💊💊 Deprescribing Algorithm for Polypharmacy

Comprehensive Medication Review — Finding the True List

Polypharmacy — conventionally defined as concurrent use of five or more medications — affects an estimated 42% of adults over 65 in high-income health systems, and the number rises sharply with each additional prescriber a patient sees. Before any clinical decision can be made about a drug regimen, the regimen itself must first be accurately reconstructed, because fragmented care across specialists, pharmacies, and over-the-counter purchases routinely produces medication lists that no single record fully captures.

• ≥5: Polypharmacy threshold (concurrent chronic medications)
• 5: Prescribers involved (across primary + 4 specialists)
• 3: Pharmacies dispensing (incomplete individual records)
• 14: Medications identified (Rx + OTC + supplements)

The brown-bag review and reconciliation workflow

Medication reconciliation begins with a structured "brown-bag review": the patient (or caregiver) brings every pill bottle, blister pack, inhaler, topical, and supplement to the visit — not a recalled list, the actual containers. This single step routinely uncovers 20–30% more products than the electronic health record shows, because samples, OTC purchases, and medications prescribed by out-of-network clinicians are systematically under-captured by any one system.

The reconciliation process cross-references three independent sources:

• EHR active medication list — often stale, containing discontinued drugs never removed • Pharmacy fill/refill history (via state PDMP or NCPDP SCRIPT feeds) — objective adherence and duplication signal • Patient/caregiver report and physical inspection of containers — captures OTC, herbal, and as-needed use invisible to claims data

Discrepancies between these sources are common: one study of hospitalized older adults found a mean of 3.3 unexplained discrepancies per patient between admission medication history and outpatient records. Each discrepancy is a candidate for either an omission error (a needed drug missing) or a duplication/prescribing-cascade error (an unneeded drug present).

Prescribing cascades and the accumulation of polypharmacy

A prescribing cascade occurs when a new drug is started to treat the side effect of an existing drug, rather than recognizing the side effect as iatrogenic and adjusting the original therapy. This is one of the principal engines of polypharmacy accumulation in older adults:

• Amlodipine → peripheral edema → misattributed and treated with furosemide, when dose reduction or a different antihypertensive class would resolve the edema directly • NSAID → hypertension → treated with an additional antihypertensive rather than discontinuing the NSAID • Metoclopramide or antipsychotic → drug-induced parkinsonism → treated with levodopa • Diphenhydramine or a first-generation antihistamine → urinary retention/constipation → treated with additional laxatives or a bladder agent

In this case, the review flags exactly this pattern: docusate was added roughly 14 months after diphenhydramine, and furosemide dose was escalated shortly after amlodipine initiation — both plausible cascade signatures worth tracing before assuming each drug is independently indicated. Each medication on a reconciled list is annotated with its original indication, start date, and prescriber, producing the indication-linked inventory that every subsequent stage of the algorithm depends on.

Risk Stratification — STOPP/START, Beers Criteria, and Burden Scores

Once the medication list is accurate, each drug is screened against explicit, criteria-based tools developed from expert consensus and pharmacoepidemiologic evidence. These tools do not replace clinical judgment, but they systematize the detection of potentially inappropriate medications (PIMs) that are easy to miss when reviewing a 14-item list one drug at a time, and they generate a reproducible burden score that can be tracked over time.

• 133: STOPP/START v3 criteria (2023 update; 40 countries validated)
• 5: AGS Beers Criteria domains (avoid / caution / interaction / renal / drug-class)
• 10: Explicit flags identified (STOPP + Beers, this regimen)
• 2.1: Drug Burden Index (DBI) (anticholinergic + sedative load)

STOPP/START and the 2023 AGS Beers Criteria

Two complementary explicit-criteria instruments anchor this stage:

STOPP/START (Screening Tool of Older Persons' Prescriptions / Screening Tool to Alert to Right Treatment), now in version 3 (2023, O'Mahony et al.), pairs 133 criteria across two directions: • STOPP criteria flag drugs that are potentially inappropriate given the patient's age, diagnoses, or concurrent therapy — e.g., "long-term PPI at full therapeutic dose for >8 weeks without a clear indication" (flags omeprazole here) • START criteria flag indicated therapies that are potentially missing — e.g., osteoporosis prophylaxis in a patient on long-term corticosteroids

The 2023 American Geriatrics Society (AGS) Beers Criteria organizes potentially inappropriate medications into five domains: drugs to avoid in most older adults, drugs to avoid in specific diseases/syndromes, drugs to use with caution, drug-drug interactions to avoid, and drugs requiring renal dose adjustment. Benzodiazepines, first-generation antihistamines, tricyclic antidepressants, and sulfonylureas — all present in this regimen — appear on the "avoid" list because of consistently demonstrated excess risk of falls, fractures, cognitive impairment, or severe hypoglycemia in adults ≥65.

Applied to this 14-drug regimen: omeprazole (long-duration PPI, no documented indication), diphenhydramine (Beers avoid, high anticholinergic burden), lorazepam (Beers avoid, benzodiazepine), amitriptyline (Beers avoid, TCA), glyburide (Beers avoid, long half-life sulfonylurea), zolpidem (Beers avoid, Z-drug), plus duplicate acid-suppression (omeprazole + famotidine) and an aspirin without a clearly re-confirmed cardiovascular indication — 10 flags in total.

Composite burden scores: ACB and Drug Burden Index

Beyond binary flags, two composite scores quantify cumulative pharmacologic load:

• Anticholinergic Cognitive Burden (ACB) scale: each drug is scored 0 (none), 1 (mild), 2 (moderate), or 3 (definite/strong anticholinergic activity) and summed across the regimen. Diphenhydramine and amitriptyline each score 3; a cumulative ACB ≥3 is independently associated with a 50% increased risk of incident cognitive impairment over 2 years in longitudinal cohort studies.

• Drug Burden Index (DBI, Hilmer et al. 2007): a validated pharmacological dose-response model summing exposure to anticholinergic and sedative drugs, DBI = Σ [D/(D+δ)], where D is the daily dose and δ is the minimum effective dose for that drug's principal anticholinergic/sedative action. A DBI above 1.0 is associated with measurably reduced physical function; this patient's DBI of 2.1 places them at substantially elevated risk of falls, frailty, and hospitalization independent of any single drug's individual risk profile.

These composite scores matter because explicit criteria alone can miss cumulative, sub-threshold contributions — a patient can carry four "mild" anticholinergic drugs, none individually flagged by Beers, yet accumulate a burden equivalent to one strong anticholinergic agent.

Explicit-criteria domains applied to this regimen

Deprescribing Prioritization — Interaction Networks and the Ariadne Principle

Flagging a drug as inappropriate is necessary but not sufficient — a 78-year-old on 14 medications cannot safely have all flagged drugs stopped simultaneously. Prioritization requires mapping how flagged drugs interact with each other and with the rest of the regimen, then applying a structured decision algorithm that weighs discontinuation benefit against withdrawal risk and the patient's own goals of care.

• 11: DDI network edges mapped (Lexicomp/Micromedex severity-graded)
• 6: Major-severity interactions (of 11 edges — CNS depressant clustering)
• 5: Deprescribing algorithm steps (Scott et al., JAMA Intern Med 2015)
• 6: Discontinuation candidates (selected for Stage 4 taper)

Building the drug-drug interaction network

Every pairwise combination among the 14 medications is checked against a clinical interaction database (Lexicomp, Micromedex, or equivalent) and severity-graded as major, moderate, or minor. The result is a graph: 14 nodes (drugs), 11 edges (clinically significant interactions) for this regimen.

The most consequential structural feature is not any single edge but the clustering: lorazepam, zolpidem, diphenhydramine, and amitriptyline form a densely interconnected subgraph of five major-severity edges — every pairwise combination among these four CNS-active, sedating, anticholinergic drugs carries an independent excess risk of falls, delirium, and respiratory depression, and their effects are pharmacodynamically additive rather than merely coincident. A network view makes this visible in a way that a flat list cannot: reviewing each drug's package insert in isolation would show four separate "use with caution" warnings, but the network reveals a single interconnected hazard requiring coordinated, not independent, tapering.

A second, smaller cluster links furosemide, metoprolol, and aspirin through moderate-severity renal/hemodynamic interactions — clinically real, but lower acuity than the CNS-depressant cluster, and appropriately triaged for monitoring and dose adjustment rather than urgent discontinuation.

The 5-step deprescribing algorithm and the Ariadne principle

Scott and colleagues (JAMA Internal Medicine, 2015) formalized a 5-step algorithm now widely adopted in geriatric and general practice:

1. Ascertain all drugs currently used and the reasons for each (completed in Stage 1) 2. Consider overall risk of drug-induced harm in this individual, weighting age, renal/hepatic function, cognitive status, and cumulative burden indices (completed in Stage 2) 3. Assess each drug for its current or future benefit versus harm, considering: efficacy for the treated condition, time-to-benefit relative to the patient's life expectancy, and whether the original indication still applies 4. Prioritize drugs for discontinuation using explicit ranking — lowest benefit-to-harm ratio, shortest time-to-benefit, and highest interaction/burden contribution are stopped first 5. Implement a discontinuation regimen with the patient, combining monitoring for withdrawal or rebound with a shared follow-up plan

Running parallel to this is the Ariadne principle (Muth et al. 2014), developed specifically for multimorbid patients on complex regimens: it insists that any single-disease guideline recommendation be re-interpreted in light of the patient's full problem list, their explicitly elicited preferences and goals of care (symptom relief vs. life prolongation vs. functional independence), and the realistic feasibility of the regimen — because a pharmacologically "optimal" 14-drug regimen that the patient cannot physically organize or afford is not, in practice, optimal at all. Combining both frameworks against the interaction network selects 6 candidates for Stage 4: omeprazole, diphenhydramine, lorazepam, amitriptyline, glyburide, and zolpidem.

The single highest-yield intervention in this regimen is not any one drug — it is the CNS-depressant cluster. Coordinated tapering of lorazepam, zolpidem, diphenhydramine, and amitriptyline as a linked group, rather than sequentially and independently, is what the interaction network reveals as the priority: in trials of structured benzodiazepine/sedative deprescribing in this exact drug cluster, coordinated withdrawal reduced fall incidence by 27–39% over 6 months, an effect not reliably seen when only one drug in the cluster is addressed at a time.

Stepwise Tapering Protocol and Withdrawal Monitoring

Abrupt discontinuation of chronically used CNS-depressant, cardiovascular, or acid-suppressive medications can precipitate withdrawal syndromes as dangerous as the original polypharmacy — benzodiazepine withdrawal seizures, rebound insomnia, cholinergic rebound, or acid hypersecretion. Each of the six discontinuation candidates therefore follows its own individualized, evidence-based taper schedule with explicit monitoring for withdrawal symptoms.

• −25%: Lorazepam taper rate (every 2 weeks (equivalent-dose diazepam bridge optional))
• 4 wks: PPI step-down duration (omeprazole → alternate-day → PRN)
• 25 mg: TCA decrement (biweekly, amitriptyline)
• 6–12 wks: Full taper horizon (per drug, longest for benzodiazepine)

Drug-specific tapering schedules

Each targeted medication is withdrawn according to a schedule calibrated to its pharmacokinetics and known withdrawal syndrome:

• Lorazepam (benzodiazepine): reduce total daily dose by 25% every 2 weeks; below 50% of original dose, slow to 10–12.5% reductions every 2–4 weeks. For patients on short half-life agents like lorazepam, an optional cross-taper to a longer half-life equivalent (diazepam) can smooth blood-level troughs and reduce interdose withdrawal symptoms. Full taper typically requires 8–12 weeks.

• Zolpidem (Z-drug hypnotic): reduce by one tablet-strength decrement every 1–2 weeks; because of its short half-life (~2.5 h) and lower physical dependence potential than benzodiazepines, most patients tolerate a somewhat faster taper, 4–6 weeks total, combined with cognitive-behavioral therapy for insomnia (CBT-I) to manage rebound sleep difficulty.

• Amitriptyline (TCA): reduce by 25 mg every 2 weeks to avoid cholinergic rebound (nausea, diaphoresis, insomnia) and mood destabilization; full taper 6–8 weeks depending on starting dose.

• Diphenhydramine: typically discontinued outright or over 1–2 weeks given short half-life and primarily as-needed use pattern, with substitution of non-pharmacologic sleep hygiene measures.

• Omeprazole (PPI): step down from full dose to alternate-day dosing for 2 weeks, then to as-needed use over the following 2 weeks, to avoid acid-hypersecretion rebound; total step-down 4 weeks.

• Glyburide (sulfonylurea): not tapered in the pharmacokinetic sense but cross-titrated to a shorter-acting, renal-safer secretagogue or non-sulfonylurea agent over 2–4 weeks with frequent glucose monitoring during the switch.

Monitoring for withdrawal and rebound

Structured taper without structured monitoring is incomplete — withdrawal symptoms are the leading reason deprescribing attempts are abandoned mid-course, often unnecessarily, when a symptom is misattributed to disease recurrence rather than recognized as a transient, expected, and manageable withdrawal phenomenon.

Validated monitoring instruments used across this taper:

• CIWA-B (Clinical Institute Withdrawal Assessment for Benzodiazepines): a structured symptom checklist administered at each taper step, scoring anxiety, tremor, sweating, and perceptual disturbance; scores are trended visit-to-visit rather than interpreted as single time points • Benzodiazepine Withdrawal Symptom Questionnaire (BWSQ): patient-reported symptom burden used between visits, including insomnia rebound and irritability • Sleep diaries: track rebound insomnia during zolpidem and diphenhydramine taper separately from anxiety-driven wakefulness during benzodiazepine taper, since the two require different reassurance and management • Home glucose logs: mandatory during any sulfonylurea cross-titration given the acute risk of both hyper- and hypoglycemia during transition

Across this six-drug taper plan, all six candidates reach ≥50% dose reduction by week 6 with no seizure activity, no severe rebound insomnia requiring taper reversal, and stable glycemic control throughout the glyburide switch — outcomes consistent with published structured-taper cohorts, where fewer than 10% of appropriately monitored tapers require pausing or reversal.

Outcome Reassessment — Closing the Deprescribing Loop

Deprescribing is not complete at the moment a taper schedule ends — it requires the same rigor applied to starting a new drug: a scheduled follow-up assessment of whether discontinuation achieved its intended benefit without unmasking harm. At 3 and 6 months, the full assessment battery from Stages 1–2 is repeated, and the regimen is either confirmed as stable, adjusted further, or — occasionally — partially reversed if an originally treated symptom recurs.

• 14 → 8: Medications, before → after (6 discontinued or substituted)
• 2.1 → 0.6: Drug Burden Index, before → after (71% reduction in sedative/anticholinergic load)
• −38%: Self-reported falls, 6-month change (vs. pre-deprescribing baseline)
• 2: Residual DDI network edges (down from 11, both low-acuity)

The follow-up assessment battery

Every instrument used to justify deprescribing is reused to verify its outcome, at 3 and 6 months post-taper completion:

• Medication count and updated STOPP/START and Beers screening — confirming no new inappropriate prescriptions have accumulated (a common failure mode: a discontinued drug's original symptom recurs and is re-treated with a different, equally inappropriate agent) • Mini-Cog and Montreal Cognitive Assessment (MoCA) — tracking whether reduced anticholinergic burden translates into measurable cognitive benefit; effects are typically modest and take 3–6 months to manifest • EQ-5D-5L quality-of-life instrument — captures patient-reported function, pain, anxiety/depression, and usual activities, since deprescribing's ultimate goal is functional and subjective improvement, not merely a lower pill count • Fall diary and Timed Up-and-Go (TUG) test — objective and patient-reported mobility/balance measures, directly targeting the CNS-depressant cluster's primary harm pathway • Renal function and HbA1c — confirming the glyburide-to-alternative switch maintained glycemic control without hypoglycemic events

In this patient, all five domains showed improvement or stability at 6 months: no new PIM was added, Mini-Cog score improved from 3/5 to 4/5, EQ-5D-5L index rose from 0.68 to 0.79, self-reported falls fell from 5 in the prior 6 months to 3, and HbA1c remained at target (7.1%) throughout the sulfonylurea transition.

Evidence base for deprescribing outcomes

This case-level result mirrors findings from larger controlled deprescribing trials and implementation programs:

• The OPTIMIZE trial (structured pharmacist-led medication review in primary care) demonstrated a mean reduction of 2–4 medications per patient with no increase in adverse events or hospitalization at 12-month follow-up • The SENATOR trial (Software ENgine for the Assessment and optimisation of drug and non-drug Therapy in Older persons) used a computerized decision-support system layering STOPP/START logic onto EHR data across multiple European hospitals, reducing potentially inappropriate prescriptions with acceptable safety • Systematic reviews of benzodiazepine/sedative-hypnotic deprescribing interventions in older adults consistently report 25–40% relative reductions in fall incidence when tapering is structured, monitored, and combined with non-pharmacologic substitution (CBT-I, physical therapy) rather than discontinuation alone • McMaster/Deprescribing.org evidence-based guidelines for PPIs, benzodiazepines, and sulfonylureas independently validate the specific taper schedules used in Stage 4, each grade A or B recommendation

The consistent pattern across this literature is that deprescribing, done through a structured algorithm rather than ad hoc drug-by-drug cessation, reduces medication burden without increasing — and often while reducing — adverse clinical events, directly countering the intuitive but incorrect assumption that "more monitoring drugs equals safer care" in a frail, polypharmacy population.

A DBI reduction from 2.1 to 0.6 is not a cosmetic number: each 1-point increase in Drug Burden Index has been associated in longitudinal cohorts with roughly a 5–10% relative increase in fall risk and measurable decline in gait speed and grip strength. Reversing 1.5 points of DBI in a single deprescribing episode places this patient's pharmacological burden back within a range associated with meaningfully preserved physical function — a magnitude of benefit that no single new drug added to a 14-item regimen could plausibly have achieved.