Delivering the right amount and balance of intravenous fluids is crucial to the survival and recovery of critically ill patients with sepsis, a devastating inflammatory response to out-of-control infection.
The standard for answering these life-and-death questions is the prospective, randomized clinical trial, which compares outcomes between a randomly selected treatment group and control group, and the cohort study, which follows a group of people, in this case, sepsis patients, over time to see how they do long term.
These studies are enormously expensive to conduct, however. To get the biggest bang for the research buck, Jonathan Schildcrout, PhD, and his colleagues at Vanderbilt University Medical Center are developing new methods to maximize what can be learned from clinical trials and cohort studies, while maintaining their rigor and reproducibility.
Their methods include the two-phase study design, applying data obtained from a large clinical trial or cohort study to a subset of patients from the original trial to answer additional research questions.
This approach “not only allows researchers to answer novel questions with high precision, compared to standard (study) designs, it frees up resources so that more questions can be answered,” said Schildcrout, professor and vice chair of Research in the Department of Biostatistics at VUMC.
Schildcrout recently received a $2.5 million second renewal of a grant from the National Heart, Lung and Blood Institute of the National Institutes of Health to continue to develop two-phase study designs.
Since 2009, grant number 2R01HL094786, titled “Outcome Dependent Sampling of Longitudinal Data: Design and Analysis,” has provided $4 million to support the project, which has generated more than 40 publications on research methods and research outcomes.
An example is the Crystalloid Liberal or Vasopressors Early Resuscitation in Sepsis (CLOVERS) study, a multicenter, randomized clinical trial that compared the effectiveness of different fluid “resuscitation” strategies in more than 1,500 patients with early sepsis.
There is some evidence that overaggressive fluid therapy may accelerate the septic degradation of the glycocalyx, the gel-like protective inner layer of blood vessels, thereby increasing the risk of acute respiratory distress syndrome (ARDS) and death.
Schildcrout and his colleagues are studying whether a protein fragment released by damaged glycocalyx, syndecan-1, could serve as a biomarker for a poor outcome.
Because it is expensive to measure, syndecan-1 levels were collected on an informative sub-sample of 600 participants from the CLOVERS trial, based on their outcomes. The study team is currently conducting analyses to see how well syndecan-1 levels associate with outcomes and the effectiveness of various resuscitation strategies.
This method, where outcomes determine which study participants’ risk factor values should be retrospectively ascertained, is called outcome-dependent sampling, and is a central tool for two-phase study research.
“Using data that already exist, outcome-dependent sampling allows us to focus our person-time and research dollars on the most informative sub-cohort for answering novel study questions,” Schildcrout said. This approach could provide insights into the development of novel interventions for sepsis patients and could be used to personalize treatment.
“In an era where pre-existing data sources such as cohort studies, randomized clinical trials, and electronic health records data can be linked to biobanks and to other expensive-to-ascertain exposure variables,” he said, “efficient study designs and analytical approaches offer exciting opportunities for cost-preserving research.”
Schildcrout’s co-investigators at VUMC include Ran Tao, PhD, assistant professor of Biostatistics, Cassianne Robinson-Cohen, PhD, assistant professor of Medicine, Sunil Kripalani, MD, MSc, professor of Medicine and director of the Center for Health Services Research and the Center for Clinical Quality and Implementation Research, and Bryan Shepherd, PhD, professor of Biostatistics and Biomedical Informatics.
