6577 No child left behind – AI’s role in equalising healthcare

Objectives 1.2 million paediatric appointments were missed in England between 2018–2019, costing the NHS £150 million and lengthening national paediatric waiting lists. Missed appointments disproportionately affect children from lower socioeconomic backgrounds, exacerbating existing health inequalities. This pilot study, coordinated by the Children's Hospitals Alliance, aimed to reduce was not brought (WNB) rates amongst paediatric patients, using an artificial intelligence (AI) tool to target vulnerable patients predicted to be at high risk of missing appointments. Methods Utilising longitudinal data from linked electronic patient records (EPR) and health inequality metrics, a design thinking and validation process identified 46 variables, which, through an algorithm, predict the likelihood of a child missing their appointment. Using historical EPR data, the AI tool was trained until population WNB rates could be predicted with 80% accuracy (20% allowed for unpredictable events). Once trained, this tool could predict an individual child's risk of missing an appointment. Over a two-month period, live data were submitted every 2–4 weeks to the AI tool, which, within 1–2 hours of submission, assigned a risk score for every child's appointment in the upcoming 4–8 weeks. Administration-led telephone calls were made 2–10 days in advance of community paediatric and rheumatology appointments to patients with an 80%+ WNB risk score. Calls where there was a concern or subsequent non-attendance allowed staff, patients, and their family/carers to identify attendance barriers. Serious concerns were escalated to senior clinicians. Results The intervention reduced WNB rates from 68% to 23% among high-risk patients, significantly improving access to healthcare for vulnerable populations. Improved patient satisfaction and staff awareness of patient needs were reported. Variable importance data analysis unveiled the drivers of missed appointments, including waiting times and previous attendance behaviours. Despite improved WNB rates, phone calls placed a significant burden on staff time and improved resource allocation is needed. Patients who lived 10–20 miles away from their appointment displayed higher WNB rates. Phone calls alone were not enough to support these patients, who may lack access to transportation. Additionally, those who gave their ethnicity as Slovak or Romany Slovak had unusually high WNB rates, suggesting further targeted interventions are needed. Conclusion The AI-driven intervention significantly reduced WNB rates among high-risk paediatric patients, proving a powerful tool in addressing healthcare disparities. This not only acutely benefitted the children involved, but also helped to alleviate the financial and operational strain on the healthcare system. More comprehensive solutions are, however, needed to fully address systemic health inequalities.