Drugs today aren’t as safe as they could be for several big reasons, including failure to see the big picture, says Steve Jolley, principal of SJ Pharma Consulting. “It took years to figure out Vioxx was killing… 2-3 patients per thousand due to heart attacks and strokes because the people taking it were likely to die of [those causes] anyway.”
Software tools by themselves are of little consequence without professionals medically interpreting the signals to determine which ones are legitimate product safety concerns, says Sally Van Doren, president and CEO of the drug safety consultancy BioSoteria. But there is a shortage of trained drug safety practitioners evaluating product safety signal detection, she says. “There’s a huge training need in this area. Pharmacovigilance, risk management, and signal detection training—including certificate programs offered by organizations like BioSoteria and the Drug Information Association—are a good start but not completely sufficient,” because certification confirms only class attendance. No competency testing is presently available or required for those actually entrusted with drug safety surveillance. The proficiency of plumbers and accountants is better defined!
Post-marketing regulatory inspections provide some benchmarks about how well product safety signals are detected, evaluated, and acted upon by pharmaceutical companies, she says. For example, the UK Medicines and Healthcare Products Regulatory Agency’s latest metric report of key pharmacovigilance inspection findings notes limited safety signal detection systems in about 13 percent of pharma inspections.
Heightened recognition of the importance of safety signal detection and evaluation has improved therapeutic risk management. Drug makers can ill afford suboptimal patient outcomes and the expense of legal damages as a result of shortcutting safety monitoring. So if preclinical or early-stage clinical studies suggest an experimental drug might be associated with kidney or liver damage, additional biomarker measures get monitored in addition to standard blood and urine tests. The expanding field of pharmacogenetics has also identified genetic differences in patients with AE susceptibility.
Signal detection of common AEs is best accomplished in a controlled clinical trial setting where the sponsor can collect details of reported AEs and calculate actual incidence. Safety problems are more likely to be detected as data accumulates and clinical development progresses, often with thousands of patients with the targeted disease, says Van Doren. Even in blinded studies, a data safety monitoring board often reviews incoming safety data for imbalances in the drug’s safety profile compared to placebo or marketed therapy.
Still, safety is a hard endpoint to evaluate relative to efficacy, says Cynthia Uber, former VP medical services at Eisai. “Safety requires exposure to broader populations than are utilized in clinical trials and can therefore be more costly and time consuming to confirm. It is also subject to a labyrinth of reporting requirements that can become labor intensive,” she says. “Safety centers are typically not revenue generating for pharmaceutical companies.”
One challenge for safety departments in past years was that those evaluating and reporting serious adverse event (SAE) data often had little access to the full clinical data set during a clinical trial, with separate databases housing different safety information. Today, with electronic data capture (EDC) systems, sponsors may interface the clinical data with a safety database, making “real time” SAEs and patient data available for analysis. Most sponsors also utilize alert systems triggered from clinical databases and sent to the safety monitor when individual patients experience predefined thresholds of certain AEs or abnormal laboratory values, Van Doren says (see “FDA Final Rule”).
The Weaknesses of AERS
FDA currently accepts only post-marketing safety reports in its Adverse Events Reporting System (AERS). But the agency anticipates electronic transmission of—and presumably an electronic repository for—IND safety reports on experimental products. In Europe, expedited safety reports of serious unexpected adverse reactions are amassed in the EudraVigilance database for both investigational products and post-authorization (marketed) products.
Although many AEs are detected by spontaneous reporting systems, these systems have limitations that hamper signal detection, says Van Doren. Beyond the study setting, only a small fraction of post-marketing AEs—even the most severe toxicities—are reported to drug manufacturers or regulators. Improving the ability of health care professionals to recognize and report AEs will undoubtedly be a focus of future technology. Applications for the iPad and similar devices will allow health care professionals and patients to send AE information directly to FDA. Drug manufacturers typically have product-specific or corporate websites or call centers for reporting of AEs, although as Van Doren notes, that information is often re-entered into a corporate safety database again when it gets reported to regulators. Paper-based reporting systems are so resource-intensive that drug manufacturers and regulators not using electronic AE capture systems have a backlog of hundreds to thousands of cases awaiting manual entry at any one time.
Another concern with AERS, according to Uber, is that it gathers information from multiple sources, potentially “over-representing” information. A physician could conceivably report a SAE to both FDA and the drug manufacturer, while the same event could be reported by an attorney without necessarily being flagged as a duplicate. “To look at data at the aggregate level and not the individual case level can lead to the wrong conclusions,” she says. “It all comes back to the accuracy and quality of the source data.”
The FDA’s remedy to under-use of its MedWatch AE reporting system is the Sentinel Initiative. Launched in May 2008, this aims to create a national electronic system for tracking reports of AEs linked to FDA-regulated products. The idea is to query diverse health care data holders—including electronic medical record (EMR) systems, administrative and insurance claims databases, and registries—to quickly evaluate potential safety issues. Data owners are tasked with mining for answers to FDA-posed questions and submitting summary results. However, it may be a few years before the Sentinel Initiative “comes good,” cautions Jolley. As of December 2011, progress had been made on EMR querying methods but “not many” new safety issued had been identified.
The FDA has been instructed by Congress to mine at least 100 million EMRs—about one-third of the U.S. population—for signs of safety problems with drugs, says Paul Watkins, director of the Hamner-University of North Carolina (UNC) Institute for Drug Sciences. But without any validated tools to do this, “most safety signals that pop out will probably be false ones,” he says. In the absence of randomization to treatments and placebo controls, interpreting the data is “extremely challenging.” The Observational Medical Outcomes Partnership had mixed results trying to find EMR evidence of known drug troubles during beta testing of the Sentinel system, he notes.
Watkins predicts the most immediate effect of the Sentinel Initiative will be to make FDA reviewers—already spooked by the recent Vioxx and Avandia disasters—“more cautious.” In the long term, once reliable approaches have been developed to screen EMRs for AEs, the Sentinel Initiative might allow a graded approval of certain drugs. One possible scenario is a new drug approved for use in health care networks where EMR data get mined in real time, allowing for rapid, real-world safety assessments.
Corporate Due Diligence
Based on audits of 50 pharma firms on three continents, SJ Pharma Consulting says there is a lot drug makers can do themselves to beef up product safety. Jolley developed a 50-page checklist of safety-related questions, best practices and regulatory guidances to help companies tease out deficiencies. Safety data exchange agreements with partners and subcontractors, for example, may be weak or absent entirely. Participants in a U.S. clinical trial were recently put at risk when a Japanese drug licensor failed to inform the sponsor of a label change indicating potential liver toxicity. The obligation to report any drug risk was never written into the safety data exchange agreement by the U.S. licensee, whom the FDA holds responsible for maintaining the integrity of the “supply chain of safety information,” says Jolley.
Companies also sometimes struggle to report unexpected SAEs within 15 days, as required by FDA. Last November, it took one sponsor’s safety group three months to learn that one of its trial participants had been hospitalized with a heart problem, because the reporting investigator used the wrong fax number. The SAE was entered into the EDC system, Jolley notes, but the information wasn’t passed on to the safety team, so unsuspecting patients continued being enrolled.
Debate continues about how useful computers are for linking AEs to drugs, because statistically suggestive causalities ultimately require a “prepared mind” to prove, says Jolley. Data mining tools are designed for use in the post-marketing environment, where both the numerator (drug-associated AEs) and denominator (the drug’s universe of users) are largely unknown. The software thus seeks clues either in public databases such as AERS or the World Health Organization’s VigiBase or in proprietary big pharma databases. The denominator is approximated by comparing the number of AEs reported for a drug to the average number of drug-event combinations reported for other drugs.
Corporate desire to ensure drugs are approved, especially among small companies, may also be compromising the clinical trial enterprise, says Jolley. Sponsors tend not to use signal detection techniques to proactively identify problems. The good news is that the European Medicines Agency now compels companies to annually submit a Development Safety Update Report (DSUR) of therapies under development, much as they do for marketed drugs. FDA plans to do likewise. DSUR contains ten elements not in the IND annual report used in the U.S., including cumulative summary tabulations of all AEs of special interest and an overall safety assessment. “This should encourage companies to take a more holistic view of the safety of drugs in development,” Jolley says.
Hiring out safety surveillance functions to less well-qualified employees of offshore contract safety organizations is another growing problem, says Jolley. For sponsors, this is an issue of control that can be mediated by better contractual language and perhaps higher allowable fees for the advanced level of medical decision making required.
Outsourcing of safety surveillance is becoming more common for drug makers, says Uber. With limited real-world data in the early post-approval phase, companies are obliged to employ the FDA’s risk evaluation and mitigation strategies (REMS) that might include restricted product distribution or creation of patient registries. Based on the sentiment of 28 organizations recently surveyed by the Tufts University Center for the Study of Drug Development, REMS may not be packing much punch: only 22% think the REMS system has improved safety. As products mature, companies can learn more from safety signaling tools. But the technology often yields a large volume of potential signals that may or may not be causally related to a drug, and can require extensive resources to investigate, says Uber.
Both FDA and National Institutes of Health (NIH) are clearly interested in improving the science behind the assessment of drug risks, and support research in priority areas of regulatory science such as adaptive trial design. The regulatory science program has recently focused on novel approaches for testing drug efficacy and safety prior to clinical trials, notably in vitro cell-based technologies that can mimic how human tissues respond to drugs. The Defense Advanced Research Projects Agency recently committed over $70 million to developing a “human on a chip” that can assess the safety of medical counter measures to a biological attack when there is no time for animal studies. Pre-clinical animal models are time-consuming and expensive, not to mention relatively poor predictors of results in humans. In fact, safety-related issues—and more specifically, actual or feared liver toxicity—are the main reason drugs fail, contends Watkins.
Cardiac arrhythmia disturbances, as evidenced by a prolonged QT interval, were until recently the most frequent culprit in drug recalls. But early-phase EKG testing on healthy volunteers has become standard practice, making liver meltdown the major unresolved safety issue. In the absence of a regulatory path for drug-induced liver injury (DILI), FDA sometimes requires expanded clinical trials to improve the odds of detecting a latent liver issue. The “disaster” is not just the time and costs of these extended trials, but potentially billions of dollars for the years of lost patent life “we all end up paying for.”
Watkins is chairman of the steering committee for the NIH-funded Drug-Induced Liver Injury Network, which since 2004 has been endeavoring to genetically characterize people who experience uncommonly severe liver reactions to a drug. Separately, the Hamner-UNC Institute has been attempting to create a computer model of DILI to make animal models more predictive —and ultimately help eliminate animals from drug development entirely. The European Union is taking the lead in this arena with its Innovative Medicines Initiative (IMI) that has set aside 2 billion euros ($2.6b) for matching grants going to pharma companies engaged in regulatory science research. The IMI’s goal, says Watkins, is to “make Europe the worldwide home for pharmaceutical research.”
Meanwhile, the Hamner-UNC Institute is about to launch a DILI-sim initiative with partners AstraZeneca, GlaxoSmithKline, Novartis, and other pharma companies, says Watkins. The goal is to model significant inter-species differences and susceptibilities to liver toxicity, to help inform first-in-man dosing. Watkins argues that part of the problem in achieving efficacy is the failure to administer a sufficiently high dose. Currently, drugs are typically given in doses “at least ten times lower than what causes liver toxicity in animals,” he says.
Corporate partners are providing financial and in-kind resources as well as data from compounds that looked promising in animals but subsequently fail in humans. The FDA, which intends to apply lessons from DILI-sim in regulatory decisions, is also supportive. The first iteration of the computer model will be distributed to partners early in 2012.
Previous research from the Hamner-UNC Institute suggests a simple urine test could predict which clinical trial participants are at risk for DILI, based on specific metabolite patterns predictive of mild liver damage associated with acetaminophen. Better biomarkers of organ toxicities would be a huge advance for trial safety, says Watkins. The four blood tests currently available for detecting liver problems have been in use for 50 years, produce false-positive results, and miss sudden-death liver injury that can inexplicably arise weeks or months after ingestion of a drug.
“Companies do not routinely collect and keep blood and urine samples from clinical trials,” says Watkins. “If they do, there’s no standard way to handle them and link [the samples] to patient data.” Watkins is leading the push to create a standardized safety database linked to a biospecimen repository. In the cardiovascular arena, FDA already prescribes how EKGs should be performed and the data warehoused. The mandate helped spur formation of the Cardiac Safety Research Consortium for ongoing evaluation of medical products. Watkins would like to launch a similar consortium, initially focused on liver safety, at the Hamner-UNC Institute.
However, increasingly robust patient privacy requirements could make creation of a biospecimen repository problematic, warns Uber. “If a study sponsor retains samples, it has to inform study participants. And if a sponsor doesn’t know how those samples may eventually be analyzed and evaluated, they will need to obtain permission for ‘research not yet specified.’ Language that provides the sponsor unlimited future access to retained biological specimens is a permission patients are likely to question, and may be reluctant to grant.” •
This article also appeared in the January 2012 issue of Bio-IT World magazine.