Treatements structuring for critical care management with application to persistent Acute Kidney Injury

Acute Kidney Injury (AKI), also referred to as Acute Renal Failure (ARF), is defined by a loss of excretory kidney function that approximately affects 50% of patients admitted to intensive care units (ICUs). The mortality rate due to AKI may be higher in elderly patients, in patients with chronic diseases, such as diabetes or heart disease, and in patients requiring dialysis support during their AKI treatment. In some cases, the mortality rate for AKI can be more than 50%. However, with timely and adequate treatment, the mortality rate for AKI can be significantly reduced. Over the past decade, an important distinction has been made between transient and persistent AKI. Although no consensus has been reached on the distinction between the two cases, persistent AKI represents an episode that lasts longer in time, i.e. more than 48 or 72 hours. In fact, several studies proved that a complete and sustained reversal of AKI episodes within this temporal threshold is related to better outcomes for ICU patients. Persistent acute kidney failure can be difficult to predict, as it can be caused by a variety of factors, such as kidney disease, underlying medical conditions, medications, or injury. However, there are some risk factors that may increase the likelihood of developing persistent AKI, such as advanced age, history of kidney disease or heart problems, or use of certain medications. Early identification of persistent AKI could allow the initiation of a dedicated evaluation and management protocol to decrease the risk of further kidney damage and related mortality. Artificial Intelligence has been widely used for AKI onset prediction, and its models could be adopted for persistent AKI prediction, too. This data science thesis aimed to analyze pharmaceutical administration events in intensive care units and extract relevant information to gain insights into patient health and treatment outcomes. Extraction of various drugs used in these events was performed and statistical analysis was conducted, including calculation of drug frequency and administration speed. The clustering of drugs was executed to identify patterns and relationships. Grouping of events into twelve-hour windows was performed to construct a comprehensive three-day pharmaceutical history for each patient and admission in intensive care. A combination of this history with temporal data of the patients’, hematochemical, physical, and comorbidity values were conducted to perform a classification of Persistent AKI (Acute Kidney Injury). The final aim was to determine if the use of certain therapies, specifically with regard to the presence and absence of certain drugs, could be useful in predicting the stage of persistent AKI. The research had the potential to help clinical decision-making and improve patient outcomes in intensive care settings.