Making the Link

In 2000, FDA pulled Glaxo's Lotronex (alosetron) from the market. Lotronex, indicated for diarrhea-predominant irritable bowel syndrome (IBS), seemed to cause ischemic colitis in some patients. It was subsequently re-released.

It was in this environment that Novartis took Zelnorm (tegaserod) before FDA for approval for use in constipation-predominant IBS patients. The drug's approval was complicated by the fact that clinical trials of 2,965 Zelnorm-treated patients and 1,740 placebo-treated patients showed a slight increase in abdominal surgeries and gall bladder removals for patients receiving Zelnorm. While the imbalance was not high enough to block FDA approval, the agency required a post-marketing study to ensure that Zelnorm did not increase the risk of abdominal surgery and gall bladder removal.

Needing sound scientific answers quickly, Novartis opted to conduct an observational study using real-world data. Based on health insurance data, epidemiologists matched patients who initiated treatment with Zelnorm with similar patients who did not. They followed health insurance claims for the group for six months to assess incidence of abdominal surgery. They obtained medical records to verify surgical outcomes; additionally, a physician survey and a case control study were performed to ensure the validity of the matching.

The two cohorts included 5,524 patients, more than were enrolled in all pre-marketing trials combined. Across both cohorts, 456 potential abdominal surgeries were identified, and medical records were obtained for approximately 85 percent of these. Overall, the study addressed FDA's safety concerns, showing no increased risk of abdominal surgery for Zelnorm-treated patients. Moreover, the study was conducted for less than $1,000 per patient, a fraction of the price of a traditional clinical trial, where per-patient costs can range from $10,000 to $30,000.

With the ability to quickly answer critical safety questions at a fraction of the expense of traditional clinical trials, is there a place in the drug safety arsenal for studies based on automated claims data?

The Evolving Safety Environment

Since 1986, 22 drugs have been withdrawn from the US market because of safety concerns. Because less than 10 percent of all adverse drug reactions are reported to FDA's MedWatch program, it may take years to accumulate sufficient data to warrant regulatory action on some drugs. Meanwhile, theoretical safety concerns may remain unanswered, and actual safety issues might progress unchecked. Since the Vioxx withdrawal, Congress has been under pressure to consider new legislation, and there has been a rising tide of criticism of FDA and the industry. It is crucial for pharma to identify new tools for the drug safety arsenal.

Risk Management

Drug safety assessment occurs both pre-marketing and post-marketing. In the pre-marketing phase, the key tool is the clinical trial, but pre-marketing trials have limitations. Small trials, for example, cannot detect undesirable side effects that occur rarely—say, at a rate of less than one per thousand. In addition, trials are generally performed on people who don't have other diseases and aren't taking additional medications. As a result, when a product launches, adverse reactions occurring at high frequencies are typically well-described. But less-frequent and serious adverse events (AEs) are not, and companies and regulators lack data on the real-world experience of patients with multiple co-morbidities or those using multiple products.

In March 2005, FDA released three guidance documents (Premarketing Risk Assessment; Development and Use of Risk Minimization Action Plans; and Good Pharmacovigilance Practices and Pharmacoepidemiologic Assessment), which established pharmacovigilance and pharmacoepidemiology as the key methods in post-marketing drug safety assessment.

Pharmacovigilance (PV) derives from the concept of public health surveillance. Drug companies are required by law to conduct PV, either in-house or through a contract research organization. Drug companies, of course, do not have direct interaction with patients and rely on spontaneous reporting of adverse events by healthcare providers. PV is the first line of defense to identify serious adverse drug reactions during the early post-marketing phase.

Pharmacoepidemiology (PE) studies the use and effects of drugs in large populations and has been used for drug safety assessment for decades. PV is designed to detect safety signals across all organ systems. PE, on he other hand, compares two or more groups of patients to answer specific research questions. For example, pharmacoepidemiological studies were used to quantify the risk of gastric or duodenal ulcer among users of non-steroidal anti-inflammatory drugs (NSAIDs).

Prospective epidemiologic studies require prohibitive amounts of time and resources. As a result, many researchers have turned to large databases that link prescription information to patient outcomes, such as the databases maintained by state Medicaid programs and various health insurers. To date, PE studies have been conducted mostly in reaction to safety signals detected through pre-marketing trials or PV, rather than as a matter of course.

Fluoroquinolone Antibiotics, Achilles Ruptures, and Claims Studies

Registries In addition to PV and PE, some drug safety programs use techniques such as monitored release, but the methodology is not feasible for all drugs, and there are few examples. The most striking of these is the recently initiated registry of all patients taking Accutane (isotretinoin), Roche's product for severe acne. Other registries also have been established for specific drugs or populations.

Problems With the Status Quo

PV systems, including FDA's MedWatch, generate tens of thousands of reports each year from healthcare providers and patients. But it is impossible to expect overworked healthcare providers to report all adverse events. It is estimated that no more than 10 percent of adverse events find their way into MedWatch. There may be duplicate reports, and many of the reports received by PV systems lack crucial information. The reports certainly do not represent a random sample of adverse events, which limits their usefulness.

It has been proposed that the Adverse Event Reporting System use automated signal detection and statistical algorithms to screen for safety signals. The problem is that mathematical models may not produce valid answers when applied to data of haphazard origins. In addition, these models have not been tested in a controlled setting. In April 2005 the Pharmaceutical Research and Manufacturers Association (PhRMA) issued a Request for Proposal to call for more rigorous evaluation of these statistical methods.

There are other limitations to the post-marketing spontaneous reporting system:

Common adverse events can be hard to spot Spontaneous reports are valuable in linking drug exposure to certain adverse events such as liver failure, agranulocytosis, or Stevens-Johnson syndrome—events that are both serious and rare in the general population. But the reporting system may not be sensitive enough to identify events that are common among high-risk groups. For example, myocardial infarction is not rare, and there are several well-known risk factors. Even if the use of a drug raises the risk of acute myocardial infarction by 50 percent, medical care providers may not recognize the increase among limited numbers of individual cases. Take Vioxx (rofecoxib) for example. Typically adverse events first come to light through case reports in the FDA system or clinical literature. In the case of Vioxx, cardiovascular safety signals were first detected in a clinical trial—and at the time there was not a single case report in the peer-reviewed literature that suspected an association between myocardial infarction and use of rofecoxib.

Post-marketing trials are slow—and single-purpose Studies can take several years to complete, and though they can answer a specific question ("Does this drug cause liver toxicity"), they probably won't provide much help when the next question ("Does it cause cardiac problems?") emerges.

Safety signals detected in pre-marketing trials may have limited utility in predicting post-marketing events For example, some drugs that subsequently proved hepatotoxic, such as troglitazone, showed increased incidence of elevation of transaminase in pre-marketing trials. But simvastatin, which also raised concerns during clinical trials because of elevated transaminase, has proven to be relatively safe after many years of post-marketing experience; in fact, low-dose simvastatin is now sold over the counter in the United Kingdom.

Drug-drug interactions are difficult to evaluate in clinical trial settings The number of potential drug combinations in a real-life population is very large, and it doesn't make sense to include all potential combinations in clinical trials unless there is a strong pharmacological basis for testing for a small number of potential interactions.

Better Databases Ahead

When a drug is approved, the manufacturer has an integrated safety database of all pre-marketing clinical trial results. But to take full advantage of these data, manufacturers need to systematically enhance and utilize the database post-approval. Too often, post-marketing safety data are not routinely integrated. The advent of electronic data capture (EDC), which encourages standardization of data, has made it easier to merge data from multiple studies, and new results can be added almost in real time. In addition, it has become easier to query databases for specific events because of the worldwide adoption of the Medical Dictionary for Regulatory Activities (MedDRA) for coding adverse events. The database could be queried periodically and systematically after a drug is introduced. Pre-marketing safety concerns, theoretical or actual, could be continually assessed.

An enhanced database of clinical trial data could also support evaluation of safety signals received as spontaneous reports. Although only a small number of serious adverse reactions are typically reported through PV systems, when such an event occurs, drug safety personnel need to evaluate quickly whether it represents an excess risk associated with the drug. Could it be explained by the background rate (i.e., the incidence of events among patients with the same disease)? Is the adverse event also associated with similar drugs (class effects)? Are there subgroups of patients with risk factors that predict occurrence of the reaction? All of these factors must be reviewed to recommend a course of action.

Even an improved clinical trials database will have limitations. In the real world, an adverse event isn't necessarily caused by the intrinsic toxicity of the drug. Many adverse events are caused by inappropriate use, such as taking the wrong dose or taking a medication in spite of contraindications. These types of inappropriate use would not be assessed in a controlled clinical trial environment. That can only happen after approval and extensive population drug exposure. And that is where claims databases have their place.

Eventually, if electronic medical records become universal, and if they are implemented correctly, they may become the tool of choice for assessing safety. In the meantime, thanks to the evolution of the healthcare financing system and advances in information technology, healthcare claims databases have become a vast repository of information about a patient's health and use of the healthcare system. The Health Insurance Portability and Accountability Act (HIPAA) allows the use of de-identified data for public health research

Scientists trained in rigorous study design, data collection, and data analysis process may be skeptical about data they have not collected first-hand. But routinely collected data, such as vital statistics, have played an important role in public-health research for years. These secondary data are often as valuable as primary data in drug safety research.

FDA started to recognize the utility of large linked databases in the mid 1980s, when the agency initiated a number of cooperative agreements to allow its officers to work with large data sources to carry out drug safety assessment. Since then, those data sources have been used in many drug safety studies. On the industry side, pharmaceutical companies have sponsored many drug safety studies using claims datasets.

In fall 2005, FDA awarded grants to i3 Drug Safety, Harvard Pilgrim Healthcare and HMO Research Network, Kaiser California, and Vanderbilt University. The agency's goal was to (1) conduct drug safety analyses to the benefit of the public's health; (2) respond expeditiously to urgent public safety concerns; (3) provide a mechanism for collaborative pharmacoepidemiological research designed to test hypotheses, particularly those arising from suspected adverse reactions reported to FDA, and enable rapid access to US population-based data sources.

Public policy must be based on solid science, but large-scale clinical trials are not the sole answer to important drug-safety questions. Scientists are trained to design perfect studies, but public health scientists also understand the need for timely data to guide regulatory decisions. The key is to strike a balance between the timeliness of data collection, the quality of the data, and the validity of inference.

STRONG CLAIMS

The major advantage of claims databases is their large sample size. For products already on the market, thousands of patients can be identified in a matter of weeks, and results can be available relatively quickly. These databases become even more powerful when they are combined with tools for automatically detecting safety signals.

Claims databases can also be used to establish registries. Traditionally, registries could be established for a rare disease, a subset of a patient population with special attributes, or patients receiving a certain drug (e.g. the clozapine and Accutane registries). With the expansion of claims databases and advances in information technology, larger registries can be built faster at substantially less expense.

A common criticism of research using claims data is the potential for "up-coding"—when a physician bills an incident as something more serious than it really is in order to ensure insurance reimbursement. But there are ways to work around up-coding. For example, a scientist trying to learn the number of hip fractures in a given population would not simply count claims for hip fracture. Instead, that scientist would look for hip fractures combined with a cluster of reimbursement codes associated with imaging studies and surgical procedures. Epidemiologists experienced in using these data know what criteria to specify.

FDA recognizes that a new standard must evolve—one of routine, proactive safety surveillance. The history of drug safety assessment tells us that this is an ongoing process. Drug companies should take a more proactive stance in setting the standard for drug safety monitoring. Let's not wait for legislation to define the future of drug safety monitoring. Using a combined arsenal of standard PV activities, enhanced with integrated clinical trial databases that are systematically maintained and mined post-approval, and supplemented with other data sources such as healthcare claims data, let's make proactive monitoring of drug safety the routine standard.

Terri Madison is president of i3 Drug Safety; Arnold Chan, MD, is a senior scientist with i3 Drug Safety and director of i3 Aperio; John Seeger, is a senior scientist with i3 Drug Safety.