The research, led by pharmacist-researcher Aili Langford of the University of Sydney and Monash University’s Centre for Medicine Use and Safety, together with colleagues including deprescribing specialists Barbara Farrell of the Bruyère Research Institute in Ottawa, Wade Thompson of the University of British Columbia, and Emily Reeve of Monash University, involved in-depth interviews with 24 Australian healthcare professionals. The sample comprised ten medical practitioners—specialists spanning cardiology, clinical pharmacology, emergency medicine, general practice, geriatrics, nephrology, and psychiatry—alongside nine pharmacists and five nursing professionals, recruited across five Australian states and territories through professional organizations, social media, and the Australian Deprescribing Network. Between March and October 2024, participants completed semi-structured interviews lasting 33 to 56 minutes, in which they dissected real deprescribing recommendations drawn from published guidelines, ranked deliberately varied versions of the same recommendation, and explained what would make them follow such advice at the bedside. The team then analyzed the transcripts with thematic framework analysis, deductively mapping the data onto the Guideline Language and Format Instrument, or GLAFI—a validated tool that specifies how the wording, structure, and layout of guideline recommendations determine whether clinicians adopt them or abandon them—while remaining alert to themes the instrument could not capture.

The stakes are anything but academic. Deprescribing—the clinician-supervised process of tapering or stopping medications that are inappropriate or no longer necessary—remains uncommon, while the global prevalence of potentially inappropriate medication use continues to climb. Single-disease guidelines, which dominate clinical practice, routinely fail to account for older adults living with multiple conditions, generating stacks of disease-specific recommendations that can cumulatively promote polypharmacy, drug interactions, and medication-related harm. Class-specific deprescribing guidelines do exist for benzodiazepines, proton pump inhibitors, antipsychotics, and antihyperglycemic agents, among others, but their advice has never been routinely woven into mainstream treatment guidelines. A recent scoping review by the same research group found that only two percent of evidence-based deprescribing recommendations were supported by high-certainty evidence—a stark measure of the evidentiary hole that guideline developers are being asked to write their way out of.

The first and most striking finding was a double standard. Participants consistently described deprescribing as a distinct, more complex, and higher-risk clinical activity than initiating a drug, and consequently demanded far more from any recommendation telling them to do it. They wanted the full decision logic spelled out—who is suitable for deprescribing, who is responsible, which medications, when, and why—but above all, how: explicit tapering regimens, monitoring parameters, co-interventions, and strategies for engaging patients in the conversation. “With deprescribing, it’s much more common than in prescribing to have to work out complex regimens…a little bit of handholding would definitely make someone more comfortable,” one physician told the researchers. GLAFI’s content checklist asks developers to state the action to be performed, by whom, for whom, and under what conditions, yet stays silent on these operational mechanics—the precise gap participants kept flagging. The authors argue the instrument itself will need extending for de-implementation, the deliberate withdrawal of established practices, which poses challenges that implementation frameworks built for rolling out new interventions were never designed to handle.

Stopping a medication also collides with deeply ingrained clinical psychology. Participants summarized the prevailing mentality bluntly: “adding things is doing good, taking things away is doing bad.” Because deprescribing is rarely incentivized in practice settings while prescribing is reinforced by clear guidance, workflow defaults, and entrenched habit, clinicians default to continuation—a pattern the researchers frame as clinical inertia. To counter it, participants favored direct, active-voice language of the form “we recommend A for B” over hedged phrasing like “consider this,” paired with explicit statements of the relative advantage of stopping, whether the benefit of deprescribing or the harm of continued use. Perceived legal exposure sharpened the reluctance: “It takes balls to deprescribe something that someone’s been on for a long time,” one physician admitted. And because deprescribing typically aims to avert future harm rather than deliver immediate benefit, the persuasive framing that props up prescribing recommendations is intrinsically harder to construct for the reverse maneuver.

Evidence, or the perceived absence of it, compounded the hesitancy. Many participants assumed deprescribing recommendations rest on expert consensus rather than robust trials—”it would be some well-intentioned group of clinicians coming up with what seems reasonable,” in one physician’s words—and several said weak, low-certainty recommendations were so off-putting that they would rather the evidence ratings be omitted entirely, even though that would violate current best practices in guideline development. The researchers argue this reflects a misunderstanding of what evidence ratings mean rather than a genuine design flaw, and calls instead for plain-language interpretation of GRADE-style ratings and targeted clinician education. Participants also exposed a striking double standard: prescribing recommendations built on trials conducted in younger, healthier populations are applied to older, more complex patients without question, while deprescribing advice lacking direct evidence is met with suspicion. Some argued that the absence of evidence for a drug’s ongoing benefit should itself justify stopping it, and that guidelines should distinguish “evidence of no benefit” from “no current evidence of benefit” so clinicians can judge for themselves—while still anchoring decisions in patient goals and preferences rather than rigid evidence thresholds.

The second theme captured a tension every guideline writer will recognize: clinicians want recommendations that are simultaneously succinct and complete. Ambiguous verbs were a particular worry. Terms such as “de-escalate,” “taper,” and “withdraw” leave it unclear whether dose reduction, abrupt cessation, or both are intended—a gap that can translate directly into harm when a drug requiring gradual weaning is stopped cold. “Taper slowly means nothing to anybody. Slowly for one person is a day and another person is a month…I think it’s quite easy for something like this to be misinterpreted or misapplied,” one pharmacist explained. Participants called for standardized, precise terminology—for example, “discontinue”—supported by glossary definitions, and they generally praised bullet points, color coding to signal evidence strength, and clinical algorithms that strip cognitive load from dense text and speed up decisions in time-poor environments.

Yet the same participants, when shown fully detailed step-by-step recommendations, judged them long, cumbersome, and overwhelming—text that nobody would actually read in a busy clinic. Recommendations offering multiple tapering regimens and co-intervention options triggered decision fatigue, delaying or deferring choices altogether. Intriguingly, some participants defended a degree of deliberate ambiguity, arguing that overly directive instructions risk a one-size-fits-all approach that tramples individual patient goals, while others complained that open-ended guidance left them feeling unsafe making the call. “The guidelines are meant to be guidelines, right? They’re meant to be giving you a direction. When it’s kind of open, I don’t feel good about making that call,” one physician said, while another bristled at the prospect of a “cookbook of medicine.” The authors suggest the resolution lies in flexibility of format rather than ambiguity of language: layered resources combining comprehensive technical documents, concise clinical tools, and plain-language summaries, plus multiple tailored options where the optimal strategy is genuinely uncertain.

Placement proved equally consequential. Some clinicians wanted deprescribing advice co-located with prescribing recommendations, so that the question of eventual withdrawal surfaces at the very moment of initiation, shifting deprescribing from a reactive response to adverse events—a fall, a hospitalization—toward proactive planning. “You might be prescribing it with no thought or intention of deprescribing it. But if you see that at the bottom of the article…you’re like, oh, hang on…I need to plan to wean this,” one physician reflected. Others preferred a standalone deprescribing section, reasoning that clinicians rarely open a hypercholesterolemia guideline to ponder statin withdrawal and that a dedicated section can hold broader principles and patient-conversation strategies beyond medication-specific advice. There was consensus, however, that paper alone is not enough: deprescribing advice must be embedded in national medicine handbooks, formularies, and prescribing and dispensing software, delivered as context-sensitive, patient-data-driven prompts engineered to avoid the alert fatigue that buries clinicians under notifications. Participants even envisioned artificial intelligence systems cross-referencing prescribing behavior against guidelines in real time, reducing reliance on external documents altogether.

Synthesizing the themes, the team distilled seven implementability design principles for deprescribing recommendations: specify the full decision logic—who, what, when, why, and how; use active, directive language with harmonized verbs; counteract clinical inertia with positive framing that states the relative advantage of stopping; standardize terminology and define it in a glossary; communicate the strength and certainty of evidence clearly, with plain-language interpretation; visualize information through bullet points, flowcharts, and algorithms; and place recommendations where clinicians will actually act on them, whether alongside prescribing guidance, in a dedicated section, or within digital workflows. The findings have already informed an international clinician survey designed to test preferred content, format, and terminology across a broader sample. The authors acknowledge the study’s limits—a qualitative design, an Australian volunteer sample, and analysis led by researchers with pharmacy backgrounds—while arguing the principles target universal features of clinical cognition, from cognitive load to the appetite for clear, directive guidance. If they hold up, guideline developers finally have a concrete recipe for writing the half of prescribing that has gone unwritten—a step that could shrink the burden of medications patients no longer need, before another 92-year-old like the nurse participant’s “Dot,” who fell while taking “a truckload of anti-hypertensives,” pays the price for medicine’s silence.