āļ„āļđāđˆāļĄāļ·āļ­āļ­āļļāļ•āļŠāļēāļŦāļāļĢāļĢāļĄ

āļāļēāļĢāļ—āļģāļ™āļēāļĒāļāļēāļĢāļšāļēāļ”āđ€āļˆāđ‡āļšāļ—āļĩāđˆāđ„āļ•āđ€āļ‰āļĩāļĒāļšāļžāļĨāļąāļ™āđ‚āļ”āļĒ AI

AI prediction of acute kidney injury (AKI) uses machine learning on electronic health record data to estimate which hospitalized patients are likely to develop AKI in the next day or two.

āļŸāļąāļ‡āļ„āļģāđāļ™āļ°āļ™āļģāļ™āļĩāđ‰āļšāļĢāļĢāļĒāļēāļĒāđ‚āļ”āļĒ AI · ~3 min

āļšāļ™āļŦāļ™āđ‰āļēāļ™āļĩāđ‰āļ­āđˆāļēāļ™ 4 āļ™āļēāļ—āļĩ
  1. āļ āļēāļžāļĢāļ§āļĄ
  2. āđ€āļˆāļēāļ°āļĨāļķāļ
  3. āļāļēāļĢāđƒāļŠāđ‰āļ‡āļēāļ™āļˆāļĢāļīāļ‡āđƒāļ™āđ‚āļĨāļāđāļŦāđˆāļ‡āļ„āļ§āļēāļĄāđ€āļ›āđ‡āļ™āļˆāļĢāļīāļ‡
  4. The Future of AI Prediction of Acute Kidney Injury
  5. āļāļēāļĢāļ™āļģāđ„āļ›āļ›āļāļīāļšāļąāļ•āļī
  6. āļŠāļģāļĢāļ§āļˆāļ•āđˆāļ­āđ„āļ›
  7. āļ„āļģāļ–āļēāļĄāļ—āļĩāđˆāļžāļšāļšāđˆāļ­āļĒ

āļ āļēāļžāļĢāļ§āļĄ

The data includes creatinine trends, urine output, medications, vital signs and diagnoses. It matters because AKI is common in hospitals and often silent until creatinine rises. Acting early, for example by stopping a drug that harms the kidneys or correcting fluid status, may prevent damage. Still, alerts on their own have not reliably improved outcomes.

āđ€āļˆāļēāļ°āļĨāļķāļ

Acute kidney injury is a sudden drop in kidney function. Clinicians define it with the KDIGO criteria: serum creatinine rises by at least 0.3 mg/dL within 48 hours, or rises to 1.5 times baseline within seven days, or urine output stays below 0.5 mL/kg per hour for six hours. The catch is that creatinine is a lagging marker. By the time it rises, the injury may be a day or more old. Prediction models try to get ahead of that lag. They learn from many past admissions which patterns came before AKI: small creatinine drifts, rising blood urea nitrogen, low blood pressure, sepsis, recent surgery, chronic kidney disease, and exposure to drugs that can harm the kidneys, such as vancomycin, aminoglycosides, NSAIDs or IV contrast. The best-known study came from DeepMind and the US Department of Veterans Affairs and was published in Nature in 2019. A recurrent neural network trained on VA records predicted a little over half of inpatient AKI episodes up to 48 hours ahead. It produced roughly two false alerts for every true one. It also performed worse for women, who made up only a small share of VA patients. That shows how the training population shapes who a model serves well. A common misconception is that DeepMind's Streams app at the Royal Free Hospital in London was AI. It wasn't. Streams sent clinicians' phones alerts from the NHS England AKI algorithm, which is a rule-based comparison of creatinine values. Streams became better known for a 2017 finding by the UK Information Commissioner's Office that the hospital had not complied with data protection law when it shared patient records with DeepMind. The bigger lesson is that prediction is not the same as benefit. Randomized trials at Yale, led by F. Perry Wilson, tested electronic alerts that fired once AKI was present. The alerts did not reduce outcomes such as dialysis or death overall. An alert helps only if someone acts on it and an effective action exists for that patient.

āļāļēāļĢāđƒāļŠāđ‰āļ‡āļēāļ™āļˆāļĢāļīāļ‡āđƒāļ™āđ‚āļĨāļāđāļŦāđˆāļ‡āļ„āļ§āļēāļĄāđ€āļ›āđ‡āļ™āļˆāļĢāļīāļ‡

A model re-scores every inpatient each time a new lab result posts. It flags a post-surgical patient on vancomycin and piperacillin-tazobactam whose creatinine has crept from 0.8 to 1.0 mg/dL, before the rise meets the formal KDIGO definition of AKI.

A cardiac surgery team uses a risk model before the operation to choose which bypass patients get hourly urine output monitoring and a plan to avoid IV contrast in the first days after surgery.

A clinical pharmacist gets a daily list of high-risk patients and checks each one for risky drug combinations, such as an NSAID with an ACE inhibitor and a diuretic, then recommends stopping or swapping one of them.

An ICU dashboard combines hourly urine output with fluid balance and warns of falling output. The nurse checks first for a blocked urinary catheter, which is a common non-kidney reason for low recorded output.

The Future of AI Prediction of Acute Kidney Injury

Several directions are being studied, including combining model risk scores with kidney stress biomarkers measured in urine, and pairing alerts with specific care bundles instead of a bare warning. Small trials of biomarker-guided care bundles after cardiac surgery have reported fewer AKI cases. It is still unclear whether AI-driven alerts improve dialysis rates, kidney recovery or survival at scale, so hospitals considering these tools should look for prospective, randomized evidence and check performance on their own patients. Fairness checks across sex, race and care setting will likely become routine, because the VA study showed how much a model can reflect its training population.

āļāļēāļĢāļ™āļģāđ„āļ›āļ›āļāļīāļšāļąāļ•āļī

āļ—āļģāđ„āļĄāļĄāļąāļ™āļ–āļķāļ‡āļŠāļģāļ„āļąāļ

  • āļšāļĢāļīāļšāļ—āđāļĨāļ°āļāļŽ. āļšāļĢāļīāļšāļ—āļ—āļēāļ‡āļ­āļļāļ•āļŠāļēāļŦāļāļĢāļĢāļĄāđ€āļ›āđ‡āļ™āļ•āļąāļ§āļāļģāļŦāļ™āļ”āļ§āđˆāļēāđāļ™āļ§āļ„āļīāļ” AI āļˆāļ°āļĢāļ­āļ”āļˆāļēāļāļāļēāļĢāļŠāļąāļĄāļœāļąāļŠāļāļąāļšāļ„āļ§āļēāļĄāđ€āļ›āđ‡āļ™āļˆāļĢāļīāļ‡āļŦāļĢāļ·āļ­āđ„āļĄāđˆ
  • āļāļēāļĢāļ„āļ§āļšāļ„āļļāļĄāļ„āļļāļ“āļ āļēāļž. āļ‚āđ‰āļ­āļˆāļģāļāļąāļ”āļ‚āļ­āļ‡āđ‚āļ”āđ€āļĄāļ™āļĄāļĩāļ­āļīāļ—āļ˜āļīāļžāļĨāļ•āđˆāļ­āļ­āļąāļ•āļĢāļēāļ‚āđ‰āļ­āļœāļīāļ”āļžāļĨāļēāļ”āļ—āļĩāđˆāļĒāļ­āļĄāļĢāļąāļšāđ„āļ”āđ‰āđāļĨāļ°āđāļšāļšāļˆāļģāļĨāļ­āļ‡āļāļēāļĢāļ„āļ§āļšāļ„āļļāļĄāļ”āļđāđāļĨ
  • āļŠāļĢāđ‰āļēāļ‡āļ—āļēāļ‡āđ€āļĨāļ·āļ­āļ. āļāļēāļĢāļ›āļĢāļąāļšāđƒāļŠāđ‰āļ—āļĩāđˆāļ›āļĢāļ°āļŠāļšāļ„āļ§āļēāļĄāļŠāļģāđ€āļĢāđ‡āļˆāļˆāļ°āļ›āļĢāļąāļšāļ„āļ§āļēāļĄāļŠāļēāļĄāļēāļĢāļ–āļ—āļēāļ‡āđ€āļ—āļ„āļ™āļīāļ„āđƒāļŦāđ‰āļŠāļ­āļ”āļ„āļĨāđ‰āļ­āļ‡āļāļąāļšāđ€āļ§āļīāļĢāđŒāļāđ‚āļŸāļĨāļ§āđŒāđāļ™āļ§āļŦāļ™āđ‰āļē

āļĢāļ°āļ§āļąāļ‡

  • āļ‚āđ‰āļ­āļāļģāļŦāļ™āļ”āļ”āđ‰āļēāļ™āļāļŽāļĢāļ°āđ€āļšāļĩāļĒāļšāļ­āļēāļˆāļ—āļģāđƒāļŦāđ‰āļ•āđ‰āļ™āđāļšāļšāļ—āļĩāđˆāđāļ‚āđ‡āļ‡āđāļāļĢāđˆāļ‡āđ€āļ›āđ‡āļ™āđ‚āļĄāļ†āļ°āđ„āļ”āđ‰
  • āļ‚āđ‰āļ­āļĄāļđāļĨāđƒāļ™āļ­āļ”āļĩāļ•āļ­āļēāļˆāđ€āļ‚āđ‰āļēāļĢāļŦāļąāļŠāļ­āļ„āļ•āļīāļ—āļĩāđˆāđ€āļ›āđ‡āļ™āļ­āļąāļ™āļ•āļĢāļēāļĒāļ•āđˆāļ­āļŠāļļāļĄāļŠāļ™āļšāļēāļ‡āđāļŦāđˆāļ‡
  • āļĢāļ°āļšāļšāđ€āļ”āļīāļĄāļŠāļēāļĄāļēāļĢāļ–āļŠāļĢāđ‰āļēāļ‡āļ›āļąāļāļŦāļēāļ„āļ­āļ‚āļ§āļ”āđƒāļ™āļāļēāļĢāļšāļđāļĢāļ“āļēāļāļēāļĢāđāļĨāļ°āļ•āđ‰āļ™āļ—āļļāļ™āđāļ­āļšāđāļāļ‡āđ„āļ”āđ‰

āļ‚āļąāđ‰āļ™āļ•āļ­āļ™āļ•āđˆāļ­āđ„āļ›

  1. āđƒāļŦāđ‰āļœāļđāđ‰āđ€āļŠāļĩāđˆāļĒāļ§āļŠāļēāļāđ‚āļ”āđ€āļĄāļ™āļĄāļĩāļŠāđˆāļ§āļ™āļĢāđˆāļ§āļĄāļ•āļąāđ‰āļ‡āđāļ•āđˆāļāļēāļĢāļāļģāļŦāļ™āļ”āļāļĢāļ­āļšāļ›āļąāļāļŦāļēāđ„āļ›āļˆāļ™āļ–āļķāļ‡āļāļēāļĢāļ›āļĢāļ°āđ€āļĄāļīāļ™
  2. āļ­āļ­āļāđāļšāļšāđ€āļŠāđ‰āļ™āļ—āļēāļ‡āļāļēāļĢāļ•āļĢāļ§āļˆāļŠāļ­āļšāđāļĨāļ°āđ€āļ­āļāļŠāļēāļĢāļ›āļĢāļ°āļāļ­āļšāļāđˆāļ­āļ™āļāļēāļĢāđ€āļ›āļīāļ”āļ•āļąāļ§
  3. āļ•āļĢāļ§āļˆāļŠāļ­āļšāļāļēāļĢāļ›āļāļīāļšāļąāļ•āļīāļ•āļēāļĄāļ‚āđ‰āļ­āļāļģāļŦāļ™āļ”āđāļĨāļ°āļ āļēāļĢāļ°āļœāļđāļāļžāļąāļ™āļ”āđ‰āļēāļ™āļ„āļ§āļēāļĄāļ›āļĨāļ­āļ”āļ āļąāļĒāļ•āļąāđ‰āļ‡āđāļ•āđˆāđ€āļ™āļīāđˆāļ™āđ†
  4. āđ€āļ›āļīāļ”āļ•āļąāļ§āđ€āļ›āđ‡āļ™āļĢāļ°āļĒāļ°āđ‚āļ”āļĒāļĄāļĩāđ€āļāļ“āļ‘āđŒāļāļēāļĢāļŦāļĒāļļāļ”āđāļĨāļ°āļĒāđ‰āļ­āļ™āļāļĨāļąāļšāļ—āļĩāđˆāļŠāļąāļ”āđ€āļˆāļ™

āļŠāļģāļĢāļ§āļˆāļ•āđˆāļ­āđ„āļ›

āļˆāļ”āļŦāļĄāļēāļĒāļ‚āđˆāļēāļ§āļŸāļĢāļĩ

āļĢāļąāļšāļŠāļĢāļļāļ› AI āļĢāļēāļĒāļ§āļąāļ™

āđ€āļĢāļ·āđˆāļ­āļ‡āļĢāļēāļ§ AI āļ—āļĩāđˆāļ•āļĢāļ§āļˆāļŠāļ­āļšāđāļĨāđ‰āļ§ 3 āđ€āļĢāļ·āđˆāļ­āļ‡āļ—āļļāļāļ§āļąāļ™āļ—āļģāļ‡āļēāļ™ āđ€āļŠāđ‰āļē, āđ€āļ‚āļĩāļĒāļ™āđ€āļ›āđ‡āļ™āļ āļēāļĐāļēāļ­āļąāļ‡āļāļĪāļĐāļ˜āļĢāļĢāļĄāļ”āļē āļŸāļĢāļĩāļ•āļĨāļ­āļ”āđ„āļ› āđ„āļĄāđˆāļĄāļĩāđ‚āļ†āļĐāļ“āļē

āļ­āļĩāđ€āļĄāļĨāļŦāļ™āļķāđˆāļ‡āļ‰āļšāļąāļšāļ—āļļāļāļ§āļąāļ™āļ—āļģāļ‡āļēāļ™ āļĒāļāđ€āļĨāļīāļāļāļēāļĢāļŠāļĄāļąāļ„āļĢāđƒāļ™āļ„āļĨāļīāļāđ€āļ”āļĩāļĒāļ§ āđ€āļĢāļēāđ„āļĄāđˆāđ€āļ„āļĒāļ‚āļēāļĒāļŦāļĢāļ·āļ­āđāļŠāļĢāđŒāļ—āļĩāđˆāļ­āļĒāļđāđˆāļ­āļĩāđ€āļĄāļĨāļ‚āļ­āļ‡āļ„āļļāļ“

āļ—āļ”āļŠāļ­āļšāļ•āļąāļ§āđ€āļ­āļ‡

Take the AI Prediction of Acute Kidney Injury quiz

āļ‚āđ‰āļ­āđ€āļŠāļ™āļ­āđāļ™āļ°āļ—āļąāļ™āļ—āļĩāļŠāļģāļŦāļĢāļąāļšāļ—āļļāļāļ„āļģāļ•āļ­āļš āđāļĨāļ°āđƒāļšāļĢāļąāļšāļĢāļ­āļ‡āļ—āļĩāđˆāļŠāļēāļĄāļēāļĢāļ–āđāļŠāļĢāđŒāđ„āļ”āđ‰āļžāļĢāđ‰āļ­āļĄ ID āļ—āļĩāđˆāļ•āļĢāļ§āļˆāļŠāļ­āļšāđ„āļ”āđ‰āđ€āļĄāļ·āđˆāļ­āļ„āļļāļ“āļœāđˆāļēāļ™āļŦāļĨāļąāļāļŠāļđāļ•āļĢ

āđ€āļĢāļīāđˆāļĄāđāļšāļšāļ—āļ”āļŠāļ­āļš

āļŠāļ™āļąāļšāļŠāļ™āļļāļ™āļāļēāļĢāļĻāļķāļāļĐāļē AI āļŸāļĢāļĩ AI Understanding āđ€āļ›āđ‡āļ™āļ­āļ‡āļ„āđŒāļāļĢāđ„āļĄāđˆāđāļŠāļ§āļ‡āļŦāļēāļœāļĨāļāļģāđ„āļĢ 501(c)(3) — āđ„āļĄāđˆāļĄāļĩāđ‚āļ†āļĐāļ“āļē āđ„āļĄāđˆāļĄāļĩāļāļģāđāļžāļ‡āđ€āļ‡āļīāļ™, āļ•āļĨāļ­āļ”āđ„āļ› āļšāļĢāļīāļˆāļēāļ„

āļ„āļģāļ–āļēāļĄāļ—āļĩāđˆāļžāļšāļšāđˆāļ­āļĒ

What is AI Prediction of Acute Kidney Injury?

AI prediction of acute kidney injury (AKI) uses machine learning on electronic health record data to estimate which hospitalized patients are likely to develop AKI in the next day or two. The data includes creatinine trends, urine output, medications, vital signs and diagnoses. It matters because AKI is common in hospitals and often silent until creatinine rises. Acting early, for example by stopping a drug that harms the kidneys or correcting fluid status, may prevent damage. Still, alerts on their own have not reliably improved outcomes.

Under the KDIGO criteria described in this guide, which creatinine change within 48 hours qualifies as acute kidney injury?

KDIGO defines AKI as a creatinine rise of at least 0.3 mg/dL within 48 hours, a rise to 1.5 times baseline within seven days, or low urine output for six hours.

Why do AKI models try to predict injury instead of waiting for creatinine to rise?

Creatinine rises after the kidneys have already been injured, so waiting for it wastes the window in which prevention could help.

In the 2019 DeepMind and VA study, roughly how many false alerts came with each correct AKI prediction?

The model predicted a little over half of inpatient AKI episodes up to 48 hours ahead, with roughly two false alerts for every true one.

Which group did the VA-trained AKI model perform worse for, and why?

Women were underrepresented in VA data, and the model did worse for them. This shows how the training population shapes who a model serves well.

What did DeepMind's Streams app at the Royal Free Hospital actually deliver to clinicians?

Streams sent alerts from a rule-based creatinine comparison to clinicians' phones. It was not a machine learning predictor, even though it is often described that way.

āđ€āļĢāļĩāļĒāļ™āļĢāļđāđ‰āļ•āđˆāļ­āđ„āļ›

āļ„āļģāđāļ™āļ°āļ™āļģāļ—āļĩāđˆāđ€āļāļĩāđˆāļĒāļ§āļ‚āđ‰āļ­āļ‡

āļ„āļģāđāļ™āļ°āļ™āļģāđ€āļžāļīāđˆāļĄāđ€āļ•āļīāļĄāļ—āļĩāđˆāđ€āļĨāļ·āļ­āļāļŠāļģāļŦāļĢāļąāļšāļŦāļąāļ§āļ‚āđ‰āļ­āļ™āļĩāđ‰

āļ•āđˆāļ­āđ„āļ›āļ„āļģāđāļ™āļ°āļ™āļģāļ•āđˆāļ­āđ„āļ›

āļāļēāļĢāļ—āļģāļ™āļēāļĒāļāļēāļĢāļšāļēāļ”āđ€āļˆāđ‡āļšāļˆāļēāļāđāļĢāļ‡āļāļ”āļ”āļąāļ™āļˆāļēāļ AI

āļ­āļļāļ•āļŠāļēāļŦāļāļĢāļĢāļĄ

āļĢāļąāļšāđƒāļšāļĢāļąāļšāļĢāļ­āļ‡ AI āļŸāļĢāļĩāļŦāļĨāļąāļāļŠāļđāļ•āļĢāđāļĨāļ°āļāļēāļĢāļŠāļ­āļšāļŠāļąāđ‰āļ™ āđ† āļŸāļĢāļĩ āđ„āļĄāđˆāļĄāļĩāļ„āđˆāļēāđƒāļŠāđ‰āļˆāđˆāļēāļĒ āđ„āļĄāđˆāļĄāļĩāļšāļąāļ•āļĢāđ€āļ„āļĢāļ”āļīāļ• āļ›āļĢāļ°āļĄāļēāļ“ 2.5 āļŠāļąāđˆāļ§āđ‚āļĄāļ‡āđ€āļĢāļīāđˆāļĄāļ•āđ‰āļ™āļŸāļĢāļĩ →