Background: The growing complexity of patient data and the intricate relationship between heart failure (HF) and acute kidney injury (AKI) underscore the potential benefits of integrating artificial intelligence (AI) and machine learning into healthcare. These advanced analytical tools aim to improve the understanding of the pathophysiological relationship between kidney and heart, provide optimized, individualized, and timely care, and improve outcomes of HF with AKI patients. Summary: This comprehensive review article examines the transformative potential of AI and machine-learning solutions in addressing the challenges within this domain. The article explores a range of methodologies, including supervised and unsupervised learning, reinforcement learning, and AI-driven tools like chatbots and large language models. We highlight how these technologies can be tailored to tackle the complex issues prevalent among HF patients with AKI. The potential applications identified span predictive modeling, personalized interventions, real-time monitoring, and collaborative treatment planning. Additionally, we emphasize the necessity of thorough validation, the importance of collaborative efforts between cardiologists and nephrologists, and the consideration of ethical aspects. These factors are critical for the effective application of AI in this area. Key Messages: As the healthcare field evolves, the synergy of advanced analytical tools and clinical expertise holds significant promise to enhance the care and outcomes of individuals who deal with the combined challenges of HF and AKI.

Heart failure (HF) and acute kidney injury (AKI) are two common conditions with significant implications for global health. Their epidemiology underscores their considerable impact on mortality, morbidity, and healthcare costs. AKI is a frequent coincidental syndrome among patients with HF (13%) [1]. When HF patients are admitted to the hospital, the incidence of new-onset AKI is around 20% [2]. Indeed, AKI is a robust independent predictor for both in-hospital and 1-year mortality for HF patients [1, 3]. AKI is independently associated with a higher risk of cardiovascular events, especially recurrent HF, after hospital discharge [4‒7]. This relationship is also associated with a higher likelihood of chronic kidney disease (CKD), expedited progression to end-stage renal disease, and a decline in health-related quality of life [2, 8].

The grim consequence of HF and AKI combination emphasizes the importance of prompt detection and timely, individualized preventive and therapeutic interventions. Thus, early detection of patients at risk of AKI is essential for improving outcomes [3]. New technologies, including AI and machine learning, can provide a platform for improved care for HF patients who are at risk of AKI development. In this review paper, we discuss the technologies used for this purpose, provide examples of successful projects, and then discuss the constraints and barriers to their implementations.

Traditional Statistical Approaches

Congestion, often seen in decompensated HF, is a major factor contributing factor to AKI development [9, 10]. The severity or persistency of congestion is linked with adverse outcomes. This notion underscores the necessity for effective decongesting strategies using timely and adequate diuretic therapies. Hemodynamic variables, such as low cardiac output and congestion, combined with the effects of medications like angiotensin-converting enzyme inhibitors and diuretics on kidney function, are pivotal in the onset of AKI.

CKD stands out as a significant AKI risk factor associated with acute decompensated heart failure (ADHF), along with other factors like age, diabetes mellitus, liver disease, right ventricular failure, and baseline left ventricular ejection fraction and diastolic function also playing crucial roles [1, 11]. Various statistical models, including multivariable logistic regression and Cox proportional analysis, have been utilized to study AKI predictors among HF patients [12]. These models aim to identify independent risk factors and assess their impact on patient outcomes. However, challenges exist with these prediction models, such as reliance on retrospective data, potential confounding factors, and the necessity for their validation across different patient populations. The selection of predictors and the integration of new biomarkers or clinical parameters can greatly affect the efficacy of these models [12]. Table 1 summarizes Studies Predicting Acute Kidney Injury in Heart Failure Patients by Traditional Statistical Approaches. Collectively, these studies offer the potential for clinical application in identifying ADHF patients at risk for AKI. Lee et al. [12] conducted a study to validate existing AKI prediction models, underscoring Wang et al. [16] and Forman et al. [13] models in determining AKI risk in ADHF patients.

Table 1.

Summary of selected studies predicting AKI in HF patients by traditional statistical approaches

Study authorPopulation and sizeAKI definitionKey parameters/biomarkersC-statistic/AUC
Forman et al. [131,004 ADHF patients An increase in serum creatinine >0.3 mg/dL History of prior HF, diabetes, systolic BP, admission serum creatinine Not reported 
No validation group 
Verdiani et al. [14394 ADHF patients An increase in serum creatinine >0.3 mg/dL Age, serum creatinine, heart rate, calcium channel blocker use, digoxin use Not reported 
No validation group 
Breidthardt et al. or Basel risk score [15657 ADHF patients An increase in serum creatinine >0.3 mg/dL CKD, serum bicarbonate, outpatient diuretic use 0.71 
No validation group 
Wang et al. [161,709 ADHF patients AKIN Age, HF functional class, admission times for ADHF, systolic BP, serum creatinine, sodium, proteinuria, intravenous furosemide use Development: 0.76 
Development: 1010 Validation: 0.76 
Validation: 699 
Zhou et al. [17507 ADHF patients development: 321 KDIGO Age, sex, CKD, serum albumin, NT-proBNP, uAGT, uNGAL Development: 0.859 
Validation: 186 Validation: 0.847 
Wang et al. [3675 ADHF patients KDIGO Age, diabetes, previous renal dysfunction, serum creatinine, BNP, serum albumin Development: 0.766 
Bootstrap validation: 0.763 
Lassus et al. [18292 ADHF patients AKIN Cystatin C 0.92 
Lee et al. [1210,364 ADHF hospitalizations (validation study) KDIGO Various prediction models Wang: 0.73 
Forman: 0.70 
Basel: 0.60 
Verdiani: 0.59 
Zhou: 0.54 
Study authorPopulation and sizeAKI definitionKey parameters/biomarkersC-statistic/AUC
Forman et al. [131,004 ADHF patients An increase in serum creatinine >0.3 mg/dL History of prior HF, diabetes, systolic BP, admission serum creatinine Not reported 
No validation group 
Verdiani et al. [14394 ADHF patients An increase in serum creatinine >0.3 mg/dL Age, serum creatinine, heart rate, calcium channel blocker use, digoxin use Not reported 
No validation group 
Breidthardt et al. or Basel risk score [15657 ADHF patients An increase in serum creatinine >0.3 mg/dL CKD, serum bicarbonate, outpatient diuretic use 0.71 
No validation group 
Wang et al. [161,709 ADHF patients AKIN Age, HF functional class, admission times for ADHF, systolic BP, serum creatinine, sodium, proteinuria, intravenous furosemide use Development: 0.76 
Development: 1010 Validation: 0.76 
Validation: 699 
Zhou et al. [17507 ADHF patients development: 321 KDIGO Age, sex, CKD, serum albumin, NT-proBNP, uAGT, uNGAL Development: 0.859 
Validation: 186 Validation: 0.847 
Wang et al. [3675 ADHF patients KDIGO Age, diabetes, previous renal dysfunction, serum creatinine, BNP, serum albumin Development: 0.766 
Bootstrap validation: 0.763 
Lassus et al. [18292 ADHF patients AKIN Cystatin C 0.92 
Lee et al. [1210,364 ADHF hospitalizations (validation study) KDIGO Various prediction models Wang: 0.73 
Forman: 0.70 
Basel: 0.60 
Verdiani: 0.59 
Zhou: 0.54 

ADHF, acute decompensated heart failure; AKI, acute kidney injury; AUC, area under the curve; BNP, B-type natriuretic peptide; BP, blood pressure; CKD, chronic kidney disease; NT-proBNP, N-terminal pro-brain natriuretic peptide; uAGT, urine angiotensinogen; uNGAL, urine neutrophil gelatinase-associated lipocalin.

Developing digital tools or applications that incorporate these risk models could provide clinicians with immediate, real-time risk assessments. Additionally, they elevate the scope for research of novel biomarkers using cutting-edge technologies like proteomics or genomics, which could lead to even more precise prediction models. The application of AI and machine learning in this context can enhance the accuracy and applicability of the models that are based on traditional statistical tools. These technologies could greatly enhance the performance of existing models, offering a more sophisticated and accurate approach to managing HF and AKI.

AI and Machine Learning in HF and AKI

The increasing availability of electronic health records and the intricate relationship between HF and AKI provide a platform for developing advanced analytical tools [19‒21]. AI and machine learning have emerged as pivotal solutions [21]. It is essential to discern that AI and machine learning, though frequently conflated, each possess distinct characteristics. AI is a broader concept where machines are designed to carry out tasks in a way that we would consider “smart” or “intelligent” [20]. Machine learning, a subset of AI, involves algorithms that allow machines to learn from and make data-based decisions. Instead of being explicitly programmed to perform a task, a machine-learning model uses patterns and inference to make predictions [19, 22, 23].

Machine learning can be further delineated into multiple subcategories based on their methodologies and utilities (Fig. 1). Supervised machine learning entails training models using datasets with designated labels; the algorithm is furnished with paired input-output data. It assimilates this information and extrapolates this acquired knowledge to novel, unobserved data [24]. Unsupervised machine learning, in contrast, deals with unlabeled data. The algorithm learns the data’s inherent structure without explicit instructions [24]. Reinforcement learning is a form of machine learning in which algorithms learn decision-making through interaction with an environment. The objective was to maximize cumulative rewards through trial and error, enabling algorithms to adapt and enhance their decision-making capabilities over time [25].

Fig. 1.

Types of artificial intelligence and machine learning in heart failure and AKI. HF, heart failure; AKI, acute kidney injury; AI, artificial intelligence; ML, machine learning; LLM, large language models.

Fig. 1.

Types of artificial intelligence and machine learning in heart failure and AKI. HF, heart failure; AKI, acute kidney injury; AI, artificial intelligence; ML, machine learning; LLM, large language models.

Close modal

More advanced machine-learning techniques, such as large language models (LLMs) and chatbots, have emerged recently [24]. Chatbots are AI systems designed to simulate human conversation audibly or with text-based tools [25‒27]. Integrating these AI and machine-learning approaches into the study of HF and AKI has yielded encouraging outcomes.

Supervised machine-learning techniques have been increasingly recognized for their potential in healthcare predictive analytics [28‒32]. One of the critical areas of application is the prediction of AKI in patients with HF [25, 27]. Supervised machine learning may offer the advantage of processing vast amounts of data, recognizing intricate patterns, and providing accurate predictions [25, 27]. These models can be trained on various data sources, including electronic health records, to predict the likelihood of AKI based on a combination of demographic, clinical, and treatment variables.

Recently, Liu et al. [27] conducted a study on developing a supervised machine-learning model to predict the likelihood of AKI in HF patients during their hospitalization among 2,678 HF patients within the MIMIC-IV database, among whom 919 developed AKI. Thirty-nine demographic, clinical, and treatment features were employed to train five distinct machine-learning algorithms, i.e., decision tree, random forest, support vector machine, K-nearest neighbor, and logistic regression. Notably, the random forest algorithm exhibited superior performance, yielding an area under the receiver operating characteristic (ROC) curve of 0.96, an accuracy of 88.36%, a sensitivity of 96.04%, and a specificity of 73.91% (Fig. 2) to predict AKI. In contrast, logistic regression reported an AUROC of 0.92, an accuracy of 86.42%, a sensitivity of 91.42%, and a specificity of 77.02%. Despite its traditional approach, logistic regression displayed a commendable ability to predict AKI. The ensemble approach of the random forest, which utilizes multiple decision trees, exhibited an enhanced performance.

Fig. 2.

Performance metrics of prediction models. The bar chart above visually represents the performance metrics (accuracy, sensitivity, and specificity) for each algorithm (random forest, support vector machine, decision tree, K-nearest neighbor, and logistic regression).

Fig. 2.

Performance metrics of prediction models. The bar chart above visually represents the performance metrics (accuracy, sensitivity, and specificity) for each algorithm (random forest, support vector machine, decision tree, K-nearest neighbor, and logistic regression).

Close modal

The study also identified several variables among the analyzed factors by the random forest algorithm’s Gini index, i.e., sequential organ function assessment score, partial pressure of oxygen, estimated glomerular filtration rate, serum bicarbonate, hemoglobin, platelet count, blood lactic acid, serum creatinine, serum magnesium, and blood glucose. While the Gini index analysis1 was performed only for the random forest, it can be postulated that key predictors like the SOFA score, eGFR, and vital signs might have been significant in the logistic regression model.

To enhance usability, the authors additionally explored a simplified model employing these ten most predictive variables, as determined by the random forest algorithm. This parsimonious model, encompassing readily available clinical features, displayed notable efficacy in predicting AKI. Furthermore, the parsimonious model’s integration into clinical practice could support patient risk stratification and inform treatment decisions.

The authors defined AKI as an increase in serum creatinine of ≥0.3 mg/dL within 48 h of ICU admission. Notably, the study did not elaborate on the severity of AKI based on creatinine variations or the need for dialysis. The study reported that the baseline estimated glomerular filtration rate (eGFR) was lower in the AKI group compared to the non-AKI group. Also, the authors did not explicitly address CKD beyond the baseline eGFR measurements. Comorbidity rates, including diabetes, coronary heart disease, hypertension, and atrial fibrillation, were found to be higher in the AKI group. Medication utilization also varied, with lower use of RAAS inhibitors and digoxin in the AKI group.

The investigation of Liu et al. [27] could benefit from additional details concerning AKI severity, CKD status, ejection fraction, and incorporating newer drugs like SGLT2 inhibitors. These missing elements hinder a comprehensive assessment of risk factors and model performance. Meanwhile, this research identifies potential limitations and areas for improvement, including the need for external validation, incorporation of novel biomarkers, evaluation of the model’s cost-effectiveness, and integration into clinical practice.

Supervised machine-learning techniques have been explored for their potential to predict hospital readmissions in HF patients. Mortazavi et al. [25] compared machine-learning methods, including random forests, boosting, and support vector machines, against traditional logistic regression models. Their research utilized the Tele-HF trial data, which included comprehensive clinical and sociodemographic information on 1,004 patients, encompassing 472 variables. These variables ranged from clinical data, such as laboratory test results and medical history, to socioeconomic and demographic factors. Interestingly, while conditions such as CKD and dialysis were considered, there was a lack of detailed quantitative metrics on kidney function, such as creatinine and eGFR levels. In this study, the models focused on predicting the likelihood of 30-day and 180-day hospital readmissions, differentiating between all-cause and HF-specific readmissions (Fig. 3). Compared with traditional logistic regression, the machine-learning models showed a modest improvement in discrimination. Their C-statistics improved by 17–25%, indicating a better ability to identify low- and high-risk groups, thereby expanding the predictive range. However, the prediction accuracy of these machine-learning models was still only average, which suggests that while these advanced analytical methods offer some improvement, there is still room for enhancement in accurately predicting hospital readmissions for HF patients. A notable omission in the study is the lack of specific mention of AKI or detailed metrics like creatinine or eGFR. The study does acknowledge the history of CKD but omits information on any acute variations in kidney function. More comprehensive data on kidney function at admission and its acute variations would be essential to ascertain the impact of kidney function fluctuation on the rehospitalization rate after admission for acute exacerbation of HF.

Fig. 3.

Heatmap of ML techniques over logistic regression. The color intensity in each cell indicates the level of percentage improvement. This heatmap uses a gradient from light to dark blue, where lighter shades represent lower improvements (or even negative values) and darker shades indicate higher improvements.

Fig. 3.

Heatmap of ML techniques over logistic regression. The color intensity in each cell indicates the level of percentage improvement. This heatmap uses a gradient from light to dark blue, where lighter shades represent lower improvements (or even negative values) and darker shades indicate higher improvements.

Close modal

Unsupervised machine learning is a subset of machine-learning techniques where algorithms are trained on data without predefined labels, allowing the model to autonomously identify patterns and structures within the data [33]. This approach contrasts with supervised learning, where models are trained on labeled data to make predictions or classifications. One of the primary utilities of unsupervised learning is clustering, where data are grouped based on similarities, often revealing hidden patterns or subgroups that might not be evident through traditional analysis [34]. By leveraging unsupervised machine learning, clinicians and researchers can potentially unearth distinct patient phenotypes or subgroups that share common clinical characteristics or outcomes. One such application is in the identification of HF patients carrying higher AKI risk. Given the intricate relationship between cardiac function and kidney health, the ability to proactively identify these high-risk patients can lead to more personalized care strategies, timely interventions, and improved patient outcomes [33].

Hong et al. [26] utilized an unsupervised machine-learning approach to identify HF patients with normal renal function susceptible to AKI. Using an unsupervised clustering algorithm on data from 5,075 hospitalizations, they discerned 2 patient clusters of “high-risk” and “low-risk” groups (Fig. 4). Notably, the high-risk group was characterized by older age, diminished cardiac function, and indications of multi-organ dysfunction. The study’s data were sourced from two primary datasets. The primary development dataset was derived from the PLA General Hospital (PLAGH) in China. Data from the Medical Information Mart for Intensive Care (MIMIC-III) database, including 1,006 HF patients, were utilized for external validation. AKI was adjudicated based on the KDIGO criteria, focusing on specific serum creatinine changes. Despite these advancements, the study had certain limitations. It successfully predicted early biochemical AKI using electronic health records (EHR) data but did not explore the progression to severe AKI and the need for dialysis. Although the high-risk group showed an AKI rate of 11.5%, the study did not provide detailed information on the severity or stages of AKI. Furthermore, while in-hospital mortality was observed to be higher in the high-risk group, the specific causes of death were not reported. Hong et al. [26] highlighted the significant potential of unsupervised machine learning in patient risk stratification. Their innovative approach shed new light on identifying high-risk HF patients, underscoring the critical role of data-driven phenotyping in risk assessment. However, there is a need for further research to expand the model’s capabilities, particularly in predicting more severe stages of AKI and understanding the clinical implications of these predictions. This direction could provide deeper insights into patient management and treatment strategies in HF cases.

Fig. 4.

Key characteristics and differences between phenotype 1 (low risk) and phenotype 2 (high risk) for AKI among patients with HF.

Fig. 4.

Key characteristics and differences between phenotype 1 (low risk) and phenotype 2 (high risk) for AKI among patients with HF.

Close modal

Urban et al. [35] analyzed 312 patients hospitalized with ADHF to identify predictors of worsening renal function (WRF) through unsupervised machine learning. The study utilized k-Medoid clustering, an unsupervised machine-learning technique that groups similar patients as cluster centers based on actual patient data. This method is preferred to the other strategy, i.e., k-means clustering2, for its enhanced interpretability in patient phenotyping. To fine-tune the clustering process, they employed the Davies-Bouldin index, aiming for optimal intra-cluster similarity and inter-cluster differences. The analysis involved 86 admission variables, categorizing patients into three distinct phenotypes (Fig. 5). The three identified phenotypes exhibited notable differences in WRF incidence. Cluster 1 included 110 older females with a first-time HF diagnosis and preserved EF. This group had high comorbidity rates, such as diabetes and hypertension, and reported a 24% incidence of WRF. Cluster 2 consisted of 44 younger males with chronic HF, reduced EF, high rates of substance abuse, and the lowest WRF incidence at 2%. Cluster 3 encompassed 158 older males with chronic reduced EF HF and a history of coronary artery disease, showing a moderate WRF incidence of 15%. This study by Urban et al. [35] showcases the utility of unsupervised machine learning in uncovering distinct AHF phenotypes, each with unique demographic and clinical profiles.

Fig. 5.

Key characteristics and differences between three unique phenotypes of patients with acute HF with different incidences of WRF.

Fig. 5.

Key characteristics and differences between three unique phenotypes of patients with acute HF with different incidences of WRF.

Close modal

Current data on the use of Reinforcement Learning (RL) in managing HF patients with AKI are limited. However, there are promising indications of the potential utilization of RL in this specific clinical context [36, 37]. RL’s capabilities can potentially revolutionize the prediction and management of AKI, a significant complication in HF patients (Table 2).

Table 2.

Potential utilization of RL in managing HF patients at high risk for AKI

AspectDescription
Predictive modeling in HF and AKI Utilizing RL’s capabilities, healthcare professionals can develop predictive models that analyze intricate patient data, including medical history, symptoms, and diagnostic outcomes. This empowers these models to anticipate the trajectory of HF and the potential onset of complications like AKI. By identifying AKI risk factors early on, interventions can be timely and tailored, potentially altering the course of the disease and preventing severe consequences. 
Optimizing AKI prevention strategies Customizing treatments for HF patients often involves managing medications like beta-blockers and ACE inhibitors. RL can aid this endeavor by optimizing drug regimens based on individual patient responses and preemptively detecting possible drug interactions. This ensures that treatment remains effective and minimizes the risk of complications like AKI. 
Real-time monitoring and AKI detection Wearable devices have transformed patient monitoring by providing real-time data. When coupled with RL algorithms, these devices can continuously monitor vital signs and symptoms, enabling the early detection of changes that might signal the onset of AKI. Such early alerts offer valuable time for medical professionals to intervene promptly and mitigate the risk of AKI development. 
Dynamic AKI management protocols The unpredictable nature of AKI necessitates adaptable management strategies that can be adjusted on the fly. RL’s capabilities can provide real-time suggestions for fluid administration, medication dosages, or the initiation of dialysis. By aligning these recommendations with the patient’s evolving clinical status, RL assists healthcare providers in making informed decisions that optimize AKI management. 
Long-term AKI effects and preventive measures After an episode of AKI, patients remain at risk for chronic kidney disease. RL’s involvement extends beyond the acute phase, vigilantly monitoring patients for signs of lasting kidney damage and recommending appropriate preventive measures. This comprehensive approach helps reduce the likelihood of AKI-related complications in the long run. 
AspectDescription
Predictive modeling in HF and AKI Utilizing RL’s capabilities, healthcare professionals can develop predictive models that analyze intricate patient data, including medical history, symptoms, and diagnostic outcomes. This empowers these models to anticipate the trajectory of HF and the potential onset of complications like AKI. By identifying AKI risk factors early on, interventions can be timely and tailored, potentially altering the course of the disease and preventing severe consequences. 
Optimizing AKI prevention strategies Customizing treatments for HF patients often involves managing medications like beta-blockers and ACE inhibitors. RL can aid this endeavor by optimizing drug regimens based on individual patient responses and preemptively detecting possible drug interactions. This ensures that treatment remains effective and minimizes the risk of complications like AKI. 
Real-time monitoring and AKI detection Wearable devices have transformed patient monitoring by providing real-time data. When coupled with RL algorithms, these devices can continuously monitor vital signs and symptoms, enabling the early detection of changes that might signal the onset of AKI. Such early alerts offer valuable time for medical professionals to intervene promptly and mitigate the risk of AKI development. 
Dynamic AKI management protocols The unpredictable nature of AKI necessitates adaptable management strategies that can be adjusted on the fly. RL’s capabilities can provide real-time suggestions for fluid administration, medication dosages, or the initiation of dialysis. By aligning these recommendations with the patient’s evolving clinical status, RL assists healthcare providers in making informed decisions that optimize AKI management. 
Long-term AKI effects and preventive measures After an episode of AKI, patients remain at risk for chronic kidney disease. RL’s involvement extends beyond the acute phase, vigilantly monitoring patients for signs of lasting kidney damage and recommending appropriate preventive measures. This comprehensive approach helps reduce the likelihood of AKI-related complications in the long run. 

Incorporating RL into HF management holds immense promise for predicting and managing the development of AKI in HF patients. By leveraging personalized insights, real-time monitoring, and dynamic intervention recommendations, RL can significantly enhance patient care and outcomes in this complex clinical scenario. Figure 6 demonstrates an example of a flow diagram loop for using RL in HF management to prevent AKI. The RL agent (the decision-maker) interacts with the environment (the healthcare setting and patient conditions) by taking actions (medical interventions) based on the current state. The environment then provides feedback in the form of rewards, indicating the effectiveness of those actions. Over time, the agent learns from this feedback, refining its decision-making process to optimize patient outcomes in HF management and AKI prevention.

Fig. 6.

Flow diagram loop for RL in HF management and AKI prevention.

Fig. 6.

Flow diagram loop for RL in HF management and AKI prevention.

Close modal

AKI in HF patients is a niche area that requires collaborative efforts between nephrologists and cardiologists. This niche area calls for innovative tools and strategies. Emerging technologies like LLMs and chatbots [38‒43] offer promising avenues to enhance patient care and are increasingly relevant for healthcare professionals in this field (Table 3).

Table 3.

Large language model and potential utilization of chatbot in HF patients with AKI

Large language model and chatbot potential utilizationDescription
Advanced diagnostic support Nephrologists and cardiologists often face diagnostic challenges in HF patients with AKI. LLM-powered chatbots can analyze vast data to provide diagnostic suggestions based on various factors, leading to more precise investigations. 
Enhanced patient education and engagement Recognizing the importance of patient education, chatbots can deliver condition-specific insights. This includes explanations on the impact of renal function on cardiac health, ensuring patients receive tailored information for their conditions. 
Collaborative treatment plans Chatbots can help formulate combined treatment plans by considering up-to-date research. This ensures that both heart and kidney health are concurrently addressed. 
Real-time Patient Monitoring Real-time monitoring is vital, given the link between renal and cardiac functions. Patients can enter daily metrics into chatbots, which can alert professionals if any alarming patterns emerge. 
Telemedicine and virtual consultations Incorporating chatbots in telemedicine can make virtual consultations more efficient. These bots can collect initial patient data, address minor issues, and escalate serious concerns to the medical professional, optimizing clinicians’ time. 
Multidisciplinary team coordination Chatbots act as coordination tools among various healthcare professionals. They can provide real-time updates on treatment plans, diagnostic results, and patient inquiries, promoting a cohesive care approach. 
Large language model and chatbot potential utilizationDescription
Advanced diagnostic support Nephrologists and cardiologists often face diagnostic challenges in HF patients with AKI. LLM-powered chatbots can analyze vast data to provide diagnostic suggestions based on various factors, leading to more precise investigations. 
Enhanced patient education and engagement Recognizing the importance of patient education, chatbots can deliver condition-specific insights. This includes explanations on the impact of renal function on cardiac health, ensuring patients receive tailored information for their conditions. 
Collaborative treatment plans Chatbots can help formulate combined treatment plans by considering up-to-date research. This ensures that both heart and kidney health are concurrently addressed. 
Real-time Patient Monitoring Real-time monitoring is vital, given the link between renal and cardiac functions. Patients can enter daily metrics into chatbots, which can alert professionals if any alarming patterns emerge. 
Telemedicine and virtual consultations Incorporating chatbots in telemedicine can make virtual consultations more efficient. These bots can collect initial patient data, address minor issues, and escalate serious concerns to the medical professional, optimizing clinicians’ time. 
Multidisciplinary team coordination Chatbots act as coordination tools among various healthcare professionals. They can provide real-time updates on treatment plans, diagnostic results, and patient inquiries, promoting a cohesive care approach. 

In the medical domain, the margin for error is slim, as even minor mistakes can significantly affect patient outcomes. For nephrologists and cardiologists who grapple with intricate conditions like HF compounded by AKI, accuracy is of the essence. Hence, tools like chatbots and LLMs must be extensively validated against real-world clinical cases to affirm their reliability. These tools must be designed to discern when to engage human expertise, especially in circumstances that are either complex or ambiguously presented.

Moreover, the active involvement of cardiologists and nephrologists in the design, development, and periodic updates of these AI-driven tools cannot be overstated. Their on-the-ground clinical experience and insights will be vital in calibrating the bot’s algorithms to align with current medical practices and mitigate the risks of inaccuracies.

Skalidis et al. [42] assessed the capabilities of ChatGPT, an AI chatbot, based on its performance on the European Exam in Core Cardiology (EECC). The bot achieved an overall accuracy rate of 58.8%. When presented with questions from previous EECC examinations, it delivered an accuracy of 61.7%. The accuracy of materials like StudyPRN and Braunwald’s used for exam preparation ranged between 52.6% and 63.8%. Meanwhile, the EECC passing score hovers around 60%, implying that ChatGPT’s performance was around or slightly above the passing rate. It is worth noting that the bot’s capabilities were limited to text-based questions. Questions incorporating images, which constitute 25% of EECC queries, were beyond its scope [42].

Several challenges become apparent when considering integrating AI and machine-learning models into research on HF patients with AKI (Table 4). At the forefront, the effectiveness of such models hinges significantly upon the quality of the data they receive. In situations where data are subpar, the resulting predictions of the model could be skewed. Moreover, the adaptability of the models across different patient groups remains a concern. For instance, a model developed with data from one particular group may not necessarily yield accurate results for another, posing constraints in concluding broader populations. This phenomenon could lead to disparity in healthcare, resulting in suboptimal care for underprivileged populations who did not participate in developing and validating these tools.

Table 4.

Challenges and possible policy development or solutions for AI and machine learning in HF patients with AKI

ChallengesPossible policy development or solutions
Dependence on accurate and specific data for HF with AKI Design and implement AKI-specific data collection standards to ensure precision and relevance 
Narrow scope of model applicability in diverse HF and AKI scenarios Use datasets from varied clinical settings and patient backgrounds; employ iterative testing across these settings 
Difficulty in interpreting AI models specific to HF and AKI complexities Advocate for transparent AI algorithms tailored to HF and AKI contexts for better clinical interpretability 
Overfitting when the model is trained on specific HF and AKI cases Employ regularization techniques and continuously validate models against diverse HF and AKI scenarios. 
Ethical dilemmas in AI application, especially regarding patient data in HF and AKI Draft strict ethical guidelines on HF and AKI patient data protection, unbiased analysis, and informed consent. 
Shortage of detailed, ailment-specific data on HF patients suffering from AKI Forge partnerships with cardiac and renal departments for richer data pools and endorse research dedicated to the intersection of HF and AKI 
ChallengesPossible policy development or solutions
Dependence on accurate and specific data for HF with AKI Design and implement AKI-specific data collection standards to ensure precision and relevance 
Narrow scope of model applicability in diverse HF and AKI scenarios Use datasets from varied clinical settings and patient backgrounds; employ iterative testing across these settings 
Difficulty in interpreting AI models specific to HF and AKI complexities Advocate for transparent AI algorithms tailored to HF and AKI contexts for better clinical interpretability 
Overfitting when the model is trained on specific HF and AKI cases Employ regularization techniques and continuously validate models against diverse HF and AKI scenarios. 
Ethical dilemmas in AI application, especially regarding patient data in HF and AKI Draft strict ethical guidelines on HF and AKI patient data protection, unbiased analysis, and informed consent. 
Shortage of detailed, ailment-specific data on HF patients suffering from AKI Forge partnerships with cardiac and renal departments for richer data pools and endorse research dedicated to the intersection of HF and AKI 

Exploring further, the complex architecture of some AI models presents a challenge regarding interpretability. This complexity can make it difficult for medical professionals to understand the reasoning behind a model’s predictions, leading to concerns about its transparency. Such intricacy also increases the risk of overfitting, where a model might show high performance with its training data but fails to generalize well to new, unseen data. In addition to these technical challenges, there are significant ethical considerations, particularly regarding data privacy and inherent biases in the data used to train these models. Adopting a rigorous and thoughtful approach to address these ethical issues effectively is essential. Another critical aspect is the lack of comprehensive data specifically concerning AKI in HF patients. This gap can hinder the development of robust and comprehensive AI models in this area. The AI and machine-learning potential in healthcare, particularly in the context of HF and AKI, is substantial. However, recognizing and proactively addressing these challenges is crucial for successfully integrating AI into healthcare research and practice. By strategically tackling these issues, we can fully leverage the capabilities of AI to advance research and improve patient outcomes in HF and AKI.

In summary, AI and machine learning present a profound potential to redefine the treatment paradigm for HF patients with AKI. This transformation integrates both supervised and unsupervised learning approaches. Furthermore, reinforcement learning demonstrates considerable promise with its innate capacities for prompt monitoring, early AKI detection, and the formulation of tailored therapeutic strategies. The rise of LLMs and chatbots is anticipated to innovate diagnostic support, patient engagement, and virtual medical consultations. Nonetheless, obstacles related to the data reliability, interpretability of AI model predictions, and ethical quandaries emphasize the need for meticulous validation and iterative refinement. The manuscript underscores the imperative of synergistic efforts between cardiologists and nephrologists, accentuating the judicious application of AI to enhance patient health outcomes.

None of the authors has any conflict of interest with this article.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

All authors contributed to the conception of the idea. Wisit Cheungpasitporn drafted the paper, and Charat Thongprayoon and Kianoush Kashani provided critical review and extensive editing.

1

The Gini index is to measure the randomness or the impurity or entropy in the values of a dataset. This is to mitigate the impurities from the root nodes (at the top of decision tree) to the leaf nodes (vertical branches down the decision tree) of a decision tree model.

2

K-means clustering is a statistical method used for partitioning a dataset into a set of k groups (or clusters), where k is a predetermined or user-defined constant. The method aims to partition the data points into clusters in which each data point belongs to the cluster with the nearest mean, serving as a prototype of the cluster.

1.
Holgado
JL
,
Lopez
C
,
Fernandez
A
,
Sauri
I
,
Uso
R
,
Trillo
JL
, et al
.
Acute kidney injury in heart failure: a population study
.
ESC Heart Fail
.
2020
;
7
(
2
):
415
22
. .
2.
Doshi
R
,
Dhawan
T
,
Rendon
C
,
Rodriguez
MA
,
Al-Khafaji
JF
,
Taha
M
, et al
.
Incidence and implications of acute kidney injury in patients hospitalized with acute decompensated heart failure
.
Intern Emerg Med
.
2020
;
15
(
3
):
421
8
. .
3.
Wang
L
,
Zhao
YT
.
Development and validation of a prediction model for acute kidney injury among patients with acute decompensated heart failure
.
Front Cardiovasc Med
.
2021
;
8
:
719307
. .
4.
Odutayo
A
,
Wong
CX
,
Farkouh
M
,
Altman
DG
,
Hopewell
S
,
Emdin
CA
, et al
.
AKI and long-term risk for cardiovascular events and mortality
.
J Am Soc Nephrol
.
2017
;
28
(
1
):
377
87
. .
5.
Go
AS
,
Hsu
CY
,
Yang
J
,
Tan
TC
,
Zheng
S
,
Ordonez
JD
, et al
.
Acute kidney injury and risk of heart failure and atherosclerotic events
.
Clin J Am Soc Nephrol
.
2018
;
13
(
6
):
833
41
. .
6.
Son
HE
,
Moon
JJ
,
Park
JM
,
Ryu
JY
,
Baek
E
,
Jeong
JC
, et al
.
Additive harmful effects of acute kidney injury and acute heart failure on mortality in hospitalized patients
.
Kidney Res Clin Pract
.
2022
;
41
(
2
):
188
99
. .
7.
Soranno
DE
,
Bihorac
A
,
Goldstein
SL
,
Kashani
KB
,
Menon
S
,
Nadkarni
GN
, et al
.
Artificial intelligence for AKI!now: let’s not await plato’s utopian republic
.
Kidney360
.
2022
;
3
(
2
):
376
81
. .
8.
Rewa
O
,
Bagshaw
SM
.
Acute kidney injury-epidemiology, outcomes and economics
.
Nat Rev Nephrol
.
2014
;
10
(
4
):
193
207
. .
9.
Schefold
JC
,
Filippatos
G
,
Hasenfuss
G
,
Anker
SD
,
von Haehling
S
.
Heart failure and kidney dysfunction: epidemiology, mechanisms and management
.
Nat Rev Nephrol
.
2016
;
12
(
10
):
610
23
. .
10.
Chahal
RS
,
Chukwu
CA
,
Kalra
PR
,
Kalra
PA
.
Heart failure and acute renal dysfunction in the cardiorenal syndrome
.
Clin Med
.
2020
;
20
(
2
):
146
50
. .
11.
Chen
JJ
,
Lee
TH
,
Kuo
G
,
Yen
CL
,
Chen
SW
,
Chu
PH
, et al
.
Acute kidney disease after acute decompensated heart failure
.
Kidney Int Rep
.
2022
;
7
(
3
):
526
36
. .
12.
Lee
TH
,
Fan
PC
,
Chen
JJ
,
Wu
VC
,
Lee
CC
,
Yen
CL
, et al
.
A validation study comparing existing prediction models of acute kidney injury in patients with acute heart failure
.
Sci Rep
.
2021
;
11
(
1
):
11213
. .
13.
Forman
DE
,
Butler
J
,
Wang
Y
,
Abraham
WT
,
O’Connor
CM
,
Gottlieb
SS
, et al
.
Incidence, predictors at admission, and impact of worsening renal function among patients hospitalized with heart failure
.
J Am Coll Cardiol
.
2004
;
43
(
1
):
61
7
. .
14.
Verdiani
V
,
Lastrucci
V
,
Nozzoli
C
.
Worsening renal function in patients hospitalized with acute heart failure: risk factors and prognostic significances
.
Int J Nephrol
.
2010
Oct 11;2011:785974.
15.
Breidthardt
T
,
Socrates
T
,
Noveanu
M
,
Klima
T
,
Heinisch
C
,
Reichlin
T
, et al
.
Effect and clinical prediction of worsening renal function in acute decompensated heart failure
.
Am J Cardiol
.
2011
Mar 1;
107
(
5
):
730
5
.
16.
Wang
YN
,
Cheng
H
,
Yue
T
,
Chen
YP
.
Derivation and validation of a prediction score for acute kidney injury in patients hospitalized with acute heart failure in a Chinese cohort
.
Nephrology
.
2013
;
18
(
7
):
489
96
. .
17.
Zhou
LZ
,
Yang
XB
,
Guan
Y
,
Xu
X
,
Tan
MT
,
Hou
FF
, et al
.
Development and validation of a risk score for prediction of acute kidney injury in patients with acute decompensated heart failure: a prospective cohort study in China
.
J Am Heart Assoc
.
2016
Nov 16;
5
(
11
).
18.
Lassus
JP
,
Nieminen
MS
,
Peuhkurinen
K
,
Pulkki
K
,
Siirilä‐Waris
K
,
Sund
R
, et al
.
Markers of renal function and acute kidney injury in acute heart failure: definitions and impact on outcomes of the cardiorenal syndrome
.
Eur Heart J
.
2010
Nov;
31
(
22
):
2791
8
.
19.
Shouval
R
,
Hadanny
A
,
Shlomo
N
,
Iakobishvili
Z
,
Unger
R
,
Zahger
D
, et al
.
Machine learning for prediction of 30-day mortality after ST elevation myocardial infraction: an Acute Coronary Syndrome Israeli Survey data mining study
.
Int J Cardiol
.
2017
;
246
:
7
13
. .
20.
Elul
Y
,
Rosenberg
AA
,
Schuster
A
,
Bronstein
AM
,
Yaniv
Y
.
Meeting the unmet needs of clinicians from AI systems showcased for cardiology with deep-learning-based ECG analysis
.
Proc Natl Acad Sci USA
.
2021
;
118
(
24
):
e2020620118
. .
21.
Bart
NK
,
Pepe
S
,
Gregory
AT
,
Denniss
AR
.
Emerging roles of artificial intelligence (AI) in cardiology: benefits and barriers in a “brave new world”
.
Heart Lung Circ
.
2023
;
32
(
8
):
883
8
. .
22.
Cassol
C
,
Sharma
S
.
Nephrology lagging behind in machine learning utilization
.
Kidney Med
.
2021
;
3
(
5
):
693
5
. .
23.
Krisanapan
P
,
Tangpanithandee
S
,
Thongprayoon
C
,
Pattharanitima
P
,
Cheungpasitporn
W
.
Revolutionizing chronic kidney disease management with machine learning and artificial intelligence
.
J Clin Med
.
2023
(
8
):
3018
. .
24.
Kaur
A
,
Singh
S
,
Chandan
JS
,
Robbins
T
,
Patel
V
.
Qualitative exploration of digital chatbot use in medical education: a pilot study
.
Digit Health
.
2021
;
7
:
20552076211038151
. .
25.
Mortazavi
BJ
,
Downing
NS
,
Bucholz
EM
,
Dharmarajan
K
,
Manhapra
A
,
Li
SX
, et al
.
Analysis of machine learning techniques for heart failure readmissions
.
Circ Cardiovasc Qual Outcomes
.
2016
;
9
(
6
):
629
40
. .
26.
Hong
C
,
Sun
Z
,
Hao
Y
,
Dong
Z
,
Gu
Z
,
Huang
Z
.
Identifying patients with heart failure who are susceptible to de novo acute kidney injury: machine learning approach
.
JMIR Med Inform
.
2022
;
10
(
10
):
e37484
. .
27.
Liu
WT
,
Liu
XQ
,
Jiang
TT
,
Wang
MY
,
Huang
Y
,
Huang
YL
, et al
.
Using a machine learning model to predict the development of acute kidney injury in patients with heart failure
.
Front Cardiovasc Med
.
2022
;
9
:
911987
. .
28.
Thongprayoon
C
,
Hansrivijit
P
,
Bathini
T
,
Vallabhajosyula
S
,
Mekraksakit
P
,
Kaewput
W
, et al
.
Predicting acute kidney injury after cardiac surgery by machine learning approaches
.
J Clin Med
.
2020
;
9
(
6
):
1767
. .
29.
Tseng
PY
,
Chen
YT
,
Wang
CH
,
Chiu
KM
,
Peng
YS
,
Hsu
SP
, et al
.
Prediction of the development of acute kidney injury following cardiac surgery by machine learning
.
Crit Care
.
2020
;
24
(
1
):
478
. .
30.
Kuno
T
,
Mikami
T
,
Sahashi
Y
,
Numasawa
Y
,
Suzuki
M
,
Noma
S
, et al
.
Machine learning prediction model of acute kidney injury after percutaneous coronary intervention
.
Sci Rep
.
2022
;
12
(
1
):
749
. .
31.
Song
Z
,
Yang
Z
,
Hou
M
,
Shi
X
.
Machine learning in predicting cardiac surgery-associated acute kidney injury: a systemic review and meta-analysis
.
Front Cardiovasc Med
.
2022
;
9
:
951881
. .
32.
Li
Q
,
Lv
H
,
Chen
Y
,
Shen
J
,
Shi
J
,
Zhou
C
.
Development and validation of a machine learning predictive model for cardiac surgery-associated acute kidney injury
.
J Clin Med
.
2023
(
3
):
12
.
33.
Thongprayoon
C
,
Vaitla
P
,
Nissaisorakarn
V
,
Mao
MA
,
Genovez
JLZ
,
Kattah
AG
, et al
.
Clinically distinct subtypes of acute kidney injury on hospital admission identified by machine learning consensus clustering
.
Med Sci
.
2021
;
9
(
4
):
60
. .
34.
Rumsfeld
JS
,
Joynt
KE
,
Maddox
TM
.
Big data analytics to improve cardiovascular care: promise and challenges
.
Nat Rev Cardiol
.
2016
;
13
(
6
):
350
9
. .
35.
Urban
S
,
Błaziak
M
,
Jura
M
,
Iwanek
G
,
Ponikowska
B
,
Horudko
J
, et al
.
Machine learning approach to understand worsening renal function in acute heart failure
.
Biomolecules
.
2022
;
12
(
11
):
1616
. .
36.
Komorowski
M
,
Celi
LA
,
Badawi
O
,
Gordon
AC
,
Faisal
AA
.
The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care
.
Nat Med
.
2018
;
24
(
11
):
1716
20
. .
37.
Roggeveen
L
,
El Hassouni
A
,
Ahrendt
J
,
Guo
T
,
Fleuren
L
,
Thoral
P
, et al
.
Transatlantic transferability of a new reinforcement learning model for optimizing haemodynamic treatment for critically ill patients with sepsis
.
Artif Intell Med
.
2021
;
112
:
102003
. .
38.
Gala
D
,
Makaryus
AN
.
The utility of Language Models in cardiology: a narrative review of the benefits and concerns of ChatGPT-4
.
Int J Environ Res Public Health
.
2023
;
20
(
15
):
6438
. .
39.
Miao
J
,
Thongprayoon
C
,
Cheungpasitporn
W
.
Assessing the accuracy of ChatGPT on Core questions in glomerular disease
.
Kidney Int Rep
.
2023
;
8
(
8
):
1657
9
. .
40.
Miao
J
,
Thongprayoon
C
,
Garcia Valencia
OA
,
Krisanapan
P
,
Sheikh
MS
,
Davis
PW
, et al
.
Performance of ChatGPT on nephrology test questions
.
Clin J Am Soc Nephrol
.
2023
;
19
(
1
):
35
43
. .
41.
Nakaya
Y
,
Higaki
A
,
Yamaguchi
O
.
ChatGPT’s ability to classify virtual reality studies in cardiology
.
Eur Heart J Digit Health
.
2023
;
4
(
3
):
141
2
. .
42.
Skalidis
I
,
Cagnina
A
,
Luangphiphat
W
,
Mahendiran
T
,
Muller
O
,
Abbe
E
, et al
.
ChatGPT takes on the European Exam in Core Cardiology: an artificial intelligence success story
.
Eur Heart J Digit Health
.
2023
;
4
(
3
):
279
81
. .
43.
Suppadungsuk
S
,
Thongprayoon
C
,
Miao
J
,
Krisanapan
P
,
Qureshi
F
,
Kashani
K
, et al
.
Exploring the potential of chatbots in critical care nephrology
.
Medicines
.
2023
;
10
(
10
):
58
. .