Patients and setting
This study was conducted at Aker University Hospital in Oslo, Norway, between1994 and 2004. During the study period, the hospital had 350 beds and served apopulation of 500,000 people for urology and abdominal vascular surgery, and180,000 people for internal medicine, general surgery and psychiatry.
Medical records for all adult (≥ 16 years) patients admitted duringthe study period with culture-verified bacteremia due to E.coli or S. pneumoniae infection were retrievedfrom the hospital’s bacteriology laboratory database. Patients who hadmore than one episode of bacteremia during the study period were registered onlyonce in the study. As we wanted to study community-acquired infections, weincluded only patients who had blood cultures drawn on the day of or day afterhospital admission. Only patients with medical records available were includedin the study.
Clinical data
The following clinical data on comorbidities, risk factors for infection,diagnoses, signs and symptoms were extracted from medical records for allpatients included in the study.
Comorbid illnesses specified in the medical records were extracted andcategorized using a predefined list. Malignant disease was registered in casesof cancer or hematological malignancy. Alcoholism was registered whenaccompanied by organ involvement or social decompensation. Chronic renal failurewas registered if repeated creatinine values > 500 μmol/L in precedingadmissions, differentiated as severe if combined with dialysis or medicationspecific for renal failure, and as moderate chronic if neither dialysis normedication specific for renal failure was recorded. Heart failure andcardiomyopathy were both registered as heart failure.
Risk factors for infection included having an indwelling urinary catheter,surgical procedure at site of infection within the two weeks prior to admission,obstruction of the gastrointestinal or urinary tracts, and chronic inflammation.Medication with implicit risk for infection included use of corticosteroids≥ a dose equivalent to 10 mg prednisolone per day, chemotherapy in the twoweeks before admission or other immunosuppressive medication on a dailybasis.
Tentative diagnoses by the admitting physicians were categorized into infection,non-specific diagnoses (including delirium and acute deterioration in theability to perform daily tasks), organ-specific diagnoses not indicating aninfection (i.e. myocardial infarction, acute abdominal pain, acute asthma), andmissing/others.
Symptoms indicative of infection preceding admission were dichotomized into“classical symptoms” and “atypical symptoms”.“Classical symptoms” included fever/chills, localized pain,nausea/vomiting, diarrhea, cough, dyspnea, expectoration, urinary urgency,painful voiding, hematuria, skin rash, coma, and seizures, whereas“atypical symptoms” included malaise, falls, dizziness, syncope,unsteadiness, immobility, acute urinary or fecal incontinence, paresis, speakingdifficulties, and confusion.
Signs of infection in the emergency department (ED) included decline in generalhealth if recorded. Findings during the physical examination indicative oflocalized pathology were recorded, and markers of systemic inflammatory responsesyndrome (SIRS) were registered according to international standards [15]. The SIRS criteria include body temperature more than 38.0°C orless than 36.0°C; heart rate more than 90 beats per minute; tachypneamanifested by a respiratory rate more than 20 breaths per minute or as a partialpressure of CO2 below 4.30 kPa; and a white blood cell count greaterthan 12,000/mm3 or below 4,000/mm3. The SIRS criteria wereconsidered not met if data were not recorded. We used two alternative cut pointsfor SIRS, ≥ 2 criteria met (SIRS-2) and ≥ 3 criteria met (SIRS-3).Cut points from the Simplified Acute Physiology Score (SAPS) [16] were used to define hypothermia (body temperature less than36.0°C), fever (body temperature ≥ 38.5°C), leukocytosis(leukocyte counts above 15,000/mm3), and leukopenia (leukocyte countsbelow 3,000/mm3). C-reactive protein (CRP) values from blood samplesdrawn on the day of admission were categorized at 80 mg/L, which is applicablefor predicting sepsis in patients with SIRS [17], and 200 mg/L, which is the suggested level for differentiatinginfection from other causes of shock [18]. We included new-onset atrial fibrillation as a marker of severeinfection, as described previously [19].
Presumed primary site of infection was identified by one of the clinicallytrained authors (ALW) based on the medical history, symptoms, physicalexamination, blood tests, X-rays, specimen cultures from other body sites thanblood, biopsies from surgical procedures, and autopsies. The sites of infectionwere categorized into urinary tract, lower respiratory tract, other (i.e.gastrointestinal tract, liver, pancreas and biliary tract, central nervoussystem), or inconclusive.
Criteria for organ failure
Criteria for organ failure within one day after admission are presented in theAdditional file 1. Whenever possible, criteria weredefined according to the Sequential Organ Failure Assessment (SOFA) score system(cut point 2 or 3) [20]. Indicators for organ dysfunction, defined in the diagnostic criteriafor sepsis in 2001 [21] and for severe sepsis and septic shock in 1992 [15], were also used. Criteria for acute renal failure were adjusted tothe modified risk, injury, failure, loss and end-stage kidney (RIFLE) criteria [22], and on clinical presentation. Since the central nervous system isincluded in organ failure scoring systems for use in sepsis [23], we included impaired consciousness as an indicator of organ failure.However, signs of delirium were not included, because data on this state werenot routinely collected upon admission. Data on liver and hematological markersas well as markers of peripheral perfusion such as serum lactate were notsystematically registered in patient records, and were therefore excluded.
Date of death
Date of death during hospitalization was extracted from patient records. Foranalytical purposes, mortality was classified into early hospital mortality(within ≤ 3 days of admission), and in-hospital deathwithin 14 days of admission. Prior to the data extraction process survivalafter discharge from hospital had been confirmed through the National PopulationRegister by the medical record staff. If death had occurred after the indexstay, they had put the date onto the records.
Statistical methods
In order to study any systematic differences in clinical presentation related tothe oldest patients, descriptive analyses were performed for three age groups(< 65 years, 65–84 yearsand ≥ 85 years). In the multivariate analyses, however,age was dichotomized (< 65 and ≥ 65) based on preliminaryanalyses. Categorical variables were presented as absolute numbers andpercentages and compared using Chi-squared tests. Normally distributed numericalvariables were compared using one-way ANOVA, and non-normal variables usingKruskal-Wallis tests and Mann–Whitney tests. The number of“classical” symptoms was dichotomized at three symptoms,“atypical” symptoms were dichotomized at one symptom.
Non-parametric correlation analysis (Spearman rho) was performed to study therelationship between CRP value at admission and the number of failing organswithin one day of admission. The associations between organ failure anddifferent cut-points of CRP and different number of SIRS criteria were exploredusing Chi-squared tests.
In order to identify factors recorded upon admission to the ED independentlyassociated with either early organ failure or in-hospital death (truncated at14 days after admission to hospital), variables significantly associatedwith these outcomes (p < 0.05) in bivariate analyses were enteredinto binary logistic regression models. Ordinal factors not linearly associatedwith either of the two outcomes were dichotomized. For variable selection, weused backward stepwise removal of variables based on likelihood-ratio judgments.Model summary given in Nagelkerke R square and model of fit given by the Hosmerand Lemeshow test were applied. We also tested for any interactions between thedichotomized age variable and the other factors in the full main effects models.To obtain the logit of the two outcomes when interactions were active, the macroModprobe developed and adjusted to SPSS by Hayes and Matthes was applied [24]. However, since the statistical power of interaction analyses isgenerally low, the effects of interacting variables on outcomes are presentedonly as directions rather than graphically or by numbers.
One-year survival by number of early failing organs, bacterial species, and agewere analyzed using Kaplan Meier survival analysis, applying the log-rank test.A Kaplan Meier plot was used to present the results graphically. All analyseswere performed with SPSS 17.0 software (SPSS, Chicago, IL).
Ethical considerations
The study was approved by the South-East Norway Regional Committee for Ethics inMedical Research. The Norwegian Data Inspectorate gave permission to carry outthe study without the patient consent. Dispensation of professionalconfidentiality was given by the Norwegian Directorate of Health.