Personalised predictions of outcomes in adolescence and adulthood for children diagnosed with Autism Spectrum Disorders: A Systematic review and IPD meta-analysis, evaluation of methodology, and external validation study.

Research Question Based on their individual characteristics and circumstances, what is the range of likely outcomes in adolescence or adulthood for a child with a diagnosis of autism spectrum disorder (ASD)?

Background Autism spectrum disorders (ASD) is a neurodevelopmental condition which effects 700,000 (1%) people in the UK. ASD is highly heterogeneous with varying levels of severity of symptoms and differences in co-occurring conditions.

Mapping out future outcomes is challenging for professionals and parents and currently no personalised prognostic tools exist to aid this task.

Prognostic modelling for children with ASD poses statistical challenges: repeated assessments are required as a child develops; outcomes are measured using continuous measures and it is important to estimate prediction intervals, describing the uncertainty in any predictions. Prediction from repeated measures of prognostic factors is referred to as dynamic prediction.

There has been limited application of methods for dynamic prediction to continuous outcomes, and the calculation of prediction intervals from dynamic prediction models.

Aims We aim to develop prognostic models to predict behavioural and emotional outcomes in adolescence and a broad range of measures summarising different aspects of outcomes in adulthood.

Methods We will consult with autistic adults, parents of autistic children and professionals working with autistic children to ensure that the prognostic tools we develop are able to meet the needs of these groups. We will conduct an individual patient data (IPD) meta-analysis to develop prognostic models.

A systematic review will identify relevant data sets for which we will seek to obtain individual participant data.

In the first instance we will develop prognostic models with prognostic factors measured at a single timepoint, using multiple linear regression. Prediction intervals will be calculated using non-parametric methods.

Methods for dynamic prediction including frequentist and Bayesian mixed effects models and methods for calculating prediction intervals will be evaluated using simulated data. The best methods will be applied to develop further prognostic models.

We will conduct external validation with new data collected from autism services at the South London and Maudsley NHS trust. Timelines for delivery Stakeholder engagement will be ongoing throughout the fellowship. In the first 9 months I will conduct the systematic review and set up the external validation study.

Months 10-24 will focus on methodological aspects and obtaining and cleaning data for the IPD meta-analysis. Prognostic models will be developed in months 24-30.

The final six months of the fellowship will consist of analysis of external validation data, writing up of the thesis and dissemination.

Anticipated impact and dissemination Where prognostic mode we develop have undergone successful external validation we will look to integrate them into routine clinical practice in parallel with the MyHealthE system, an existing digital platform for collecting data on mental health symptoms. Methodology for dynamic prediction with continuous outcomes will have applications across many clinical areas.

Dissemination will include presentations at national and international meetings, publications, and short videos explaining the output of the fellowship. Autistic people and parents of autistic children will be consulted to ensure that communication is appropriate.