Agitation in Patients With Autism Spectrum Disorder Admitted to Inpatient Pediatric Medical Units

Through a retrospective chart review, we identified every patient >8 and <19 years of age with a documented ASD diagnosis admitted to the general pediatric medical floors as well as the PICU at MassGeneral Hospital for Children, a tertiary care academic medical center in Boston, Massachusetts, over a 5-year period from 2009 to 2013. This was done by using keyword searches of the electronic medical record, which initially identified 705 potential admissions. These charts were then manually reviewed in detail to verify documentation of the diagnosis. Patients who had clear, consistent documentation in the medical record of a diagnosis recorded as autism, ASD, pervasive developmental disorder not otherwise specified (PDD-NOS), or Asperger disorder were included. Cases in which the diagnosis was described inconsistently in the chart or given in vague terms, such as “developmental delay” or “possible autism,” were not included.

Patients who were admitted primarily for psychiatric reasons, such as those awaiting inpatient psychiatric placement, were excluded from the study. There were 299 medical admissions that met these criteria during the study period. We then performed a detailed chart review of each of these admissions. Data gathered included reason for admission, level of pain at the time of admission, duration of hospitalization, comorbid psychiatric diagnoses, past psychiatric history, history of aggression, history of self-injurious behavior, history of sensory sensitivities, and degree of intellectual disability. When the medical record contained information regarding past neuropsychological testing, this information was included. Otherwise, degree of intellectual disability was estimated on the basis of documentation of the patient’s cognitive capabilities and language functioning. For example, patients who were described as nonverbal were estimated to have severe intellectual disability for the purposes of this study. Similarly, when documentation of outpatient mental health or pediatric treatment was included in the record, this information was used in the determination of variables such as past psychiatric history and history of psychiatric diagnoses. Severity of agitation was estimated on the basis of the potential for harm, degree of disruption, and responsiveness to intervention, which were based on the documentation and descriptions of behavior in the medical record.

The goals of the current analysis were to (1) estimate the risk of agitation and report the frequency of its consequences in inpatient hospital medical admissions for young persons with ASD and (2) identify risk factors associated with agitation in inpatient admissions for young persons with ASD.

We use risk factor to indicate any characteristic of a patient that can be identified at the time of admission (the start of the risk period for agitation) and that is associated with a change in risk of agitation during hospitalization, even if the association is correlational rather than causal. Because differences in risk between admissions are clinically important both within and across patients, because the presence of some candidate risk factors varied across hospitalizations from the same patient as well as across patients, and because some individuals had admissions both with and without agitation episodes, we modeled associations at the admission level rather than the individual level and allowed data from multiple admissions from the same patient. We classified covariates as either a priori control covariates (those that would be included in models for risk of agitation regardless of statistical significance because they had strong potential to act as confounders; age at admission and sex), primary candidate risk factors (those that we hypothesized to have strong associations with agitation on the basis of clinical experience; history of agitation, documented sensory sensitivities, and psychiatric comorbidity), or exploratory candidate risk factors (those with potential associations with agitation but with less basis in clinical experience; past psychiatric admissions, intellectual disability, acute pain at admission, number of psychotropic medications at admission, and Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition [DSM-IV] diagnosis). Reason for admission was considered as a candidate risk factor in a post hoc analysis because of its highly significant but unanticipated association with agitation.

To quantify and test the significance of the association between candidate risk factors and agitation during hospitalization, we used the modified Poisson regression approach to performing relative risk regression for binary data,⁷  together with the method of generalized estimating equations to account for clustering of hospitalizations within individuals, and an exchangeable correlation structure.⁸  In our first step, we quantified the association of each of our primary and exploratory candidate risk factors with agitation in a model controlling for age at admission and sex (first-stage models). In a second step, we used a modified backward stepwise selection procedure to select among our primary and exploratory candidate risk factors individually significant at the α = .10 level in the first step for inclusion in a second-stage model. Discrimination of our second-stage model was assessed by using area under the curve (AUC) and calculated by using methods for clustered data.⁹ 

In the interest of parsimony, for covariates categorized as none, mild, moderate, or severe (history of agitation and intellectual disability), we initially fit models assuming the same change in risk between the mild and moderate and between the moderate and severe categories on the log scale. No equivalent assumption was made about change in risk between the none and mild categories. We assumed a change in risk with number of psychotropic medications that could be adequately captured by a classification of none, 1 or 2, and ≥3 and assumed a linear increase with age. We then assessed for evidence of violation of these assumptions by comparing the fits of our chosen models with the fits of less restrictive models (eg, allowing unequal changes between the mild and moderate and between the moderate and severe categories, adding a squared term for age). Reason for admission was categorized as neurologic conditions (including seizures), behavioral change, or other.

Characteristics of the sample were compared between patients with and without agitation by using Welch’s t test and Fisher’s exact test and between admissions with and without agitation by using repeated-measures linear and risk regression with robust SEs. Regression models were fit by using PROC GENMOD in SAS (version 9.4; SAS Institute, Inc, Cary, NC), and AUC was calculated by using the funcs_clusteredROC function¹⁰  for R (version 3.5.0).¹¹  Wilson’s method was used to obtain the confidence interval (CI) for the percentage of patients experiencing agitation; the CI for the percentage of hospitalizations was obtained by fitting an intercept-only regression model with independence as the working correlation structure. The risk ratio (relative risk) was used as our measure of relative association. Fisher’s exact test was used in post hoc tests for associations between history of agitation and other candidate risk factors. Statistical significance required test-wise 2-tailed P < .05.

Hospitalization records were available for 299 admissions for 168 patients aged >8 and <19 years with a confirmed ASD diagnosis. Patients contributed data from up to 15 admissions, although most patients contributed data from only 1 admission. Demographics and clinical characteristics of the patients, overall and by presence of agitation during ≥1 hospitalization, are presented in Table 1. These characteristics did not differ across hospitalizations for patients with multiple hospitalizations. Characteristics of hospitalizations, overall and by presence of agitation, are presented in Table 2. These characteristics could vary across hospitalizations for patients with >1 hospitalization.

One or more episodes of agitation occurred during 37 of 299 hospitalizations (12.4%; 95% CI 8.9%–17.3%) and for 31 of 168 patients (18.5%; 95% CI 13.3%–25.0%). The number of episodes of agitation per hospitalization ranged from 1 to 32; of 37 hospitalizations with agitation, 15 (41%) had 1 episode, 13 (35%) had 2 or 3 episodes, and 9 (24%) had ≥4 episodes. Psychotropic medications in addition to the patient’s usual medication regimen were administered on an as-needed or emergent basis during 25 of 37 (68%) of hospitalizations with agitation episodes, compared with 15 of 262 (6%) hospitalizations without agitation episodes. A clear precipitant of agitation episodes, such as pain or constipation, was documented for fewer than half of the hospitalizations with agitation. Of the 37 hospitalizations during which an episode of agitation occurred, 2 resulted in injury to the patient, and 1 resulted in injury to staff. Further information on these and other potential correlates and consequences of agitation episodes is given in Table 2.

Frequencies, risk ratios for agitation during hospitalization, and P values for primary and exploratory candidate risk factors and control covariates specified a priori are presented in Table 3. History of agitation and documented sensory sensitivities (primary candidate risk factors) were significantly associated with risk of agitation in both first- and second-stage models. No other primary or exploratory candidate risk factors were selected for inclusion in the second-stage model. History of agitation had the strongest association with agitation during hospitalization of any candidate risk factor; after controlling for age at admission and sex, the risk of agitation during hospitalization was estimated to be 26.7 (95% CI 6.5–110.0) times higher for hospitalizations of patients with a history of severe agitation than for hospitalizations of patients with no history of agitation. This estimate decreased to 21.9 (95% CI 5.4–88.3) after controlling for documented sensory sensitivities. Risk ratios for documented sensory sensitivities were 3.1 (95% CI 1.6–5.8) in the first-stage model and 2.3 (95% CI 1.3–3.8) after adjusting for history of agitation. Psychiatric comorbidity was not significantly associated with agitation in the first-stage model and was not eligible for inclusion in the second-stage model.

Of the exploratory candidate risk factors, only past psychiatric admission was significantly associated with agitation during hospitalization in first-stage models. However, after adjusting for history of agitation and documented sensory sensitivities, the association was no longer significant, and the estimated risk ratio was <1 (0.8; 95% CI 0.4–1.5). Intellectual disability was eligible for inclusion in the second-stage model but was not significantly associated with agitation after controlling for history of agitation and sensory sensitivities (P = .18), with an estimated risk ratio for severe intellectual disability of 1.5 (95% CI 0.7–3.3). In post hoc analyses, reason for admission was strongly associated with agitation after controlling for age at admission and sex (P ≤ .001); risk of agitation was reduced for neurologic conditions but elevated for behavior change relative to other conditions. After controlling for history of agitation and documentation of sensory sensitivities, the reduction in risk for neurologic conditions remained substantial, but the increase in risk for behavioral change was attenuated (P = .005; estimated risk ratio 0.4 [95% CI 0.2–0.8] for neurologic conditions; estimated risk ratio 1.3 [95% CI 0.7–2.4] for behavior change).

The estimated AUC for our second-stage model (with age at admission, sex, history of agitation, and documented sensory sensitivities as covariates) was 0.85 (95% CI 0.78–0.92). The model-estimated risk of agitation for each admission, along with the observed frequency of agitation for each quintile of model-estimated risk (indicated by crosses), is shown in Fig 1. The model appears to overestimate the risk of agitation for patients in the fourth quintile but, otherwise, has reasonable fit.

In this retrospective review, we found that at least 1 documented episode of agitation occurred in 12.4% of inpatient pediatric medical admissions of patients with ASD during the study period (a small but significant minority) and in 18.5% of patients in our sample who were hospitalized. Given that not all patients met age criteria during the entire 5-year study window and that some patients may have had additional admissions at other hospitals, this may underestimate the percentage of admitted youth with ASD who will have ≥1 episode of agitation while hospitalized. Physical restraints were used in almost one-third (30%) of hospitalizations with episodes of agitation, and hospitalizations with agitation were far more likely to involve as-needed doses of psychotropic medications. There were only 2 documented incidents of significant injury to the patient as a result of agitation, and there was 1 episode of injury to hospital staff.

Perhaps not surprisingly, we found that the strongest predictor of agitation during a hospital admission was a past history of agitation, with more severe agitation, characterized by the severity and frequency of the past episodes documented in the chart, leading to a greater likelihood of agitation during admission. After we adjusted for age, sex, and sensory sensitivities, patients with a history of severe agitation had 21.9 times the risk of having an episode of agitation during hospital admission, whereas the risk was 11.6 times higher for those with a history of moderate agitation and 6.1 times higher for those with a history of mild agitation. These findings underscore the importance of obtaining a good history of a patient’s past behaviors to best anticipate potential difficulties during a hospitalization.

A history of sensory sensitivities was also significantly associated with an increased risk of agitation during hospitalization, with a 2.3-fold increase in risk after adjusting for age at admission, sex, and history of agitation. Although this does not necessarily reflect a causal association, this is a notable finding because it has been shown that hypersensitivity to sensory stimuli is a common feature in children with ASD and can be a source of significant distress.¹⁴  A hospital admission is typically associated with a barrage of sensory stimuli, such as the sound of alarms and monitoring devices, bright lights, and the smell of disinfecting agents, that can be unpleasant even to an individual without specific sensitivities. Gathering information about a history of these sensitivities may allow for preventive measures, such as involvement of occupational therapy, and the use of techniques and materials such as noise canceling headphones, sunglasses, and sensory-based coping strategies.

Having ≥1 psychiatric admission was strongly and significantly associated with agitation during hospitalization in the first-stage model but not after adjustment for other candidate predictors, including history of agitation. Attenuation of estimates of relative risk after adjustment for history of agitation was expected given the strong association of agitation history with agitation during hospitalization and the potential for factors associated with agitation during hospitalization to also be associated with agitation history. For example, post hoc tests confirmed a significant association of past psychiatric admission with history of agitation (P < .001; Fisher’s exact test); 43% of those with ≥1 past psychiatric hospitalization (but only 13% of those without past psychiatric admission) had a history of severe agitation, whereas 58% of those without a previous psychiatric admission (but only 4% with ≥1 psychiatric admission) had no history of agitation. Documentation of sensory sensitivities was also strongly and significantly associated with agitation history. Unexpectedly, however, psychiatric comorbidity, a primary candidate risk factor in our analysis that also had a strong and significant association with history of agitation, had no significant association with agitation during hospitalization, even before adjustment for history of agitation, and had a relatively modest risk ratio estimate of 1.2.

Overall, results from our second-stage model suggest the potential for even a small number of risk factors assessed at the time of admission to be informative in identifying children with ASD at elevated risk for agitation. Our estimated AUC of 0.85 for our second-stage model would generally be considered in the good range, and our plot revealing low (<0.02) model-estimated and observed risks for children in the first 2 quintiles of risk and high (>0.25) model-estimated and observed risks in the highest quintile of risk suggests clinically meaningful risk differences between children who would be identified as low risk and high risk by our model. Identification of patients at risk can be better informed by studying a larger set of risk factors and by studying risk in a more diverse population of children with ASD.

Limitations of our study include reliance on medical records review, inclusion of hospitalizations and patients from a single hospital, and the small number of hospitalizations with agitation episodes identified. It is likely that there are factors associated with risk of agitation that we were not able to assess in this retrospective chart review. For example, painful or invasive procedures may increase the stress on patients and increase likelihood of agitation. The availability of parents or caregivers in the hospital also is likely to impact how a patient with ASD is able to cope with the stress of hospitalization. Patients of parents who spend less time at the bedside because of the demands of work or the care of other children may have a greater risk of agitation. Patients’ expressive and receptive language capabilities, independent of their intellectual functioning, may also be related to agitation risk because greater language ability may facilitate communication and may be useful in deescalating patients. Finally, involvement from outpatient providers, such as behavioral therapists, may be helpful in reducing agitation risk. These factors would be important to assess in future prospective studies of agitation risk during hospitalization. Results from our post hoc analyses suggest that reason for admission, and admission for neurologic conditions in particular, may also be a useful factor to study. The reduced risk of agitation observed for these patients in our study may be due to the substantial proportion of them who had short admissions, during which they were briefly monitored and then discharged.

The small sample size, and consequently the small number of hospitalizations with agitation episodes identified, resulted in limitations on our statistical analysis. The study was powered to detect only strong associations. This is reflected by the wide CIs for many of our relative risk estimates and the lack of significant associations for some candidate risk factors with relative risk estimates substantially >1. A lack of statistically significant association in this study should not be equated with absence of clinically meaningful association. Low-frequency categories for our outcome and some candidate risk factors limited our ability to build numerically stable models with multiple predictors and made allowing interactions among covariates prohibitive. We also did not validate the results of our analysis using an independent sample and so were not able to assess the impact of overfitting.

Although the majority of patients with ASD admitted to pediatric medical units will not have episodes of agitation during the admission, a significant subset of patients will experience these difficulties, placing themselves and hospital staff at risk. We found that agitation occurred more commonly in patients with a past history of agitation as well as in those with sensory sensitivities. Agitation also occurred more commonly in patients with past psychiatric admissions, although risk was no longer elevated after accounting for history of agitation and sensory sensitivities. We found no significant associations with psychiatric comorbidity, intellectual disability, acute pain at the time of admission, use of psychotropic medications before admission, or DSM-IV ASD diagnosis in this study, although we could not rule out clinically meaningful associations.

Identification of risk factors associated with behavioral dysregulation during medical hospitalization may allow for better screening at the time of admission to identify high-risk patients and intervene to reduce their risk of agitation as well as to improve the quality and safety of inpatient medical care for children with ASD. For the highest-risk patients, early involvement of support services, such as consultation with psychiatry and occupational therapy, and implementation of patient-specific care plans¹⁵  may be helpful in reducing risk. Future research should be directed at identifying risk of agitation in patients with ASD in a proactive manner through the development and implementation of screening tools to facilitate interventions to reduce such risk.

We thank the Nancy Lurie Marks Family Foundation for their support of this research. We dedicate this article to the memory of Ms Rao Hureau, who played an invaluable role in this project.

ASD

autism spectrum disorder

AUC

area under the curve

CI

confidence interval

DSM-IV

Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition

PDD-NOS

pervasive developmental disorder not otherwise specified