Antibiotic stewardship interventions in hospitals in low-and middle-income countries: a systematic review.

OBJECTIVE: To review the effectiveness of antibiotic stewardship interventions in hospitals in low- and middle-income countries.
METHODS: We searched MEDLINE®, Embase®, Cochrane Central Register of Controlled Trials and regional indexes for studies of interventions to improve appropriate prescribing and use of antibiotics for hospitalized patients in low- and middle-income countries. We included controlled trials, controlled before-and-after studies and interrupted time-series studies published up to December 2017. We report prescribing, clinical and microbiological outcomes using a narrative approach.
FINDINGS: We screened 7342 original titles and abstracts, assessed 241 full-text articles and included 27 studies from 2 low-income and 11 middle-income countries. We found a medium (11 studies) or high risk (13 studies) of bias. Generally, all types of interventions (structural, persuasive and enabling) and intervention bundles were reported to improve prescribing and clinical outcomes. However, the studied interventions and reported outcomes varied widely. The most frequent intervention was procalcitonin-guided antibiotic treatment (8 of 27 studies, all randomized controlled trials). The intervention was associated with a relative risk for patients receiving antibiotics ranging between 0.40 and 0.87.
CONCLUSION: The majority of studies reported a positive effect of hospital antibiotic stewardship interventions. However, we cannot draw general conclusions about the effectiveness of such interventions in low- and middle-income countries because of low study quality, heterogeneity of interventions and outcomes, and under-representation of certain settings. To strengthen the evidence base, action needs to be taken to address these shortcomings.

Introduction

Antibiotic resistance is a problem of global importance. Representative data on the extent of the problem in low-and middle-income countries are relatively scarce, but high levels of resistance are increasingly being reported worldwide.– Misuse and overuse of antibiotics in humans and animals is one of the main drivers of antibiotic resistance., Antibiotic stewardship, that is, interventions designed to optimize use of antibiotics, is therefore one of the key actions of the World Health Organization (WHO) Global Action Plan to contain antibiotic resistance., Stewardship interventions are typically classified as structural (such as the introduction of new diagnostic tests to guide antibiotic treatment), persuasive (such as expert audit of prescriptions and feedback advice to prescribers), enabling (such as guidelines or education on antibiotic use) or restrictive (such as expert approval for use of certain antibiotics). Often, different interventions are combined in antibiotic stewardship bundles.

Several systematic reviews showed that antibiotic stewardship interventions for hospitalized patients increased compliance with local antibiotic policies and improved clinical patient outcomes.– These reviews included mainly or exclusively papers from high-income countries. Whether these results also apply to low- and middle-income countries is unclear. The organization of health-care system, availability of diagnostic testing and appropriate antibiotics, infection prevention and control practices and prescribing practices (such as over-the-counter availability of antibiotics) differs markedly between high-income countries and low- and middle-income countries. These differences may affect the implementation and effectiveness of antibiotic stewardship interventions in these settings.

Many hospitals in low- and middle-income countries are setting up antibiotic stewardship programmes. To better inform the selection of antibiotic stewardship interventions, we systematically reviewed the literature for studies that describe the effect of these interventions on clinical, microbiological or antibiotic prescribing outcomes in hospitalized patients in low- and middle-income countries.

Methods

The review protocol including the complete search strategy has been registered at the PROSPERO international prospective register of systematic reviews (CRD42016042019).

We included studies on antibiotic stewardship interventions for hospitalized patients in low- and middle-income countries. Stewardship interventions were defined as any intervention aiming to improve appropriate prescribing of antibiotics. A summary of the search strategy is shown in Box 1. Low- and middle-income countries were defined according to the World Bank criteria. To be included, studies had to report at least one prescribing outcome (such as defined daily doses per 100 bed-days), clinical outcome (such as mortality) or microbiological outcome (such as proportion of bacterial isolates with antibiotic resistance). We included (non)randomized controlled trials, cluster randomized controlled trials, controlled before‒after studies and interrupted time-series studies if these contained at least three points of comparison pre-and post-intervention. Studies were excluded if they included residents of long-term health-care or nursing facilities; studied malaria, human immunodeficiency virus, mycobacterial or fungal infections, Helicobacter pylori eradication, or care pathways (such as malnutrition bundles); compared antibiotic regimens; were written in language other than English, Dutch, French, German, Portuguese or Spanish; or had no full-text article available.

We searched the following databases from inception to 5 December 2017: Cochrane Central Register of Controlled Trials, EMBASE®, MEDLINE®, regional databases of the Global Index Medicus and the World Health Organization’s Virtual Health Library. The combination of the following and related terms was used: “low- and middle-income country”, “antibiotic”, “stewardship”, “inpatient” and terms related to study design such as “clinical trial”, “randomized controlled trial”, “interrupted time series”, “controlled before after”. Syntax and wording was adapted to the different libraries. Moreover, we searched reference lists of selected studies and of relevant reviews and consulted experts for additional literature. The full search strategy can be viewed online.

Titles and abstracts were independently screened for eligibility by two authors. In case of disagreement, consensus was sought after reading the full-text article. The study selection was piloted by screening 630 abstracts and 44 full-text articles. These results were discussed among a panel of experts, after which the eligibility criteria were fine-tuned.

Two researchers extracted the data using an electronic form. The authors of original studies were not contacted in cases of incomplete or missing data. Data that were analysed inappropriately in the original studies were excluded. The quality of the studies was evaluated at the study level by two researchers independently. We used the 2017 quality criteria for randomized controlled trials and quasi-experimental studies of the Effective Practice and Organisation of Care Review Group. Reporting was done in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. For controlled trials, intention-to-treat analyses were reported unless indicated otherwise. If the original paper did not mention a relative risk (RR), we calculated a RR and 95% confidence interval (CI) if the necessary data were available. Due to the heterogeneity of the interventions and their reported outcomes, we present our findings using a narrative approach. Because of the large number of reported outcomes, we were unable to report all. We therefore selected the outcomes that were reported most frequently across the studies. We grouped studies by intervention type: structural, persuasive, enabling or intervention bundle.

Results

We screened 7342 abstracts, selected 241 full-text articles and included 27 studies:– 12 interrupted time-series, 9 randomized controlled trials, 3 cluster randomized controlled trials and 3 non-randomized controlled trials (Fig. 1). The studies were performed between 1996 and 2015 in 13 different countries. Two countries were considered low-income at the time of the study, one country transitioned from low to lower-middle income and the remaining were middle-income countries. Nine studies were conducted in multiple hospitals (range 2–65) but the majority was single-centre (18 studies). The interventions were implemented hospital-wide (10 studies) or on specific wards (17 studies) and targeted therapeutic prescriptions (20 studies), surgical prophylaxis (3 studies) or a combination of those (4 studies; Table 1).

Fig. 1

Flowchart of the selection of studies included in the review of antibiotic stewardship interventions in hospitals in low-and middle-income countries

Table 1

Characteristics of studies included in the review of antibiotic stewardship interventions in hospitals in low-and middle-income countries

Authors, year	Study design	Country	Setting	Participants	Intervention type	Intervention details	Target illness
Weinberg et al., 200139	Interrupted time-series	Colombia	2 referral hospitals	Surgeons performing caesarean sections	Bundle	Guidelines on surgical antibiotic prophylaxis; structural changes (availability of prophylactic antibiotics on site); audit and feedback to physicians and nurses at hospital and individual level	Surgical site infections after caesarean section
Perez et al., 200340	Interrupted time-series	Colombia	2 university hospitals	Hospital A: all prescribers; hospital B: anaesthesiologists	Bundle	Prescription form with (un)restricted antibiotics; educational campaign; reminders in the workplace	NR
Gülmezoglu et al., 200727	Cluster randomized controlled trial	Mexico and Thailand	22 non-university maternity hospitals	Physicians, midwives, interns, students	Structural	Access to WHO’s online Reproductive Health Library44 and workshops on its use	Surgical site infections after caesarean section
Hadi et al., 200834	Interrupted time-series	Indonesia	1 teaching hospital	All prescribers of 5 internal medicine wards	Enabling	Antibiotic guidelines; education for prescribers	NR
Özkaya et al., 200926	Non-randomized controlled trial	Turkey	1 university hospital	Paediatric emergency department residents	Structural	Antibiotic initiation guided by influenza rapid test versus no laboratory investigation	Mild influenza-like illness
Rattanaumpawan et al., 201032	Non-randomized controlled trial	Thailand	1 public university hospital	All prescribers	Persuasive	Audit and feedback to prescribers by infectious diseases specialist	NR
Long et al., 201118	Randomized controlled trial	China	1 university hospital	Emergency department physicians	Structural	Antibiotic initiation and discontinuation guided by serum procalcitonin level versus routine carea	Community-acquired pneumonia
Maravić-Stojković et al., 201120	Randomized controlled trial	Serbia	1 tertiary hospital	Cardiac surgery and intensive care unit staff	Structural	Antibiotic initiation guided by serum procalcitonin level versus routine care (based on clinical signs, C-reactive protein levels and leukocyte count)	Infections after coronary artery bypass grafting or valve surgery
Shen et al., 201133	Cluster randomized controlled trial	China	1 tertiary hospital	All prescribers of 2 pulmonary wards	Persuasive	Audit and feedback to prescribers by clinical pharmacist	Respiratory tract infections
Opondo et al., 201137	Cluster randomized controlled trial	Kenya	8 district hospitals	Nurses, clinical and medical officers	Bundle	Guidelines for treatment of non-bloody diarrhoea; education for prescribers; audit and feedback to prescribers on hospital performance	Non-bloody diarrhoea
Bucher et al., 201225	Randomized controlled trial	Peru	1 public hospital	Paediatric emergency department physicians	Structural	Antibiotic initiation guided by faecal rotavirus rapid test in combination with a faecal leukocyte test versus faecal leukocyte test only	Acute diarrhoea
Magedanz et al., 201241	Interrupted time-series	Brazil	1 university hospital	All prescribers of the cardiology department	Bundle	Restriction of certain antibiotics; audit and feedback to prescribers by (i) infectious diseases specialist and (ii) pharmacist	NR
Qu et al., 201224	Randomized controlled trial	China	1 municipal hospital	Intensive care unit staff	Structural	Antibiotic initiation and discontinuation guided by serum procalcitonin level versus standard 14 days of antibiotics	Severe acute pancreatitis
Ding et al., 201317	Randomized controlled trial	China	1 tertiary hospital	Respiratory ward physicians	Structural	Antibiotic initiation and discontinuation guided by serum procalcitonin level versus routine care (based on clinical experience, sputum bacteriology results and leukocyte count)	Acute exacerbation of idiopathic pulmonary fibrosis
Aiken et al., 201336	Interrupted time-series	Kenya	1 public referral hospital	Nursing, medical and operating theatre staff	Bundle	Guidelines on surgical antibiotic prophylaxis; clinician education; patient education posters; audit and feedback to prescribers	Surgical site infections
Oliveira et al., 201323	Randomized controlled trial	Brazil	2 public university hospitals	Intensive care unit staff	Structural	Antibiotic discontinuation guided by serum procalcitonin level versus C-reactive protein test	Sepsis or septic shock
Tang et al., 201321	Randomized controlled trial	China	1 university hospital	Emergency department physicians	Structural	Antibiotic initiation guided by serum procalcitonin level versus routine carea	Acute asthma exacerbation
Chandy et al., 201435	Interrupted time-series	India	1 private tertiary hospital	All prescribers	Enabling	Implementation and dissemination of antibiotic prescribing guidelines	NR
Long et al., 201419	Randomized controlled trial	China	1 university hospital	Emergency department physicians	Structural	Antibiotic initiation guided by serum procalcitonin level versus routine carea	Acute asthma exacerbation
Najafi et al., 201522	Randomized controlled trial	Islamic Republic of Iran	1 university hospital	Intensive care unit staff	Structural	Antibiotic initiation guided by serum procalcitonin level versus routine carea	Severe inflammatory response syndrome
Bao et al., 201542	Interrupted time-series	China	65 public hospitals (30 tertiary; 35 secondary)	All prescribers	Bundle	Implementation of a nationally imposed multifaceted antibiotic stewardship programme	NR
Sun et al., 201543	Interrupted time-series	China	15 public tertiary hospitals	All prescribers	Bundle	Implementation of a nationally imposed multifaceted antibiotic stewardship programme	NR
Gong et al., 201638	Interrupted time-series	China	1 tertiary paediatric hospital	Paediatricians	Bundle	Antibiotic guidelines and prescribing restrictions; audit and feedback to prescribers by pharmacists and infection control physicians; financial penalties according to number of noncompliant prescriptions	NR
Brink et al., 201629	Interrupted time-series	South Africa	47 private hospitals	Physicians, other clinical staff and managers	Persuasive	Audit and feedback to prescribers by a pharmacist	NR
Li et al., 201730	Non-randomized controlled trial	China	6 university hospitals	Physicians of 8 intensive care units	Persuasive	Audit and feedback to prescribers by a pharmacist versus no intervention	NR
Tuon et al., 201728	Interrupted time-series	Brazil	1 university hospital	All prescribers	Structural	Mobile phone application providing antibiotic prescribing guidance	NR
Wattal et al., 201731	Interrupted time-series	India	1 tertiary hospital	Surgeons of 45 units	Persuasive	Audit and feedback to prescribers; focus group discussions per specialty	NR

NR: not reported; WHO: World Health Organization.

a The content of routine care was not specified.

Risk of bias assessment

For the 12 interrupted time-series studies the risk of bias was low (3 studies), medium (8 studies) or high (1 study; Fig. 2). The main risks of bias were that the intervention was not independent of other changes (5 studies) and that incomplete data were not adequately addressed (5 studies). For the 15 (non)randomized trials the risk of bias was medium (3 studies) or high (12 studies). The main risks of bias included the absence of baseline outcome measurement (14 studies), lack of protection against contamination (prescribers could have been involved in treatment of both the intervention and control group; 11 studies), non-random or unclear randomization methods (8 studies) and incomplete data not being adequately addressed (7 studies).

Fig. 2

Assessment of risk of bias in studies included in the review of antibiotic stewardship interventions in hospitals in low-and middle-income countries

a The criteria were: A: intervention independent of other changes; B: shape of intervention pre-specified; C: intervention unlikely to affect data collection; D: knowledge of allocated interventions adequately prevented during study; E: seasonality taken into account; F: incomplete outcome data adequately addressed; G: study free from selective outcome reporting; H: adequate allocation sequencing; I: adequate allocation concealment; J: baseline outcome measures similar; K: baseline characteristics similar; L: any blinding reported; M: study protected against contamination.

b The risk of bias was considered low if all criteria were scored as low, medium if one or two criteria were scored as medium or high, and high if more than two criteria were scored as medium or high.

Structural interventions

Structural interventions were reported by 12 studies,– eight of which were randomized controlled trials of the effect of using serum procalcitonin levels to guide antibiotic treatment (Table 2).– Five of these studies reported antibiotic use as the outcome. All of them found a significant decrease in the percentage of patients receiving antibiotics in the procalcitonin group compared with routine care or C-reactive protein testing. RR ranged between 0.40 and 0.87.– Five studies reported patient deaths as the outcome and found no significant effect of procalcitonin-guided antibiotic use on in-hospital or 30-day mortality.,,–

Table 2

Outcomes of interventions to improve appropriate prescribing and use of antibiotics in hospitals in low-and middle-income countries: controlled trials

Intervention type and study design	Study duration, weeks	No. of patients	Data summary	Outcome measure	Effect size	P
Structural intervention
Procalcitonin guidance
Randomized controlled trial18	201	172	No. of patients receiving antibiotics: 72/86 in procalcitonin group; 79/86 in routine care group	RR of receiving antibiotic (95% CI)	0.87 (0.79 to 0.96)	0.01
Randomized controlled trial20	NR	205	No. of patients receiving antibiotics: 19/102 in procalcitonin group; 48/103 in routine care group	RR of receiving antibiotic (95% CI)	0.40 (0.25 to 0.63)	0.01
			No. of deaths: 3/102 in procalcitonin group; 3/103 in routine care group	RR of in-hospital death (95% CI)	0.88 (0.33 to 2.35)	0.80
Randomized controlled trial17	154	78	No. of patients receiving antibiotics: 26/39 in procalcitonin group; 35/39 in routine care group	RR of receiving antibiotic (95% CI)	0.74 (0.58 to 0.95)	0.01
			No. of deaths: 21/39 in procalcitonin group; 20/39 in routine care group	RR of death after 30 daysa (95% CI)	1.11 (0.76 to 1.64)	0.42
Randomized controlled trial24	133	71	No. of deaths: 7/35 in procalcitonin group; 8/36 in standard 14 days of antibiotics group	RR of in-hospital death (95% CI)	0.90 (0.37 to 2.22)	0.99
Randomized controlled trial23	141	97	No. of deaths: 21/50 in procalcitonin group; 21/47 in routine care group	RR of in-hospital death (95% CI)	0.92 (0.59 to 1.44)	0.84
Randomized controlled trial21	283	265	No. of patients receiving antibiotics: 59/132 in procalcitonin group; 95/133 in routine care group	RR of receiving antibiotic (95% CI)	0.63 (0.50 to 0.78)	0.01
Randomized controlled trial19	133	180	No. of patients receiving antibiotics: 44/90 in procalcitonin group; 79/90 in routine care group	RR of receiving antibiotic (95% CI)	0.56 (0.44 to 0.70)	0.01
Randomized controlled trial22	52	60	No. of deaths: 5/30 in procalcitonin group; 4/30 in routine care group	RR of in-hospital death (95% CI)	1.25 (0.37 to 4.21)	0.71
Rapid diagnostic testing
Non-randomized controlled trial26	21	97	No. of patients receiving antibiotics: 34/50 in influenza rapid diagnostic test group; 47/47 in routine care group	RR of receiving antibiotic (95% CI)	0.68 (0.56 to 0.82)	0.01
Randomized controlled trial25	26	201	No. of patients receiving antibiotics: 29/100 in faecal leukocyte + rotavirus rapid test group; 50/101 in faecal leukocyte test only group	RR of receiving antibiotic (95% CI)	0.59 (0.41 to 0.84)	0.03
Library access plus workshops
Cluster randomized controlled trial27	43 to 52b	1000 to 1022 per hospital	Mean % of operations with antibiotic prophylaxis:Mexico: 25.8 in intervention group; 6.5 in control group Thailand: 26.0 in intervention group; 14.7 in control group	% of operations with antibiotic prophylaxis: difference in adjusted rate (95% CI)	Mexico: 19 (−8 to 46)	0.12
					Thailand: 5 (−18 to 27)	0.66
Persuasive intervention
Audit and feedback on individual patient cases
Non-randomized controlled trial32	17	948	Mean no. of days of hospitalization: 30.4 in intervention group; 30.7 in control group	Mean difference in hospital length of stay (95% CI), days	−0.3 (−3.3 to −3.0)	0.80
			Mean no. of days of treatment: 12.7 in intervention group; 16.4 in control group	Mean difference in treatment duration, days	−3.7 (−5.2 to −2.2)	0.01
Cluster randomized controlled tria33	43	436	Mean no. of days of hospitalization: 14.2 in intervention group; 15.8 in control group	Mean difference in hospital length of stay (95% CI), days	−1.6 (−2.9 to −0.3)	0.03
Non-randomized controlled trial30	9	874	Median no. of days of treatment: 4.0 in intervention group; 5.0 in control group	Difference in median no. of days of treatment	1.0	0.03
Intervention bundle
Treatment guidelines plus education plus audit and feedback
Cluster randomized controlled trial37	77	1160	No. of patients receiving antibiotics for inappropriate indication: 313/594 in intervention group; 437/566 in control group	Absolute risk reduction for receiving antibiotic for inappropriate indication (95% CI)	41 (−6 to 88)	0.08

CI: confidence interval; DDD: defined daily doses; NR: not reported; RR: relative risk.

a Per protocol analysis.

b Different collection periods in different hospitals.

Note: Intention-to-treat analysis results are reported unless indicated otherwise. When significant P-values were not specified, we assumed P < 0.05 as significant.

A non-randomized controlled trial among 97 patients in a Turkish emergency department studied the effect of introducing a rapid diagnostic test for influenza-like disease. A lower percentage of tested patients were prescribed antibiotics compared with patients given clinical examination only (RR: 0.68; 95% CI: 0.56 to 0.82). In a randomized controlled trial among 201 patients in a Peruvian emergency department, use of a rapid test for rotavirus was associated with fewer patients receiving antibiotics (RR: 0.59; 95% CI: 0.41 to 0.84).

In a cluster-randomized controlled trial in Mexico and Thailand health-care staff were given access to the WHO’s online Reproductive Health Library and workshops on its use. Thereafter, it was left open to the 22 participating hospitals whether certain activities, including antibiotic stewardship, were implemented. After 10‒12 months, no significant difference was found in the proportion of caesarean sections in which antibiotic prophylaxis was given, when comparing the 22 intervention hospitals to the 18 control hospitals (difference in adjusted rate in Mexico was 19.0%; 95% CI: −8.0 to 46.0% and in Thailand was 4.6%; 95% CI: −17.7 to 26.9%).

One interrupted time-series study evaluated the implementation of an antibiotic treatment guide through a free-of-charge mobile application (Table 3). Twenty-four months after the intervention there were significant increases in the defined daily doses per 1000 bed-days of recommended antibiotics (amikacin and cefepime) and a significant decrease in non-recommended antibiotics (piperacillin; P = 0.02). Use of other non-recommended antibiotics (meropenem, ciprofloxacin and polymyxin) did not decrease significantly.

Table 3

Outcomes of interventions to improve appropriate prescribing and use of antibiotics in hospitals in low-and middle-income countries: interrupted time-series studies

Intervention	Study segments (duration in weeks)	No. of data points per segment (no. of observations per data point)	Outcome measure	Effect sizea	P
Structural interventions
Mobile phone application28	S1: Pre-intervention (52)	12 (NR)	DDD per 1000 bed-days	Baseline trend NR	N/A
	S2: Post-intervention (52)	12 (NR)		Trend increased for amikacinb	0.02
				Trend increased for cefepimeb	0.01
				Trend decreased for piperacillinb	0.02
				Trend decreased for meropenemb	0.44
				Trend decreased for polymyxinb	0.34
				Trend decreased for ciprofloxacinb	0.08
Persuasive interventions
Audit and feedback on individual patient cases29	S1: Pre-intervention (70)	16 (NR)	DDD per 100 bed-days	Baseline level NR	N/A
				Baseline trend +0.064/month	0.62
	S2: Implementation (104)	24 (NR)		Level change NR	N/A
				Trend change −0.56/month	0.01
	S3: Post-intervention (86)	20 (NR)		Level change NR	N/A
				Trend change −0.20/month	0.05
Audit and feedback at department level31	S1: Pre-intervention (52)	12 (NR)	DDD per 100 bed-days	Baseline level: NR	N/A
				Baseline trend: increasing in 1/35 wardsb	0.05
	S2: Post-intervention (13)	3 (NR)		Level decreased in 3/35 wardsb	0.05
Enabling interventions
Treatment guidelines34	S1: Pre-intervention (16)	9 (14)	DDD per 100 bed-days	Baseline level: NR	N/A
				Baseline trend: −1.0 per 14 days	0.53
	S2: Guideline development (14)	6 (14)		Level change: −31.9	0.03
				Trend change +2.1 per 14 days	0.52
	S3: Guideline declaration (8)	4 (26)		Level change: −29.2	0.11
				Trend change: −9.5 per 14 days	0.14
	S4: Teaching sessions (8)	4 (27)		Level change: +38.2	0.05
				Trend change: +10.0 per 14 days	0.21
	S5: Refresher course (8)	5 (15)		Level change: −2.4	0.88
				Trend change: −9.8 per 14 days	0.15
Treatment guidelines35	S1: Pre-intervention (86)	20 (NR)	DDD per 100 bed-days	Baseline level: 56.9	N/A
				Baseline trend: +0.95 per month	0.01
	S2: Guideline preparation and booklet dissemination (94)	22 (NR)		Level change: NR	N/A
				Trend change: +0.21 per month	0.03
	S3: No new intervention (104)	24 (NR)		Level change: NR	N/A
				Trend change: +0.31 per month	0.01
	S4: Guideline revision and booklet dissemination (104)	24 (NR)		Level change: NR	N/A
				Trend change: +0.05 per month	0.64
	S5: Guideline revision and booklet with electronic dissemination (86)	20 (NR)		Level change: NR	N/A
				Trend change: −0.37 per month	0.01
Intervention bundles
Treatment guidelines plus structural changes39	Hospital A
	S1: Pre-intervention (13)	3 (308)	% of operations with surgical site infection	Baseline level: 13.9	N/A
				Baseline trend: NRc	NR
	S2: Guideline introduction with structural changes (30)	7 (272)		Level change: −9.8	0.01
				Trend change: NRc	NR
	S3: Post-intervention (21)	5 (217)		Level change: NRc	NR
				Trend change: NRc	NR
	Hospital A
	S1: Pre-intervention (13)	3 (308)	% of caesarean sections with administration of antibiotic prophylaxis	Baseline level: 47.5	N/A
				Baseline trend: NRc	NR
	S2: Guideline introduction with structural changes (30)	7 (272)		Level change: +31.6	0.01
				Trend change: NRc	NR
	S3: Post-intervention (21)	5 (217)		Level change: −4.9	0.01
				Trend change: NRc	NR
	Hospital B:
	S1: Pre-intervention (13)	3 (396)	% of caesarean sections with administration of antibiotic prophylaxis	Baseline level: 5.1	N/A
				Baseline trend: NRc	NR
	S2: Guideline introduction with structural changes (39)	9 (1026)		Level change: NRc	NR
				Trend change: +5.4 per month	0.01
	S3: Post-intervention (52)	12 (709)		Level change: +7.1	0.05
				Trend change: −4.1	0.01
	Hospital A
	S1: Pre-intervention (13)	3 (308)	% of caesarean sections with administration of antibiotic prophylaxis within 1 hour of delivery	Baseline level: 32.5	N/A
				Baseline trend: NRc	NR
	S2: Guideline introduction with structural changes (30)	7 (272)		Level change: 62.2	0.01
				Trend change: NRc	0.01
	S3: Post-intervention (21)	5 (217)		Level change: NRc	NR
				Trend change: NRc	NR
	Hospital B
	S1: Pre-intervention (13)	3 (396)	% of caesarean sections with administration of antibiotic prophylaxis within 1 hour of delivery	Baseline level: 30.8	N/A
				Baseline trend: +18.4 per month	0.01
	S2: Guideline introduction with structural changes (39)	9 (1026)		Level change: NRc	NR
				Trend change: −18.7 per month	0.01
	S3: Post-intervention (52)	12 (709)		Level change: +15.2	NR
				Trend change: NRc	NR
Prescription form plus education plus reminders40	S1: Pre-intervention (103)	103 (NR)	% of operations with incorrect timing of antibiotic prophylaxis	Baseline level: NR	N/A
				Baseline trend: NR	N/A
	S2: Post-intervention (42)	42 (NR)		Level change: −20	0.01
				Trend change: NR	NR
Antibiotic restrictions plus audit and feedback41	S1: Pre-intervention (129)	30 (NR)	Antibiotic use, DDD per 100 bed-days	Baseline level: NR	N/A
				Baseline trend +1.2 per month	0.01
	S2: Antibiotic restrictions plus audit and feedback by infectious diseases specialist (94)	22 (NR)		Level change: −1.3	0.8
				Trend change: −2.7 per month	0.01
	S3: Antibiotic restrictions plus audit and feedback by pharmacist (86)	20 (NR)		Level change: +4.7	0.4
				Trend change: +1.2 per month	0.01
Treatment guidelines plus education plus audit and feedback36	Timing study
	S1: Pre-intervention (26)	26 (NR)	% of operations with incorrect timing of antibiotic prophylaxis	Baseline level: 99%	N/A
				Baseline trend: NR	N/A
	S2: Post-intervention (40)	40 (NR)		Level decreasedb	0.01
				Trend decreasedb	0.01
	Infection study
	S1: Pre-intervention (26)	6 (223)	% of operations with surgical site infection	Baseline level: NR	N/A
				Baseline trend: −0.5 per month	0.49
	S2: Post-intervention (39)	9 (223)		Level change: NR	0.05
				Trend change: −0.7 per month	0.03
Multifaceted antibiotic stewardship programme42	Outcome A
	S1: Pre-intervention (52)	12 (NR)	% of patients receiving antibiotic	Baseline level: NR	N/A
				Baseline trend +0.3 per month	> 0.05
	S2: Implementation (52)	12 (NR)		Level change: −2.3	> 0.05
				Trend change: −2.3 per month	0.01
	S3: Post-intervention (104)	24 (NR)		Level change: −2.7	0.05
				Trend change: +1.9 per month	0.01
	Outcome B
	S1: Pre-intervention (52)	12 (NR)	Antibiotic use, DDD per 100 bed-days	Baseline level: NR	N/A
				Baseline trend: −0.4 per month	0.2
	S2: Implementation (52)	12 (NR)		Level change: +2.8	> 0.05
				Trend change: −2.2 per month	0.01
	S3: Post-intervention (104)	24 (NR)		Level change: −7.1	0.01
				Trend change: +2.4 per month	0.01
Multifaceted antibiotic stewardship programme43	S1: Pre-intervention (334)	26 (58)	% of patients receiving antibiotic	Baseline level: 74.7	N/A
				Baseline trend: −0.3 per quarter	0.01
	S2: Post-intervention (78)	6 (750)		Level change: −7.3	0.04
				Trend change: −1.5 per quarter	0.07
Treatment guidelines plus antibiotic restrictions plus audit and feedback38	S1: Pre-intervention (17)	4 (375 985)	% of patients receiving antibiotic	Baseline level: 59.0	N/A
				Baseline trend: −3.0 per month	0.01
	S2: Guidelines and restrictions (21)	5 (424 702)		Level change: +3.0	0.2
				Trend change: −0.4 per month	0.6
	S3: Financially punished audit and feedback (60)	14 (446 727)		Level change: −9.0	0.01
				Trend change: +3.0 per month	0.01

DDD: defined daily doses; N/A: not applicable; NR: not reported; RR: relative risk; S: segment.

a In interrupted times-series studies the linear curve which summarizes the outcome data in each study segment can be defined by its level (y-intercept) and trend (slope). Level change reflects the difference of the level of the current segment compared with the level of the previous segment. Trend change reflects the difference of the trend of the current segment compared to the trend of the previous segment.

b The authors reported no values for level or trend changes.

c The authors reported that there were no significant changes but with no values for levels or trend changes.

Persuasive interventions

Four studies evaluated the effect of audit and feedback to prescribers on individual patient cases by pharmacists (3 studies) or infectious diseases specialists (1 study):,,, A non-randomized controlled trial including 577 patients in eight intensive care units reported a decrease of duration of antibiotic treatment of −1.0 day (P = 0.03) (Table 2). Another non-randomized controlled trial of 948 patients in a public university hospital reported a decrease of duration on antibiotic treatment of −3.7 days (P < 0.01) and a decrease in mean length of hospital stay of −1.6 days (P = 0.03). A cluster randomized trial found no significant difference in mean length of hospital stay among 436 patients (0.3 days; P = 0.8). An interrupted time-series study in 47 private hospitals in South Africa found a decreasing trend of antibiotic use during the implementation phase of the intervention (−0.56 defined daily doses per 100 bed-days per month; P < 0.01; Table 3). The trend was sustained in the 20 months post-implementation (−0.20 defined daily doses per 100 bed-days per month; P < 0.05).

An interrupted time-series study evaluated the effect of audit and feedback at the departmental level in 35 surgical wards. Three months after the intervention a significant decrease in defined daily doses per 100 bed-days was reported in 3 out of 35 wards (immediate decreases of −66.5%, −46.1% and −26.4% respectively; P  <  0.05).

Enabling interventions

Two interrupted time-series studied the effect of enabling interventions on antibiotic prescribing (Table 3)., A study in an Indonesian hospital subsequently studied the development of treatment guidelines which were officially presented, followed by education and then refresher education. The authors reported a significant decrease of −31.9 defined daily doses per 100 bed-days (P  =  0.03) after guideline development and a significant increase of +38.2 defined daily doses per 100 bed-days (P  <  0.05) after education. The net effect of the intervention remains unclear. Another study in an Indian hospital evaluated the effect of an antibiotic policy guideline which was first developed and introduced, then revised and made available as booklet and lastly revised and made available through the intranet. The authors initially reported a baseline rising trend in antibiotic use of +0.95 defined daily doses per 100 bed-days per month (P  <  0.01) which levelled off after the first two interventions and declined by −0.37 defined daily doses per 100 bed-days per month (P  <  0.01) after the last intervention.

Intervention bundles

Eight studies evaluated bundles combining different interventions.– A cluster randomized controlled trial in eight Kenyan hospitals compared a bundle containing guidelines, education and face-to-face feedback to prescribers with a similar, but less intensive bundle (fewer hours of training, written feedback; Table 2). Comparing prescriptions for 594 children in intervention hospitals and 566 children in control hospitals showed that the intensive bundle was associated with a non-significant absolute risk reduction in inappropriate use of antibiotics for non-bloody diarrhoea of 41% (95% CI: −6 to 88%).

The other seven studies all used an interrupted time-series design (Table 3). One study in two Colombian hospitals implemented antibiotic prophylaxis guidelines for caesarean sections, immediate availability of antibiotics in the operating theatre and feedback to surgeons about surgical site infections. The study reported a significant increase in the percentage of caesarean section births in which prophylaxis was administered (immediate increase by +31.6% in hospital A; P <  0.01 and gradual increase by +5.4% per month in hospital B; P <  0.01), an increase in antibiotic administration within 1 hour of delivery (immediate increase by 62.2% in hospital A only; P <  0.01) and a significant decrease in the monthly rate of surgical site infections with 9.8% (P <  0.01) in hospital A.

In another study in a Kenyan hospital, surgical antibiotic prophylaxis guidelines were implemented, combined with training, personal feedback to prescribers and patient information posters. The proportion of operations with incorrect timing of antibiotic prophylaxis significantly decreased (no values reported) and the percentage of surgical site infections decreased after the intervention by −0.7% per month (P  =  0.03).

Another Colombian study introduced an antibiotic prescription form containing a list of restricted antibiotics with information on dosing intervals and an educational campaign. The study found a decrease of 20% (P  <  0.01) in the proportion of operations with incorrect timing of surgical prophylaxis.

In a Chinese study, guidelines and antibiotic restrictions were introduced, followed by individual prescriber audit and feedback, with financial penalties and revocation of prescribing privileges in case of non-compliance. Before the intervention the proportion of patients on antibiotic treatment was decreasing significantly by −3% per month from a baseline level of 59% (P  =  0.01). After the first intervention, no significant changes were reported. After the second intervention, a sudden drop of −9% (P  =  0.01) was observed, followed by a steady increase of +3% per month (P  =  0.01) in the next 14 months. The net effect of the intervention bundle remains unclear.

A study in a Brazilian cardiology hospital first introduced restriction of certain antibiotics with individual audit and feedback to prescribers by an infectious diseases specialist and subsequently more comprehensive audit and feedback by a pharmacist. Before the intervention, the total antibiotic consumption significantly increased during 30 months (+1.2 defined daily doses per 100 bed-days per month; P  <  0.01). This trend decreased after the first intervention (−2.7 per month; P  <  0.01) and increased after the second (+1.2 per month; P  <  0.01). The net effect of the intervention bundle remains unclear.

Two Chinese studies looked at the implementation of a multifaceted national antibiotic stewardship programme, containing structural changes, antibiotic restriction, education, guidelines, and audit and feedback, in 65 and 15 secondary and tertiary public hospitals respectively., Participation was compulsory and financial punishment for hospitals and disciplinary actions for individual prescribers could be imposed. Both studies reported a significant decrease in antibiotic use after the intervention. One study reported a decreasing trend of −2.2 defined daily doses per 100 bed-days per month (P  <  0.01). The other study reported a decrease in the proportion of patients receiving antibiotics (−7.3%; P  =  0.04).

Discussion

In this systematic review the majority of the included studies reported a positive effect of antibiotic stewardship interventions for hospitalized patients. This is in line with previously published systematic reviews on stewardship interventions in hospitals, which did not focus specifically on low- and middle-income countries.– However, we cannot make general recommendations to guide the selection of antibiotic stewardship interventions due to limitations of the included studies, including the low quality of methods, variations and shortcomings in outcome reporting, under-representation of certain settings, heterogeneity of the interventions and variations in implementation strategy.

When screening titles and abstracts, we found 153 articles that reported on stewardship activities in a hospital setting, but 126 of those were excluded because of the study design (mainly bias-prone uncontrolled before‒after studies). So, although antibiotic stewardship is taking place and is being studied in low- and middle-income countries, most studies fall short methodologically. The studies we did include were also generally of low quality. For those with a randomized study design, a major risk of bias was contamination, meaning that prescribers could be involved in treatment of both the intervention and control groups. Because it may not be feasible to randomize individual prescribers, wards or hospitals to overcome this bias, interrupted time-series design has been recommended as an alternative. In interrupted time-series, data are collected continuously, and trends and outcome levels are compared before and after the intervention. To minimize bias and confounding, interrupted-time-series should meet certain requirements: a minimum of 12 data points before and after intervention, 100 observations per data point and the use of analytic techniques or models. These requirements were seldom met by the included studies. Poor quality of methods is a consistent theme among reviews of antibiotic stewardship in countries of all income levels and this issue needs to be addressed to strengthen the evidence base.,,

Many of the included studies focused on a quantitative reduction in antibiotic prescribing. However, stewardship is not merely concerned with a reduction in antibiotic use, but in finding the balance between the potency of antibiotics and their potentially hazardous effects. The goal is to improve patient outcomes, decrease antibiotic resistance and increase cost‒effectiveness of care. Therefore, it is recommended that clinical outcomes (including adverse events), microbiological and cost‒effectiveness outcomes are reported in all stewardship studies., Most of the studies included in this review failed to do so. There is an ongoing debate about which parameters should be reported to accurately reflect the above-mentioned outcomes., This generally leads to a wide variety of reported parameters, as we observed in our review. This lack of uniformity limits comparison and aggregation of data. Also, for low- and middle-income settings, the measurement of certain clinical or microbiological outcomes, for example infection with Clostridium difficile, may be challenging if not impossible. Defining feasible outcome measures that can be uniformly applied in low- and middle-income countries should be prioritized. In the meantime, parameters that are easy to assess, such as mortality or hospital length of stay, should be reported by every stewardship study.

The majority of studies were performed in tertiary care centres in urban areas in middle-income countries, which limits the generalizability of the results. Large differences exist in terms of resources, organization, prescription practices and financing between countries and between facilities within countries. The intervention most frequently studied in our review was the implementation of procalcitonin testing. Although this intervention showed promising results, it may not be feasible to implement in many health-care settings in low- and middle-income countries. In addition, good quality evidence from non-tertiary or rural hospitals in low-income countries is lacking. Studies focusing on these settings should therefore be prioritized.

The effectiveness of the interventions varied across the studies, even those that implemented similar interventions. This is likely due to differences in the intervention or the implementation strategy, which may have been adapted to fit local circumstances. A detailed description of the intervention and the implementation strategy is therefore mandatory to interpret the study findings. Stewardship interventions in hospitals usually aim to change individual prescriber’s behaviour. This behaviour is influenced by social norms, attitudes and beliefs. These are therefore important determinants of the effectiveness of the intervention and should be an integral part of studies of stewardship interventions. For this reason, collaboration with behavioural scientists has been recommended. None of the included studies reported behaviour determinants.

Our review has several limitations. We defined a broad search strategy, allowing different settings, participants, interventions and outcomes to be included. This strategy provides a good overview of what evidence is available, but limits the generalizability of the findings. Moreover, to ensure the validity of the results, studies had to fulfil high methodological standards to be included. This led to discarding numerous lower quality studies. Also, we did not include studies that only reported cost (effectiveness) as an outcome, as these require a different analysis model. Lastly, due to publication bias (not reporting negative results) and language restrictions we may have missed certain studies.

We conclude that, based on the currently available evidence, general recommendations regarding the effectiveness of antibiotic stewardship interventions in low- and middle-income countries cannot be made. As many hospitals in low- and middle-income countries are setting up antibiotic stewardship programmes, what should be the way forward? On the basis of our findings, we suggest the following actions should be prioritized to strengthen the evidence base: (i) provision of methodological and statistical support for commonly used, complex study designs such as interrupted-time-series; (ii) seeking consensus on relevant and feasible outcome measurements for low- and middle-income countries; (iii) performing methodologically solid studies in settings such as non-tertiary, rural and public hospitals in low-income countries; and (iv) accurate descriptions of interventions, implementation strategies and inclusion of behavioural aspects. While awaiting the effect of these actions, the current lack of evidence should not prevent health-care workers from engaging in stewardship. Evidence and examples both from high- and low-and middle-income countries can inspire and provide guidance in the meantime.–