© The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org.
Creating Standard Cost Measures Across Integrated Health Care Delivery Systems
Affiliations of authors: Clinical Research Unit, Kaiser Permanente Colorado, Denver, CO (DPR); HealthPartners Research Foundation, Minneapolis, MN (MJG, MVM); Henry Ford Health System, Detroit, MI (JEL); Kaiser Permanente Northwest, Center for Health Research, Portland, OR (RM, MCH); Center for Health Studies, Group Health Cooperative, Seattle, WA (PAF)
Correspondence to: Debra P. Ritzwoller, PhD, Kaiser Permanente, 580 Mohawk Dr., Boulder, CO 80301 (e-mail: debra.ritzwoller{at}kp.org).
 |
ABSTRACT |
|---|
 TopNotes Abstract IntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
Background: Economic analyses are increasingly important in medical research. Accuracy often requires that they include large, diverse populations, which requires data from multiple sources. The difficulty is in making the data comparable across different settings. This article focuses on how to create comparable measures of health care resource use and cost using data from seven health plans and delivery systems participating in the Cancer Research Network's HMOs Investigating Tobacco study. Methods: We used a data inventory to identify variation in data capture across sites and used data dictionaries to develop algorithms for assigning standardized cost to the three major components of health care use: outpatient, inpatient, and pharmacy. Results: The plans included in this study varied from fully integrated, closed-panel models to plans and delivery systems that include network or independent physician association components. Information derived from the data inventory and data dictionary instruments demonstrated a substantial variation in both the content and capture of data across all sites and across all components of usage. The methods we employed for cost allocation varied by usage component and were based on our ability to leverage the data points available to best reflect actual resource use. Conclusions: The importance of this article is the method of ascertaining, cataloging, and addressing the within- and between-plan differences in health care resource use. Second, the decisions we made to address the differences between health plans provide other researchers a starting point when creating a cost algorithm for multisite retrospective research.
|
INTRODUCTION |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
Increasingly, decision makers want to understand the costs associated with the delivery of medical care. They often want to understand whether patient survival, quality of life, satisfaction, or cost differ by treatment modality or by how health care is organized and delivered. Addressing such questions requires an understanding of the costs associated with the resources used to deliver and manage health care, often across diverse health care settings. Within large, prospective studies as well as within retrospective studies, it is rarely feasible to directly observe cost data (1). When such cost data are available, variation both among organizations and over time within an organization can obscure real cost differences. From a research perspective, the goal is to obtain cost estimates that are consistent across place and time.
Cancer-specific data on the cost of treatment have been used both for the purpose of constructing aggregate economic burden of illness measures and as inputs into many cost-effectiveness analyses of cancer prevention, screening, and treatment interventions (2). Sources of data for these estimates have included resource based estimates collected in the context of single or multiple institution clinical trials, data from nationally representative surveys such as the Medical Expenditure Panel Survey, and data from health insurance claims systems, most notably, using linked data between SEER and Medicare (3–5). Because the Medicare program does not cover nondisabled persons under the age of 65 years and because detailed Medicare claims data have not been available in the past for Medicare-managed care enrollees, cancer cost of treatment estimates have also been constructed using the data systems of individual HMOs (6–8). Although patterns of cost generated SEER–Medicare and HMO data were roughly consistent, there has been no effort in the past to systematically examine how HMO-based cost estimates might differ from Medicare estimates at the level of specific resource and service use and/or by unit costs. Likewise, there has been no effort to examine how these components might differ across HMOs, especially across HMOs with differing plan structures and administrative data systems.
The purpose of this article is to describe the process of creating comparable cost measures using the automated data routinely available within and across multiple health plans and integrated delivery systems. Our strategy is to identify comparable measures of resource use and to derive standardized estimates of cost. If there is variation in how resources are measured, we may not be able to detect real differences in resource use and cost.
To illustrate the challenges in constructing comparable costs estimates as well as possible strategies to overcome these challenges, we use data from seven health plans with integrated delivery systems that are participating in the National Cancer Institute–funded Cancer Research Network (CRN). The health care delivery systems participating in the CRN are: Group Health Cooperative, Harvard Pilgrim Health Care, Henry Ford Health System/Health Alliance Plan, HealthPartners Research Foundation, the Meyers Primary Care Institute of the Fallon Healthcare System/University of Massachusetts, and Kaiser Permanente in six regions (Colorado, Georgia, Hawaii, Northwest [Oregon and Washington], Northern California, and Southern California). The 11 health plans have nearly 10 million enrollees. The CRN conducts collaborative research on variations in cancer prevention and treatment policies and practices. We use these data to develop algorithms for consistently assigning standardized cost to health service utilization. The CRN was funded by the NCI to take advantage of the natural laboratories and large defined population base of these nonprofit integrated healthcare organizations (http://www.crn.cancer.gov). Most CRN members are long-standing organizations with a stable presence in their communities. The participating plans all began as group/staff model HMOs, but they have all changed structures in recent years. As with the managed-care industry nationally, several CRN member plans are expanding and merging to become more heterogeneous in their practice arrangements. Specifically, several plans now have network or independent physician association (IPA) components.
The immediate product of this research is an analysis of the health care costs of smokers, former smokers, and never smokers identified under CRN Project HIT (HMOs Investigating Tobacco) (9–11). Our objective was to understand the differences in health care resource use and cost among the three cohorts of patients (smokers, former smokers, and nonsmokers) included in Project HIT by pooling data across health plans. We did not aim to generate plan-specific cost estimates, and therefore we sought to limit the effect of site variation.
|
VARIATION IN RESOURCE USE IDENTIFICATION |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
There are several sources of variation that are common in multisite research. Fishman et al. (12) have suggested that there are three major sources of resource cost variation: approaches to care, costing method, and coding system. Our goal in this analysis was to seek methods that would minimize the cost variation associated with the last two sources. At the health plan level, these sources of variation can be linked to two general organizing concepts, integration and incentives.
Health plans and delivery systems vary substantially in their organizational structure. The traditional closed-panel health plan is tightly integrated with substantial centralization of administration, medical management, information systems, and other aspects of care. All or nearly all care is provided in plan-owned-and-operated facilities by clinicians seeing only patients insured by the system's financial arm. These plans may be contrasted with organizations like HealthPartners or Henry Ford Health System that provide services through a range of community providers, pharmacies, and hospitals.
These differences in system integration generate differences in the resource use identification and cost estimation. More integrated systems can create a single electronic patient record with standardized test ordering and procedure coding. A drawback of this can be the creation of site-specific procedure codes that cannot be consistently valued compared with similar procedures at other sites. Conversely, in less integrated delivery systems, each clinic is likely to have a different information system for capturing the information from a patient medical encounter. Most commonly, less integrated systems like HealthPartners and Henry Ford Health System collect data in the same form that Medicare uses, a standard claim form (13). Claims, although providing fairly detailed information, are constrained by limits of the coding systems in place.
System ownership also plays a role in measuring resource use. When an integrated delivery system owns a facility, it may create its own system of cost accounting to track resource usage. For example, if one study site uses community hospitals paid on a Diagnosis-Related Groups (DRG) basis while another site's hospital care is provided in an owned hospital, the reported costs in the latter are usually a function of the internal cost accounting methods, whereas in the former the DRG-based resource use is a function of national average usage.
Provider incentives also affect measurement of relative resource use. In capitated environments, a busy clinician has limited incentive to fully code each procedure that was performed during a clinical encounter. Since additional coding does not increase physician's income, the extra time may not be spent on coding additional procedures. Conversely, a physician who sees mainly fee-for-service (FFS) patients has an incentive to code every reimbursable procedure (14,15).
The clinician coding for FFS reimbursement is more likely to reflect all procedures, but there is also an incentive to upcode. In that case, resource use is overstated to enhance reimbursement (16). The most common historically cited example is the disaggregation of procedures that are performed simultaneously into two separate procedure codes or the incorrect use of evaluation and management codes assigned to each face-to-face clinical encounter (17).
Time is the other variable that can seriously affect resource identification. In addition to the well-known economic issues of inflation and discounting, the much more important questions relate to changes within a health plan. Since health plans participating in research take place in highly competitive and rapidly changing environments, researchers should not assume that systems will remain static. For example, at many plans in our study there was a transition from paper medical records to electronic medical records (EMRs) during the study period. In other plans, mergers and changes in business models caused major changes in organizational design, incentives, and data flow.
As well described in Meenan et al. (18), administrative data derived from health plans and delivery systems can generally be categorized as either claims based or encounter (transaction) based. Both integrated and network-based health delivery systems often use claims data to track usage. Claims systems generally use nationally standardized record layouts and coding systems for key variables: ICD-9 diagnostic coding; Current Procedural Terminology, version 4 (CPT-4) procedure coding; and National Drug Code (NDC) for prescriptions. With claims data, recorded procedures can be submitted by providers on a per-procedure basis or bundled by multiple dates of service. Every reimbursable service performed has a separate procedure, codes are often chosen to maximize reimbursement, and multiple face-to-face encounters with different medical professionals may be bundled in a single submission. Health plans and delivery systems often use a paid amount derived from claims as a cost proxy; however, this approach may not approximate the true resource cost of providing these services. Furthermore, linking specific services across providers to underlying events such as inpatient stays, visits, and episodes remains problematic (18,19).
Encounter systems capture services provided or produced internally, including outpatient and pharmacy in systems that do not contract with a pharmacy benefit manager. Encounter systems record services performed by physicians and other allied health professionals. Prior to the advent of EMRs, diagnosis and procedure codes were often provided on preprinted encounter forms chosen from among the most common codes. Within many EMRs it is an option to freely input text for details of the event, rather than use options within the system that map to specific diagnoses or procedure codes. Often, in all systems, there is no indication of a principal procedure or principal diagnosis.
|
METHODS |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
We used data from seven of the nine plans that participated in the HIT project. They include Group Health Cooperative (GHC) based in Seattle, WA; Henry Ford Health System/Health Alliance Plan of Michigan (HFHS); HealthPartners Inc., in Minnesota (HP); and the Kaiser Permanente health plans of Colorado (KPCO), Hawaii (KPH), Northern California (KPNC), and the Northwest (KPNW). At least one economist or cost management expert from each site was included in the CRN economics investigative team. Table 1 describes the 1998 characteristics of each plan included in this analysis.
|
A sample of more than 20 000 adult enrollees was included in our study population. Eligible subjects were all adults aged 25 years and older who were continuously enrolled in one of the seven participating health plans for 2 years as of August 1999. Subjects making primary-care visits were surveyed regarding their current smoking status and receipt of plan smoking cessation services (3,4). Table 2 provides descriptive data associated with the study population (11,12).
|
In this analysis, we define resource use as the direct monetary costs of treating an individual with an illness: hospitalizations, physician services, and prescription drugs incurred by the health plan or delivery system. This definition of resource use excludes some costs borne by the patient, such as transportation and time lost to normal activities but may include patient copayments or coinsurance.
Our method for developing the multisite costing algorithm for this population was centered on two types of instruments: a data inventory and detailed data dictionaries for inpatient services (including hospital outpatient but excluding long-term care such as skilled nursing), outpatient services (primary, specialty and emergent care), and outpatient pharmacy. These three components of health care account for approximately 90% of the cost of care in our delivery systems. The data inventory was designed to answer who, where, what, why, and when. Who relates to both the clinician and to the patient. Where relates to the place of service, such as inpatient, outpatient, or emergency department. What describes the procedures that were administered during an encounter. Why is the diagnosis. When describes the dates of service. The inventory sought to find out how, and in what detail, could we identify these elements at each of the seven participating health plans. Table 3 is an example of an unidentified site data inventory. To support data collection derived from the HIT parent project, we determined that January 1, 1995, through December 31, 2001, would be the optimal period that the most consistent data could be collected within and across the participating plans.
|
We sought to capture the details associated with as many of the factors as possible that may contribute to resource use variation in outpatient settings. These factors include: 1) service location, which can vary from an emergency department to an injection room where a nurse administers influenza vaccinations; 2) provider type, ranging from nurse practitioner or physician's assistant to cardiac surgeon; 3) department or specialty, in which much of the encounter system information is captured in department-specific databases; 4) diagnoses and procedure codes, since differences in resource intensity are identified with coded procedures in claims-based systems, but are less likely to be coded in a standard metric in encounter-based systems, due to less incentive for complete coding. The costing algorithm goal is to assign a value to what happened during each visit, so we deemed each of these variables important to capture for valid resource allocation.
Variables in the outpatient data dictionary included the following: service date, location (ED, medical office, hospital outpatient), provider type (MD, NP, PA, RN), MD type (e.g., primary, specialty), length of visit in minutes (either by Evaluation and Management codes or by internal codes), principal diagnosis and procedure, secondary diagnoses and procedures, estimated internal cost (if available), billed charges (if claim derived), and paid amount charges (if claim derived).
The cost algorithm must assign a standard value to each service performed. In claims-based systems, CPT-4 codes map to the Medicare Resource Based Relative Value System (RBRVS). The RBRVS assigns a single relative value unit to each procedure code. For Medicare payments and many other insurance contracts, this relative value unit (RVU) is multiplied by a fee schedule to derive a cost. Inflation can be removed by using a single year's fee schedule as the multiplier. Encounter systems present different problems. Aside from coding incentives, there is neither a standard set of weights that can be assigned nor a direct method of inflation adjustment. For the encounter sites, we incorporated the time of the visit, as noted in minutes. To make this information comparable, we mapped time to the CPT-4 Evaluation and Management (E&M) codes. For encounter data, we also took into account the variation in the distribution of visits by location, department, and provider type.
Inpatient stays are a critical component of any costing algorithm. Medicare and many health plans and delivery systems separate the facility cost from the professional or physician costs of a stay. The DRG payments used by Medicare apply only to the facility costs and do not capture resources associated with physician time. Thus, the inpatient data dictionary includes both facility and professional/technical services delivered during an inpatient hospital stay. The data dictionary for the standardized inpatient files included the following: service begin and end date, discharge status (e.g., home, discharged to skilled nursing facility, hospice) principal diagnosis and procedure, secondary diagnoses and procedures, DRG (if available), hospital affiliation (owned, external with contract, external with no contract), health plan cost (estimated by internal cost system), billed amount, and paid amount (i.e., by contract).
The outpatient pharmacy is a critical and rapidly growing component of resource use. However, health plan's pharmacy cost data are subject to substantial restrictions. Contracts with drug manufacturers often provide discounts in return for an agreement not to disclose the acquisition cost of the drug. Moreover, the lack of information regarding dispensing costs makes the acquisition cost an inaccurate measure. Also, plans using community pharmacies will not know the acquisition cost. Thus, the largest challenge to the pharmacy working group was to assign standardized costs to pharmacy data that recognized all three of these barriers but still generated meaningful cost values that were consistent with the experience of the health systems participating in the CRN.
The data dictionary for standardized pharmacy files includes all outpatient pharmacy dispenses. The following fields are included: dispense date, national drug code, days supplied, therapeutic class, dispensing notes (if available), quantity supplied, days supplied, unit codes, route of administration (e.g., inhaled, oral), product name, internal cost (combined acquisition and dispensing, if available), and average wholesale price (AWP).
Once data were collected from the participating plans, the working groups compared the supplied files across sites for completeness and then modified the preliminary cost algorithms. Each working group sought to find methods of creating the most detailed, yet still accurate, resource use estimates for each person receiving services at a participating health plan. Due to the proprietary nature of health plan and delivery system data, we present a subset of results derived from these instruments in tabular form anonymously by site.
|
RESULTS |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
Outpatient Data
The data inventory indicated that the most integrated plans could identify the facility, department, and provider type. However, in the mixed-model health plans and delivery systems, very little information was available for these variables largely because there is no differentiation in the dominant CPT-4 codes by department, specialty, or training level. Claims-based systems had limited ability to identify specialty without substantial review. Therefore, we limited our definition of place of service to medical office, hospital outpatient, or emergency department. The data inventory also showed the claims-based systems to be advantageous with respect to the capture of resource intensity because they, at least theoretically, coded each service that was performed using standardized coding.
Substantial variation was found in outpatient data derived from the data dictionary. Data on more than one million outpatient encounters were evaluated. Five of seven sites captured plan-specific costs for internally produced visits; four of seven sites captured paid amount for externally provided visits; and five of seven sites captured provider type (MD, NP, other), specialty (PC, OB/GYN, Spec), and place of service (medical office, ED/UC, Hosp) for internally derived visits. As noted in Table 4, the percentage of observations containing one or more E&M codes or capture of visit time (in minutes) varied by site with a range of 2%–99%, with an average of 52%. The percentage of observations with a noted principal diagnosis varied from 0 at one site to 99% at another, with an average of 66%. The average number of observations with a noted principal procedure varied from 0 at one site to 100% at another, with an average of 56%.
|
Given the diversity of outpatient data, the only feasible approach was to create synthetic cost estimates by leveraging the data points at some sites to impute or back fill data points that were missing at others. Optimally, an observation associated with an outpatient event would contain information on the service date, type of visit or procedure, place of service, provider type, and cost—either plan specific or paid amount derived from a claim. When plan-specific cost information was unavailable for a given visit or procedure, using either available CPT-4 code or by mapping the estimated length of a visit to an E&M code, we then estimated costs using a method based on resource-based relative value units. If both CPT-4 and cost were missing, we imputed cost based on place of service and provider type and/or specialty of provider. If one site's outpatient costs were extremely high, relative to average, we made adjustments based on age, sex, relative case-mix, or the prevalence of comorbidities, along with adjustments based on average costs for that type of event derived from pooled data derived from all sites.
Inpatient Data
In our data inventory, we discovered substantial variation in how different health plans identify the services delivered during an inpatient stay. For facility costs, the data elements that were captured most consistently across health plans and systems for both owned and contracted hospitals included discharge diagnoses and length of stay, as measured by admit and discharge dates. The availability of DRG codes, revenue codes, and procedures performed by hospital staff varied across all sites. Incomplete capture of professional services was a key obstacle in tabulating complete inpatient costs. Many of the integrated plans did not separately capture physician services provided in the inpatient setting. At other health plans, incomplete data on encounter location made it impossible to say with certainty which professional services were associated with an inpatient stay. Like outpatient data, a major source of variation was facility ownership. In five of seven plans, inpatient services were provided at both plan-owned and contracted community hospitals.
We collected data on 14 644 inpatient hospital stays over the 5-year study period. There was variation across sites with respect to the capture of DRGs, diagnoses, and procedures (Table 4). Although six of the seven sites captured DRGs for inpatient stays, the variation ranged from 43% to 99% across sites. Principal diagnosis was available at all sites for most inpatient stays. However, much higher variation existed with respect to the capture of secondary diagnoses with a range of 44% to more than 90% of cases by site. Principal procedure codes were also available at all sites for most inpatient admissions. However, the degree of completeness across all observations ranged by site from 64% to 92%. Capture of secondary procedures was much less complete with less than 15% of all sites capturing up to four procedures per inpatient stay.
We made the following decisions on how to deal with these inpatient cost data issues. First, because DRGs were either available for all health plans or could be assigned using available data elements and DRG grouping software, we calculated per-DRG facility costs among all plans that provided facility cost estimates. Second, unlike traditional DRGs, which are used only to pay facility bills for inpatient stays, we included professional services related to the inpatient stay in the per-DRG cost. The professional services component of the per-DRG costs were tabulated from health plans with records of professional service costs. The per-DRG costs were then applied to inpatient stays of all participating health plans so that inpatient stays from all plans would reflect a cost estimate for professional services. Third, for the health plans and delivery systems that had separate professional services claims, the simplifying assumption was made that all professionals services that occurred on the day of an inpatient stay, inclusive of admit and discharge dates, were related to the inpatient stay and that all other professional services were not (these were then included in outpatient costs). Fourth, to maintain as much patient-level variation in inpatient costs as possible, we assigned costs that were based, when possible, on length of stay. Most surgeries occur on the first or second day of an inpatient stay. For DRGs with at least 100 stays in the study sample, we estimated a separate cost for stays with various duration by stratifying the stays within the DRG by duration and calculating a separate mean cost for each duration strata.
Pharmacy
Our data inventory and analysis of the data collected revealed that pharmacy data were the most complete and consistent of all the health care use data from all sites. The 20 664 subjects included in project HIT had 1 992 213 pharmacy dispenses for an average of 13.8 fills per year per subject. As described in Table 4, more than 90% of records had a valid National Drug Codes (NDC), dispense date, days supplied, and dosage.
For our pharmacy cost algorithm, we rejected any cost assignments based on the average wholesale price because many health systems can often buy drugs at substantial discounts. Thus, their relative price advantage may change the mix of drugs that are prescribed and will affect the system's overall cost structure. Our solution was to estimate unit costs for all agents in a therapeutic class weighted by the mix of generic and patent drugs on the health systems' formulary within that class. The cost of any specific dispense was then multiplied by the dose and days supplied to determine the total cost.
When details about pharmacy use such as dose or days dispensed were missing, we used the mean values of other dispenses for the same agent to impute values. If details on the specific agent could not be imputed, but the therapeutic class was provided, we assigned the mean unit cost value for other drugs in the same class. When information about the specific agent and therapeutic class were both missing, we assigned the unit cost for all of the dispensed drugs.
|
DISCUSSION |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
Variation between and within health plans provides substantial challenges to economic research. In this article, we have described a method used in the CRN to standardize costs across a diverse collection of health systems. We believe the method provides a reasonable compromise between the competing needs for preserving real variation and removing noise. This is one solution for a diverse group of plans at a particular point in time. We stress that, as we pursue further research in these plans, the inventory will need to be updated and the optimal choices may differ in future studies. The importance of this article is the method of ascertaining, cataloging, and addressing the within- and between-plan differences. Secondarily, the decisions we made to address the differences between health plans provide other researchers with a starting point when creating a cost algorithm for multisite retrospective research.
Researchers need to recognize the complexity of identifying costs consistently. The same variable, with the same name, at two different plans may represent two entirely different concepts. The importance of local experts with historical knowledge of changes at the plan cannot be underestimated. The health plan market is too competitive and diverse to understand multiple plans without these partners.
During the course of our research, CRN member health plans have made substantial investment in administrative and clinical data systems. These improvements include expansion of a common EMR system among the Kaiser Permanente sites. Several CRN plans are also implementing Decision Support Systems to support cost management consistent with a system developed at Group Health Cooperative. A Decision Support System is intended primarily for administrative use, but over time it will become the primary cost source for economic analyses within and across many CRN sites.
Regulatory changes are also motivating data standardization. In particular, Medicare is now requiring HMOs to provide inpatient and outpatient data in claims forms. Employers are also demanding more information in standard formats, which may drive further data standardization. These market changes might allow standardization of economic analyses by using claims information that is not possible today.
Other collaborative projects have sought to address the problems of pooling data across health plans, but these projects have largely ignored the complexity of economic data. For example, the HMO Research Network Center for Education and Research in Therapeutics (CERT) developed a standard data format for capturing drug exposure and outcome information (20). Each CERT health plan maintains its own files that are stored in a standard format. Similarly, the Vaccine Safety Datalink, a collaboration of eight HMOs, has developed standard files for tracking eligibility, vaccines, and outcomes (21). As in the CERT, no economic information is captured. The CRN has developed a virtual data warehouse (VDW), which consists of standardized data definitions similar to those developed for this project, along with file specifications, computer programs, and data quality documentation. The VDW allows data from unique local systems to be translated into standardized data bases with a common format across sites (22).
The limitations of both outpatient and inpatient data stem from the various incentives to produce data that are an accurate record of resources used in a visit. Clinicians in each system face challenges posed by the specific incentive structure. For clinicians seeing a mix of FFS and capitated patients, the incentive to code for reimbursement outweighs the lack of additional remuneration for a capitated patient. In integrated systems, the incentive structure is different, driven by factors other than the number of RVUs generated during a visit. The weakness of our cost algorithm is that we cannot fully reconcile these two incentive structures. Without physically reviewing the physician's notes and other chart-based records, there is no way to directly resolve this conflict.
Pharmacy data presented fewer challenges, but we still need to recognize the limits of these data. First, pharmacy databases typically include only outpatient dispensings. The data do not include inpatient medications and, at some plans, do not include drugs, such as chemotherapy agents, administered within clinics. Also, the data represent only filled prescriptions, not what the physician ordered. If a patient chooses either to fill the prescription under alternative insurance or does not fill the prescription, this information is missing. These, and other gaps such as the lack of information on over-the-counter medications, need to be recognized by economic researchers.
Our data algorithm does not capture the true economic cost of resource use. Although this is a laudable goal when creating economic analyses from the societal perspective, we are aware of no one who has achieved this even within a single-site study. The principal difficulty is that it is nearly impossible, in pooling data from diverse sites, to determine where there are differences in a health plan's cost structure for a specific service. In our study, which uses a health insurer perspective, we have sought to achieve a reasonable middle ground across diverse health plans. This middle ground should be broadly generalizable but does not directly measure the economic cost of resource use in any one plan.
There are many services not included in our study. For example, we have not addressed long-term nursing care, home health, hospice, or durable medical equipment. For some populations, these might represent important cost elements. Also, some specialized patient management services like disease and case management are not included. These services are aimed at the most complex patients and need further exploration.
Finally, although future changes in health plans, health insurance, and data are important, investigators working with historical data should make sure that the data in past inventories are unchanged. Health plan administrative data systems do not necessarily leave historical data unchanged. As new systems come online, they may delete data elements or merge files to make data consistent across time. This is especially true in the archiving process where the business model may decide that some data elements are unnecessary. These will then disappear from the data, making historical data less complete than it once was.
|
NOTES |
|---|
Funded by grant U19 CA79689 from the National Cancer Institute. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NCI or NIH.
This study would not have been possible without the active cooperation of the researchers and staff of the participating health plans and research centers: Center for Health Studies, Group Health Cooperative of Puget Sound; HealthPartners Inc., and HealthPartners Research Foundation; Henry Ford Health Systems, Clinical Research Unit, Kaiser Permanente Colorado, Division of Research, Kaiser Permanente Northern California; and Center for Health Research Kaiser Permanente Northwest and Kaiser Permanente Hawaii.
|
REFERENCES |
|---|
|
TopNotesAbstractIntroductionVARIATION IN RESOURCE USE...METHODSRESULTSDISCUSSIONReferences |
|---|
(1) Etzioni R, Riley GF, Ramsey SD, Brown M. Measuring costs: administrative claims data, clinical trials, and beyond. Med Care 2002;40(6 Suppl):1163–72.
(2) Brown, ML, Lipscomb J, Synder C. The burden of illness of cancer: economic cost and quality of life. Annu Rev Public Health 2001;22:91–113.[CrossRef][ISI][Medline]
(3) Fryback DG, Craig BM. Measuring economics outcomes on cancer. In: Lipscomb J, Donaldson MS, Eds. Cancer outcomes research: the arenas of application. JNCI Monogr 2004;33:134–41.
(4) Howard DH, Molinari N, Thorpe KE. National Estimates of Medical Costs Incurred by Nonelderly Cancer Patients. Cancer 2004;100:883–91.[CrossRef][ISI][Medline]
(5) Brown ML, Riley GF, Schussler N, Etzioni RD. Estimating health care costs related to cancer treatment from SEER-Medicare. Med Care 2002;40(suppl):IV104–17.
(6) Barlow WE, Taplin SH, Yoshida CK, Buist DS, Seger D, Brown ML. A cost comparison of mastectomy versus breast conserving therapy for early-stage breast cancer. J Natl Cancer Inst 2001;93:447–55.
(7) Fireman B, Quesenberry C, Somkin C, Jacobson A, Baer D, Brown, ML, et al. The cost of care for cancer in a health maintenance organization. Health Care Finance Rev 1997;18:51–76.
(8) Taplin SH, Barlow W, Urban N, Mandelson MT, Timlin DJ, Ichikawa L, et al. Stage, age, comorbidity, and direct costs of colon, prostate, and breast cancer care. J Natl Cancer Inst 1995;87:417–26.
(9) Rigotti NA, Quinn VP, Stevens VJ, Solberg LI, Hollis JF, Rosenthal AC, et al. Tobacco-control policies in 11 leading managed care organizations: progress and challenges. Eff Clin Pract 2002;5:130–6.[Medline]
(10) Solberg LI, Quinn VP, Stevens VJ, Vogt T, Rigotti NA, Zapka JZ, et al. Tobacco control efforts in managed care: what do the doctors think? Am J Manag Care 2004;10:193–8.[ISI][Medline]
(11) Quinn VP, Stevens VJ, Hollis JF, Rigotti NA, Solberg LI, Aickin M, et al. Compliance with the National Tobacco Treatment Guideline: making progress and promoting patient satisfaction in non- profit health maintenance organizations. Am J Prev Med 2005;29:77–84.[CrossRef][ISI][Medline]
(12) Fishman PA, Hornbrook, MC, Meenan RT, Goodman MJ. Opportunities and challenges for measuring cost, quality, and clinical effectiveness in health care. Med Care Res Rev 2004;61(3 Suppl):124S–43S.[Abstract]
(13) Hornbrook MC, Goodman MJ, Fishman PA, Meenan RT. Health-based payment and computerized patient record systems. Eff Clin Pract 1998;1:66–72.[Medline]
(14) Saver BG. Commentary on "Opportunities and challenges for measuring cost, quality, and clinical effectiveness in health care": the fault lies not in our stars but in our system. Med Care Res Rev 2004;61(3 Suppl):151S–60S.[CrossRef][ISI][Medline]
(15) Saver BG, Ritzwoller DP, Maciosek M, Goodman MJ, Conrad DA, Finkelstein E, et al. Does payment drive procedures? Payment for specialty services and procedure rate variations in 3 HMOs. Am J Manag Care 2004;10:229–37.[ISI][Medline]
(16) Lasker RD, Marquis MS. The intensity of physicians' work in patient visits—implications for the coding of patient evaluation and management services. N Engl J Med 1999;341:337–41.
(17) Lorence DP, Spink A. Regional variation in medical systems data: influences on upcoding. J Med Syst 2002;26:369–81.[CrossRef][Medline]
(18) Meenan RT, Goodman MJ, Fishman PA, Hornbrook MC, O'Keeffe-Rosetti MC, Bachman DJ. Issues in pooling administrative data for economic evaluation. Am J Manag Care 2002;8:45–53.[ISI][Medline]
(19) Hornbrook MC, Goodman MJ, Fishman PA, Meenan RT, O'Keeffe-Rosetti M, Bachman DJ. Building health plan databases to risk adjust outcomes and payments. Int J Qual Health Care 1998;10:531–8.
(20) Fishman PA, Goodman MJ, Hornbrook MC, Meenan RT, Bachman DJ, O'Keeffe Rosetti MC. Risk adjustment using automated ambulatory pharmacy data: the RxRisk model. Med Care 2003;41:84–99.[CrossRef][ISI][Medline]
(21) Platt R, Davis R, Finkelstein J, Go AS, Gurwitz JH, Roblin D, et al. Multicenter epidemiologic and health services research on therapeutics in the HMO Research Network Center for Education and Research on Therapeutics. Pharmacoepidemiol Drug Saf 2001;10:373–7.[CrossRef][ISI][Medline]
(22) DeStefano, F. The Vaccine Safety Datalink project. Pharmacoepidemiol Drug Saf 2001;10:403–6.[CrossRef][ISI][Medline]
(23) Hornbrook MC, Hart G, Ellis JL, Bachman DJ, Ansell G, Greene SM, et al. Building a virtual cancer research organization. J Natl Cancer Inst Monogr 2005;35:12–25.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  K. R. Yabroff, E. B. Lamont, A. Mariotto, J. L. Warren, M. Topor, A. Meekins, and M. L. Brown Cost of Care for Elderly Cancer Patients in the United States J Natl Cancer Inst, May 7, 2008; 100(9): 630 - 641. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Lipscomb Estimating the Cost of Cancer Care in the United States: A Work Very Much in Progress J Natl Cancer Inst, May 7, 2008; 100(9): 607 - 610. [Full Text] [PDF]
|
|
|
|
 |
|
 G. J. Spalding, M. Hartrumpf, T. Sierig, N. Oesberg, C. G. Kirschke, and J. M. Albes Cost reduction of perioperative coagulation management in cardiac surgery: value of 'bedside' thrombelastography (ROTEM) Eur. J. Cardiothorac. Surg., June 1, 2007; 31(6): 1052 - 1057. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||