• Decrease font size
  • Return font size to normal
  • Increase font size
U.S. Department of Health and Human Services

Drugs

  • Print
  • Share
  • E-mail

AASLD-FDA-NIH-PhRMA Hepatotoxicity Steering Group Meeting, 2006 Presentations: Can we replace opinion consensus with a Bayesian process?

Tim Davern, MD
University of California San Francisco
Can we replace opinion consensus with a Bayesian process? [PDF]

Causality assessment in drug induced liver injury (DILI) provides an estimate of the probability that a suspect drug was involved in an observed episode of liver injury. An accurate and quantitative assessment of this probability is essential for pharmacogenomic and other mechanistic studies, as well as for appropriate patient care. Unfortunately, few useful biomarkers are available to aid causality assessment, which thus relies heavily on circumstantial evidence and exclusion of competing causes of liver injury. Causality assessment is particularly challenging when liver injury occurs in the background of underlying liver disease or multiple potentially toxic drugs.

In an effort to formalize the process of causality assessment and reduce interobserver variability in the diagnosis of DILI, several instruments have been developed, the most widely used of which is the Roussel Uclaf Causality Assessment Method (RUCAM). Although still popular, the RUCAM instrument is marred by the seemingly arbitrary selection and scoring of its components, its relative inflexibility and inability to deal well with missing data. Recognizing the limitations of the RUCAM and similar causality assessment instruments, many studies rely upon the consensus opinion of experts – also called “global introspection” - as the “gold standard” for diagnosis of DILI. In fact, the Drug Induced Liver Injury Network (DILIN) utilizes such a process currently for causality assessment in both retrospective and prospective studies. However, expert consensus opinion is not readily available to the practicing clinician, is not very reproducible, and its components are often opaque (i.e., not clearly stated or quantified).

A Bayesian approach to causality assessment, in addition to being logical and scientifically elegant, should address these shortcomings of consensus opinion. In addition, unlike the RUCAM, such an approach takes into account the prior probability of DILI and drug-specific risk factors, deals well with missing data, and is quite flexible. The prior probability of DILI is calculated based on the incidence of liver injury from a given drug in exposed compared to unexposed individuals, data ideally derived from randomized, placebo controlled trials of the drug. The posterior probability is calculated by modifying this prior probability by the probability of other potential causes of liver injury expressed as likelihood ratios. Likelihood ratios express the probability that the particular features of a particular case are caused by the drug compared with the probability that they have another cause. Likelihood ratios can theoretically be developed to incorporate a wide spectrum of clinicopathological information and thus capture characteristic or “signature” features of liver injury from particular drugs. In essence, the Bayesian approach translates the overall probability of an event in the general population to the probability of the event in a particular individual. However, such an approach is labor intensive and time consuming to develop, requiring the establishment of a large computerized database of drug-specific prior probabilities and likelihood ratios, and in some cases these data may simply not exist.

We clearly need better methods for quantitatively assessing the likelihood that an observed episode of liver injury is caused by a given drug and not by some other etiology. Development of novel and improved causality assessment instruments is in fact one of the mandates of the DILIN. We offer, for consideration and debate, a proposal that a Bayesian approach, despite its limitations, may represent an important advance in causality assessment for DILI.

Biographical Sketch

Dr. Davern attended the College of the Holy Cross, received his medical degree from Columbia University College of Physicians and Surgeons, and subsequently trained in Internal Medicine at Beth Israel Hospital, Boston. Dr. Davern completed his gastroenterology/ hepatology training at the University of California San Francisco and has been on the faculty since 1997. He is currently an Associate Professor of Medicine. Dr. Davern’s research interests include drug induced liver disease, acute liver failure, and hepatic gene therapy. He is a principal investigator at one of the 5 clinical sites comprising the Drug Induced Liver Injury Network (DILIN), and is also a site PI in the US Acute Liver Failure Study Group.