Disease and mortality trajectories of cognitively able autistic individuals in mid- and later adulthood | BMC Medicine

Study design

Data for this study were based on the UK Biobank, a prospective cohort recruiting over 500,000 individuals aged 40 to 69 years across the UK between 2006 and 2010 [8]. Data on diagnosis and death was identified from national registers in the UK (details in Table S1) [8]. Participation of the study needed voluntarily informed consent and travel to designated locations to complete comprehensive questionnaires and assessments. This process inherently selected individuals with higher cognitive and functional abilities, leading to what has been termed a “healthy volunteer” effect in prior research [9]. Thus, the study population was considered as “cognitively able autistic adults.” Autistic individuals were defined by (1) an inpatient diagnosis of ASD coded by ICD-10 (F84.0/1/5/8/9) or (2) a self-reported “diagnosis of ASD by a professional,” which was collected in 2016 (responded by ~ 50,000 participants) and 2022 (responded by ~ 170,000 participants), respectively. For each autistic individual, we randomly selected up to 10 non-autistic participants individually matched by birth year (continuous), sex (male or female), Townsend deprivation index (continuous), and respondence status of the self-reported ASD question (yes or no). As the participants who responded at follow-up were necessarily alive and on average healthier than the non-respondents [10], the respondence status was involved in the matching procedure (workflow for details, Fig. S1).

This study included three parts: first, we investigated whether ASD patients had a higher mortality compared with their matched non-ASD individuals. The follow-up started at the date of recruitment or the date of response to the diagnostic question, whichever came later. The follow-up ended on the date of death or 30th December 2022, whichever came first. Data on ASD collected in 2022 were not involved in this analysis because of the short follow-up period. Second, we investigated disease trajectories that linked ASD to mortality using inpatient diagnoses, which were first available in 1997. Follow-up ended on the date of death or 30th December 2022, whichever came first. Third, we investigated the association between ASD and risk of multimorbidity, including three main chronic conditions (cardiovascular disease/hypertension [CVD/HTN], type 2 diabetes mellitus/disorders of lipoprotein metabolism [T2D/DLP], and depression/anxiety [DEP/ANX]) identified in the trajectory analysis. The chronic diseases were identified from both the inpatient and primary care registers [11]. Follow-up started at the first date recorded in the medical record system after the 10th birthday or recruitment time, whichever came first. Follow-up ended on death or 30th December 2022, whichever came first.

Diagnoses of medical conditions and mortality

All UK Biobank participants were regularly linked to the National Death Registries of the UK to update the mortality data [8].The participants were considered to be included in the hospital inpatient data since 1997 [8].

Diagnosis of medical conditions and the date of diagnosis in the trajectory analysis were identified from the primary and secondary diagnoses recorded with ICD-10 in the UK Biobank inpatient data (Data-Field 41,270). The ICD-10 codes were used according to an established method considering the clinical or biological similarities, resulting in a total of 470 3-digits codes [12]. Only inpatient diagnosis was included in trajectory analysis because it had generally higher validity than the diagnosis in the primary care settings [13]. Based on findings from the disease trajectory analyses, we were further interested in the patterns of multimorbidity of three main chronic conditions: CVD/HTN, T2D/DLP, and DEP/ANX [14,15,16]. For chronic diseases, diagnoses in the primary care settings have been shown to exhibit comparable or better validity when determine the time of disease onset [11]. Thus, diagnoses of these chronic conditions and dates of diagnoses were identified from the first occurrence data (Category 1712), which integrated information from the primary care data (Category 3000), the hospital inpatient data (Category 2000), death register records (Field 40,001, Field 40,002), and self-reported medical condition (Field 20,002). Diagnostic codes for the three chronic conditions were detailed in Table S2.

Covariates

Covariates for disease and multimorbidity trajectory analyses included birth year (continuous) or age (continuous), sex (male or female), Townsend deprivation index (continuous), and ethnicity/race (white or not-white). Due to the limited number of non-white individuals (< 5% in autistic participants), specific analyses in the disease trajectory included exclusively white individuals, violating the positivity assumption. Thus, we did not adjust for ethnicity/race in disease trajectory analyses. Additional covariates at baseline were described and adjusted for in the survival analysis, including current smoker (yes or no), current drinker (yes or no), body mass index (kg/m²), number of days per week of moderate physical activity with at least 10 min (continuous), diagnosed intellectual disability (ID; ICD-10 codes F70/F79), and age-adjusted Charlson comorbidity index (0, 1, or ≥ 2) [17]. ID was identified from the first occurrence data. Notably, among the 20 individuals diagnosed with ID, six had completed university education. This finding suggests the possibility of misclassification in ID diagnoses, which may reflect the historically common misdiagnosis of ID and autism [18]. As autistic individuals are commonly comorbid with cognitive problems, we additionally performed a variety of analyses to assess the cognitive functioning, including fluid intelligence scores, reaction time, and prospective memory (specific definition detailed in the supplementary methods).

Statistical analysis

Continuous variables were compared with t-test and categorized variables with chi-square test. Cox regression was applied to estimate the hazard ratio (HR) and its confidence interval (CI) of mortality associated with ASD, adjusting for sex, birth year, Townsend deprivation index, respondence status of the self-reported ASD question in 2016, and the covariates as described above.

Disease and mortality trajectory analyses

We evaluated the pattern of progression of medical conditions associated with ASD following an established method [12]. Briefly, we identified (1) diagnoses associated with ASD from 470 medical conditions (of which we selected conditions that occurred in at least 30 autistic individuals [~ 5% of total]) by logistic regression; (2) the sequence of each pair of diagnoses in step 1 by binomial test (> 50%); and (3) the magnitude of associations between each sequential pair in step 2 by logistic regression. Bonferroni corrections for multiple testing were applied in each step. Due to the incomplete records of the diagnoses during childhood in the study population (the earliest record in the inpatient register was of 30 years old), we were unable to determine the exact date of the diagnosis for ASD for Cox regression. Nevertheless, as ASD mostly developed during childhood [19], we a priori assumed that ASD developed before all outcomes in step 2 and measured the associations by odds ratios (ORs) estimated by logistic regression. For comparability, all-cause mortality was also regarded as one condition in this analysis and the associations between other conditions and mortality were also measured by ORs. To enhance interpretability, we categorized the observed disease and mortality trajectories into distinct clusters based on established clinical patterns from previous literature [3, 12, 20].

Multimorbidity trajectory analyses

To evaluate the pattern of progression of multimorbidity associated with ASD, we considered a multistate model that characterized the rate of accumulation of three chronic conditions during the development of multimorbidity and estimate hazard ratios (HRs) for ASD with adjustment for age, sex, and Townsend deprivation index [21]. In this model, states represented all possible combinations of the three conditions (8 states) or death that was the terminal state and could be reached from any other state (Fig. 3a). A state-to-state transition was assumed from a state with fewer conditions to a state with more conditions. Each individual was presumed to have no underlying diseases at the initial state. We assumed that the three chronic conditions were persistent throughout life once diagnosed. For two or more events happening on the same day, the transition would be counted in multiple transitions to the next states, while the individual would move to a state that matched the individual’s condition after transition. According to the results from disease trajectory analyses and prior knowledge, we consider the following conditions in the multimorbidity: depression, anxiety, CVD/HTN (including ischemic heart disease and hypertension), type 2 diabetes mellitus, disorders of lipoprotein metabolism, and cancer. Cancer was excluded from the multistate model due to lack of significant association with ASD (Table S3). To ensure adequate sample sizes for each disease state in the multistate model, we consolidated other conditions according to the ICD-10 classification into three states: depression/anxiety, CVD/HTN, and diabetes mellitus/disorders of lipoprotein metabolism. For example, if individual A was diagnosed with CVD/HTN and T2D/DLP at day 10, this transition would contribute to both “none to CVD/HTN” and “none to T2D/DLP.” At the same time, individual A would move from the state “none” to “CVD/HTN + T2D/DLP” at day 10; thus, individual A would not contribute to any transition from “CVD/HTN to CVD/HTN + T2D/DLP” and “T2D to CVD/HTN + T2D/DLP.” Individual event-time models were established for each transition, with the time parameter representing the duration from one diagnosis to the subsequent diagnosis or death [21,22,23]. We employed the “flexsurv” package in the R software environment for implementation.

To account for potential misclassification of ASD due to self-reported diagnosis, we additionally restricted to only inpatient diagnosis of ASD for sensitivity analysis. Given evidence that autistic individuals were disproportionately affected by COVID-19 in terms of mental and physical health outcomes, we conducted sensitivity analyses restricting the follow-up period to December 31, 2019 (pre-pandemic) to evaluate potential temporal effects. Sensitivity analyses of multimorbidity trajectory analyses were also performed to remove self-reported diagnoses to exclude the effect of self-reported time errors on diseases progression analyses. Statistical analyses were conducted using R (version 4.2.2) during the time period October 2023 to March 2024.