Study setting
This study was conducted at the Emergency Department (ED) of a major urban teaching hospital and referral center in Beirut, Lebanon. The hospital is one of many in the Beirut area located in the center of the city, at a distance from the sites of violent events ranging between 1.4km to 12.5km. The ED was the most proximal site for only one of the events (Dec 27, 2013), with more proximal hospitals receiving the majority of the direct casualties for the remaining events. The ED is organized into a high acuity unit, a low acuity unit and a pediatrics unit; core ED personnel includes American Board-certified/eligible physicians in Emergency Medicine, as well as physicians without specific emergency training, but with extensive experience in emergency medicine. The number of visits to the ED is about 55,000 per year. This study was deemed exempt from human subject research by our Institutional Review Board.
Study design
We used a variant of case-control design to compare patient acuity and disposition in weeks where events happened, compared to weeks where no events took place. For each event, we defined an "event," or “case” week as the week starting on the day of the event. We defined two "no-event", or “control” weeks: the week before the event and the same week in the preceding year. We assumed that events occur randomly, and that event (case) and no-event (control) weeks are similar in all respects, except the occurrence of the events.
To study the impact of events on utilization of EDs, we compared ED daily visit volumes 30 days pre-event with ED daily visit volumes 30 days post-event. For this analysis, we considered only events preceded by at least two months of calm to allow for patterns of ED utilization to stabilize back to non-conflict routine. Three events fit these criteria (event 1:July 9, 2013; Event 2: Nov 19, 2013; Event 3: June 24,2014) and were used for the trend analysis.
Study protocol
In this analysis, we focused on the most recent years of violent events happening in Beirut (2013 and 2014). Supplementary Table 1 provided a summary of the events and the weeks included in the study [see Additional file 1].
We retrieved de-identified data from medical records for patients who visited the Emergency Department (ED) during the selected periods in 2012, 2013, and 2014. There were 7,874 visits to the ED during event weeks, and 15,193 during no-event weeks, for a total of 23,067 visits for which individual medical records were reviewed. The following data were extracted from each record: age, gender, nationality, residence, admitting diagnosis, how bill was paid (a variable associated with employment and socio-economic status), length of stay, discharge information (admitted or discharged), and discharge diagnosis.
Trained staff recoded the physicians' diagnoses to International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-CM9) codes (21). To reduce the number of codes and make comparative analyses more manageable, these diagnoses were further classified into single-level categories, using Clinical Classifications Software (CCS), which is the standard used for many analyses [24]. Thus, the 1,864 ICD-9-CM diagnoses were collapsed into 218 codes, and all the diagnoses used here refer to the single-level CCS categories that the ICD-9 coded diagnoses mapped to, rather than to the clinical definitions themselves.
For each patient who visited the ED during the study weeks, we also included information on the severity of the case. We used the emergency severity index (ESI), a well-validated triaging score which relies on trained nurses to make an acuity judgment based on the likelihood of immediate threats to life or organs, and to predict the number of resources that would be required in order to stream patients to appropriate care [25, 26]. A score of 1 or 2 indicates high acuity, a score of 3 intermediate, and a score of 4 or 5 low acuity; thus the scores were re-categorized into these three levels.
Statistical analyses
Descriptive statistics were used to compare the volume of visits and patients' characteristics for event and no-event weeks, using chi square, Fisher's exact tests, t-tests, or Cochrane-Armitage as appropriate. We used all the variables available in the medical record to compare patients in event and no-event weeks. Logistic regression was used to assess the significance of differences in the frequencies of CCS codes between event and no-event weeks.
Individual control charts (ICR) were used to assess impact of events on ED visit trends and assess for special cause variation (non-routine events). Baseline values were computed using daily ED visit data 30 days prior to each event with control limits set at 3 standard deviations (SD) above and below the center line, using Quantum XL. Time related variation was based on 2 rules: rule 1, where 6 or more consecutive points steadily increase or decrease; rule 2, where 15 consecutive points fall within +/- 1 SD on either side of the center line. Baseline data was compared to daily ED visit data up to 30 days post-event. When a period of calm was followed by a series of back to back events that were separated by less than one month, the post event period included all ED visits up to 30 days post last event in the series.
In addition, daily ED visit trends were further analyzed using interrupted time-series analysis for the period 30 days before the event, and 30 days after the event; segmented regression analysis was conducted using the newey command (considering Newey-West standard errors) in STATA version 15 (StataCorp LLC., College Station,TX). Statistical significance was defined as P < 0.05.