Mortality associated with the perioperative management of patients with proximal femur fractures: A machine learning prediction model.

BACKGROUND

Proximal femur fractures (PFF) are common in the elderly, representing a significant public health issue. This study aims to define the epidemiology and morbidity of PFF and identify factors associated with 90-day mortality in patients with osteoporotic PFF.

METHODS

We conducted a retrospective, bicentric, observational study in Marseille from November 2018 to June 2023, including patients operated for osteoporotic PFF. Clinical, biological, therapeutic, and socio-economic data were collected to analyse their influence on 90-day mortality and construct a mortality predictive model using a neural network.

RESULTS

During the study period, 2442 patients were included, the mean age at diagnosis being 81 (13.7) years. The 279 (11.4%)non-survivors at 90 days were older (87.2 (0.8) vs. 79.8 (0.3) years; p < 0.0001), predominantly males (36.6% vs. 27.8%; p = 0.003), and had higher Charlson score (1[0-3] vs. 1[0-2]; p < 0.0001) and ASA score (3[3-3] vs. 3[2-3]; p < 0.0001). No significant differences were found in the use of cement, type of anaesthesia, and socio-economic level. A neural network predictive model for 90-day mortality included age, gender, ASA score, perioperative confusion, and haemoglobin, creatinine, and albumin at inclusion. The model performed with an area under the receiving operating characteristic curve of 0.91 [0.91-0.92], sensitivity of 50.5%, specificity of 75.0%, positive predictive value of 20.6%, and negative predictive value of 92.1% on the test set.

CONCLUSION

Our results provide interesting elements to optimize the perioperative management of these patients. The future perspectives include validating the predictive model on external cohorts and integrating new variables to improve its accuracy.