2.1. Aim: To describe the demographic profile of ED users in a tertiary hospital over a 5-year period and to investigate differences in outcomes by sex and age.
2.2 Design: Observational cross-sectional analytic study.
2.3. Study period: 1st January 2009 to 31st December 2013.
2.4. Setting: São Paulo, capital of São Paulo State is the largest city in Brazil, with an estimated population of 12 millions of people. Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (HCFMUSP) is a teaching hospital complex with 2,200 beds, serving as the referral center for the whole state. The Instituto Central, a tertiary university hospital, is the main unit within the Hospital das Clínicas complex. It has a capacity of approximately 1,000 beds, with 105 intensive care beds split between 10 intensive care units (ICUs). The ED of the Instituto Central of Hospital das Clínicas (PSHC) receives over 100,000 cases annually, with roughly 10,000 inpatient admissions. Cases are accepted in the following areas: general medicine, general surgery, trauma, neurology, neurosurgery, vascular surgery, urology and gynecology. The emergency services for obstetrics, ophthalmology and otolaryngology are located in the adjacent outpatient unit, and are not seen in PSHC.
2.5. Participants
2.5.1. Eligibility criteria
We considered for inclusion all patients aged 18 years or older attending PSHC. Those with obstetric, ophthalmological or otolaryngological problems were not eligible, since they were treated in the adjacent unit. We excluded records if they were incomplete or inconsistent, or if there were duplicates of the unique hospital attendance number or admission authorization form number. Furthermore, we excluded cases that resulted in hospital transfers; outcomes other than discharge or admission and patients that left without a medical consultation (LWBS) or against medical advice (LAMA).
2.5.2. Participant selection
On arrival at PSHC, the patient (or those accompanying them) provides their personal data and reason for attendance which are recorded in the electronic registration. The system generates a unique number for each separate PSHC attendance. We considered an ED visit to be complete if a medical evaluation was finished and an outcome recorded electronically (discharge, admission, hospital transfer or other). Patients requiring more than 12 hours of observation were admitted.
For each admission, the responsible doctor fills out an admission authorization form with the patient’s data; this generates a new admission authorization form number for the billing system.12,13 To conclude the admission, the doctor has to select the main ICD-10 code as well as the outcome (discharge, death, hospital transfer, self-discharge or other). If the patient dies in ED, it is standard practice to admit them on the system with the outcome coded as death.
To be eligible for inclusion, a medical evaluation had to be completed during the visit, and its outcome recorded as either admission or discharge. In addition, when the ED visit resulted in admission, we only included those with a final outcome coded as discharge or death. We excluded all other types of ED attendance and admission.
2.6. Variables
We analyzed eligible PSHC visits with respect to the following variables: age and sex of the patient; year of attendance; mode of presentation to PSHC being either ‘spontaneous’ (without prior evaluation by another service) or ‘referred’ (having already accessed a different health service, and arriving by ambulance or helicopter); time of ED visit (day shift 7AM-7PM or night shift 7PM-7AM); and ED outcome (admission or discharge). In cases resulting in admission we analyzed additional factors. These were type of hospitalization (surgical, clinical or other); main procedure (surgical, clinical or transplant-related); length of hospital stay (LOS); use or not of the ICU; length of ICU stay (ICU-LOS); and final admission\ outcome (discharge or death).
The categorical variables were the following: age-group, year of presentation, mode of presentation, reason for attendance, time of ED visit, type of hospitalization, main procedure, and ICU admission. We stratified age into the following groups: young adults (18-39 years), adults (40-59), young-older adults (60-79) and old-older adults (80-109). We categorized the reasons for attendance as either ‘external causes’, ‘general and localized symptoms’, ‘evaluation requested by another service’, ‘scheduled attendances’, or ‘other’. The continuous variables were age, LOS (the interval between the admission and discharge billing dates), and ICU-LOS (ICU days billed). Furthermore, we stratified LOS into six categories: 0-1, 2, 3, 4-7, 8-20, >=21 days of hospitalization. Hospital admissions lasting one day or less were grouped as one category (0-1).
The primary dichotomous outcomes were hospitalization (admission vs ED discharge), and mortality (death vs hospital discharge). The secondary dichotomous outcome was ICU admission (or not). The primary aim was to investigate associations between demographic characteristics (age and sex) and the outcome variables.
2.7. Data source
We retrieved routinely collected data from administrative electronic registers maintained by HCFMUSP, then consolidated them to produce a single dataset. ED attendance data are recorded in the hospital information system, and admissions data in the hospital billing system. In some cases, it was necessary to recode entries, depending on how they were recorded in the electronic system. Otherwise, we obtained data directly from the hospital databases.
2.8. Potential biases and analytic issues
This is an observational analytic study of electronic health data collected routinely for administrative purposes and for documentation of clinical care. Routinely collected health data are defined as those collected without an a priori research question.14 Guidelines such as STROBE and its extension, the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD), were developed to enhance the quality of observational research and the transparency of results.15,16 We used the STROBE and RECORD statements as reporting guidelines.
The study covers a five-year period, and some patients had multiple ED visits and admissions. We identified individuals with more than one ED visit, ranking them by total number of attendances over the 5-year period. As such, we determined an upper-limit for inclusion in the study.
The reasons for attendance were varied and numerous; 72 were recorded in the hospital information system. To facilitate the analysis, we assigned broader categories (‘external causes’, ‘general and localized symptoms’, ‘evaluation requested by another service’ , ‘scheduled attendances’, or ‘other’). The category of ‘scheduled attendance’, which describes non-emergency visits (e.g. returning for test results), represents neither ‘spontaneous’ nor ‘referred’ modes of ED presentation (see Variables in the main text), and was therefore defined as missing data.
During the study period, there were changes in the triage processes at PSHC. A new triage system based on individual clinical risk was implemented. We analyzed year of attendance and mode of presentation in order to identify any effect due to these changes.
The high number of study subjects (ED visits) demands a measure of effect size, such as an Odds Ratio (OR) (or log OR), to estimate the magnitude of effect or association between two or more variables.17,18,19 The effect size together with its confidence interval provides an estimate of the magnitude of an effect of interest and the precision of that estimate.20,21 Generalized linear mixed models (GLMM) for a given dichotomous outcome (dependent variable), using binomial probability distribution and logit link function, allow an estimate of ORs (with 95% confidence intervals) for independent variable categories in relation to respective reference categories.
2.9. Statistical analysis
We calculated summary statistics for PSHC visits and admissions. Categorical variables are presented as total count (n) and percentage. Continuous variables are presented as mean (standard deviation) or median (maximum and minimum values). Descriptive statistics were further stratified according to year, sex and age-group. For the multivariate analysis, generalized linear mixed models were built in order to investigate variables associated with the primary and secondary dichotomous outcomes. All three models had binomial probability distribution and logit link function Results are presented as odds ratios (OR) with 95% confidence intervals (CI). The significance level was set at 5% with Bonferroni correction. All analyses were conducted using SPSS Statistics version 25.0 (IBM, Corp., Armonk, NY).