Towards a Balanced View of Benefits and Harms in Deprescribing Trials

Abstract

A friend's nonagenarian mother recently experienced a probable adverse drug withdrawal event (ADWE). She had increasing weakness, exhaustion, and falls. Her new geriatrician, following a “less is more” philosophy, raised concerns about hypotension and frailty and enthusiastically stopped several medications because of concerns polypharmacy was causing her symptoms. The geriatrician deprescribed several antihypertensives and a diuretic. She also deprescribed dronedarone—an anti-arrhythmic used for rhythm control in atrial fibrillation. Within weeks, the friend's mother became weaker and more tired rather than less. She developed a rapid heart rate, leg swelling, and shortness of breath. She was readmitted to the hospital in atrial fibrillation and decompensated heart failure, and after unsuccessful attempts at diuresis as an inpatient, she unfortunately died.

ADWEs are (i) physiological withdrawal reactions (e.g., flu-like symptoms when stopping serotonergic antidepressants) or (ii) the re-emergence of symptoms of underlying disease (e.g., depressive symptoms) when discontinuing or reducing the dose of a drug [1]. In this issue of JAGS, Lee et al. [2] raise concerns that we are inadequately monitoring for ADWEs in deprescribing research. In their systematic review of 139 randomized controlled trials (RCTs) of deprescribing interventions, they found less than 1 in 10 reported on ADWEs. Of the few studies reporting ADWEs, they found ADWEs were slightly more likely to occur in participants receiving a deprescribing intervention. Their results suggest we may be systematically overlooking and underestimating the harms of deprescribing.

Why this oversight in research and, in our case, clinical practice? One possibility is the natural bias to believe that what we do helps. This also happens in drug initiation trials, where it is well documented that researchers neglect harms too [3]. This bias motivated recommendations by the CONSORT group (Consolidated Standards of Reporting Trials) in 2004 for reporting harms in clinical trials [4]. They suggested researchers explicitly declare if they are studying (i) benefits and harms or (ii) benefits only. If studying harms, researchers should explicitly describe how data on harms are collected, define and classify expected versus unexpected harms by severity, and consider whether a study is powered to detect a meaningful difference. The statement also recommended researchers try to determine if participants left the study because of adverse events (AEs); for deprescribing trials, this would mean collecting data on whether participants restarted medications and why. In other words, proof that deprescribing is safe requires that researchers consider, collect, and analyze data about possible harms with the same rigor as they treat possible benefits.

Yet another explanation for this oversight is that we lack clarity in how we think about the balance of risks and harms with deprescribing, as evidenced by the variability in ascertainment methods Lee et al. found among trials measuring ADWEs. To clarify the conceptual challenges, it is useful to first consider how we determine and think about harms in drug initiation trials. In drug initiation trials, we give drugs directly to patients. We use protocols and knowledge of expected harms to ask questions of participants about any AEs they have experienced. We inquire about temporal relationships between drug use and symptoms, dose response, and plausible mechanisms to adjudicate which AEs were directly caused by drug use as opposed to an underlying condition or other illness. These data are then incorporated into product monographs to inform clinician prescribing; done right, it lets us know that a drug has, for example, a 1% risk of major bleeding or a 10% risk.

It is very hard to create analogous “deprescribing safety monographs” for several reasons. First, most deprescribing trials target polypharmacy. This is based off evidence that cumulative drug exposure, drug–drug interactions, and risk of adverse drug reactions increase with medication count [5]. The corollary is that multiple drug classes (e.g., antihypertensives, diuretics, and antiarrhythmics) must be deprescribed for effect. How do we attribute harm when multiple drugs are deprescribed? There is no consensus for this, nor on whether harms should be reported separately by class or in aggregate. Deprescribing multiple drugs also makes data collection onerous. Each drug can have unique ADWEs and expected timelines for presentation; detailed inquiry about each requires lengthy follow-up visits.

Second, deprescribing trials are often pragmatic, which makes results difficult for clinicians to interpret and apply to their own practice. In a pragmatic trial, adherence to the intervention is flexible [6]; the intervention might only advise clinicians to deprescribe and leave the final decision up to them. Outcomes, then, depend on the intervention itself, fidelity of implementation, and the implementation context [7]. This affects interpretation. Suppose, for example, a trial of pop-up reminders to deprescribe medications found no evidence of harm. Was this because deprescribing was safe, clinicians ignored the prompts, or because clinicians were cautious and overrode unsafe recommendations? It is hard to know without even more data collection, and hard to apply to the clinical practice of deprescribing, which is often stepwise involving one medication change at a time and serial assessment [8].

Third, the focus of what we want to learn from a deprescribing trial differs from a drug initiation trial. With new drugs, we want to discover unknown risks [9]. This is why validated causal attribution tools (such as the Naranjo ADWE Probability Scale highlighted by the authors) [10] ask questions about timing, competing causes, response to drug discontinuation, and response to rechallenge; it confirms if a symptom or sign was caused by a drug even if rare. But with deprescribing, we rarely struggle to discern if a symptom is an ADWE (as in our case of dronedarone). Rather, we struggle to decide whether we should stay the course for overall clinical benefit or whether we should restart the drug. Unfortunately, existing causal attribution tools fail to collect information to help resolve this dilemma. Moreover, questions in these tools are often inapplicable in deprescribing trials. For example, we cannot ask “did the AE appear after the suspected drug was withdrawn?” in control groups that continued treatment as usual. This was perhaps observed in Lee et al.'s finding that most control groups in deprescribing RCTs reported no ADWEs when using these tools.

We therefore need further methodologic development to align the measurement of harms in deprescribing research with the clinical practice of deprescribing. Some of that work is underway; the Measures Workgroup of the US Deprescribing Research Network recently published recommendations to guide ADWE measurement in deprescribing studies.

One key recommendation is that we must acknowledge the difference between single drug and polypharmacy deprescribing studies [11]. Single drug deprescribing studies—especially those with well-defined enrollment criteria and that implement blinding through masked tapers—lend themselves better to the methods employed in drug initiation trials and produce estimates more relevant for clinical practice [12]. For these single drug studies, we should develop consensus definitions for ADWE severity and consider how we might elicit ADWEs in controls. We also need ascertainment tools and methods that distinguish transient withdrawal effects from more enduring re-emergence of symptoms because this distinction dictates management. If the symptom is a withdrawal effect, it can be overcome with careful tapering and time. If a symptom is a re-emergent condition, it requires restarting treatment. Drawing this distinction is not easy in many cases; for example, symptoms of opioid withdrawal are very difficult to separate from symptoms of underlying chronic pain.

For pragmatic randomized polypharmacy trials, we argue that it may not be as important to collect data to attribute ADWEs. Such studies are essentially investigating system-level interventions for which standardized assessment of major AEs through routine clinical data (e.g., mortality and hospitalization, as occurred in our clinical case) and clinician-reported serious AEs may suffice [11]. The goal of such trials is to determine net benefits and harms, which can be rigorously achieved through randomization, inclusion of two-sided outcomes that can detect both improvement and deterioration (such as quality of life scales), blinded or routine outcome assessment to reduce differential misclassification bias, and prompts to ensure participating clinicians report all serious AEs. Researchers studying polypharmacy interventions explicitly indicate which drug classes were targeted by their interventions and include deprescribing outcomes by drug or at least drug class. This transparency would help clinical readers appraise the likelihood of benefits and risks for their own practice—for example, if the most commonly deprescribed medications from a polypharmacy intervention were laxatives (as often happens when deprescribing during a transition of care) [13], we would not expect significant benefits or harms.

In conclusion, deprescribing is a complex process, and a “less is more” philosophy can lead to tunnel vision about its potential harms. This philosophy alone cannot be the basis for the science of deprescribing nor our clinical decisions to stop medications, as Lee et al. highlight in this systematic review and as we present in our case. By adopting best practices in the conduct of clinical trials and developing better methods for recognizing harms of deprescribing, we hope the field can move towards a more balanced and rigorous view of its risks and benefits.

All authors meet the criteria for authorship stated in the Uniform Requirements for Manuscripts Submitted to Biomedical Journals. All authors were responsible for the conception and design of the manuscript and revising it critically for important intellectual content. K.L. was responsible for drafting the work. All authors approve the final version of this manuscript and agree to be accountable for all the aspects of this work.

T.S.A. reports grant funding for a pilot deprescribing clinical trial from the National Institute on Aging through the US Deprescribing Network (NIA R33 AG086944).

This publication is linked to a related article by Lee et al. To view this article, visit https://doi.org/10.1111/jgs.19457.