Imagine living in a world where everyone says the sky is green. You’ve always seen it as blue, but no one believes you—they dismiss your confusion, call you ‘oversensitive’ or accuse you of not trying hard to fit in, making you doubt yourself overtime. This is what undiagnosed autism could feel like: a lifelong disconnection where your innate way of experiencing the world is misunderstood, even to yourself. Many autistic people navigate life without knowing why social interactions feel so frustrating and exhausting, just as a colour-blind person not realising that the same colour to them is in fact two different colours to others until they are handed a pair of corrective lenses.
💭So What is Autism?
Autism Spectrum Disorder (ASD), often known as autism, is a lifelong brain development condition marked by challenges in social communication and interaction, as well as limited and repetitive patterns of behaviour.
Still unclear? Watch this one-minute video about autism!!!🧐
🥺Challenges for Autistic People
- ⬆️mental health difficulties: attention-deficit hyperactivity disorder (ADHD), anxiety, sleep-wake problems etc. are more common in autistic people
- ⬆️physical health problems: chronic pain, heart disease
- ⬆️other disadvantages: discrimination, social exclusion, unemployment
🤓Fun Facts:
✅~1.1% of adults in the UK are on the autism spectrum−that’s about 1 in every 100 people. But the actual number is estimated to be much higher → under-diagnoses (Autism is more common than you think!)
✅Autism can be spotted in children as early as their first year of life from clinical signs, but on average children are only diagnosed until 3 – 4 years of age. → Two-year delayed diagnosis
✅59% – 72% (435,700 – 1,197,300) of autistic people, 0.77% – 2.12% of the English population may be undiagnosed → under-diagnoses again!!!
✅Only 4.9% of referrals got their first appointment in the recommended 13 weeks. → Long waiting time for diagnosis
Under-diagnosis and Early Detection of Autism
Consider the extra difficulties that autistic people have to face, and the fact that support will be inaccessible if they are undiagnosed, life will be a LOT harder!!! Therefore, under-diagnosis is a big problem!
The two-year delayed diagnosis is a missed opportunity for treatment because the younger you are, the better the outcome of early behavioural interventions. Therefore, early detection of autism is important!
Standard autism screening tools are time-consuming → imaging waiting for 13 weeks before doing a bunch of questionnaires and finally seeing an autism specialist. Therefore, an efficient early detection technique is desired!
Machine Learning is a SOLUTION
Machine learning is a technique that uses data to train a model to recognise patterns or predict outcomes. It has been widely used in healthcare to detect diseases, refine screening tools and improve diagnoses accuracies. In autism research, machine learning models have been used to predict whether a person has autism or not with behavioural and developmental data. Although a few studies found promising results, more evidence is needed, especially from longitudinal data.
🔎Our Research Aim
To test the accuracy of machine learning models in predicting autism outcomes using early developmental data, and to see if using developmental information from different developmental periods (e.g., Age 3 vs. Age 7) affect accuracy.
📋What did we do?
- Analysed data of 18827 children from a British birth cohort (Millennium Cohort Study)
- Chose autism-related factors such as items in the Strength and Difficulties Questionaire (SDQ) from different time points (e.g., age 3, 5, 7, etc.) to train the model for predicting whether the children will be diagnosed with autism by a professional.
- Parent-reported autism diagnosis was used an indication of autism for the model
- Four different machine learning algorithms were used and compared to choose the best performing model.
- For each algorithm, different models with factors from different time points were built. For example, a model contain only factors from one time point and another model have factors from two different time points.
📉What did we find?
(Figure above shows the results of the best performing algorithm.)
1️⃣ Using data from 9 months old alone is not accurate in predicting autism (prediction is only slightly above chance)
2️⃣Adding data from more time points (until 11 years old) increases accuracies of predictions → increasing trend in all metrics!!!
3️⃣The most accurate model includes factors from all time points and has an accuracy of 89.4%.
4️⃣Factors such as emotional, concentration and behavioural difficulties, sex, and frequency of non club/class physical activities are the most important factors amongst all in predicting autism.
👀Why does this matter?
- Early detection of autism is possible! Using only developmental data before 3 years old has an accuracy of 70% in prediction. This can help reduce the number of undiagnosed people in the future generations if this technique is applied in public health services. High risk population can be identified and can be arranged for autism screening. MOST IMPORTANTLY, it reduces suffering of autistic people as support could be reached early and treatment and interventions will be more effective.
- Efficient and cost-effective technique. Important predictors can be grouped into a questionnaire for primary care staffs to record during daily visits, or even in online healthcare systems such as the NHS app. This saves health care resources and time spent by both parents filling in the standard screening questionnaires and healthcare professionals reviewing them.
- A more systematic approach in identifying high risk people: solves the problem of too many referrals and insufficient availability for autism diagnosis appointments.
- Application in clinical practice often requires an accuracy of at least 90%, which could be achieved with more research. (OUR result is promising, nearly there!)
References
O’Nions, E., Petersen, I., Buckman, J. E., Charlton, R., Cooper, C., Corbett, A., … & Stott, J. (2023). Autism in England: assessing underdiagnosis in a population-based cohort study of prospectively collected primary care data. The Lancet Regional Health–Europe, 29.
NICE CKS (2024, June). Autism in adults: How common is it?. NICE Clinical Knowledge Summaries Site. https://cks.nice.org.uk/topics/autism-in-adults/background-information/prevalence/.
NHS digital. (2024, November 14). Autism Statistics, October 2023 to September 2024.https://digital.nhs.uk/data-and-information/publications/statistical/autism-statistics/october-2023-to-september-2024.
Word Count: 952