William H. Crown

It is widely recognized that the pharmaceutical industry faces serious financial challenges. Large numbers of blockbuster drugs are losing patent protection and going generic. The pipeline of new drugs is too sparse to fill the gap and generate a platform for future growth. Moreover, many of the new products are biologics with much narrower target patient populations and comparatively higher prices relative to traditional pharmaceuticals. 

As the number of such specialty pharmacy products has grown, payers in the United States have become increasingly concerned about cost, demanding higher levels of evidence about the economic value of biopharmaceutical products from manufacturers. At the same time, the health care system is aggressively moving toward a consumer-directed care model in which the consumer (and the cost of treatment) is given much more weight in the treatment decision. Adapting to these changes requires a wholesale rethinking of drug development and marketing strategy on the part of today's biopharmaceutical companies. 

Faced with this scenario, the pharmaceutical industry has moved to accelerate drug development timelines and to design studies containing health economic and patient-reported outcomes in order to provide payers with the evidence of the economic value of their products that payers are demanding. One important aspect of demonstrating value is to develop products that provide treatments for conditions that previously had no treatments, or which are much more effective or safer than existing products. 

Assessing unmet need can begin very early in the clinical development process—as early as Phase I or II—and provide valuable information that informs early strategic decisions about potential indications to pursue. For example, suppose a company has a compound in development that has the potential to treat a variety of mental illnesses. Very early in the clinical development process the company could examine sources such as medical and pharmaceutical claims data from large insurers to determine the relative sizes of the markets for major depression, obsessive compulsive disorders, social phobias, etc. 

Moreover, it would be possible to determine from these data the various treatments that are currently being used for these conditions; measures of unmet need such as medication switching or discontinuation; and rates of mental health hospitalization. These same data could reveal the medical specialties of the physicians treating the patients, as well as their geographic locations. They could also provide profiles of the patients in terms of their medical co-morbidities, concomitant medications, and health care utilization. 			

Although companies are increasingly building health economics into their early clinical development processes, many still focus on the clinical efficacy and safety criteria needed to obtain market approval. 

These companies then typically develop decision analytic models in which clinical results from such trials are combined with health economics and outcomes data obtained from database studies or meta-analyses of the literature. Models such as these are useful to help payers assess the cost-effectiveness and budget impact of new treatments relative to existing therapies. However, a significant limitation of such models is that they do not tie together the health economic, outcomes, and safety data for the same individuals—rather, they bring together information from several different sources to determine the expected impacts of covering the product. 

Moreover, most clinical trials that incorporate health economic and patient-reported outcomes are powered to detect statistically significant differences in the primary clinical endpoints but seldom to detect differences in health economic or patient-reported outcomes. This is because the wider variation in health economic and patient-reported outcomes requires substantially larger sample sizes to enable hypothesis testing at comparable levels of statistical significance. Unfortunately, cost-effectiveness conclusions based upon such trials may be misleading because there is no guarantee that sample sizes are large enough to ensure that randomization can adequately balance the treatment cohorts in terms of the health economic outcomes. 

Partly because of the substantial cost associated with conducting trials powered to detect health economic and patient-reported outcomes, and partly because payers are demanding real-world data on clinical effectiveness and safety, there is a growing emphasis on postlaunch surveillance studies. 

Like clinical trials, such prospective real-world studies represent substantial research efforts that must be carefully designed to maximize the probability of answering the study's research question. For example, such studies could be used to collect data on health-related quality of life among patients receiving different treatments for a particular condition in actual clinical practice. Similarly, real-world prospective studies can obtain information from patients about their reasons for medication refill decisions as well as from physicians about factors influencing their prescribing decisions. 

Given the significant investment required to execute clinical trials and prospective studies, it is instructive to consider how secondary use of post-launch health data can contribute to the planning, design, and implementation of clinical trials and real-world prospective studies to increase the likelihood of "success." Such data have a number of desirable properties, including large sample size; the ability to observe the use of drugs in real-world settings (in particular, patients with multiple co-morbidites being treated with concomitant drugs); and the ability to observe treatment patterns in actual clinical practice. 			

The use of health care data can influence the planning and execution of clinical trials in many ways, including study design, development of the research protocol/clinical research form (CRF), and physician and site recruitment. Use of health care data may slow the development of the study design and protocol somewhat, but will generally lead to an improved design that will accelerate CRF development. It may also aid in more efficient recruitment of investigators and patients, help determine the sample size needed to detect statistically significant differences in health economic and patient-reported outcomes, and has a variety of other uses, including determining the length of follow-up needed to detect certain endpoints, such as hospitalization risk. Overall, such evidence-based trial design should increase the likelihood of timely recruitment and successful trial completion. 

Most medical claims databases contain geographic codes (e.g., three-digit zip codes) that make it feasible to calculate prevalence rates of particular diseases or the use of particular drugs in a given geography. This is helpful in identifying and prioritizing geographic areas that are ripe for patient recruitment. For example, many companies have clinical trial investigator lists. Overlaying the geographic locations of the investigator list with the geographic concentrations of patients helps to identify areas (and, therefore, the investigators) where patient recruitment will be most efficient. 

In some large payer organizations such as certain large health plan settings, it is possible to identify particular physicians treating patients of interest without ever identifying individual patients. With appropriate Institutional Review Board (IRB) approvals, participating physicians can then recruit patients for studies and obtain patient consent at the time of recruitment. This is an efficient mechanism for identifying and recruiting patients, and ultimately results in improved identification of physicians and sites, improved quality of investigators, and reduced time for recruitment and training. 

Medical claims databases also contain a great deal of information on real-world patterns of health care utilization and costs for patient cohorts of interest. Data on rates of hospitalization and the variation in total health care costs can be valuable for calculating sample sizes needed to detect statistically significant differences in health care outcomes. 			

Clinical trials and prospective studies that focus on the collection of health economic data present a number of unique challenges. Generally, researchers assume that 20% of patients account for 80% of health care utilization and cost (hence, "the 80/20 rule"). This "skewness" requires that health economic outcomes studies must often be substantially larger than studies focusing on efficacy and safety outcomes. The relatively small numbers of patients with very high expenditures need to be balanced across treatment cohorts in order to avoid biasing conclusions about cost effectiveness. 

Given this, it is important to estimate the variance in health care expenditures among a population that closely resembles the patient population that will be entered into the clinical trial in order to properly power the trial to detect differences in economic outcomes across treatment cohorts. This often can be accomplished using retrospective claims data from large national health plans or other sources. 

In addition, many features of a study protocol can be tested—including evaluation of inclusion and exclusion diagnoses, inclusion and exclusion drugs, wash out periods, length of follow-up, etc. Of course, it is not always possible to model all features of a study protocol in this fashion. Some features, such as blood pressure or body mass index, might not be present in administrative claims databases. Nevertheless, the "simulation" of costs associated with patients similar to those that are planned to be studied can provide useful guidance on the required sample sizes necessary to have the statistical power to detect differences in health economic endpoints across treatment groups. 

For studies conducted within large national health plans or other payer organizations, the medical claims offer another important advantage: The claims capture tremendous amounts of detail on patterns of health care utilization that does not need to be collected via survey from patients (who have notoriously poor recall). This enables the survey question to focus more selectively on data not otherwise captured, such as health care not reimbursed by insurance (e.g., over-the-counter pharmacy utilization), health-related quality of life, and other patient reported outcomes. 

In turn, data collection becomes more efficient and less costly, thereby enabling a given clinical trials budget to be spread over a greater number of patients—improving the power of the study to detect statistically significant differences in health economic or patient reported outcomes across treatment cohorts. With patient consent (to comply with regulations such as the Health Insurance Portability and Accountability Act), it even becomes possible to "follow" the patients in the claims for extended periods after the trial has ended—enabling long-term health economic and safety outcomes to be assessed. 

Table 1 illustrates these concepts with an empirical example. Suppose we want to evaluate the feasibility of a clinical trial protocol for osteoarthritis (OA). Using a database of de-identified medical claims, we first identify all patients in the database, age 40 or older, with a diagnosis code for OA. This yields an initial sample size of 27,015. 

Patient Records for Forecasting a Protocols Potential				

Next, we exclude patients currently being treated with certain OA drugs, reducing the sample to 21,791 (or 80.7% of the original). Further restricting the sample to patients on a particular class of drug and excluding patients exceeding a maximum dose causes a major drop in sample size to 3923 patients (18% of the original patient pool). Finally, excluding patients with hypersensitivity to drugs, a history of certain co-morbidities, and a history of substance abuse yields a final potential sample of 3281 patients. 

It is useful for clinical development teams to understand that only about 12% of potential patients (3281/27,015) meet the contemplated study criteria. The impact of changes in inclusion or exclusion criteria on estimated sample size can be assessed for parameters such as gender; age cutpoints; primary and co-morbid diagnoses (using ICD-9 codes); drug therapy, dose or duration; and use of particular diagnostic tests or procedural interventions. 

Linking Investigator Site Records for Added Value			

Linkage of physician identifiers to other databases provides even more value (see Figures 1–3). In the context of a large health plan or provider group, it is possible to go beyond the de-identified patient data and have physicians (or the health plan) recruit patients for studies. These studies must follow standard IRB and patient informed consent processes to ensure compliance with regulations governing clinical research and treatment of protected health information. 

Identifying Investigators by Primary Specialty				

Health economic data and analysis can (and should) play a substantial role in the clinical development of biopharmaceutical products. Although there is substantial variation by country, payers around the world are demanding evidence of economic value, safety, and effectiveness to support coverage and benefit design decisions. The applications' evidence range widely: 					

the use of data analysis very early in the clinical development process to inform strategic decisions about the indications to be pursued, determining the sample size needed to detect statistically significant differences in health economic outcomes

the use of claims data to assess and improve the feasibility of clinical trial protocols

the actual identification of physicians and investigator sites.

Tapping into Investigator Experience						

Such evidence-based trial design should become a routine part of the clinical trial planning process. 

As a result of changes in the marketplace for biopharmaceutical products, payers and regulators are requiring companies to provide much more information about the economic value and safety of their products in actual clinical practice. Phase IV is the first opportunity to observe a drug's use in routine clinical practice. As a consequence, Phase IV clinical studies provide the first chance to observe the impact of a product upon medication adherence and switching patterns, use of health care services, and drug safety in large numbers of patients who often have other co-morbidities and who are being treated with concomitant medications. 

A major challenge with Phase IV studies is the need to control for the myriad other factors, besides drug treatment, that can influence patient behaviors and outcomes. In the hierarchy of evidence, most observers would agree that large, simple Phase IV trials specifically designed to measure health economics and outcomes research endpoints would probably do the best job of this. But because such trials are expensive to undertake, nonrandomized study designs are more common. Such studies can provide reliable evidence, provided that they are properly designed and analyzed using appropriate multivariate statistical methods. 

In general, the more robust the data collection effort, the better the job that such methods can do in controlling for confounding factors, such as unobserved disease severity, which may be the real reason behind variation in patient outcomes. Health economics has much to offer along the entire clinical development continuum—from strategic decision-making early in clinical development to the generation of evidence after the product is on the market. 

, PhD, is vice president, global health economics and outcomes, i3 Innovus. 

, MD, MPH, is therapeutic area director, primary care, with Roche Laboratories Inc., Nutley, NJ.

Articles

Increased cost concerns set the stage for health data to play a major role in drug development.

William H. Crown

Articles

Health Economics & Analysis in Clinical Development