Center for Health Research, Kaiser Permanente (KKB, NAP), Kaiser Foundation Health Plan of the Northwest (LLN), and Office of Research Development, School of Nursing, Oregon Health & Science University (NAP), Portland. 

This study was supported in part by grant 4-00500 from the Robert Wood Johnson Foundation Health Care Financing and Organization and by the Kaiser Permanente Northwest Social HMO Medicare demonstration's longitudinal consortium database and health status survey. 

Walter Leutz, PhD, Schneider Institute for Health Policy, Heller School for Social Policy and Management, Brandeis University, Waltham, MA 02454. E-mail: leutz@brandeis.edu. 

POLICY

Annual US health-related productivity losses are estimated to reach some $260 billion, attributable not only to absenteeism but also to presenteeism (being present at work but working at a reduced capacity). The search for remedies has been hampered by the lack of accurate estimates of the loss of productivity and its true costs. To date, little effort has been made to assess the availability of measurement instruments or the validity and reliability of those that exist. 

To systematically review the instruments used to measure productivity loss and its costs and to assess limitations in current research. 

A systematic search was conducted of the published and gray-market research literature from 1995 through 2005 on methods for estimating productivity loss and monetizing that loss. 

Twenty survey instruments were identified that assess the effect of health problems on absenteeism or presenteeism by attempting to quantify self-perceived or comparative impairment or by measuring unproductive work time. Some of the methods have been validated. The challenges of measuring presenteeism far exceed those of measuring absenteeism primarily because many jobs do not have easily measurable output. Methods to estimate the cost of lost productivity were also identified; however, none have been validated, to our knowledge. 

The greatest impediment to estimating the cost of productivity lost to illness is the lack of established and validated methods for monetization. The issues raised in this review are intended to stimulate future research to validate and improve such methods. 

This article reviews available methods to measure and monitor health-related productivity loss, a major concern to employers and policy makers as productivity losses may cost $260 billion annually and may exceed direct medical costs: 

Several methods to estimate the effect of productivity loss on cost have been developed, but none of them have been firmly established and validated to our knowledge. 

Approximately $260 billion in output is lost each year in the United States because of health-related problems, according to the Commonwealth Fund.

Seemingly harmless conditions such as back pain are suspected to translate into substantial losses because of their high prevalence. Stewart and coauthors

estimated the productivity loss attributable to common pain conditions at $61.2 billion annually. Such revelations, particularly at a time when competition has employers searching for ways to cut costs, have sparked the interest of employers, policy makers, and others interested in this issue. 

suggest that the cost of lost productivity may be several times greater than the direct medical costs; furthermore, presenteeism (being present at work but working at a reduced capacity) may account for a larger proportion of losses than absenteeism (being absent from work). For example, in a meta-analysis

of 7 studies that estimated productivity losses, the overall cost of presenteeism was found to account for one fifth to three fifths of the total US dollars lost to 10 costly conditions (which also included absenteeism costs and direct medical costs). Another study

found that days lost because of presenteeism were 7.5 times the number of days lost due to absenteeism when 17 common conditions were considered. 

Awareness of the magnitude of these losses has driven employers and policy makers to search for remedies. Employers believe that better management of chronic conditions might help decrease their costs substantially. For some employers, such data have triggered a paradigm shift in that they have begun to regard their employees as critical assets, with the health-related benefits their organizations provide (such as group health insurance, wellness programs, and disability insurance) as an investment in those assets rather than just costly benefits. Policy makers are intrigued by the opportunity to align the social welfare objective of improving care for chronic conditions with the need to make the business case for quality improvement. 

To pursue those opportunities, instruments that measure the effect of health on productivity and estimate its financial effect with reasonable accuracy are needed. To date, little effort has been made to assess the availability, validity, or reliability of instruments for productivity measurement. This article summarizes the findings of a systematic review of such instruments and interviews with eminent researchers in the field. Our aim was to inform the debate by describing the existing types of instruments and monetary conversion methods and by highlighting limitations and trends in current research. This information can help readers understand what is behind the numbers reported on health-related productivity losses and put such estimates into perspective. The review should also contribute to focusing the research agenda in this important and growing field. 

We identified instruments for measuring the effect of ill health on productivity because of absence from work (absenteeism) or because of reduced performance while at work (presenteeism) through searches of the published and unpublished literature and governmental and corporate communications from 1995 through 2005 using the following search terms: 

. We contacted experts and searched the references of identified articles for additional leads. We retrieved supporting material, such as information on assessment of reliability and validity, for each identified instrument. Finally, we reviewed methods to derive monetary estimates of productivity loss from those instruments and conducted interviews with 5 recognized experts in the field of estimating cost of lost productivity to help put the findings into perspective and to shed light on current research trends. 

We identified 17 survey instruments that assess the effect of respondents' health problems on absenteeism or presenteeism (

). One instrument, the Stanford Presenteeism Scale, exclusively addresses presenteeism, but it is commonly combined with questions on absenteeism. The instruments vary substantially in length (range, 3-44 questions) and scope. Some address only specific conditions, some address a range of conditions, and others address all conditions. We identified several methods for estimating the cost of lost work time. The challenges involved in measuring presenteeism and its costs are far greater than those involved in measuring absenteeism because reduced performance on the job is less tangible than absence. Therefore, most of the findings we report herein pertain to measuring presenteeism. 

Absenteeism is measured by asking respondents how much time they missed from work because of ill health. The recall periods range from 1 week to 3 months. Because such self-reported data have been found to be reliable and valid when the recall periods are short (ie, 1-2 weeks), they can serve as a reasonable substitute for lost time data, which most companies do not routinely collect.

Results derived based on longer recall periods should be viewed with caution. 

Measuring presenteeism is complex. Some attempts have been made to measure presenteeism directly (eg, by telephone call volume per employee in a call center).

However, generating objective data for other types of work would require developing methods to suit the particular characteristics of a given firm, workplace, and profession or job description and collecting data on a regular basis. Furthermore, developing such methods for knowledge-based occupations might be impossible because such workers often produce no easily quantifiable output. 

To overcome these obstacles, researchers have developed instruments that can be applied to various professions and employers.

These instruments focus on the following 3 modes of conceiving presenteeism: (1) assessment of perceived impairment, (2) comparative productivity, performance, and efficiency (with those of others and with one's norm), and (3) estimation of unproductive time while at work. 

The most common approach to measuring presenteeism is assessment of perceived impairment, accomplished by asking employees how much their illnesses hinder them in performing common mental, physical, and interpersonal activities and in meeting job demands. Tools that use this approach include the Health and Productivity Questionnaire, Health and Work Questionnaire, Stanford Presenteeism Scale, Work Limitations Questionnaire, and Work Productivity and Activity Impairment Questionnaire. 

Questions about perceived impairment can range from the general to the specific. The following example of a general question is found in the Stanford Presenteeism Scale: "Despite having my [health problem], I felt energetic enough to complete all my work," to which the employee is invited to respond using a 5-point scale of responses ranging from "strongly disagree" to "strongly agree." An example of a specific question in the Work Limitations Questionnaire requires a respondent to rank on a 5-point scale the difficulty he or she had in using the "upper body to operate tools or equipment." 

Comparative Productivity, Performance, and Efficiency.

 Measuring comparative productivity, performance, and efficiency is another way to capture presenteeism. This method, which is used by the Health and Productivity Questionnaire and the Health and Work Questionnaire, seeks to understand how an employee's performance differs from that of others or from his or her usual performance. 

On a 10-point scale that ranges from "worst ever" to "best possible," the Health and Work Questionnaire asks respondents to rate the overall quality and amount of work produced in the preceding week and how efficiently it was performed. Using a 10-point scale that ranges from "worst performance" to "best performance," the Health and Productivity Questionnaire asks respondents to rate the job performance of workers in similar positions, their usual performance in "the past year or two," and their overall performance during the recall period (4 weeks). The Health and Productivity Questionnaire and the Health and Work Questionnaire include these comparative performance questions in addition to questions about perceived impairments. 

Compared with measures of perceived impairment, measures of self-reported performance have 3 main advantages for expressing presenteeism as a single meaningful number. First, the attempt to benchmark one's perceived performance provides a reference against which loss can be measured. Questions about perceived impairment do not include any conception of what is a standard or usual level of impairment. Second, when based on a 10-point performance scale or a percentage scale, the results can more easily be incorporated into a monetization formula than agreements or disagreements with statements about perceived impairment. Third, attempts have been made to validate employees' self-reported performance evaluation by comparing them with their supervisors' assessments.

Estimation of Unproductive Time While at Work. 

Estimation of unproductive time to assess presenteeism (ie, asking employees to estimate lost time, as is done for absenteeism) is attempted by only a small number of instruments. For example, the Work Productivity Short Inventory asks employees to estimate how many unproductive hours they spent at work during the recall period. Although this approach would lead to the easiest monetization, no study (to our knowledge) has shown that employees can accurately transform their perceived impairments into a temporal measure. 

Validation Studies for Presenteeism Measures

The validity of productivity survey instruments is difficult to establish. While it is conceptually straightforward to validate self-reported measures of absenteeism against factual data of workplace presence or absence, validating presenteeism poses significant challenges because of the nature of the data being collected. For certain types of employment and occupation, such as call centers, employee activity logs are maintained. However, for most jobs there is no true account of productivity with which to assess an employee's performance. Nevertheless, researchers have attempted to validate presenteeism instruments. A detailed summary of published validation studies can be found in the online 

Perhaps more complicated than trying to account for the total time lost by quantifying presenteeism is trying to understand the many competing methods for monetizing (ie, estimating the cost of) lost productivity. These methods comprise the following 3 main types: (1) salary conversion methods, which use survey responses and salary information to estimate productivity loss; (2) introspective methods, which use survey responses as a basis for thought experiments to give businesses an idea of the magnitude of their lost productivity; and (3) firm-level methods, which attempt to monetize productivity losses based on the cost of countermeasures used to deal with absenteeism and presenteeism. 

Salary conversion methods attempt to estimate productivity losses based on self-reported lost time or decreased productivity. The simplest version is the human capital approach (HCA), which expresses the loss as the product of missed workdays multiplied by daily salaries.

 Originally developed for monetizing absenteeism, the method has been extended to presenteeism losses by using self-reported unproductive hours or self-reported percentage reduction of performance instead of missed days.

The obvious attractions of this method are its computational ease, its intuitive plausibility, and its consistency with economic theory (assuming perfectly competitive labor markets) that wages should reflect a worker's marginal contribution to a firm's output. Although its validity has not yet been assessed (to our knowledge), there was consensus among the experts we interviewed that the HCA provides at least a lower-bound estimate for the true cost of lost productivity (telephone interviews with Sean Nicholson [February 22, 2005] and Thomas Parry, PhD [February 15, 2005]). To increase the accuracy of the estimate of productivity losses, one expert suggested also including the cost of fringe benefits (telephone interview with Ronald C. Kessler, PhD [February 8, 2005]). The HCA is the method typically used in studies reporting the economic effect of health-related productivity losses. Depending on the available data sources, authors have used actual salaries of the respondents,

An extension of the HCA is the team production model developed by Pauly and colleagues,

who argue that simple salary-based conversion is appropriate for workers performing discrete tasks in isolation, but the model fails to take into account the interdependence of job functions in the modern economy. For example, if the only surgeon in a hospital stayed home sick, the entire operating room would remain idle, causing much greater losses than just the surgeon's salary. The authors proposed to operationalize this interdependence by the following 3 criteria: (1) the replaceability of an employee, (2) the extent to which an employee works as part of a team, and (3) the time sensitivity of an employee's work.

Initial empirical work by Nicholson and colleagues

derived a set of multipliers that can be applied to a worker's salary for 35 different job categories. Simple jobs, like that of a fast-food cook, have a multiplier of 1.00, suggesting that the productivity loss equals the actual salary, while more demanding occupations, such as construction engineering, have higher multipliers that reflect overall effect on the firm. Different multipliers are used for short-term (3 day) and long-term (2 week) absences. Ongoing work aims at a larger set of multipliers and at methods to capture the interaction between medical conditions and job characteristics (telephone interview with Sean Nicholson, PhD [February 22, 2005]). This approach has 2 practical challenges. First, a large library of multipliers would have to be created and maintained. Second, the method is based entirely on individual-level characteristics and does not take firm-level factors into account. For example, it is conceivable that the absence of an analyst would have different implications for a consulting firm than for a not-for-profit research organization. Other firm-level factors, such as unionization and competitive position, may also modify the effect of loss of productivity in a given job category. 

A more fundamental challenge was posed by Koopmanschap and colleagues,

who argued that the HCA overestimates the true absence-related productivity losses because short-term absences might be partially compensated with greater effort or unpaid overtime, whereas longer-term absences would lead to replacement of workers with new hires. Based on those considerations, the authors proposed the friction cost method that aims at estimating only the actual lost production, as opposed to the potential lost production estimated by the HCA. They tested their method on national data from the Netherlands and found the estimates of lost productivity to be consistently lower than those derived by the HCA.

We identified no attempt to apply the method to US data or to data at the company level. Other authors have challenged the friction cost method as inconsistent with concepts of standard economic theory, such as opportunity cost and profit maximization.

However, this discourse remains largely theoretical at this point because neither of the salary conversion methods has been evaluated empirically, to our knowledge. 

Introspective methods reflect an attempt to overcome the theoretical and practical challenges of converting self-reported productivity reduction into monetary units. Some researchers have argued that conversion should be abandoned in favor of providing guidance to firms on deriving their own estimates: for example, managers would be provided with an analysis of the productivity survey and asked to consider questions such as "How much would you be willing to pay a contractor who can raise everyone's productivity by 20%?" or "How many full-time employees could you cut if the productivity of your chronically ill workers increased by 20%?" (telephone interview with Ronald C. Kessler, PhD [February 8, 2005]). Another approach is to encourage managers to estimate the revenue that various staff members contribute and to use this number for conversion (telephone interview with Ronald C. Kessler, PhD [February 8, 2005]). The aim of such thought experiments is to illustrate the magnitude of the problem rather than to derive precise estimates. Although helpful, their validity remains untested (to our knowledge), and their results have not been compared with those of the HCA approach as far as we know. 

Firm-level methods represent a logical extension of the introspective methods and use a topdown approach that assesses firm-level information to derive cost estimates for lost productivity. These methods are based on the premise that managers have a good sense of how their company's productivity is affected by health-related problems and use countermeasures to deal with them. For example, they may have redundant staff to compensate for absences, or they may hire temporary workers or offer overtime payment to maintain output. Alternatively, they could forgo revenues. Economic theory suggests that a competitive firm combines these different strategies to maximize profits. Therefore, information about a firm's cost for those countermeasures can be used to approximate its lost productivity. The attraction of this approach is that it does not require detailed individual-level data and that the cost of many of the countermeasures (such as the fees paid to temporary employment agencies) is easy to quantify. The downsides are that some of the cost may be intangible and that forgone revenue estimation must rely on a manager's perceptions. It may also prove difficult to elicit countermeasures to presenteeism as opposed to absenteeism because presenteeism is not immediately visible to a firm and may not provoke a conscious response. Furthermore, the correct attribution of the cost items to health-related productivity losses needs to be assured, as (for example) part of the temporary staff could also be part of a firm's usual staffing mix. 

As for other firm-level methods, empirical evidence remains sparse. One study

used staffing cost to cover short-term disability absences to estimate productivity losses, but no published evidence was found of attempts to generalize this approach into a broader framework for measurement. 

The interest in measuring and monetizing the effect of health on corporate productivity has resulted in the development of numerous instruments to capture this important concept. Most have undergone validity testing and have gained acceptance as reliable tools for research and benefits decisions. Among those instruments, the biggest gap remains the lack of an established and validated method to derive monetary estimates of the cost of lost productivity. Although many users are comfortable in applying a salary conversion method and believe that it provides at least a lower-bound estimate, a more rigorous evaluation of this method and its alternatives is warranted before it becomes the basis for potentially far-reaching policy and managerial decisions. A first step could be to benchmark the different methods used to determine whether and to what degree estimates based on different methods differ. 

Conducting such research is by no means a straightforward task because direct measurement of job productivity is difficult, particularly in knowledge-based occupations. This mandates that a research agenda should involve multiple disciplines or job descriptions. It should also reflect the interest of various stakeholders, as (for example) the weight of evidence that businesses will require for operational decisions will differ from the standards of the research community. Any method will also have to be tailored to the precise question it aspires to answer. For example, the social welfare perspective on measuring cost is different from a corporate perspective, and various sectors of the economy may require different approaches. The review herein should stimulate interest in this field and contribute to endeavors that push it forward. 

We thank Ronald C. Kessler, PhD, Sean Nicholson, PhD, and Thomas Parry, PhD, for contributing their expertise to this review. We are indebted to Ron Loeppke, MD, MPH, for his input and guidance and Emily Taylor for her help with preparing the manuscript. 

This study was supported by CorSolutions, Inc through a contract with RAND Corporation. 

Soeren Mattke, MD, DSc, RAND Corporation, 1200 S Hayes St, Arlington, VA 22202. E-mail: mattke@rand.org. 

METHODS

A Review of Methods to Measure Health-related Productivity Loss

To examine recent trends in childhood immunizations recommended by the Advisory Committee for Immunization Practices measured by the Health Plan Employer Data and Information Set (HEDIS) and to describe the factors associated with higher rates over time. 

The HEDIS performance measures from 1999 to 2002 and plan characteristics include approximately 400 enrollees per plan each year. 

Longitudinal regression analysis of commercial managed care organizations' HEDIS measures. The outcome measure was the proportion of children aged 24 to 35 months in the plan who received 4 doses of diphtheriatetanus-pertussis vaccine, 3 doses of polio vaccine, 1 dose of measles-mumps-rubella vaccine, 3 doses of Haemophilus influenzae type b vaccine, and 3 doses of hepatitis B vaccine. 

The mean immunization rate for health insurance plans increased from 65.7% in 1999 to 67.9% to 2002. Plans that reported publicly had higher childhood immunization rates than plans that did not report publicly (

< .001). Plans with higher proportions of Hispanics or African Americans had lower childhood immunization rates (

< .001). Immunization rates varied significantly by type of visit; plans with higher proportions of children making visits to their primary care physician had higher rates of immunization (

Managed care organizations' performance measured by childhood immunization rates varies by organizational and demographic factors. Our findings suggest that plans should ensure efficient and accurate data collection systems and should encourage their providers to assess for immunizations at sick-child and well-child care visits. 

Health plan decision makers should be aware of the following regarding childhood immunization rates: 

Plans that reported publicly had higher childhood immunization rates than plans that did not report publicly (

Immunization rates varied significantly by the type of visit. Plans with higher proportions of children making visits to their primary care physician had higher rates of immunization, whereas plans with higher proportions of infants making sick-child care visits (outpatient or emergency department) had lower rates of immunization (

Measures of quality give purchasers of healthcare and consumers of this care information about how insurance plans perform. The Health Plan Employer Data and Information Set (HEDIS) from the National Committee for Quality Assurance (NCQA) provides comparative information across health plans to measure the quality of care and preventive services for health plan beneficiaries. More than 66% of commercial health plans report to NCQA, representing 85% of the commercially enrolled managed care organization (MCO) population, or approximately 70 million to 73 million enrollees.

Results have shown that differences in childhood immunization rates vary depending on the collection methods (eg, National Immunization Survey, National Health Interview Survey, or HEDIS).

Investigations using HEDIS data have described factors associated with immunization rates (eg, physician credentials)

; however, little is known about factors associated with immunization rates accounting for time. For HEDIS, reporting these associations can help administrators of health insurance plans in identifying characteristics that are more predictive of delivery of immunization services. This knowledge can assist in the development of strategies to maintain high immunization rates, as well as inform potential purchasers of insurance plans about what plan characteristics are associated with sustained high immunization rates over time. The method of longitudinal regression analysis has been rarely used when examining HEDIS data, although its necessity has been acknowledged.

Because 6 or more encounters with a healthcare provider are needed in the first 2 years of life to fully immunize a child, timely receipt of childhood immunizations, captured in the immunization rates, are an indication of how well the healthcare delivery system is working. Influential employers such as General Motors, a company that has given employees significant financial incentives to choose health plans with better HEDIS scores, emphasize childhood immunization performance among important measures to assess.

In the present study, longitudinal regression analysis was used to examine the factors associated with higher childhood immunizations rates reported by public reporting and nonpublic reporting commercial health plans to the NCQA. 

The analyses included the HEDIS data sets from 1999 to 2002 for public reporting and nonpublic reporting commercial health insurance plans (for care received from 1998 to 2001). Plans included in the analyses were operating in all US states (except Montana), in Guam, and in Washington, DC. The sample sizes were 423 plans in 1999, 383 plans in 2000, 371 plans in 2001, and 332 plans in 2002. Variables examined from the Consumer Assessment of Health Plans survey included the proportions of respondents by race/ethnicity.

Variables were examined from 8 HEDIS measurement categories.

The NCQA accreditation status changed during the years, so plans were considered accredited if their status was excellent, commendable, full accreditation, or 1-year accreditation; otherwise, the plan was not considered accredited by the NCQA. Because this study constituted analysis of secondary data without identifiers, it did not require a review by the Centers for Disease Control and Prevention Institutional Review Board, and consent was not required. 

For HEDIS clinical measures (eg, immunizations) that require medical record review, a sample of 411 enrollees is required to obtain a valid rate. If the plan has fewer than 411 eligible persons for a particular measure, all eligible enrollees are sampled. For measures that use administrative data only (eg, child access to a primary care provider), the health plan samples the entire eligible population. For the Consumer Assessment of Health Plans survey, a minimum of 100 responses is required for reporting particular rates. 

For immunization measures, children were systematically selected from the enrollment cohort who reached their second birthday in the reporting year and who met the continuous enrollment criterion of no more than 1 gap in enrollment of up to 45 days during the 12 months before their second birthday. Children identified as having contraindications (eg, anaphylactic reaction to a vaccine or its components) were excluded from all immunization statistics.

Childhood immunization statistics from plans were calculated and reported (1) by using only administrative records specific to the plan (administrative method) or (2) by using administrative data supplemented with medical record data (hybrid method), which may contain information about immunizations received outside of the plan. To avoid ascertainment bias and to maintain data comparability, only plans using the hybrid method were included for immunization measures. Few plans that used the administrative method were excluded (16 plans in 1999, 5 plans in 2000, 11 plans in 2001, and 8 plans in 2002). 

The plan characteristics examined were those groups of measures, defined by NCQA, that have been shown previously in the literature to be associated with receipt of immunizations

 or those that could theoretically be associated with preventive services, including receipt of immunizations. These domains include access and availability of care and health plan stability. Health plan descriptive characteristics included NCQA accreditation status, region of operation, public reporting status, vaccine policy of states of operation, enrollment size (below or above the median), race/ethnicity proportions (based on the sample of Consumer Assessment of Health Plans survey respondents), and MCO type (eg, health maintenance organization, point of service, or combined health maintenance organization and point of service). The Vaccines for Children program is a federal entitlement program that supplies public and private healthcare providers with federally purchased vaccines at no cost for administration to eligible children. Vaccine policies include (1) Vaccines for Children program only, (2) Vaccines for Children program and underinsured, (3) universal,

and (4) multiple states, in which the insurance plan operates in several states with various vaccine policies. Because of policy changes, the vaccine policy variable was coded for the immunization policy in effect in the state of operation for each year that a plan reported to the NCQA. Variables assessing access and availability of care were calculated and reported using the administrative method and included the proportion of board-certified providers per number of enrollees, proportion of total providers per total commercial enrollees, proportion of children aged 25 months to 6 years who made visits to their primary care physician during the reporting year (visits are for well-child or sick-child care), and proportion of outpatient sick-child care visits and emergency department visits made by children younger than 1 year annually. Variables used to assess health plan stability included the number of years in business by MCO type and the practitioner turnover rate for the reporting year. Plans reporting to be in business for more than 25 years were considered as being in business for 25 years. 

Because the NCQA releases data aggregated by managed care plans rather than by enrolled individuals, the unit of analysis was the plan. Childhood immunization statistics were calculated as the proportion of children sampled in the plan who received 4 doses of diphtheria-tetanus-pertussis vaccine, 3 doses of polio vaccine (inactivated from 1999 to 2002 and oral vaccine in 1999), 1 dose of measles-mumps-rubella vaccine, 3 doses of 

type b vaccine, and 3 doses of hepatitis B vaccine as recommended by the Advisory Committee for Immunization Practices. Immunizations were only included if they were determined to be valid according to HEDIS criteria (

). Because of the vaccine shortages in 2001 and 2002, plans had the option to report the previous year's immunization statistics if the plan or entity did not change (ie, did not merge or split into different plans) and they had reported in the previous year.

To include all available data, all variables of interest reported by these plans were included except for the redundant immunization statistics reported during 2 consecutive years. 

Data were examined using SAS software version 9.1.3 (SAS Institute, Cary, NC). Data were modeled longitudinally using PROC MIXED to fit linear mixed models that included a random intercept, an unstructured covariance, and a robust variance estimator, the sandwich estimator, which was used to calculate valid standard errors while accounting for misspecification of the covariance and a normal distribution. Plan characteristics were fixed effects. An advantage of fitting linear mixed models is that PROC MIXED uses all available data in the analysis.

Therefore, the plans that opted to report the previous year's immunization data during the years of shortage were included in the model using their independent variables, although the dependent variable data were missing. In the bivariate analyses, variables of interest were modeled with the year of report and an interaction term (eg, MCO type) to determine if they were significantly associated with the change in proportion immunized over time at 

< .05. Testing for interactions among variables was performed (NCQA accreditation status with region, public reporting status, and proportion of visits with primary care physicians), and region was tested with race/ethnicity proportions. A significant interaction term consisting of an independent variable with time (year or report) indicates that one group's rate of change is faster than another group's rate of change. Backward elimination was used to determine the multivariate model; fit statistics were used to determine the best fit of the model, and variables significant at 

The rates for full immunization remained below 70% during the study period. The mean + standard deviation (SD) rates were as follows: 65.7% + 13.5% (range, 0.0%-93.6%) for 1999, 63.3% + 14.9% (range, 5.5%-92.9%) for 2000, 65.6% + 13.7% (range, 11.9%-90.5%) for 2001, and 67.9% + 12.2% (range, 17.6%-88.6%) for 2002 (

Two hundred forty-seven plans reported for 1 year, 132 plans reported for 2 years, 110 plans reported for 3 years, and 147 plans reported for 4 years. Health plan characteristics are given in Table 2 and 

for continuous and categorical variables, respectively. The proportion of publicly reporting plans increased from 67.7% in 1999 to 83.7% in 2002 (Table 3). Similarly, the proportion of NCQA-accredited plans increased from 48.3% in 1999 to 65.2% in 2002. The proportions of enrollees by race/ethnicity remained steady during the 4 years (Table 2). 

Factors Associated With Higher Immunization Rates

Bivariate Longitudinal Regression Analysis. 

Many analyzed plan characteristics were significantly associated with immunization rates over time (

). Time effects were significant for all the independent variables; interactions between independent variables and time were not significant except for region and proportion of Pacific Islanders. 

Multivariate Longitudinal Regression Analysis. 

To account for the variance within plans across years, we allowed the intercept to vary randomly. Plan characteristic variables were fixed effects. Factors that remained in the final multivariate model at 

< .05 included region, year of report, NCQA accreditation status, number of years in business, proportion of Hispanics, proportion of Pacific Islanders, proportion of African Americans, enrollment size (≥ 120 000 vs < 120 000), reporting status (ie, public vs nonpublic), and proportion of children aged 25 months to 6 years who made visits to their primary care physician during the measurement year. The only significant interaction detected was between year of report and region (

). Although immunization rates dropped from 1999 to 2000, they increased significantly from 1999 to 2001 (

= .04) and increased nonsignificantly from 1999 to 2002 (

= .65). Publicly reported plans had 3.2 percentage point higher immunization rates than nonpublicly reported plans. The NCQA-accredited plans had 2.6 percentage point higher immunization rates than the nonaccredited plans. For each 10 percentage point increase in the proportion of members in health plans that were of Hispanic or African American race/ethnicity, plans had a 2.0 percentage point decrease in immunization rates, and for Pacific Islanders, plans had a 6.0 percentage point increase in immunization rates. For each 10 percentage point increase in the proportion of children aged 25 months to 6 years visiting their primary care physician in the year of report, plans had a 2.0 percentage point increase in immunization rates. Insurance plans with at least 120 000 enrollees had a 2.2 percentage point increase in immunization rates compared with plans that had fewer than 120 000 enrollees. Insurance plans that reported a region of operation other than the northeast had significantly lower immunization rates. The only significant interaction found was that the central region's fully immunized proportion dropped significantly and the northeast's fully immunized proportion increased faster during 2001 as opposed to 1999. 

After controlling for covariates, the mean predicted immunization rates were 65.36% for 1999, 63.45% for 2000, 66.58% for 2001, and 67.10% for 2002. Although the mean + SD rate decreased by -1.96 + 0.86 in 2000 from 1999, it was not statistically significant (

April 2007

Impact of Chest Radiography Screening on Healthcare Spending

Selection Bias Between 2 Medicare Capitated Benefit Programs

A Review of Methods to Measure Health-related Productivity Loss

Managed Care Organizations' Performance in Delivery of Childhood Immunizations (HEDIS, 1999-2002)

Overcoming Information Asymmetry in Consumer-directed Health Plans

Employee Costs Before and After Treatment Initiation for Bipolar Disorder

Impact of Telephonic Interventions on Glycosylated Hemoglobin and Low-density Lipoprotein Cholesterol Testing