The model relied on statistics and machine learning to identify pain points, potentially saving lives.
More than a million people die or are injured each year due to adverse drug reactions (ADRs). Approximately 2 million of these incidents occur in hospital settings, leading to more than 100,000 deaths — despite the fact that the drugs in question are being prescribed by doctors at regular dosages.
Many of these deaths and injuries could be prevented if it weren’t for low rates of spontaneous reporting in the pre- and post-market stages of drug development. In fact, more than 90 percent of new ADRs remain unreported — a major problem, especially when it comes to predicting the less common adverse effects of specific drugs.
Seeking to correct this issue, researchers developed a new deep neural network (DNN) model that can both identify the ADRs of existing drugs and predict the ADRs of new drugs.
As reported in a recent study, published in the Journal of Medical Internet Research, the new DNN model used machine learning and statistical methods to identify the ADRs of 746 existing drugs and predict the ADRs of 232 new drugs. The model examined the chemical and biological information of drugs from 2009 to 2012. None of the findings were reliant on past human reportage.
The DNN model has the potential to help physicians beyond the initial point of prescribing and administering drugs. Using the new information, doctors can better evaluate the risk of ADRs, prescribe the least risky medicine for their patients, monitor for possible ADRs, and take an appropriate course of action when ADRs do occur.
In addition to aiding doctors in saving lives, there are potential financial benefits to this new DNN model.
“Developing new drugs costs a lot of money and time,” says lead researcher Jung-Hsien Chiang, Ph.D., of Taiwan’s National Cheng Kung University. “If the pharmaceutical companies can get the additional information during the new drug development stage through our DNN model, it can help them to make a safer drug and reduce the risk when it launches to market.”
In terms of next steps, Chiang believes the new DNN model must incorporate more data in order to expand the number of drugs and ADRs it can detect. This will include collecting additional medical literature and retraining the model.
“We also need to reconsider the risk of ADRs for drug-to-drug,” says Chiang. To accomplish this, researchers will examine complicating factors, such as the potential for additional adverse effects when prescribing multiple drugs simultaneously.
While more research is needed, this new DNN model could represent a large step forward in terms of patient safety and pharmacovigilance.
Navigate the digital transformation with confidence. Register for our newsletter.
Machine Learning Could Boost Feasibility of Neutron Imaging
Reinventing Clinical Decision Support
Cellphones Can Boost Patient-Centered Care, Drug Adherence Rates