Discussion

Although our intervention did not further reduce the median IV antimicrobial duration, our ability to further reduce IV duration was likely constrained by strong baseline AMS performance. Our baseline median was already comparable to the postintervention mean IV duration achieved in other face-to-face, general medicine AMS programs. ^{Reference Manuel, Burnand and Bady9} This aligns with the typical 48–72 hour period required to obtain preliminary microbiology results, evaluate inflammatory markers, and assess clinical stability. Additionally, some IV-to-oral switch opportunities may have been delayed, particularly over weekends when the AMS service was not available, further limiting the potential for earlier transition. As this intervention did not aim to influence prescribing at the point of initiation; optimizing prescribing at this earlier decision point represents an important next step for reducing total treatment duration.

Our pharmacist-led digital model achieved a 78% recommendation acceptance rate, comparable to other in-person handshake stewardship interventions (80%–91%) in general medicine cohorts. ^{Reference Morita and Mueller18–Reference Stromberg, Marx and Mavrogiorgos20} Similar acceptance rates (80%) were reported by Neuner et al. with thrice-weekly ID pharmacist and ID physician review, ^{Reference Neuner, Atkinson and Ilges19} while higher rates were seen in more intensive daily models (90–91%) described by Morita et al. and Sapozhnikov et al. respectively. ^{Reference Morita and Mueller18,Reference Sapozhnikov, Hoff, Adams, Santarossa, Kalathiya and Albarillo21} While some studies have indicated preference for face-to-face feedback, ^{Reference Morita and Mueller18} our digital model achieved engagement with a more scalable alternative. This model enabled review of a larger number of patients across multiple wards, not only a single patient care unit, ^{Reference Morita and Mueller18} supporting broader stewardship reach.

The intervention was associated with reduced LOS during the intensive phase, particularly among patients without bacteremia. This is a meaningful finding in this patient population, where hospital admissions are often prolonged. ^{Reference Newnham22–Reference Swamy, Sood and Kapil24} This aligns with prior focused AMS initiatives demonstrating shorter LOS in patients with cellulitis, ^{Reference Ritchie, Cutfield and Lee25} older adults, ^{Reference Mauro, Kannangara, Peterson, Livert and Tuma26} and general hospital inpatients. ^{Reference Wathne, Harthug and Kleppe27} We hypothesize that frequent IV-to-oral switch recommendations (89%) increased prescriber confidence for earlier discharge, without increasing readmissions, a conclusion further supported by the reduction in total antibiotic consumption observed in the long-term evaluation. A limitation was the lack of detailed data capturing decision-making contributing to LOS.

Antibacterial utilization trends indicate the intervention reinforced and sustained high-quality prescribing practices. Unlike other studies showing reduction in broad-spectrum antimicrobials, ^{Reference Neuner, Atkinson and Ilges19,Reference Sapozhnikov, Hoff, Adams, Santarossa, Kalathiya and Albarillo21,Reference Tang, Gupta and Majid28,Reference Kaplun, Monteforte, Tirmizi, Abate, Psevdos and Go29} our site already tightly regulated these agents, with notable decline observed only in piperacillin-tazobactam. Instead, sustained decreases were seen in total antibacterial use following implementation, including narrow-spectrum IV agents (flucloxacillin and ampicillin) without compensatory broad-spectrum increases. This suggests the intervention may have promoted early cessation of therapy.

Implementation of this digital review and communication model enabled consistent, multidisciplinary input while avoiding logistical challenges of daily in-person reviews, which have limited sustainability in other pilots. ^{Reference Morita and Mueller18–Reference Sapozhnikov, Hoff, Adams, Santarossa, Kalathiya and Albarillo21,Reference Tang, Gupta and Majid28} Nonetheless, the model still required substantial time and resources which we recognize may not be practical in other settings. An implementation science-informed approach may help tailor the intervention model to local contexts, identify barriers and facilitators to uptake, and guide adaption of less resource-intensive strategies such as targeting specific infections/antimicrobials or employing less demanding models of audit and feedback to better balance sustainability and workforce capacity. ^{Reference Monsees, Wirtz and Krein30}

Limitations of this study include its before-and-after design, which is vulnerable to unmeasured confounding and secular trends. The unchanged median IV duration likely reflects the timing of AMS reviews: recommendations were issued around midday, after the third IV dose had usually been administered, meaning the day was still counted as an IV day even when a switch occurred promptly. This highlights a broader limitation of DOT-based metrics, which are sensitive to dosing times rather than clinical decision-making. Given that total antibacterial use declined after implementation, measures such as total duration of therapy or time to switch may better capture the impact of AMS interventions in future work. There is increasing evidence that shorter therapy durations are as effective as longer durations for common infections seen in general medical units such as community acquired pneumonia. ^{Reference Dinh, Ropers and Duran31}

In summary, this study demonstrates that a pharmacist-led digital AMS model is feasible and associated with high engagement, sustained low use of IV antimicrobials, and reductions in both total antibacterial use and LOS. Future AMS initiatives should consider alternative outcome measures, such as total duration of therapy, to further assess their impact on LOS.