Temporal Variations in and Predictive Values of ABG results prior to In-Hospital Cardiac Arrest

Abstract

In-hospital cardiac arrest (IHCA) has been understudied relative to out-of-hospital cardiac arrest. Further, studies of IHCA have mainly focused on a limited number of pre-arrest patient characteristics (e.g., demographics, number and types of comorbidities). Arterial blood gas (ABG) analysis, one of the most common diagnostic tests for assessing and managing critically or acutely ill hospitalized patients, reflects pathophysiological changes associated with adverse events or complications, including IHCA. Yet the predictive and prognostic values of patterns of pre-arrest ABG parameters for IHCA have not been fully studied. The purpose of this retrospective pilot cohort study was to investigate temporal variations in and predictive values of pre-IHCA ABG values among patients with a history of cardiopulmonary disease. Eligible patients had a history of structural heart disease, heart failure, or pulmonary disease. Patients were excluded if their IHCA was due to trauma, drug overdose, hypothermia, drowning, chronic terminal illness such as cancer or human immunodeficiency virus, or bleeding not caused by hemorrhage in the brain or heart. Also collected were dates, times, and causes of mechanical intubation prior to IHCA and cause of mortality. Co-primary outcomes were initial rhythms of IHCA and return of spontaneous circulation (ROSC). We conducted a pilot study and the ABG results (pH, partial pressure of carbon dioxide [PCO2], partial pressure of oxygen [PaO2], bicarbonate [HCO−3], and lactate) from each of the 3 days prior to IHCA were extracted from the electronic health records of patients (N = 44) who had experienced IHCA at a single medical center. To characterize differences in ABG parameters among study days, coefficients of variation (CVs) were compared using the modified likelihood ratio test (MLRT) using the worst ABG values. Linear regression models were run for the continuous ABG parameters and logistic regression models for the dichotomous ABG variables. Overall model effect and least squares means, SDs, mean differences within and between days (with 95% confidence intervals), p-values and effect sizes are reported for continuous variables. For categorical variables, estimates and standard errors, 95% confidence intervals, Wald X2 variables and p-values are presented. The CVs for pH, PaCO2, and HCO3− differed significant between study days (p <.05). The least squares means with 95% confidence intervals for pH and lactate differed significantly in days (p <.01). Moderate to large effect sizes were obtained for all ABG parameters. Arterial lactate predicted initial rhythm (shockable versus non-shockable) and ROSC, while pH and HCO−3 predicted ROSC. Results demonstrate, for the first time, the presence of significant variability in ABG parameters across 72hr prior to IHCA and the predictive potential of these parameters for initial rhythms of IHCA and ROSC. While validation in a larger sample is necessary, this study confirms the feasibility and potential value of exploring temporal patterns of pre-arrest ABG values from the EHR. Findings of future larger studies on pre-arrest patterns of ABG parameters and other laboratory values may be used to design models that better predict risk for IHCA and guide patient care in the pre and intra-arrest periods.