Heart attacks remain a leading cause of death. NTT is building a digital twin to tilt the odds in the ICU.

Cardiovascular disease continues to kill more people than anything else on Earth, and it has for more than three decades running. Despite significant progress, the latest Global Burden of Disease analysis, published in JACC in September 2025, attributes roughly one in three deaths worldwide (19.2 million in 2023) to CVD. That is up from 13.1 million in 1990.

It is for that reason then that NTT Research’s Medical & Health Informatics (MEI) Lab, part of the company’s Silicon Valley basic-research push, is focused on the cardiovascular system. Leading MEI is Joe Alexander, M.D., Ph.D., an Auburn-trained chemical engineer turned Johns Hopkins physician-scientist who did cardiovascular clinical research stints at Merck and nearly two decades at Pfizer, joined NTT Research in 2020 as Distinguished Scientist to lead the Bio Digital Twin initiative. He became director of the lab in July 2021.

Acute care first

Otto, the lab’s mechanistic cardiovascular simulator, represents the heart, vasculature, and linked pulmonary and renal subsystems as a causal, physiology-based circuit model rather than a deep-learning black box. The use case the team has built around it is the Autonomous Closed-Loop Intervention System (ACIS), a decision-support and (eventually) closed-loop control system for managing patients in the hours and days after a heart attack or acute heart failure episode. “In the acute situation, we have access to lots of data to inform our models, and we only need to model a four- or five-day hospital stay,” Alexander said. “We don’t need to project far into the future.”

The multiple-day time horizon offers a relatively simple frame for “a challenging problem,” as Alexander noted. One of the core tools feeding the digital twin is instrumentation. In the ICU, the model could be fed by dense, real-time physiological data. As Jon Peterson, Ph.D., a distinguished scientist in the lab and Alexander’s longtime collaborator going back to their Johns Hopkins days, noted: “These patients are instrumented with indwelling catheters. They’re generating information that we can’t get from someone ambulatory, walking around.”

Virtual populations

The MEI Lab is building a virtual population of patients spanning the realistic range of cardiovascular states for a given acute condition, then locating the incoming patient within it. “Our approach involves creating virtual populations, from which we then move on to create populations specific to our acute conditions, for example, acute myocardial infarction or acute heart failure,” Alexander said. “When a patient rolls in the door, we find within this virtual population which of these patients best match the patient in front of us.” The Otto system narrows the field to a band of similar cases that may help predict response to care.

NTT’s approach is distinct from digital-twin efforts that aim for higher-fidelity, molecular-scale replication. As Peterson put it, “At this point these models are not molecular. I mean, they’re mechanistic and they’re based in physiology,” he continued. “Our guiding principle is, what’s the simplest model we can use that gives us the information we need?”

A physiology-based model light enough to run quickly can be queried with the same variables a critical-care physician already tracks at the bedside. In addition, it can be interrogated for what happens if a given drug or dose is adjusted.

Specific ICU targets

After a heart attack, in the Intensive Care Unit (ICU) or Cardiac Care Unit (CCU), clinicians have to stabilize the patient’s vital signs while giving the damaged heart room to recover.

The Autonomous Closed-Loop Intervention System is thus designed around a specific set of physiological targets, including cardiac output, arterial pressure and left atrial pressure. The last of these helps keep fluid from backing up into the lungs. “This autonomous closed-loop intervention system is one where, for example, you type in the desired cardiac output, the desired arterial pressure, the desired left atrial pressure so that blood or fluids don’t accumulate in the lungs,” Alexander said. “You type in the desired condition, and then the system runs automatically to achieve that.” The harder layer, Alexander said, is doing all of that while keeping the heart’s own workload down, a problem the team frames as minimizing myocardial oxygen consumption.

Automated blood-pressure control already exists in anesthesiology, but controlling several hemodynamic variables simultaneously while also minimizing the metabolic burden on an injured heart remains challenging. Alexander described, in an earlier CXO Talk interview, a sort of feedback loop. Clinicians enter desired endpoints, drug-filled syringes deliver the agents and the model adjusts dosing as the patient’s measured response comes back in.

Decision support is a near-term focus

The MEI Lab’s nearer-term goal is clinical decision support. Peterson said the model is meant to help physicians think through treatment choices by surfacing recommendations.

The technology isn’t live in hospitals quite yet. “We are not yet ready [for human trials],” Peterson said. The proof-of-concept milestone came in 2024, when the team showed the closed-loop system could manage induced heart failure in a dog while holding to its oxygen-consumption constraint. “For the first time ever, we demonstrated the proper functioning of an autonomous closed loop intervention system to treat heart failure in an experimental animal, a dog, that had experimentally induced heart failure,” Alexander said in late 2025 on CXOTalk. “The system compensated and automatically treated that dog while minimizing myocardial oxygen consumption.”

Alexander expects the lab to have human trial results within the coming years, with Japan’s National Cerebral and Cardiovascular Center (NCVC) as the clinical partner. “NCVC is the referral center for all of Japan for cardiovascular cases,” Alexander said. “So we’re moving from animals to patients.”