Psychiatric Hospitalizations Among Autistic and Non-Autistic Older Adults

Study design

We used a retrospective cohort study design of secondary data. We observed beneficiaries for a 365-day baseline period and a subsequent measurement period of at least 365 days, resulting in a study period of at least two years. We extracted beneficiaries’ demographic characteristics from the baseline period. We identified outcomes of interest during the measurement period.

Data source

We derived data from Centers for Medicare and Medicaid (CMS) Limited Data set Standard Analytical Files. These files contain de-identified billing claims for health care encounters for 100% of Medicare fee-for-service beneficiaries during the years 2016–2022. Medicare is a government-funded health insurance program that covers inpatient and outpatient health care services for American adults aged 65 years and older. Inpatient files contain billing claims linked to hospital admission encounters. Outpatient files contain billing claims from settings such as emergency departments, primary care offices, and outpatient clinics. ³⁷ Medicare provides health insurance coverage for approximately 99% of American adults aged 65 years and older; ³⁸ therefore, this dataset is broadly representative of the older adult population of the United States.

Sample

The study sample consisted of an Autistic older adult cohort and a PC cohort from the United States. We included Autistic older adults in this study if they were (a) continuously enrolled in Medicare for a minimum of 48 months; (b) aged 65 years or older within the study period; and (c) had at least one inpatient claim or two outpatient claims with an Autism diagnosis anytime during the baseline period. We identified an Autism diagnosis using International Classification of Diseases, 10^th edition codes (ICD-10 F84.0, F84.1, F84.5, or F84.9).

The PC cohort met the same inclusion criteria as the Autistic older adult cohort, except they had no medical encounter claim with an Autism diagnosis code at any point during the study period. We used group frequency matching in tandem with variable ratio propensity score matching to match at most two PC beneficiaries with one Autistic beneficiary. We matched beneficiaries exactly on baseline age, sex, and race. We performed greedy nearest neighbor matching, with a caliper of 0.2 without replacement, on baseline longitude/latitude coordinates for patient county of residence, Health Maintenance Organization status, presence of baseline psychiatric diagnoses, and Charleson Comorbidity Index ³⁹ (CCI). Baseline psychiatric diagnoses consisted of a series of 11 binary variables, which indicated the presence of at least one inpatient or outpatient encounter during the baseline period with a diagnostic code from the Mental, Behavioral, and Neurodevelopmental Disorders (MBD) category in the Healthcare Cost and Utilization Project (HCUP) Clinical Classifications Software Refined (CCSR) ⁴⁰ including mood disorders, personality disorders, and substance use disorders (see Supplementary Table S1). Baseline psychiatric diagnoses were not mutually exclusive.

Measures

Our primary independent variable for this study was a diagnosis of Autism. Our outcome variables for this study were psychiatric hospitalizations, LOS, and psychiatric readmissions. We measured psychiatric hospitalization as a binary variable, indicating whether a beneficiary had at least one inpatient hospitalization with a psychiatric condition listed as the principal diagnosis. We identified psychiatric conditions using the HCUP-CCSR, which organizes ICD-10 codes into clinically meaningful categories based on body systems. ⁴⁰ We then used General Equivalence Mappings to identify corresponding ICD-9 codes. ⁴¹ We considered diagnoses within the “MBD” category (e.g., including mood disorders, personality disorders, substance use disorders, etc.) as psychiatric conditions. We identified any hospitalization with an MBD code as the principal diagnosis as a psychiatric hospitalization. Supplementary Table S1 provides a complete list of MBD codes we used in this study. We defined LOS as the number of days between admission and discharge for the index hospitalization, which refers to the first psychiatric hospitalization during the measurement period. We measured psychiatric readmission as binary indicators of whether a beneficiary was rehospitalized for a psychiatric condition within 30-, 90-, or 180-days after discharge from the index psychiatric hospitalization. We treated each psychiatric readmission interval as a mutually exclusive binary variable.

Covariates included sex, race, age, U.S. region, rurality, CCI, baseline psychiatric diagnoses, prior history of a psychiatric hospitalization, and diagnosis of intellectual disability. Rurality was determined by utilizing the beneficiary’s Social Security Administration code for their county of residence during the baseline period and was then categorized based on United States Department of Agriculture rural–urban continuum codes. ⁴² We defined prior history of a psychiatric hospitalization as having at least one psychiatric hospitalization during the baseline period. We identified an intellectual disability diagnosis based on having at least one inpatient or two outpatient encounters with an ICD-10 code of F70–F79.

Data analysis

We used descriptive statistics to summarize all demographic characteristics. We described continuous variables as median and interquartile range and described categorical variables as frequency and percentage. We performed five separate regression models to compare the cohorts on all outcome measures. Specifically, to compare the odds of psychiatric hospitalization between Autistic older adults and PCs, we used multivariable logistic regression. Among beneficiaries who had a psychiatric hospitalization, we used negative binomial regression to compare LOS between Autistic older adults and PCs. In addition, among beneficiaries who had a psychiatric hospitalization, we performed three separate multivariable logistic regression models to compare cohorts on the odds of psychiatric readmission at 30, 90, and 180 days. Given the rare occurrence of readmissions following hospitalizations, we performed a series of sensitivity analyses serving as robustness checks to our primary findings. All regression models adjusted for age, sex, race/ethnicity, rurality, U.S. region, CCI, baseline psychiatric diagnoses, and prior history of inpatient psychiatric hospitalizations. Administrative claims data contain minimal missing information due to required billing fields; thus, we used complete case analysis when missing values occurred, with the exception of race, which was missing in fewer than 2% of observations and was combined with other race.

We conducted additional analyses to assess the influence of co-occurring intellectual disability on study outcomes. For these analyses, we replicated all models after stratifying the sample based on whether the Autistic older adult beneficiary had a co-occurring intellectual disability. For all models, we compared Autistic beneficiaries with and without a co-occurring intellectual disability to their respective matched PCs. We used SAS statistical software version 9.4 for all data analyses and assessed significance at the 0.05 level.

Odds of psychiatric hospitalization

Contrasting with prior studies using younger samples,^28–30 we found no significant difference in the odds of psychiatric hospitalization between Autistic older adults and PCs prior to stratifying for intellectual disability. However, differences in methodology, sample, and data source limit direct comparisons between our findings and prior studies.

Pertaining to methodology, we used propensity score matching to identify a subset of PCs who were well-balanced with our Autistic older adult cohort on demographic characteristics and factors potentially associated with psychiatric hospitalizations, such as psychiatric diagnoses at baseline. Conversely, other studies used non-matched cohorts or entire system-level samples.^28,29,31,34 The use of propensity score matching, in addition to ensuring more balanced cohorts, helps mitigate the confounding effects of demographic and clinical covariates that could impact the likelihood of psychiatric hospitalizations.

Pertaining to study samples, prior studies focused on young adult samples^28,29,31,32 or combined Autistic older adults with younger age groups. ³⁴ Autistic young adults often experience health care challenges unique to their life stage, including a loss in school-based services and discontinuities in care, as patients shift from pediatric to adult health services.^30,47,48 Moreover, the transition to adulthood is a period where the onset of psychiatric conditions is common ⁴⁹ and is often marked by changes in daily routines and responsibilities, which can exacerbate existing mental health challenges in persons with disabilities. ⁵⁰ Therefore, our findings, in conjunction with prior studies, may suggest that Autistic young adults are more likely to have psychiatric hospitalizations than their peers, whereas Autistic older adults are not. Alternatively, given underdiagnosis of autistic people in prior generations, ⁵¹ our sample may over-represent Autistic older adults with intellectual disability and under-represent Autistic older adults without intellectual disability; in part, this could explain differences between our findings and studies of younger Autistic populations.

Last, pertaining to data source, prior studies used regional^31,32 or local single-site data, ²⁸ which reflect patterns of psychiatric hospitalizations within specific geographic regions. In contrast, we used Medicare claims data, a comprehensive U.S. national data source. This allowed us to examine a geographically generalizable sample of Autistic older adults.

Length of stay

We found that Autistic older adults in our sample experienced a 23% longer LOS (an approximate difference of 3 days in our sample) for psychiatric hospitalizations than PCs, a statistically significant difference. This finding is consistent with existing literature, which has documented longer lengths of stay among younger Autistic individuals during psychiatric hospitalizations.^29,34,52 Longer psychiatric hospitalizations can reflect appropriate clinical care or potential gaps in service delivery, highlighting the need for nuanced interpretation.

Longer hospitalizations may be necessary for patients who require more extensive discharge planning or for patients receiving complex psychiatric care. ⁵³ Autistic adults can benefit from tailored psychiatric interventions, ⁵⁴ which may require additional time to implement. This could potentially explain the longer LOS observed in our sample. Our data do not provide information about the quality of care or autistic adults’ perceptions about whether longer lengths of stay were beneficial, but this is an important direction for future work.

In contrast, longer hospitalizations may indicate ineffective service delivery. Payer and hospital systems generally aim to decrease LOS due to shortages of psychiatric beds ²² and to reduce costs. ⁵⁵ One study reported that training pediatric psychiatric providers in Autism-specific interventions significantly reduced LOS, ⁵⁶ suggesting that inadequate provider knowledge or lack of appropriate interventions may contribute to longer psychiatric hospitalizations for this population. While it was beyond the scope of this study to examine the appropriateness of LOS and quality of interventions provided during Autistic older adults’ psychiatric hospitalizations, this is an important area for future work.

Psychiatric readmissions

We found that Autistic older adults in our sample had significantly higher odds of psychiatric readmission at 90 and 180 days post-discharge, but not at 30 days. While Rast et al. reported higher 30-day readmissions among a younger sample of Autistic adults, their study examined all-cause readmissions rather than psychiatric readmissions. ³⁴ Therefore, our findings provide a more targeted estimate of psychiatric readmissions in Autistic older adults.

One possible explanation for higher odds of psychiatric readmission may be inadequate care following discharge. In the general population, care coordination, ⁵⁷ aftercare (e.g., referral to a community-based agency, follow-up, and receipt of services), ⁵⁸ and high mental health care continuity ⁵⁸ following a psychiatric hospitalization reduce psychiatric readmissions. Autistic adults often experience poor continuity of health care services ¹⁸ and experience well-documented barriers to accessing community-based mental health care.^21,24 Therefore, this finding may underscore the importance of improving community-based mental health care follow-up and efforts to improve mental health care continuity for autistic adults who have a history of psychiatric hospitalization.

However, psychiatric readmission is complex and impacted by numerous clinical factors and social determinants of health.^57,59–61 Particularly, racial minority status, ⁶⁰ homelessness, ⁵⁷ social isolation, ⁵⁹ substance misuse, ⁵⁷ and suicidal ideation ⁵⁷ are positively associated with psychiatric readmissions, whereas higher education is a protective factor against readmission. ⁶¹ Autistic older adults are less likely to attain postsecondary education ⁶² and are more likely to experience suicidal ideation ⁶³ and feelings of social isolation, ⁶⁴ which may contribute to the higher odds of psychiatric readmission at 90 and 180 days observed in this study. Examining the intersection of Autism and social determinants of health is therefore essential to developing a more comprehensive understanding of psychiatric readmission. Further research is needed to clarify the mechanisms underlying increased odds of psychiatric readmission among Autistic older adults among Autistic older adults and to assess the broader implications for care delivery.

Impact of intellectual disability status

We observed notable differences between Autistic older adults with and without intellectual disability. Autistic older adults without intellectual disability had significantly higher odds of experiencing a psychiatric hospitalization and significantly longer LOS relative to their matched PCs. In contrast, Autistic older adults with intellectual disability had significantly lower odds of experiencing a psychiatric hospitalization and showed no significant differences in LOS relative to their matched PCs. These findings were contrary to our original hypothesis. We expected Autistic older adults with intellectual disability would have greater odds of psychiatric hospitalizations, readmissions, and longer LOS relative to PCs. This hypothesis was based on prior research examining emergency department utilization ³⁶ and increased medical complexity of this patient population. ⁶⁵

We posit that the reasons for differences in psychiatric hospitalizations of Autistic older adults with and without intellectual disability is likely multifactorial. One possible factor is eligibility for public benefits and support services. In the U.S., Autistic individuals with intellectual disability often qualify for programs such as Home and Community-Based Services (HCBS), ⁶⁶ which provide noninstitutional, government-funded support to enable one’s function and participation across their adult lifespan. ⁶⁷ HCBS are administered at the state level to improve the health and welfare of vulnerable persons (e.g., developmental disability, traumatic brain injury), but in many states, Autistic individuals without intellectual disability may not qualify for these services. ⁶⁸ More states are expanding eligibility for HCBS to include Autism, ⁶⁹ but waitlists to receive such services are often lengthy, ⁷⁰ and it is possible that mental health conditions may deteriorate while they are waiting for services. Autistic adults without intellectual disability may not receive governmental support until older adulthood when age-related support programs become widely available. ⁶⁷ We postulate that lack of access to HCBS throughout adulthood may play an important role in shaping long-term patterns of health care utilization.

Another possible explanation is that medical providers may have greater difficulty accurately identifying psychiatric symptoms in patients with intellectual disability due to diagnostic overshadowing or communication differences. ⁷¹ Examining causal mechanisms explaining why Autistic older adults’ psychiatric hospitalization varies as a function of intellectual disability was beyond the scope of our study but is an important direction for future work. Similarly, inpatient mental health treatment settings may be ill equipped to adequately serve the potentially more complex mental health, communication, and behavioral needs of autistic adults with an intellectual disability. Therefore, providers may refer autistic adults with intellectual disability to other congregate care settings (e.g., group homes) that are more equipped to provide adequate behavioral and/or communication supports.

Methodologic considerations

We acknowledge several limitations of the present study. First, the unique characteristics of the sample limit the generalizability of our findings. These findings are nationally representative of the U.S. but may not be generalizable to other countries due to differences in health care systems. In addition, findings may not generalize to future generations of U.S.-based Autistic older adults. All beneficiaries in our sample included were born before 1950, meaning they reached adulthood before Autism was formally recognized as a diagnosis in the Diagnostic and Statistical Manual of Mental Disorders in 1980. ⁷² As a result, all Autistic individuals in this study were diagnosed in adulthood and likely did not have access to evidence-based therapies or early intervention services in childhood. Similarly, key U.S.-based disability rights laws, including the Americans with Disabilities Act ⁷³ and Individuals with Disabilities Education Act ⁷⁴ were not enacted until these individuals were already adults. Also, current diagnostic criteria for Autism are more inclusive and capture a wider range of presentations, whereas earlier criteria were more restrictive and more likely to identify individuals with higher support needs and/or co-occurring intellectual disability. Thus, Autistic older adults without intellectual disability may be underrepresented in this sample.

Second, our data period includes the COVID-19 pandemic. The impact of COVID-19 on the mental health needs of Autistic people is undetermined; however, research has found an increase the incidence of depression and decrease in health care service use among older adults directly following the pandemic. ⁷⁵ As a result, future cohorts of Autistic older adults may have experienced different developmental courses and life experiences, leading to different health care needs and service use patterns.

Third, there are intrinsic limitations with the use of medical billing data. Psychiatric hospitalizations, LOS, and psychiatric readmissions are influenced by numerous social determinants of health that cannot be captured through billing data alone.^57,76–78 These should be considered in future studies using different data sources. This data source does not provide information on specific interventions received during psychiatric hospitalizations, which could impact outcomes. Similarly, this data source does not capture patient perspectives or satisfaction with hospitalization, both of which are important for understanding the quality and effectiveness of care. Fourth, we did not examine or control for hospital characteristics, including hospital type (e.g., public, private, or forensic) or the availability of disability-specific care. In addition, the very low event rate of psychiatric readmissions may have limited statistical power and reduced the stability of the estimates; therefore, we recommend interpreting the stratified findings with caution. Finally, we employed a cross-sectional design. Capturing psychiatric hospitalizations over time with a longitudinal design was beyond the scope of the present study but is an important direction for future research.

Despite these limitations, this study has several strengths. A key strength is the use of U.S. national Medicare claims data, which provide an objective and standardized measure of inpatient psychiatric hospitalization. ⁷⁹ Unlike self-reported data, claims data captures data for individuals who are unable to participate in survey assessments, minimizes recall bias, and ensures consistency in identifying service use. In addition, this dataset contains a large, U.S. national sample of Autistic older adults; therefore, the findings generalize to the broader U.S. population of fee-for-service Medicare-enrolled Autistic older adults rather than being limited to a single region or state. The large sample size further enabled us to match Autistic older adults with population-based comparison groups on demographic and clinical characteristics. Matching allows for robust comparisons between cohorts, allowing us to attribute differences more confidently to the distinguishing characteristic (i.e., Autism diagnosis).