Artificial Intelligence enabled Dyskalemia using Electrocardiogram (AIDE) trial

The Artificial Intelligence enabled Dyskalemia using Electrocardiogram (AIDE) trial (registered at clinicaltrial.gov: NCT05118022) is a pioneering clinical study designed to improve the early detection and treatment of dangerous potassium imbalances in patients. Potassium is a vital electrolyte for heart function, and when its levels become too high (hyperkalemia) or too low (hypokalemia), it can lead to severe health risks, including heart rhythm disturbances and even cardiac arrest. Traditionally, detecting these imbalances relies on blood tests, which can delay treatment, or on manual interpretation of ECGs—a method that may miss subtle signs of dyskalemia. To address these challenges, the AIDE trial integrated an AI-powered system into the electronic health records of emergency departments. This system analyzes patients’ ECGs in real time and automatically alerts physicians when it detects patterns suggesting a significant potassium imbalance. By providing an immediate “pop-up” notification, the AI tool is intended to prompt faster clinical decisions and timely intervention, particularly for patients with hyperkalemia, where rapid treatment is critical. In this randomized controlled trial, physicians were divided into two groups: one received the AI alert during patient care, while the other did not. The trial’s primary goal was to determine whether the AI alert could lead to more prompt and effective treatment for patients with abnormal ECG findings associated with dyskalemia. Early results indicated that doctors who received the alert were more likely to initiate treatment quickly, potentially reducing the risk of serious complications. Ultimately, the AIDE trial aims to demonstrate that incorporating AI into everyday clinical practice can enhance patient safety by supporting healthcare professionals with accurate, real-time diagnostic information. This innovative approach marks an important step toward harnessing advanced technology to improve outcomes in emergency medicine.

This repository is carried out using the software environment R version 3.4.4. The overall file structure is as follows:

AIDE
├── code
│   ├── ...
├── data
│   ├── ...
├── result
│   ├── ...

The relevant syntax for data analysis is provided below.

Analysis for the randomized controlled trial

This repository provides codes for all pre-specified analysis.

Table 1

Table 1 provides an overview of the patient information collected during the AIDE trial. In simple terms, it acts as a snapshot of who the patients are and how they were grouped in the study. The de-identified data is stored in data, and the respective analysis scripts are stored in code directory.

Figure 2

Figure 2 is a visual summary of how well the AI system (AIDE) predicts blood potassium levels using ECG data. It shows several panels that together illustrate the system’s accuracy compared to standard laboratory tests. All these analyses were performed using data from patients whose blood tests were taken within one hour of their ECG. This ensures that the comparison is fair and reflects real-time decision-making in an emergency setting. The de-identified data is stored in data, and the respective analysis scripts are stored in code directory.

Figure 3

Figure 3 displays the primary analysis of the trial, focusing on how the AIDE alert influences the timing of treatment for potassium imbalances in patients. In simple terms, this figure compares two groups: one where physicians received the real-time AI alert (intervention group) and one where they did not (control group). In summary, Figure 3 demonstrates that the AIDE alert system effectively prompts physicians to act more quickly in cases of hyperkalemia, which is critical in preventing serious complications. The de-identified data is stored in data, and the respective analysis scripts are stored in code directory.

Figure 4

Figure 4 is designed to compare and assess both safety outcomes and overall clinical results between patients whose care was guided by the AI alert (intervention group) and those who received standard care (control group). Essentially, this figure answers the question: "Does using the AI alert improve patient care without causing extra harm?" Overall, Figure 4 helps us understand whether the early interventions prompted by the AI alert are both safe and effective, ensuring that faster treatment does not come at the expense of patient safety. The de-identified data is stored in data, and the respective analysis scripts are stored in code directory.

Other supplemental data

We also provide other supplemental data in data, and the respective analysis scripts are stored in code directory. Please see and try to repuduce the analyses.

Related publications

The paper of AIDE trial is currently under review.