AI Is Checking Notes and Raising Standards in Paediatric Care

Advertisement feature produced by Nature Research Custom Media and CHOC. 

A child unable to describe their pain, a parent filling in the gaps — paediatric health care often relies on narratives rather than hard data. Historically, handwritten clinical notes were used to document evolving symptoms and behaviours in young patients. Now, researchers are using AI and advanced statistical methods to analyse these complex records, extracting patterns and indicators that can improve clinical decision-making.

“AI will never replace clinicians, no matter how advanced it becomes,” says Louis Ehwerhemuepha, director of computational research at Children’s Hospital of Orange County (CHOC). “But a clinician equipped with insights from an AI or statistical model will achieve better outcomes than one who overlooks these tools.”

….

[M]aking sense of clinical notes and other forms of unstructured data — information lacking the organization and formatting that guides analysis — requires advanced computational techniques. While presenting his initial COVID findings at the nearby University of California, Irvine, Ehwerhemuepha encountered Annie Qu, a statistician whose expertise in natural language processing (NLP) spans biomedical studies, public health and beyond. Intrigued by CHOC’s vast trove of clinical notes, particularly for patients with mental health conditions, Qu was up for the challenge.

“Using NLP and large language models (LLMs) is especially valuable for mental health data,” says Qu. “Unlike physical conditions, where lab tests can give objective results such as blood pressure or glucose levels, mental health is much harder to measure — there’s no test for why someone feels depressed.”

Ehwerhemuepha and Qu co-founded an NLP lab at CHOC. Their first task was to understand the COVID-19 pandemic’s impact on children’s mental health. Using dynamic topic modelling, an NLP technique, they analysed psychiatric notes, summarizing and tracking mental health topics across four key periods — before lockdowns, before vaccine release, before schools reopened, and after reopening.[3] This approach used the longitudinal clinical note data, enabling psychiatrists and researchers to identify persistent mental health issues such as depression, anxiety and suicidal intention over time.

Read the full article in Nature Research.