Machine learning algorithms for risk factor selection with application to 60-day sepsis morbidity risk for a geriatric hip fracture cohort.

BACKGROUND

Sepsis after hip fracture in elderly people is a risk factor for mortality. The purpose of this study was to screen for risk factors for 60-day sepsis morbidity after hip fracture and to establish a predictive model using various machine learning algorithms.

METHODS

A total of 697 patients who were older than 65 years of age were selected; 590 patients were divided into training and test sets (at a ratio of 7:3), and 107 individuals were used for external validation. Least absolute shrinkage and selection operator (LASSO), the Boruta algorithm, and random survival forest (RSF) were used to screen for risk factors for sepsis incidence. Correlation coefficients and variance inflation factors (VIFs) were calculated to assess multicollinearity. Cox proportional hazards model analysis was used to establish a model. A restricted cubic spline (RCS) was used. The discrimination, calibration, and decision curve analysis (DCA) results were evaluated. The prediction model that was constructed was compared with those of the qSOFA.

RESULTS

There was no multicollinearity between variables (r < 0.6, variance inflation factor (VIF) < 5). Pneumonia, nutritional status, age, diabetes status, hemoglobin level, and BMI were the final variables. The C-indices of the training and test sets and the external validation set were 0.83 (95% CI 0.78 ~ 0.88), 0.78 (95% CI 0.70 ~ 0.85), and 0.76 (95% CI 0.70 ~ 0.83), respectively. The calibration and DCA results indicated that the model had good predictive value. The RCS revealed a linear relationship between age, BMI, hemoglobin and sepsis incidence (p > 0.05). The discrimination ability and clinical applicability of the model were superior to those of the qSOFA.

CONCLUSIONS

This study, which uses various feature selection methods, can effectively construct a 60-day sepsis morbidity prediction model after hip fracture in elderly people and could be superior to the qSOFA.

TRIAL REGISTRATION: Date of registry: 02/08/2023; Trial number: ChiCTR2300074246.