Introduction

Individuals with spinal cord injury (SCI) experience a disproportionately high rate of health care utilization, including hospitalization and emergency department (ED) visits [1,2,3,4,5,6]. In the past decade, there has been a considerable body of literature dedicated to better understanding healthcare utilization among those with acute [5, 7,8,9,10,11] and chronic SCI [4, 6, 12,13,14,15,16,17,18,19,20], however, a notable gap remains regarding the understanding of ED utilization and subsequent hospitalizations, particularly among adults with chronic SCI living in the United States. Specifically, while there is data on the prevalence, causes of (i.e., diagnoses), and reasons for ED utilization, there is limited population-based research on the predictors of ED utilization, specifically modifiable, behavioral predictors.

In the United States, data from the CDC suggests a general trend of increasing ED utilization since the 1990s [21, 22]; in 2018, 21% of the US adult population experienced an ED visit in the past year, 8% had two or more visits, and 12% resulted in a hospital admission [21, 22]. Adults with SCI experience much higher rates of ED utilization and are at greater risk of both ED visits and ED-related hospitalization. The reported rates of ED utilization in the past 12 months among adults with chronic SCI range from 37 to 48% [6, 12, 15, 17, 20], with a considerable portion of those having more than one ED visit per year [6, 12, 20]. For those with SCI who experience an ED visit, a large percentage are subsequently admitted to the hospital [12, 15, 17]. Studies have found that nearly 20% of adults with SCI report at least one ED-related hospitalization, but specifically among those who report an ED visit, roughly 50% report an ED hospitalization [12, 17]. In a comparison to a nationally representative sample, one study found that after controlling for sex, age, and race/ethnicity, individuals with SCI were 151% more likely to visit the ED and 249% more likely to have had an ED hospitalization. When looking only at those with an ED visit, the SCI sample was 118% more likely to be hospitalized [12].

A number of studies have begun to examine the causes or reasons for ED utilization after SCI. The primary causes of ED visits include acute conditions such as genitourinary (e.g., urinary tract infection), musculoskeletal (e.g., pain), and respiratory conditions (e.g., pneumonia), other infections and skin breakdown, trauma and accidents [5, 10, 23,24,25,26]. Using ICD-9 codes, Sikka et al. [10] found that essential hypertension and a behavioral factor, non-dependent abuse of drugs, primarily tobacco use disorder, were the main diagnoses associated with ED visits.

According to one Canadian study [24], while many visits may be classified as high acuity (50%), many are classified as less or non-urgent (33%), or as potentially preventable (17%), suggesting a need to better understand the drivers of ED utilization. Recently, a study examined the underlying reasons for ED utilization among adults within the SCI Model System, compared to a nationally representative sample, based on data from the 2017 National Health Interview Survey [27]. Individuals with SCI were statistically significantly more likely report reasons including: “The emergency room is your closest provider” (86% vs. 37%), “The problem was too serious for the doctor’s office or clinic” (85% vs. 54%), “You get most of your care at the emergency room” (82% vs. 14%), “Your health provider advised you to go” (52% vs 22%), and “You arrived by ambulance or other emergency vehicle” (50% vs 20%) [12]. On the other hand, individuals in the general population were more likely to report that they had no other place to go (41%) compared to the SCI sample (27%). There were no significant differences between the two samples for the following reasons: “Your doctor’s office or clinic was not open”, and “Only a hospital could help you”. This study not only found a higher prevalence of ED visits and related hospitalizations for people with SCI compared with the general population, but a distinctly different pattern of reasons for ED utilization, highlighting three key themes related to severity, frequency and proximity, and availability of alternative care. These findings call attention to the need to better address factors related to ED utilization.

Only a few studies have assessed predictors of ED utilization and subsequent hospitalization, focusing mainly on sociodemographic and injury-related characteristics. Race-ethnicity, injury severity and functional independence at initial rehabilitation discharge, education, income, and rurality are associated with higher ED visits [2, 15, 17, 24, 28]. Additionally, one study found that having more ED visits during the previous year was predictive of future ED visits [25]. Specifically, studies have found that non-Hispanic White individuals, and those with more education and higher income are less likely to report an ED visit [15, 17]. Individuals with greater injury severity, high cervical level injuries, and low functional scores are more likely to report an ED visit [2, 17]. Those who live in more rural locations are also more likely to utilize the ED [2, 28].

Considering the high rates of ED utilization and subsequent hospitalization compared to the general population, it is important to better understand the predictors of ED utilization and subsequent hospitalizations. In particular, there are a limited number of predictor variables assessed in previous studies related to ED utilization. Our purpose was to identify the demographic, injury, and behavioral predictors of ED visits and ED hospitalizations among people with SCI.

Methods

Participant

The current study was part of a prospective cohort study designed to investigate secondary health conditions after SCI. The original study data collection began in 1997–1998, with subsequent assessments in 2007–2009, and 2012–2016. We used measures collected during the third time period (2012–2016) for this cross-sectional study because it had the largest study sample, including population-based cohorts. After obtaining institutional review board approval, we mailed a self-report assessment (SRA) to our prospective participants during the 2012–2016 study period (the assessment questionnaire is available upon request). A total of 4670 participants completed the SRA (60.4% response rate). There were 2516 participants identified from a rehabilitation specialty hospital in the Southeastern USA (rehabilitation/specialty cohort) based on their medical diagnosis of patient record, 1018 identified from a Southeastern state-based surveillance system, and 1136 identified from a state-based surveillance system and the Midwestern USA. Both state-based surveillance systems used population-based registry of traumatic SCI, identifying all civilians with SCI from non-federal hospitals in both states by using International Classification of Diseases-9th Revision, Clinical Modification codes of 806 and 952.

All participants met the following criteria: at least 18 years of age and 1-year post traumatic SCI, not reporting fully recovered from their neurologic impairment. Participants with missing values (n = 522, 11%) and those who reported full recovery (n = 91, 2%), were excluded from the analyses; our final study sample was 4057.

Procedures

Four to six weeks before mailing the SRA, we sent an introductory letter describing the study and alerting prospective participants of the forthcoming SRA materials. Non-respondents to the first mailing received follow-up phone calls and a second mailing. We offered $50 remuneration to those who completed the SRA.

Measures

We measured the number of ED visit and ED hospitalization by asking the participants, “In the past year (12 months), how many total times (if any) did you go to the emergency room because of illness or injury?” and “In the past year (12 months), how many times have you been hospitalized after going to the emergency room (i.e., you went to the emergency room because of illness or injury and were directly admitted to the hospital)?” We then dichotomized the two variables (yes/no) as the primary outcomes.

Demographic and injury characteristics included chronologic age, years post injury, sex (male/female), race/ethnicity (Hispanic, non-Hispanic Black, and non-Hispanic White and others), injury level (cervical level, thoracic level, and other level), ambulatory status (non-ambulatory, ambulatory with assistance from others, and ambulatory without assistance), and annual household income (<$25,000, $25,000–$75,000, and ≥$75,000). We also used the three participant pools (Midwestern surveillance cohort, Southeastern surveillance cohort, and rehabilitation/specialty cohort) as a covariate.

We measured eight behavioral indicators that presented a combination of risk and protective behaviors or behavioral proxies, including: 1) self-perceived body weight (average weight, underweight, and overweight); 2) self-reported healthy diet (poor/fair diet vs. good/excellent diet); 3) average number of alcoholic drinks per occasion (question adapted from Behavioral Risk Factor Surveillance System [29]; 4) use of a non-medical substance (yes/no) in past three months for which the participant did not have a prescription, including cannabis, cocaine, amphetamine type stimulants, inhalants, sedatives or sleeping pills, hallucinogens, and opioids (question adapted from Alcohol, Smoking, and Substance Involvement Screening Test [30]; 5) the frequency of prescription medication use for treating pain, spasticity, sleep problems, and depression (never, rarely, weekly, and daily) (question adapted from previous study) [31]; 6) prescription medication misuse, assessed as the use of one of the aforementioned prescription medications to treat symptoms other than for which they were intended (yes vs. no); 7) smoking status (never smoking regularly, ever smoking regularly but not smoking now, and smoking regularly now) (question adapted from Behavioral Risk Factor Surveillance System [29]; and 8) the frequency of planned exercise (at least once a week, 2–3 times per month, once a month or less).

Analyses

All analyses were conducted using SAS 9.4. Then descriptive statistics of demographic and injury characteristics were generated for the three study cohorts (Midwestern surveillance cohort, Southeastern surveillance cohort, and rehabilitation/specialty cohort). We compared the differences among the three cohorts by using Pearson’s Chi-square test for categorical variables and using the Kruskal–Wallis test for two non-normally distributed continuous variables (age, and years post injury).

We had two dichotomous outcome variables: ED visit and ED hospitalization, and we developed one multiple logistic regression model for each of them. Both logistic regression models included the same set of eight behavioral variables, and eight covariables: study cohort, chronologic age, years post injury, sex, race/ethnicity, injury level, ambulatory status, annual household income. The multiple logistic regression models generated adjusted odds ratios (OR) for each of the variables.

Since our sample was from three different research cohorts, we ran an unconditional means models to get the inter-correlation coefficient (ICC). It was 0.02, which means 2% variance is at the second level. Thus, we did not develop the multilevel model for the small variance. We also conducted sensitivity analyses to compare the demographic and injury characteristics between our study sample and the 522 participants who had missing values, and then used multiple imputations for the two logistic regression models to see if there was any significant changes to the results.

Result

Descriptive

Among the 4057 participants, 1664 participants (41%) reported having at least one ED visit and 858 participants (21%) reported having at least ED hospitalization in the previous 12 months. There were some significant differences in the demographic and injury characteristics among three study cohorts (Table 1). The rehabilitation/specialty cohort had a lower percentage of individuals with cervical injury level (19% compared to 25% and 27%) and higher percentage of non-ambulatory participants (75% compared to 41% and 45%). The rehabilitation/specialty cohort were three years younger than the Midwestern surveillance cohort, but with three years more post injury than the Midwestern surveillance cohort and seven years more post injury than the Southeastern surveillance cohort. The Southeastern surveillance cohort had the highest percentage of non-Hispanic Black participants (33%) and highest percentage of low household income participants (61%).

Table 1 Demographic and injury characteristics of 3 participant cohorts.
Full size table

ED visits

The multiple logistic regression model indicated that both the Midwestern and Southeastern surveillance cohorts were more likely to visit the ED [OR = 1.36 (CI: 1.13–1.62), 1.58 (CI: 1.13–1.62) respectively] compared to the rehabilitation/specialty cohort (Table 2). We found participants who could walk independently [OR = 0.63 (CI: 0.53–0.74)] and who had higher household income [OR = 0.77 (CI: 0.66–0.91), 0.71 (CI: 0.59–0.87) respectively] were less likely to visit ED, while non-Hispanic Black participants [OR = 1.58 (CI: 1.32–1.89)] were more likely to visit ED. The frequency of using prescription medication to treat sleep problems, pain, and depression were associated with greater odds of at least one ED visit [OR = 1.08 (CI: 1.01–1.15), 1.23 (CI: 1.16–1.30), and 1.15 (CI: 1.08–1.22) respectively]. Prescription medication misuse, regular smoking currently, and less planned exercise were associated with greater odds of ED visit [OR = 1.18 (CI: 1.01–1.37), 1.20 (CI: 1.00–1.44), and 1.18 (CI: 1.02–1.37) respectively]. Several behavioral factors were not significantly related to ED visits, including self-perceived body weight, healthy diet, average drinks number, and non-medical substance usage.

Table 2 Logistic regression model for ED visit.
Full size table

ED hospitalizations

We also found Midwestern and Southeastern surveillance cohorts were more likely to have at least one ED-related hospitalization [OR = 1.25 (CI: 1.00–1.55), 1.71 (CI: 1.38–2.11) respectively] than the rehabilitation/specialty cohort (Table 3). Being a male [OR = 1.22 (CI: 1.01–1.47)], older [OR = 1.01 (CI: 1.01–1.02)], and non-Hispanic Black [OR = 1.39 (CI: 1.13–1.71)] was associated with greater odds of ED hospitalization. Those with fewer years post injury [OR = 0.99 (CI: 0.98–1.00)], who could walk independently [OR = 0.46 (CI: 0.37–0.56)], and whose income was between $25K and $75K [OR = 0.79 (CI: 0.65–0.96)] were less likely to have ED hospitalization. The frequency of using prescription medication to treat sleep problems, pain, and depression were also associated with greater odds of ED hospitalization [OR = 1.12 (CI: 1.04–1.21), 1.15 (CI: 1.08–1.24), and 1.16 (CI: 1.09–1.24) respectively]. Prescription medication misuse [OR = 1.27 (CI: 1.07–1.52)] was associated with greater odds of ED hospitalization. We did not find a significant association between ED hospitalization and self-perceived body weight, healthy diet, smoking, average drinks number, and non-medical substance usage.

Table 3 Logistic regression model for ED hospitalization.
Full size table

The sensitivity analyses found participants who had missing values were more likely to be older, non-Hispanic Black, and with lower household income. We applied multiple imputations to use the all the participants for both logistic models, and we did not find significant changes to our study results (both sensitivity analyses and multiple imputations results are available upon request).

Discussion

The unique contributions of the current study include the focus on both ED visits and ED hospitalizations, the comparison of outcomes among rehabilitation/specialty and population-based cohorts, and the identification of multiple sets of covariates for each ED outcome. We assessed a much wider array of behavioral factors than has been used in previous research.

While the current study supports earlier findings, there are some distinct differences to note. In a previous analysis of ED visits and subsequent hospitalizations [17], we found several differences in the predictors of ED visits (n = 1579), conditional ED-related hospitalizations (i.e., hospitalizations among those who had at least one ED visit, n = 581), and unconditional ED hospitalizations (i.e., hospitalizations among all participants, n = 1579). For example, race-ethnicity was not related to unconditional hospitalizations, but significant relationships were found between race-ethnicity and both ED visits and conditional ED hospitalizations. Non-Hispanic White participants were 9% less likely to report an ED visit. However, among those who had at least one ED visit, non-Hispanic Whites were 10% more likely to have an ED hospitalization. The findings suggested that other participants, primarily non-Hispanic Blacks, were more likely to utilize the ED for treatment, potentially for less severe conditions that did not necessitate subsequent hospitalization. In the current analysis, we found non-Hispanic Black participants were 58% more likely to report an ED visit, consistent with previous research, however, these individuals were also found to report greater odds of ED hospitalization [OR = 1.39 (CI: 1.13–1.71)), a change from the earlier findings. Additionally, while the previous research found age unrelated to having at least one ED visit or unconditional ED hospitalization, the oldest participants had an increased risk of conditional ED hospitalization, suggesting more adverse outcomes among older adults with SCI. In the present analyses, age was not associated with ED visits, but was associated with slightly increased risk of ED hospitalization. The current analyses support the previous findings related to injury level. As for socioeconomic factors, the earlier study [17] found greater income was related to a decreased risk of ED visit, but no difference in ED hospitalization, contrary to our current findings.

Clinical and public health implications

The pattern of significant findings may serve to guide rehabilitation and public health professionals. First, there clearly is a socioeconomic and racial ethnic component to both ED visits and hospitalizations, with those having higher income having better outcomes and those who are non-Hispanic Black experiencing poorer outcomes. The extent to which this relates to access to care and use of the ED versus other forms of healthcare is not clear from the current findings. Second, as individuals age, both based on chronologic age or years postinjury, there is a greater risk of hospitalization following an ED visit. This may be an indication of more acute conditions and greater vulnerability to long-term adverse outcomes. Third, being treated for secondary health conditions with prescription medication is associated with a substantially higher risk of ED visits and hospitalizations. Although we also cannot determine whether it is the medication per se, versus the underlying condition, misuse of the medications, as defined by use for conditions for which they were not prescribed, was significantly related to both ED visits and ED hospitalizations. All these factors should be considered when developing treatment plans and, in particular, the presence of extensive use of prescription medications to treat secondary conditions should be a red flag for future adverse outcomes. Another implication is that lifestyle factors (smoking and exercising) associated with ED outcomes indicated the significance of working to minimize high-risk behaviors and promote healthier lifestyle. We would like to recommend health care professionals assess health behaviors at patients’ initial discharge and at the following visits in order to target those individuals at higher risk of ED visits and hospitalizations for early interventions. The greater likelihood of both ED visits and hospitalizations among those identified through population-based state surveillance underscores the importance of public health to the well-being of people with SCI. Specifically, population-based systems capture those people who fall through the cracks of the traditional rehabilitation specialty system. Therefore, the poorer outcomes suggest that outcomes identified within specialty care settings may not be representative of those who are not privileged to receive specialty services. The findings also present an opportunity for the development of tools to identify people who are at high risk for ED visits and hospitalizations living in the community.

Methodologic considerations

There are both strengths and limitations. The sample was relatively large for statistical modeling of a diverse set of characteristics. This included multiple behavioral characteristics that are important to understand in relation to all health outcomes. Second, utilizing both rehabilitation/specialty and population-based cohorts for direct comparisons helps better understand the study findings. Population-based cohorts capture individuals who fall through the cracks of the traditional healthcare delivery system. Third, the combination of both ED visits and hospitalizations allowed for direct comparisons of the behavioral predictors of each outcome.

On the other hand, all data were self-report. Therefore, we do not know the extent to which the self-report matched the actual number of ED visits and hospitalizations. Previous research has indicated that self-reported ED visits correspond closely with administrative billing data, whereas general reports of hospitalizations do not correspond as well [20]. Because hospitalizations were ascertained directly in relation to the ED visits, this should not be a major issue in the current analysis. Self-report may have unknown biases in reporting, such as measuring self-perceived body weight, rather than actual weight. Additionally, due to the nature of the cross-sectional design, causality cannot be inferred. Third, even though participation was strong, selective bias of non-response (40%) is always of concern. We did not have sufficient sociodemographic or injury data sufficient to compare responders and non-responders. Lastly, the findings are restricted to two geographic regions of the USA. The data was not intended to be normative, but rather to be used for identification of risk and protective factors, so this limitation would have minimal effect on the current study.

Future research

Additional research is needed to identify longitudinal relationships between demographic, injury, socioeconomic, and behavioral factors with future patterns of ED visits and hospitalizations. There is also a need to establish the causes and costs of ED visits to help determine the cost effectiveness of potential interventions. Whereas linking specific risk factors to ED outcomes will allow for identification of high-risk behaviors and factors that may become the focus of intervention, identifying the cost of ED visit will allow for evaluation of the relative costs associated with multiple health risk behaviors.

Conclusion

The prevalence of both ED visits and ED-related hospitalizations remained high among people with chronic SCI. Participants identified from the population-based surveillance had greater probability of ED visits and hospitalization than those identified from rehabilitation specialty hospitals. The study also indicated the need to prevent medication misuse, to quit smoking, and to have more planned exercises.