Data and sample
We utilized the 2013–2017 public use data files of the National Health Interview Survey (NHIS), an annual, cross-sectional household survey which is the principal source of information on the health and healthcare access of the civilian, noninstitutionalized US population [8]. For each sampled household, interviews are conducted (mostly face-to-face) with an adult family member who answers questions about the demographic and health status characteristics of each family member. For this study, we linked data on medically attended injury/poisoning episodes that occurred to any family member within the 91-day reference period to their demographic and other health data. Combining all 5 years of NHIS data resulted in a sample of 495,663 individuals aged < 1 to 85+ years (NHIS public use data sets do not provide chronological age of those aged > 85 years). Of these, 104,340 were aged > 60, and of them, 1800 had other injuries (e.g., due to motor vehicle or other mobility means, cutting/piercing, burns, poisoning) than falls, and 1840 (representing 1.16 million individuals over the 5-year study period) had fall injuries. We focused on those with fall injuries to address study questions. In the case of those with more than one fall injury episode during the 91-day period, the most recent episode was used for analysis.
Measures
Healthcare utilization
For each injury episode, questions were asked about whether or not injury care was received through a call to a medical professional, at a doctor’s office or clinic, at an ED, and/or at any place else. Questions were also asked about whether the person was hospitalized and the number of nights hospitalized. Responses to these questions led to identifying the following three groups of fall victims: those without ED visit or hospitalization (reference group), those who had an ED visit only, and those who were hospitalized.
Diagnosed chronic illnesses and sensory and memory problems that cause limitations
Diagnosed chronic illnesses that caused limitations included arthritis, cancer, diabetes, high blood pressure, heart disease, lung cancer, and stroke. We also included vision, hearing, and memory problems that caused limitation as possible correlates of fall injuries.
Other health conditions
For descriptive purposes only, we present number of activities of daily living (ADL) impairments (0–6; feeding, bathing, getting dressed, toileting, transferring to/from bed or chairs, and getting around in the home), difficulty walking without equipment, and limitations caused by depression/anxiety (yes = 1, no = 0 for each). Past-year healthcare use is also presented to describe the sample. All of these may have been affected by fall or other injuries among the injured individuals.
Fall injury site and broken bones or fractures
For each fall (or other) injury episode, respondents were asked to list up to four parts of their body hurt due to the injury and “how” each body part was affected. We collapsed the answers regarding fall injury sites into eight categories (e.g., hip, head, face, lower and upper limbs). Broken bones/fractures were distinguished from all other types of fall injury (e.g., sprains, cuts, scrapes, bruises, burns).
Fall location
Response categories for location were at the injured person’s home or outside (sidewalk, parking lot, sports facility, shopping mall, and so forth). Because most older adults’ falls occurred at home, we categorized the responses into (1) inside the home, (2) at home but outside (e.g., yard, patio), and (3) away from home. Questions were also asked about whether the fall involved the floor/level ground, stairs/steps, bathtub/shower toilet, ladder/scaffolding, sports field/court/rink, and so forth. As these categories are highly correlated with fall location, we present them for descriptive purposes only.
Cause of fall
The response categories were slipping/tripping, loss of balance or dizziness, bumping into an object or another person, being shoved or pushed by another person, jumping or diving, or other. For parsimony, we categorized them into slipping/tripping, loss of balance or dizziness, and other.
Socioeconomic variables
These included age (60–69; 70–79; and 80+ years); gender, race/ethnicity; marital status, living arrangement (alone vs. with someone), education (college degree vs. no college degree), family income to poverty ratio, and health insurance types (Medicare, Medicaid, Veterans Administration and other insurance for military personnel, and/or private health insurance).
Data analysis
All analyses were conducted with Stata/MP 15’s svy function to account for NHIS’ stratified, multistage sampling design. First, to describe the sample, we used χ2 and one-way ANOVA tests to compare socioeconomic, clinical, and healthcare utilization characteristics of the three groups of older fall victims by their healthcare utilization (no ED visit or hospitalization, ED visit only, and hospitalization). We also used χ2 tests to compare injury site, fracture, location, and cause among the three groups. We did not adjust reported p values due to a number of considerations [9]; however, it is important to acknowledge that 5% of tests represent a Type I error. To test H1 (correlates of ED visit only and hospitalization vs. no ED visit or hospitalization), we used multinomial logistic regression analysis. To identify a parsimonious model, we used backward elimination and excluded the following nonsignificant socioeconomic factors: race/ethnicity, marital status, past-year work status, family income to poverty ratio, and health insurance type. Survey years (2013 vs. 2014, 2015, 2016, and 2017) were also excluded from the final model because they were nonsignificant as a covariate of healthcare utilization. Variance inflation factor diagnostics, using a cut-off of 2.50 [10], showed that multicollinearity among the included covariates was not a concern. Results of the multinomial logistic regression model are presented as relative risk ratios (RRR) with 95% confidence intervals (CI). Statistical significance was set at p < .05.