In this retrospective, cohort study, we reviewed the APACHE III database of patients admitted to the MICU of Mayo Medical Center, Rochester, Minnesota. Mayo Medical Center is a tertiary, teaching institution with two hospitals comprising approximately 1,900 in-patient beds. The study was approved by the Mayo Foundation Institutional Review Board. The MICU was a closed unit throughout the study period. It had a 15-bed capacity at the beginning of the study period. The capacity was increased to 19-bed in August 2002 and 24-bed in December 2002. A critical care service team consisting of attending intensivists, critical care fellows, residents, pharmacists, nurses and respiratory therapists staffed the MICU. The non-physician staffing was consistent throughout the study period. The nurse to patient ratio was 1 to 1 or 1 to 2. Nurses, pharmacists and respiratory therapists participated during the daily rounds. All attending intensivists had internal medicine background and critical care or pulmonary/critical care subspecialty training. Fellows and internal medicine residents provided 24-hour in-house coverage. The attending intensivists did not routinely stay in-house at night but were available by phone and came to the ICU as needed. Clinically important decisions in the ICU were made by, or under the direct supervision of, the attending intensivists.
Patients who did not authorize their medical records to be reviewed for research were excluded. Data retrieved included demographics, MICU admission diagnosis; Acute Physiology Score (APS), APACHE III score, and hospital predicted mortality rate based on the first MICU day values; length of MICU stay and hospital mortality. The hospital predicted mortality rates were calculated based on the admission diagnoses, APACHE III score and location prior to MICU admission, using software provided by Cerner Corporation (Kansas City, Missouri) . Subgroup analyses were performed based on the severity of illness and the length of MICU stay. The severity of illness was categorized into high and low using the median predicted hospital death as a cutoff point. The MICU length of stay was categorized into long and short using the median MICU length of stay as a cutoff point.
We developed and started implementing four evidence-based protocols as follows: lung protective strategy for ALI in February 2002, activated protein C for severe sepsis/septic shock in October 2002, intravenous insulin for hyperglycemia control in September 2003 and a protocol for sedation/analgesia in April 2004. The protocols were developed with the participation of all MICU staff including physicians, nurses, respiratory therapists and pharmacists. The protocol for lung protective strategy was based on providing tidal volume not greater than 6 mL/kg ideal body weight in patients with ALI or Acute Respiratory Distress syndrome (ARDS). The activated protein C protocol was applicable for adults with severe sepsis/septic shock and multiple organ failure with no risk factor for bleeding and who opted for full resuscitation and life support. The hyperglycemia control protocol was activated if patients' glucose was > 150 mg/dL. A continuous intravenous insulin infusion was titrated to maintain blood glucose level between 100 and 119 mg/dL. There was also a protocol for the treatment of hypoglycemia based on symptoms or blood glucose level < 60 mg/dL. The sedation/analgesia protocol had two parts, one for patients anticipated to remain intubated for 48 hours or less and another protocol for those expected to remain intubated longer than 48 hours. The less than 48 hours sedations/analgesia protocol used morphine or fentanyl for analgesia and propofol or midazolam for sedation. The longer than 48-hour protocol used morphine or fentanyl for analgesia and lorazepam for sedation. Both analgesia/sedation protocols used numeric pain scales and Richmond Agitation-Sedation Scale (RASS) for titration and the continuous infusion of the opioids and sedatives was interrupted daily. (The sedation/analgesia protocols are available as additional file 1 and 2.)
The development of each protocol had taken several months before implementation. Once implemented, no significant modifications were made in any one of the protocols during the study period. We considered the times from January 2000 through January 2002 as the pre-protocol period, from February 2002 through March 2004 as the transition period (since introduction of the three protocols started during this period) and from April 2004 through June 2005 as the protocol period (all four protocols were implemented). During the pre-protocol period, the implementation of evidence-based practice was based on the individual physicians' knowledge and discretion. During the development of the protocols, all MICU staff became more familiar with the available evidence. With the activation of the protocols, the elements that constitute evidence-based practice were easily available in order-set forms and were implemented with the active participation of intensivists, fellows, residents, nurses, respiratory therapists and pharmacists.
The 28-day ICU free days were calculated by subtracting the actual ICU length of stay in days from 28. The 28-day ICU free day was considered 0 if a patient died before hospital discharge or stayed in the ICU for > 28 days [21, 22]. We used the 28-day ICU free days to avoid the confounding effect of mortality. This number measured the time interval that a patient was both alive and did not require ICU support.
We summarized data as mean (standard deviation) (SD), median (interquartile range) (IQR) or percentages. Student's t, Kruskal-Wallis, Mann-Whitney U, and chi square tests were used to compare differences between groups. In order to determine the impact of the protocol on severity adjusted patient outcome, we created a multiple logistic regression model consisting of hospital mortality as a dependent variable and the APACHE III predicted hospital mortality rate and the three study periods as independent variables. All independent variables were entered into the model simultaneously. The pre-protocol period was considered as reference in this logistic regression model. When appropriate, the 95% confidence intervals (CI) were calculated. We considered P values < 0.05 as statistically significant. We used StatView version 5.0 (SAS Inc, Cary, North Carolina) and Confidence Interval Analysis version 2.0.5 (Trevor Bryant, University of Southampton, United Kingdom) computer softwares for statistical analysis. We used variable life-adjusted display (VLAD) to show the differences between the cumulative and actual deaths during the three periods of the study [23, 24].