Gaining insight, changing medical practice from large data stores
Using a large collection of data such as that contained in Epic’s Cosmos can give clinicians a quicker way to affirm others’ research or test hypotheses.
While I’m not a research scientist, I can validate the fact that using aggregated data, in the form of the Cosmos database from Epic, has affected how I’ve practiced medicine. More importantly, I feel that it has the potential to democratize data exploration and knowledge discovery.
If you do not already know, Cosmos contains a data set that represents the largest integrated database of clinical information in the United States – it holds records from more than 167 million Epic patient records with in excess of 5.7 billion encounters – it has records from patients in all 50 states. It’s built on a data platform optimized to answer clinical questions – it enables clinicians and researchers at organizations using Cosmos to ask questions without needing to file a request with someone else in that organization.
Cosmos can play a role as a rapid knowledge generation tool that can help clinicians iterate and discover ways to increase medical knowledge base. But I believe its true value is more than just creating volumes of research publications – it can be a trusted digital colleague that a clinician can use to ping hunches or confirm findings of a publication. After all, research is necessary, but it only matters if it can affect patient care.
I find something empowering about doing self-research to better inform care. Even when I’m not necessarily generating new insights, I use Cosmos to confirm or perhaps refute things in recent studies or publications. In my experience, that investment of time and effort makes it more likely to impact care delivery.
Recent uses of research
Cosmos’ usefulness in this arena was most apparent to me during the COVID pandemic.
One query I ran was a result of realizing that influenza rates were significantly lower in the winter of 2021-2022, even as COVID patient numbers were declining. A query of Cosmos confirmed that influenza rates were up during the winter of 2021-2022, cases were still far below the two previous winters, when preventive measures and social distancing wasn’t being practiced.
Another query confirmed that while the Omicron variant was more transmissible, its virulence and severity were much lower than the Alpha and Delta strains. The Cosmos data query showed that the percentage of deaths of patients recorded as having COVID was 1.4 percent in January 2022, compared with a peak of 2.6 percent the previous year. Other Cosmos data indicated rapid declines in hospitalization rates, compared with rates prior to the omicron variant.
As an example of how the database enabled the impact of vaccination on cardiovascular outcomes among COVID patients, I queried Cosmos after reading Johns Hopkins research this past March on certain complications. Cosmos data showed that, when vaccinated, patients who got COVID had no change in rates of pericarditis and arrhythmias, but vaccination was associated with a 50 percent reduction in myocarditis and 25 percent reduction in venous thromboembolism in COVID patients. Those results were based on about 1.5 million patients in the database with COVID infections.
While these insights were interesting, they really didn’t have an enormous impact on my practice in the emergency department. I started considering ways to use the platform. For me, the best uses weren’t necessarily coming up with novel insights but confirming ones that were already published or known.
Impacting treatment
For instance, consider fluoroquinolones, a class of antibiotics that rose to prominence in the 1990s and gained increasing popularity in the 2000s. They became popular because they killed everything -- using a fluoroquinolone seemed to make prescribing antibiotics really simple. But concerns grew about increased resistance to antibiotics and a particularly high rate of C. difficile that made fluoroquinolones undesirable first-line agents.
Even with those concerns, physicians only marginally decreased use of fluoroquinolones, because, again, they were a one-stop antibiotic shop that killed anything. However, there were reasons to be cautious. Yes, there were the standard arguments that such usage increased resistance and caused the aforementioned C. difficile. But there were increasing issues and concerns that were quite unique to fluoroquinolones.
For example, a recent article described some concerning side effects of quinolone antibiotics, such as a tendency for patients on quinolones to develop neuropathy. I decided to see if this trend was confirmed in a Cosmos query, and I found that patients on quinolones seemed to have a correlation with new neuropathy diagnosis: 1.8 percent vs. 0.255 percent – results that were consistent with the study. While the absolute incidence was not high, a nearly 2 percent incidence of a completely avoidable long-term side effect shouldn’t be ignored.
Another fairly well-known complication of fluoroquinolone usage was tendon rupture, particularly Achilles tendon rupture. This issue is particularly pronounced when the antibiotic is combined with a corticosteroid medication, such as prednisone. In examining my own historical patterns of ordering fluoroquinolone, I recalled that I had started seeing drug-drug interaction warnings when ordering fluoroquinolones concurrently with steroids. I typically ignored the warning and prescribed the medications together without a problem. I knew many of my colleagues had prescribed this combination as well. How bad was the problem and should it result in a change of practice?
The Cosmos database held some answers. Query results suggested that steroids with quinolones resulted in a significant increase in tendon ruptures, compared with just prescribing quinolones alone. Although I had already started to avoid quinolones, these sorts of findings reinforced and accelerated this trend in my practice pattern.
On the flip side, the absolute incidence of tendon rupture was still quite low. The numbers provided ammunition in instances where I could have a reasonable discussion about a treatment regimen with a patient if reasons existed in which fluoroquinolones were the only option because of allergies or another clinical reason. Thus, grounding existing and emerging medical research in patient data contained at a massive scale can influence care. But what has it historically taken to advance medical care in the past (putting research into practice) and how can that process be accelerated? I’ll look at that in Part 2.
John Lee, MD, has been a clinical informaticist since 2006 and has been Chief Medical Information officer at two health systems. In 2019, he was honored with the prestigious HIMSS Physician Executive of the Year award. He is a firm believer that the key to solving the myriad problems we experience in our healthcare system is efficient and transparent delivery of information.