We utilized the United States Consumer Product Safety Commission National Electronic Injury Surveillance System (NEISS) database for this study [10]. The NEISS collates data from approximately 100 participating hospitals that have been selected as a probability sample of all 5000+ emergency departments in the wider United States and United States territories and then extrapolates the data using strata-specific weights to generate national estimates. The sampling frame consists of five strata for hospital types based on hospital size and patient demographics. One stratum includes only children’s hospitals, and the remaining four strata are categorized based on emergency department visits: small (1–16,830), medium (16,831-28,150), large (28,151-41,130), and very large (41,131+). We utilized strata- and hospital-specific weights provided by NEISS to project national estimates. The methodology on the estimation of these weights is published elsewhere [11].
The data collected includes a general diagnosis, specific consumer product code, patient demographics, and brief narratives that describe other aspects of the patient visit in a de-identified manner. For each NEISS designated hospital, a specifically trained physician coordinator compiles the data to ensure a nationally standardized data collection. Additionally, NEISS incorporates sample weights and cluster variables to enable variance calculation and confidence interval estimates for data. Reports run through the NEISS provide the coefficient of variation, which is used to calculate the 95% confidence interval of the estimate [10]. This database has previously been used to characterize nationwide trends of consumer products across numerous specialties [8, 9, 12,13,14].
We queried the NEISS database for visits specifically related to “Scooters/skateboards, powered” (Code 5042). Within the results, only entries containing the word “scooters” were selected to filter out extraneous results and skateboard accidents. We analyzed data for years dating from 2013 to 2018 to capture and evaluate current estimates and trends. We then filtered entries with injuries specifically requiring hospital admission. Emergency room visits due to e-scooter injuries that did not require admission were excluded, as the primary focus was to define the burden of serious injuries and hospital admissions. The incidence, patient demographic characteristics (i.e. age and sex), and injury characteristics (injury location, disposition, injury diagnosis) were collected from the entries meeting our search criteria. This study qualifies as non-human subject research and was exempt from institutional review board approval because the data is derived from a publicly available database offered by the United States Consumer Product Commission.
The patients were grouped into clinically relevant age groups, including toddlers and young children (0–4 years old), children (5–9 years old), adolescents (10–14 years old), young adults (15–19 years old), adults (20–39 years old), middle-aged adults (40–64 years old), and senior adults (65+ years old). To control for variations within a given year, annual data was broken into two groups for trend analysis; data from 2013 to 2015 comprised the first group, while 2016–2018 comprised the second group within the study.
We then conducted a statistical analysis using SPSS Version 23 (IBM, Armonk, NJ, USA). Individual cases were aggregated on a yearly basis to generate annual rates, with national extrapolation at the national level based on NEISS provided risk weights. For descriptive analysis, total number and percentage were reported for categorical variables, mean and standard deviation for continuous parametric variables, and median and interquartile range [IQR] for continuous non-parametric variables. For descriptive analysis, age was examined in different age grouped cohorts.
Our primary outcome of interest was to identify factors that lead to hospital admission following e-scooter related injuries. We stratified patients on hospital admission status and performed a univariate analysis to compare all baseline variables. To identify independent factors associated with our primary outcome of interest, we constructed a multivariate logistic regression model. Our dependent variable was the presence of hospital admission, and as independent covariates we included all variables that had P value < 0.1 on univariate analysis. For analysis, age was examined as a continuous variable in this model. We utilized conditional backward selection to determine independent associations. To confirm the validity of our model, we performed appropriate regression diagnostics, including calculating the Hosmer-Lemeshow goodness-of-fit test, testing for outliers, and using classification tables to compare the predicted vs. actual outcomes. On univariate analysis, to compare both patient cohorts, we used the χ2 test and the Fisher’s exact test for categorical variables, the Mann-Whitney U test for nonparametric continuous variables, and the independent-samples t-test for parametric continuous variables.