719

Abstract

Introduction: Accurate data collection is paramount to obtain unbiased point estimates. Studies in endotracheal intubation (ETI) often use retrospective methods. Hypothesis: Compared to data collected prospectively in the same cohort of patients, data collected retrospectively have a high degree of missingness and a low degree of agreement. Methods: Data were obtained from a completed randomized controlled trial comparing direct laryngoscopy to videolaryngoscopy for ETI in critically ill patients. Prospective data were collected at the time of ETI by research coordinators. Medical records of these patients were then reviewed retrospectively by a single investigator. Missingness was compared using McNemar’s test. Cohen’s Kappa was calculated for dichotomous variables. Bias and 95% limits of agreement (LOA) were calculated using Bland & Altman analysis. Results: Number of ETI attempts was recorded retrospectively in 39 of 40 cases (98%) with perfect agreement in 28 of 39 (72%). Overall correlation coefficient was 0.66. Data on glottic view were available in 39 of 40 (98%) of prospective and 29 of 40 (73%) of retrospective data (p=0.002) with a correlation coefficient of 0.94. Lowest mean arterial pressure (MAP) was missing for 4 of 40 (10%) and 15 of 40 (38%) of prospective and retrospective data respectively (p=0.007). Compared to prospective data, retrospective MAP data had a bias of -11 mmHg with a 95% LOA of -69 to 46 mmHg. Lowest arterial oxygen saturation (SaO2) was missing for 5 of 40 (13%) and 15 of 40 (38%) of prospective and retrospective data respectively (p=0.02). Bias comparing retrospective to prospective SaO2 data was -3% with a 95% LOA of -19 to 14%. Data on the ETI technique used were documented in 39 of 40 (98%) prospectively and 33 of 40 (83%) retrospectively (p=0.03) with perfect agreement. Conclusions: Data relating to number of ETI attempts demonstrated a low degree of missingness and good agreement. However, variables such as glottic view, MAP, SaO2, and ETI technique had significant missingness and/or poor agreement when comparing retrospective to prospective data collection. Impact of missingness and bias needs to be considered when interpreting retrospective studies on ETI.