Prolonged intensive care unit (ICU) stays, variably defined as > 48 h to > 14 days, are a known complication of cardiac surgery.1-8 Prolonged stays are associated with higher resource utilization and higher mortality.2,3,9-12 Although there are several cardiac surgery risk models that can be used preoperatively to identify patients at risk for prolonged ICU stay, factors that influence outcomes for patients who experience prolonged ICU stays are poorly understood.2,13-19 Little information is available to inform discussions between health care practitioners (HCPs) and patients throughout a prolonged ICU stay, especially those ≥ 7 days.
As cardiac surgical complexity, patient age, and preexisting comorbidities have increased over time, so has the need to provide patients and HCPs with data to inform decision making, enhance prognostication, and set realistic expectations at varying time intervals during prolonged ICU stay. The purpose of this study was to evaluate short- and long-term outcomes in cardiac surgery patients after prolonged ICU stays at relevant time intervals (7, 14, 21, and 28 days) and to determine factors that may predict a patient’s outcome after a prolonged ICU stay.
The University of Michigan Health System Institutional Review Board approved this study and waived informed consent. We merged the University of Michigan Medical Center Society of Thoracic Surgeons (STS) database, which is updated periodically with late mortality, with elements of the electronic health record (EHR). Adult patients were included if they had cardiac surgery at the University of Michigan between January 2, 2001, and December 31, 2011. Late mortality was updated through December 1, 2014. Data are presented as frequency (%), mean (SD), and median (IQR) as appropriate. Bivariate comparisons between survivors and nonsurvivors were done with χ2 or Fisher exact test for categorical data, Student t test for continuous normally distributed data, and Wilcoxon rank sum test for continuous not normally distributed data. To determine factors associated with operative mortality (death within 30 days of surgery or hospital discharge, whichever occurred later), we used logistic regression with forward selection. All available factors were initially entered in the models.
Separate logistic models were created based on all data available at days 7, 14, 21, and 28. Final models consisted of factors with statistically significant P values (< .05) and adjusted odds ratios (AORs) with 95% CIs that excluded 1. To determine factors associated with late mortality, we used a Cox proportional hazard model, which used data available at discharge and STS complications. As these complications did not include their timing, they could only be used in models created at discharge and not for days 7, 14, 21, and 28 models. Final models consisted of factors with P values < .05 and 95% CIs of the AORs or the hazard ratios (HRs) that excluded 1. As the EHR did not start recording data until January 2, 2004, and its capture of data remained incomplete for several years, rather than imputing these missing data or excluding these patients, we chose to create an extra categorical level for each factor to represent missing data. For continuous factors with missing data, we first converted the continuous data to terciles and the missing data became the fourth level.20,21
The discrimination of the logistic models were determined by the c-statistic and for the Cox proportional hazards model with the Harrell concordance index (C index). Time trends were assessed with the Cochran-Armitage trend test. P < .05 was deemed statistically significant. Statistics were calculated with SPSS versions 21-23 or SAS 9.4.
Of 8309 admissions to the ICU after cardiac surgery, 1174 (14%) had ICU stays ≥ 7 days, 386 (5%) ≥ 14 days, 201 (2%) ≥ 21 days, and 80 (0.9%) ≥ 28 days. The prolonged ICU study population was mostly male, White race, with a mean (SD) age of 62 (14) years. Patients had a variety of comorbidities, most notably 61% had hypertension and half had heart failure. Valve surgery (55%) was the most common procedure (n = 651). Twenty-nine percent required > 1 procedure (eAppendix 1, available at doi:10.12788/fp.0300).
The operative mortality