Learning to forget: deimplementation and the science of sustainability in healthcare

Sustainability is not static maintenance; it is evolution. Health systems that can adapt, discard and relearn, retaining what matters and deliberately forgetting what no longer serves patients, can sustain high-value care and continually move beyond low-value practices. Healthcare prides itself on being a learning system. We collect data, consider the context, analyse outcomes and iterate improvement cycles. Yet, true sustainability—the ability to embed and maintain improvement over time—requires more than learning: it demands deliberate unlearning. Outdated order sets, decision rules and routines have the potential to shape care long after evidence has evolved.1 A paradox of improvement is that systems cannot sustain what is new until they have let go of what is old. Change does not fail only because people fail to learn—there are many reasons, one being that systems fail to forget. The PREDICT bronchiolitis programme implemented across Australasian emergency departments, as described by Ramsden and colleagues in this issue of the journal, offers rare evidence of system-level unlearning.2 3 Over successive knowledge translation cycles, PREDICT achieved sustained reductions in low-value bronchiolitis treatments, such as salbutamol, corticosteroids, antibiotics and chest X-rays, through multifaceted, theory-informed interventions. Success required more than teaching clinicians what to do; it required helping the system forget outdated practices. Outdated order sets were retired, guidelines embedded into workflow and new norms reinforced by leadership until they became standard. Without deliberate unlearning, even evidence-based interventions risk being dragged backward by entrenched routines. The evidence base makes it clear that adopting new practices alone does not guarantee sustainability. There is good evidence that multifaceted interventions combining education, stakeholder engagement, audit and feedback, and decision support …