AGS Beers Criteria: The Five Tables, the 2023 Update, and How Hospitals Use It

The AGS Beers Criteria® is an explicit, evidence-graded list of medications that are potentially inappropriate for adults aged 65 and older. It is maintained by the American Geriatrics Society, which has been its steward since 2010 and publishes updates on a regular cycle. The current edition is the 2023 update, published in the Journal of the American Geriatrics Society (2023;71(7):2052–2081, doi:10.1111/jgs.18372) — the seventh overall revision and the fourth produced under AGS stewardship.

For hospital patient-safety, quality and pharmacy staff, the Beers Criteria is three things at once: a prescribing-review reference, the technical backbone of two widely reported health-plan quality measures, and — when it is used carelessly — a source of real friction with prescribers. This guide covers what the criteria actually contain, what the 2023 update changed, how the criteria are applied in medication-safety work, and the limits the AGS panel itself puts on them in writing.

What the Beers Criteria is, and what it is not

The criteria began in 1991 as the work of the late Mark H. Beers, MD, and colleagues at UCLA, developed to identify medications whose potential harm outweighed the expected benefit in nursing home residents. A 1997 update, also led by Dr. Beers, expanded the scope to all older adults; an interprofessional group revised it in 2003, and the AGS assumed stewardship in 2010, producing updates in 2012, 2015, 2019 and 2023.

The AGS states the intent of the criteria in three parts: to reduce older adults’ exposure to potentially inappropriate medications (PIMs) by improving medication selection; to educate clinicians and patients; and to serve as a tool for evaluating quality of care, cost and patterns of drug use in older adults. The primary target audience is practising clinicians, but the AGS explicitly names healthcare consumers, researchers, pharmacy benefits managers, regulators and policymakers as secondary users.

Scope: who and where the criteria apply

Two scope boundaries are frequently missed and are worth putting in any local policy that references the criteria:

• Age. Unless a specific criterion says otherwise, the criteria apply to adults 65 years and older. Some individual criteria carry their own age cutoff (several apply specifically to adults 75 and older) where the evidence supports one.
• Setting. The criteria are intended for all ambulatory, acute and institutionalised settings of care except hospice and end-of-life care. The 2023 panel restates this explicitly in the context of antipsychotics and other symptom-control drugs: decision-making at end of life “may require other considerations.”
• Country. The criteria are designed for the United States. The panel lists cross-national differences in drug availability as an acknowledged limitation and advises clinicians elsewhere to adapt to their local formulary context.

The structure: five categories, plus three supporting tables

The 2023 update keeps the same five general categories used in 2019, presented as Tables 2 through 6, with three additional supporting tables and one summary box.

Every criterion carries a quality of evidence rating and a strength of recommendation rating, derived from an approach adapted from GRADE and the American College of Physicians’ evidence framework. The two do not always align: the panel notes that strength-of-recommendation ratings incorporate expert opinion, clinical judgement and context, so a criterion can carry a strong recommendation on low-quality evidence, or the reverse. If your organisation builds alerts or audit logic off the criteria, read both ratings rather than treating every row as equivalent.

What “avoid” actually means

This is the single most consequential thing to get right, and the source of most of the disputes the criteria generate on the wards. The 2023 panel is unambiguous:

“Avoid” is not defined as an absolute contraindication unless specified in the medication’s label.

The word carries the same meaning it did in the original 1991 criteria — the medication should be avoided except under unusual circumstances, such as when a safer alternative has failed to achieve the desired therapeutic outcome. The panel’s expectation is that a PIM, when chosen, is chosen through shared decision-making that acknowledges its potential harms. Table 4’s separate “use with caution” recommendation exists precisely to mark drugs that raise concern but not to the level of “avoid” — because the evidence is limited or inconsistent, because the harm relative to alternatives is not large enough, or because extenuating clinical circumstances are common.

The panel also draws a distinction that matters enormously in practice: for a number of criteria, the recommendation is to avoid initiating the drug in a non-user while considering deprescribing in a current long-term user — two different actions, not one. Aspirin for primary prevention, non-vaginal estrogen and warfarin are all handled this way in the 2023 edition.

And in the panel’s own words on how the criteria should be enforced: they “are not to be used in a punitive manner.”

What changed in the 2023 update

The 2023 panel — twelve interprofessional members from medicine, nursing and pharmacy, ten of whom had served on the 2019 panel, plus ex-officio representatives from CMS, NCQA and the Pharmacy Quality Alliance — reviewed literature published between June 2017 and May 2022. The search returned 33,965 references; 1,574 went to full-text review and 451 were abstracted into evidence tables. A modified Delphi process, internal AGS review and a public comment period (58 comment forms, 200-plus comments) produced the final text.

The changes most likely to affect hospital medication-safety work:

Anticoagulation

• Rivaroxaban moved from “use with caution” to avoid for long-term treatment of nonvalvular atrial fibrillation and VTE, on evidence of higher major and gastrointestinal bleeding risk in older adults than other DOACs — particularly apixaban, but also dabigatran.
• Warfarin was added to Table 2: avoid starting warfarin as initial therapy for VTE or nonvalvular atrial fibrillation unless alternatives are contraindicated or there are substantial barriers to their use. Continuing a long-term user with well-controlled INR is treated differently, and the panel explicitly flags out-of-pocket cost and formulary restriction as equity barriers to DOAC access.
• Dabigatran remains “use with caution” (Table 4) for the same long-term indications.
• Apixaban‘s renal criterion was removed from Table 6 given evidence for its safe use in end-stage renal disease.

Other substantive changes

• Aspirin for primary prevention of cardiovascular disease moved from Table 4 to Table 2, with the recommendation now to avoid initiating it, aligning with the US Preventive Services Task Force. Deprescribing should be considered for those already taking it for primary prevention.
• Sulfonylureas: the “avoid” recommendation was expanded from long-acting agents to all sulfonylureas as first- or second-line monotherapy or add-on therapy, on cardiovascular event, all-cause mortality and hypoglycaemia risk. If one must be used, a shorter-acting agent is preferred.
• Estrogen: avoid initiating oral or transdermal estrogen in older women; consider deprescribing in current users. Low-dose intravaginal estrogen remains appropriate for its major indications.
• SGLT2 inhibitors: a new “use with caution” criterion, citing increased risk of urogenital infection and euglycaemic diabetic ketoacidosis, with a recommendation to monitor actively early in treatment. The panel is careful to note it recognises the class’s value.
• Ticagrelor was added alongside prasugrel in Table 4, with caution particularly at 75 and older.
• Anticholinergic burden: the Table 2 rationale was expanded to recognise cumulative exposure across concurrent agents, not just single-drug risk — a change echoed in Tables 3 and 5.
• Drug–drug interactions: skeletal muscle relaxants were added to the CNS-active drug categories that trigger the “three CNS-active drugs” criterion, and SSRIs were added to the warfarin interaction list.
• Delirium: opioids were added to the list of drugs that can exacerbate delirium in Table 3.
• Baclofen was added to Table 6 (avoid at eGFR <60 mL/min, encephalopathy risk); NSAIDs at CrCl <30 mL/min moved from Table 3 to Table 6.
• Decluttering: for the first time the panel systematically assessed US utilisation, moving drugs with fewer than roughly 4,000 Medicare Part D beneficiaries aged 65+ in 2020 off the main tables into Table 8. Certain over-the-counter drugs below that threshold were deliberately retained.

The 2025 Alternatives List

In July 2025 the AGS announced a companion resource, the AGS Beers Criteria® Alternatives List, published in the Journal of the American Geriatrics Society as alternative treatments to selected medications in the 2023 criteria. It provides evidence-based pharmacologic and non-pharmacologic alternatives across more than twenty common conditions, including insomnia, anxiety, delirium and chronic pain. This addresses the most common practical complaint about the criteria — that they identify what to stop without saying what to start — and it is worth pairing with the criteria themselves in any local deprescribing protocol.

Using the criteria in hospital medication-safety review

The AGS frames the criteria as “a starting point for a comprehensive process of identifying and improving medication appropriateness and safety,” not as the process itself. In practice, that means it usually functions as a screening layer feeding a pharmacist-led review rather than as a decision rule.

Admission and transition points

The highest-yield application is at transitions, where a complete list of what the patient is actually taking already has to be assembled. Running a Beers screen against the reconciled list during medication reconciliation surfaces candidates for review at the one moment when a prescriber, a pharmacist and an accurate list are all available at once. Discharge is the second such point — and the point at which a stop decision has to be communicated to a community pharmacy, which the AGS panel notes is a systemic weak spot (it points to the CancelRx script standard as one fix).

Retrospective surveillance

Beers-listed medications make good trigger-tool signals. A PIM dispensed to a patient 65 or older, particularly one on the drug–disease table with a matching diagnosis, is a reasonable trigger for chart review inside an adverse drug event surveillance programme. When a suspected event is found, causality assessment is a separate step — the Naranjo algorithm is the usual instrument for that, and its output is independent of whether the drug appears in Beers.

Clinical decision support and formulary governance

Encoding the criteria into EHR alerts is where most organisations create their own problems. Table 3 and Table 6 criteria depend on data the alert may not have — an active diagnosis, an indication, a current creatinine clearance — and the panel says so directly: a criterion “may not be easily applied as a quality measure, particularly when such subgroups cannot be easily identified through structured and readily accessible electronic data.” An alert that cannot see the exception fires on the exception, and the result is the alert-fatigue pattern already familiar from smart pump drug library governance. Age-triggered interruptive alerts on high-prevalence classes are the usual culprits.

The more durable route is governance rather than interruption: use the criteria as an input to formulary management decisions, default order-set contents and standard order sentences, so the safer option is the path of least resistance rather than the one that requires clicking through a warning.

Overlap with existing safety programmes

Several Beers criteria sit directly on top of programmes most hospitals already run. The anticoagulation criteria intersect with an anticoagulation management program and with National Patient Safety Goals content on anticoagulant therapy. The falls-and-fractures criterion — which names anticholinergics, selected antidepressants, antiepileptics, antipsychotics, benzodiazepines, Z-drugs and opioids — is a medication-review component of a fall prevention program, complementing rather than replacing risk screening with instruments such as the Morse Fall Scale. The delirium criteria pair naturally with CAM-ICU screening, and the dementia criteria with cognitive screening using the MoCA or a shorter instrument. The ciprofloxacin, nitrofurantoin and trimethoprim-sulfamethoxazole entries are a natural handoff to an antimicrobial stewardship program. And the criteria are a standing item worth mapping against the ISMP Targeted Medication Safety Best Practices self-assessment.

Because risk of drug-related harm rises with frailty and multimorbidity rather than with chronological age alone, pairing a Beers screen with a frailty measure such as the Clinical Frailty Scale gives a more defensible prioritisation than an age cutoff on its own. The panel makes the same point: two older adults of the same age can have markedly different risks.

Beers in quality measurement

Two NCQA HEDIS measures are built directly on the criteria:

• Use of High-Risk Medications in Older Adults (DAE) — potentially inappropriate medication use in all older adults, assessed from prescription data.
• Potentially Harmful Drug-Disease Interactions in Older Adults (DDE) — potentially inappropriate medication use in older adults with specific diseases or conditions.

NCQA states both measures are based on recommendations in the AGS Beers Criteria. These are health-plan measures rather than hospital measures, so the direct accountability sits with the plan; the practical consequence for hospitals and health systems is that plan-side pressure on PIM rates propagates back through prescriber networks, medical group reporting and shared-savings arrangements. NCQA, CMS and the Pharmacy Quality Alliance all held ex-officio seats on the 2023 update panel, which is why the criteria and the measure specifications track each other closely — but the measures are a subset of the criteria, deliberately narrowed to what can be computed from claims.

The gap between the two is where misuse begins. The panel writes that quality measures “must be clearly defined, easily applied, and measured with limited information and, thus, although useful, cannot perfectly distinguish appropriate from inappropriate care.” Treating a HEDIS numerator hit as a prescribing error, or a Beers-listed prescription as a reportable deviation, inverts the panel’s stated intent.

Limitations, stated by the panel

The AGS lists its own limitations plainly, and quoting them is usually more persuasive with prescribers than restating them second-hand:

1. Thin evidence base in the target population. Few clinical trials enrol enough older adults for age-specific analysis, so the panel often relied on observational studies and meta-analyses for evidence of harm.
2. Lack of diversity in study populations. Underrepresented and understudied populations are inadequately enrolled in the underlying trials, and even where diverse populations were included there was often inadequate power for subgroup outcomes.
3. US-only drug scope. The criteria include only medications available in the United States.
4. Search limits. The literature search excluded unpublished studies, non-English publications, white papers, abstracts, technical reports and other grey literature.
5. It is a blunt instrument. The panel’s own phrase — it cannot “delineate all specialized use cases and possible exceptions.”

The AGS also publishes a short set of principles for how patients, clinicians, health systems and payors should use the criteria. The load-bearing ones: medications in the criteria are potentially inappropriate, not definitely inappropriate; read the rationale and recommendation statements for each criterion, because the caveats there matter; optimal application means identifying PIMs and offering safer alternatives; and access to Beers-listed medications should not be excessively restricted by prior authorisation or coverage policy.

Beers and STOPP/START

The Beers Criteria is not the only explicit PIM tool. The STOPP/START criteria, developed in Ireland and widely used across Europe, cover similar ground with two structural differences: they are built around European prescribing patterns and drug availability rather than the US market, and they include START criteria for potential prescribing omissions — indicated treatments the patient is not receiving — which the Beers Criteria does not address at all. Organisations that want to catch under-treatment as well as over-treatment generally run both, or supplement Beers with a separate omissions review. The AGS itself does not position the two as competitors.

Frequently asked questions

What is the most current version of the Beers Criteria?

The 2023 AGS Beers Criteria, published in the Journal of the American Geriatrics Society 2023;71(7):2052–2081. AGS updates on a regular cycle — 2012, 2015, 2019 and 2023 under its stewardship — so verify the edition before citing it in policy. A companion Alternatives List was published in 2025 but does not replace the 2023 criteria.

Does “avoid” mean the drug is contraindicated?

No. The AGS panel states explicitly that “avoid” is not an absolute contraindication unless the medication’s own label says so. It means the drug should be avoided except in unusual circumstances, chosen through shared decision-making that recognises the potential harms.

How many tables are in the Beers Criteria?

The criteria proper are organised into five categories — potentially inappropriate medications (Table 2), drug–disease and drug–syndrome interactions (Table 3), drugs to use with caution (Table 4), drug–drug interactions (Table 5), and renal dosing (Table 6). Table 7 lists strongly anticholinergic drugs, Table 8 lists drugs moved off the main tables for low US use, and Box 1 consolidates the anticoagulant recommendations.

Do the Beers Criteria apply to hospice patients?

No. The criteria explicitly exclude hospice and end-of-life care settings. This exclusion is worth writing into any local screening logic, since an unfiltered age-based screen will otherwise flag entirely appropriate comfort-care prescribing.

At what age do the criteria apply?

65 and older, unless a specific criterion states otherwise. Several criteria carry their own cutoff at 75. The panel cautions that age alone is a poor risk stratifier — frailty, multimorbidity, and cognitive and functional status matter more — and encourages clinicians to use common sense in applying the criteria.

Can the criteria be used to audit or discipline prescribers?

The AGS says no: the criteria “are not to be used in a punitive manner.” They are designed to support, not replace, shared clinical decision-making, and the panel notes that structured data often cannot identify the exceptions a given criterion allows.

What is the difference between the Beers Criteria and the HEDIS DAE and DDE measures?

The Beers Criteria is the clinical reference; DAE and DDE are NCQA health-plan measures built on a claims-computable subset of it. DAE covers PIM use in all older adults; DDE covers PIM use in older adults with specific diseases or conditions. Because the measures must run on prescription and diagnosis data alone, they cannot capture the clinical exceptions the criteria permit.

Where do I get the criteria?

The full 2023 criteria are published in JAGS (doi:10.1111/jgs.18372). AGS also produces a pocket card and a mobile app via GeriatricsCareOnline.org, and patient-facing education materials through HealthinAging.org.

Related reading

• Medication Reconciliation: The Five-Step Process and What the Evidence Shows
• Adverse Drug Event Surveillance: Trigger Tools and Pharmacy-Based Screening
• Naranjo Algorithm for Adverse Drug Reactions
• Anticoagulation Management Program: Protocols, Monitoring, and DOAC-Era Reversal
• Fall Prevention Program Design
• Hospital Formulary Management
• ISMP Targeted Medication Safety Best Practices
• Patient Safety cluster overview