Time to Awakening and Self-Fulfilling Prophecies After Cardiac Arrest.

OBJECTIVES: Withdrawal of life-sustaining therapies for perceived poor neurologic prognosis (WLST-N) is common after resuscitation from cardiac arrest and may bias outcome estimates from models trained using observational data. We compared several approaches to outcome prediction with the goal of identifying strategies to quantify and reduce this bias. DESIGN: Retrospective observational cohort study. SETTING: Two academic medical centers ("UPMC" and "University of Alabama Birmingham" [UAB]). PATIENTS: Comatose adults resuscitated from cardiac arrest. INTERVENTION: None. MEASUREMENTS AND MAIN RESULTS: As potential predictors, we considered clinical, laboratory, imaging, and quantitative electroencephalography data available early after hospital arrival. We followed patients until death, discharge, or awakening from coma. We used penalized Cox regression with a least absolute shrinkage and selection operator penalty and five-fold cross-validation to predict time to awakening in UPMC patients and then externally validated the model in UAB patients. This model censored patients after WLST-N, considering subsequent potential for awakening to be unknown. Next, we developed a penalized logistic model predicting awakening, which treated failure to awaken after WLST-N as a true observed outcome, and a separate logistic model predicting WLST-N. We scaled and centered individual patients' Cox and logistic predictions for awakening to allow direct comparison and then explored the difference in predictions across probabilities of WLST-N. Overall, 1,254 patients were included, and 29% awakened. Cox models performed well (mean area under the curve was 0.93 in the UPMC test sets and 0.83 in external validation). Logistic predictions of awakening were systematically more pessimistic than Cox-based predictions for patients at higher risk of WLST-N, suggesting potential for self-fulfilling prophecies to arise when failure to awaken after WLST-N is considered as the ground truth outcome. CONCLUSIONS: Compared with traditional binary outcome prediction, censoring outcomes after WLST-N may reduce potential for bias and self-fulfilling prophecies.

A novel methodological framework for multimodality, trajectory model-based prognostication.

INTRODUCTION: Prognostic tools typically combine several time-invariant clinical predictors using regression models that yield a single, time-invariant outcome prediction. This results in considerable information loss as repeatedly or continuously sampled data are aggregated into single summary measures. We describe a method for real-time multivariate outcome prediction that accommodates both longitudinal data and time-invariant clinical characteristics. METHODS: We included comatose patients treated after resuscitation from cardiac arrest who underwent ≥6 h of electroencephalographic (EEG) monitoring. We used Persyst v13 (Persyst Development Corp, Prescott AZ) to generate quantitative EEG (qEEG) features and calculated hourly summaries of whole brain suppression ratio and amplitude-integrated EEG. We randomly selected half of subjects as a training sample and used the other half as a test sample. We applied group-based trajectory modeling (GBTM) to the training sample to group patients based on qEEG evolution, then estimated the relationship of group membership and clinical covariates with awakening from coma and surviving to hospital discharge using logistic regression. We used these parameters to calculate posterior probabilities of group membership (PPGMs) in the test sample, and built three prognostic models: adjusted logistic regression (no GBTM), unadjusted GBTM (no clinical covariates) and adjusted GBTM (all data). We compared these models performance characteristics. RESULTS: We included 723 patients. Group-specific outcome estimates from a 7-group GBTM ranged from 0 to 75%. Compared to unadjusted GBTM, adjusted GBTM calibration was significantly improved at 6 and 12 h, and time to an outcome estimate <10% and <5% were significantly shortened. Compared to simple logistic regression, adjusted GBTM identified a substantially larger proportion of subjects with outcome probability <1%. CONCLUSIONS: We describe a novel methodology for combining GBTM output and clinical covariates to estimate patient-specific prognosis over time. Refinement of such methods should form the basis for new avenues of prognostication research that minimize loss of clinically important information.