Characteristics of Patients Associated With Restraint Use at a Midwest Children’s Hospital

We conducted a retrospective single-center study of hospital encounters between January 1, 2017 and December 31, 2021 for patients aged 5 to 20 years. Our health care system includes 2 freestanding children’s hospitals with 370 beds (with no associated inpatient psychiatric, behavioral, or adolescent-specific units).

The primary outcome was mechanical restraint during hospitalization, identified by nursing documentation of restraint use in the electronic health record (EHR). As in other pediatric hospitals, a mechanical restraint in our setting involves restricting a child to their bedframe using 4-point polyurethane limb restraints with or without a chest strap.⁵  Manual holds without the use of mechanical restraint were not included in this study given the lack of documentation of use.

We extracted demographic information from the EHR, including patient age and weight at time of hospital admission, sex, hospital payer, and race and ethnicity. Sex- and age-adjusted weight z-scores were computed and included in modeling because of known provider bias toward children with obesity, including ascribing more severe mental health pathology to obese children than those without obesity.^18,19  Payer information was used to classify each patient as having or not having any record of public insurance during the study period.

Race and ethnicity data are entered in our EHR on the basis of patient/guardian responses to questions from a Patient Access Representative. These responses are used to classify each patient according to race (1 variable) and as having or not having Hispanic/Latino ethnicity (a second variable). The coding options for race include Asian, Black or African American, Hispanic, American Indian or Alaskan Native, Native Hawaiian or Pacific Islander, and White, along with declined/refused, multiracial, other, respondent not available, and unknown to respondent.

Race and ethnicity are social constructs subjective to those reporting, and we did not attempt to differentiate between the 2 in our analyses. Instead, we created a single race and ethnicity variable and classified patients primarily using our EHR’s “race” variable (which includes Hispanic as an option, as noted above). For example, patients classified in our EHR as having Black race were also classified as Black on our race and ethnicity variable, regardless of whether they were classified as Hispanic/Latino on our EHR’s ethnicity variable. In cases in which the EHR race variable had missing data (eg, because the respondent declined/refused to provide the information) or was coded “unknown to respondent,” we used the EHR ethnicity variable to identify patients classified as Hispanic/Latino and included these patients in our race and ethnicity variable’s “Hispanic” category.

Clinical patient factors examined as potential correlates of restraint included the presence of various mental health disorders, identified by their diagnosis codes during any visit (including clinic and ED encounters) during our study period. Mental health disorder codes were identified on the basis of the Child and Adolescent Mental Health Disorders Classification System, a classification scheme that aligns billing codes with the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition psychiatric diagnosis groups.^20,21  Given the numbers in our sample, we combined certain conditions and limited our attention to the following groups: anxiety and obsessive compulsive disorders, autism spectrum disorder, bipolar disorders, communication disorders, depressive disorders, developmental, motor, and learning disorders, disruptive, impulse control, and conduct disorders, eating disorders, intellectual disability, schizophrenia, and psychotic disorders, substance-abuse related illnesses, suicidality/self-injury, and trauma-related disorders.

For each hospitalization, we captured if the patient experienced restraint during their immediately preceding hospitalization during the study period. In addition, we extracted the following hospitalization characteristics: hospital length of stay (LOS), admission through our hospital system’s ED (vs transfer or direct admission), mental health disorder codes associated with hospitalization, psychiatry or psychology consultation, admitting hospital service, and season of the year. We included LOS to account for complex, extended hospitalizations (eg, psychiatric boarders who remained in our hospital because of a lack of external placement). Notably, in our data, LOS includes time spent in our ED immediately preceding a hospitalization. Admission through our hospital system’s ED (transfer or direct admission) was included because we believe the source of admission may be a proxy for patient acuity/complexity. Additionally, in our hospital system, patients who are medically stable and are awaiting placement at different facilities (ie, inpatient psychiatry) are not accepted as direct admissions. Psychiatry/psychology consult was included because we thought it again may serve as a proxy for patient mental health acuity/complexity.

Missing values for weight z-score at hospitalization were filled in, when possible, by the patient’s mean weight z-score across hospitalizations during the study period. Then, for purposes of statistical modeling, we used the mice package in R to conduct multiple imputations by chained equations at the patient level to address missing values for mean weight z-score, public insurance, and race and ethnicity category. The resulting 25 complete sets of patient-level data were merged with the observed visit-level data for analysis.

For each complete data set, we used logistic regression to model the odds of restraint at the hospitalization level as a function of patient and hospitalization characteristics. Explanatory variables included patient age, sex, race and ethnicity category, any record of public insurance during the study period (yes/no), any of 13 mental health disorders recorded at any previous visit during the study period (yes/no for each diagnosis), admitting service (general pediatrics only vs specialty pediatrics/surgery), and adjusted weight z-score at time of visit. To adjust for dependence between outcomes for patients with repeat hospitalizations during the study period, we modeled a first-order autocorrelation on the logit scale by including restraint at the previous visit (yes/no) as an explanatory variable. Thus, we allowed for a correlation between a patient’s restraint outcomes at consecutive visits and, by extension, for weaker correlations between outcomes at nonconsecutive visits, thereby adjusting for within-patient clustering of observations.

For analysis purposes, children were categorized as Black, Hispanic, multiracial, other/unknown to respondent, or White. The category for “other/unknown” was constructed by the authors because of the small number of restraints in the component subgroups. To allow for age and weight z-score to have nonlinear associations with odds of restraint, we computed a single-knot natural spline basis for each variable and included the resulting variables in the model. Length of stay was included as an exposure variable. Seasonality was captured by including indicator variables for winter (December–February), spring (March–May), and summer (June–August), with fall (September–November) as the referent. Statistical analyses are described in more detail in the Supplemental Information.

To mitigate any demographic bias associated with mental health diagnoses, we performed a sub-analysis of restraint use limited to patients with a mental health diagnosis recorded during the study period.

We identified 29 808 children who experienced 46 302 unique hospitalizations during the study period. Within this cohort, 225 children experienced mechanical restraint across 275 hospitalizations (a rate of 5.9 restraint episodes per 1000 hospitalizations). Among these 225 children, 195 had 1 restraint-associated hospitalization, whereas 30 patients experienced 2 or more restraint-associated hospitalizations. Patient characteristics for our cohort are found in Table 1. The median patient age was 12.5 years (interquartile range 8.5–15.6), and one-half (50%) identified as male sex. The majority (66%) of patients identified as White. One-third of all patients had a mental health disorder code identified during the study period. Median hospital LOS was 1.9 days (interquartile range 1–3.7). Mental health disorder codes were associated with 33% of all hospitalizations (Table 2). Rates of missing data were low. Weight was missing for 2.6% of visits overall and for 6 (2.2%) of the 275 visits involving a restraint. Race and ethnicity were missing for 1.3% of patients overall and for 2 (0.9%) of the 225 patients experiencing restraint. Public insurance status was missing for 0.1% of patients overall and for no patients who experienced restraint.

In unadjusted comparisons between restrained patients and nonrestrained patients (Table 1), restrained patients were typically older (median age 14.7 vs 12.5 years), and more likely to be male (62% vs 50%, odds ratio [OR] 1.6 [95% confidence interval (CI) 1.2–2.1]). Patients with a history of public insurance use accounted for 66% of restrained patients versus 45% of nonrestrained patients (OR 2.3 [95% CI 1.8–3.1]). Black children (32% of restrained patients vs 14% of nonrestrained patients, and multiracial children (8% of restrained patients vs 5% of nonrestrained patients) had odds of restraint 3.0 (95% CI 2.2–4.0) and 1.9 (95% CI 1.1–3.0) times higher, respectively than odds of restraint for White patients. Patients with a mental health disorder code were much more likely to experience restraint than patients without such a code (OR 44.0 [95% CI 24.7–88.9]).

In unadjusted comparisons of hospitalizations with and without restraint use, odds of restraint were higher with admission via the ED (64% of hospitalizations with restraint vs 38% of those without, OR 2.9 [95% CI 2.3–3.8]), coding of a mental health disorder (95% of hospitalizations with restraint vs 33% of those without, OR 38.2 [95% CI 23.2–68.6]), occurrence of a psychology or psychiatry consult (57% of hospitalizations with restraint vs 9% of those without, OR 13.2 [95% CI 10.4–16.8]), and admission by the general pediatrics service (84% of hospitalizations with restraint vs 42% of those without, OR 7.1 [95% CI 5.2–9.8]; Table 2).

Logistic regression model results are shown in Table 3, which includes adjusted odds ratio (aOR) estimates and their 95% CIs. We found sex, race and ethnicity category, restraint use during the immediately preceding hospital visit, and a previous mental health disorder code were associated with a patient’s odds of restraint. After adjusting for covariates, male patients had nearly double the odds of restraint as female patients (aOR 1.9 [95% CI 1.5–2.5]). Black patients had increased odds of restraint (aOR 1.9 [95% CI 1.4–2.6]) compared with White patients. Children described as multiracial also had nearly twice the odds of restraint as White patients (aOR 31.8 [95% CI 1.1–2.8]). By contrast, Hispanic patients had less than one-half the odds of restraint as White children.

Patients restrained at the immediately preceding hospital visit had 8.6 times (95% CI 4.8–15.5) the odds of restraint as those without a record of restraint at the preceding visit. schizophrenia spectrum and other psychotic disorders (aOR 5.4 [95% CI 2.7–10.4]), disruptive, impulse control and conduct disorders (aOR 4.7 [95% CI 2.8–7.8]), suicidal or self-injury (aOR 2.8 [95% CI 1.7–4.6]), and substance abuse-related illnesses (aOR 2.3 [95% CI 1.3–4.3]), were associated with the highest odds of restraint use.

Our sub-analysis of patients with a mental health disorder included 9919 patients (20 280 hospitalizations). Patient and hospitalization characteristics for this cohort are included in Supplemental Table 4. Patients with mental health disorders were restrained at a rate of 13.1 restraints per 1000 hospitalizations. Factors associated with higher restraint risk were similar to those in our primary analyses and included male sex, children categorized as Black or multiracial, and restraint at preceding hospitalization.

In our single-center retrospective study of hospitalized patients, we found restraint use occurred with a prevalence of 5.9 per 1000 hospitalizations and 13.1 per 1000 hospitalizations for patients with at least 1 mental health disorder. Our work, although limited by single-center design, gives a novel description of several diagnostic and sociodemographic factors associated with restraint use among children presenting to a general pediatric hospital setting, regardless of diagnosis or reason for admission.

In our adjusted analyses, we found restraint at the preceding hospitalization was the strongest correlate of restraint use at the index hospitalization. Restraint use at preceding hospitalization is arguably a proxy for the many patient and systems factors that increase the risk of patient behavioral dysregulation and restraint implementation. Previous studies have revealed that a small number of patients are associated with a disproportionately large number of restraint events. A pediatric psychiatric facility found that ∼5% of children accounted for ∼50% of restraint episodes.¹³  Similarly, a behavioral unit located inside a children’s hospital found that 67% of hospitalizations in which restraint use occurred involved at least 2 restraint episodes.²²  These findings suggest that children with a history of restraint use may especially benefit from the early involvement of behavioral specialists and personalized treatment plans. Similarly, debriefing with staff and/or family after restraint use for these patients may help promote staff wellness, as well as identify any anchoring biases staff may have for the patient (e.g., “verbal de-escalation didn’t work last time, so we’ll probably need to restrain her again if she gets upset”).

We found odds of restraint use were highest with psychotic disorders, disruptive disorders, suicidality, substance-abuse disorders, and autism spectrum disorders. Our findings echo studies from other settings, including a pediatric psychiatric hospital and psychiatric ED.^23,24  Unique factors, such as sensory sensitivities and social communication differences may underlie increased restraint risk for youth with neurodevelopmental disabilities like autism or intellectual disability, but provider unfamiliarity with alternative strategies to prevent and support behavioral dysregulation is among several barriers to reducing restraint risk for this population.³  There are emerging efforts on guiding the health care system to consider the needs of this patient population proactively, but more research is needed.^25–27  Routine development of individualized care plans in partnership with the patient and caregivers (guided by staff or consultants with expertise in disabilities and behavioral supports) would promote wellness across health care settings and likely benefit patients regardless of diagnosed or identified developmental or mental health needs.^2,28  Additionally, proactive development of individualized behavioral support plans may reduce stigma around behavioral dysregulation that may also underly decision-making around restraint use.

We found increased odds of restraint use among patients with suicidality/self-harm. This was unexpected, as many clinicians at our institution perceived patients with a primary presentation of suicidality to be less likely to engage in disruptive behavior toward others compared with other patient populations. Our results support the previous literature²³  and highlight the substantial risk for unsafe behaviors among patients with depression or suicidality. We initially speculated the strong association between restraint use and suicidality was because of the high frequency at which suicidality was comorbid to other mental health conditions; however, in regression modeling the effect of suicidality is estimated after adjusting for the effects of the other diagnoses. Additionally, among patients with comorbid health conditions, a diagnosis of suicidality further increased the odds of restraint. For example, among patients with disruptive, impulse control, and conduct disorders, 10% without suicidality/self-harm were restrained vs. 18% of those with suicidality/self-harm. Similar increased restraint use was seen for patients with comorbid suicidality/self-harm and bipolar and related disorders (2% restrained without suicidality/self-harm vs 13% with) and with comorbid suicidality/self-harm and ADHD (2% without suicidality vs 11% with suicidality). Given the imprecision in mental health characterizations based on billing diagnoses used in our dataset, the authors of future research should attempt to examine what additional combinations of comorbid mental health/developmental conditions drive our observed associations.

We found racial inequity in restraint use, with estimated odds of restraint for Black children being nearly twice as high as those for White children after adjusting for age, sex, insurance status, and mental health conditions. Although these results echo findings from acute mental health settings and the adult restraint literature, the disproportionate restraint of Black children in a general pediatric setting has not been previously reported. Similarly, our finding that males experienced more restraint episodes than females is consistent with research in other pediatric settings. We speculate that the observed racial inequity in restraint use may reflect interactions between the well-established implicit and systemic racial biases in health care,²⁹  mental health stigma in health care,³⁰  and longstanding racial inequity in accessing mental health supports.³¹  Our results unfortunately add to the evidence for racial inequity in restraint use across medical settings.^7,8  Although quality improvement (QI) demonstrations have proven successful at achieving decreases in pediatric restraint use, changes in racial inequity within these QI studies were unaddressed.^5,6  As suggested by previous literature on the association of QI initiatives and health disparities,³²  it is likely that embedded equity metrics within efforts to reduce restrictive practices for pediatric patient populations are needed. Linking efforts to reduce restraint use with broader institutional efforts focused on health equity and provider bias may help promote buy-in from organizational leaders.^33–35 

Our study should be interpreted in the context of several limitations. Restraint episodes are the product of many interacting child factors (behavioral and medical conditions, psychosocial history, acute pain/illness, etc.) and contextual factors (hospital environment, staff training, etc.) that we cannot parse out in this data set. Moreover, we did not examine the use of chemical restraints in our hospital setting, given the lack of a clear definition³⁶  and ambiguous recording of the intent of pharmacological interventions in the medical record. Future work, including qualitative studies of patient/family/staff experience, is critically needed to better understand the intersection of patient escalation and staff response. Although it was outside the scope of this paper, the examination of interactions between mental health conditions is needed to better understand how comorbid mental health conditions affect restraint use and other clinical outcomes. As a single-center study, our results may not be generalizable to all hospitals; however, the methodology is generalizable, and we hope others will conduct similar efforts at other hospitals. Multicenter studies of pediatric restraint are needed to better understand variability in the prevalence of restraint use and provide a more robust understanding of what geographic, systems, and demographic factors influence the odds of restraint.

At our hospital, several initiatives to improve the care of patients with behavioral health needs were launched in the last year of the study period, including the implementation of a behavioral emergency response team, expanded staff training on enhanced behavioral de-escalation training, increased sensory/environmental and behavioral support procedures, and routine stocking of staff behavioral protection equipment (eg, Kevlar sleeves and gloves) to reduce staff injury risk and the need for patient restraint. Is it unclear if these initiatives measurably impacted the population of patients receiving restraints during our study period. Efforts are ongoing, and we hope these results invite other institutions to undertake their own efforts to reduce reactive, restrictive behavior management strategies.