Predicting outcomes for autistic children
The aim of this project is to develop models that make personalized predictions of outcomes in adolescence and adulthood for an autistic child and test their accuracy in a number of different settings.
About the study
Understanding what the future may hold is important to set expectations, identify where support is needed and tailor interventions. There have been numerous longitudinal studies, following autistic children into adolescence and adulthood, however it is difficult to translate information on risk and resilience factors into a description of potential outcomes for a particular child. To make this information available to autistic people, their families, and the professionals they work with, we will develop prediction models which can be integrated into clinical practice. Models will be developed from the largest longitudinal follow up dataset of autistic children, formed by a collaboration of existing studies.
To ensure that models are relevant to clinical practice and are developed with the best methodology, this project is being overseen by Professor Andrew Pickles from the Department of Biostatistics and Health informatics, Kings College, London and Professor Emily Simonoff from the Department of Child and Adolescent Psychiatry, Kings College London.
The strength of combining studies
We are inviting all longitudinal studies with relevant data, identified via a systematic review (PROSPERO CRD42021286449), to join our collaboration. By using all data available globally, we will develop models that can be applied in clinical practice across the world. Comparing model performance across different studies will allow us to see whether our models will be applicable in different populations. The large sample size from the combined dataset, will allow for more precise model estimation, leading to more accurate predictions.
What outcomes are we looking at?
In adulthood we will be predicting outcomes relating to mental health, quality of life, employment and education, living situation and friendships. In adolescence, our first target outcome is behavioral and emotional problems, measured using the Strengths and Difficulties Questionnaire (SDQ) and the Child Behavior Checklist (CBCL), two measures commonly used in clinical practice.
What predictive factors will we using to model outcomes?
As predictor variables in the model, we will be using data that is routinely collected in clinical practice. This will make any models developed readily applicable within existing services. Alongside demographic data we will use parent report measures and assessments made by clinicians. This will include assessments of IQ, Adaptive functioning, autism traits, and childhood behavioral and emotional problems.
Accounting for uncertainty
Even after taking into account childhood variables there may be considerable variability in outcomes. An analysis of the Early Diagnosis Cohort we have published found that only very uncertain predictions are possible for mental health and quality of life, areas of great importance to autistic people. Alongside any predictions, we will present descriptions of their uncertainty, showing the potentially wide, range of likely outcomes. We have written a blog exploring some of the implications of this uncertainty.
Information for Collaborators
What will collaborating involve?
Collaborating studies will share, anonymous, individual level data to be used to develop prediction models. Any data processing and cleaning will be done by the Predicting Outcomes for Autistic Children study team. Collaborators will be invited to contribute to the design and conduct of the project including to developing a pre-registered analysis. Any publications will be co-authored by all collaborators. Beyond this project, the pooled dataset will be a valuable resource and there may be opportunities for future analysis lead by different collaborating groups.
Eligibility criteria for collaborating studies
Studies are eligible to participate if they have:
- Over 50 autistic participants with
- At least one assessment made prior to age 14.
- Follow up data collected two years or more after baseline assessment
To be eligible for the modelling of outcomes in adulthood they must include:
- A measure of IQ prior to age 14
- Outcomes measured at age 18 or older including at least one of employment and education, living situation, friendships, mental health or quality of life. IQ or adaptive functioning.
To be eligible for the modelling of behavioural and emotional problems in adolescence they must include:
- Used either the SDQ or CBCL to assess behaviour and emotional problems
- Have one assessment prior to age 14 and one after age 14.
Getting in contact
If you are interested in joining the collaboration please contact Gordon Forbes (email@example.com)
This project is funded by the National Institute of Health and Care Research NIHR301522