In intensive care units (ICUs), Acute Kidney Injury (AKI) is a frequent, typically multifactorial complication that can substantially alter the clinical course and prognosis of critically ill patients. In its early phases, however, AKI can be difficult to distinguish from a physiological response to clinical stress. A transient reduction in kidney function may in fact represent an adaptive and reversible phenomenon, making it harder to immediately recognize true structural kidney damage (Kellum, 2015).
This initial uncertainty can delay the identification of AKI, particularly in cases where the episode does not resolve quickly. For clinicians working in intensive care, the challenge is not only to recognise AKI, but also to understand its duration. This means that outcomes are determined not only by how severe AKI is at onset, but by how long kidney dysfunction persists.
When a severe stage of AKI continues over several days, the impact on prognosis is significantly worse than in short-lived episodes (Kellum, 2015). In a large study of more than 32,000 ICU patients, four days of stage 3 AKI were associated with about 30% risk of death or dialysis at one year, whereas reaching a similar level of risk required more than seven days of stage 1 AKI (Kellum, 2015). These data indicate that it is not only the initial stage that drives outcome, but that AKI duration is a crucial determinant of the clinical trajectory.
This is why early recognition must focus not only on the presence of AKI, but also on the risk that it will become persistent.
The literature describes two main patterns of AKI evolution over time: a transient form, with recovery within 3 days, and a persistent form, in which kidney dysfunction does not resolve within 3-7 days (Zappalà et al., 2025). Persistent episodes are associated with higher mortality, increased hospital costs and an elevated risk of developing or accelerating chronic kidney disease (Zappalà et al., 2025).
In a cohort of 16,968 patients with stage 2-3 AKI, 59% recovered kidney function by hospital discharge (Kellum et al., 2017). Early and sustained recovery was associated with the most favorable prognosis, with approximately 90% survival at one year, whereas in patients without recovery, or with recurrent deterioration after an initial improvement, one-year survival fell to around 40% (Kellum et al., 2017).
Predicting the onset of AKI is already complex, and estimating its risk of persistence is even more challenging, given heterogeneous pathophysiological mechanisms and the limitations of traditional markers such as serum creatinine and urine output. In this context, artificial intelligence offers a relevant operational advantage. Machine learning models can integrate routine data from electronic health records (EHRs) in real time to estimate the risk of persistent AKI with hourly updates (Zappalà et al., 2025). This approach does not rely on a single assessment at admission, but enables dynamic monitoring that follows the patient’s clinical evolution and continuously updates their risk profile.
A concrete example is the external validation of the PersEA model, a machine learning algorithm developed by U-Care Medical that updates the probability of persistent AKI every hour using routine clinical variables, including serum creatinine, urine output and selected laboratory parameters (Zappalà et al., 2025). In an independent Mayo Clinic cohort of 4,479 patients, PersEA demonstrated high performance, with an auROC of 0.977 and an auPR of 0.668 (Zappalà et al., 2025).
In clinical practice, the model enables early identification of patients at higher risk of developing persistent AKI, allowing clinicians to intensify monitoring, optimise haemodynamic management, adjust exposure to potentially nephrotoxic drugs and more appropriately plan the initiation of renal replacement therapy (RRT).
Because AKI duration is a key driver of outcomes, alongside initial severity (Kellum, 2015), and early recovery is associated with a markedly better prognosis (Kellum et al., 2017), integrating predictive models into intensive care can support timely decision-making at a stage when clinical intervention can still meaningfully impact the course of an AKI episode.
If you want to integrate an AI-driven tool into your intensive care unit to anticipate the risk of persistent AKI, discover the potential of the U-Care Renal Platform.