Significant Analytic Research Results

Power and Sample Size Assessments

Assessments of statistical power and sample size are important considerations in the design of health evaluation studies.  The work of Gardiner, et al. in this area³⁷ extending that of others,^32,39-41 provides a formal statistically rigorous approach to the problem.  It provides a framework, based on the distribution of the net health cost for deriving statistical power and sample size expressions for testing hypotheses on the CER. It also provides a test of the joint hypothesis of effectiveness and cost-effectiveness, and compares sample sizes needed for achieving a stipulated power for cost-effectiveness with that needed for demonstrating effectiveness alone.³⁸ They find that in commonly encountered circumstances a power analysis to demonstrate cost-effectiveness would require a substantially large number of patients than that needed to show effectiveness alone.  In the context of treatment trials, this raises the dilemma of continuing a study to gather data to test economic hypotheses after there is evidence of a statistically significant and meaningful difference in treatment efficacy. Because their methods permit hypothesis testing on the CER in trials powered for effectiveness, they can be used to compare observed power tests on the CER.  They also address the close relationship between the CER and net health benefit or net health cost in formulating these tests of hypotheses. Their sample size formulae have used extensively to study scenarios for designing randomized controlled trials for eliciting cost-effective evidence.⁴² A review of methods for assessing statistical power and sample size for cost-effectiveness studies was recently published in Expert Review of Pharmacoeconomics & Outcomes Research.³⁸  

Reporting the Precision of Estimated Cost-Effectiveness Ratios

By identifying the CER as a statistical parameter, inference on the CER can be initiated within the framework of a sampling design in which costs and benefits are assessed in a sample of patients from two competing interventions. The sampling distribution quantifies the degree of uncertainty in an estimated CER.  For example, it is informative to report both the estimated CER and its 95-percent confidence interval.  Tests of hypotheses on the CER can also be formulated, and issues of statistical power and sample size for cost-effectiveness studies can be addressed.

An important element in reporting results of CEA is to gauge the precision of estimates of summary statistics such as the CER.  Statistically this can be achieved by estimating the standard error of the estimated CER or providing a confidence interval for the CER.  An enormous amount of literature has been published to address this problem.^{25,32-35,43-48} Gardiner, et al. compare three of the popular parametric techniques for constructing confidence intervals for the CER.¹⁷ They demonstrate relationships between the three approaches and shows how the interpretation of the CER could be compromised when the incremental effectiveness is not statistically significant.  Additional research by Gardiner's group^49,50 and that of other investigators^45,51,52 has revealed through simulation studies the importance of using the appropriate method in constructing confidence intervals for the CER.  In recognition of their work on estimation of the CER and its potential uses in public health policy, the editors of the critically acclaimed Handbook in Statistics invited their contribution to a volume addressing Bioenvironmental and Public Health Statistics.¹¹

In their articles^17,49 Gardiner, et al. compare various techniques of obtaining confidence intervals for the CER. These articles point to the need for examining effectiveness before cost-effectiveness.  Without statistically significant effectiveness between two competing treatments, assessing variation in the CER is less important.  From the point of view of a decision-maker, treatments that are equivalent in their effectiveness would be judged on their costs alone, with the choice being the treatment with the lower average cost. Gardiner, et al.'s theoretical approach to estimation of the CER clearly indicates that meaningful confidence intervals for the CER do not exist unless the difference in effectiveness is statistically significant. This also brings into consideration the distinction between clinical significance and statistical significance. The latter depends on the method of analysis and more importantly on sample size.

Current Research on the Development and Application of Longitudinal Models for Inference in CEA

Cost-effectiveness analysis in heart disease

Coronary heart disease, the most common form of heart disease among Americans, is associated with considerable morbidity and is the leading cause of mortality in the U.S..^53,54 In the U.S. alone, the prevalence is 4.6 million, with an incidence rate of 550,000 new cases a year and approximately 957,000 hospitalizations a year.  Costs related to heart failure are extremely expensive, and comprise $20.3 billion in direct costs and $2.2 billion in indirect costs, for a total of $22.5 billion.  This figure may be an underestimate since a portion of the costs for coronary artery disease are likely to be the result of heart failure. 

Heart failure is a complex disease process. Treatment of heart disease is complicated because more than a single form of therapy is often needed depending upon the extent of disease, comorbidity, patient age and gender.  In consideration of a therapy for heart failure, there may be no clear starting point or stopping point (other than death).  The natural history of heart disease and its management may vary substantially. The patient's condition may remain stable for a while then decline, resulting in hospitalization and intensified therapy.  This can lead to a worsened health state and associated high costs.

  Led by Dr. Joel Kupersmith, MD, an eminent cardiologist, the research team of Rovner, Holmes-Rovner and Gardiner investigated the economics of heart disease with a comprehensive review of the effectiveness and cost-effectiveness of treatments and technologies.^13,55-56 Their research team produced a major three-part review that has to date garnered over 115 citations in the professional literature.  Following this review they undertook an evaluation of the cost-effectiveness of the implantable cardioverter defibrillator (ICD). Using data from previously published studies, they evaluated the effectiveness and the cost-effectiveness of the ICD compared to electro-physiology guided drug therapy and developed a statistical methodology to support this investigation.^26,27 The ICD is used to treat patients with ventricular arrhythmias who are at risk of sudden death. Several studies^57-60 have demonstrated the benefits of ICDs in secondary prevention for patients who have previously experienced serious ventricular arrhythmias, and also in primary prevention for patients at risk of ventricular arrhythmia. Ventricular fibrillation, one form of arrhythmia, is a result of multiple rapid and chaotic electrical signals from different areas of the ventricles.  Cardiac arrest soon results when the heart ceases to supply blood to the body. Unless very quickly terminated, ventricular arrhythmias can cause irreversible brain damage or sudden death.

The ICD is designed to detect ventricular tachycardia or ventricular fibrillation and restore normal rhythm, either through rapid pacing or by delivery of appropriate electrical shock.  Several randomized clinical trials (RCTs) have been conducted on the ICD including the Multicenter Automatic Defibrillator Implantation Trial (MADIT),^60,61 the Anti-Arrhythmics versus Implantable Defibrillator Study (AVID)^58,59, the Canadian Implantable Defibrillator Study (CIDS)^57,62 and the Multicenter Unsustained Tachycardia Trial (MUSTT). These trials have demonstrated the benefits of ICDs in improving survival in several classes of patients, and as a result use of the ICD continues to increase.^63-65  In recent years the use of the ICD has become much more widespread and beyond RCTs. It is unclear if its cost-effectiveness is maintained when applied to much broader patient populations. Regression-based models for analysis of health care costs and outcomes from competing interventions offer an exciting approach to answering this question.  This is the focus of Gardiner's current research on the development and application of longitudinal models for inference in CEA.²⁰  

Statistical Methods for Cost-effectiveness Analysis of the ICD

The statistical methodology that Gardiner, et al. needed to evaluate the cost-effectiveness of the ICD was not available at the time of their study.  Then available techniques did not address the longitudinal aspects of their data, the presence of censored survival outcomes and the integration of costs accumulating over time. Gardiner, et al. therefore developed de novo a technique based on survival analysis to address this problem.^25-27,66 The approach was the first to address construction of statistical confidence intervals for the CER from survival data. Their work has been cited in the peer-reviewed literature, and is mentioned in at least two comprehensive monographs, in the report on the Panel on Cost-Effectiveness in Health and Medicine⁷ and Modeling in Medical Decisionmaking—A Bayesian Approach.⁶⁷ These citations include both statistical and medical journals such as American Heart Journal, Journal of the American Statistical Association, Circulation, Medical Care, Pharmacoeconomics, Statistics in Medicine, Statistical Methods for Medical Research, and World Journal of Surgery. The broad coverage of these journals and monographs is evidence of the richness of their methods, and success in their quest to translate this research into policy and practice.  In addition, their comprehensive evaluation of the cost-effectiveness of the ICD compared to conventional electro-physiology guided drug therapy²⁷ has received also received considerable attention.

 Several studies suggest that ICDs might have favorable cost-effectiveness ratios.^{58,62,65,68,69}  However, this depends on the alternative treatment strategy to which the ICD is compared, the classes of patient groups studied (e.g., extremely high risk patients, elderly patients), the type of costs included, the length of the study, and perspective of the analyses. Previous investigations suggested appreciable gains in life-expectancy with ICDs but in recent studies this gain is more modest. For example, over a time span of 6.3 years the Canadian Implantable Defibrillator Study⁵⁷ (CIDS) reports average cost per patient of the ICD was $57,015 compared to treatment by amiodarone costing $25,090 per patient (3 percent discount rate, 1999 dollars). However, life-expectancy of 4.58 years under the ICD and 4.35 years for amiodarone was not statistically significant. This small difference in life expectancy produced a CER for ICD therapy versus amiodarone of $138,803/yr which is unattractive by current standards.  An economic substudy of the Anti-Arrhythmics versus Implantable defibrillator Study⁵⁸ (AVID) reveals that at 3 years ICD cost averages $85,522 compared to $71,431 for a patient under anti-arrhythmic drugs (amiodarone or sotalol).  Survival benefit was also quite small—0.21 years in favor of the ICD, giving a CER of $66,677 (3 percent discount rate, 1997 dollars).

These studies indicate the need for standardization in reporting the results of economic evaluation studies, since conclusions can vary with patient groups studied and perspectives taken. As noted by Thompson⁷⁰ one of the challenges facing CEA is the formulation of models that can reveal which patient characteristics drive costs and outcomes. For example, the wide variation in CER estimates raises the question whether subgroups of patients exist, defined by risk factors, clinical and demographic characteristics, in whom the ICD could be cost-effective. Gardiner's research on statistical methodology for CEA addresses formulating regression models that could inform identification of patient characteristics and resource-use elements that influence both costs and outcomes, and the cost-effectiveness of competing interventions. This will improve standardization in reporting the results of economic evaluation studies.

Hospitalizations for Heart Disease

Hospital costs constitute a significant portion of the overall expenditure in health care.  As a result of escalating costs, knowledge of the correlates of length of stay (LOS) and in-hospital cost are important for decisions regarding allocation of resources.  Based on reports from the Healthcare Cost and Utilization Project (HCUP), congestive heart failure, coronary atherosclerosis, chest pain, irregular heartbeat, stroke, and heart attack comprise 18 percent of all hospital stays for women and 23 percent of all hospital stays for men.  Operations performed on the cardiovascular system account for nearly 3.3 million of approximately 36.4 million hospital discharges in 2000, with average hospital charges of $30,433. While costs of hospitalizations have increased over the past decade, hospitals have responded to cost containment pressure by reducing the length of hospital stays.  Hospital charges generally cover all services rendered to the patient including nursing and surgical care, medications, laboratory analysis and diagnostic tests.

Escalating costs of healthcare as well as the need for cost-containment policies have brought into focus methods for analyzing medical costs and health care utilization.  Gardiner's research team address this important issue through regression models that permit estimation of mean charges as a function of patient hospital stay and adjust for the influence of patient characteristics and treatment procedures on LOS and charges.^23,24  The methods are applied to assess mean LOS and mean charge by cardiac procedure in a cohort of patients hospitalized for acute myocardial infarction, while adjusting for the impact of patient demographic and clinical factors on LOS and charge. These data were taken from the Michigan Inter-Institutional Collaborative Heart (MICH) Study. ⁷¹

The MICH study was designed as a prospective investigation of health care utilization and patient outcomes in admissions for acute myocardial infarction (AMI) to 5 mid-Michigan hospitals. The first phase covered admission from January 1, 1994, through April 30, 1995.  In 1997, a second phase was conducted to examine similar outcomes after changes in medical management and treatment options for AMI were instituted in these hospitals.  The main objective of these studies was to assess sources of variability by race and gender in the use of invasive cardiac procedures–cardiac catheterization (CATH), percutaneous transluminal coronary angioplasty (PTCA), and coronary bypass grafting (CABG), in the treatment of AMI.  The research team investigated whether changes in treatment patterns from the first phase to the second were accompanied by any differences in long-term survival. 

Analyses of hospital charges and length of stay involve several challenges. First, charges and length of stay have skewed distributions that make traditional analyses based on sample means inappropriate. This results in misleading interpretations. Second, the presence of appreciable patient heterogeneity in the sample makes statistical comparisons difficult between dissimilar groups.  Third, when comparing charges and utilization by cardiac procedure, it is important to account for the varying durations of stay that would be correlated with costs.

Gardiner, et al. approach these problems by developing a regression-based methodology.^23,24 They illustrate its application using data drawn from two hospitals in the MICH study.  The technique estimates the relative impact of primary procedures-CABG, PTCA, and CATH, including precision of estimates, the influence of patient demographic characteristics and comorbidities on both hospital charge and length of stay.  Their results indicate that the presence of comorbidities such as diabetes, congestive heart failure, and peripheral vascular disease increase costs, but they do so through increased utilization manifested by increased LOS.

An important issue in analyses of hospital stays is how patients who survived their hospital stay should be compared to those who do not.  Both cost and LOS can be very different in these subgroups. In their study in AMI patients, compared to patients who survived their hospital stay, those who died had higher charges but shorter stays.  Overall, patients who underwent CABG surgery had higher charges and length of stay than patients who had PTCA, or those had only diagnostic cardiac catheterization. These conclusions underlie the need for careful analyses of hospital charge data in relation to length of stay because of their high correlation.

In summary, the investigators develop a method to estimate the cumulative cost of health interventions over a specified duration while controlling for a mix of patient-specific variables using data of total cost and associated length of treatment.  Their method allows greater use of total cost data, typically found in hospital records and claims files, that has not been previously attempted in cost analyses.

VIII.    An International Collaboration

Since 2000, Gardiner has collaborated with an international team of investigators assembled by the World Health Organization (WHO) in Geneva, Switzerland for a planning a study of depression in 8 community settings across the world (U.S., Australia, Brazil, Turkey, Mexico, Nigeria, China, India).  The objectives were to assess costs, health outcomes and cost-effectiveness of interventions designed at both patient and providers in recognizing, managing, and treating depression. 

The WHO cites depression as one of the leading causes of global disease burden which is expected to become the second leading cause within the next two decades. International epidemiologic research has demonstrated the substantial burden that mental and substance use disorders impose on individuals, communities, and health services.  If left untreated or inadequately treated, these disorders lead to an increased likelihood of poorer outcomes in comorbid conditions, psycho-social impairment, increased disability days, and ultimately increased health care costs. However, only a very small fraction of healthcare resources in most developing countries is directed to identifying and treating these disorders.  The lack of trained professionals, barriers to effective treatments, and social stigma associated with mental health disorders have all contributed to the problem.  Research in the U.S., Europe, and in some developing nations has indicated that treatments for depression and substance abuse disorders can be delivered in a primary care setting (as opposed to specialty mental health care) with subsequent improvements in general health outcomes, in mental health functioning and health-related quality of life. Building upon a relationship between the World Health Organization and the National Institutes of Mental Health and Drug Abuse, this study plans to develop and test primary care-based interventions to reduce the burden associated with depression and substance use disorders in 5 WHO member countries (Turkey, Mexico, Nigeria, China, India) representing 4 out of 6 WHO global regions. This investigation shows promise for a wealth of knowledge and experience that would be gained in designing interventions that could deliver effective and cost-effective treatment strategies within primary care.

In the proposed study several aspects of AHRQs guidelines for diagnosis and treatment of depression in primary care were used in developing the treatment interventions.⁷² It calls for randomization of patients to four mutually exclusive treatment groups. Group I: patients treated as usual (TAU), Group II: patients whose providers receive training in evidence based management of depression (EBM), Group III: patients with proactive case management by a nurse depression care manager (DCM), and Group IV: patients whose providers receive training in EBM and augmented by a DCM.  The second part of the study will assess the cost-effectiveness of EBM alone and DCM alone compared to TAU, and the combined treatment EBM + DCM, compared to the single treatments. 

Involvement in this study was triggered by previous published work on power and sample size assessments for cost-effectiveness studies that came to the attention of the WHO investigators.³⁷ This again demonstrates the importance and impact of their research on statistical methods for CEA which has led to this very exciting and important participation. Although adequate funding to conduct this study has been difficult to garner, the WHO investigators have begun some preliminary organizational work to help launch the study when resources become available.

Identification of Major Problematic Areas in CEA and How Gardiner's Research Is Addressing These Issues

As a result of my review of the literature, I have identified what I consider to be the major problematic areas in CEA. These are as follows:

• The central problem seems to be a lack of standardization in CEA.
• CEAs can be complex and difficult to conduct due to inadequate representation of cost and effectiveness data.  Many cost effectiveness studies use complex models that rely on numerous assumptions where evidence is lacking or inconsistent. 
• Current methods generally focus on a single measure of cost or health outcome and do not fully exploit the longitudinal character of data needed for CEAs and its impact on summary measures such as the CER as well as median cost and survival rates These measures are paramount to predicting resource utilization and informing policy on the allocation of health care dollars.
• CEAs are often reported in a way that makes it difficult for users to understand how results are obtained.
• There are many difficulties in statistical analysis for CEA.  Rigorous statistical techniques must be developed to analyze jointly both costs and patient outcomes.
• When differences in approach, assumptions, methods, and quality lead to conflicting conclusions, potential users may be confused and credibility of the CEA undermined.
• Inadequate attention to the design of cost-effectiveness studies can lead to inconsistencies.

With AHRQ's support, the research team led by Gardiner has taken several steps in addressing these issues:

• Despite the rapid development of techniques for conducting economic evaluation studies in medicine and health, the statistical methodology to support these studies is in the developmental stages.  Gardiner's research formulates statistical models that inform identification of patient characteristics and resource-use elements that influence both costs and outcomes.
• Recognizing the natural setting in which cost and health outcomes would manifest over time, his current research addresses development of longitudinal models that incorporate covariate information and permit estimation of their impact on summary measures such as the CER and NHC.
• The Australian Pharmaceutical Benefits Advisory Committee guidelines on conduct of CEA³ advise adoption of methods that are "responsive in differences in health states between individuals and to changes in health states over time experienced by any one individual."  In addition, they also advise consideration of the impact of patient heterogeneity and sensitivity results of a CEA.  Gardiner's interdisciplinary team of statisticians, health economists, health services researchers and clinical investigators has built a repertoire of publications addressing both applied and methodological issues in CEA.
• Many methodological developments are theoretically sound and can be tested on simulated data.  In practice, these sophisticated methods have limited use unless they address the inherent problems in data sets commonly available to researchers from clinical and epidemiologic studies.  These include problems with patient heterogeneity, skewness in cost distributions, incomplete follow up, truncation, censoring and sample selection. Gardiner's research team uses multi-state models for the dynamics of movement of patients through health states.  They recognize the natural setting in which health outcomes and costs arise in practice, accounting for issues of censoring, truncation, and sample selection.
• The longitudinal framework that underlies their analytic techniques can be used to provide a complete specification of alternative models for estimating health care costs and outcomes.
• Gardiner's research methods and models offer practitioners of CEA a powerful set of tools for the improvement of statistical analysis of cost, health care utilization and cost-effectiveness data.

The flexible framework for stochastic CEA being developed by Gardiner and colleagues draw upon the following features:

• Recognizes that costs are stochastic and incurred at random times in random amounts as patients' transition between health states and sojourn in health states.
• Exhibits the role of discounting costs as they manifest over time.
• Defines all the summary measures used in CEA as statistical parameters arising from the underlying probability model.  These include, net present value, quality-adjusted life years, cost-effectiveness ratio, net health benefit and net health cost.
• Incorporates patient-specific explanatory variables (covariates) into the analysis, allowing for the assessment of their influence of summary measures used in CEA.
• Addresses the impact of sampling plans under which the data on costs and outcomes ensue in the longitudinal model, including the role of censoring and outcomes.
• Formalizes statistical inference on net present value, quality-adjusted life years, cost-effectiveness ratio, and net health cost by providing a rigorous basis for their estimation, as well as derivation of their statistical distributions.  This allows for quantifying uncertainty in estimates through standard errors and confidence intervals.  
• Permits testing of hypotheses on net present value, quality-adjusted life years, cost-effectiveness ratio, and net health cost.  Given the data gathering mechanisms for costs and outcomes, this provides a comprehensive scheme for statistical inference based on these entities.
• Addresses design issues in CEA such as assessment of statistical power and sample size for planning of cost-effectiveness studies.

Summary and Significance of Gardiner's Research Findings Related to Translating Research into Policy and Practice (TRIPP)

• A comprehensive review³⁸ was published in 2004 addressing statistical issues in assessing statistical power and sample size for cost-effectiveness studies. This was at the request of the editors of Expert Review of Pharmacoeconomics & Outcomes Research following their earlier work in Health Economics.³⁷ Their work in this area has led to collaboration with an international team of researchers in designing a study for evaluating the effectiveness and cost-effectiveness or treatment strategies for decreasing the burden of depression in developing nations.
• The issue of testing of hypotheses on cost-effectiveness ratios (CER) and assessing statistical power and sample size is addressed in an article in Health Economics. Following the pioneering work of O'Brien, et al.,³² this was the first attempt to place hypothesis testing on CERs on a formal statistically sound framework.
• Gardiner's method incorporates the correlation between cost and effectiveness measures, and leads to substantially lower sample size requirements than methods that ignore the correlation. It extends work by several other researchers.^39-41,73
• An important element in reporting the results of CEA is to gauge the precision of estimates such as the CER. Gardiner's work compares three of the popular parametric techniques for constructing confidence intervals for the CER.¹⁷ His work demonstrates relationships between the three approaches and show how the interpretation of the CER can be compromised when the incremental effectiveness is not statistically significant.
• In recognition of the work of Gardiner, et al. on statistical inference on the CER and its potential use in public health policy, the editors of the critically acclaimed Handbook in Statistics invited their contribution to a volume addressing Bioenvironmental and Public Health Statistics.¹¹  A summary is presented of how uncertainty in estimated parameters can be assessed by their sampling error and conventional statistical inferential techniques.  These techniques can then be applied to problems of estimation, tests of hypotheses, sample size, and power determinations in planning economic evaluation studies of health care programs.
• Gardiner's statistical method for assessing the cost-effectiveness of the ICD was the first to address construction of statistical confidence intervals for the CER from survival data.^25-27 His articles in the American Heart Journal and Medical Decisionmaking have received over 55 citations in professional literature.
• Gardiner's team continues their research in CEA to address the formulation of regression models that could inform identification of patient characteristics and resource-use elements that influence both costs and outcomes, and the cost-effectiveness of competing interventions. Applications are contemplated in cardiovascular studies and in cancer treatments studies. The methods could ultimately improve standardization in reporting the results of economic evaluation studies, and provide objective means for assessing subgroups in which an intervention could be cost-effective.
• Through experience with analyses of length of stay and cost in hospitalizations for heart failure, Gardiner, et al. have developed methods^20,23,24 to estimate the cumulative cost of health interventions over a specified duration while controlling for a mix of patient-specific variables. This method blends statistical and econometric techniques to address issues in the analysis of health care costs and allows for a greater use of total cost data, typically found in hospitalization records and claims files, that has not been previously attempted.
• Gardiner's research proposes a unified framework to estimate summary measures commonly used in cost analyses and CEA.  These include life-expectancy, quality-adjusted life years, net present values, cost effectiveness ratios, net health cost and net health benefit.  Since patient demographics, clinical variables and intervention characteristics can affect these summary measures, regression models have been developed that incorporate covariate information into structural equations for cost and outcome measures.
• These regression models are uniquely designed to account for costs engendered at transition times between health states (e.g., changes in health state that trigger resource use), and costs of sojourn in health states (e.g., resource use while in remission, relapse, or different treatment phases). For health outcomes such as quality of life assessments, their longitudinal models incorporate patient heterogeneity and address the issue of censoring commonly found in these types of studies.
• In summary, several aspects and complexities in the analyses of healthcare costs and outcomes are incorporated into these models, and collectively these new methods promise useful application in CEA. Demonstration of their methods in practice with clinical and epidemiologic data is an equally important goal of their endeavors.

Future Plans related to Translating Research into Policy and Practice (TRIPP)

Conducting a CEA requires the proper analysis of healthcare costs, utilization and health outcome data.  Statistical methodology to support the analyses however, is still in its developmental stages.  In a recent editorial in Statistical Methods for Medical Research, Thompson⁷⁰ underscores the importance of "formulating regression models for mean costs and effectiveness, in adjusting for confounders in observational studies of cost-effectiveness, and in determining the most essential patient characteristics or resource-use elements which drive costs."  He also stresses that "an area of development for the future will be an attempt to identify covariates which define subgroups of whom patients for whom an intervention is most cost-effective."  As a result of demonstrating the breadth and depth of their statistical methods in CEA, Gardiner's current research on use of regression models in cost-effectiveness analysis will continue to address these challenges.

The guidelines of the Pharmaceutical Advisory Committee in Australia³ advice adoption of methods for CEA that are "responsive to differences in health states between individuals, and to changes in health states over time experienced by any one individual."  Statistical models proposed by Gardiner, et al. are ideally designed to address these issues. Recognizing the natural setting in which cost and health outcomes will manifest over time, they will continue to develop longitudinal models that incorporate covariate information and permit estimation of their impact measures such as the cost effectiveness ratio and net health cost. Multi-state models will describe he dynamics of movement of patient through health states, accounting for issues of censoring, truncation and sample selection. 

Many sophisticated methodological developments are theoretically sound and can be tested on simulated data.  However these methods have limited use unless they address the inherent problems encountered in data sets commonly available to researchers in clinical and epidemiological studies.  These include heterogeneity and skewness in cost and outcome distributions, incomplete or censored observations in longitudinal studies of health care utilization, unobserved patient heterogeneity, and sample selection.  Gardiner, et al. plan to address these methodological gaps using data from a diversity of studies which will cover randomized clinical trials, community-based studies and administrative national and State databases. Specifically, they will use Medicaid and Medicare claims data, and the Nationwide Inpatient Sample (NIS) from the Healthcare Cost and Utilization Project (HCUP) to estimate costs of care in cancer and heart disease.

Their research will continue to add to existing research by developing, testing and implementing a methodologically rigorous unified framework for statistical inference in cost-effectiveness studies.  In addition, this research will contribute significantly toward an international effort to develop rigorous statistical methods for analyses of costs and outcomes.

Return to Contents
Proceed to Next Section