In this week’s round of insights, we discuss three of the latest findings in the application of artificial intelligence (AI) in healthcare.
AI can help identify rare disease trial candidates
A recent study from Cleveland Clinic and Dyania Health demonstrated that a medically trained large language model within electronic health records (EHRs) could rapidly screen patient data to identify candidates for a trial targeting transthyretin amyloid cardiomyopathy (ATTR-CM). ATTR-CM is a progressive heart disease that can lead to stiff heart walls and heart failure.
The AI system reviewed 1,476 patient records in just one week. It evaluated trial eligibility using thousands of criteria and generated clear justifications for each decision. Clinician reviewers confirmed that the system achieved 96.2% overall accuracy as well as 100% accuracy in its explanations. The AI-assisted approach successfully identified 29 potential participants who had not been found through traditional recruitment pathways.
AI may advance precision medicine in skin diseases
AI promises to improve the management of inflammatory skin diseases, including psoriasis, atopic dermatitis, alopecia areata and vitiligo.
A recent review highlighted how AI and machine learning (ML) tools are expanding diagnostic capabilities and enabling personalised treatments. Computer models trained on clinical photography, dermoscopy images, histology slides, and EHR data can help classify a range of dermatologic conditions.
Beyond diagnosis, AI is increasingly used to identify disease subtypes and predict treatment responses. The authors also noted that AI may support drug discovery by identifying potential therapeutic targets.
Machine learning can potentially predict chronic kidney disease
In a recent study analysing clinical and laboratory data from more than 1,000 patients, researchers developed a machine learning (ML) model that can predict chronic kidney disease (CKD).
The model identified key predictors of CKD risk, including haemoglobin levels, blood urea, sodium concentration, red blood cell count, potassium levels and hypertension status. One of the algorithms of the model achieved 98% accuracy and demonstrated high reliability in distinguishing between CKD and non-CKD cases.
Early-stage CKD often presents with minimal symptoms; predictive tools could have a major role in supporting an early diagnosis and intervention.
For more industry insights visit elion.nz/insights/
Want to get in touch? Contact us today at https://elion.nz/get-in-touch/
