Drug development is an infamously expensive, risky and time consuming process, and many pharma companies are turning their attentions to new indications for existing drugs. For the first time, researchers have found a way to trawl electronic health records (EHRs) for signs that a drug may have untapped benefits. Lead study author Josh Denny, of the Vanderbilt University Medical Center Tennessee, explains how this data can be used in the search for clues.
The pitfalls of drug discovery are well known. A time-consuming and labour-intensive process, developing a new drug costs in the region of several billion dollars. According to the National Center for Advancing Translational Sciences, the process typically takes around 14 years and has a failure rate surpassing 95%.
This being the case, it’s easy to see why ever more pharma companies are looking to repurpose existing drugs. Every so often, a commercially available medicine will have a slew of hidden benefits. Exploring these benefits – and testing for new indications – is quicker and cheaper than starting the process from scratch.
“It’s becoming more and more expensive to bring new drugs to market, and a lot of drugs fail because of lack of efficacy or adverse reactions,” says Josh Denny, associate professor of Biomedical Informatics and Medicine at Vanderbilt University Medical Center, Tennessee. “So if you have existing drugs that you know are well tolerated and are already in the marketplace, finding new indications can certainly be beneficial.
‘From the drug company’s perspective, it can certainly lead to a larger market for their drug, and from a clinical standpoint it would be great to repurpose safe medication for new diseases.”
Of course, you cannot simply attempt to repurpose a drug at random. In order to justify the cost of clinical trials, you need to have evidence that the drug may be suited to different applications. And in order to generate that evidence, it is necessary to perform preliminary studies.
Generally this involves looking at health care billing data, performing computer simulations and testing in the lab or on animals. Most recently, though, a new technique has emerged – analysing electronic health records (EHRs) for clues that a drug may be worth a second look.
Mining the data
Josh Denny, along with colleagues at Vanderbilt University and the Mayo Clinic, trawled through the EHRs of 32,000 cancer patients, dating back to the mid-1990s. The study, published in the Journal of the American Medical Informatics Association in November, explored the secondary benefits of the diabetes drug metformin. It is thought to be the first to use EHRs to validate a ‘repurposing signal’.
Metformin has long been hypothesised to improve cancer survival rates. A number of studies have suggested that the drug, which is safe and well-tolerated in diabetics, slashes the risk of mortality for several different types of cancer. Most of these studies, however, have simply relied upon large-scale claims data. The Vanderbilt researchers felt they could explore the phenomenon in more detail and with greater validity.
“Because we have the full EHR data, we’re able to look at whether or not the patients are smokers, we’re able to look at all the different kinds of cancer, we’re able to look at what stages of cancer they had when they were diagnosed, and we’re able to follow them for mortality,” explains Denny. “So we can see whether the effect is just for certain kinds or stages of cancer, and the environmental risk factors too.”
The researchers looked at five-year cancer survival rates across the cohort, with and without exposure to metformin. They then repeated the same process in a separate cohort, using data from 79,000 cancer patients at the Mayo Clinic.
Adjusting the results for confounding variables, they found that metformin use was associated with a 23% drop in mortality from all causes. This included a decreased mortality risk from breast, colorectal, lung and prostate cancers. Strikingly, diabetic patients receiving metformin were 39% less likely to die over the time period than those receiving insulin alone.
The next step for EHRs
The team used a wide range of techniques – some well-established; some pioneering – to source the information they needed.
“I believe the approach to EHR data should always be multimodal, using all the information you have,” says Denny. “We use billing code data, we use laboratory data, we use things like electronic prescribing records, and then we also use natural language processing. For the natural language processing, we’ve developed a couple of computer programs that can analyse the text and pull things out. We’ve been able to use these technologies well enough to answer the relevant questions.”
This study, of course, is just one piece of the puzzle for metformin, but it does add to a growing body of evidence about its potential efficacy as a cancer drug. A randomised controlled clinical trial is currently underway to examine the possibilities further.
More broadly, the techniques pioneered here may prove invaluable for further studies of this kind. As EHR networks become more widespread and comprehensive, there will be scope to use the records as a rich source of longitudinal data.
“Not all electronic health records at this stage have enough depth and data to perform studies like this, but I think ours is very robust,” says Denny. “And with the national and international adoption of EHRs increasing dramatically, these studies will become easier and easier to do over time. We won’t just be restricted to the few academic centres that have longitudinal data – you’ll be able to do it across healthcare systems and other non-profits.”
The next step is to build on the research by seeking out repurposing signals for different drugs. The researchers believe there may be many other therapies, yet to be determined, that have secondary indications. They are also using EHRs in genetic studies, with a view to discovering new associations between various genes and diseases.
“We want to do this with other drugs,” explains Denny. “And we’re also looking at pharmacogenomics to predict drug safety. It’s a work in progress!”
This article appears in the March 2015 edition of Pharma Technology Focus