Research Spotlight: Dr. Sameer Kadri discusses critical care medicine, big data
Dr. Sameer S. Kadri, associate research physician and head of the Clinical Epidemiology Section in the Clinical Center's Critical Care Medicine Department, has leveraged big data generated from routine care at hundreds of hospitals to better understand and improve the management of the severely ill as well as inform healthcare policy.
Kadri founded and leads the NIH Antimicrobial Resistance Outcomes Research Initiative (NIH–ARORI), a collaborative between the NIH Clinical Center, intramural National Institute of Allergy and Infectious Diseases scientists, the Centers for Disease Control and Prevention (CDC) and Harvard Medical School. Through this initiative he developed and tested a simple but clinically relevant classifier of antimicrobial resistance called "Difficult-to-treat Resistance" or "DTR" that helps bedside clinicians quickly determine if they have a safe and effective antibiotic option for patients infected with resistant pathogens.
"I am humbled by the lethality of infection that I witness each day as a critical care provider, which fuels my drive to search for ways to enhance survivability. Although electronic health record data are expansive and granular, they tend to be messy. My team and I are applying recent advances in data analytics and methodology to curate, standardize and model several terabytes of data to understand survival signals that may inform treatment strategies."
He is currently working with the CDC to estimate the United States burden of Difficult-to-treat Resistance and study the impact of antibiotic stewardship in culture negative sepsis. He is working with the FDA to scrutinize the need for novel antibiotics.
Below, Kadri talks about his research and innovation in using large databases to answer important clinical questions with scientific rigor that cannot be addressed by randomized clinical trials:
What kinds of questions in critical care medicine cannot be studied through clinical trials? Is there anything particular about critical care medicine that causes this?
Critical care medicine deals with the management of acutely ill patients at a high risk of death who tend to require intensive monitoring and organ support in an Intensive Care Unit (ICU). Randomized controlled trials (RCTs) remain the gold standard for testing the efficacy and safety of new treatment interventions. In RCTs, consented subjects are allocated at random (purely by chance) to receive an investigational treatment and are compared to recipients of a placebo or a standard intervention.
However, RCTs cannot answer every question in critical care medicine for a variety of reasons, namely when (1) recruitment is low, (2) trial populations differ from real-world populations, (3) financial incentives to conduct a trial are lacking, and (4) when there are special situations requiring immediate response.
RCTs are a costly endeavor and are often supported by the private sector who are almost exclusively interested in investigating new drugs and devices with the goal of revenue. Hence trials integral to understanding ICU care but unlikely to generate revenue are difficult to get off the ground. Unique aspects of critical care medicine call for non-trial-based research to complement areas where RCT evidence is non-existent or suboptimal.
How can large databases substitute or provide answers that a conventional clinical trial cannot?
Thanks to mandated electronic health record use and technological advances in informatics and analytics, large granular healthcare datasets today can offer a window into contemporary real-world healthcare practices and enable observational research in areas of critical care medicine where conventional clinical trials have not been feasible. They contain data from inpatient and outpatient care of millions of patients. They are an excellent resource to study rare conditions and rare treatments. However, they can be relatively heterogenous and need to be painstakingly cleaned and standardized in order to be ready for analysis.
It is impossible to look backwards in existing data and identify the better of two treatments administered with the same degree of fairness as is offered by randomization and the experimental design of an RCT. Sophisticated statistical modelling techniques such as propensity matching, inverse probability weighting and instrumental variable analysis help to mitigate this bias but still cannot replicate what an RCT can do. Yet the evidence obtained from such analyses can provide proof of concept, identify high-risk groups, assess rare populations and the impact of healthcare policy, and generate preliminary data that can build a strong case for a trial.
Was there any particular research project that you could use as an example?
As a critical care and infectious diseases I was particularly intrigued by patients with flesh-eating bacterial infection (necrotizing fasciitis) and shock who would often die despite prompt antibiotics and surgical control, enhancing interest in adjunctive therapies. One such therapy made from pooled human blood, intravenous immune globulin (IVIG), had been evaluated in a trial in Europe that was prematurely terminated after enrolling 21 patients owing to poor recruitment.
In a database of US teaching hospitals, we found 161 (4%) of 4127 patients with debrided necrotizing fasciitis and shock received IVIG. We were in fact unable to identify a survival benefit attributable to IVIG. Importantly, we showed that a future trial would require 888 patients in each arm to be adequately powered to assess IVIG's impact on survival in this condition. Given how rare this condition is, such a trial would be incredibly difficult to operationalize.
What would be the next step in this use of databases in critical care medicine?
I don't think that databases will ever replace clinical trials. They are complementary and occupy unique roles in healthcare outcomes research. What will increase with time is the breadth, granularity and hopefully cleanliness of available data. As the world is coming to realize the benefits of large database research, a need for guidelines and policy to standardize curation and analysis of data has been recognized, as has the need to integrate genomic and clinical data as we enter the phase of personalized medicine.
- Robert Burleson