Rural-residing children in the United States experience higher suicide mortality than urban-residing children but are underrepresented in research. We examined changes in emergency department (ED) utilization and subsequent hospitalization for suicide or self-harm in a rural hospital after the coronavirus disease 2019 pandemic onset.
This retrospective cohort study involved children aged 5 to 17 years visiting a rural, Northeastern hospital from January 1, 2017 to May 31, 2021. We used autoregressive integrated moving average modeling, an interrupted time series analysis, to examine monthly changes after the pandemic onset (March 2020) in ED visits with a primary mental health diagnosis, number of mental health visits with a suicide or self-harm diagnosis, proportion of patients with suicide or self-harm admitted to hospital, and length of stay for suicide or self-harm.
Prepandemic, there was an average of 20.6 visits per month for mental health conditions, with 23.3 visits per month postpandemic (P = .16). Monthly visits for suicide or self-harm were greater in the postpandemic (15.6 visits per month) versus prepandemic months (11.4 visits per month, P < .01). In autoregressive integrated moving average modeling, pandemic onset related to an additional 0.83 (95% confidence interval: 0.31 to 1.36) primary mental health visits with suicide or self-harm diagnoses per month. Of these visits, there was an immediate, absolute increase of 39.6% (95% confidence interval: 26.0% to 53.1%) in the proportion resulting in admission; admission rates declined in subsequent months. Pandemic onset was not associated with significant changes in the number of visits for mental health conditions or length of stay.
Pediatric ED visits for suicide or self-harm increased at a significant rate during the coronavirus disease 2019 pandemic and a greater proportion resulted in hospitalization, highlighting the acute mental health needs of rural-residing children.
Rural-residing children in the United States experience higher suicide mortality than their urban-residing counterparts.1,2 During the first year of the coronavirus disease 2019 (COVID-19) pandemic, the global prevalence of depressive and anxiety symptoms in youth was higher than prepandemic estimates, and the proportion of emergency department (ED) visits for pediatric mental health conditions increased.3,4 Research about the health system impacts of the COVID-19 pandemic has not focused on suicide and self-harm in rural-residing children, who may have experienced greater vulnerability because of existing disparities in access to mental health resources.5 Therefore, we aimed to examine changes in ED utilization and subsequent hospitalization for suicide or self-harm in a rural hospital after the onset of the pandemic.
Methods
We analyzed medical records from an academic medical center with a general (nonpediatric) ED and no pediatric psychiatric beds located in a rural region of the Northeast United States. Data spanned from January 1, 2017 to May 31, 2021 and included youth aged 5 to 17 years. We used the Child and Adolescent Mental Health Disorders Classification System International Classification of Diseases, Tenth Revision codes to identify encounters with a principal diagnosis of any mental health condition.6 Among those, we classified encounters with a principal or secondary diagnosis of suicide or self-injury, or accidental or undetermined poisoning as suicide or self-harm. We defined the onset of the COVID-19 pandemic as March 2020, when it was declared a national emergency.7 Our primary outcomes were monthly changes in the (1) ED encounters with a primary diagnosis of a mental health condition, (2) mental health visits with a suicide or self-harm diagnosis, (3) proportion of these suicide or self-harm encounters admitted to hospital, and (4) length of stay (LOS) for suicide or self-harm encounters (reported as the geometric mean). As supplemental outcomes, we assessed the total monthly number of ED visits in children 5 to 17 years of age and the proportion of these with a principal diagnosis of a mental health condition.
We compared cohort characteristics pre- and postpandemic using χ2 tests for categorical variables and t tests or Wilcoxon rank tests for continuous variables. Race and ethnicity were self-reported per the hospital’s standard operating procedures. We used autoregressive integrated moving average (ARIMA) modeling8 to compare observed and counterfactual values (ie, expected values if no pandemic occurred) of each outcome relative to the pandemic onset. ARIMA is a type of interrupted time series analysis that can accommodate autocorrelation and seasonal cycles in the data when forecasting trends after some event. Models considered the following changes postpandemic onset for each outcome: an immediate, temporary change that reverted back to the expected value (ie, a pulse transfer function), an immediate, sustained change that did not revert back to the expected value (ie, a step transfer function), and a differential rate of change in the outcome (ie, a slope transfer function).
For each outcome, we visualized the temporal patterns in the data and determined the best-fitting ARIMA model via an automated algorithm.8 We confirmed that the model residuals were uncorrelated using the Ljung-Box test (all P values >.05) and plotted the observed and predicted trends in the outcome after the pandemic onset.8 All analyses were completed by using R (version 1.4.1106); P < .05 was the threshold for statistical significance.
This study was approved with a waiver of informed consent by the hospital’s institutional review board.
Results
During the study period, January 2017 to May 2021, there were 17 199 ED visits by children 5 to 17 years of age, of which 1131 (6.6%) were for primary mental health conditions: 781 (5.9%) prepandemic versus 350 (9.0%) postpandemic (P <.001). Prepandemic, there were an average of 20.6 (standard deviation [SD] = 5.54) mental health visits per month, and 23.3 postpandemic (SD = 6.51, P = .16). Of these, there were an average of 11.4 (SD = 5.04) visits per month with suicide or self-harm prepandemic versus 15.6 (SD = 4.78, P = .008) postpandemic, representing an increase from 55.3% of mental health visits to 66.9% (P < .01). Sociodemographic characteristics of children experiencing suicide or self-harm were similar during the pre- and postpandemic periods, but the proportion with any cooccurring diagnosis of depression (67.6% vs 79.1%, P = .002) and anxiety (30.8% vs 40.6%, P = .014) was significantly greater postpandemic (Tables 1 and 2).
Volume of Pediatric ED Visits at a Rural Hospital Between January 2017 to February 2020 and March 2020 to May 2021
. | Prepandemic (Jan 2017–Feb 2020) . | Postpandemic (Mar 2020–May 2021) . | Pa . |
---|---|---|---|
Total ED visits, 5–17 y old | 13 292 | 3907 | — |
Mean monthly ED visits (SD) | 333 (74.4) | 302 (70.9) | .16 |
Primary mental health visits | 781 | 350 | — |
Mean monthly mental health visits (SD) | 20.6 (5.54) | 23.3 (6.51) | .16 |
Suicide or self-harm visitsb | 432 | 234 | — |
Mean monthly suicide/self-harm visits (SD) | 11.4 (5.04) | 15.6 (4.78) | .008 |
. | Prepandemic (Jan 2017–Feb 2020) . | Postpandemic (Mar 2020–May 2021) . | Pa . |
---|---|---|---|
Total ED visits, 5–17 y old | 13 292 | 3907 | — |
Mean monthly ED visits (SD) | 333 (74.4) | 302 (70.9) | .16 |
Primary mental health visits | 781 | 350 | — |
Mean monthly mental health visits (SD) | 20.6 (5.54) | 23.3 (6.51) | .16 |
Suicide or self-harm visitsb | 432 | 234 | — |
Mean monthly suicide/self-harm visits (SD) | 11.4 (5.04) | 15.6 (4.78) | .008 |
IQR, interquartile range; —, not applicable.
Derived via t tests or Wilcoxon rank tests for means and medians of continuous variables and v2 tests for categorical variables. P values were not derived for absolute counts nor non-mutually exclusive characteristics.
Encounters with a principal diagnosis for any mental health condition and a principal or secondary diagnosis for suicide or self-harm.
Characteristics of Children and Adolescents With Suicidality or Self-Harm Cared for in the ED of a Rural Hospital Between January 2017 to February 2020 and March 2020 to May 2021
. | Prepandemic, n = 432 . | Postpandemic, n = 234 . | Pa . |
---|---|---|---|
Age, y, median (IQR) | 15.0 (13.0–16.0) | 14.0 (13.0–16.0) | .01 |
Sex, female, n (%) | 277 (64.1) | 154 (65.8) | .73 |
Co-occurring conditions, n (%) | |||
Depressive disorders | 292 (67.6) | 185 (79.1) | .002 |
Anxiety disorders | 133 (30.8) | 95 (40.6) | .01 |
Trauma and stressor-related disorders | 49 (11.3) | 40 (17.1) | .05 |
Substance-related disorders | 33 (7.6) | 19 (8.1) | .95 |
Other conditions | 100 (23.1) | 62 (26.5) | .39 |
Race/ethnicity, n (%)b,c | 0.71 | ||
White, non-Hispanic | 372 (86.1) | 206 (88.0) | |
Hispanic or Latino | 31 (7.2) | 13 (5.6) | |
Other | 19 (4.4) | 10 (4.3) | |
Primary payor, n (%)b | 0.86 | ||
Commercial | 227 (52.5) | 126 (53.8) | |
Government | 201 (46.5) | 107 (45.7) | |
Rurality of home address, n (%)b,d | 0.19 | ||
Urban | 26 (6.0) | 13 (5.6) | |
Large rural | 197 (45.6) | 125 (53.4) | |
Small town/rural | 205 (47.5) | 96 (41.0) | |
Transfer discharge disposition, n (%)e | 183 (42.4) | 78 (33.3) | .03 |
Length of stay, h, geometric mean (SD)f | 19.0 (54.8) | 21.7 (97.4) | .09 |
Admitted from ED, n (%) | 46 (10.6) | 85 (36.3) | <.001 |
. | Prepandemic, n = 432 . | Postpandemic, n = 234 . | Pa . |
---|---|---|---|
Age, y, median (IQR) | 15.0 (13.0–16.0) | 14.0 (13.0–16.0) | .01 |
Sex, female, n (%) | 277 (64.1) | 154 (65.8) | .73 |
Co-occurring conditions, n (%) | |||
Depressive disorders | 292 (67.6) | 185 (79.1) | .002 |
Anxiety disorders | 133 (30.8) | 95 (40.6) | .01 |
Trauma and stressor-related disorders | 49 (11.3) | 40 (17.1) | .05 |
Substance-related disorders | 33 (7.6) | 19 (8.1) | .95 |
Other conditions | 100 (23.1) | 62 (26.5) | .39 |
Race/ethnicity, n (%)b,c | 0.71 | ||
White, non-Hispanic | 372 (86.1) | 206 (88.0) | |
Hispanic or Latino | 31 (7.2) | 13 (5.6) | |
Other | 19 (4.4) | 10 (4.3) | |
Primary payor, n (%)b | 0.86 | ||
Commercial | 227 (52.5) | 126 (53.8) | |
Government | 201 (46.5) | 107 (45.7) | |
Rurality of home address, n (%)b,d | 0.19 | ||
Urban | 26 (6.0) | 13 (5.6) | |
Large rural | 197 (45.6) | 125 (53.4) | |
Small town/rural | 205 (47.5) | 96 (41.0) | |
Transfer discharge disposition, n (%)e | 183 (42.4) | 78 (33.3) | .03 |
Length of stay, h, geometric mean (SD)f | 19.0 (54.8) | 21.7 (97.4) | .09 |
Admitted from ED, n (%) | 46 (10.6) | 85 (36.3) | <.001 |
IQR, interquartile range.
a Derived via t tests or Wilcoxon rank tests for means and medians of continuous variables and χ2 tests for categorical variables. P values were not derived for absolute counts nor non-mutually exclusive characteristics.
b <5% of data missing.
c Those identifying as Asian, American Indian/Alaska Native, Black non-Hispanic, and Native Hawaiian/Other Pacific Islander combined to protect patient identity because of small cell sizes.
d Determined based on Rural-Urban Commuting Area of residence.
e Discharge disposition of transfer defined as: discharge to Psych Hospital/Distinct Part of Hospital, Rehab Center in a Facility, Other Short Term General Hospital, Long Term Care Hospital, Intermediate Care Facility, Another HCF/Not Defined.
f Defined as hours from arrival to hospital discharge.
Changes in the Characteristics of ED Visits and Hospital Admissions for Mental Health Diagnoses Associated With the Onset of the COVID-19 Pandemic (March 2020): Results of Autoregressive Integrated Moving Average Modeling (ARIMA)
. | Final Model Specification . | Model Intercept (95% CI) . | Transfer Functions for Postpandemic Onset (95% CI) . | ||
---|---|---|---|---|---|
Primary Outcomes | Step | Pulse | Slope | ||
Number of monthly mental health encounters | ARIMA (0,0,0) (1,0,0)a | 20.7 (18.5 to 23.0) | −1.14 (−6.34 to 4.06) | Not included | 0.67 (−0.02 to 1.37) |
Number of monthly mental health encounters for suicide or self-harm | ARIMA (0,0,0) (0,0,0) | 11.4 (9.94 to 12.8) | −2.43 (−7.41 to 2.56) | Not included | 0.83 (0.31 to 1.36)* |
Proportion of monthly mental health encounters for suicide or self-harm resulting in admission | ARIMA (0,0,0) (0,0,0) | 11.1 (7.23 to 15.0) | 39.6 (26.0 to 53.1)* | Not included | −1.59 (−3.02 to −0.17)* |
Geometric mean length of stay for suicide or self-harm | ARIMA (0,0,0) (0,0,1)b | 21.5 (19.0 to 24.0) | 0.88 (−6.78 to 8.54) | Not included | 0.11 (−1.03 to 1.24) |
Supplemental outcomes | |||||
Number of monthly total pediatric ED visits among 5- to 17-y-olds | ARIMA (0,1,1) (0,0,0)c | Not included | Not included | −119 (−185 to −53.5)* | 4.51 (−11.6 to 20.6) |
Proportion of monthly pediatric encounters for a mental health condition | ARIMA (0,1,1) (1,0,0)d | Not included | 4.69 (1.85 to 7.53)* | Not included | −0.06 (−0.46 to 0.34) |
. | Final Model Specification . | Model Intercept (95% CI) . | Transfer Functions for Postpandemic Onset (95% CI) . | ||
---|---|---|---|---|---|
Primary Outcomes | Step | Pulse | Slope | ||
Number of monthly mental health encounters | ARIMA (0,0,0) (1,0,0)a | 20.7 (18.5 to 23.0) | −1.14 (−6.34 to 4.06) | Not included | 0.67 (−0.02 to 1.37) |
Number of monthly mental health encounters for suicide or self-harm | ARIMA (0,0,0) (0,0,0) | 11.4 (9.94 to 12.8) | −2.43 (−7.41 to 2.56) | Not included | 0.83 (0.31 to 1.36)* |
Proportion of monthly mental health encounters for suicide or self-harm resulting in admission | ARIMA (0,0,0) (0,0,0) | 11.1 (7.23 to 15.0) | 39.6 (26.0 to 53.1)* | Not included | −1.59 (−3.02 to −0.17)* |
Geometric mean length of stay for suicide or self-harm | ARIMA (0,0,0) (0,0,1)b | 21.5 (19.0 to 24.0) | 0.88 (−6.78 to 8.54) | Not included | 0.11 (−1.03 to 1.24) |
Supplemental outcomes | |||||
Number of monthly total pediatric ED visits among 5- to 17-y-olds | ARIMA (0,1,1) (0,0,0)c | Not included | Not included | −119 (−185 to −53.5)* | 4.51 (−11.6 to 20.6) |
Proportion of monthly pediatric encounters for a mental health condition | ARIMA (0,1,1) (1,0,0)d | Not included | 4.69 (1.85 to 7.53)* | Not included | −0.06 (−0.46 to 0.34) |
Each ARIMA model included visits measured at the per month level as the time interval, measured from January 2017 to May 2021. Statistically significant findings (P < .05) are indicated by an asterisk. Step transfer functions indicate a sustained change in the outcome value postpandemic onset. Pulse transfer functions indicate a temporary change in the outcome value postpandemic onset. Slope transfer functions are the rate of change in the outcome postpandemic onset. All transfer functions were modeled in comparison with the expected trends based on the historical data.
a Model included an autoregressive, 12-month seasonal component with a lag of 1 (coefficient = 0.41; 95% CI 0.14 to 0.67).
b Model included a moving average, 12-month seasonal component with a lag of 1 for the error terms (coefficient = −0.28; 95% CI −0.64 to 0.08).
c Model included a nonseasonal differencing component of lag 1 and a moving average with lag of 1 for the error terms (coefficient = −0.51; 95% CI −0.76 to −0.26). This model contains a pulse function for the time period of March 2020 to June 2020 and not a step function as the data were more indicative of a temporary change in visits versus a sustained change.
d Model included a nonseasonal differencing component of lag 1, a moving average with lag of 1 for the error terms (coefficient = −0.65; 95% CI: −0.94 to −0.36), and an autoregressive, 12-month seasonal component with a lag of 1 (coefficient = 0.48; 95% CI: 0.17 to 0.78).
In the ARIMA analysis, although the onset of the COVID-19 pandemic was associated with a temporary, dramatic decrease in the monthly number of pediatric ED visits for any cause (pulse transfer function covering March 2020 to June 2020 = −119 visits/month; 95% confidence interval [CI]: −185 to −53.5; Supplemental Fig 2A, Table 3), there was not a significant change in the number of ED encounters for mental health diagnoses over that same timeframe (step transfer function = −1.14 visits per month; 95% CI: −6.34 to 4.06; Fig 1A, Table 3). Correspondingly, there was a 4.69% (95% CI: 1.85 to 7.53; step transfer function) immediate, absolute increase in the monthly proportion of encounters for mental health conditions (Supplemental Fig 2B, Table 3). After the pandemic onset, there was a statistically significant and positive rate of change in the number of mental health encounters with suicide or self-harm, with an additional 0.83 visits per month (95% CI: 0.31 to 1.36; Fig 1B, Table 3). A greater proportion of these encounters resulted in hospital admission during the pandemic (step transfer function = 39.6% absolute change in proportion admitted; 95% CI: 26.0% to 53.1%; Fig 1C, Table 3); there was a significant decrease in the slope change in the months after March 2020 representing the reduction from the initial peak (−1.59; 95% CI: −3.02 to −0.17). The geometric mean LOS of encounters for suicide or self-harm was not significantly different than predicted (Fig 1D, Table 3).
Observed versus predicted outcomes January 2017 to May 2021. (A) Number of ED visits among 5- to 17-year-olds with a primary mental health diagnosis, (B) number of mental health ED visits for suicide or self-harm, (C) proportion of mental health ED visits with a diagnosis of suicide or self-harm admitted to hospital, (D) geometric mean length of stay for suicide or self-harm.
Observed versus predicted outcomes January 2017 to May 2021. (A) Number of ED visits among 5- to 17-year-olds with a primary mental health diagnosis, (B) number of mental health ED visits for suicide or self-harm, (C) proportion of mental health ED visits with a diagnosis of suicide or self-harm admitted to hospital, (D) geometric mean length of stay for suicide or self-harm.
Discussion
At the onset of the COVID-19 pandemic, we observed a sharp decrease in the number of pediatric ED visits whereas the number of visits for primary mental health conditions was not statistically different than predicted. The number of mental health ED encounters for suicide or self-harm increased at a significant rate and a greater proportion of these visits resulted in hospital admission, compared with predicted trends. These findings may reflect the relative paucity of community-based mental health services in our rural setting, and a corresponding dependence on the ED to address mental health needs.
The United States Surgeon General’s recent advisory on youth mental health during COVID-19 identifies rural-residing youth as a group facing a higher risk of challenges.9 In alignment with this report, our findings suggest the consequences of the COVID-19 pandemic, which created additional stress for rural pediatric populations that have greater access to firearms, higher rates of ED visits for firearm-related self-harm, and less access to community mental health resources than their urban peers.5,10,11
As in our hospital, several studies have revealed that the number of pediatric ED visits and hospital admissions during the COVID-19 pandemic decreased compared with data from previous years.12–14 Consistent with a previously published study at a US freestanding children’s hospital, we also observed a significant increase in the proportion of children and adolescents presenting with mental health concerns.4 However, in both the pre- and postpandemic periods, the overall proportions of ED visits for mental health were greater at our hospital than at the children’s hospital in this previous study. We also found that a greater proportion of patients with a suicide or self-harm diagnosis was admitted from the ED, shedding light on the changing composition of patients who the pediatrics service has cared for during the COVID-19 pandemic.
The literature on COVID-19 and youth mental health contains some heterogeneity in findings. A study in California revealed a greater proportion of all ED visits for suicidal thoughts or behaviors postpandemic onset; however, incidence rates decreased significantly from March 2020 to May 2020.15 Similarly, a cohort study in Ontario, Canada revealed that ED visits for self-harm or overdose decreased significantly during the pandemic.16 In a Northeastern US free-standing children’s hospital, the monthly number of visits for suicide ideation or attempts increased at a positive rate during the pandemic (March 2020 to February 2021), but the slope change was not significantly different from the prepandemic trend.17 As the pandemic has highlighted, national crises impact youth differently depending on their community’s preexisting advantages and challenges.
Limitations of this study include its retrospective, single-center design, which impacts generalizability to other hospitals. However, rural hospitals are typically underrepresented in research, thus motivating the focus of our work. Like all analyses of health systems data, this research assumes accurate coding of mental health diagnoses. In defining LOS, we only had access to hours spent at the presenting hospital, which may not represent a patient’s full time in care because more than one-third of youth experienced an interhospital transfer. Although our data come from a comparatively lower-volume, nonpediatric ED, we used 38 months of data before the onset of the pandemic and robust modeling to capture trends in health system utilization. Based on our findings, we cannot make conclusions about the incidence of suicide or self-harm; however, we are able to offer a lens into changes in patient presentations to the ED and the inpatient unit.
In conclusion, the number of ED encounters for suicide or self-harm increased significantly after COVID-19 onset, and a greater proportion of these patients were admitted to hospital, illustrating the profound impact of mental health conditions in this rural setting. Additional resources and clinician training are warranted as EDs and inpatient pediatrics services care for increasing numbers of youth experiencing mental health crises during and after the COVID-19 pandemic.
FUNDING: No external funding.
CONFLICT OF INTEREST DISCLOSURES: The authors have indicated they have no potential conflicts of interest to disclose.
Ms Arakelyan conceptualized and designed the study, drafted the initial manuscript, conducted analyses, and reviewed and revised the manuscript; Dr Emond conceptualized and designed the study, reviewed the analytical plan, and reviewed and revised the manuscript; Dr Leyenaar conceptualized and designed the study, supervised and facilitated data collection, and reviewed and revised the manuscript; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
Comments