CHILDREN AT RISK: AUTISM SPECTRUM DISORDER

Injuries Among Children With Autism Spectrum Disorder Anjali Jain, MD; Donna Spencer, PhD; Wenya Yang, MPA, MA; Jonathan P. Kelly, MPP; Craig J. Newschaffer, PhD; Jonathan Johnson, MS; Jaclyn Marshall, MS; Francisca Azocar, PhD; Loni Philip Tabb, PhD; Taylor Dennen, PhD From the Lewin Group, Falls Church, Va (Dr Jain, Ms Yang, Mr Kelly, Ms Marshall, and Dr Dennen); OptumInsight Life Sciences, Eden Prairie, Minn (Dr Spencer and Mr Johnson); Drexel University School of Public Health, Philadelphia, Pa (Dr Newschaffer and Dr Tabb); and OptumHealth Behavioral Solutions, San Francisco, Calif (Dr Azocar) The National Institute of Mental Health was provided with a copy of the final draft of the article, but the content of the article is solely the responsibility of the authors and does not represent the views of the funding institute, the National Institutes of Health, or the federal government. The authors declare that they have no conflict of interest. Address correspondence to Anjali Jain, MD, The Lewin Group, 3130 Fairview Park Dr, Suite 500, Falls Church, VA 22042 (e-mail: anjali.jain@ lewin.com). Received for publication August 30, 2013; accepted March 28, 2014.

ABSTRACT OBJECTIVE: We compared risk of injury among children with autism spectrum disorder (ASD) to those without ASD, adjusting for demographic and clinical characteristics. METHODS: We used claims data from 2001 to 2009 from a commercial health plan in the United States. A validated ASD case identification algorithm identified 33,565 children (ages 0–20 years) with ASD and 138,876 children without. Counting process models tested the association between ASD status and injury episodes with separate regressions run for children during different age periods. RESULTS: Unadjusted results demonstrated that children with ASD had a 12% greater injury risk than children without ASD (hazard ratio [HR] ¼ 1.119; P < .001). After including demographic variables, the HR was 1.03 (P < .05); after controlling for co-occurring conditions, such as seizures, depression, etc, HR decreased to 0.889 (P < .001). For the age period analysis, HR values were as follows: for 0 to 2 years, HR 1.141; 3 to 5

WHAT’S NEW In a large sample of commercially insured children from 2001 to 2009, younger children with autism spectrum disorder (ASD) and children with ASD and seizures, depression, visual impairment, and attentiondeficit disorders are at the greatest risk of injury.

INJURIES ARE A leading cause of morbidity and mortality among children and are considered almost entirely preventable.1 In 2009, injuries resulted in over 9000 deaths in children in the United States2 and about 9.2 million emergency department visits.3 Numerous studies report that children with developmental disabilities are at increased risk of injury, likely related to the physical, mental, and social vulnerabilities associated with their

years, HR 1.282; 6 to 10 years, HR not significant; and 11 to 20 years, HR 0.634 (P < .05 for all significant results). CONCLUSIONS: Children with ASD have more injuries than children without ASD. After controlling for demographic factors and co-occurring conditions, children with ASD are at lower risk of injury, suggesting that co-occurring conditions or the ways these conditions interact with ASD is related to injuries. Clinicians should understand that injury risk in children with ASD may be driven by co-occurring conditions. Treating these conditions could thus decrease injury risk as well as have other benefits. Injury prevention interventions are especially warranted for younger children with ASD and those with seizures, depression, visual impairment, or attentiondeficit disorders. KEYWORDS: administrative claims; autism spectrum disorder; child injury; commercial insurance

ACADEMIC PEDIATRICS 2014;14:390–397

conditions.4–10 Only 2 recent studies, however, have examined and characterized injury risk specifically among children with autism spectrum disorder (ASD), and neither of these identified characteristics of children most at risk despite the vast heterogeneity present among children with ASD.11,12 Many children with ASD exhibit behaviors such as wandering or bolting— even going missing for a time—that, coupled with the difficulties responding appropriately to social cues, increase their risk of injury that can be severe or even fatal.13 Little is known about whether current injury prevention programs and interventions have helped children with ASD to the same extent as other children or whether more tailored approaches might be needed. We used medical claims data for a large sample of commercially insured children to investigate injuries

ACADEMIC PEDIATRICS 390 Copyright ª 2014 Published by Elsevier Inc. on behalf of Academic Pediatric Association

Volume 14, Number 4 July–August 2014

ACADEMIC PEDIATRICS

INJURIES AMONG CHILDREN WITH AUTISM SPECTRUM DISORDER

among children with ASD (n ¼ 33,565) compared to children without ASD (n ¼ 138,876) and determined whether risk of injury varied by age and the presence of cooccurring conditions. These data allowed us to explore injuries among a large and heterogeneous group of children with ASD. They included information about both inpatient and ambulatory services—important because most childhood injuries do not result in hospitalization.1 Identifying risk factors for injuries among children with ASD is part of addressing the special primary health care needs of children with ASD.

PATIENTS AND METHODS We conducted a retrospective observational study using a proprietary administrative claims database associated with a large US commercial health plan. Using claims linked to enrollment information, we identified 33,565 children with ASD and a random sample (selected by means of a random number generator) of 138,876 children without ASD as an approximately 4:1 comparison group. All children were enrolled in a plan with continuous medical, pharmacy, and behavioral health care coverage for at least 6 months between January 2001 and December 2009. Children with ASD were identified using a validated algorithm14 requiring 2 or more claims with a diagnosis code in any position for autistic disorder or other specified or unspecified pervasive developmental disorder (International Classification of Diseases, 9th edition, Clinical Modification [ICD-9-CM], 299.0x, 299.8x, and 299.9x) on separate dates of service. Also, eligible children with and without ASD were between the ages of 0 and 20 as of their first day (ie, index date) of enrollment during the study period and were without claims for childhood disintegrative disorder (ICD-9-CM code 299.1x) or Rett syndrome (ICD-9CM code 330.8x). Also, the comparison children could not be a family member of a child with ASD. Because 2 claims for ASD in any position generated a positive predictive value approaching 90%,14 children with only 1 ASD claim (n ¼ 12,671) were excluded. Each child’s total observation time was the sum of all of his or her enrollment times between 2001 and 2009 during which he or she had simultaneous medical, pharmacy, and behavioral health coverage. Only claims with dates of service occurring during the study time frame were used in the analysis. Subjects were required to have at least 1 period of 6 months of continuous enrollment but may have had more enrollment time (a longer single period, multiple enrollment periods, or both) with all 3 types of coverage during the study. Gaps in enrollment had to be at least 33 days to be considered discontinuous. Eighty percent of children had only 1 period of continuous enrollment during the study. Of those who had more than 1 period of enrollment, over 90% had only 1 additional period. The mean enrollment lengths during the study were 43.5 months and 30.5 months for children with and without ASD, respectively. Longer enrollment among children with ASD was anticipated because families with chronic health conditions are

391

more likely to seek, stay with, and return to health insurance coverage.15 The study was considered exempt by the New England Institutional Review Board and was approved by OptumInsight’s disclosure limitation program in November 2011. STUDY VARIABLES The primary study outcome was injury, measured as a count variable by summing each child’s number of injury episodes during study enrollment. Both unintentional and intentional injuries were included. Injury episodes were measured using the episode treatment group (ETG) methodology and software. The ETG methodology and a map of diagnostic codes to the 4 ETGs studied is available online (http://www.optuminsight.com/transparency/etglinks/episode-treatment-groups/). ETGs aggregate medical and pharmacy claims data into meaningful episodes of care for a clinical condition and prevent both overcounting of injuries (where multiple services are associated with a single injury) and undercounting of injuries (where a single event may have resulted in multiple injuries). A child could have independent yet overlapping ETG injury episodes. For example, a car accident could have caused 2 injury episodes (fracture and burn) with the same start date. Codes used to define injuries and injury ETGs are provided in Appendix Tables 1 and 2. Covariates were identified through a review of the literature11 and included health conditions that co-occur with ASD and have been independently associated with injury risk: attention-deficit disorders (ADD), anxiety, depression, learning/intellectual disabilities, visual impairment, and epilepsy/seizures (Appendix Table 3). For anxiety, depression, and learning/intellectual disability, subjects had at least 2 medical claims with the relevant diagnosis codes in any position at least 30 days apart. For ADD and seizures, subjects had either 2 or more claims with the relevant diagnosis codes in any position at least 30 days apart or 1 claim with the diagnosis code in any position along with a pharmacy claim for an ADD medication or anticonvulsant, respectively. For visual impairment, subjects had at least 1 medical claim with the relevant diagnosis code in any position. Other covariates were the child’s sex, age, race/ ethnicity, household income, US geographic region, and enrollment time. Age was the child’s age at the first day of enrollment during the study (ie, index date). Race/ ethnicity and income variables were derived through a marketing database (KBM Group; http://www.kbmg. com/), which relies on self-report, modeling, US Census data, and imputed data. Although these data have application to health research, limitations include potential inaccuracies in socioeconomic status assignment, inability to determine when data were imputed, relatively high rates of missing data, and predefined categorizations (eg, income level). Race/ethnicity and income variables were populated for approximately 65% to 70% of children in this study. Those with missing data were included in the “unknown” category.

392

JAIN ET AL

ACADEMIC PEDIATRICS

STATISTICAL ANALYSIS All analyses were conducted by SAS 9.2 software (SAS, Cary, NC). To compare injury risk between children with and without ASD, multivariable regression analyses were conducted using counting process models, an extension of the Cox proportional hazard model applied to recurrent count data where each subject contributes to the risk for an event under observation at the time the event occurs. The model yields hazard ratios that are interpreted as log relative risks of the occurrence of an injury event. Counting process models can be applied where subjects are observed for discontinuous risk intervals (such as 2 separate enrollment periods with both included in the model).16 A sandwich estimate of the covariance matrix was used to account for intrasubject correlation of these risk intervals. As with other approaches based on a Cox model, the counting process model requires a proportionality assumption, which was found to hold based on inspection of the survival function and the log negative log of the survival function across time by sample. The Wald test of global fit was examined, and all models were statistically significant at the .001 level. Three models are presented: 1) an unadjusted model estimating only the association of ASD versus no-ASD group on injury risk, 2) a model adjusting for demographic characteristics only, and 3) a model adjusting for both demographic characteristics and co-occurring conditions. Potential interactions were explored between ASD status and sex, age, and each co-occurring condition to examine whether the effect of ASD on injury risk differed across key subgroups. To explore the dependence of the ASD association with injury on age, we also stratified the samples into 5 age periods (0 to 2, 3 to 5, 6 to 10, 11 to 20, and 21þ years) and ran separate models for subsamples with enrollment time during these ages. Children who had at least 1 day of enrollment during an age period were included in the age period analysis and were observed for as long as they were enrolled during the age period. Individuals included in the 21þ age period were aged 20 or younger at study start (as were all subjects) but turned 21 during the study. Over 60% of all children were included in only 1 age period. Although injuries as acute events may be less susceptible to surveillance bias (where individuals with ASD and more contact with the health care system might be more likely to have claims for injuries), we tested for surveillance bias by exploring whether ASD risk estimates changed after adjustment for the number of preventive care visits during subjects’ study enrollment.

RESULTS BASELINE STATISTICS A total of 15,023 (44.8%) of children with ASD had at least 1 injury episode compared to 43,762 (31.5%) of children without ASD. Table 1 presents the unadjusted count of injury episodes and the demographic, enrollment, and clinical characteristics for both groups. Although the mean number of injury episodes was less than 1.00 for both

groups, it was higher for children with ASD than children without ASD (0.87 and 0.55, respectively); this difference held across all age periods. The median number of episodes overall was 0, and the 75th percentile was 1.0 for both groups, with the maximum number of episodes reaching 25 and 20 for the 2 samples, respectively (data not shown). Adjusting for enrollment time, the overall rate of injury episodes per year was 0.24 for children with ASD compared to 0.22 for the comparison group (P < .001; data not shown). As shown in Table 1, sex composition differed, with the male-to-female ratio around 4:1 in the ASD group and 1:1 in the comparison group. On average, children with ASD were younger at study start (mean 6.73 years vs 8.66 years). When race/ethnicity information was available (61.6% of children with ASD and 51.5% of comparison children), more children in the ASD sample were white (86.1% vs 78.7%), and slightly more comparison children were African American/black (6.8% vs 3.3%) and Hispanic (10.4% vs 6.6%). When household income data were available (58.4% of children with ASD and 45.4% of comparison children), slightly higher percentages of children without ASD fell into income groups lower than $75,000. Table 1 also shows the distributions of the co-occurring conditions. Among children with ASD, the most common conditions were ADD (38.8%), anxiety (16.4%), and depression (12.1%), while all co-occurring conditions were observed in less than 5% of children without ASD. MULTIVARIABLE RESULTS Table 2 presents the results of the unadjusted and adjusted counting process models examining risk of injury for children with and without ASD. In the unadjusted model, children with ASD have a 12% greater injury risk than children without ASD (hazard ratio [HR] 1.119; 95% confidence interval [CI] 1.099, 1.140). With inclusion of demographic covariates, the HR decreased to 1.029 but remained statistically significant (95% CI 1.009, 1.050). When the cooccurring conditions were incorporated into the model, the risk estimate fell to 0.890 (95% CI 0.870, 0.911), suggesting that children with ASD had an 11% lower risk of injury than children without ASD. Each co-occurring condition included in the model was independently associated with increased risk of injury, with the highest risks associated with seizures (HR 1.433; 95% CI 1.365, 1.505), depression (HR 1.298; 95% CI 1.254, 1.344), and visual impairment (HR 1.197; 95% CI 1.141, 1.256). Interactions between ASD status and sex, age, and cooccurring conditions were explored (data not shown). The interaction between ASD and seizures was not statistically significant. However, the relationship between ASD and injury risk varied significantly by whether a child had depression or ADD (P values on interaction terms

Injuries among children with autism spectrum disorder.

We compared risk of injury among children with autism spectrum disorder (ASD) to those without ASD, adjusting for demographic and clinical characteris...
189KB Sizes 0 Downloads 9 Views