Study design and setting
This is an observational multicenter study on an existing database of prospectively collected data as part of an ongoing quality improvement program in three Dutch EDs, which has been described in detail previously (online supplementary file 1) [17]. In the Leiden University Medical Centre (LUMC), data were collected from 1 April 2011 to 1 February 2016, in the Rijnstate Hospital from 1 March 2012 to 1 April 2013, and in the Albert Schweitzer Hospital (ASZ) from 1 September 2015 to 1 December 2015.
For the first aim of the study, we divided the patients into two age groups: < 70 years and ≥ 70 years. We compared how much fluids were administered per initial SBP categories in both groups.
For the second aim of the study, we first had to stratify patients according to their initial SBP, because low SBP is an important indicator to initiate fluid resuscitation. We stratified patients into a group with a low SBP (≤ 120 mmHg) or a “seemingly normal” SBP (> 120 mmHg), based on the study by Oyentunji et al. [16], in which it has been suggested that 120 mmHg is a hypotension threshold for older patients. Secondly, fluid volume itself is a measure of disease severity not captured in initial disease severity scores because it partially reflects response to ED treatment. In multivariable regression models, fluid volume administered in the ED has been found to be an independent predictor of mortality [18]. Hence, patients who received high fluid volumes could not simply be compared to those who received low fluid volumes. It has also been suggested that older and younger patients receive similar amount of fluids [11]. Therefore, to investigate whether the current fluid resuscitation strategy, in which older and younger patients receive similar amount of fluids, is associated with a higher mortality in older patients, we had to divide patients into three fluid volume groups: 0–1 L, 1–2 L and > 2 L. Older patients were compared to younger patients in the same fluid category and same initial SBP category. The stratification is shown in Fig. 1.
Selection of participants
Consecutive patients ≥ 17 years and urgent triage categories who were hospitalized with a suspected infection after receiving intravenous antibiotics in the ED were included. There were no exclusion criteria.
The study was approved by the medical ethics committee of the LUMC.
Methods and measurements
Demographics, comorbidities, laboratory values, vital signs, treatment administered (including antibiotics, intravenous fluids, and oxygen), disposition from the ED, and outcomes were collected as described previously [17].
SBP was measured non-invasively with the MP52 monitor (Philips, Eindhoven, The Netherlands) in LUMC, with the M8007A Intellivue (Philips, Eindhoven, The Netherlands) in ASZ, and with the Infinity C700/M540 (Dräger medical systems, Telford, USA) in Rijnstate Hospital. Initial SBP was measured within half an hour after ED registration and divided in four categories: ≤ 100, 101–120, 121–140, and ≥ 140 mmHg, based on the threshold for hypotension in the quick Sequential Organ Failure Assessment Score (qSOFA) and the mean SBP in the general Dutch population of ≥ 70 years [13, 19].
The Predisposition, Infection, Response and Organ failure (PIRO) score was used as a measure of disease severity, taking into account demographics, comorbidities, and acute physiology parameters [20, 21].
Admission to a medium or intensive care unit (MICU) was used as a reflection of response to ED treatment and disease severity, not captured in the initial PIRO score.
Fluid administration
All types of fluids, i.e., colloids and crystalloids (mostly NaCl 0.9%), were used. If patients had arrived by ambulance, the amount of fluids in the ambulance was taken into account and added to the amount given during the total ED stay. If a registration form was missing, it was assumed that no fluid was given. Fluid volumes administered after ED departure were not registered.
The primary outcome measure was in-hospital mortality.
Data analysis
Sample size
The second aim of the study required the largest sample size and was therefore used to calculate the necessary number of events. Approximately 5–9 events per variable have been shown to be acceptable in association studies [22]. We corrected for 5 confounders; hence, we needed approximately 25 events per group. Because the second part of the study was intended as hypothesis generating, smaller number of events was considered to be acceptable as long as the regression model did not give unacceptably high 95% confidence intervals (CI).
Descriptive statistics
Data were presented as mean (standard deviation (SD)) when normally distributed. Skewed data were presented as median (interquartile range (IQR)). Categorical data were presented as number (%).
Main analysis
For the first aim of the study, the association between the initial systolic blood pressure and the amount of fluid administered in the ED was assessed through a line graph with 95% CI bars, comparing older and younger patients.
For the second aim of the study, patients were divided into two initial SBP categories: ≤ 120 and > 120 mmHg. The patients in the two SBP categories were thereafter split into three different fluid categories. An association model [23], as opposed to a prediction model, was developed per fluid category in order to explain the relation between age and outcome in patients, who received similar fluid volumes and presented in the same initial SBP category. Based on previous studies [17, 18], the following confounders were entered into the model: type of hospital (urban vs academic), oxygen administration, “Do Not Resuscitate (DNR) code”, the Predisposition and Infection (PI) score and Response and Organ Failure (RO) score [20], and MICU admission. The PI score reflects the potentially non-modifiable aspects, taking age, comorbidities, and type of infection into account, while the RO score represents the modifiable aspects, based on acute physiology parameters and extent of organ failure [21, 24]. These variables all met the criteria to be a potential confounder as they are associated with the primary determinant and the outcome [25]. To investigate if the primary association of interest could be corrected for less confounders (because of the limited number of event per variable), we first entered each of the predefined potential confounders into the model with age and outcome. The variable resulting in the largest change in the regression coefficient of the association age and outcome was then added to the model, which subsequently became the new starting model. This procedure was repeated until the addition of a new variable changed the regression coefficient of the primary association of interest with less than 10%, which was considered irrelevant. Multivariable logistic regressions were performed separately for the three fluid categories, each time comparing the older group to the younger group, with the younger group as reference. This data analysis design (Fig. 1) will answer the question of whether older and younger patients receiving similar fluid volumes at the same initial SBP affects in-hospital mortality in older patients.
Odds ratios (OR) are reported with 95% confidence intervals (95% CI). An α of 0.05 was used to distinguish statistically significant results. Data were processed using SPSS (SPSS, version 23.0, IBM, New York, USA).
Sensitivity analyses
We performed several sensitivity analyses: First, to investigate the impact of the number of variables in the final model on effect size, we did a sensitivity analysis in which we excluded the variables with the smallest impact on the regression coefficient (see Additional file 1). Secondly, chronic heart disease was also added into the association model building process to see if it had an influence on the association age and outcome (see Additional file 2).