Artificial intelligence and clinical guidelines to find at-risk COPD patients for treatment optimisation

Abstract

<bold>Artificial intelligence and clinical guidelines to find at-risk COPD patients for treatment optimisation</bold> Despite the rising COPD prevalence and burden, many patients remain unidentified, untreated or under-treated, which negatively impacts patients’ lives and healthcare systems. This is partly due to time-consuming and error prone manual review and documentation processes, which takes ~37% of clinicians’ time (bit.ly/4hWABmE). A possible solution is to configure guideline-based AI to help find untreated and under-treated at-risk COPD patients in the electronic health records for clinical review and treatment optimisation. This analysis aimed to evaluate whether AI, configured on GOLD recommendation, could accurately find COPD patients and those at high risk of exacerbation (HRE COPD). Once configured, the AI was applied to a retrospective US dataset of 29,339 ICU patient records, i.e. structured and unstructured data (e.g. clinical notes). The AI flagged COPD and HRE COPD patients. A sample of the patients was then manually reviewed by clinicians (‘gold standard dataset’) to evaluate the AI’s performance, measuring the proportion of accurately/inaccurately classified patients. The AI configuration to final evaluation took 10 weeks. When applied to a retrospective dataset, the AI found ~5 times (462.1%) more COPD patients than ICD codes with 83.5% precision. Moreover, the AI found ~2 times more HRE COPD patients than ICD codes, with 87.9% precision. This analysis provides a leading example of how applying AI can help optimise patient management and reduce healthcare system burden by finding untreated and under-treated at-risk COPD patients with higher performance than ICD searches.

Ähnliche Arbeiten