Towards semi-personalized (cluster) medicine with the optimal osteoporotic fracture prediction: artificial intelligence- or current human- based models? The OsteoLaus Project
Description: Osteoporosis is a major public health issue expected to become even more important in the next decades with population aging, which is specifically truer for Switzerland. Clinical consequences of osteoporosis are fragility fractures. Hip fractures are the most serious, associated with a significant increase in death and institutionalization. Although much effort has been made in recent years to improve fracture risk detection and patient management, we are still missing about 30-50% of atraumatic fractures, and are lacking a more precise individual information to better manage patients.OsteoLaus, a population-based cohort of Swiss postmenopausal women, is by far the only Swiss study with comprehensive data on bone health. Its continuation up to a follow-up period of ten years is of major importance for the validation of the calibration (predicted versus observed) of the current used fracture prediction tool, FRAX, in the Swiss clinics. Moreover, the extension of the follow-up period enables us to study occurrences and associations with a greater power.Using two large population based cohort followed prospectively (OsteoLaus and Rotterdam cohorts totalizing more than 8,300 individuals), we aim at developing a new fracture model based on artificial intelligence (Machine learning - ML) for fracture prediction. This model will include the best fracture predicting variables (chosen from more than 600 parameters available in both studies) with the hypothesis that it will upgrade fracture prediction to a more specific and complex-but-sophisticated level. The ML approach we are suggesting to use will `teach` us which individual`s risk factors profile corresponds to a certain risk profile. In this way, individuals will be clustered based on their risk profile as defined by their risk factors; and therefore be managed/treated based on their profile. Given the economic and social burden of fractures and their incidence, it is a need in the field to provide information on the cost effectiveness (CE) of any suggestive method/tool to be included in the osteoporosis management. Therefore, we will compare the newest AI approach with the most currently used fracture model (FRAX) in terms of clinical costs, personal costs and cost effectiveness performance. In this way, we can provide a full information and decision-making strategy for the actual implementation of our results in clinical practice. Finally, hypothesizing that the ML derived model would be the most cost effective approach, we will categorize individuals in different clusters based on their risk profile as defined by their risk factors. For instance, we claim that two women with an apparent equal level of fracture risk may not be treated similarly as their treatment should depend on the causes of their risk (e.g. normal bone quantity but an impaired bone structure and unhealthy lifestyle or impaired bone parameters but with a healthy lifestyle – both give the same risk of fracture). Treating our patients based on where the causes for their increased fracture risk lie, would bring osteoporosis management a step closer to a more personalized approach in its management.This project, the first of its kind in the osteoporosis field, will enable a strong interdisciplinary collaboration between doctors, physicists, and statisticians directed towards a major public health problem. It will elaborate on an advanced direction of fracture risk assessment and holds promise to improve the management of osteoporosis in general.
Amount: CHF 904,000
Funding type: research
Funder scheme: Project funding
Start date: 2019-10-01
End date: 2024-09-30
Investigator: Didier, Hans (lead)
Institutions: University of Lausanne, University of Lausanne – LA
Subfield: Orthopedics and Sports Medicine
Field: Medicine
Domain: Health Sciences
Funded outputs count: 10
Funder award ID: 188886
OpenAlex ID: G955999310
