🦠 Prospective Audit & Feedback Stewardship Loop

Auto-Flagging Antibiotic Orders for Prospective Review

Prospective audit and feedback (PAF) begins before any human ever opens a chart: a rules engine — increasingly embedded directly in the EHR pharmacy module — continuously screens active antibiotic orders and routes a defined subset into the stewardship pharmacist's daily worklist. The CDC's Core Elements of Hospital Antibiotic Stewardship Programs and the 2016 IDSA/SHEA stewardship guideline (Barlam et al., Clinical Infectious Diseases) both identify PAF, alongside preauthorization, as the two "Level I" evidence-based interventions every acute-care program should implement.

• Strong / High: IDSA/SHEA evidence grade (2016 guideline, PAF vs. no PAF)
• 20–50: Typical daily flag volume (per 300-bed hospital)
• Action: CDC Core Elements category (one of 7 core elements)
• <24 h: Median time to first review (from order flag to pharmacist eyes)

What triggers an automatic flag

Flagging logic is built from a combination of formulary restriction lists, pharmacy dispensing triggers, and time-based rules, typically configured inside the EHR's clinical decision support (CDS) layer (Epic Bugsy/Stewardship module, Cerner St. John, or third-party platforms such as Sentri7, TheraDoc, or DoseMeRx):

• Broad-spectrum / high-risk agents — carbapenems (meropenem, ertapenem), anti-pseudomonal beta-lactams (piperacillin-tazobactam, cefepime), vancomycin, daptomycin, linezolid, and IV fluoroquinolones are flagged on first dose or within 24–48 hours • Formulary-restricted antimicrobials — agents requiring stewardship pre-approval or automatic review per the hospital's restricted-drug list (often includes colistin, ceftaroline, ceftazidime-avibactam, and newer beta-lactam/beta-lactamase-inhibitor combinations) • The "antibiotic time-out" at 48–72 hours — IDSA/SHEA and CDC both recommend a structured reassessment once culture data typically results; day-3 triggers catch any order still active without a documented reassessment • Bug-drug mismatches — pharmacy CDS cross-references the microbiology result against the active order and flags any organism resistant (or intrinsically non-susceptible) to the prescribed agent • Redundant or duplicate coverage — two agents with overlapping anaerobic or Gram-negative spectrum ordered concurrently (e.g., piperacillin-tazobactam plus metronidazole) • Renal/hepatic dose mismatch — CrCl-adjusted dosing rules flag orders where the dose or interval is inconsistent with current renal function • Prolonged duration — any course exceeding the guideline-typical ceiling for the documented indication (e.g., >7 days for uncomplicated cystitis, >8 days for ventilator-associated pneumonia per IDSA/ATS guidance) without an extension order

Programs that route flags automatically through EHR-integrated CDS rather than manual daily chart pulls report reviewing 30–60% more eligible orders with the same pharmacist FTE — the single highest-leverage infrastructure investment a stewardship program can make, per the CDC 2019 Core Elements update.

Why PAF over preauthorization

Preauthorization (requiring stewardship sign-off before an antibiotic can even be dispensed) produces faster upfront restriction of use, but PAF is generally preferred as the default strategy because it intervenes after the clinical picture has matured — culture data, source control status, and clinical trajectory are all visible by the time of review — and because it preserves prescriber autonomy, which correlates with higher long-term buy-in and durability of the intervention. The IDSA/SHEA guideline notes both strategies are effective, and many mature programs run them in combination: preauthorization for the highest-risk/highest-cost agents, PAF for the broader antibiotic population.

The Stewardship Pharmacist Chart Review

Once a chart is in the worklist, the stewardship pharmacist (often working under a collaborative practice agreement with a supervising ID physician) performs a structured review that typically takes five to ten minutes per patient. This is the diagnostic core of the intervention: every recommendation made downstream is only as good as the data reconciliation performed here.

• 5–10 min: Median review time (per flagged chart)
• 25–40: Charts reviewed / pharmacist FTE / day (typical caseload)
• ~60–75%: Documented indication present (at time of initial order, pre-PAF)
• ~80%: Culture data available by day 3 (of bacterial infections)

The five-point structured review

A disciplined PAF review walks the same checklist every time, which is what makes the intervention scalable across a large pharmacist team:

1. Indication — is a clear infectious indication documented in the chart (progress note, admission diagnosis, or order comment)? Vague indications ("fever," "prophylaxis — unspecified") are themselves a common target for feedback 2. Microbiology — are culture and susceptibility results available? Blood, urine, respiratory, and wound cultures are cross-checked against the current regimen; a negative culture at 48–72 hours with clinical improvement is one of the strongest de-escalation triggers 3. Renal and hepatic function — creatinine clearance trend (Cockcroft-Gault or CKD-EPI) is recalculated against the ordered dose and interval; acute kidney injury during vancomycin or aminoglycoside therapy is a high-priority flag 4. Duration to date and anticipated stop date — is there a documented or implied endpoint consistent with syndrome-specific guidance (e.g., IDSA CAP 2019: minimum 5 days; IDSA/ATS HAP/VAP 2016: 7 days for most pathogens)? 5. Drug interactions and redundancy — QT-prolongation stacking (fluoroquinolones, azoles, macrolides), warfarin potentiation (metronidazole, fluoroquinolones, TMP-SMX), and anaerobic/Gram-negative double coverage are systematically screened

Most programs complete this review using a standardized note template or structured field set inside the stewardship software so that data elements are captured consistently for later program reporting.

Data sources beyond the chart

A thorough review pulls from more than the active medication list: the microbiology lab's antibiogram context (is this organism's local susceptibility pattern typical or an outlier?), prior culture history (colonization vs. new infection), allergy/adverse-reaction documentation (true penicillin allergy vs. historical intolerance — a major driver of unnecessary broad-spectrum use), and nursing administration records to confirm the ordered regimen is actually the one being given. Many programs also review source-control status (has an abscess been drained, a line been pulled, an obstruction been relieved?) since antibiotics alone rarely resolve infection without adequate source control.

Formulating a Specific, Actionable Recommendation

Generic feedback ("consider narrowing") is rarely acted upon. The recommendations that move acceptance rates are concrete, tied to a specific data point, and propose an exact next step — a named drug, a dose, a route, or a stop date — that the prescriber can accept with a single click or a one-line order change.

• 25–40%: De-escalation opportunity rate (of PAF-reviewed regimens)
• ~20%: IV-to-PO eligible on review (of hospitalized courses)
• 30–50%: Duration excess identified (of reviewed courses)
• ↑ %T>MIC: Extended-infusion beta-lactam benefit (esp. for elevated MIC pathogens)

The four recommendation archetypes

Nearly every PAF recommendation falls into one of four categories:

• Dose optimization — adjusting for renal/hepatic clearance, switching to extended or continuous infusion for time-dependent beta-lactams (piperacillin-tazobactam, meropenem) to maximize %T>MIC in critically ill or elevated-MIC scenarios, or correcting a subtherapeutic vancomycin AUC/MIC target (AUC24 of 400–600 mg·h/L per the 2020 ASHP/IDSA/PIDS/SIDP vancomycin consensus guideline) • De-escalation — narrowing empiric broad-spectrum therapy once culture and susceptibility data return; classic example: stopping empiric vancomycin plus piperacillin-tazobactam and switching to cefazolin once blood cultures grow methicillin-susceptible Staphylococcus aureus • IV-to-PO conversion — switching to an equally bioavailable oral agent once the patient is hemodynamically stable, tolerating oral intake, and the infection does not require sustained high tissue concentrations; fluoroquinolones, linezolid, metronidazole, fluconazole, TMP-SMX, and doxycycline all have near-complete oral bioavailability and are prime conversion candidates • Duration recommendation — proposing a specific stop date anchored to a guideline: 5 days for uncomplicated CAP with early clinical stability (IDSA/ATS 2019), 7 days for most Gram-negative bacteremia with source control (per the 2023 IDSA-endorsed shorter-course literature), 3–7 days for uncomplicated cystitis

Anchoring the recommendation to evidence

The strongest recommendations cite the exact supporting data inline — "Blood culture 8/14 grew MSSA, susceptible to cefazolin; no additional coverage needed" is far more persuasive and more often accepted than an unqualified "please narrow." Many programs use standardized SBAR-style (Situation-Background-Assessment-Recommendation) note templates specifically because SBAR structuring has been shown in nursing and quality-improvement literature to increase the clarity and actionability of clinical communication, translating directly into higher PAF acceptance rates.

Delivering Non-Restrictive, "Suggest Not Require" Feedback

The defining feature of PAF — what separates it from restrictive preauthorization — is that the recommendation is communicated as a suggestion the prescriber retains full authority to accept, modify, or decline. This preserves clinical autonomy and, counterintuitively, tends to produce more durable behavior change than mandates, because prescribers internalize the reasoning rather than simply complying.

• ~85–90%: Direct verbal contact acceptance (phone/in-person vs. note-only)
• ~50–65%: EHR note-only acceptance (asynchronous, no live contact)
• ICU/oncology: "Handshake stewardship" model (embedded co-rounding programs)
• <24 h: Recommended response window (from feedback to documented decision)

Communication channels, ranked by impact

Published stewardship literature consistently shows a gradient of effectiveness across delivery channels:

• In-person or "handshake stewardship" — an ID physician or pharmacist walks rounds directly with the primary team (common in ICUs, oncology, and transplant units); this face-to-face model, popularized by programs at institutions such as the University of Utah, produces the highest acceptance rates because it allows real-time clarification and relationship-building • Telephone call — a direct conversation with the ordering provider or resident; still highly effective and scalable across a whole hospital, not just high-acuity units • EHR best-practice advisory or secure in-basket note — asynchronous and highly scalable, but acceptance rates are measurably lower than live contact because the message can be deferred, missed amid alert fatigue, or lack the nuance of a conversation • Combined/tiered approach — many mature programs default to an EHR note for lower-acuity recommendations and escalate to a phone call or in-person visit for higher-stakes or previously-ignored recommendations

The language matters as much as the channel: IDSA/SHEA explicitly frames PAF communication as "suggest," "recommend considering," or "would favor" rather than "must," "require," or "will be discontinued." This non-restrictive framing is what allows PAF to run without formulary lockouts and is a core reason it is rated Level I evidence alongside preauthorization rather than as a lesser alternative.

Documentation and closed-loop communication

Regardless of channel, every recommendation is documented in the chart (a dedicated stewardship note type in most EHRs) and logged in the stewardship program's tracking software, both for legal/clinical documentation purposes and to feed the acceptance-tracking analytics described in the next stage. Closed-loop communication — confirming the prescriber received and understood the recommendation, not just that a message was sent — is considered a best practice, particularly for handshake and phone-based delivery, where a brief teach-back ("so the plan is stop day 5, correct?") measurably reduces miscommunication-driven overrides.

Acceptance and Override Tracking

Every recommendation resolves into one of a small number of tracked outcomes, and structured logging of both acceptance and the reason for any override is what turns individual interventions into a learning system — feeding quarterly antimicrobial stewardship committee review, prescriber-level feedback, and refinement of the flagging rules themselves.

• 70–90%: Published PAF acceptance rates (range across program literature)
• Clinical judgment: Most common override reason (physician disagrees with data interpretation)
• ~10–15%: Partial acceptance rate (modified vs. fully accepted/declined)
• Committee review: Repeat-override provider flag (triggers escalation/education)

The acceptance taxonomy

Stewardship software (Sentri7, TheraDoc, Epic Bugsy, or locally built registries) typically codes each recommendation outcome into a small structured taxonomy:

• Fully accepted — the exact recommendation is implemented, usually within 24 hours • Partially accepted — a modified version is implemented (e.g., prescriber narrows to a different but still appropriate agent than the one suggested) • Declined / overridden — the recommendation is not implemented, with a required reason code • Deferred — the team indicates they will reassess at a defined future point (e.g., "will reassess after ID consult tomorrow")

Acceptance rates in the published literature cluster between 70% and 90%, with early studies such as Standiford et al. (2012, Infection Control & Hospital Epidemiology) reporting acceptance around 80% and substantial associated reductions in antimicrobial expenditure. Rates vary by institution culture, recommendation type (duration and IV-to-PO recommendations tend to have higher acceptance than de-escalation recommendations that touch active clinical decision-making), and delivery channel.

Override reason coding and its uses

Common structured override reasons include: clinical judgment/disagreement with the data interpretation, patient too unstable to narrow, awaiting additional culture or imaging data, ID consult already involved with a differing plan, and allergy or intolerance concerns not fully captured in the chart. This taxonomy serves three purposes: it identifies systematic gaps in the flagging logic (a recurring "awaiting culture" override may mean the flag is firing too early), it surfaces individual prescribers or services with disproportionately low acceptance for targeted education, and it provides the denominator data stewardship committees need to demonstrate the program's clinical reasoning is sound even when a particular recommendation is not implemented.

Cumulative Days-of-Therapy Saved and Program-Level Impact

A single accepted recommendation might shave two or three days off one patient's antibiotic course. Aggregated across thousands of reviews per year, PAF programs produce measurable, reportable reductions in total antimicrobial exposure, cost, and — over longer time horizons — downstream harms like Clostridioides difficile infection and the emergence of resistant organisms.

• 10–30%: DOT reduction, mature programs (vs. pre-implementation baseline)
• $200–900: Antimicrobial cost avoidance (per patient reviewed (program studies))
• ↓ up to ~30%: C. difficile infection reduction (associated with sustained PAF + de-escalation)
• <1 year: Program payback period (typical, pharmacist FTE vs. drug savings)

Rolling individual interventions into program metrics

Every accepted recommendation is tagged with an estimated days-of-therapy (DOT) impact — a duration shortened from 10 days to 7 contributes 3 DOT saved; a de-escalation from a two-drug regimen to one contributes 1 DOT saved per day remaining on therapy. These are summed continuously and typically reported as DOT saved per 1,000 patient-days, mirroring the same NHSN Antimicrobial Use denominator used for national benchmarking, so a program can directly compare its trajectory to its own historical baseline and to peer institutions.

Mature, well-resourced PAF programs commonly report a 10–30% reduction in broad-spectrum antimicrobial DOT within 12–24 months of full implementation, alongside measurable cost avoidance in the hundreds of dollars per patient reviewed once pharmacist labor costs are netted against drug expenditure savings — a return on investment that is one of the most consistently reproduced findings in the health-economics stewardship literature.

Downstream clinical and safety outcomes

Beyond direct antimicrobial savings, sustained PAF programs are associated with reductions in Clostridioides difficile infection rates (antibiotic exposure, particularly to fluoroquinolones, clindamycin, and broad-spectrum beta-lactams, is the dominant modifiable risk factor for CDI) and with slower emergence of resistant organism colonization on the units where the program is most active. These outcomes, alongside DOT and cost trends, are the standard content of the quarterly report the antimicrobial stewardship committee — a body required under CMS Conditions of Participation and Joint Commission accreditation standards for essentially all US acute-care hospitals — reviews to sustain institutional support and staffing for the program.

CDC modeling underlying the national "Get Smart" and AR Solutions initiatives estimates that if every US acute-care hospital ran a fully implemented PAF-based stewardship program, total inpatient antibiotic use could fall by roughly 20%, meaningfully slowing the population-level emergence of resistant organisms — the same logic that makes PAF, not just preauthorization, a required CMS Condition of Participation element for hospital antibiotic stewardship programs since 2020.