Back
Retrospective observational cohort. We included 2,934 patients hospitalized at the National Institute of Respiratory Diseases in Mexico City, between 2020 and 2022 with confirmed diagnoses of COVID-19 and AKI. The research was approved by the institutional ethics committee (C38-23). Data preprocessing was done with 3 steps: validation, standardization, and integration of AKI features. Clinical variables that are likely to impact AKI mortality were integrated by an experienced nephrologist and relevant variables were selected through the use of an algorithmic method of feature selection for regression using neighborhood components, ensuring that ≥ 70% of the values were met. The data were analyzed using standard statistical methods and the Correlation Matrix was calculated. Statistical significance was established with a value of p <0.05. The construction and grouping of the clusters were carried out using the K-means method and the Matlab® program (MathWorks, version 2022) was used to analyze the data.
A total of 278 clinical variables were collected, 22 were chosen by the principal investigator and only 10 were used for clustering using target variable survival. The variable identified with the greatest weight was the in-hospital length of Stay (LoS). We found a positive Pearson Correlation between in-hospital Renal Replacement Therapy (RRT) and the severity of AKI and a negative association between baseline Glomerular Filtration Rate (GFR) and age, both with moderate scores. A total of 3 SP were identified with different clinical manifestations (n=1269, 894, and 771 patients in clusters 1, 2, and 3, respectively). SP2 was characterized by patients with critical illness, including higher requirement of mechanical ventilation, and use of vasopressor or nephrotoxic drugs, and it was related to less probability of recovery and a long hospital length of stay days. This cluster has an impact on mortality with a greater percentage of deceased patients than SP3 and SP1 (54.4% vs 43.8% vs 25.6%). SP3 was characterized by a higher burden of chronic diseases, more severe forms of AKI, and a higher need for iCn-hospital RRT. SP3 was observed as less vulnerable to risk factors and with better outcomes.
Our approach based on the classification of clinical data from patients with AKI and severe COVID-19 suggests the presence of 3 clusters with clinical relevance and SP2 had worse outcomes in comparison with SP3 and SP2. The use of AI serves as a risk predictor, outcome, and prognosis. The identification of patterns is useful for implementing individualized therapeutic strategies in heterogeneous clinical syndromes toward precision medicine. External validations are required to confirm our observations.