# AI in nephrology, onco-nephrology, cancer care, and medicine

**Prepared as a physician-facing briefing, June 2026**

## Executive summary

AI is already being used in nephrology, but mostly as a clinical assistant rather than an autonomous doctor. The most mature applications are: predicting acute kidney injury, estimating CKD progression, analyzing dialysis-session data, quantifying pathology and imaging, supporting transplant decisions, summarizing charts, and triaging risk. Recent nephrology reviews describe the field as moving from experimental models toward clinical decision support, but with the same unresolved problem that has dogged alerts for years: a prediction is only useful if it changes management safely and measurably. ([Kidney Medicine][1])

For cancer-related renal disease, the most interesting area is **onco-nephrology**, where kidney injury is often multifactorial: the malignancy itself, chemotherapy, immune checkpoint inhibitors, antimicrobials, contrast, obstruction, tumor lysis, volume depletion, sepsis, surgery, radiation, and comorbid CKD can all overlap. AI is a natural fit because these cases are data-dense, time-sensitive, and full of competing explanations. Current examples include machine-learning prediction of AKI in cancer patients, risk prediction for severe cisplatin-associated AKI, explainable AI for AKI during immune checkpoint inhibitor therapy, and models that help identify patients with monoclonal gammopathy of renal significance who may need biopsy. ([PMC][2])

The biggest near-term promise is not “AI replaces the nephrologist.” It is more practical: AI watches the chart continuously, notices patterns earlier, reduces clerical burden, finds similar patients, flags nephrotoxic combinations, helps decide when a biopsy or ultrasound is worth it, and gives the clinician a better first draft of the differential.

---

## 1. What “AI” means in this context

In medicine, “AI” usually means one of several things:

| Type of AI                      | What it does                                                                                    | Nephrology examples                                                                                    |
| ------------------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ |
| **Predictive models**           | Forecasts an event from labs, vitals, medications, notes, comorbidities, and prior trajectories | AKI prediction, CKD progression, graft failure, dialysis hypotension                                   |
| **Computer vision**             | Reads images or pathology slides                                                                | Quantifying glomerulosclerosis, interstitial fibrosis, tubular atrophy, ultrasound fibrosis prediction |
| **Natural language processing** | Extracts meaning from clinical text                                                             | Finding AKI causes in notes, summarizing consults, identifying trial eligibility                       |
| **Generative AI / LLMs**        | Produces text, explanations, summaries, draft plans                                             | Chart summaries, patient instructions, prior-authorization drafts, literature synthesis                |
| **Multimodal AI**               | Combines text, images, labs, genomics, waveforms, and pathology                                 | Tumor board tools, onco-nephrology risk dashboards, transplant virtual biopsy                          |

The important shift is that older clinical decision support was usually rule-based: “creatinine rose by X, alert the team.” Newer AI can work across time and across modalities. It can notice that a patient’s creatinine, magnesium, albumin, cisplatin dose, neutrophil trend, recent contrast, and PPI exposure together imply a higher-risk trajectory than any one variable alone.

---

## 2. Current uses of AI in nephrology

### 2.1 Acute kidney injury prediction

AKI prediction has been one of the flagship nephrology AI use cases. A landmark 2019 Nature study from DeepMind and the U.S. Veterans Affairs system trained a deep-learning model on large-scale EHR data to predict AKI before it was clinically obvious. The model reportedly predicted 55.8% of inpatient AKI episodes and 90.2% of dialysis-requiring AKI episodes up to 48 hours in advance, although it also produced false alerts, roughly two false positives for every true positive. ([IDEAS/RePEc][3])

That is impressive, but the clinical lesson is subtle. Predicting AKI is not the same as improving outcomes. A randomized BMJ trial of electronic AKI alerts found no meaningful reduction in progression, dialysis, or death: 21.3% in the alert group versus 20.9% in usual care met the primary outcome. ([PubMed][4]) A 2024 JAMA trial of a “kidney action team” intervention improved delivery of recommendations but still did not significantly reduce a composite outcome of worsening AKI stage, dialysis, or mortality. ([JAMA Network][5])

The takeaway for your father: **AKI prediction is technically real, but implementation is the hard part.** The model needs to answer: What is the cause? What action should be taken? Who owns the action? Will the alert reduce harm or just add noise?

### 2.2 CKD progression and population risk

AI models are being used to predict CKD progression, kidney failure, hospitalization, mortality, and dialysis initiation. Recent systematic reviews focus on how models use EHR variables, labs, demographics, medication history, and comorbidities to identify patients likely to decline. ([PMC][6])

The practical value is population management. A health system could identify patients with stage 3 or 4 CKD who are likely to progress, then prioritize nephrology visits, SGLT2 inhibitor evaluation, blood-pressure optimization, medication review, transplant education, or vascular-access planning. A 2026 JAMIA study on ESRD outcome prediction emphasized integrated multisource data, interpretability, and bias mitigation as important for CKD management. ([OUP Academic][7])

### 2.3 Dialysis: intradialytic hypotension, hypertension, dry weight, and operations

Dialysis produces dense time-series data: blood pressure, ultrafiltration rate, blood flow, dialysate composition, symptoms, session interruptions, medication timing, and lab trends. That makes it well suited to AI.

Recent models have focused on predicting intradialytic hypotension and hypertension in real time. One Scientific Reports study used explainable deep learning to predict both hypotensive and hypertensive events during hemodialysis sessions. ([Nature][8]) Another recent review describes “precision dialysis” as the use of big data and AI to help personalize ultrafiltration, dry weight, and treatment parameters. ([PMC][9])

Near-term uses include safer ultrafiltration targets, earlier nursing interventions, better dry-weight estimation, detection of access problems, and staffing or chair optimization. These are not glamorous, but they matter because dialysis care is repetitive, measurable, and operationally complex.

### 2.4 Transplant nephrology

AI is being studied for donor-recipient matching, graft survival prediction, waitlist mortality, rejection risk, allocation policy, and transplant biopsy interpretation. A 2025 review on AI-driven kidney allocation concluded that AI and ML models may improve prediction of graft survival, recipient outcomes, and waitlist mortality, although fairness and explainability remain essential. ([PMC][10])

One especially interesting 2024 Nature Communications study developed a “virtual biopsy” system for kidney transplant patients. It used 14,032 day-zero kidney biopsies from 17 centers and 11 donor parameters to predict Banff lesions and the percentage of sclerotic glomeruli. The authors reported good discrimination, calibration, and robustness, and described a ready-to-use online application for physicians. ([Nature][11])

This does not eliminate biopsy, but it points toward a future where AI helps decide which biopsies are most needed, how to interpret borderline findings, and how to combine donor features, histology, and outcomes.

### 2.5 Renal pathology and imaging

Digital pathology is one of the clearest areas where AI can augment nephrology. The kidney biopsy is information-rich, but many features are labor-intensive and subject to interobserver variability. AI can quantify structures at scale.

Examples include deep-learning models to quantify glomerulosclerosis in whole-slide kidney biopsy images and models that estimate interstitial fibrosis and tubular atrophy from ultrasound images. ([JAMA Network][12]) Recent renal pathology reviews describe AI as moving from segmentation and counting tasks toward broader diagnostic support for glomerular disease and transplant pathology. ([Lippincott Journals][13])

For a nephrologist, the key idea is that pathology AI is not only about diagnosis. It can make pathology more quantitative: percent sclerosis, fibrosis burden, lesion maps, reproducible scoring, comparison to cohorts, and possibly prognosis tied to histologic patterns.

---

## 3. AI in onco-nephrology and cancer-related renal disease

Onco-nephrology may be one of the best subspecialty fits for AI because cancer patients accumulate risks quickly. A single patient may have metastatic disease, CKD, cisplatin exposure, immune checkpoint inhibitor therapy, IV contrast, antibiotics, PPI use, hypoalbuminemia, magnesium wasting, sepsis, and obstruction. A human consultant can reason through this, but the EHR does not make it easy.

### 3.1 Predicting AKI in cancer patients

A 2018 PLOS One study specifically addressed AKI prediction in cancer patients using heterogeneous and irregular clinical data. That phrase matters because cancer care does not generate clean, evenly spaced data. Patients cycle through infusion visits, admissions, surgeries, scans, antibiotics, and changing therapies. ([PMC][2])

The immediate application is a cancer-center AKI risk dashboard. It could monitor creatinine slope, eGFR, urine findings, volume status proxies, medication exposures, chemotherapy timing, contrast, sepsis markers, and obstruction risk, then flag not merely “AKI risk” but a ranked differential: cisplatin toxicity, ICI nephritis, prerenal azotemia, ATN, obstruction, tumor lysis, myeloma-related disease, or drug-induced interstitial nephritis.

### 3.2 Cisplatin-associated AKI

Cisplatin nephrotoxicity is one of the most concrete onco-nephrology AI-adjacent examples. A 2024 BMJ cohort study derived and externally validated a simple risk score for severe AKI after intravenous cisplatin. The study used a multicenter cohort of adults receiving first IV cisplatin between 2006 and 2022, with 11,766 patients in the derivation cohort and 12,951 in external validation. Severe cisplatin-associated AKI occurred in 5.2% of the derivation cohort and 3.3% of the validation cohort. Predictors included age, hypertension, diabetes, creatinine, hemoglobin, white blood cell count, platelets, albumin, magnesium, and cisplatin dose. The primary model had a C-statistic of 0.75, outperforming earlier models with C-statistics around 0.60 to 0.68. ([www.PhysiciansWeekly.com][14])

This is not “AI magic,” but it is exactly the kind of clinically useful model that can be embedded into oncology workflows. Before cisplatin, the system could estimate severe AKI risk, recommend hydration and magnesium protocols, suggest nephrology input for high-risk patients, identify alternatives when appropriate, and guide post-infusion lab timing.

A useful way to explain this to your father: this is like turning decades of clinical intuition into a risk-stratified workflow, then applying it consistently before the injury happens.

### 3.3 Immune checkpoint inhibitor kidney injury

Immune checkpoint inhibitors have created a new renal complication space. The American Society of Onco-Nephrology position statement notes that ICIs can cause a spectrum of kidney immune-related adverse events, most commonly acute tubulointerstitial nephritis, but also glomerular disease and electrolyte disturbances. ([PubMed][15])

This is an ideal AI problem because the clinical question is not simply “does creatinine rise?” It is: Is this ICI nephritis, ATN, prerenal AKI, obstruction, infection-related AKI, PPI-associated AIN, NSAID toxicity, or something else? The distinction matters because it affects steroids, cancer therapy interruption, rechallenge, biopsy, and prognosis.

A 2024 PLOS One study used interpretable machine learning to analyze AKI in patients receiving immune checkpoint inhibitors. The model predicted AKI within seven days using records from 616 ICI-treated patients, then used SHAP-based explanations and clustering to identify patient-specific drivers of the prediction. ([PLOS][16])

A 2024 meta-analysis found that all-cause AKI occurred in 7.4% of ICI-treated patients and ICI-related AKI in 3.2%. It also found associations between AKI and concomitant medications such as PPIs and NSAIDs, with higher odds for both all-cause and ICI-related AKI. ([OUP Academic][17])

The future use case is clear: an AI model could help decide which ICI-treated patients need urgent biopsy, which can be managed by holding nephrotoxins and rechecking labs, and which are high-risk for true immune-mediated nephritis. It could also support safer ICI rechallenge decisions after AKI.

### 3.4 Myeloma, monoclonal gammopathy, and MGRS

Renal disease from plasma-cell and monoclonal gammopathy disorders is another high-value area. The International Myeloma Working Group recommends a structured evaluation of renal impairment in multiple myeloma, including creatinine, eGFR, free light chains, 24-hour urine protein, electrophoresis, immunofixation, and biopsy in selected cases such as significant albuminuria or lower free-light-chain levels. ([International Myeloma Foundation][18])

A Mayo Clinic MGRS prediction tool uses clinical and laboratory features to estimate the probability that a kidney biopsy will show an MGRS lesion. In reported performance, a probability threshold of 0.10 had sensitivity of 98.9% and specificity of 50.5%; a threshold of 0.25 had sensitivity of 88.0% and specificity of 70.2%. ([Mayo Clinic][19])

This is highly relevant to cancer-associated nephrology because the key decision is often whether kidney findings reflect a monoclonal process requiring hematologic treatment, a coincidental MGUS plus unrelated CKD, diabetic kidney disease, amyloid, cast nephropathy, light-chain deposition disease, or another lesion. AI can help triage biopsy decisions and avoid both underdiagnosis and unnecessary invasive procedures.

### 3.5 Other cancer-related kidney injury use cases

AI is also likely to matter in:

**Tumor lysis syndrome:** predicting which patients need aggressive prophylaxis, rasburicase, closer electrolyte monitoring, or admission.

**Obstructive nephropathy:** combining imaging reports, hydronephrosis history, creatinine trajectory, tumor location, and symptoms to flag patients needing ultrasound, CT review, stent, or nephrostomy.

**VEGF inhibitor and TKI toxicity:** detecting early proteinuria, hypertension, thrombotic microangiopathy signals, and risk of progressive CKD.

**CAR-T and cytokine-release syndromes:** predicting AKI risk from inflammatory markers, hemodynamics, nephrotoxins, and ICU-level events.

**Pediatric oncology AKI:** pediatric oncology AKI is commonly associated with nephrotoxic drugs, but also with tumor infiltration or compression, chemotherapy, radiotherapy, immunotherapy, surgery, dehydration, and infection. ([Springer][20])

---

## 4. Potential uses: what a cancer-center onco-nephrology AI system could do

The most useful system would not be a chatbot that gives generic advice. It would be an integrated clinical assistant inside the cancer center’s EHR.

### 4.1 A renal risk “command center” for oncology patients

For every active cancer patient, the system could continuously calculate:

* AKI risk in the next 24 to 72 hours
* severe AKI risk after cisplatin or other nephrotoxic therapy
* ICI nephritis probability versus other AKI etiologies
* obstruction probability
* tumor lysis risk
* CKD progression risk after cancer therapy
* risk of needing dialysis during a hospitalization
* renal survivorship risk after cure or remission

The system would not merely alert. It would suggest the next useful action: repeat urinalysis, check urine microscopy, stop PPI or NSAID, adjust antimicrobial dosing, order renal ultrasound, modify cisplatin hydration, consult nephrology, consider biopsy, or discuss cancer-therapy modification.

### 4.2 A better differential diagnosis engine for AKI

The most interesting AI use in onco-nephrology is causal ranking. For example:

> “This patient’s AKI is temporally close to ICI cycle 4, but the model finds stronger support for volume depletion plus vancomycin/piperacillin-tazobactam exposure. No eosinophilia, no pyuria, no PPI exposure, no new proteinuria. Consider repeat labs and urine sediment before steroids.”

Or:

> “Creatinine rise after cisplatin cycle 2, new hypomagnesemia, low albumin, high baseline risk score, no obstruction on recent imaging. High likelihood of cisplatin-associated tubular injury.”

This would be more useful than a generic AKI alert.

### 4.3 AI-assisted renal biopsy decision-making

For cancer patients, biopsy decisions can be difficult because bleeding risk, thrombocytopenia, prognosis, and treatment urgency all matter. AI could help estimate the expected diagnostic yield and likely management impact of biopsy.

Examples:

* ICI AKI: biopsy versus empiric steroids
* MGRS: biopsy versus observation
* nephrotic syndrome in malignancy: membranous nephropathy, amyloid, minimal change, TMA, diabetic kidney disease
* post-transplant cancer patients: rejection, BK, drug toxicity, recurrent disease

The MGRS prediction tool and transplant virtual-biopsy work are early examples of this direction. ([Mayo Clinic][19])

### 4.4 Cancer-therapy planning with renal risk built in

Oncology decisions often include renal tradeoffs. AI could present oncologists and nephrologists with patient-specific risk under different choices:

* cisplatin versus carboplatin
* ICI rechallenge versus discontinuation
* contrast-enhanced imaging now versus alternative imaging or prophylaxis
* nephrotoxic antimicrobial choices
* VEGF inhibitor continuation with proteinuria
* dose adjustment under dynamic eGFR estimates

The goal is shared decision-making with quantified risk, not algorithmic control.

### 4.5 Survivorship nephrology

Cancer survivors increasingly live long enough for CKD, hypertension, proteinuria, and cardiovascular risk to matter. AI could identify survivors who need long-term kidney monitoring after nephrectomy, cisplatin, ifosfamide, radiation, stem-cell transplant, CAR-T complications, or recurrent AKI. This is a major future area because survivorship is fragmented across oncology, primary care, nephrology, and cardiology.

---

## 5. AI in medicine and cancer generally

### 5.1 The FDA landscape: mostly imaging so far

As of the FDA’s current public AI-enabled medical device work, the agency maintains an AI-enabled medical device list and emphasizes safety, effectiveness, transparency, and lifecycle management. ([U.S. Food and Drug Administration][21])

A 2025 npj Digital Medicine analysis of 1,016 FDA AI/ML device authorizations found that image analysis dominated the field. Among 736 unique devices, 84.4% were based on images, 14.5% on signals, 0.7% on omics, and 0.4% on tabular EHR data. Radiology was the dominant specialty. The same analysis found no large-language-model devices in the FDA-authorized set as of its December 2024 cutoff. ([Nature][22])

This matters because the public imagination is focused on chatbots, but regulated medical AI is still mostly radiology, cardiology signals, pathology, and workflow tools.

### 5.2 Cancer screening and diagnosis

The National Cancer Institute describes AI as relevant across cancer mechanisms, screening, diagnosis, drug discovery, surveillance, and care delivery. ([Cancer.gov][23]) In cancer, the most mature clinical applications are imaging and pathology: mammography, lung nodule detection, prostate MRI, colonoscopy polyp detection, dermatology image classification, digital pathology, and biomarker discovery. Reviews of oncology AI repeatedly find that diagnosis, radiology, and pathology make up the largest share of real-world oncology AI applications. ([PMC][24])

### 5.3 Radiotherapy planning

Radiation oncology has a practical AI use case: auto-contouring. Treatment planning requires segmentation of tumors and organs at risk. Deep-learning auto-segmentation can reduce manual contouring time and improve consistency, especially for head and neck, prostate, thoracic, and pelvic cancers. Recent reviews describe clinical use of deep learning for radiotherapy auto-segmentation. ([PMC][25]) A Mayo Clinic study reported major time savings for head and neck segmentation using deep-learning automation without compromising contour accuracy. ([Mayo Clinic][26])

This is a good example to show your father because it is not abstract. AI saves clinician time on a defined, repetitive, high-skill task, while the physician remains responsible for review.

### 5.4 Clinical trial matching

Trial matching is another strong oncology AI use case. Cancer trial eligibility criteria are long, complex, and often hidden in unstructured text. TrialGPT, published in Nature Communications in 2024, used large language models for patient-to-trial matching. It performed trial retrieval, eligibility matching, and ranking; its rankings correlated strongly with human judgments, and a user study found it reduced screening time by 42.6%. ([Nature][27])

For a cancer center, this is operationally important. A patient may have a rare mutation, renal impairment, prior ICI exposure, or exclusion criteria buried in notes. AI can screen more patients more consistently, then hand the final decision to a research nurse, oncologist, or trial team.

### 5.5 Drug discovery and toxicity prediction

AI is being used in target discovery, protein structure prediction, compound screening, toxicity prediction, biomarker selection, and trial design. In kidney care, one particularly relevant subfield is drug-induced kidney injury prediction. Reviews describe AI and ML models for predicting drug-induced kidney injury from chemical, biological, and clinical data. ([MDPI][28])

For oncology, this could eventually help identify cancer therapies less likely to cause tubular injury, TMA, podocytopathy, electrolyte disorders, or progressive CKD.

### 5.6 Generative AI: documentation, summarization, education, and second-pass reasoning

Large language models are being tested for clinical note generation, summarization, handoff support, literature review, patient communication, and administrative work. Reviews of clinical text summarization describe rapid growth in LLM use for synthesizing notes and records. ([JMIR][29]) A 2025 study compared ambient LLM-generated clinical notes with physician-authored reference notes across five specialties, reflecting how fast this area is moving into real clinical workflows. ([Frontiers][30])

For nephrology and oncology, the obvious uses are:

* summarizing long hospitalizations before consult
* extracting creatinine baselines and AKI timelines
* listing nephrotoxic exposures
* drafting consult notes
* translating patient instructions into plain language
* summarizing pathology, imaging, genomics, and treatment history
* helping prepare tumor board or nephrology conference cases

But LLMs must be supervised. A 2026 JAMA Network Open study of multiple LLMs on diagnostic vignettes found weaknesses in differential diagnosis and uncertainty handling, even when models performed better on final diagnosis or management questions. ([JAMA Network][31])

---

## 6. Major cautions

### 6.1 Prediction is not outcome improvement

The AKI alert literature is the warning sign. AI can predict AKI earlier, but unless the alert is specific, actionable, trusted, and connected to a workflow, it may not improve dialysis, mortality, or progression outcomes. ([PubMed][4])

### 6.2 Cancer patients are distribution-shift machines

Cancer care changes quickly. New immunotherapies, antibody-drug conjugates, CAR-T products, bispecifics, targeted therapies, supportive-care protocols, and trial regimens can make last year’s model stale. A model trained at one cancer center may fail at another because patient mix, formularies, lab timing, imaging frequency, and admission practices differ.

### 6.3 Explainability matters more in nephrology than in many fields

A nephrologist does not just need a risk score. He needs the why. Is the model worried because of creatinine slope, albumin, cisplatin dose, NSAID exposure, urine protein, hydronephrosis, hypotension, or vancomycin level? Explainable models, like the SHAP-based ICI-AKI work, are more clinically persuasive because they show which features drove the prediction. ([PLOS][16])

### 6.4 Bias and access

CKD, ESRD, transplantation, and cancer outcomes are deeply shaped by race, socioeconomic status, referral patterns, insurance, geography, and access to specialty care. AI can either reduce disparities by finding missed risk earlier, or amplify disparities by learning from biased historical care.

### 6.5 LLM hallucination

Generative AI can sound confident and be wrong. In clinical medicine, this is dangerous. The best current use is drafting and summarizing under physician review, not unsupervised diagnosis or treatment.

---

## 7. Suggested way to explain this to your father

I would show him three concrete examples first, then broaden out.

### Example 1: AKI prediction

Start with the DeepMind/VA AKI model. It shows that AI can see deterioration before a physician might, but then show the randomized AKI alert trials to make the point that alerts alone do not save patients. ([IDEAS/RePEc][3])

The discussion question for him:

> “What action would have to follow an AKI prediction for this to actually change outcomes?”

That will get him thinking like a chief, not like a gadget buyer.

### Example 2: Cisplatin AKI risk

Then show the BMJ cisplatin-associated AKI risk score. This is closer to immediate clinical use: before a nephrotoxic cancer therapy, identify who is at high risk and change monitoring, hydration, magnesium, or treatment planning. ([www.PhysiciansWeekly.com][14])

The discussion question:

> “Would this have changed which patients you wanted to see before chemotherapy?”

### Example 3: ICI nephritis and competing etiologies

Then show the immune checkpoint inhibitor AKI work. This is intellectually interesting because the model is not merely predicting creatinine rise. It is trying to help interpret a messy, high-stakes differential. ([PubMed][15])

The discussion question:

> “Could AI help decide who needs biopsy versus empiric steroids versus simple medication cleanup?”

### Then broaden to medicine

Once he sees the nephrology examples, the broader medicine examples will make more sense: radiology reads, radiation auto-contouring, pathology quantification, trial matching, drug discovery, and chart summarization. FDA-authorized AI devices are still overwhelmingly imaging-based, which is useful context because it separates real deployment from chatbot hype. ([U.S. Food and Drug Administration][21])

---

## 8. Bottom line

The best way to describe AI in nephrology today is:

> AI is becoming a second layer of clinical attention. It is good at watching many variables at once, noticing risk earlier, quantifying images and slides, summarizing messy records, and standardizing repetitive decisions. It is not yet good enough to replace judgment, explain tradeoffs, or own the consequences of treatment.

In onco-nephrology, the opportunity is especially strong because kidney injury in cancer is high stakes, data-rich, and causally complex. The killer application is probably not a generic AI doctor. It is a cancer-center renal intelligence system that tracks AKI risk, explains likely causes, anticipates nephrotoxicity, helps triage biopsy, supports therapy decisions, and follows survivors for CKD.

For your father, I would frame it this way: AI is not replacing the nephrologist’s mind. It is replacing the parts of the hospital that currently force that mind to hunt through fragmented data, delayed signals, and incomplete histories.

[1]: https://www.kidneymedicinejournal.org/article/S2590-0595%2824%2900138-9/fulltext?utm_source=chatgpt.com "Artificial Intelligence in Nephrology: Clinical Applications ..."
[2]: https://pmc.ncbi.nlm.nih.gov/articles/PMC6053162/?utm_source=chatgpt.com "Predicting acute kidney injury in cancer patients using ... - PMC"
[3]: https://ideas.repec.org/a/nat/nature/v572y2019i7767d10.1038_s41586-019-1390-1.html?utm_source=chatgpt.com "A clinically applicable approach to continuous prediction of"
[4]: https://pubmed.ncbi.nlm.nih.gov/33461986/?utm_source=chatgpt.com "Electronic health record alerts for acute kidney injury - PubMed"
[5]: https://jamanetwork.com/journals/jama/article-abstract/2825492?utm_source=chatgpt.com "Early, Individualized Recommendations for Hospitalized ..."
[6]: https://pmc.ncbi.nlm.nih.gov/articles/PMC11166249/?utm_source=chatgpt.com "Predicting the Progression of Chronic Kidney Disease - PMC"
[7]: https://academic.oup.com/jamia/article/33/1/26/8223429?utm_source=chatgpt.com "Enhancing end-stage renal disease outcome prediction"
[8]: https://www.nature.com/articles/s41598-023-45282-1?utm_source=chatgpt.com "Real-time dual prediction of intradialytic hypotension and ..."
[9]: https://pmc.ncbi.nlm.nih.gov/articles/PMC11342780/?utm_source=chatgpt.com "Precision Dialysis: Leveraging Big Data and Artificial ... - PMC"
[10]: https://pmc.ncbi.nlm.nih.gov/articles/PMC12619412/?utm_source=chatgpt.com "Artificial intelligence–driven kidney organ allocation - PMC - NIH"
[11]: https://www.nature.com/articles/s41467-023-44595-z "A Machine Learning-Driven Virtual Biopsy System For Kidney Transplant Patients | Nature Communications"
[12]: https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2775298?utm_source=chatgpt.com "Deep Learning Model to Quantify Glomerulosclerosis in ..."
[13]: https://journals.lww.com/kidney360/fulltext/2024/07000/role_of_artificial_intelligence_in_kidney.19.aspx?utm_source=chatgpt.com "Role of Artificial Intelligence in Kidney Pathology:..."
[14]: https://www.physiciansweekly.com/post/simple-risk-score-predicts-severe-cisplatin-linked-acute-kidney-injury "Simple Risk Score Predicts Severe Cisplatin-Linked Acute Kidney Injury | www.PhysiciansWeekly.com"
[15]: https://pubmed.ncbi.nlm.nih.gov/39455026/?utm_source=chatgpt.com "a position statement from the American Society of Onco ..."
[16]: https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0298673&utm_source=chatgpt.com "Interpretable machine learning-based individual analysis of ..."
[17]: https://academic.oup.com/ckj/article/17/1/sfad292/7453684?utm_source=chatgpt.com "a meta-analysis of occurrence rate, risk factors and mortality"
[18]: https://www.myeloma.org/resource-library/imwg-recommendations-managing-multiple-myeloma-related-renal-impairment?utm_source=chatgpt.com "IMWG Recommendations: Renal Impairment in MM"
[19]: https://mayoclinic.elsevierpure.com/en/publications/the-mayo-mgrs-prediction-tool-calculates-the-risk-of-finding-mono?utm_source=chatgpt.com "The Mayo MGRS Prediction Tool calculates the risk of ..."
[20]: https://link.springer.com/article/10.1186/s12882-025-04085-4?utm_source=chatgpt.com "Urinary biomarkers in prediction of subclinical acute kidney ..."
[21]: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices "Artificial Intelligence-Enabled Medical Devices | FDA"
[22]: https://www.nature.com/articles/s41746-025-01800-1 "How AI is used in FDA-authorized medical devices: a taxonomy across 1,016 authorizations | npj Digital Medicine"
[23]: https://www.cancer.gov/research/infrastructure/artificial-intelligence?utm_source=chatgpt.com "Artificial Intelligence (AI) and Cancer"
[24]: https://pmc.ncbi.nlm.nih.gov/articles/PMC12128506/?utm_source=chatgpt.com "Current AI technologies in cancer diagnostics and treatment"
[25]: https://pmc.ncbi.nlm.nih.gov/articles/PMC11839850/?utm_source=chatgpt.com "Deep learning for autosegmentation for radiotherapy ... - PMC"
[26]: https://www.mayoclinic.org/medical-professionals/cancer/news/time-savings-of-76-percent-using-deep-learning-automation-in-head-and-neck-cancer/mac-20547794?utm_source=chatgpt.com "Time savings of 76% using deep learning automation in ..."
[27]: https://www.nature.com/articles/s41467-024-53081-z?utm_source=chatgpt.com "Matching patients to clinical trials with large language models"
[28]: https://www.mdpi.com/1424-8247/17/11/1550?utm_source=chatgpt.com "Artificial Intelligence and Machine Learning Models for ..."
[29]: https://www.jmir.org/2025/1/e68998?utm_source=chatgpt.com "Scientific Evidence for Clinical Text Summarization Using ..."
[30]: https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1691499/full?utm_source=chatgpt.com "Assessing the quality of AI-generated clinical notes"
[31]: https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2847679?utm_source=chatgpt.com "Large Language Model Performance and Clinical ..."