This AI Could Help Diagnose ADHD Sooner

Early ADHD diagnoses could help cut crime and drastically improve lives, but a cash-starved NHS is struggling to keep up

As a child, Casey*, 38, always found it hard to concentrate. Her old school reports said the same thing. “Casey manages to talk through the entire class. She somehow gets her work done, but she distracts everybody else,” read one primary school report she recently dug out the back of her cupboard.

These behaviours continued into adulthood. She was impulsive and bought expensive items on a whim: clothes she barely wore or a – £5000 camper van she drove three times. Colleagues wondered whether she had dyslexia because of misspelled words in the odd email. She looked into the possibility, but it didn’t seem to fit. At age 36, Casey began to question whether she had undiagnosed attention deficit hyperactivity disorder (ADHD).

Her behaviours seemed to match some of the symptoms she was reading about on the NHS website at the time. “It was like reading an autobiography,” Casey remembers. She visited her GP but was told that no ADHD services for adults existed where she lives. She appealed for help through her MP and wrote to ADHD charities for further assistance.

She was frustrated, but had little choice but to wait. At the time, though, Casey was in a violent relationship with a partner of around two years. He was verbally abusive, manipulative and physically assaulted her, she says. On several occasions, she alleges, the police were called to their property, but he remained in the house.

In December 2019, the police were called to Casey’s home, which ended with Casey being picked up and carried away by the police. As Casey was being carried down the stairs, she lashed out and bit an officer arresting her. She was charged with assault on an emergency officer and went to court in February 2020. Too ashamed to tell her parents and with her partner refusing to go with her, Casey went to court with only her solicitor.

Casey is no longer with her abusive former partner and last September, at the age of 38, was diagnosed with ADHD. She has since undergone treatment, but wonders how things could have been different if it came sooner. “I keep thinking to myself: how have I got to nearly 40 and nobody noticed?” she says. “I wonder if I didn’t have ADHD, would I have dealt with the police coming to my property in the same way? Would I have been more rational? The fact this happened is something that will stay with me forever.”

ADHD affects around three to five per cent of children and two per cent of adults in the UK, but it’s a disorder that remains greatly underdiagnosed. Of the estimated 1.5 million adults with the condition, only 120,000 are formally diagnosed. There are traits that are synonymous with ADHD: hyperactivity or an inability to pay attention are common associations but many are less known, such as excessive talking or little sense of danger.

ADHD also influences criminal behaviour. Research dating back to 2012 has found that people with ADHD are more likely to commit crime than adults without the condition. Around 25 per cent of adult prisoners meet diagnostic criteria for ADHD. For those in the Children and Young People Secure Estate (CYPSE), the prevalence rate is approximately 30 to 40 per cent. In the general population, it’s five per cent.

Susan Young is a clinical and forensic psychologist who, as part of her job, has spent over 20 years exploring the links between ADHD and criminality. Young began her work in the mid-90s at Maudsley Hospital in south London, and helped set up the first clinic for adults with ADHD. Over time, people who had been assessed weren’t showing up for further appointments. She wondered why. They hadn’t forgotten, Young discovered. They were in prison.

In 2016, Young examined the role of ADHD in behavioural disturbances in prison and violent and nonviolent offending. 196 male prisoners from Aberdeen prison completed a Symptom Checklist –
90, a screening questionnaire for psychiatric symptoms. 27 met the screening criteria for ADHD.

Her findings also suggested the vast majority of crimes were reactive. “People with ADHD are not going to be the people who are planning and organising a crime,” she says. “It’s opportunistic offending. It’s people who lose their temper and get involved in a fight in a pub.”

Studies suggest that early intervention and medicinal treatment for ADHD is linked to a 32 per cent drop in offences by men and 41 per cent for women. But in the UK, ADHD resources are painfully thin. GPs are stretched and specialist centres are scarce. Around 21,000 people are awaiting formal diagnosis on waiting lists in the UK.

Typically, assessments are done face-to-face with a GP where diagnosis is made via a series of tests. Physical examinations can be made, and sometimes interviews are conducted with partners, parents and teachers. But ADHD is a complex condition to diagnose. Bipolar disorder, autism or even low blood sugar can mimic ADHD symptoms. ADHD in adulthood also doesn’t always present itself in the same way as it does in children.

Can machine models help? Casey’s ADHD diagnosis was supported by an online screening tool called the Do-It Profiler that helps identify factors that may be related to neurodisability. It’s currently being used in prisons, mostly in facilities across Wales and Scotland, where it has recognised that “around one in three people” coming through the prison system are neurodivergent, says Amanda Kirby, CEO of Do-IT Solutions, the company behind the screening tool.

Young is currently part of a team investigating similar interventions using a computer-based tool called the QBTest. Whereas the Do-It Profiler assesses for a multitude of neurodivergent conditions, the QBTest is ADHD-specific. The QBTest measures core signs of ADHD â hyperactivity, impulsivity and inattention â through a computer-based test that monitors subtle changes in expression and movement through motion tracking. These results are then compared to average data of a person of the same age and gender.

The QBTest aims to speed up the assessment process. It can also save considerable amounts of money, in excess of £80,000 per year, per clinic, they say, but is not yet widely available through the NHS. But experiments with automated assessments are being explored within the NHS by Grigoris Antoniou and Marios Adamou to make assessments cheaper and more widely available. And machine-learning may be the key.

Antoniou is a professor at the department of computer science at the University of Huddersfield, Adamou a consultant psychiatrist from the South West Yorkshire NHS Trust. Three years ago, they started looking at new approaches to assessing adult ADHD, one that combined Antoniou’s background in semantic technologies with Adamou’s experience in the NHS. Together, they devised a decision tool for ADHD through hybrid research, one that uses clinical information and machine-learning. To the pair’s knowledge, this is a world first.

It began by training a prediction model using clinical data of past cases and knowledge from clinical experts. The artificial intelligence system then takes this data to generate three clear cut outcomes based on a screening: has ADHD, does not have ADHD, or consult a medical expert. Their results were published in November 2020, and the test was successful in identifying cases to an accuracy of 95 per cent.

The project is currently going through a clinical trial across south west Yorkshire, and so far it’s working. If widely used, says Adamou, this technology will facilitate roughly double the amount of assessments in the same amount of time and will provide major cost savings for the NHS.

The tool has two main clinical outputs: to organise cases and signpost to potential diagnostic outcomes. These outputs allow allocation of cases to clinical experts more efficiently, and facilitate consistency in the diagnostic approach and outcome.

But there are limitations. The tool is not to be used for automated diagnosis of ADHD, but rather provides a recommendation. When services are used to “cutting corners,” says Adamou, uptake of this technology may be slow. Doctors may also be resistant to change. “Some people have an idea that healthcare should be done by humans for humans,” says Antoniou.

Adamou and Antoniou stress that their machine-learning algorithm is a tool that can increase the productivity of a clinical team, not replace them. “What we’re selling people is a technology that will save money, cut waiting lists and will not make anyone redundant,” says Antoniou.

But improving access to treatment is a moral obligation for Adamou. “It’s seen as acceptable to be waiting three years for an ADHD assessment, but it wouldn’t be for something like dementia,” he says. “It’s not good for any mental health disorder to go untreated. For ADHD, the treatment is available and it’s very good, but people aren’t able to access that because they’re poor. This tool won’t alleviate that stigma or injustice, but it will ameliorate it.”

*Name has been changed

Original at