University of Tennessee Faculty Member's Research Aims to Improve Drug Safety Monitoring
In the U.S., all new pharmaceuticals undergo rigorous safety testing and approvals before being sold to the public. Despite these measures, some adverse drug reactions (ADRs) are not detected until after medications are on the market, ranging from common issues to life-threatening consequences.
In one case, Vioxx was approved in the U.S. for public use in 1999 before being pulled from the market five years later. The arthritis pain medication had a serious ADR: It caused increased risk of heart illnesses, with one study estimating Vioxx may have been responsible for up to 140,000 cases of serious coronary heart disease before it was withdrawn.
All new medications are monitored for ADRs after they are released on the market to ensure safety. However, to confirm true correlation between drugs and ADRs requires a sufficient sample size to establish statistical significance. Waiting to collect larger samples can lead to longer delays and more occurrences of harm. A new study1 by a research team including Wenjun Zhou, a faculty member in the Department of Business Analytics and Statistics at the University of Tennessee, Knoxville, Haslam College of Business, proposes a method to increase the speed and accuracy of this critical monitoring.
The Importance of Early, Accurate Detection
"Like raising the alarm of smoke before a fire spreads, detecting ADRs early is the surest way to safeguard patients, prevent harm and maintain trust in drug safety," says Zhou, Lawson Professor of Business, Martin Lee and Carol Fri Robinson Faculty Fellow and co-author of "Early Detection of Adverse Drug Reactions in Postmarket Monitoring."
Postmarket drug safety monitoring often involves collecting reports of ADRs that occur after patients take certain drugs. Systems such as the FDA's Adverse Event Reporting System (FAERS) and the HHS's Vaccine Adverse Event Reporting System (VAERS) gather these reports, enabling statistical analyses to identify significant safety signals. Once the correlation is quantified, researchers rank the drug-ADR pairs from the most correlated to least. Strongly correlated drug-ADR pairs are subject to further investigation to determine their causal relationships.
"Usually, the analytical step is not perfect," Zhou says. "If there are many false positives, investigating them wastes time, leading to missed opportunities to identify the truly dangerous pairs. We need this analytical step to be both perceptive and accurate."
New Correlation Method Improves Detection of Drug Side Effects
Estimating correlations with small samples is difficult because wide confidence intervals can make nearly all pairs seem correlated. Since interval width depends on variability, addressing this variability more effectively can help make better use of limited data.
In their study, Zhou and her coresearchers introduced error-controlled correlation (ECC), a new framework designed to spot potential ADRs earlier in postmarket monitoring. ECC offers more practical correlation estimates that help rank drug-ADR pairs and adjust for variability in the data, a key advantage when information is limited in the early stages of monitoring. By accounting for this variability, ECC provides more stable and reliable estimates of the relationships between drugs and reported reactions.
"Using small data sets, traditional correlation analysis may lead to many false positives, but ECC accounts for this variability by making dynamic adjustments," Zhou explains.
The researchers deployed ECC across five of the most popular drug-ADR correlation methods, effectively identifying pairs in each, demonstrating that it can be applied to most types of correlation measures. This makes ECC an effective tool for monitoring drug safety.
Finally, with error rates controlled, ECC produces a numeric value that allows researchers to rank drug-ADR pairs more fairly. This ranked list may offer a more fruitful ordering of verifying true drug-ADR associations without encountering a high number of false positive signals.
"Our study shows that the ECC equals traditional methods' results using just one-tenth of the data," Zhou says. "Using ECC should enable materially earlier postmarket ADR detection, which could prevent serious public health safety concerns when new drugs are put on the market."
Honoring Business Analysts' Work
The work by Zhou and her colleagues is just one example of how data analysis is making a growing impact worldwide. At Haslam, the Department of Business Analytics and Statistics is dedicated to preparing the next generation of data professionals with the skills and expertise to lead in the field. At the same time, its faculty are producing research, like Zhou's, with real-world applications that make a difference in everyday life. Efforts like these are among the reasons November 14 is celebrated this year as Global Business Analysis Day.
Provided by University of Tennessee at Knoxville
