Data collection
All data were retrieved from the patient data management system (Metavision, iMDsoft, Düsseldorf, Germany). Demographic data included age, gender, Simplified Acute Physiology Score as marker of disease severity (SAPS-3) [15], admission type (medical, elective surgery, emergency surgery including trauma), an extensive comorbidity profile and the reason for admission, divided into 8 medical and 8 surgical categories (Additional file 2: Table S2). We also collected data on renal function by assessing the RIFLE score based on creatinine level [16]. Because this score can be troubled by methodological difficulties, e.g., the incomplete availability of baseline creatinine, we also calculated the RIFLE score based on urine production using an automated search for the 6- and 12-h time frame during which the smallest amount of urine was produced on the first day of admission [16]. Both scores were used separately in the models due to the much higher sensitivity of the urine-based score.
As regards laboratory data, we used only data from each patient’s first day of admission for statistical analysis. Since we were interested in chloride level, at least one measurement had to be taken during the first day. If multiple measurements were available, the highest value was used. Other variables included plasma biochemistry data (sodium, potassium, albumin, phosphate, calcium, magnesium) and arterial blood gas analysis (pH, pCO2, lactate). We only included laboratory data that had been taken within a certain time from the referenced chloride value: 1 h for sodium, potassium and blood gas parameters, and 24 h for albumin, phosphate, calcium and magnesium, although in the majority of cases all measurements were performed on the same sample.
To calculate the physicochemical acid–base parameters, we used generally accepted formulas and defined apparent SID as follows: SIDa = [Na+] + [K+] + [Ca++] + [Mg++] ? [Cl?] (all values by convention in mEq/L) [12]. Since lactate is fully ionized in the pH range of 6–8, it also acts as a strong anion. We decided, however, to exclude lactate from the calculation of SIDa because we wished to specifically investigate the impact of the electrolytes. The inclusion of lactate could have led to ambiguous conclusions considering its well-established role as a marker of poor clinical outcome [17]. Effective SID was defined as the electrical charge attributed to the routinely measured weak anions and calculated as follows: SIDe = bicarbonate + albumin charge + phosphate charge (all values in mEq/L), where bicarbonate = 0.0301*pCO2*10pH?6.1, albumin charge = [albumin] in g/L*(pH*0.123–0.631) and phosphate charge = [phosphate] in mmol/L*(pH*0.309–0.469) [18, 19]. Strong ion gap (SIG), the sum of routinely unmeasured anions, was defined as SIG = SIDa ? lactate ? SIDe.
Arterial blood gases were analyzed on a point-of-care Rapidlab 1265 (Siemens Laboratory Diagnostics, Beersel, Belgium). Biochemistry analyses were performed using direct potentiometry (Vitros 950, Johnson Johnson, Ortho Clinical Diagnostics, Beerse, Belgium) until April 2010, after which the central laboratory changed to indirect potentiometry (Dimension Vista 1500, Siemens Laboratory Diagnostics, Beersel, Belgium). To match samples taken before and after this transition, Passing-Bablok regression equations were applied to the measurements taken before April 2010. To exclude residual confounding [20, 21] due to this transition, we also added the potentiometry type as a covariate in the statistical models and double-checked all conclusions using analyses with point-of-care data that were not influenced by a change in measurement technique (Additional file 3: Table S3, Additional file 4: Table S4).