Associate Professor Lahn Straney is an epidemiologist with expertise in healthcare quality, outcomes measurement, biostatistics and applied analytics. He holds a PhD in Epidemiology and Biostatistics and is an Adjunct Associate Professor at Monash University and Visiting Senior Research Fellow at the University of Leeds.
Lahn has published more than 80 peer-reviewed papers and has worked across research examining healthcare quality, patient outcomes, risk adjustment and the use of data to support better decision-making.
His current work is focused on aged care, where he applies epidemiological and analytical methods to questions of care quality, safety, workforce, medication use, resident experience and service performance. Alongside his academic roles, he is Chief Scientific Officer at Evaran and founder of MOA Benchmarking, where he works with aged care providers to strengthen the use of data for quality improvement, governance and evaluation.
Beyond generative AI: predicting adverse outcomes using existing aged care data
Background: Predictive analytics offers aged care providers an opportunity to identify changing resident risk earlier and to compare outcomes more fairly. However, many current quality indicators are reported without adjustment, or with only limited adjustment for differences between residents. At the same time, discussion of artificial intelligence in aged care is often dominated by large language models, despite established analytical methods offering a more conservative and transparent approach to prediction.
Approach: We created a resident-level analytical dataset linking demographic information, functional status and routinely collected National Aged Care Quality Indicator Program data. Preliminary models were developed for five outcomes across four quality domains: falls, pressure injuries, significant and consecutive unplanned weight loss, and decline in activities of daily living. Logistic regression, random forest and gradient-boosting models were trained on historical resident-quarter data and evaluated against a subsequently observed quarter. The framework was designed to support two related purposes: predicting risk for individual residents and, in future, accounting for differences in resident risk when comparing outcomes between homes.
Preliminary results: The analysis included more than 24,000 residents. Predictive performance varied by outcome. The best models achieved C-statistics of 0.82 for pressure injuries and 0.72 for falls, demonstrating useful discrimination using existing structured data. Performance was more modest for decline in activities of daily living and significant unplanned weight loss, and limited for consecutive weight loss. More complex machine-learning models did not consistently outperform logistic regression. This is important because the simpler model makes the contribution of individual factors visible and open to clinical scrutiny. Previous events, functional status, hospital and emergency department use, incontinence, weight loss and other clinical indicators contributed differently across outcomes.
Implications: These preliminary findings establish that routinely collected aged care data contain meaningful prospective information about resident risk. The approach does not require generative AI, analysis of unstructured notes or additional data collection. Models can be developed using historical data, tested against later observations and then applied prospectively, with their assumptions, contributing factors and performance available for review and challenge.
This provides a practical foundation for governed use of predictive analytics in aged care. Risk estimates can prompt clinical assessment and preventive action without replacing professional judgement. As the breadth, consistency and timeliness of available data improve, the models can be extended, recalibrated and evaluated prospectively.