RESUMO
Intradialytic hypotension (IDH) is a common complication in patients undergoing hemodialysis therapy. No consensus on the definition of intradialytic hypotension has been established so far. As a result, coherent and consistent evaluation of its effects and causes is difficult. Some studies have highlighted existing correlations between certain definitions of IDH and the risk of mortality for the patients. This work is mainly focused on these definitions. Our aim is to understand if different IDH definitions, all correlated with increased mortality risk, catch the same onset mechanisms or dynamics. To check whether the dynamics captured by these definitions are similar, we performed analyses of the incidence, of the IDH event onset timing, and checked whether there were similarities between the definitions in those aspects. We evaluated how these definitions overlap with each other and we evaluated which common factors could allow identifying patients at risk of IDH at the beginning of a dialysis session. The definitions of IDH we analyzed through statistical and machine learning approaches, showed a variable incidence on the HD sessions and had different onset time. We found that the set of parameters relevant for the prediction of the IDH was not always the same for the definitions considered. However, it can be observed that some predictors, such as the presence of comorbidities such as diabetes or heart disease, and a low pre-dialysis diastolic blood pressure, have shown universal relevance in highlighting an increased risk of IDH during the treatment. Among those parameters, the one that showed a major importance is the diabetes status of the patients. Diabetes or heart disease presence are permanent risk factors pointing out an increased IDH risk during the treatments, while, pre-dialysis diastolic blood pressure is instead a parameter that can change at every session and should be used to evaluate the specific risk to develop IDH for each session. The identified parameters could be used in the future to train more complex prediction models.