ideasIdea

Algorithm-triggered goals-of-care conversations before crisis

When a prediction model says a patient has a high chance of dying within a year, their team is prompted to have a structured conversation about what matters to them, while there is still time to act on it.

Most patients with advanced cancer do not have documented conversations about goals until the final weeks. A randomised trial at Penn used a machine-learning mortality prediction to nudge oncologists, quadrupling serious illness conversations. The proposal is to make this standard: validated risk model, default prompt with a Serious Illness Conversation Guide, documentation template, and measurement of goal-concordant care and end-of-life chemotherapy use.

Hypothesis

Triggered conversations increase documented goals within 3 months of a high-risk flag from under 20% to over 50%, reduce systemic therapy in the last 14 days of life, and improve family-reported quality of dying.

Rationale

Clinicians overestimate prognosis and avoid the conversation; a neutral prompt gives permission and a time. The behaviour change has been demonstrated; the outcomes now need scale.

What would test it

Multi-centre stepped-wedge implementation with end-of-life quality metrics and bereaved-family surveys as primary outcomes.

Maturity

being tested at scale

Who has to act

clinic

Cost to try

Small (under $1M)

Years to first evidence

Bottlenecks it attacks

Patients lack understanding, navigation and agency · Most patients cannot understand their options, find trials, or push back, so decisions are made for them.
Pain relief and palliative care are unavailable to most · Most people who die of cancer worldwide do so without adequate pain relief.

Connected

3top

Pages like this

not linked directly; found by shared links

institutions

Abramson Cancer Center, University of Pennsylvania

bottlenecks

Pain relief and palliative care are unavailable to most Patients lack understanding, navigation and agency