Study design
Our study was conducted at the ED of the Leiden University Medical Center, a tertiary university teaching and level 1 trauma hospital in the Netherlands. Patients aged 70 years and older that had attended the ED between 1 January 2012 and 31 December 2012 were included retrospectively. The Medical Ethics Committee of the Leiden University Medical Center waived the obligation of approval as data were collected in the past as part of routine clinical care.
Health care in the Netherlands
The Netherlands is a small and highly populated country in Europe measuring 41.5 thousand square kilometres [11] and counting 16.7 million people in 2012 [12]. Standard medical care is equally accessible for every Dutch citizen through legally mandatory health insurance. Primary care is provided by general practitioners (GPs). Specialist care can only be accessed after referral by a GP. One of the exceptions are EDs of hospitals, where a substantial proportion of patients are self-referred [13]. The Leiden University Medical Center is a tertiary referral centre in Leiden. The ED is one of two level 1 trauma EDs that together serve a catchment area of 400.000 inhabitants, both urban and rural. The population is predominantly Caucasian and includes all social classes. Our ED is equipped with 15 rooms of which three are specially designed to accommodate trauma victims. Patients are triaged by an ED nurse. Within hours self-referred patients are evaluated by an ED physician or ED resident. Out of hours self-referred patients are primarily evaluated by a GP and if indicated subsequently referred to an ED physician or ED resident. Referred patients are directly allocated to a resident of the appropriate medical specialty present at the ED. After evaluation patients are either treated at the ED and discharged home or admitted to an inpatient ward. Patients with an electrocardiogram indicative for myocardial infarction bypass our ED and are immediately referred to the catheterisation laboratory [14]. As a consequence, they are not included in the present study.
Selection of study population
Patients were identified in our computerised patient record system (ChipSoft-EZIS®, version 5.2, 2006–2014, Amsterdam, The Netherlands, www.chipsoft.nl). Several steps of exclusion criteria were applied. Our study was aimed at a selection of older patients that may benefit from additional interventions during or following an ED visit. First, medical records based upon unjustified ED use were excluded. Unjustified ED use was defined as ED use for any other reason than acute medical care, such as outpatient check-ups on weekends, plaster cast readjustments, performed blood tests for other medical departments and patients who decided to leave the ED before medical attention was bestowed. We believe these are not representative for the acutely presenting older patient visiting the ED and may disturb associations between predictors and outcome results. Second, patients who deceased in the ED and patients receiving cardiopulmonary resuscitation therapy upon arrival were excluded from analysis since prognosis of these patients is known to be poor and these patients fall outside the scope for identifying new predictors [15]. As we used retrospective data, we were unable to assess whether an ED visit was the first or one of many visits. Patients may have visited other hospitals as well as ours or made visits outside our selected timeframe. Therefore, we included only the first ED visit of each patient in 2012.
Potential predictors
Apart from demographic characteristics (age and gender), we selected routinely collected parameters that may reflect severity of disease as presented in the acute situation. We investigated time and way of arrival, presenting complaint, consulting medical specialty, vital signs, pain score and laboratory parameters. These data were automatically generated from the digital patient records and outliers were manually checked for validity by a researcher. Triage category was not included since we were interested in universal predictors and hospitals differ in the triage systems they use.
Time of ED visit was determined from ED registration time and subdivided in three categories, day (08.00 h–15.59 h), evening (16.00 h–23.59 h) and night time (00.00 h–07.59 h). Way of arrival at the ED was mutually exclusively noted as self-referral, brought in by ambulance, referral by a GP, internal referral from another department or referral by another hospital. Patients categorised as self-referral or referral by a GP visited the ED with private transportation. By contrast, patients who arrived by ambulance were categorised as brought in by ambulance regardless of whether the ambulance was ordered by a referring GP or because of an emergency call. Dutch ambulance staff is trained to judge the accuracy of emergency calls at the scene. Ambulance staff will only transport such patients to the hospital if they consider the referral justified. At our hospital, triage is based on the Manchester Triage System (MTS) [16]. This system uses flow charts for 55 disease presentations to determine the level of urgency and associated target time a patient should receive care from a physician. The presenting complaints of our study population were categorised according to these MTS disease presentations [16]. Disease presentations occurring in less than 3 % of patients were merged as ‘other’. The medical specialty a patient was assigned to was categorised as surgical or non-surgical [17]. Finally, we listed clinical measurements that were recorded in the ED: vital signs, pain score and laboratory results. At triage, an ED nurse determined which clinical measurements were medically indicated according to protocols. They were measured at triage or soon after a patient was placed into a treatment room. Laboratory testing is performed on indication and either ordered by an ED nurse or consulting physician. The first set of vital signs assessed in the ED was recorded. Vital signs were categorised according to the Modified Early Warning Score and included systolic blood pressure, heart rate, respiratory rate and body temperature [18]. Oxygen saturation was recorded as well [19]. Categories containing less than 1 % of patients were combined with adjacent categories, but not with the reference category, in order to minimise the number of categories. Pain was evaluated using the Numeric Rating Scale (NRS) rating from 0 to 10 and categorised as no or light (NRS 0–3), mild (NRS 4–6) and serious (7–10) pain according to the Dutch guidelines for pain classification in emergency settings [20]. Blood pressure, heart rate, respiratory rate and oxygen saturation were measured using a medical monitor (IntelliVue MP50®, Eindhoven, The Netherlands, www.philips.nl/healthcare). Body temperature was determined by a tympanic thermometer (Genius 2®, Mansfield, USA, www.covidien.com). Registered laboratory results were haemoglobin, thrombocytes, leukocytes, C-reactive protein, sodium, potassium, creatinine, urea, troponin T and non-fasted glucose. Vital signs and laboratory parameters will only be assessed if there is a medical indication to do so. If data on vital signs were missing, they were either not measured or they were measured but not recorded in the medical chart correctly. It is impossible to categorise this in a retrospective manner. Therefore, we assumed that missing vital signs meant that there was no indication to perform these measurements.
Primary end point
Our primary outcome measure was mortality in the first three months after ED admittance. Beyond this time period, the association of predictors measured at baseline and mortality is likely to be obscured by the occurrence of new medical events. Mortality data were acquired from the municipal personal records database on 1 May 2014.
Statistical methods
Data are displayed as mean and standard deviation if normally distributed and median and interquartile range if not normally distributed. To investigate the association between predictors and mortality we used Cox proportional hazards models. We performed uni- and multivariate Cox regression analysis. In the univariate models only one parameter was entered as independent variable. In the multivariate analyses, multiple parameters were entered as independent variables simultaneously to assess which were independent predictors of mortality. Our study was aimed at potential predictors assessed upon or soon after arrival at the ED. Results of laboratory testing became available at least one hour after withdrawal, but laboratory testing is usually ordered in the first few minutes after a patient is placed into a treatment room. Therefore, we added merely the medical indication to perform laboratory testing to the set of predictors in the multivariate model. As an in-depth analysis we have additionally analysed the univariate association of individual laboratory results with mortality using univariate Cox regression. The level of significance was set at P < 0.05. All statistical analyses were performed using IBM SPSS Statistics package (version 20).
