Back
Metabolic risk factors for the development of cardiovascular disease are extremely common in renal graft recipients. Conditions related to abnormal glucose regulation, dyslipidemia, metabolic syndrome (MS), obesity and bone diseases can interfere negatively in the post-operative outcomes and decrease graft survival. In this set, the weight gain after the transplantation appears to be multifactorial and the creation of a prediction model for this outcome should provide adjustment to correlated variables. Previously different multivariate linear regression models reported female gender, lower pre-transplant weight, live donor, and fewer hospitalizations as independent risk factors for 5% weight gain at 12 months after transplantation. Considering the ease of dealing with a large amount of data and the effectiveness to create predictive models of complex outcomes, such as weight gain post-transplant, the use of machine learning algorithms could help health professionals in developing care protocols that facilitate a better approach to preventing this negative outcome in clinical practice. This study aims to identify variables related to weight gain one year after kidney transplantation using a machine learning approach.
Retrospective cohort study, based on secondary data from 374 kidney transplant patients in a hospital in southern Brazil between January 2006 and July 2013. Socio-demographic, clinical and anthropometric parameters were evaluated. The elastic net algorithm was used for the analysis of machine learning. All experiments were performed in R (version 3.6.3) with the aid of the caret library. All variables with more than 15% missing data were excluded from the analysis. The dataset was divided into training (75%, N = 282) and testing (25%, N = 92) sets to develop the final prediction model.
Weight gain was observed in 72.45% of patients one year after transplantation, with 31.55% experiencing a weight gain ≥10% of their pre-transplant weight. The prediction model for the outcome of percentage weight gain at 12 months included five variables and the most important were female sex (increased risk) and deceased donor (reducing risk). Age, pre-transplant weight and polycystic kidneys as the etiology of kidney disease were of little importance (risk reduction). The accuracy of the model was low, with a weak correlation (Pearson's correlation 0.28; p = 0.01). The mean absolute error (MAE) was 7.25%, signifying a clinically relevant variation.
Although the model's predictive power using a machine learning algorithm was not satisfactory, the study underscores the necessity for individualized, multidisciplinary interventions to prevent weight gain post-transplant, especially for female recipients. In addition, machine learning algorithms can provide versatile and viable tools to create predictive models in kidney transplantation and should be used in future studies.