Study population
This was a multi-site retrospective observational cohort study consisting of adult
patients with sepsis admitted to the Foothills Medical Centre (FMC), Peter Lougheed
Centre (PLC) or Rockyview General Hospital (RGH) through the ED between January 1,
2006 and September 30, 2009 in the Calgary zone of Alberta Health Services (AHS).
The Calgary zone of Alberta Health Services provides virtually all acute hospital
care to the residents of the cities of Calgary and Airdrie and surrounding communities
in the province of Alberta, Canada.
Sepsis and severe sepsis were defined as described in the Canadian Institute for Health
Information “In Focus: A National Look at Sepsis†4]. Sepsis was defined as having one of the following ICD-10-CA codes: A03.9, A02.1,
A20.7, A21.7, A22.7, A23.9, A24.1, A26.7, A28.0, A28.2, A32.7, A39.2, A39.3, A39.4,
A40.–, A41.–, A42.7, B00.7, B37.7, P36.–, P35.2, P37.2 and P37.5. The diagnosis type
was modified to only include the main diagnosis (M), pre-admission comorbidity (1)
or second pre-admission comorbidity (3) to better identify patients with sepsis at
the time of hospital admission. Additional information on ICD-10-CA codes can be referenced
in the International Statistical Classification of Diseases and Related Problems documentation
14]. Severe sepsis was defined as sepsis with the addition of organ dysfunction occurring
in at least one of the following six systems from the following associated ICD-10-CA
codes, Hematologic D69.5, D69.6, D65; Cardiovascular R57.–, I95.1, I95.8, I95.9; Hepatic
K72.0, K72.9, K76.3; Neurologic F05.0, F05.9, G93.1, G93.4, G93.80; Renal N17.–; and
Respiratory J96.0, J96.9, J80, R09.2.
Data sources
The study population was extracted from three administrative databases by a trained
and experienced health analyst within Alberta Health Services independent of the investigators.
These databases were the Inpatient Discharge Abstract Database (DAD) which includes
all Health Records related information such as ICD10 diagnosis and in-hospital outcome,
the Admission, Discharge, Transfer (ADT) database which includes all physical bed-management
patient transfers in the acute care hospital and the Ambulatory Care Classification
System database (ACCS) which provides information on ED patient encounters in the
acute care hospitals. The DAD system, which provides the coded ICD10 diagnosis data
extracted from the patient’s chart, is maintained by professionally trained health
records coders and has been described previously 15].
Each database had a common unique patient identifier. All data were collected following
removal of any identifiable information such as name or address. The three databases
were then linked together using the patient identifier with the hospital date of admission,
and following data linkage, the individual patient identifier was encrypted to ensure
anonymity. All databases were imported into a Postgres relational database (http://postgresql.org) for data management and the creation of the study population. The Postgres database
was then interfaced to the statistical software, R version 2.10 (http://r-project.org), through ODBC (Open Data Base Connectivity) for the statistical analysis.
Operational definitions
The ACCS database had detailed information about the patient’s ED visit. This data
included time of arrival, triage level using the CTAS (Canadian Triage and Acuity
Scale) score 16] determined in accordance with their published guidelines, and time of first ED physician
assessment. The ED wait time used in our analysis was dichotomized as less than or
equal to 7Â h or greater than 7Â h; this variable was calculated as the difference from
when the patient was admitted to the ED and the time that the patient was admitted
to the hospital.
The ADT database included information on patient movement (flow) including time stamps
for admission/discharge/transfer into the hospital and all units throughout the hospital.
The time stamps are automatically generated through a bed management system linked
by an HL7 interface engine to the clinical information systems. ICU occupancy was
calculated from the ADT system at the time of first ED physician assessment assuming
that the physician was the most likely factor to determine the need for admission.
Outcome was extracted from the DAD database, and was defined as either all cause in
hospital mortality during the same admission or discharge alive irrespective of location
that the patient was discharged to (i.e. home, convalescent care, et cetera). The
DAD was also used to define sepsis and severe sepsis as described above. We applied
the ICD-10 coding algorithm to the DAD using the methods described by Quan et al.
17], and previously applied in critically ill patients 18] to determine the Charlson index score 19].
Statistical analysis
Analysis was performed using R version 2.14 (www.r-project.org). Normally or near normally distributed data were reported as Mean?±?SD and any comparisons
used the Student t test. Non-normally distributed data were reported using the Median and Interquartile
Range (IQR) and any comparisons used the Mann-Whitney U test. Categorical data were assessed using Fisher exact test for pair wise comparisons.
Logistic regression models were developed to examine the effect of independent risk
factors on in-hospital mortality.
Ethics approval was obtained from the Conjoint Health Research Ethics Board at the
University of Calgary. Individual consent was not required as this was a cohort study
with a large number of deaths (obtaining consent not practical) and all data made
available to the investigators was anonymized.