People who need medication rarely take just one pill. Many of those who have been prescribed medications for health reasons take upwards of five medications per day. According to the U.S. Centers for Disease Control and Prevention, the percentage of persons using at least one prescription drug in the past 30 days, in the U.S., is 48.9 percent; and the percentage of people using three or more prescription drugs (‘polypharmacy’) in the past 30 days stands at 23.1 percent; whereas those using five or more prescription drugs in the past 30 days is 11.9 percent. The figures for the U.S. will mirror those of many other high-income countries.
From the ubiquitous aspirin to the most sophisticated prescription medicine on the market, all medications come with side effects. Many are minor, some are just an inconvenience, and a few are serious. For example, effectively any medication can lead to nausea or an upset stomach. Others can cause allergies, like rashes or dry mouths. On the more serious end of the scale, internal bleeding can occur.
While the side-effects of a specific drug is generally known, the dilemma for the the medic is that knowing what the side-effects are of these drugs in relation to each other is often too complex to predict and there are few clinical trials in this area. This is unlikely to change in this area, since running such trials into multi-drug combinations is impractical. To understand the complexity there are some 5,000 licensed medications, with 1000 known side effects. This means there are 125 billion possible side effects once difference drugs are paired up.
However, a solution is at hand. Computer scientists from Stanford University have worked out out how to predict side effects of combination medicines using artificial intelligence. The new system is called Decagon, and it is designed to aid doctors with making more informed decisions about which drugs to prescribe. The platform also has the potential to offer researchers pathways to find new combinations of medicines to help to treat complex diseases.
Decagon was trained up using deep learning, testing out the platform on those drug combinations that were known. Currently the AI can only assess pairs of drugs; the aim in the future is to use the system for three or more drugs in combination. The research was recently presented to the International Society for Computational Biology in Chicago, in July 2018.
The research has also been published in the journal Bioinformatics, under the heading “Modeling polypharmacy side effects with graph convolutional networks.”
