ARPA-H puts up to $9.5 million behind testing whether an agent can adjust heart-failure care

Heart-failure management, in Tempus's own description, is repeated adjustment: medication, diet, exercise and follow-up care, with decisions shaped by symptoms, vital signs, laboratory results and other medical conditions ^[6]. Of that set, one signal arrives continuously from the patient's home, and it is heart rate ^[5]. The rest lands in the electronic record whenever a clinician orders, measures or types it. So the agent acts on a fresh heart rate and on record values that move only when a clinician orders, measures or types them, and ADVOCATE's stated answer to that gap is escalation to a human clinician ^[10].

Tempus has shipped detection for years ^[7]. The company says its ECG-Low EF algorithm is FDA-cleared to detect signals associated with reduced ejection fraction ^[8]. In a notify product, a wrong output costs a clinician's review time; in an agent that makes defined changes to appointments, medication, diet and exercise, a wrong output is a change to someone's care ^[9][10]. The whole safety argument is the list of changes the agent is allowed to make alone, and that list is missing from the account of the award ^[24].

The money is small next to what Tempus has already raised: $9.5 million is about 2.3 percent of the $410.7 million in gross proceeds from its June 2024 Nasdaq IPO ^[22][16]. Against the program it is larger, about 28 percent of the $33.7 million ARPA-H committed in ADVOCATE's first year and about 15 percent of the $62.7 million four-year ceiling ^[18][19]. All three of those are ceilings. Public materials do not disclose the payment schedule, contract conditions or amount disbursed so far ^[3]. More than half the program total is committed in year one, leaving up to $29 million for the remaining three ^[20].

The filing deadline is the one that binds: twenty-four months from contract award is half the four-year program ^[21], and a prospective multicenter study has to be designed, enrolled, run and analysed inside that window to appear in the package at all. ARPA-H's rule sets a deadline; it does not create a completed regulatory pathway, and the award does not establish that Olivia changes prescriptions, provides autonomous treatment or improves outcomes ^[11][14].

Two other named patient-facing teams answer the same clock with different architectures. Atman Health is building a voice-first system that changes its questions as it gathers information, and UpDoc keeps its conversational interface separate from a clinician-built rules engine ^[13]. runtimewire's summary of the award says the study could produce evidence for clinical and regulatory review while no authorization, treatment authority or patient benefit has been established ^[15]. If the trial does report a benefit, it transfers to another health system only where the record supplies the same fields at the same latency and the patients are on the same home monitor.