Table 1.
Abbreviations used in the text.
Fig 1.
The flow diagram describes the study cohort utilized for Aim 1 and Aim 2 according to renal function trajectories.
After the exclusion of records not fulfilling the inclusion and exclusion criteria, we utilized 34,579 records for Aim 1 (described in the box with a light green outline) and 1,362 records for Aim 2 (patients with AKI during the first 24 hours of hospitalization described in the light blue box).
Fig 2.
illustrates the execution day and the number of patients used for each developed ML model.
Model AKI_HOSP_T0 predicts whether an admitted patient will develop an AKI during their stay. This model is executed at admission time and is developed using the entire dataset (N = 34,579). Models AKD_T1-4 predict whether patients who experience AKI within the first 24 hours of hospitalization will progress to AKD. AKD_T1 is based on patients still hospitalized on day 1 (24 hours – N = 1,362). AKD_T2 is based on patients still hospitalized on day 2 (48 hours – N = 1,263). AKD_T3 is based on patients still hospitalized on day 3 (72 hours – N = 1,157). AKD_T4 is based on patients still hospitalized on day 4 (96 hours – N = 1,053). Fig legend: AKI: acute kidney injury; NAKI: non-acute kidney injury; AKD: acute kidney disease; OUT: patients discharged or dead.
Table 2.
Demographic, administrative, and laboratory data of the entire study cohort (utilized for Aim 1).
Table 3.
Diagnostic accuracy calculated on the external validation set of the ML model predicting the risk of developing an AKI during the hospitalization based on data available before and at hospital admission (model: AKI_HOSP_T0). AKI: acute Kidney Injury, ROC: Receiving Operator Curve, PRC: Precision-Recall Curve, AUC: Area Under the Curve, PPV: Positive Predictive Value, NPV: Negative Predictive Value.
Fig 3.
Receiving Operator Curve (ROC) of the ML model predicting an AKI during the entire hospitalization (AKI_HOSP_T0).
The predictions are made at the time of patient admission.
Fig 4.
SHAP analyses of the ML model running at admission and predicting an AKI during the entire hospitalization (AKI_HOSP_T0).
Variables are listed in descending order of importance, with the most important variable at the top. The relative importance of each variable is indicated in parentheses. Fig legend: sCr: serum creatinine; [lab hist (1y back)] sCr: maximum value of serum creatinine in the time window of 365 to 7 days prior to hospitalization; [lab hist (1y back)] sCr Last: last value of serum creatinine in the time window of 365 to 7 days prior to hospitalization.
Table 4.
Diagnostic accuracy of the ML models in predicting AKI to AKD transition among subjects experiencing AKI during the first day of hospitalization calculated on the external validation set. AKI: acute Kidney Injury, ROC: Receiving Operator Curve, PRC: Precision-Recall Curve, AUC: Area Under the Curve, PPV: Positive Predictive Value, NPV: Negative Predictive Value.
Fig 5.
Receiving Operator Curve (ROC) of the ML models to predict the risk of AKD transition at day 1, 2, 3, and 4 (AKD_T1 through T4).
Fig 6.
Represents a single case of a patient developing AKI during the first 24 hours of a hospital stay.
Fig 7.
represents the patient-level SHAP analyses of the ML model running at admission and predicting AKI during the hospitalization
(A) and the risk of AKD transition at day 1, 2, 3, and 4 (B-E). Model A predicted AKI (PRED = 1). Model B predicted no transition to AKD (PRED = 0). Models C-E predicted the transition to AKD (PRED = 1). Color legend: In red are the factors positively associated with the risk of the outcome of interest; in green are the factors negatively associated with the outcome of interest. Fig legend: sCr: serum creatinine; ALATGPT NTH: alanine aminotransferase; STD: Standard deviation. Between the square brackets, the time point considered is indicated. For example, [lab hist (1y back)]sCr maximum refers to the maximum serum creatinine value in the 365 to 7 days before hospitalization. Delta signifies a change that occurs during hospitalization. The time points used to calculate the delta are indicated in round brackets. For example, [lab 0–24] sCr Delta (2nd-1st) refers to the change in serum creatinine (micromol/l) between day 2 and day 1.