Temporal patterns of respiratory rate exposure and mortality in critically ill patients: a retrospective multi-database cohort study

Abstract

To investigate the associations between respiratory rate (RR) exposure patterns—both intensity and duration—and mortality in critically ill patients. This retrospective cohort study analysed 160,075 adult ICU patients from three international databases (HiRID, MIMIC-IV, eICU). The primary analysis used a landmark design: RR exposure and all covariates were ascertained during the first 24 h after ICU admission, and outcome follow-up began at the 24-hour landmark. Heatmap visualisation examined RR intensity-duration patterns. Cox and logistic regression models related the proportion of monitoring time spent in each RR range to mortality, adjusted for demographics, disease severity, laboratory values, comorbidities, treatment limitations, ventilation and sedation. Restricted cubic splines assessed the dose-response relationship for time-weighted mean RR. Sensitivity analyses examined alternative exposure windows, competing risks, multiple imputation, hospital-level clustering and a harmonised in-hospital mortality outcome. A greater proportion of time in the 12–20 breaths/min range was associated with lower 28-day mortality in MIMIC-IV (HR 0.988, 95% CI 0.987–0.989 per 1 percentage point) and lower in-hospital mortality in eICU (OR 0.991, 95% CI 0.990–0.991). Risk rose progressively across the tachypnoeic ranges (>35 breaths/min: HR 1.026, 95% CI 1.023–1.028). Spline analysis identified a nadir of risk at a time-weighted mean RR of 16.3 breaths/min, with a markedly asymmetric relationship ( P for non-linearity <0.001): the adjusted hazard ratio reached 1.38 at 20 and 2.19 at 24 breaths/min but rose only modestly below the nadir. The bradypnoea-mortality association seen with whole-stay exposure (HR 1.029) was absent under the landmark design (HR 0.990) in both databases. Tachypnoeic associations were robust to stratification by ventilation and sedation, clustering and multiple imputation. Respiratory rate showed an asymmetric, non-linear relationship with mortality, with lowest risk at approximately 16–17 breaths/min. The cumulative burden of tachypnoea during the first 24 h was consistently associated with higher mortality across three databases and robust to controls for reverse causation, ventilation and sedation; the apparent association for bradypnoea was not, and likely reflects terminal-phase respiratory depression. Respiratory rate should be regarded as a monitoring and early-warning signal that integrates both magnitude and duration, rather than as a therapeutic target.