Presentation
Mitigating Self-Fulfilling Prophecies in Neurocritical Care Research
DescriptionThe goal of much clinical research (both trials and observational) is to estimate the relationship between one or more clinical characteristics or intervention with patients' outcomes. Unfortunately, we make these estimates from data contaminated by myriad treatment decisions. In randomized trials, these include post-randomization interventions that by definition cannot be balanced by us of prior randomization as an instrument. Risks of bias are even greater in observational studies. Traditional methods to make unbiased causal estimates (e.g., causal inference methods or randomization) cannot be used when an intervention is deterministic of outcome (e.g., withdrawal of life-sustaining therapies (WLST)). This creates a ubiquitous risk of self-fulfilling prophecies to which our neurocritically ill patients are uniquely vulnerable.
Supported by NINDS, we will present interim results from an R01 (R01NS124642) that aims to innovate novel methods that mitigate the risk of self-fulfilling prophecies by incorporating expert knowledge into prediction models. We reviewed >1,400 cases from comatose patients resuscitated from cardiac arrest who died after WLST with a pool of 35 international experts (3-5 experts per case, ~1,500 total expert-hours). Together with analytical colleagues, we developed multiple intuitive statistical approaches to infuse this expert knowledge into predictive models. We believe our work and results are of interest and broadly relevant to neurocritical care researchers and NCS conference-goers.
Supported by NINDS, we will present interim results from an R01 (R01NS124642) that aims to innovate novel methods that mitigate the risk of self-fulfilling prophecies by incorporating expert knowledge into prediction models. We reviewed >1,400 cases from comatose patients resuscitated from cardiac arrest who died after WLST with a pool of 35 international experts (3-5 experts per case, ~1,500 total expert-hours). Together with analytical colleagues, we developed multiple intuitive statistical approaches to infuse this expert knowledge into predictive models. We believe our work and results are of interest and broadly relevant to neurocritical care researchers and NCS conference-goers.
Speaker
Event Type
Breakout Session
TimeWednesday, October 16th9:45am - 10:05am PDT
LocationHarbor Ballrooms D-I
Science of Neurocritical Care
Basic/Neurocritical Care 101
General Critical Care
Intracerebral Hemorrhage
Multimodal Neuromonitoring (invasive/non-invasive)
Nursing Pharmacology
Pharmacist Practice
Stroke
Subarachnoid Hemorrhage
Traumatic Brain Injury
Intermediate