2.1 Study Design and Setting

We undertook a retrospective, population-based, incident cohort study of children and adolescents aged 0–19 years old diagnosed with ASD to estimate total cumulative health care costs to the health care system, using data from Ontario, Canada’s most populous province (2019 population 14.6 million).

2.2 Data

We employed health records obtained through the administration of Ontario’s health insurance system and made available at ICES, an independent, non-profit research institute in Toronto, Ontario. The ICES data repository contains individual-level linkable and longitudinal data on most publicly funded health care services for all legal residents of Ontario. The following health services databases were used in this analysis: Discharge Abstract Database, Ontario Mental Health Reporting System, Continuing Care Reporting System, National Rehabilitation Reporting System, National Ambulatory Care Reporting System, Ontario Health Insurance Plan claims database, Ontario Drug Benefit Program database, and Home Care Database. In addition, the Registered Persons Database, a population-based registry, which contains data on individuals who contacted the health care system, was used to obtain patients’ date of birth, sex, eligibility for universal health care and status changes, and postal code of residence, while the 2011 and 2016 Census data were employed to obtain neighbourhood-level data, such as household income expressed in quintiles, and level of rurality of residence. A full description of each database can be found in Appendix Table A1 (see electronic supplementary material [ESM]). All databases were linked using unique encoded identifiers and analysed at ICES, in compliance with Ontario privacy legislation. The use of these data was authorised under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a Research Ethics Board.

2.3 Patient Population

All children and adolescents (ages 0–19 years old) with a valid health card number, residing in Ontario and diagnosed with ASD were included in the analysis. This age range was chosen based on a meta-analysis, which found a mean age at diagnosis of 60.48 months (i.e., 5 years) (95% CI 50.12–70.83) with a range of 30.90–234.57 months (i.e., 2.6–19.5 years old) [6]. To identify all incident cases of children and adolescents with ASD, a previously validated algorithm [7] was employed, where individuals were selected into the cohort if they had an ASD diagnostic code (ICD-9 code 299.x/ICD-10 code F84.x [autism, atypical autism, Asperger’s syndrome, other pervasive developmental disorders, pervasive developmental disorders unspecified], any diagnosis) for a single hospital discharge, emergency department visit or outpatient surgery, or three ASD physician billing codes (ICD-9 code 299 – child psychoses) within 3 years since birth. To ensure an incident cohort was obtained, all health records for each individual since database inception or date of birth, whichever came first when looking back, were examined. Incident cases were estimated from 2010 to 2019, for the total sample and by sex.

Incident cases of children and adolescents with ASD were characterised in terms of their socio-demographic characteristics—sex, age, neighbourhood income quintile and rural residence—and presence of chronic physical, mental and behavioural health conditions and intellectual disabilities common among children and adolescents with and without ASD—asthma, cancer, diabetes mellitus, ADHD, mood and/or anxiety disorders, learning and developmental disorders, Down syndrome, and Prader-Willi syndrome—ascertained either through disease registries (the Pediatric Oncology Group of Ontario Networked Information System and the Ontario Cancer Registry), validated algorithms or algorithms defined elsewhere [8,9,10,11,12,13].

2.4 Estimation of Observation Time

The number of days alive (and eligible for public health insurance) were estimated from diagnosis until the end of the observation period (i.e., December 31, 2020) for each child/adolescent. The sample was then divided into two groups: (i) censored (i.e., those who were alive during the entire observation period and whose death was not observed) and (ii) non-censored (i.e., those who died, moved out of the province, or who lost eligibility for public health insurance before the end of the observation period and for whom death or end of observation window was observed). See Appendix Table A3 in the ESM for the number of censored and non-censored groups of patients each year.

2.5 Estimation of Health Care Costs Incurred by the Public Third-Party Payers

A costing algorithm was used to estimate total direct patient-level health care costs of children and adolescents with ASD borne by the public third-party payers, the Ontario Ministries of Health and Long-term Care [14]. The algorithm’s costing methodology uses a bottom-up/micro-costing approach to cost services at the patient level, which identifies individual episodes of care or utilisation in the health care system and respective prices, or costs paid. Given Ontario’s public health insurance system, providers in a private marketplace rarely set prices; therefore, costs or amounts paid by the third-party payer were used. Where individual unit costs were not available (e.g., long-term care), a top-down approach, which allocates corporate aggregate costs to individual visits or cases/episodes of care, was employed. For hospital/institution-based care, such as hospitalisations and emergency departments, a measure of utilisation (which accounts for resource intensity using appropriate weights) was multiplied by a unit cost (e.g., cost per standard hospital stay, cost per Comprehensive Ambulatory Classification System weighted case). For other services, such as physician visits and outpatient prescription drugs, utilisation (i.e., number of visits, units) was multiplied by a unit cost (i.e., fee paid) obtained from Ontario Health Insurance Plan and Ontario Drug Benefit Program claims data, respectively, while for home care, utilisation was multiplied by a unit cost (i.e., cost per visit) obtained from the Ontario Ministry of Health [14]. Costs captured by the algorithm account for over 90% of all government-paid health care services, given the availability of relevant databases at ICES, and include costs directly related to ASD as well as other non-ASD related costs [14]. Further details on the costing methodology can be found elsewhere [14]. Costs were categorised into several health service categories—psychiatric hospitalisations, medical hospitalisations, other hospital/institution-based care (i.e., complex continuing care, inpatient rehabilitation and long-term care), hospital outpatient clinic visits, emergency department visits, other ambulatory care (i.e., same-day surgery, dialysis clinic visits and cancer clinic visits), physician services, outpatient prescription drug covered under the public provincial drug plan and home care—and reported in 2021 Canadian dollars (CAD, where 1 CAD = 0.7978 USD in 2021).

2.6 Estimation of Cumulative Mean Health Care Costs Incurred by the Public Third-Party Payers

One way to estimate mean cumulative health care costs when censoring is present is to reweight each case so that it represents not only itself but also some number of incomplete/censored cases [15]. Several estimators have been proposed to estimate cumulative mean health care costs [15]. Bang and Tsiatis proposed a consistent estimator based on the inverse probability weighting technique, which can be used to estimate cumulative costs [16]. The total cost of children and adolescents who died or were lost to attrition and those with complete cost information during the analysis period was weighted by one over the Kaplan-Meier (i.e., 1/KM) survival probability estimator, with reverse censoring (where deaths/attritors are denoted as 0 and censored cases as 1). The weighted costs were then summed and divided by the total study sample size to determine the mean total cost estimate in the presence of censoring. However, when the patient’s cost history is available, the Bang and Tsiatis estimator is not efficient since it does not use the cost information from censored observations. A more efficient (but also consistent) estimator is the one proposed by Zhao and Tian [17]. Given that we are estimating costs of children and adolescents, who typically do not have a long history of health care costs, and among whom death is rare, the estimator proposed by Zhao and Tian did not seem appropriate. Therefore, cumulative health care costs from diagnosis to death or end of observation period were estimated using the Bang and Tsiatis estimator, which employed data on observation time and health care costs incurred by the public third-party payers. Total mean cumulative health care costs for the first year after diagnosis, including date of diagnosis (i.e., 1-year costs), and respective 95% confidence intervals (CIs) were estimated for the full sample as well as by health service, sex and age group at diagnosis. In addition, total mean cumulative health care costs, undiscounted and discounted at 1.5% (in line with the Canadian Agency for Drugs and Technologies in Health recommendations [18]), were estimated for the 5- and 10-year periods, for the full sample, by sex and by age group at diagnosis. As a sensitivity analysis, we also estimated cumulative costs using the Zhao and Tian estimator as well as cumulative costs for the fully observed observations to assess the robustness of our findings.