Context-specific impact of antimicrobial stewardship on antibiotic use and antibiotic resistance in hospitals in a lower-middle income country - results from an implementation study with a controlled interrupted time series design in Vietnam

29 30

31 High-quality evidence regarding the impact of anti microbial stewardship (AMS) is 32 limited in Asia. In this study, we aimed to determine the effe cts of a pharmacist-led 33 prospective audit with feedback intervention , as part of an AMS programme following 34 national guidelines, in two provincial-level general hospitals in Vietna m, a lower-middle-35 income coun try. 36 37

38 We performed controlled interrupt ed time-series analyses to evaluate the impact of an 39 AMS interven tion on antibiotic use in days of therapy per 1000 patie nt-days, antibiotic 40 non-susceptibility percentage and patient out comes . In each hospital , four wards 41 received the interven tion and four wards acted as controls. Pre-interv ention periods 42 began in January 2019 and contin ued to May 2020 (Hospital 1) and July 2020 (Hospital 43 2), followed by a 12-month post-in tervention period. 44 45

46 In Hospital 1, th e interven tion was associated with a reduction in the level of antibiotic 47 use (95.9, 95% CI [10. 9, 180 .8]), alt hough there was no evidence for a change in trend 48 (0.9[-3.6 , 5.4]) . In con trast, in Hospital 2, there was no evidence for a change in either 49 level (6.3[-83.7 , 96.3]) or trend (-2. 1[-4.8 , 0.6]). I n Hospital 1, we observed a decreasing 50 trend in antibiotic non-suscep tibility among hospital-acquired Escherichia coli to 51 aminoglycosides (odd s ratio: 0.87[0.78, 0 .97]), but incre asing for Pseudomonas 52 aeruginosa to carbapenems (1.1 1[1.00, 1 .22]) and Acinet obacte r spp. to aminoglycosides 53 (1.07[1.00 , 1.27]) . In Hospital 2, evi dence indicated decreasing trends in Acinetobact er 54 spp. to carbapenems (0.96[0.88 , 1. 00]), ciprofloxacin (0.93[0 .85, 1 .00]), and piperacillin-55 tazobactam (0 .94[0.7 8, 1.0 0]), but i ncreasing for P. aeruginosa to ami noglycosides 56 (1.07[1.00 , 1.20] , ciprofloxacin (1.4 5[1.18, 1 .77]), and cef tazidime (1.0 3[1.00, 1 .19]). We 57 did not find evidence that the in te rvention was associated with chan ges in mortality or 58 hospitalisation costs. 59 60

61 The impact of AM S varied between the two hospitals, highlighting c ontex t-specific 62 implemen tation ch allenges and th e necessity to monitor changes in antibiotic resistance 63 over time to tailor interven tions th at respond to local resi stance epidemiology. 64 65 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 3 Key words 66 Antimicrobial stewardship 67 Impact 68 Asia 69 Vietna m 70 Antibiotic use 71 Antimicrobial resistance 72 Mortality 73 Hospitalization cost 74 75 76 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 4 WHAT IS ALREADY KN OWN ON THIS TOPIC 77 • Previous systematic reviews and recent interrup ted time series (ITS) studies have 78 shown large variations in the impact of antimicrobial stewardship (AMS) on 79 hospital antibiotic use and resistance. Few previous ITS studies of AMS have used 80 a control in their analysis. 81 82 WHAT THIS STUDY ADDS 83 • Using a strong quasi-experimental study design with a control group , we 84 generated empirical evidence on t he multi- facet ed effects of an AMS intervention 85 on antibiotic use and resistance in hospitals in a middle-income coun try in Asia. 86 • The impleme ntation of prospectiv e audit and feedback in the conte x t of 87 established AMS programmes following the national guidelines had different 88 effects on total an tibiotic use and antibiotic non-suscep tibility propo rtions 89 among com mon hospital-acquired pathogens found in Vietna m. 90 91 HOW THIS STUDY MIGHT AFF EC T RES EAR CH, PRACTI CE O R POLI CY 92 • This study contributes evidence from a strong study design on the e ffects of AMS 93 implemen tation in hospitals in a m iddle-income coun try in Asia, which can inform 94 future AM S programmes in similar settings. 95 • It highlights the need to mo nitor t he emergence and spread of antibiotic 96 resistance in hospital settings in Asia to support development and design of 97 novel interventio ns for AMS and other programmes to respond to th e fast-98 changing resistance profiles of bac terial pathogens in hospitals. 99 100 101 102 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 5

103 Widespread use of antibiotics in humans, food production ani mals and spillover into the 104 environmen t has accelerated t he e mergence and trans mission of dru g-resistant 105 bacteria 1 . Anti microbial stewardship (AMS) interventions are designed to target 106 inappropriate antibiotic use to reduce selective pressure on resistant bacteria 2 . A recen t 107 meta -analysis of the global impact of AMS programmes reported that such 108 interventions were associated with an estimated me an reduction in t he proportion of 109 patients receiving antibiotic prescriptions by 10% (95% CI: [4%, 15%]) and a mean rate 110 ratio of 0.72 (95%CI: [0. 56, 0. 92]) in the consump tion rate measured b y defined daily 111 doses per 100 patient-days 3 . Impo rtantly, AM S programmes need to monitor the i mpact 112 on mortality to ensure interve ntio ns do not harm patients as well as impact on antibiotic 113 use and resistance. Unfortu nately , such evidence for hospital inpatien ts is currently 114 insufficient 4 5 . A meta -analysis including 221 studies across 34 countries by Davey et al 115 found that mortality risks were sim ilar between intervention and con t rol groups 4 . 116 117 The two study designs conside red appropriate for evaluating the impact of AMS 118 interventions are randomised cont rolled trials and quasi-experimenta l studies (non-119 randomised controlled trials, controlled before-and-after designs, and interrupted time 120 series) 4 . Recent evidence of t he eff ect of AMS interv entions on resistance outco mes 121 comes from studies with weak desi gns. For example , between 2012 a nd 2017, only 8 of 122 26 studies used interrupted time series, and none included a control group in their 123 analysis 5 . In recent studies, only on e study in Canada used a control g roup and 124 demonstrated th e impact of a com prehensive AMS programme with a sustained 125 reduction of hospital-acquired anti biotic-resistant organisms 6 . Large heterogeneity in 126 study designs, AMS interventions, and how resistance was evaluated (denominators of 127 outcom e measures), along with un controlled contex t-specific confo u nding factors, have 128 contributed to variations in the re ported impact of AMS programme s on resistance 5 . 129 130 In this study, we aimed to evaluat e the impact of an AM S interven tio n in two provincial 131 general hospitals (Hospital 1 and Hospital 2) in Vietna m, a lower-mi ddle-income coun try 132 in Asia. W e hypothesised that t he AMS interven tion would reduce an tibiotic use without 133 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 6 negatively impacting patien t outc omes, and reduce the proportions of patients carrying 134 resistant organisms for clinically i mportant bac terial species. Vietna m developed its first 135 national action plan for controlling antimicrobial resistance and initiat ed discussions on 136 AMS impleme nta tion in a local hospital network in 2013 7 . Since t hen , several guidelines 137 for antibiotic treat ment a nd AMS implemen tation in hospitals have b een issued and 138 local hospitals have started their AMS programmes following guideli nes from the 139 Ministry of Health (MoH) 8 . We pre viously reported on our implemen tation research to 140 assess the feasibility of AMS interv entions in these two provincial hospitals in 141 collaboration with the Duke An tim icrobial Stewardship Outreach Ne t work 9 . The 142 theoretical fra mework for AMS implementa tion in this research was based on the 143 assumption tha t hospitals are complex adaptive systems and that AMS tea ms could 144 leverage their unique charact eristics and interconnec tions to develop a locally feasi ble 145 and sustainable programme. Prosp ective audit and feedback (PAF) w as chosen as the 146 core AMS interven tion implem ent ed at these two hospitals based on evidence from a 147 previous systematic review on effective behaviour change in terventio ns for antibiotic 148 prescribing in hos pitals 4 . 149 150

151 Study setting and population 152 The study was implemented in two provincial hospitals in Vietnam: H ospital 1 (1000 153 beds) and Hospital 2 (2000 be ds). We selected 8/26 clinical wards in Hospital 1 and 8/27 154 clinical wards in Hospital 2, equally divided between interven tion and control groups. 155 Ward selection and assignment we re described previously 9 . Briefly, w ards were selected 156 based on two criteria: (1) higher-than-average antibiotic use in th e ho spital based on 157 pharmacy-reported data and (2) willingness of the ward head to pa rticipate. All 158 inpatients in the study wards during evaluation periods were include d. Allocations of the 159 wards in the two hospitals were si milar in terms of clinical specialties, with four 160 intervention versus con trol ward p airs, as shown in Figure 1. Detailed characteristics of 161 these two hospitals are presented in Table 1S (Supplemen tary Data). 162 163 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 7 Before the interven tion, an AM S te am was formally established at the two hospitals with 164 the implem enta tion of the pre-au t horisation policy (which required d octors to obtain 165 approval from the head of the clinical ward and a d irector board re pr esentative before 166 using antibiotics on the restricted antibiotic list) following the 2016 national AMS 167 guidelines 10 . Infectio n prevention and control (IPC) guidance from th e MoH has been 168 issued since 2009 ( 18/2009/TT-BYT) with documents and training provided by the MoH 169 to the hospitals since then 11 . The 2 016 AMS guidelines included the role of IPC staff in 170 the AMS co mmi ttee and described the set of responsibilities in implementi ng protocols 171 for the isolation of patients with m ultidrug-resistant organisms, alon g with basic IPC 172 measures such as hand hygiene, u se of personal p rotective equipme nt, sterilisation of 173 medical equipment , enh anced mo nitoring, and outbreak investigatio ns. Although t here 174 were cases of COVID-19 in 2020 in some specific areas of Vietnam, th e COVID -19 175 pandemic did not affect th e two h ospitals until late April 2021 in Hos pital 1 and July 176 2021 in Hospital 2, towards the en d of the intervention periods 12 . 177 178 Interventions 179 At the beginning of the project , an AMS team was established to collect baseline data 180 for assessments of needs, gaps, str engths, and weaknesses to inform planning of the 181 intervention 9 . As part of the PA F a ctivity, clinical pharma cists made weekly visits to the 182 intervention wards to review antibiotic prescriptions for patients and provide 183 recomme ndations for improveme nt where needed. During t his intervention period, both 184 hospitals still maintained the pre-a uthorisation policy and routine hospital-level IPC 185 activities in all clinical wards a s usu al, including the study wards. Figure 1S outlines the 186 timeline of project activi ties implement ed before, during, and after t he interven tion 187 period, when PAF started on 01 Ju ne 2020 in Hospital 1 and 29 July 2020 in Hospital 2. 188 189 During the one-year int ervention period, the PAF ac tivity at Hospital 1 was led by two 190 clinical pharmacists who conducte d a total of 1,890 PAF reviews, whil e Hospital 2 191 assigned four clinical pharmacists to conducted a total of 1,628 PAF r eviews 9 . 82 192 recomme ndations were made am ong the reviews at Hospital 1 (75% were accepted by 193 the treati ng doctors), and 128 were made at Hospital 2 (33% accept e d) . Common 194 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 8 recomme ndations included de-esc alation of antibiotics, microbiology and additional 195 testing, m edication switches, a nd documen tation of an tibiotic indications in the charts. 196 Recom mendations were com muni cated through face -to- face discussions, notes 197 attach ed to patient medical charts, or phone calls. AMS teams also prepared a monthly 198 summary PA F report for each inter vention ward in both hospitals, and clinical 199 pharmacists presented this report in ward meetings and attended routine clinical ward 200 rounds. 201 202 In addition to PAF , doctors in the interven tion wards also participated in evaluation 203 activities at baseline and during the intervention period conducted b y the AMS tea m to 204 inform planning and monitoring, i ncluding retrospective review of antibiotic 205 prescriptions 13 and repeated surve ys on knowledge, attitude, and practices (KAP) related 206 to antibiotic use, AMR , and AMS ; K AP surveys were also completed b y doctors in the 207 control wards (Figure 1S, Supplem entary Data) . Training was delivere d by experts from 208 national universities for medicine a nd pharmacy and hospitals for tro pical diseases, 209 focusing on antibiotic treat men t f or common infec tions, surgical prophylaxis, antibiotics, 210 and the use of microbiological tests and interpretation of microbiology results. 211 212 Patient and public involment 213 Patien ts and/or the public were not involved in the design, or conduct, or reporting, or 214 dissemination plans of this researc h. 215 216 Outcome measures 217 Primary outco mes: 218 • Antibiotic use: The primary outco m e is the amount of an tibiotic use in Days of 219 Therapy (DOT) per 1000 patient-d ays on a weekly time interval. Raw patient-lev el 220 antibiotic prescription data from hospital information systems (HI S) were 221 extract ed together with patien t administrative , diagnosis di scharge o utcom es and 222 bed day information. From t hese data, we calcula ted the an tibiotic use indicators 223 overall and by Anatomical Therapeutic Chemical (ATC) classification of chemical 224 therapeutic subgroup defined by the World Health Organization ( W HO) . Pa tients 225 could move between study wards; each patient was coun ted only once under 226 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 9 each grouping (and antibiotics used by the patient on a specific day were 227 counted for the ward where the pa tient stayed on tha t day). We also described 228 antibiotic use before and after the start of the AMS in terventio n by calculating 229 the proportions of patients admitt ed to each study ward who used at least one 230 antibiotic, and proportions of all used antibiotics by AWaRe (Access, Watc h, 231 Reserve, and Ot her) groups (2021 version) defined by WHO 14 . 232 233 • Antibiotic no n-susceptibility among hospital-acquired isolates: Microbi ology data 234 were deduplicated, i.e. if a patient had several specimens collected within 30-235 days, then the duplicate results we re excluded . Hospital-acquired isolates were 236 defined as those from specimens sampled at least 48 hours after hospital 237 admission (counted for the first positive sample of the same specime n type and 238 bacterium only). Hospital-acquired isolates were identified and analy s ed for the 239 change in the non -susceptibility proportion after interven tion. We me asured the 240 proportions of antibiotic non-susceptibility in five common organism s identified 241 in routine clinical investigations (specimens from all bodily sites, excluding 242 specimens for screening purposes) : Escherichia coli, Klebsiella spp ., Ac inet obacte r 243 spp ., Pseudomo nas aeruginosa, an d Staphylococcus aureus . Raw data for antibiotic 244 susceptibility testing results were e xtracted from the WH ONET dat abase of each 245 hospital and interpreted using the AMR R package 15 (interpretation u sing CLSI 246 guidelines 2023). Non-susceptibility proportions were calculated as the ratio of 247 the nu mber of non-susceptible isolates to the nu mber of tested isolates for a 248 specific organism (isolates were de duplicated by patient and specimen type). We 249 reported microbiology data follow ing the recomm endations of the MICRO 250 framework (Supplemen tary Data) f or the following pathogen-drug combinations 251 which were considered relevant in the local epidemiological context : 252 • E.coli and Klebsiella spp.: third-generation cephalosporin (ceftriaxone or 253 ceftazidime) , a minoglycoside (gentamicin and one of amikacin or tob ramycin) , 254 fluoroquinolone (ciprofloxacin), ca rbapenem (one of ertapene m, imi penem , 255 meropenem , or doripenem); 256 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 10 • P. ae ruginosa and Acinetobac ter sp p.: third-genera tion cephalosporin 257 (ceftazidime), a minoglycoside (one of amikacin or tobramycin), 258 fluoroquinolone (ciprofloxacin), ca rbapenem (one of imipene m, m eropenem, 259 or doripenem), piperacillin-tazobactam , aztreona m , colistin ; 260 • S. aureus: Methicillin-resistan t (MR SA) (oxacillin or cefoxitin). 261 Secondary outco mes: 262 • In-hospital mortali ty: In- hospital m ortality per 1000 admitted patients on a 263 weekly time interval, including pati ents who died in hospital and thos e who were 264 discharged to die at home. 265 • Cost of hospitalisation : Includes all costs incurred during each hospit al admission 266 as recorded in the medical record of each patient af ter hospital discharge. This is 267 direct medical costs (including all types of costs: drugs, medical servi ces, 268 procedures, consumables, tes ts, be d and room services) pa id to the hospital, 269 either by the patient ou t of pocket or by a third-party payer ( such as health 270 insurance). Direc t non- medical costs and indirect costs were not included. All 271 costs were converted from Vietn a m Dong to US Dollar in 2021 value s, with costs 272 incurred in 2019 and 2020 adjusted to the equivalent values in 2021 using Gross 273 Domestic Product deflation rates 16 . 274 275 For antibiotic use, mor tality, and h ospitalisation costs, before-inteve ntion tim e series 276 were available from 1 Jan 2019 to 1 Jun 2021 in Hospital 1 and from 1 Jan 2019 to 29 Jul 277 2021 in Hospital 2. Antibiotic non - susceptibility data were available f rom 1 Jan 2014 to 278 31 Dec 2021 in Hospital 1, excep t f or S. aureus from 1 Jan 2017 to 31 Dec 2021 and from 279 1 Dec 2017 to 31 Dec 2021 in Hos pital 2. 280 281 Statistical methods 282 All analyses were conducted separ ately for each hospital. The proportion of patients 283 with any antibiotic use during their hospital stay, DOT per 1000 patient-days, D OT 284 percentage by AWaRe classificatio n, in-hospital mor tality per 1000 admitted patien ts, 285 and cost of hospitalisation were su mmarised by pre- and post-intervention periods for 286 all study ward s and for each ward pair. Out come variables were visualised and 287 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 11 decomposed to examine pote ntial patterns, trends, and seasonality. Based on the 288 available observations and perio di c patterns, t he aggregate unit , i.e. week or month, for 289 each outco me was determined . 290 291 An interrupted tim e series (ITS) design was used to compare longitudinal changes in the 292 post-interven tion period to a hypo thetical scenario in which t he inter vention did not 293 occur. W e then perfor med a contr olled interrupted time series (CITS) analysis that 294 incorporated both control and inte rvention groups into an ITS model. Segmen ted 295 regression models we re used to es timate the effe cts of the int erventi on for both ITS and 296 CITS, which are shown in the suppl emen tary data. Specifically, a ntibiotic non-297 susceptibility was modelled us ing l ogistic regress ion, while other out comes were 298 modelled using linear regres sions. 299 300 Segmen ted regression model assu mptions were checked by examini ng the residuals, 301 particularly temporal correlation using ACF/PACF plots and Ljung-Box test. For linear 302 segmented regression, an Autoreg ressive Integrated Moving Average (ARIMA) model 303 was added to the regres sion model to adjust for non-stationarity, au t ocorrelation, and 304 seasonality. ARIMA model selectio n was performed using an automated process in the R 305 function au to.ari ma(). 306 307 For logistic regres sion, the models were first run without any lags of antibiotic non-308 susceptibility proportions. In case of auto-correlation in t he simulate d residuals, the 309 models were re-run afte having added lags of antibiotic non-suscepti bility propo rtion as 310 suggested by the ACF plots. In case of presence of overdisper sion, the logistic 311 regression models we re re-run replacing the binomial distribution by a quasibinomial. 312 Finally, for all models with less tha n 10 events (antibiotic use or antib iotic non-313 susceptibility) or non-events per co-variable, the models were re-run with L 2 314 regularization (ridge) in order to a void biases and large confidence i ntervals for the 315 parameter estima tes. Th e optimal value of the shrinkage parameter was searched by 316 cross-validation as implemented by the cv.glmn et() func tion of the gl mnet R package . 317 (See additional information in t he methods provided in Supplementa ry Data). 318 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 12 All analyses were conducted in R (v.4.3 .1; R Core Team 2022) . 319 320 This study report follow s the criteria described for nonrandomised evaluations of 321 behavioral and public health interv entions 17 (see TREND S tate men t Checklist in 322 Supplemen tary Data) . 323 324

325 Figure 1 illustrates the inpatient ad missions to the eight study wards i n each hospital, 326 both prior to and following the implementa tion of the AM S programme. I n total , there 327 were 45,623 patients in the in terve ntion group and 51,444 patients in the control group 328 in Hospital 1, while Hospital 2 reported 18,842 and 25,401 patie nts, r espectively. A 329 summary of patient de mographic characteristics by study group, war d pair and hos pital 330 is available in Table 2S (Supplementary Data). 331 332 In Hospital 1, th e most com mon di agnoses in the intervention group included 333 injury/poisoning (18.3%), gastroint estinal (13.6%) , infectious (12 .5%), and respiratory 334 (10.9%) . In the co ntrol group, the predominant diagnoses were respi ratory (29.3%), 335 infectious (14.8 %), digestive (13.4 %), cardiovascular (9.8%) , and genitourinary (8.0%). I n 336 Hospital 2's intervention group, th e leading diagnoses were injury/p oisoning (38.3%), 337 cardiovascular (12.2%), gastrointes tinal (9.5%) , and respiratory (8.2%). In the control 338 group, the most frequen t diagnoses were oncological (29.7%), gastrointestinal (26.2 %), 339 infectious (14.4 %), and respiratory (9.1%) (Table 3S, Suppleme ntary D ata). 340 341 Antibiotic use 342 Proportion of patients with any antibiotic use 343 Higher proportions of patients with any antibiotic use were observed in the interven tion 344 group compared to the control group across most diagnostic categories in Hospital 1, 345 except for infec tious, genito- urinar y, and respiratory d iagnoses durin g both periods. 346 Conversely, in Hospital 2, the proportions of antibiotic use were lowe r for circulatory, 347 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 13 gastrointestinal, and not -elsewher e-classified diagnoses in the interv ention group (Table 348 3S, Supplemen tary Data) . 349 350 The proportion of patients with any antibiotic use decreased slightly i n the post-351 intervention period for both interv ention and control group wards in Hospital 1, with 352 respective absolute reductions of 1.3% (from 71 .4% to 70 .1%) and 0. 6% (from 67.0 % to 353 66.4%) (Table 1). In con trast, Hospital 2 experienced an absolute reduction in antibiotic 354 use in intervention wards of 4.4% ( from 68.3% to 63.9%) and an abso lute increase of 355 4.3% (from 72 .0% to 76 .3%) in con trol wards (Tab le 1). 356 357 Proportion of AWaRe antibiotic classification 358 There was an increase in the proportion of Access antibiotics from 11.3% to 19 .1% in th e 359 intervention group and from 14.4 % to 19.7% in the con trol group at Hospital 1. 360 Conversely, the use of Access antibiotics fell at Hospital 2, declining from 21.9% to 361 14.9% in t he interven tion group and from 21.0% to 15. 6% in the co nt rol group. 362 Additionally, there was a rise in th e use of other antibiotic categories that were classified 363 as “Not recommended” or unclassified by the WH O in the A WaRe cla ssifications version 364 2021 (including cefoperazone/ beta-lacta mase inhibitor and ticarcillin/ beta-lacta mase 365 inhibitor), increasing from 16.9% t o 26.2% and from 16.9 % to 22. 4% , respectively (Table 366 1). We observed decreases in the use of certain subgroups, accompanied by increases in 367 others, depending on the available antibiotic agents at each hospital , with more 368 pronounced changes noted at Hospital 2 compared to Hospital 1 (Table 2S, 369 Supplemen tary Data) . In the in terv ention groups, there were decreases in the use of 370 second-generation cephalosporins (from 13.8% to 5 .2%) and glycopeptides (from 3.7% 371 to 2.3%) at Hospital 1, and in penic illin/beta-lacta mase inhibitors (from 27.9 % to 24.6 %), 372 fourth-gen eration cephalosporins (from 5.6% to 2 .5%) , and fluoroquinolones (from 373 22.6% to 16 .5%) at Hospital 2.374 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 14 Table 1. Study participants, antibiot ic use, mortality, costs of hospitaliz ation and antibio tic 375 non -susceptibility among hospital- acquired isolates before and af te r t he start of AMS 376 inte rv en tion 377 Study varia ble s Hos pita l 1 Hos pita l 2 Interven tion Control Interven tion Control Before After Before After Before After Before After Number of patie nt s All s tu dy wards 27,341 18,282 30,245 21,199 10,994 7,848 15,403 9,998 ICU pair 15,569 10,644 8,359 6,126 2,256 1,444 2,799 2,125 Surg ical pa ir 4,979 3,462 2,081 1,574 3,700 2,730 4,575 3,190 Internal pair 1 3,130 2,954 13,146 8,933 3,472 2,322 4,289 2,917 Internal pair 2 6,602 3,418 9,074 6,156 1,806 1,450 4,220 2,040 Antib ioti c use ( % ) All s tu dy wards 71.4 70.1 67.0 66.4 68.3 63.9 72.0 76.3 ICU pair 71.1 63.9 52.0 49.5 93.7 91.6 88.6 94.3 Surg ical pa ir 79.1 77.5 92.0 90.0 82.5 80.7 88.7 89.8 Internal pair 1 79.2 77.1 80.7 82.5 37.4 26.8 50.1 50.1 Internal pair 2 49.2 57.9 49.7 47.8 68.2 65.7 63.5 72.6 DOT1000 All s tu dy wards 804 875 907 934 551 490 361 427 ICU pair 957 968 964 945 979 861 941 1178 Surg ical pa ir 866 778 1361 1497 587 531 591 693 Internal pair 1 1100 1219 978 1016 215 190 117 122 Internal pair 2 491 571 635 604 561 527 472 581 AWaRe group (% DOT) for all study wards Acces s 11.3 19.1 14.4 19.7 21.9 14.0 21.0 15.6 Watch 88.1 80.0 84.6 78.9 60.1 58.7 58.6 57.8 Reserve 0.4 0.5 0.7 0.9 1.0 1.1 3.5 4.2 Other* 0.1 0.4 0.3 0.4 16.9 26.2 16.9 22.4 Mortal ity § All s tu dy wards 6.7 9.6 45.2 58.8 43.8 41.7 92.3 81.5 ICU pair 5.0 8.2 160.2 199.8 194.1 213.3 368.3 382.1 Surg ical pa ir 1.8 0.6 2.4 1.3 6.5 3.3 7.4 11.6 Internal pair 1 27.2 28.1 0.5 0.4 7.8 4.7 36.4 19.5 Internal pair 2 2.0 2.0 16.2 16.6 8.9 6.9 26.1 27.0 Hos pita liza tion co st # All s tu dy wards 183 (69-263) 192 (82-277) 97 (54-198) 106 (61-207) 491 (240-1027) 470 (249-922) 402 (193-835) 520 (265-979) ICU pair 233 (186-324) 237 (186-323) 206 (112-382) 206 (111-382) 1575 (829-2571) 1453 (771-2341) 1575 (829-2571) 1210 (599-2127) Surg ical pa ir 146 (81-310) 142 (78-279) 176 (82-387) 178 (82-410) 521 (275-1003) 507 (266-908) 434 (237-760) 560 (326-909) Internal pair 1 145 (82-250) 150 (80-264) 67 (41-106) 79 (48-121) 291 (165-494) 284 (166-454) 300 (187-529) 287 (180-521) Internal pair 2 49 (35-71) 58 (42-87) 116 (65-234) 114 (66-222) 508 (256-880) 519 (282-812) 255 (123-504) 357 (198-633) Antib ioti c non -s us cept ibi lity amo ng ho spi tal ac quire d iso late s ∏ . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 15 Study varia ble s Hos pita l 1 Hos pita l 2 Interven tion Control Interven tion Control Before After Before After Before After Before After E. coli Amin og lyc os ides 242/657 (37) 65/145 (45) 395/806 (49) 104/176 (59) 12/48 (25) 3/14 (21) 27/74 (36) 7/26 (27) Carbapenem s 78/657 (12) 23/145 (16) 216/806 (27) 54/176 (31) 14/144 (10) 0/14 (0) 5/74 (7) 2/26 (8) Ci prof lo xaci n 335/476 (70) 122/144 (85) 515/590 (87) 157/167 (94) 91/113 (81) 27/34 (79) 44/55 (80) 16/18 (89) 3 rd genenerati on cephal os pori ns 311/657 (47) 123/145 (85) 508/806 (63) 166/176 (94) 89/144 (62) 27/45 (60) 43/74 (58) 13/26 (50) Klebsiella spp . A m i no gl y co si d e s 33/141 (23) 26/49 (53) 106/455 (23) 73/96 (76) 150/249 (60) 31/67 (46) 51/119 (43) 21/41 (51) Carbapenem s 32/141 (23) 23/49 (47) 95/455 (21) 68/96 (71) 130/249 (52) 21/67 (31) 50/119 (42) 19/41 (46) Ci prof lo xaci n 46/105 (44) 34/45 (76) 154/268 57 81/96 (84) 166/205 (81) 35/56 (62) 69/91 (76) 23/30 (77) 3 rd genenerati on cephal os pori ns 60/141 (43) 43/49 (88) 164/455 (36) 86/96 (90) 174/249 (70) 36/67 (54) 70/119 (59) 23/41 (56) P. aeruginosa Amin og lyc os ides 15/50 (30) 11/20 (55) 69/150 (46) 35/41 (85) 206/348 (59) 44/87 (51) 45/92 (49) 13/22 (59) Carbapenem s 21/50 (42) 5/20 (25) 74/150 (49) 28/41 (68) 174/348 (50) 42/87 (48) 49/92 (53) 12/22 (55) Ci prof lo xaci n 22/42 (52) 13/20 (65) 63/109 (58) 28/40 (70) 193/293 (66) 46/75 (61) 47/83 (57) 12/19 (63) Ceftaz idi me 12/43 (28) 8/20 (40) 53/108 (49) 25/40 (62) 91/316 (29) 14/61 (23) 44/86 (51) 4/15 (27) Piperaci ll in- tazobactam 11/43 (26) 6/20 (30) 38/110 (35) 20/41 (49) 30/309 (10) 7/77 (9) 6/83 (7) 1/20 (5) Acinetobacter spp. Amin og lyc os ides 120/140 (86) 50/60 (83) 172/310 (55) 65/70 (93) 244/310 (79) 88/103 (85) 153/198 (77) 47/60 (78) Carbapenem s 78/140 (56) 39/60 (65) 153/310 (49) 49/70 (70) 251/310 (81) 93/103 (90) 165/198 (83) 52/60 (87) Ci prof lo xaci n 118/124 (95) 57/58 (98) 162/179 (91) 60/67 (90) 245/282 (87) 90/97 (93) 160/177 (90) 51/56 (91) Ceftaz idi me 112/122 (92) 53/58 (91) 153/183 (84) 63/69 (91) 267/289 (92) 84/87 (97) 160/179 (89) 50/54 (93) Piperaci ll in- tazobactam 102/124 (82) 44/60 (73) 151/183 (83) 52/70 (74) 219/254 (86) 89/93 (96) 142/156 (91) 51/54 (94) S. aureus MRSA 74/85 (87) 50/57 (88) 210/278 (76) 88/112 (79) 71/93 (76) 19/25 (76) 60/79 (76) 11/12 (92) 378 * Antimic robials classified und er this g roup followin g WH O AW AR E’s 2021 v ersion in clude: cefo per azon e and bet a-lact am ase 379 inhibitor, ticar cillin and beta-l act am ase inhibitor. 380 § In-hospital mortality w as c alcul ated as those with o utcome re cord ed as “deat h” and “going home to di e” out of the p atients 381 admitted to the cor respondin g study wa rds ( exp ress ed as p er 1000 admissions). 382 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 16 # Costs are presen ted in US Dollar media n (inte rqu a rtile r ange) per da y of hospit alization, c alcul ated fo r individual pa tients, and with 383 costs in 2019 an d 2020 adjust ed to th e costs of 202 1 using GDP defl ation r ates. 384 ∏ Antibiotic non-susc eptibility d ata ar e pr esented as the pro portion of n umbe r of non-susc eptible isolat es over all isolates 385 (per cent age ). 386 387 388 Based on the frequency of observa tions for each outcom e, we deter mined to apply a 389 weekly interval for antibiotic use, mortality, and hospitalisation costs , while adopting a 390 monthly int erval for antibiotic non -susceptibility, given that the nu m ber of antibiotic 391 non-susceptible isolates is relatively low. 392 393 Days of antibiotic therapy per 1000 patient-days (DOT1000) 394 From the su mmary report, t he ove rall DOT1000 increased in the post -interven tion 395 period for b oth the interven tion a nd control groups in Hospital 1. In contrast, Hospital 2 396 experienced a decline in the post-i nterven tion period for the interven tion group, while 397 the control group showed an increase (Table 1; for comprehensive details of numerators 398 and denominators, refer to Table 2 S, S upplemen tary Data) . 399 400 ITS/CITS r esults for D OT1000 in the inter ve ntio n group 401 ITS/CIT S models provided evidenc e of consistent changes in D OT100 0 for Hospital 1 402 regarding overall antibiotic use an d certain subgroups. Specifically, t he interven tion was 403 associated with an immediate (i.e . l evel) reduction in overall antibiotic use by 95.9 (CITS 404 95% CI: [10.9 , 180.8] ; ITS : 93. 3 [0.7, 186.0]) and a reduction in the D OT 1000 level of 405 glycopeptide antibacterials by 52.7 (CITS [38.0, 67 .5]; IT S: 60 .2 [40.6 , 7 9.8]). In con trast, 406 the D OT1000 level of beta-lac tam ase resistant penicillins increased by 7.3 (CITS [0.2, 407 14.3]; IT S: 8 .4 [2.7 , 14.0]) . The slope of the DOT1000 for second-gen eration 408 cephalosporins indicates a long-term effect , with an average decreas e of 1.3 (CITS [0.5 , 409 2.0]; IT S: 1 .9 [1.1 , 2.8]) for each additional week, reflecting a declining long-term tre nd 410 over time. Conversely, th e slope of the DOT1 000 for penicillin/beta-la ctamase inhibitors 411 increased by 1.9 (CITS [0.9, 2.9] ; IT S: 1. 6 [1.0, 2 .2]). The slopes of aminoglycosides, 412 polymyxins, and imidazole derivatives increased in ITS models by 0.4 (ITS [0.1 , 0.7]), 0 .1 413 (ITS [0.03 , 0.2]), a nd 0.5 (ITS 95 % CI: [0.1 , 0.9]), respectively , while the slope of 414 glycopeptide antibacterials DOT10 00 decreased in the CITS model by 0.6 (CITS [0.1 , 1.0]) 415 (Figure 2; Figure 4S , Table 11S, Su pplementary Data) . 416 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 17 417 For the ward pair subg roup analysi s, the interve ntion was associated with a decrease in 418 overall antibiotic DOT1000 levels in the trau matology ward by 213.0 (ITS [74.8 , 351.3]) , 419 an increase in DOT1000 levels in t he general medicine ward by 127.9 (ITS [6.2, 249 .6]), 420 and a decreasing slope in DOT100 0 for the respiratory/musculoskeletal system ward by 421 3.1 (CITS [1.1 , 5.0]) (Figure 5S , Table 11S, S upplemen tary Data). 422 423 In Hospital 2, th e results from ITS and CITS were inconsistent , with some evidence of 424 changes observed only in s pecific antibiotic subgroups. Notably, ther e was strong 425 evidence indicating tha t the slope of aminoglycosides DOT1000 decreased by 0.6 (CITS 426 [0.4, 0. 8]), whereas the D OT1000 f or penicillin/beta-lacta mase inhibit ors exhibited a 427 significant reduction only in the CI TS model by 3.7 (CITS [1.7 , 5.6]), a nd the slope of 428 carbapenem D OT1000 showed a significant increasing effec t only in the ITS model by 429 0.3 (ITS [0. 1, 0. 5]). In t he ward pair analysis, only the high-quality general medicine ward 430 demonstrated an increase in D OT1 000 by 87.6 (ITS [16.0, 159 .2]) (Figu re 2; Figure 5S, 431 Table 11S, S upplemen tary Data). 432 Antibiotic non-susceptibility 433 Descriptive statistics, enco mpassing the frequency and percentage of non-susceptibility 434 to antibiotics among hospital-acq uired isolates, as well a s the intervention and control 435 ITS/CIT S models, are presented for the interven tion group (Table 1; Figure 3&4). A 436 comprehensive report of the descriptive statistics relating to other an tibiotic non-437 susceptibility and the ITS/CITS mo dels conducted on the control group is provi ded in 438 Supplemen tary Data . 439 E. coli 440 Among hospital-acquired E. coli is olates, the percen tages of non-sus ceptible isolates 441 were observed to be higher during the post-interve ntion period in Hospital 1, whereas 442 the opposite trend was noted in Hospital 2 (Table 1; Figure 3; Table 1 0S, Supplemen tary 443 Data). However, in t he CITS model s, the AMS in terventio n resulted in a slope re duction 444 for aminoglycoside non-susceptibility (odds ratio [OR] 95%CI: 0.87 [0 . 78, 0.97]) in 445 Hospital 1. In Hospital 2, the level decreased for aminoglycosides (0.81 [0.35, 1.0 0]) and 446 third-generation cephalosporins non-susceptibility (0.42 [ 0.15, 1 .00]) , but increased for 447 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 18 carbapenems non-suscep tibility (2.21 [1.00, 8.3 1]; no evidence of cha nges in the slopes 448 (Figure 2; Table 12S, Suppleme ntar y Data). 449 Klebsiella spp. 450 The percentages of Klebsiella spp. non-susceptible to antibiotics ex hi bited patterns 451 similar to those observed in E. coli, characterised by a higher percentage of non-452 susceptible isolates in the post-intervention phase for Hospital 1, an d a lower 453 percentage for Hospital 2 (see Table 1; Figure 3; Table 10S, Suppleme ntary Data). 454 Analysis using ITS/CITS models showed consistent changes only in the levels of non-455 susceptibility, with a decrease in the level of non-susceptibility to carbapenems (CITS 456 0.77 [0.33 , 1.00]) and ciprofloxacin (0.50 [0.17 , 1.00]) in Hospital 1. Co nversely, in 457 Hospital 2, the in terventio n was as sociated with a decreased level of non-susceptibility 458 to carbapenems in the IT S model ( ITS 0.27[0 .10 , 0.71]) but increased in the CITS model 459 (1.06 [1.00 , 1.95]), and increased in both models for ciprofloxacin (ITS 2.01 [ 1.00, 4 .18], 460 CITS 1.36 [1.0 0, 2. 50]) (Figure 2; Ta ble 12S, Supple ment ary Data). 461 P. aeruginosa 462 Similar trends were observed in the percentages of non-susceptible P. ae ruginosa 463 isolates across b oth hospitals (Tabl e 1; Figure 3; Table 10S , Supplem e ntary Data). I n 464 Hospital 1, the results from t he ITS /CITS analyses indicated an increase in the slope of 465 non-susceptibility to carbapenems in the CITS model (1.11 [1. 00, 1 .22 ]), and a decrease 466 for aminoglycosides but in the ITS model only (0.98 [0.80, 1 .00]). Con versely, Hospital 2 467 exhibited mixed results between b oth models in the slopes of non-susceptibility to 468 ceftazidime (CITS 1 .03 [1.00 , 1.19] ; ITS 0.84 [0 .58, 1 .13]), a minoglycosides (CITS 1.07 469 [1.00, 1 .20]; IT S: 0 .73 [0.58 , 0.88]) , and ciprofloxacin (CITS 1.45 [1.18 , 1. 77]; ITS : 0.63 [0 .49, 470 0.78]) (Figure 2; Table 12S , Supple ment ary Data). 471 Acinetobacter spp. 472 The percentages of non-susceptibi lity showed varying patterns in the descriptive 473 analysis. Notably, the non-suscep tibility percentages were heterogeneous in Hospital 1, 474 whereas all antibiotic non-suscepti bility percentages were elevated in Hospital 2 (Table 475 1; Figure 3; Table 10S , Supple ment ary Data). The CITS models indicated an increase in 476 the level (11.93 [1.00 , 94. 25] and slope of non-susceptibility to amino glycosides (1.07 477 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 19 [1.00, 1 .27]), and in the level of no n-susceptibility to piperacillin tazobactam (1. 23 [1.00 , 478 2.52]) in Hospital 1. In con trast, a r eduction was noted in the slope of non-susceptibility 479 to carbapenems (CITS 0. 96 [0.88 , 1.00]), ciprofloxacin (CITS 0 .93 [0.85 , 1.00]) and 480 piperacillin tazobactam (CITS 0.94 [0.78, 1 .00]) in Hospital 2 (Figure 2; Table 12S, 481 Supplemen tary Data) . 482 MRSA 483 The percentages of MRSA rem aine d stable during the post-interventi on period in both 484 hospitals (Table 1; Figure 3; Table 10S&1 2S , Supplem entary Dat a), ex cept for the 485 decreasing slope only observed in the IT S model of Hospital 2 (0.44 [0.26, 1 .00]) (Figure 486 2; Table 12S, Suppleme ntary Data) . 487 488 In-hospital mortality and costs 489 There was a non-significant decrea se in the level of mortality with inc reasing slope in 490 Hospital 1 intervention wards overall, while increasing levels and s lop es were observed 491 in Hospital 2 (Table 2S; Table 13S, Supplemen tary Data) . The IT S/CITS models for 492 mortality show inconsistent results for the specific wards of both hospitals (Figure 4; 493 Table 13S, S upplemen tary Data), w ith insignificant changes reported in all CITS models. 494 495 Considering the costs for monthly activities of the AMS team (coordi nator/pharmacists) 496 and on-site training as regular activities as part of the AMS programme at each hospital , 497 the one-year costs for AMS imple ment ation were approximately 2,8 09 USD for Hospital 498 1 and 2,283 USD for Hospital 2 (Ta ble 1S, Supple mentary Da ta). 499 500 Regarding hospitalization costs, in both ITS and CITS models, we observed consistent 501 decreasing levels and s lopes overa ll and across most study wards in both hospitals. 502 Particularly, t he cost of hospitalisation in the interven tion surgical ward decreased in 503 slope by 5.5 (CITS [3.1, 8.0]) US Dol lar in Hospital 1 (Figure 4; Table 14 S, S upplemen tary 504 Data). 505 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 20

506 We exa mined the i mpact of a PA F intervention on a ntibiotic use and antibiotic non-507 susceptibility for multiple pathogen-drug pairs among common hos pital-acquired 508 bacterial isolates. This approach acknowledged the potential “squeezing the balloon” 509 phenomeno n in AMS programmes 18 , wherein limiting th e use of specific antibiotics may 510 lead to counteracti ng and uninten ded changes in the use of other antibiotics and drug 511 resistance mecha nisms. Our findin gs support this as sumption, with o bserved decreases 512 in some antibiotic groups but increases in others, reflecting the co m plexity of antibiotic 513 use and resistance in provincial ho spitals with high antibiotic consum ption. 514 515 The observed impact models varied between the two hospitals, refle cting the con tex t-516 specific nature of the imple men tat ion. In Hospital 1, IT S and CITS mo dels provi de 517 evidence tha t overall antibiotic consumption im mediately decreased in the interven tion 518 group with consistent immediat e and long-term impac ts on antibioti c non-susceptibility 519 for hospital-acquired E. coli. In contrast, th e impact of AM S on antibi otic use was more 520 limited in Hospital 2, with some ev idence of positive long-term effec t s on non-521 susceptibility of Acinetobacte r spp. These heterogeneous findings and the lack of 522 significant interven tion effec ts in various subgroups or outcome indicators were likely 523 influenced by confounders, such a s differences in communi ty antibiotic pressure, patient 524 characteristics, baseline resistance profiles, resource availability (number of clinical 525 pharmacists dedicated to the PAF activity), hospital cult ure and staff engagemen t, and 526 the behaviour and collaboration of doctors and pharmacists towards the AMS 527 intervention , which could have i m pacted both the int ervention a nd the antibiotic use 528 outcom es. Furt hermore , the ef fect s on antibiotic consumption likely contributed to th e 529 inconsistent impac ts on antibiotic non-susceptibility. 530 531 One of th e inclusion criteria for selecting the wards was the willingness of the ward head 532 to participate. This may have helpe d increase the collaboration of doctors in the 533 intervention wards and their comp liance with the reco mme ndations of the AMS te am in 534 the PA F activity , and thus possibly limits the generalizability of the study. However, t he 535 PAF ac tivity reached only a small number of patients during the impl emen tation period 536 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 21 (1,890/45,6 23 or 4.1% in Hospital 1, and 1,628/18 ,842 or 8.6% in Hospital 2). This 537 suggests the change in practice following the PAF activi ty is likely to be from the 538 systems level rather than th e individual patient level, particularly the i mpact on reducing 539 overall antibiotic use in Hospital 1. The presence of clinical pharmacists on the clinical 540 wards to regularl y audit antibiotic prescriptions may have increased the compliance of 541 doctors to prescription guidelines in general and in the surgical ward (traumatology) in 542 particular, poten tially providing a generalizable evidence for the impact of this non-543 restrictive AMS interve ntion for ot her settings. 544 545 For mortality, t he results of our CITS models indicate tha t AMS imple ment ation at t he 546 hospital-wide level could improve antibiotic use without ca using negative consequences 547 on patient out comes. T he increasing trends in mortality were found in ITS models for 548 both interven tion and control ICUs, but not in th e CITS models, highli ghting the 549 importance of using a control 19 when evaluati ng impact of interve nt ions. Further 550 examina tion revealed increased post-interven tion mort ality for most diagnoses among 551 ICU patients in both hospitals, part icularly for di seases of the respirat ory sy stem and 552 abnormal clinical and laboratory fi ndings (Table 15S, Supplemen tary Data). Non etheless, 553 further study is needed to fully inv estigate the impac t of AMS in ICU patients, 554 considering the differential effects that might happen to patien ts of different diagnoses. 555 The limited nu mber of data points in our current datasets does not provide sufficient 556 power to identify the changes in specific clinical diagnoses over time in ICUs. 557 558 Generally, AMS programm es in Vietnam ese hospitals must follow national guidelines, 559 which require forming an AMS co mmit tee , assigning roles, developi ng hospital-specific 560 policies, and implementing restrict ed antibiotic lists 10 20 . Most hospitals responded to 561 the nation al guidelines by quickl y convening com mit tees, while specific actions and 562 interventions for moni toring and improving antibiotic use and resistance remained 563 limited due to poor leadership co mmit men t, lack of dedicated staff , and weak IT 564 capacity and lab resources 8 . Many hospitals established pre-authorization systems for 565 restricted antibiotics to ensure co mpliance with natio nal guidelines, often tied to social 566 health insurance rei mbursemen t, p articularly for expensive and potentially overused 567 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 22 medical services 21 . In this cont ext , clinical pharmacist-led audits of an tibiotic 568 prescriptions, with constructive fee dback to doctors, offer a proactive intervention to 569 enhance collaboration a mong heal thcare professionals. Our feasibility study showed 570 that t his approach could be integr ated into routine clinical pharma cy work 9 . The 571 intervention involved reviewing ac tive antibiotic prescriptions during pharmacist visits, 572 providing recommendations for inappropriate practices, and assessin g changes in total 573 antibiotic use and specific groups. We measured cha nges in both tot al antibiotic use 574 and specific groups to as sess the i nterven tion’s impact. I ncreased collaboration among 575 prescribing doctors, clinical pharm acists, microbiologists, and infectio us disease 576 specialists helped optimise antibio tic use for individual patients 18 . 577 578 Literat ure supports the causal relationships between antibiotic use in hospitals and 579 resistance prevalence among hosp ital-acquired isolates antibiotics 22 . Mixed results from 580 ITS and CITS models in our study s uggest the presence of history bia s and other events 581 impacting the co ntrol group 19 . Pre vious ITS studies, often lacking co ntrol groups, 582 reported mixed results as shown i n a systematic review of studies by 2018 5 as well as in 583 more recent studies in the U S 23-26 , Spain 27-29 , Germa ny 30 , Japan 31 , Brazil 32 , Korea 33 , and 584 China 34 . To our knowledge, only one observational study in a 627-bed hospital in 585 Canada used communi ty-acquired isolates as a control time series 6 . T his study 586 demonstrated AM S impact in redu cing the incidence of hospital-acq uired multidrug-587 resistant organisms by 12.6%. How ever, this approach could not acco unt for the 588 concurrent in terven tions, such as I PC or other programmes within hospital settings, that 589 could affect hospital-acquired resistance. 590 591 The significant strength of our stu dy is the inclusion of a control gro up, allowing to 592 account for conc urrent eve nts, suc h as the COVI D-19 pande mic and I PC measures. In 593 particular, AMS ac tivities and staff atten tion to optimal an tibiotic prescribing may have 594 been negatively affected in t he last few weeks of the intervention per iod in Hospital 1 595 due to the early effects of the four th wave of the CO VID -19 pandemi c in Vietna m in 596 April – May 2021 12 . We assumed such non-in terven tion even ts had broadly similar 597 effects on both th e interven tion an d control groups. Along with the s trengths of 598 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 23 employing an implement ation research design as reported previousl y 9 , we 599 demonstrated th e practicality of u sing a participatory action researc h approach to 600 evaluate a healt hcare interve ntion that could not be et hically assesse d through 601 traditional randomised controlled trials 35 . We also showed that beha viour change 602 interventions , such as prospective audit and feedback, can be safely integrated into 603 routine clinical practice withou t ca using negative consequences on patient ou tcomes in 604 resource-limited settings like Vietn am, u nder existing national leadership and guidance. 605 606 Limita tions include the relatively s hort timefra me for the pre- and post-interven tion 607 periods, which restricted our abilit y to fully capture seasonal and hist orical trends in 608 antibiotic use, especially when ther e were frequent shortages and stock-outs of 609 antibiotics, as well as changes in d rug bidding cycles and health insurance policies that 610 could affect prescribing practices i n each hospital. There was an incre asing trend in the 611 use of fixed-dose combinations in Hospital 2 which could raise concerns about 612 substitution effec ts as a result of the AMS interve ntion . Fu ture studie s with longer pre- 613 and post-interven tion periods will help investigate in-depth t hese specific changes over 614 time to inform the design of interventions. Despite effor ts to extrac t data for a longer 615 pre-interven tion period, we were unable to use data before 2019 for analysing antibiotic 616 and clinical outcomes in the co ntr ol and intervention groups due to changes in the H IS . 617 However, since t hese system-level changes are likely to have had simi lar effects on both 618 the interve ntion and con trol group s, we expect the esti mat es under the controlled 619 analyses (CITS models) to be relati vely robust to such system-level changes. Longer ti me 620 series for antibiotic non-susceptibility data were available from labor atory systems, 621 though missing patient identifiers impacted model performanc e. Nev ertheless, our 622 analysis used more data points than the mini mu m suggested for interrupted time-series 623 from a simulation-based power calculation 36 . Another limit ation , intri nsic to our 624 implemen tation study design, is that there can be a “spillover effect”, i.e. an uni ntended 625 impact of the in terven tion on the prescribing p ractices of the contro l wards. This may 626 have diluted our estimated impac t of the interven tion on antibiotic u se in the 627 intervention group in the CIT S mo dels. 628 629 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 24 In addition, microbiology data quality (which could be affected by wa rd-specific 630 variations in specimen collection p ractices and testing frequency), t h e absence of 631 molecular data on resistance me ch anisms and potential misclassificat ion of hospital-632 acquired infections limits our ability to link antibiotic use changes to resistance 633 dynamics. Finally, we were not able to collect more granular data for cost analyses as 634 only the total cost of hospitalization was available in the cost data extracted from HI S. I n 635 order to collect more detailed costs, we are now conducting an additional study on a 636 specific patient population and will use these data to inform our upc oming cost analyses 637 for the AMS programme . 638 639 Our study provides initial evidence on the complex eff ects of AMS int erventions on 640 antibiotic use and resistance, and on how quickly changes in antibiotic use lead to 641 reverse resistance in a particular or ganism 37 . These findings could be generalised to 642 similar settings with limited resources in Asia. A combination of strat egies, including 643 drug discovery, res istance monitor ing and novel interventions, is nec essary to respond 644 to current resistance pheno types and to anticipate th e evolution of a ntibiotic resistance 645 in hospital settings. This is particul arly challenging for hospitals in lo w- and middle-646 income coun tries, which have li mit ed resources, where infectious diseases are prevalent, 647 antibiotic use is high, and environment al reservoirs accelerate resistance spread. 648 Surveillance data on t he mec hanis ms of emergence and trans mission of antibiotic 649 resistance within and between hos pital reservoirs, along w ith extend ed evaluation 650 periods, are needed to further clarify the impact of AMS in terven tions and antibiotic use, 651 and to inform more effec tive, large -scale control and policy measures across hospitals 38 . 652 653 In conclusion, this study highlights the interdependen t changes in an tibiotic use and 654 antibiotic resistance driven by AMS programmes, e mploying pharmacist-led prospective 655 review of antibiotic prescriptions a nd feedback to doctors in provinci al-level hospitals in 656 Vietna m . Our findings confirm th at developing and strengthening the surveillance of 657 antibiotic resistance, alongside AMS impleme nta tion and IPC measur es, is essential to 658 monitor and respond to resistance dynamics in hospitals effectively. These efforts will 659 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 25 help to ensure that in terventions k eep pace with the rapid evolution of antibiotic 660 resistance. 661 662 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 26

663 We acknowledge the support and collaboration of the Medical Servic es Administration 664 of the Viet Na m Ministry of Healt h and the World Health Organizatio n Office in Vie t 665 Nam during the planning of this study. 666 667 Fu nding 668 Pfizer Independent Gra nts for Lear ning & Change (IGLC) provided pr oject funding, 669 administered through The Join t Commission; H. V.T .L was supported by the National 670 Institu te for Healt h Research (NI H R) (using the UK’s Official Development Assistance 671 (ODA) Fu nding) and Wellcome (Gr ant Referen ce Number : 216367/Z/ 19/Z) under the 672 NIHR -Wellco me P artnership for Global Health Research . HCT acknow ledges funding 673 from the MRC Centre for Global In fectious Disease Analysis (reference MR/R015600/1), 674 jointly funded by the UK Medical Research Council (MRC) and the UK Foreign, 675 Commonwealt h & Develop ment O ffice (FCDO) , under the MRC/ FCD O Concordat 676 agreement and is also part of the EDCTP2 program supported by the European Union. 677 The views expressed are those of t he authors and not necessarily tho se of Wellcome, the 678 NIHR or the Depar tme nt of Heal th and Social Care. 679 680 Competi ng inte rests 681 None declared. 682 683 Cont ri but ors 684 VTL H , HRvD , EDA and DJA obtain e d funding and contributed to all aspects of the study 685 design. VTL H and HRvD had over all responsibi lity for the study. LMQ, NTT H , V HV , CMD, 686 VTH DE, PNT supervised and coord inated the runni ng of the study with support from 687 EDA, NTCT, TA Q , LNM H and NH K. VTL H , L QT and VTTD an alysed the data with 688 supervision and input from MC, BC, TK, and HCT . VT LH was responsible for the drafting 689 of the man uscript. All authors gave approval for the final version of t he manuscript . 690 691 Patient co nsent fo r p ublic atio n 692 Not required for this research. 693 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 27 Ethics app roval 694 The study protocol was app roved by the Oxford University Tropical Research Ethics 695 Commit tee (O xTREC Referen ce 52 6-19), and the E thics Commi ttee of National Hospital 696 of Tropical Diseases (08/HĐĐĐ-N ĐT 31 May 2019). The conduct of t his study conformed 697 to the principles embodied in the Declaration of Helsinki. 698 699 ORCID IDs 700 Le Q uynh Trang 0009-0001 -9758- 2873 701 Vu Hai Vinh 0000-00 01-6130 -7864 702 Elizabeth Dodds Ashley 0000-0002-4213-6104 703 Deverick J. Anderson 0000-0001 -6 882-5496 704 Ben S. Cooper 0000-0002-9445 -72 17 705 Marc Choisy 0000-0002-5187-639 0 706 H. Rogier van Doorn 0000-0002 -9 807-1821 707 Vu Thi La n Huong 0000 -0002-957 9-5576 708 709 Data availability 710 De-identified data may be obtaine d from the hospitals participating in this study when a 711 data sharing agreement is in place. 712 713 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 28 List of figures: 714 Figure 1. Pa tien t locatio n befo re an d after th e AMS implementa tion sta rt at Hospital 1 (Jan 715 2019 – May 2021) and Hospital 2 (J an 2019 – Jul 2021). There wer e 4 pairs in each 716 Hospital: 1) ICU pair: surgical ICU versus inter nal ICU (both hospitals); 2) Surgical pair: 717 traumatology ve rsus nephro-ur olog y (Hospital 1) and traumatology versus 718 gastroe nt erology (Hospital 2); 3) Internal pair 1: respirat ory ve rsus paediatrics (Hosp ital 1); 719 gen eral inte rnal medicine (high-quality services) versus infectious diseases (Hospital 2); 720 and 4) Inte rnal pair 2: infec tious dis eases versus gen eral int ernal medic ine (Hospital 1); 721 gen eral inte rnal medicine (n ormal services) ve rsus oncology (Hospital 2). n: numbe r of 722 patient . 723 724 725 Figure 2. Results of ITS and CITS models for antibio tic use and antibiot ic non-susceptibility 726 among the hospital-acquired common pathoge ns identifi ed from rou tine microbiology in 727 the int er ve ntio n group at two hospitals. 1 Odds ratio betwe en non -susceptibility vs 728 susceptibility. Bold val ues: statistically s ignificant estimate ; Gre en shades: decr easing 729 tre nds; Red shades: incr easing tre n ds; Dark shades: consiste nt significant r esults for 730 ITS/CITS models. Pip-tazobactam: Piperacillin-tazobac tam. 731 732 Figure 3. P ropo rtio n of antibio tic n on-susceptibility for main pathoge n -drug pairs in 733 inte rv en tion and co ntr ol groups bef ore and af ter the start o f AMS int er ve ntio n at two 734 hospitals; H1: Hospital 1; H2: Hosp i tal 2; I: In te rv en tion ; C: C on trol ; MR SA: Methicillin-735 resistant Staphylococcus aureus; Those columns with * present r esults for hospital acquired 736 isolates, the r emaining columns are for all isolates in the correspo ndin g group. 737 738 739 Figure 4. Results of ITS and CITS models for in-hospital mortality and cost of 740 hospitalization in the int erv en tio n group at two hospitals. Bold values: statistically 741 significant estimate ; Gre e n shades: decreasing tr e nds; Red shades: incr easing tre nds. 742 743 744 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 29 Refere nces 745 1. Ahmad N, Joji RM, Shahid M. Evolution and implementation of One Health to control the 746 dissemination of antibiotic-resistant bacteria and resistance genes: A review. Front Cell 747 Infect Microbiol 2022;12:1065796. doi: 10.3389/fcimb.2022.1065796 [published Online 748 First: 2023/02/03] 749 2. Monnier AA, Schouten J, Le Maréchal M, et al. Quality indicators for responsible antibiotic 750 use in the inpatient setting: a systematic review followed by an international 751 multidisciplinary consensus procedure. J Antimicrob Chemother 2018;73(suppl_6):vi30-752 vi39. doi: 10.1093/jac/dky116 753 3. Zay Ya K, Win PTN, Bielicki J, et al. Association Between Antimicrobial Stewardship Programs 754 and Antibiotic Use Globally: A Systematic Review and Meta-Analysis. JAMA Netw Open 755 2023;6(2):e2253806. doi: 10.1001/jamanetworkopen.2022.53806 [published Online 756 First: 20230201] 757 4. Davey P, Marwick CA, Scott CL, et al. Interventions to improve antibiotic prescribing practices 758 for hospitals inpatents. Cochrane Database Syst Rev 2017;2: CD003543 doi: 759 10.1002/14651858.CD003543.pub4. 760 5. Bertollo LG, Lutkemeyer DS, Levin AS. Are antimicrobial stewardship programs effective 761 strategies for preventing antibiotic resistance? A systematic review. Am J Infect Control 762 2018;46(7):824-36. doi: 10.1016/j.ajic.2018.01.002 [published Online First: 2018/02/24] 763 6. Peragine C, Walker SAN, Simor A, et al. Impact of a Comprehensive Antimicrobial 764 Stewardship Program on Institutional Burden of Antimicrobial Resistance: A 14-Year 765 Controlled Interrupted Time-series Study. Clinical Infectious Diseases 2020;71(11):2897-766 904. doi: 10.1093/cid/ciz1183 767 7. Wertheim H, Chandna A, Phu V, et al. Providing impetus, tools and guidance to strengthen 768 national capacity for antimicrobial stewardship in Viet Nam. PLoS Med 769 2013;10(5):e1001429. doi: 10.1371/journal.pmed.1001429 770 8. Huong VTL, Ngan TTD, Thao HP, et al. Improving antimicrobial use through antimicrobial 771 stewardship in a lower-middle income setting: a mixed-methods study in a network of 772 acute-care hospitals in Viet Nam. J Glob Antimicrob Resist 2021;27:212-21. doi: 773 10.1016/j.jgar.2021.09.006 [published Online First: 2021/10/04] 774 9. Huong VTL, Ngan TTD, Thao HP, et al. Assessing feasibility of establishing antimicrobial 775 stewardship programmes in two provincial-level hospitals in Vietnam: an 776 implementation research study. BMJ Open 2021;11(10):e053343. doi: 777 10.1136/bmjopen-2021-053343 [published Online First: 2021/10/03] 778 10. Viet Nam Ministry of Health. Quy/i(Qt đ/i(Qnh v/i(Q vi/i(Qc ban hành tài li/i(Qu "H/i(Q/i(Qng d/i(Qn th/i(Qc hi/i(Qn 779 qu/i(Qn lý s/i(Q d/i(Qng kháng sinh trong b/i(Qnh vi/i(Qn" [Decision on the issuance of the document 780 "Guideline on implementing antimicrobial stewardship in hospitals"]. Hanoi: Viet Nam 781 Ministry of Health, 2016. 782 11. Lien LTQ, Chuc NTK, Hoa NQ, et al. Knowledge and self-reported practices of infection 783 control among various occupational groups in a rural and an urban hospital in Vietnam. 784 Scientific Reports 2018;8(1):5119. doi: 10.1038/s41598-018-23462-8 785 12. Nguyen T P, Wong ZS, Wang L, et al. Rapid impact assessments of COVID-19 control 786 measures against the Delta variant and short-term projections of new confirmed cases 787 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 30 in Vietnam. J Glob Health 2021;11:03118. doi: 10.7189/jogh.11.03118 [published Online 788 First: 2022/01/07] 789 13. Ngan TTD, Quan TA, Quang LM, et al. Review of antibiotic prescriptions as part of 790 antimicrobial stewardship programmes: results from a pilot implementation at two 791 provincial-level hospitals in Viet Nam. JAC-Antimicrobial Resistance 2023;5(1):dlac144. 792 doi: 10.1093/jacamr/dlac144 793 14. Sharland M, Zanichelli V, Ombajo LA, et al. The WHO essential medicines list AWaRe book: 794 from a list to a quality improvement system. Clinical Microbiology and Infection 795 2022;28(12):1533-35. doi: https://doi.org/10.1016/j.cmi.2022.08.009 796 15. Berends MS, Luz CF, Friedrich AW, et al. AMR: An R Package for Working with Antimicrobial 797 Resistance Data. Journal of Statistical Software 2022;104(3):1 - 31. doi: 798 10.18637/jss.v104.i03 799 16. Turner HC, Lauer JA, Tran BX, et al. Adjusting for Inflation and Currency Changes Within 800 Health Economic Studies. Value in Health 2019;22(9):1026-32. doi: 801 10.1016/j.jval.2019.03.021 802 17. Des Jarlais DC, Lyles C, Crepaz N. Improving the reporting quality of nonrandomized 803 evaluations of behavioral and public health interventions: the TREND statement. Am J 804 Public Health 2004;94(3):361-6. doi: 10.2105/ajph.94.3.361 [published Online First: 805 2004/03/05] 806 18. Burke JP. Antibiotic Resistance—Squeezing the Balloon? JAMA 1998;280(14):1270-71. doi: 807 10.1001/jama.280.14.1270 808 19. Lopez Bernal J, Cummins S, Gasparrini A. The use of controls in interrupted time series 809 studies of public health interventions. International Journal of Epidemiology 810 2018;47(6):2082-93. doi: 10.1093/ije/dyy135 811 20. Viet Nam Ministry of Health. Quy/i(Qt đ/i(Qnh v/i(Q vi/i(Qc ban hành tài li/i(Qu "H/i(Q/i(Qng d/i(Qn th/i(Qc hi/i(Qn 812 qu/i(Qn lý s/i(Q d/i(Qng kháng sinh trong b/i(Qnh vi/i(Qn" [Decision on the issuance of the document 813 "Guideline on implementing antimicrobial stewardship in hospitals"] Numbered 814 5631/QD-BYT. In: Health Mo, ed. Ha Noi: Ministry of Health, 2020. 815 21. Ha NT, Anh NQ, Van Toan P, et al. Health Insurance Reimbursement to Hosptials in Vietnam: 816 Policy Implementation Results and Challenges. Health Serv Insights 817 2021;14:11786329211010126. doi: 10.1177/11786329211010126 [published Online 818 First: 2021/04/30] 819 22. McGowan JE, Jr. Antimicrobial resistance in hospital organisms and its relation to antibiotic 820 use. Rev Infect Dis 1983;5(6):1033-48. doi: 10.1093/clinids/5.6.1033 [published Online 821 First: 1983/11/01] 822 23. Zeana C, Palmieri FE, Gupta V, et al. Association between fluoroquinolone utilization rates 823 and susceptibilities of gram-negative bacilli: Results from an 8-year intervention by an 824 antibiotic stewardship program in an inner-city United States hospital. Sci Prog 825 2021;104(2):368504211011876. doi: 10.1177/00368504211011876 [published Online 826 First: 2021/04/29] 827 24. Belsky B, Minson Q. Short- and long-term impact of a multifaceted approach targeting 828 fluoroquinolone use in a community hospital: an interrupted time-series analysis. Int J 829 Clin Pharm 2022;44(3):741-48. doi: 10.1007/s11096-022-01405-8 [published Online 830 First: 2022/04/23] 831 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 31 25. Hecker MT, Son AH, Murphy NN, et al. Impact of syndrome-specific antimicrobial 832 stewardship interventions on use of and resistance to fluoroquinolones: An interrupted 833 time series analysis. Am J Infect Control 2019;47(8):869-75. doi: 834 10.1016/j.ajic.2019.01.026 [published Online First: 2019/03/10] 835 26. Kinnear CL, Patel TS, Young CL, et al. Impact of an Antimicrobial Stewardship Intervention 836 on Within- and Between-Patient Daptomycin Resistance Evolution in Vancomycin-837 Resistant Enterococcus faecium. Antimicrob Agents Chemother 2019;63(4) doi: 838 10.1128/aac.01800-18 [published Online First: 2019/02/06] 839 27. Álvarez-Marín R, López-Cerero L, Guerrero-Sánchez F, et al. Do specific antimicrobial 840 stewardship interventions have an impact on carbapenem resistance in Gram-negative 841 bacilli? A multicentre quasi-experimental ecological study: time-trend analysis and 842 characterization of carbapenemases. Journal of Antimicrobial Chemotherapy 843 2021;76(7):1928-36. doi: 10.1093/jac/dkab073 844 28. López-Viñau T, Peñalva G, García-Martínez L, et al. Impact of an Antimicrobial Stewardship 845 Program on the Incidence of Carbapenem Resistant Gram-Negative Bacilli: An 846 Interrupted Time-Series Analysis. Antibiotics (Basel) 2021;10(5) doi: 847 10.3390/antibiotics10050586 [published Online First: 2021/06/03] 848 29. Rodríguez-Baño J, Pérez-Moreno MA, Peñalva G, et al. Outcomes of the PIRASOA 849 programme, an antimicrobial stewardship programme implemented in hospitals of the 850 Public Health System of Andalusia, Spain: an ecologic study of time-trend analysis. Clin 851 Microbiol Infect 2020;26(3):358-65. doi: 10.1016/j.cmi.2019.07.009 [published Online 852 First: 2019/07/20] 853 30. Schönherr SG, Ranft D, Lippmann N, et al. Changes in antibiotic consumption, AMR and 854 Clostridioides difficile infections in a large tertiary-care center following the 855 implementation of institution-specific guidelines for antimicrobial therapy: A nine-year 856 interrupted time series study. PLoS One 2021;16(10):e0258690. doi: 857 10.1371/journal.pone.0258690 [published Online First: 2021/10/15] 858 31. Akazawa T, Kusama Y, Fukuda H, et al. Eight-Year Experience of Antimicrobial Stewardship 859 Program and the Trend of Carbapenem Use at a Tertiary Acute-Care Hospital in Japan-860 The Impact of Postprescription Review and Feedback. Open Forum Infect Dis 861 2019;6(10):ofz389. doi: 10.1093/ofid/ofz389 [published Online First: 2019/10/30] 862 32. Zequinao T, Gasparetto J, Oliveira DDS, et al. A broad-spectrum beta-lactam-sparing 863 stewardship program in a middle-income country public hospital: antibiotic use and 864 expenditure outcomes and antimicrobial susceptibility profiles. Braz J Infect Dis 865 2020;24(3):221-30. doi: 10.1016/j.bjid.2020.05.005 [published Online First: 2020/06/07] 866 33. Kim YC, Kim EJ, Heo JY, et al. Impact of an Infectious Disease Specialist on an Antimicrobial 867 Stewardship Program at a Resource-Limited, Non-Academic Community Hospital in 868 Korea. J Clin Med 2019;8(9) doi: 10.3390/jcm8091293 [published Online First: 869 2019/08/28] 870 34. Wang H, Wang H, Yu X, et al. Impact of antimicrobial stewardship managed by clinical 871 pharmacists on antibiotic use and drug resistance in a Chinese hospital, 2010-2016: a 872 retrospe ctive observational study. BMJ Open 2019;9(8):e026072. doi: 873 10.1136/bmjopen-2018-026072 [published Online First: 2019/08/05] 874 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint 32 35. Leykum LK, Pugh Ja Fau - Lanham HJ, Lanham Hj Fau - Harmon J, et al. Implementation 875 research design: integrating participatory action research into randomized controlled 876 trials. Implement Sci 2009;4(69) doi: doi: 10.1186/1748-5908-4-69 877 36. Zhang F, Wagner AK, Ross-Degnan D. Simulation-based power calculation for designing 878 interrupted time series analyses of health policy interventions. Journal of Clinical 879 Epidemiology 2011;64(11):1252-61. doi: 10.1016/j.jclinepi.2011.02.007 880 37. Baym M, Stone LK, Kishony R. Multidrug evolutionary strategies to reverse antibiotic 881 resistance. Science 2016;351(6268):aad3292. doi: 10.1126/science.aad3292 [published 882 Online First: 2016/01/02] 883 38. Bonomo RA, Perez F, Hujer AM, et al. The Real Crisis in Antimicrobial Resistance: Failure to 884 Anticipate and Respond. Clinical Infectious Diseases 2024:ciad758. doi: 885 10.1093/cid/ciad758 886 887 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 27, 2025. ; https://doi.org/10.1101/2025.08.25.25334341doi: medRxiv preprint