Antibiotic prescription practices in primary care in low- and middle-income countries: A systematic review and meta-analysis.

BACKGROUND: The widespread use of antibiotics plays a major role in the development and spread of antimicrobial resistance. However, important knowledge gaps still exist regarding the extent of their use in low- and middle-income countries (LMICs), particularly at the primary care level. We performed a systematic review and meta-analysis of studies conducted in primary care in LMICs to estimate the prevalence of antibiotic prescriptions as well as the proportion of such prescriptions that are inappropriate. METHODS AND
FINDINGS: We searched PubMed, Embase, Global Health, and CENTRAL for articles published between 1 January 2010 and 4 April 2019 without language restrictions. We subsequently updated our search on PubMed only to capture publications up to 11 March 2020. Studies conducted in LMICs (defined as per the World Bank criteria) reporting data on medicine use in primary care were included. Three reviewers independently screened citations by title and abstract, whereas the full-text evaluation of all selected records was performed by 2 reviewers, who also conducted data extraction and quality assessment. A modified version of a tool developed by Hoy and colleagues was utilized to evaluate the risk of bias of each included study. Meta-analyses using random-effects models were performed to identify the proportion of patients receiving antibiotics. The WHO Access, Watch, and Reserve (AWaRe) framework was used to classify prescribed antibiotics. We identified 48 studies from 27 LMICs, mostly conducted in the public sector and in urban areas, and predominantly based on medical records abstraction and/or drug prescription audits. The pooled prevalence proportion of antibiotic prescribing was 52% (95% CI: 51%-53%), with a prediction interval of 44%-60%. Individual studies' estimates were consistent across settings. Only 9 studies assessed rationality, and the proportion of inappropriate prescription among patients with various conditions ranged from 8% to 100%. Among 16 studies in 15 countries that reported details on prescribed antibiotics, Access-group antibiotics accounted for more than 60% of the total in 12 countries. The interpretation of pooled estimates is limited by the considerable between-study heterogeneity. Also, most of the available studies suffer from methodological issues and report insufficient details to assess appropriateness of prescription.
CONCLUSIONS: Antibiotics are highly prescribed in primary care across LMICs. Although a subset of studies reported a high proportion of inappropriate use, the true extent could not be assessed due to methodological limitations. Yet, our findings highlight the need for urgent action to improve prescription practices, starting from the integration of WHO treatment recommendations and the AWaRe classification into national guidelines. TRIAL REGISTRATION: PROSPERO registration number: CRD42019123269.

Introduction

Antimicrobial resistance (AMR) is a major health threat globally [1]. Growing morbidity and mortality rates due to resistant infections in humans are expected worldwide, along with a substantial economic impact in terms of productivity losses and healthcare expenditures [2,3].

Several factors are known to play a role in the development and spread of AMR, with inappropriate use of antibiotics being one of its most important drivers [4]. Gathering data about resistance as well as antibiotic use is 1 of the top 5 priorities of the Global Action Plan on Antimicrobial Resistance by the World Health Organization (WHO) [5].

A multinational survey conducted across 76 countries to determine the magnitude of antibiotic consumption and its trend over time revealed a dramatic increase between 2000 and 2015 (+65% globally), mostly driven by a sharp rise in low- and middle-income countries (LMICs) (+114%), where the levels of antibiotic consumption are high and rapidly approaching those observed in high-income countries (HICs) [6]. However, this analysis was based on drug sales data, thus providing limited information regarding providers’ prescription habits.

The high level of antibiotic consumption in LMICs is because of multiple factors, including the high burden of infectious diseases, lack of regulations (or weak enforcement) to prevent over-the-counter sale of antibiotics, inadequate training of healthcare professionals, and the limited availability of essential diagnostics, which favors empirical use of antibiotics [1,7,8]. Besides misuse (i.e., prescription without clinical indication), another huge concern is the inappropriate use of antibiotics in terms of choice of a suitable molecule, dosage, and duration of treatment according to the site of infection and patient’s characteristics.

Most studies investigating the magnitude and determinants of antibiotic use have focused on HICs, and those from LMICs have been carried out predominantly in hospital settings [9-12], leaving a number of unanswered questions about current practices at the primary healthcare level, where the bulk of antibiotic use takes place.

Of note, there is a paucity of information regarding the degree and pattern of antibiotic use in outpatient primary healthcare facilities, i.e., any service (other than pharmacies) providing care for people making an initial contact with a health professional. Having this information will be helpful to design and implement effective stewardship interventions and policies in LMICs.

We conducted a systematic review of the literature to assess the extent and patterns of antibiotic prescription and their determinants at the primary healthcare level in LMICs, as well as the proportion of such prescriptions deemed to be inappropriate.

Methods

The protocol for this systematic review was registered in the International Prospective Register of Systematic Reviews (PROSPERO) (identifier: CRD42019123269) and followed the PRISMA guidelines [13]. The PRISMA checklist and PROSPERO protocol are provided as S1 PRISMA Checklist and S1 PROSPERO Protocol.

Search strategy and selection criteria

We performed a systematic review of cross-sectional studies that were conducted in primary care in LMICs and reported the proportion of individuals receiving any antibiotic or the proportion of drug prescriptions that included an antibiotic. We also examined randomized and non-randomized trials as well as other observational studies to determine whether potentially relevant information (e.g., results from preliminary field assessments including cross-sectional drug prescription data) was provided. Conference proceedings and abstracts, commentaries, editorials, reviews, mathematical modeling studies, economic analyses, qualitative studies, and studies published in predatory journals as defined by Beall [14] were excluded. Studies conducted solely in an inpatient setting, those that focused on veterinary use of antibiotics, and those that only enrolled patients belonging to special cohorts (e.g., patients with cystic fibrosis or neutropenia or other underlying conditions that may justify an increased empirical use of antibiotics, or patients receiving antibiotics as part of prophylactic regimens) were also ineligible. No restrictions were applied with regards to the population characteristics in terms of age, sex, pregnancy status, or HIV status.

For the purpose of the study, we considered as “primary care” any care provided by any health professional (other than pharmacists) with whom patients have their initial contact, in the public or private sector, including primary care delivered in hospital settings wherever appropriate. In cases of uncertainty, we contacted the study authors for clarification. Antibiotics were defined as any agents included in the J01 group of the ATC (Anatomical Therapeutic Chemical) classification system [15]. Inappropriate prescriptions were recorded when such assessment was performed in the original studies. Countries were classified as low, lower-middle, upper-middle, or high income following the World Bank categorization based on gross national income per capita (GNI) of the study start year [16]. GNI thresholds for the definition of such categories, which have changed slightly over time, are provided in S1 Table. Given that there is no international standard definition of “urban” and “rural” areas, we classified the study settings in accordance with the authors’ statements. If not explicitly stated by the investigators, we categorized as “urban” any site with a minimum population of 2,000 inhabitants, i.e., the most frequently used cutoff [17].

The search strategy was built in collaboration with a medical librarian (GG), using key terms for “antibiotic,” “primary healthcare,” “prescribing,” and “LMICs” (both as a group and as individual countries, adopting a filter that was developed according to the World Bank categories). Medline (PubMed), Embase (Ovid), Global Health (Ovid), and CENTRAL (Cochrane Library) were systematically searched from 1 January 2010 until 4 April 2019. We also reran our search on 11 March 2020 using PubMed only; for feasibility reasons, the update could not be conducted through all data sources used in the initial search. Studies conducted before 1 January 2010 were excluded. The start date of our search was established after the conduction of an exploratory review of the literature showing that only a small number of studies were performed before 2010 in relevant settings, in the face of the exponentially higher number of total records identified through our search strategy, which would have posed substantial feasibility issues with very little benefit. Additionally, as patterns of antibiotic prescribing have changed substantially over time, including older studies would have been of limited value for understanding the current situation. No language restrictions were applied. The full search strategies for each database are presented in S1 Text.

Study screening and data extraction

Search results were imported into a citation manager (EndNote X9, Clarivate Analytics), and duplicates were removed. Three authors (GS, PA, and VN) independently screened citations by title and abstract against predefined eligibility criteria. The full-text review of all selected records was performed by 2 authors (GS and PA). An electronic data extraction form was piloted on 5 randomly selected papers and then used by 2 reviewers (GS and PA) to extract information from all eligible publications. At each stage of the screening and data extraction process, disagreements were resolved through discussion, and, if necessary, a third author (SG) was consulted to reach consensus. Study authors were contacted to request clarifications or additional data if needed. A detailed description of the screening and data extraction process is provided in S2 Text along with interrater agreement statistics.

Assessment of study quality and publication bias

A modified version of a tool developed by Hoy and colleagues was utilized to evaluate the risk of bias of each included study (S2 Table) [18]. Our checklist included 8 methodological items (rated as low or high risk of bias), plus a summary item on the overall risk of study bias (rated as low, moderate, or high); no numeric scores were applied. All findings from this assessment were recorded in the data extraction form by the same independent reviewers. As a sensitivity analysis, we excluded studies with a high overall risk of bias.

No formal assessment of publication bias could be performed since traditional approaches such as funnel plots and tests for asymmetry are considered unsuitable for prevalence studies [19].

Statistical analysis

Depending on the type of data available from individual studies, we calculated either the proportion of patients evaluated in a given health facility or by a certain provider who received antibiotics or the proportion of all drug prescriptions containing any antibiotics, along with their Clopper–Pearson (or exact) 95% confidence intervals (CIs) [20]. The 2019 WHO Access, Watch, and Reserve (AWaRe) framework was used to classify antibiotics according to their potential for selecting resistance [21]. Access-group antibiotics are first-line and narrow-spectrum agents such as penicillin, amoxicillin, and trimethoprim-sulfamethoxazole. Watch-group antibiotics are broad-spectrum agents with higher resistance selection such as second- and third-generation cephalosporins, and fluoroquinolones. Reserve-group antibiotics include last-resort antibiotics such as colistin. Fixed-dose combinations of antibiotics (e.g., ciprofloxacin/ornidazole) were classified as “discouraged” antibiotics, in line with WHO recommendations.

Random-effects meta-analyses were performed to estimate pooled proportions after Freeman–Tukey transformation to normalize the outcome [22]. To assess the between-study heterogeneity, we used the I2 statistic and calculated prediction intervals (i.e., a type of confidence interval that provides the 95% range of true values to be expected in similar studies) [23,24]. Random-effects meta-regression with Knapp–Hartung adjustment (aimed to accommodate high degrees of heterogeneity) was employed to investigate the sources of heterogeneity. Categorical predictors for facility location (urban/rural), healthcare sector (public/private), age group (adults/children/all), type of patients (i.e., patients seeking care for any reason or individuals with a specific condition, e.g., diarrhea), and source of prescription information were considered for building the model. If collinearity issues were observed, variables with the lowest number of missing values were prioritized and included in the model.

Subgroup analyses were conducted to investigate potential differences across levels of country income and types of patients involved (with a focus on studies where all patients attending 1 or more facilities were considered without placing restrictions based on their clinical presentation).

Sensitivity analyses were done by repeating analyses without studies that (i) were conducted in Iran as they were all based on administrative data from national registers; (ii) did not report details on the population and/or health facility location; (iii) were conducted in low-income countries; (iv) were based on the standardized patient methodology, in which antibiotics were deemed inappropriate by indication; (v) were deemed to be low quality (i.e., overall risk of study bias scored as “high”).

All analyses were conducted in Stata (version 14; StataCorp) [25,26].

Results

Our initial search yielded 9,604 unique citations, and an additional 590 were retrieved through our search update. A total of 48 studies (all cross-sectional) were finally included in the analyses (Fig 1) [27-74]. All included publications were in English language, except for 1 that was in Spanish. A summary of the main study characteristics is presented in Table 1, and the full dataset used for analyses is provided as S1 Data. Most studies were conducted in lower-middle- or upper-middle-income countries (22 and 19, respectively), while only 6 were in a low-income country. Additionally, 1 study was carried out in 3 countries (1 low income and 2 lower-middle income) [70]. Both public and private healthcare services were involved in 10 of the 48 (20.8%) included studies, whereas 26 (54.2%) studies were focused on the public sector, 4 (8.3%) were focused on the private sector, and 8 (16.7%) did not provide this information; none of the studies mentioned any involvement of informal practitioners. Facilities located in urban areas were more represented than those located in rural areas (17/48 studies [35.4%; 95% CI: 22.2%–50.5%] versus 10/48 studies [20.8%; 95% CI: 10.5%–35.0%]), with 13 (27.1%) studies involving both settings and 8 (16.7%) not reporting sufficient details. While 9 (18.8%) studies only included individuals presenting with 1 prespecified condition (i.e., acute respiratory illness, diarrhea, or fever), the other studies did not apply restrictions on the reason for seeking care and/or the final diagnosis (if any) and likely included patients with various conditions. None of the studies focused solely on dental care; although it is possible that patients seeking dental care were included in some studies, this group likely represented a negligible proportion of the total sample. Of note, no clinical information was reported in most studies.

Fig 1

PRISMA diagram.

Table 1

Characteristics of studies identified through systematic review.

Income level	Study	Country	Health sector	Facility location	Number of facilities involved	Data source	Age group	Denominator*
Low	Baltzell 2019 [68]	Malawi	Private	Rural	NA	Medical records	NA	9,924 (P)
	Mukonzo 2013 [27]	Uganda	Both	Both	1	Medical records, prescription audit	All	173 (P)
	Nepal 2020 [73]	Nepal	Public	Urban	NA	Prescription audit	All	950 (P)
	Savadogo 2014 [28]	Burkina Faso	Public	Urban	2	Medical records	Children	376 (P)
	Worku 2018 [29]	Ethiopia	Public	Urban	6	Medical records, prescription audit	All	898 (D)
	Yebyo 2016 [30]	Ethiopia	Public	Rural	4	Medical records	Adults	414 (P)
Lower-middle	Abdulah 2019 [31]	Indonesia	Public	NA	25	Prescription audit	Adults	10,118 (D)
	Adisa 2015 [32]	Nigeria	Public	Urban	8	Prescription audit	Adults	400 (P)
	Ahiabu 2016 [33]	Ghana	Both	Both	4	Medical records	All	1,600 (D)
	Akl 2014 [34]	Egypt	Public	Urban	10	Medical records	NA	1,000 (D)
	Atif 2016 [35]	Pakistan	NA	Urban	10	Prescription audit	NA	1,000 (D)
	Beri 2013 [36]	India	Private	Urban	20§	Provider interview	All	400 (P)
	Chem 2018 [37]	Cameroon	Both	Both	26	Medical records	All	30,096 (D)
	El Mahalli 2011 [38]	Egypt	Public	Urban	2	Medical records	Children	300 (P)
	Graham 2016 [39]	Zambia	NA	NA	90§	Provider interview	Children	537 (P)
	Jose 2016 [40]	India	Public	Rural	1	Prescription audit	Children	552 (D)
	Kasabi 2015 [41]	India	Public	NA	20	Medical records	NA	600 (P)
	Mekuria 2019 [72]	Kenya	Private	Urban	4	Prescription audit	All	17,382 (P)
	Ndhlovu 2015 [42]	Zambia	Both	Both	148	Patient interview, medical records	All	872 (P)
	Omole 2018 [43]	Nigeria	Both	Rural	NA	Prescription audit	NA	4,255 (D)
	Oyeyemi 2013 [44]	Nigeria	Public	Urban	4	Medical records	All	600 (D)
	Raza 2014 [45]	Pakistan	Both	Urban	NA	Prescription audit	NA	1,097 (D)
	Sarwar 2018 [46]	Pakistan	Public	Both	32	Prescription audit	NA	6,400 (D)
	Saurabh 2011 [47]	India	NA	Rural	4	Prescription audit	NA	600 (D)
	Saweri 2017 [48]	PNG	Public	Both	7	Ad hoc form	All	6,008 (P)
	Sudarsan 2016 [49]	India	Public	Urban	1	Prescription audit	NA	360 (D)
	Yousif 2016 [50]	Sudan	Both	NA	220§	Prescription audit	NA	19,690 (D)
	Yuniar 2017 [51]	Indonesia	Both	NA	56	Prescription audit	NA	1,657 (D)
Upper-middle	Ahmadi 2017 [52]	Iran	Public	Rural	103	Prescription audit	NA	352,399 (D)
	Alabid 2014 [53]	Malaysia	Private	Urban	70	Patient interview	Adults	140 (P)
	Bielsa-Fernandez 2016 [54]	Mexico	NA	Urban	109§	Provider interview	All	1,840 (P)
	Gasson 2018 [55]	South Africa	Public	Urban	8	Medical records	All	654 (P)
	Greer 2018 [56]	Thailand	Public	Both	32	Medical records	All	83,661 (P)
	Lima 2017 [57]	Brazil	NA	NA	20	Prescription audit	NA	399 (D)
	Liu 2019 [71]	China	Public	Both	65	Prescription audit	All	428,475 (D)
	Mashalla 2017 [58]	Botswana	Public	Urban	19	Prescription audit	All	550 (D)
	Ab Rahman 2016 [59]	Malaysia	Both	Both	545	Medical records	All	27,587 (P)
	Sadeghian 2013 [60]	Iran	NA	NA	NA	Prescription audit	NA	4,940,767 (D)
	Safaeian 2015 [61]	Iran	NA	Both	3,772§	Prescription audit	NA	7,439,709 (D)
	Sánchez Choez 2018 [62]	Ecuador	Public	Both	1	Prescription audit	All	1,393 (P)
	Sun 2015 [63]	China	Public	Both	24	Prescription audit	All	1,468 (D)
	Wang 2014 [64]	China	Public	Both	48	Medical records	All	7,311 (D)
	Xue 2019 [65]	China	Public	Rural	NA	SP exit interview	All	526 (P)
	Yin 2015 [66]	China	Both	Urban	2,501	Prescription audit	NA	42,200 (D)
	Yin 2019 [74]	China	Public	Rural	8	Prescription audit	All	14,526 (D)
	Zhan 2019 [69]	China	Public	Rural	17	Prescription audit	All	1,720 (D)
	Zhang 2017 [67]	China	Public	Rural	20	Prescription audit	Children	9,340 (D)
Multiple	Kjærgaard 2019 [70]	Kyrgyzstan, Uganda, Vietnam	NA	NA	NA	Medical records, provider interview	Children	699 (P)

*Denominator used to calculate the outcome (i.e., total number of patients evaluated [P] or total number of drug prescriptions [D]).

§Number of healthcare providers involved.

NA, not available; PNG, Papua New Guinea; SP, standardized patient.

Importantly, almost all the studies identified through our systematic review only assessed drug prescription and did not account for direct dispensing of unlabeled medicines, which is likely a common practice [75]. This may underestimate the true antibiotic prescribing proportion.

Study quality

Fig 2 displays the summary of the risk of bias assessment, while the individual studies’ quality assessment results are presented in S3 Table. The overall risk of study bias was scored as high for 21/48 studies (43.8%), moderate for 11 (22.9%), and low for 16 (33.3%). The proportion of studies assigned to the high risk group was higher among those conducted in low- and lower-middle-income countries (14/28; 50%) and lower among those performed in upper-middle-income countries (7/19; 36.8%). No major changes were observed in terms of overall study quality over time, although this could be due to the limited number of studies. In general, the biggest issues were observed with regards to external validity: Some form of random sampling or a census was seldom performed, and the study population was rarely representative of the target, mostly due to the fact that prescriptions were often selected from one or a few facilities in circumscribed areas. The case definition was considered inadequate for studies that did not record clinical details about patients receiving prescriptions. The risk of bias concerning the data collection method was deemed to be low for studies that used medical records or similar sources to retrieve prescription information. This choice was made based on the fact that medical records and drug prescription audits constitute good sources to estimate the proportion of antibiotic prescribing, although they are generally poorly suited for an accurate evaluation of appropriateness of prescription. On the other hand, studies using patient or provider questionnaires were considered at high risk of bias given the potential for recall bias and Hawthorne effect [76,77].

Fig 2

Summary of study risk of bias assessment.

Prevalence of antibiotic prescription

Among the 21 studies that reported the total number of patients attending a certain facility at the time of data collection [27,28,30,32,36,38,39,41,42,48,53-56,59,62,65,68,70,72,73], the average proportion of individuals receiving an antibiotic prescription ranged widely, from 19.6% (95% CI: 14.0%–26.4%) to 90.8% (95% CI: 89.3%–92.0%) [27,54]. Among the 27 studies in which the denominator was the total number of drug prescriptions [29,31,33–35,37,40,43–47,49–52,57,58, 60,61,63,64,66,67,69,71,74], the proportion of prescriptions containing antibiotics varied between 17.8% (95% CI: 14.2%–21.9%) and 79.2% (95% CI: 74.4%–82.7%) [46,57]. We could not identify any specific pattern in the distribution of antibiotic prescription rates across levels of country income, partly due to small sample sizes. As very few studies were conducted solely in the private health sector, no comparisons could be made against public facilities. Similar considerations apply to the health service location (i.e., urban versus rural areas). Furthermore, we did not observe any specific variation over time in the proportion of patients receiving antibiotics, either overall or after stratifying by country income level.

Since almost all patient–provider encounters included in studies using patients as the denominator resulted in a treatment prescription, prevalence estimates can be considered comparable to those derived from the 27 studies using drug prescriptions as the denominator. The pooled proportion of patients who received antibiotics resulting from a meta-analysis of all studies was 52% (95% CI: 51%–53%), and both stratum-specific pooled proportions for studies using one or the other type of denominator were reasonably close to the overall estimate (Fig 3). As expected, very high levels of between-study heterogeneity were observed (I2 values were above 98% overall, in subgroup analyses, and in sensitivity analyses), thus limiting the reliability of our pooled estimates. However, the 95% prediction interval calculated in the primary analysis was quite narrow, ranging from 44% to 60%, indicating that a new potential observation in a similar setting would likely yield a proportion of patients receiving antibiotics close to 50%. The prediction interval is wider than the conventional confidence interval owing to the fact that it accounts for uncertainty about both the population mean and the distribution of values.

Fig 3

Forest plot of antibiotic prescription prevalence across all studies stratified by type of denominator used (i.e., either total number of patients or total number of drug prescriptions).

CI, confidence interval; ES, effect size; PNG, Papua New Guinea.

Subgroup analyses (e.g., after stratification by country income level, type of denominator, or type of patients examined) and sensitivity analyses yielded similar point estimates, but confidence and prediction intervals became much wider (S1–S4 Figs). Unsurprisingly, given the results of subgroup meta-analyses, none of the coefficients of our meta-regression models was statistically significant, and the overall model could only explain a negligible proportion of the observed heterogeneity (S4 Table).

Inappropriate antibiotic prescription

As previously mentioned, we recorded the proportion of inappropriate prescriptions when available in individual studies. In most cases, the authors made their judgment based on national and/or international guidelines for treatment of key conditions. Among the 9 studies that assessed the rationality of antibiotic prescriptions [36,39,46,53,55,62,64,65,67], the proportion judged inappropriate ranged widely, reflecting the significant differences in study designs as well as in the sets of criteria that were adopted to determine the outcome (Table 2). The lowest level of inappropriate prescription (7.9%; 95% CI: 4.6%–12.5%) was reported in a study conducted in Zambia that included 537 children aged <5 years presenting with an acute respiratory syndrome, of whom 37.6% (95% CI: 33.5%–41.9%) were given antibiotics [39]. All antibiotic prescriptions were classified as inappropriate in 3 studies: 2 of them employed standardized patients portraying conditions that did not require antibiotics such as common cold, watery diarrhea, presumptive tuberculosis, and chest pain indicative of angina, with an overall antibiotic prescription prevalence of about 41%–42% [53,65]; the other study was performed in China and included 9,340 drug prescriptions issued for children with acute respiratory tract infection of likely viral etiology, 36.6% (95% CI: 35.7%–37.6%) of whom received an antibiotic [67]. The proportion of inappropriate antibiotic prescriptions exceeded 50% in the remaining 5 studies.

Table 2

Main findings of studies that assessed inappropriate antibiotic prescription.

Study	Country	Country income	Healthcare sector	Sample size	Type of patients	Antibiotic prescriptions n (%; 95% CI)	Inappropriate antibiotic prescriptions n (%; 95% CI)
Beri (2013) [36]	India	Lower-middle	Private	400	Patients of all ages with any clinical presentation	315 (78.8; 74.4–82.7)	179 (56.8; 51.2–62.4)
Graham (2016) [39]	Zambia	Lower-middle	Not reported	537	Children under age 5 years with acute respiratory illness	202 (37.6; 33.5–41.9)	16 (7.9; 4.6–12.5)
Sarwar (2018) [46]	Pakistan	Lower-middle	Public	6,400	Patients with any clinical presentation	5,069 (79.2; 78.2–80.2)	4,238 (83.6; 82.6–84.6)
Gasson (2018) [55]	South Africa	Upper-middle	Public	654	Patients with any clinical presentation	449 (68.7; 64.9–72.2)	305 (67.9; 63.4–72.2)
Sánchez Choez (2018) [62]	Ecuador	Upper-middle	Public	1,393	Patients of all ages with upper respiratory tract infection	523 (37.5; 35.0–40.1)	472 (90.2; 87.4–92.7)
Wang (2014) [64]	China	Upper-middle	Public	7,311	Patients of all ages with any clinical presentation	3,868 (52.9; 51.8–54.1)	2,344 (60.6; 59.0–62.1)
Alabid (2014) [53]	Malaysia	Upper-middle	Private	140	Adult SPs with common cold	58 (41.4; 33.2–50.1)	58 (100)
Xue (2019) [65]	China	Upper-middle	Public	526	Adult and child SPs with 1 of the following: diarrhea (viral gastroenteritis), chest pain (suspicious for angina), fever and cough (presumptive TB)	221 (42.0; 37.8–46.4)	221 (100)
Zhang (2017) [67]	China	Upper-middle	Public	9,340	Children with upper respiratory tract infection	3,425 (36.7; 35.7–37.7)	3,425 (100)

CI, confidence interval; SP, standardized patient; TB, tuberculosis.

Information regarding individual antibiotics was available from 16 studies in 15 countries. Of note, 11 of these studies included patients seeking care for any reason, while the remaining 5 studies focused on a specific condition (i.e., respiratory tract infection [4 studies] or diarrhea [1 study]) (Table 3). Access-group antibiotics accounted for the majority of prescriptions (more than 60%) in 13 studies from 12 countries, whereas Watch-group antibiotics accounted for high proportions of prescriptions among studies from Mexico (90.3%; 95% CI: 88.8%–91.7%), China (78.4%; 95% CI: 75.7%–81.0%), and Pakistan (47.8%; 95% CI: 46.5%–49.1%) (Table 3) [46,54,63].

Table 3

AWaRe classification of antibiotic prescriptions in a subset of studies included in analysis.

Study, total number (n) of antibiotics prescribed or dispensed	Country	Patients’ clinical presentation	Access-group antibiotics (%)	Watch-group antibiotics (%)	Reserve-group antibiotics (%)	Discouraged antibiotics (%)
Abdulah (2019) [31], n = 2,389	Indonesia	Any	1,667 (69.8)	287 (12.0)	NA	NA
Sarwar (2018) [46], n = 5,853	Pakistan	Any	3,055 (52.2)	2,798 (47.8)	0	0
Sánchez Choez (2018) [62], n = 553	Ecuador	Acute respiratory syndrome	463 (83.7)	90 (16.3)	0	0
Worku (2018) [29], n = 553	Ethiopia	Any	431 (77.9)	122 (22.1)	0	0
Gasson (2018) [55], n = 519	South Africa	Any	361 (69.6)	158 (30.4)	0	0
Chem (2018) [37], n = 12,350	Cameroon	Any	11,109 (90.0)	1,241 (10.0)	0	0
Mashalla (2017) [58], n = 289	Botswana	Any	240 (83.0)	49 (17.0)	0	0
Ab Rahman (2016) [59], n = 6,009	Malaysia	Any	3,879 (64.6)	2,073 (34.5)	NA	NA
Adisa (2015) [32], n = 303	Nigeria	Any	224 (73.9)	61 (20.1)	0	18 (5.9)
Yebyo (2016) [30], n = 373	Ethiopia	Acute respiratory syndrome	312 (83.6)	61 (16.4)	0	0
Ndhlovu (2015) [42], n = 561	Zambia	Any	490 (87.3)	42 (7.5)	0	0
Sun (2015) [63], n = 978	China	Acute respiratory syndrome	174 (17.8)	767 (78.4)	NA	NA
Bielsa-Fernandez (2016) [54], n = 1,718	Mexico	Diarrhea	166 (9.7)	1,551 (90.3)	1 (0.06)	0
Mukonzo (2013) [27], n = 9,683	Uganda	Any	7,735 (79.9)	1,908 (19.7)	NA	NA
Nepal (2020) [73], n = 479	Nepal	Any	299 (62.4)	165 (34.4)	NA	NA
Mekuria (2019) [72], n = 13,646	Kenya	Acute respiratory syndrome	8,461 (62.0)	4,880 (35.7)	NA	278 (2.0)

Denominator for percentage calculations is the total number of antibiotics dispensed/prescribed. Access-group antibiotics are first-line and narrow-spectrum agents such as penicillin, amoxicillin, and trimethoprim-sulfamethoxazole. Watch-group antibiotics are broad-spectrum agents with higher resistance selection such as second- and third-generation cephalosporins, and fluoroquinolones. Reserve-group antibiotics include last-resort antibiotics such as colistin. Discouraged antibiotics are fixed-dose combinations such as ciprofloxacin/ornidazole.

NA, not available.

Discussion

To our knowledge, this is the first comprehensive analysis of antibiotic prescriptions in primary care in LMICs. We found that the proportion of patients seeking care for any reason who were prescribed antibiotics in this context often exceeded 50%. Although the interpretation of our pooled estimates is limited by the considerable between-study heterogeneity, values were quite consistent across settings. Available studies from LMICs often suffer from several methodological issues and report scanty details concerning patients’ clinical features that would help accurately judge the appropriateness of prescription. The number of health facilities involved in individual studies is often very small, particularly in low-income countries (a total of 13 facilities across 4 studies that reported this information), indicating major discrepancies in the quality of information among geographic areas. Although all included studies examined prescription data in primary care facilities, we recognize that primary care entails a wide range of facility types, each with its own peculiarities and challenges. This variegated scenario prevented us from conducting specific subgroup analyses that could inform targeted antibiotic stewardship strategies. Two studies, both conducted in an Iranian province, had a very large sample size because prescription details were captured through an electronic data collection system that is available nationwide. However, clinical information on patients receiving each prescription is much more challenging to obtain from this system, thus hindering a thorough assessment of inappropriate drug use.

WHO recommends that the proportion of patients receiving antibiotics in an outpatient setting should be less than 30% [78]. However, this threshold was established somewhat arbitrarily more than 2 decades ago, due to a lack of evidence on prescription practices and actual needs according to patients’ clinical features. If accurate and nationally representative prescribing data were available for individual countries, these could be used as a benchmark to define condition-specific ideal prescribing proportions that account for context-related variables.

High infectious disease burden in LMICs could potentially explain the high prevalence of antibiotic use; however, our results raise concerns about potential misuse of antibiotics based on a subset of studies that assessed the rationality of antibiotic prescriptions. For example, high levels of antibiotic prescriptions (41%–42%) were reported in 2 standardized patient studies in Malaysia and China, where nobody should have received antibiotics, by design [53,65]. In a study conducted in Mexico, 69% of patients had had watery diarrhea for less than 48 hours, but almost everybody received antibiotics instead of rehydration alone [54]. Similarly, in a nationwide health facility survey in Zambia, 72.2% of patients met the criteria for suspected malaria, for which antibiotics are not appropriate treatment, but nonetheless more than half were given antibiotics [42]. Studies focused on individuals with upper respiratory symptoms such as common cold or pharyngitis reported unacceptably high antibiotic prescribing proportions, ranging from 36.7% to 55.3% [39,62,63,67].

To promote the optimal use of antibiotics and assist antibiotic stewardship efforts, WHO introduced the AWaRe classification in 2017 [21]. The classification underlines that, where appropriate, narrow-spectrum antibiotics included in the Access group should be preferred over broad-spectrum antibiotics from Watch and Reserve groups in order to limit the selection and spread of antibiotic resistance. Accordingly, WHO recommends that Access-group antibiotics should constitute at least 60% of overall antibiotic use [21]. Only 16 of the 48 studies identified through our systematic review reported detailed information on individual antibiotic drugs, and all but 3 had at least 60% of antibiotics being from the Access group [21]. Three studies with a high proportion of Watch-group antibiotics were from Mexico, China, and Pakistan; however, we cannot generalize these estimates to overall antibiotic consumption in these countries based on only 1 study in each country. Interestingly, a recent study that analyzed pediatric antibiotic sales data using AWaRe categories in 70 countries showed a high proportion of Watch-group antibiotics in China, Pakistan, and Mexico [79].

A recently published umbrella review on antibiotic use for adults in primary care (though focused on dental care) identified several factors that appear to affect prescribing behaviors in HICs, such as socio-cultural context, financial incentives, personal beliefs, patients’ attitudes, and AMR awareness [80]. Similar considerations likely apply to prescription practices in LMICs, although a deeper understanding of underlying determinants remains challenging. Among the biggest issues is the poor documentation of clinical reasons leading to antibiotic prescription, as observed in other settings [81]. Reaching a definitive diagnosis is often a huge challenge in resource-constrained areas, where point-of-care diagnostic tests for the most common conditions observed in primary care are frequently lacking [82].

Along with potential antibiotic misuse, therapeutic schemes may be inappropriate because of inadequate choice of antibiotic or incorrect dose or duration. However, a thorough assessment of prescription practices that includes such considerations is made particularly difficult by the variability in national treatment guidelines regarding antibiotic regimens [83]. In an attempt to foster the harmonization of such guidelines and minimize differences across countries, WHO recently released antibiotic treatment guidelines for 26 common infectious syndromes encountered in primary care and inpatient settings [84]. These guidelines currently indicate when and what antibiotics should be prescribed, and further work on harmonizing dose, duration, and formulation is ongoing [21].

In summary, the pooled estimate of antibiotic prescription in primary care settings across LMICs was 52%, but there was significant between-study heterogeneity. Further, the true extent of misuse was hard to discern, given the lack of data on appropriateness and the low quality of studies included. Future studies should use methodologies such as standardized patients, where the diagnosis is fixed by design, or include thorough laboratory testing to match diagnoses with antibiotic use. Accurate prescription audit tools are difficult to implement in most LMICs owing to the limited availability of electronic records. Also, the paucity of clinical details that can be captured through medical records (paper-based or not) makes it even harder to determine the appropriateness of prescription [85].

There is a need for better quality data to accurately measure the magnitude of antibiotic prescribing and dispensing by healthcare professionals at the primary care level accounting for local epidemiologic patterns. Global burden of disease data [86] combined with nationally representative AMR surveillance data [87] could be utilized to estimate the amount and type of antibiotics needed in a country, which could then be compared with existing national antibiotic consumption databases [6]. Meanwhile, LMICs should adapt the WHO infection treatment guidelines and incorporate the AWaRe categorization into their national antibiotic treatment guidelines to improve antibiotic prescribing. This will help countries to prioritize surveillance and stewardship efforts aimed at curbing the spread of AMR and preserving the efficacy of currently available antibiotics.

Dataset used for analyses.

(XLSX)

Click here for additional data file.

Forest plot of proportion of patients receiving antibiotics, restricted to studies including patients seeking care for any reason.

(TIF)

Click here for additional data file.

Forest plot of proportion of patients receiving antibiotics stratified by country income level (LIC = low-income country; LMIC = lower-middle-income country; UMIC = upper-middle-income country).

(TIF)

Click here for additional data file.

Forest plot of proportion of patients receiving antibiotics, including all studies except those conducted in Iran.

(TIF)

Click here for additional data file.

Forest plot of proportion of patients receiving antibiotics, excluding studies whose overall risk of bias was scored as “high.”

(TIF)

Click here for additional data file.

(DOC)

Click here for additional data file.

(PDF)

Click here for additional data file.

World Bank criteria for the definition of countries’ income level 2010–2018.

(DOCX)

Click here for additional data file.

Risk of bias assessment tool (adapted from Hoy et al. [18]).

(DOCX)

Click here for additional data file.

Risk of bias assessment of all studies included in final synthesis.

(DOCX)

Click here for additional data file.

Results of meta-regression analysis.

(DOCX)

Click here for additional data file.

Search strategies employed.

(DOCX)

Click here for additional data file.

Selection process and data extraction.

(DOCX)

Click here for additional data file.

29 Jan 2020

Dear Dr Pai,

Thank you for submitting your manuscript entitled "Antibiotic prescription practices in primary care in low- and middle-income countries: a systematic review and meta-analysis" for consideration by PLOS Medicine.

Your manuscript has now been evaluated by the PLOS Medicine editorial staff [as well as by an academic editor with relevant expertise] and I am writing to let you know that we would like to send your submission out for external peer review.

However, before we can send your manuscript to reviewers, we need you to complete your submission by providing the metadata that is required for full assessment. To this end, please login to Editorial Manager where you will find the paper in the 'Submissions Needing Revisions' folder on your homepage. Please click 'Revise Submission' from the Action Links and complete all additional questions in the submission questionnaire.

Please re-submit your manuscript within two working days, i.e. by .

Login to Editorial Manager here: https://www.editorialmanager.com/pmedicine

Once your full submission is complete, your paper will undergo a series of checks in preparation for peer review. Once your manuscript has passed all checks it will be sent out for review.

Feel free to email us at plosmedicine@plos.org if you have any queries relating to your submission.

Kind regards,

Adya Misra, PhD,

Senior Editor

PLOS Medicine

6 Mar 2020

Dear Dr. Pai,

Thank you very much for submitting your manuscript "Antibiotic prescription practices in primary care in low- and middle-income countries: a systematic review and meta-analysis" (PMEDICINE-D-20-00248R1) for consideration at PLOS Medicine.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with an academic editor with relevant expertise, and sent to independent reviewers, including a statistical reviewer. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Mar 27 2020 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Adya Misra, PhD

Senior Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

Abstract- the last sentence of the methods and findings section should include a limitation of your study design

Abstract- perhaps structure abstract according to PRISMA guidelines

Literature search- Please update this to current date, as we are coming up to a year since the last search

At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary

Reviewers have noted the definition of primary care must be defined for clarity, please include this in the introduction and methods section

PRISMA checklist- We ask that you include a sentence in your methods section that the study followed PRISMA guidelines and that a completed checklist has been provided as SI file xx. In addition, please use paragraphs and sections instead of page numbers as they are likely to change

Methods- if all study types were included in the search this sentence is confusing “We performed a systematic review of cross-sectional studies”. Please correct and clarify as needed

Please provide a rationale for excluding studies from “predatory” journals, especially as lists of predatory publishers have been deemed controversial and biased

Methods- please provide a brief summary of how antibiotic prescriptions were deemed to be inappropriate, providing references to tools/checklists as needed

The conclusions need to be toned down, since the reliability of the pooled estimates is low owing to high heterogeneity between included studies

Comments from the reviewers:

Reviewer #1: This study reviews and meta-analyses antibiotic prescription practices in primary care in low- and middle income countries. While it is an interesting topic, I do think the manuscript should be improved before it would be acceptable for publication.

1. Abstract. It is unclear why the authors focus on the prevalence of antibiotic prescriptions. Prevalence gives the proportion of a population who have a specific characteristic in a given time period, e.g. the proportion of the population that is using antibiotics at a given day. It would be more interesting to look at the incidence of antibiotic prescriptions or e.g. of those presenting with symptom x, what proportion of patients gets an antibiotic.

2. I don't understand why the authors try to model a pooled estimate across all years and countries. Surely there are strong trends among prescribing trends in LMIC (which probably also differ per region), as also highlighted by the authors in the introduction. Thus the pooled estimate is impossible to interpret and highly dependent on the number of studies and samples sizes in different periods.

3. Abstract. The proportion of prescriptions ranged from 8 to 100%. Surely this 100% is not correct, a number that is obtained using questionable methods. Given this, please also add a comment on the quality of the included studies.

4. Introduction. 'An estimated 10 million deaths per year are expected to….' No, they are not expected to die, there are many problems with the report that came up with this number. See, for example https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5127510/

5. Introduction. 'The high level of antibiotic consumption in LMICS…' Is it hight, or is it just coming closer to HIC?

6. 'Most studies investigating the magnitude and determinants of antibiotic use have been focused on high-income countries (HIC) and particularly on hospital settings.' While the first claim is true, there are many studies estimating the magnitude and determinants of (inappropriate) antibiotic use in primary care/outpatient setting, e.g.

https://www.bmj.com/content/364/bmj.l440.full

https://academic.oup.com/jac/article/73/suppl_2/ii36/4841818

https://www.sciencedirect.com/science/article/pii/S2589537018300531

https://academic.oup.com/jac/article/73/suppl_2/ii27/4841819

https://www.bmj.com/content/362/bmj.k3155

https://www.tandfonline.com/doi/full/10.1080/02813432.2017.1288680

In fact, in the UK we probably have a better understanding of prescribing practice than in the hospital setting.

Therefore, the sentence of paucity of information of antibiotic use in outpatient primary healthcare facilities, only applies potentially to LIMC.

7. Related to above, if there truly is a paucity of information, there is no point in performing a systematic review. In that case it may be better to invest time in original research and collecting new data.

8. Please clarify why papers published in predatory journals were excluded without looking at the actual quality of the paper. Wouldn't it be possible that a useful high-quality paper would be published in one of these journals, especially if open access fees may be lower than for other journals for authors from LMIC?

9. '…as urban any site with a minimum population of 2,000 inhabitants..' This is per what? Per square mile/km? If it is 2000 persons in a very large geographical area, that area shouldn't be classified as urban.

10. I have my doubts about the search term 'primary healthcare'. I think this term wouldn't pick up many papers that were performed in primary care/outpatient/community setting. It is also unclear why the authors haven't used MeSH terms as well.

11. When discussing the reason for limiting to studies preformed >=2010, the authors should also add that in contrast to the effect of interventions which can usually be estimated using studies from different years, meta-analysing something that is almost guaranteed to chance over time, especially given the knowledge we have about trends in antibiotic prescribing, it doesn't make even sense to include relatively old studies in a pooled estimate.

12. Please specify missing % for the different variables

13. Summarising inappropriateness proportions across very different conditions, that likely have different 'ideal' prescribing proportions, e.g. see https://academic.oup.com/jac/article/73/suppl_2/19/4841820

https://academic.oup.com/jac/article/73/suppl_2/ii11/4841821, isn't very informative. Separate estimates should be provided for different conditions.

14. The study seems to be comparing apples and oranges with 8 studies including only subjects presenting with one pre-specified condition (with likely different ideal prescribing proportions see comment above) and other studies that did not apply any restrictions on reason for seeking care. These studies are not comparable and shouldn't be combined.

15. Why report confidence intervals for study bias?? I don't think confidence intervals make sense to report here, it would be sufficient to just report the percentages without CI.

16. I don't understand why the author mention a formal comparison of proportion of high-risk studies across different types of countries? Why would this be of interest?

17. '…. And the study population was rarely representative of the target….' The target of the current study, or the target of the original study?

18. Please add a reference for the Hawthorne effect.

19. 'the proportion of prescriptions containing antibiotics varied between 17.8% .. and 79%..' This is so dependent on other drugs and indications, hardly informative at all…

20. 'we could not identify any specific pattern in the distribution of antibiotic rates across levels of country income'. Could be added that this may be due to small sample sizes.

21. I2 values are above 98%, again confirming that one may be comparing oranges with apples. Simply using a random effects model and providing prediction intervals doesn't overcome this problem. The sentence 'indicating that a new potential observation in a similar setting would likely yield a proportion of patients receiving antibiotics close to 50%' is therefore also problematic.

22. All antibiotic prescriptions were classified as inappropriate in three studies. This is for the Chinese study likely based on inappropriate methodology to determine inappropriateness. E.g. RTI with a likely viral etiology, can still include a certain proportion of patients that would legitimately receives antibiotics, e.g. see https://academic.oup.com/jac/article/73/suppl_2/19/4841820 & https://academic.oup.com/jac/article/73/suppl_2/ii11/4841821

23. 'however Watch-group antibiotics accounted for high proportions among Mexico (90.3%). This extremely high percentage doesn't really line up with other studies (on restrictive populations): https://www.ncbi.nlm.nih.gov/pubmed/30522834. Of course this is a different population, but I find it at least surprising if use of Watch antibiotics would be more common among less severely ill patients in Mexico. Furthermore, the 90% is also way off the % estimated in another recent study: https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(18)30547-4/fulltext

24. 'The WHO recommends that the proportion of patients receiving antibiotics in an outpatient setting should be less than 30% [68]. However, this threshold was estabilished somewhat arbitrarily more than two decades ago, due to the lack of evidence on prescription practices and actual needs according to patient's clinical features.' There are examples in UK and EU where condition-specific ideal prescribing proportions were estimated (and compared with real prescribing data):

https://academic.oup.com/jac/article/73/suppl_2/19/4841820 & https://academic.oup.com/jac/article/73/suppl_2/ii11/4841821

https://qualitysafety.bmj.com/content/20/9/764

25. 'A recent published umbrella review'. Please specify that this focused on dental care.

26. 'or include thorough laboratory testing to match diagnosis with antibiotic use'. This would likely result in unfair comparisons as such testing is typically not available (not even in HIC setting) and in face of uncertainty it would be acceptable that a certain proportion of patients would be treated with antibiotics even if it would not be correct according to strict laboratory tests.

Reviewer #2: Interesting review, I have a few comments regarding the selection of the studies:

1)The authors said "Conference proceedings and abstracts, commentaries, editorials, reviews, mathematical modelling studies, economic analyses, qualitative studies, and studies published in predatory journals as defined by Beall [13] were excluded": Why is the rationale to exclude Conference proceedings and abstracts? sometimes these studies never get a full manuscript, because the researchers didn't have enough time, so they publish in conference abstract book. Of course, these studies have higher risk of bias and low quality, but can provide extra-evidence. Similar situation are papers from predatory journals (n=5): why not evaluate them and report the quality of them? Sometimes some researchers (specially at LMICs) didn't know the problem regarding predatory journals.

2)Figure 2: I'm a bit surprise about the low risk of "Reliability & validity of method used for measuring prescriptions" and "avoidance of inappropriate exclusions", because one of the most frequent data source were medical records and, in general, the quality from these records is not too high. I suggest to discuss a bit about it.

3)Figure 3: I suggest to don´t present an overall OR, at least you can present in sub-groups based in the type of denominator. I suggest only present as was presented in the supplementary material, because the study population is different

4)Meta-regression and heterogeneity: I suggest to evaluate the studies used for meta-regression, maybe only use studies with similar denominator

Reviewer #3: This is an important study that has the potential to contribute in informing policy and practice in tackling antimicrobial resistance as it attempted to analyse prescription practice in LMICs. It is properly conducted and very well written. The dearth and poor quality of the studies in these settings as demonstrated by the review limits the drawing of any strong conclusion as stated by the authors on the practice and appropriateness of antibiotic prescriptions. More importantly, the interpretation of such results should be carefully crafted and contextualized as every year inadequate access to antibiotics kills nearly 6 million people, including a million children who die of preventable sepsis and pneumonia mostly in LMICs. Therefore, interpretation of the data requires caution in the face of this reality

Major comments:

1. The operational definition of primary care as given by the authors is the foundation of the review. However, the definition is bundled into non-specific categorisation that its application limits ultimate utility of the study for any meaningful policy and practice implications. The level of tier of the health system (e.g. referral/tertiary/provincial/district hospital; health center/polyclinic/health station/clinic) in a country determine and regulate the system that determine prescription practice (e.g. the cadre of health worker; type and class of medicines/antibiotics should be available and used). Therefore, presenting the data by specific health facility would help to inform any potential antimicrobial/antibiotic stewardship programmes in LMICs and enhance the relevance and utility of the study.

2. The WHO AWaRe classification of antibiotics as the authors mentioned is to assist antibiotic stewardship efforts and promote optimal use so as to prevent the development of drug resistance. The classification underline that the narrow-spectrum antibiotics in the Access group are the preferred treatment option for most infections (e.g Respiratory tract) and are also thought to have a lower ecologic impact regarding the selection and spread of antibiotic resistance than broader-spectrum agents. Therefore, Access group antibiotics should therefore constitute the majority of antibiotic use in the outpatient setting and overall. As part of the outcome measurement of the current Global Programme of Work of WHO, countries should strive to ensure that Access group antibiotics constitute more than 60% of the overall antibiotic use. Authors should frame their analysis, presentation and interpretation likewise.

3. While recognising the paper will undergo statistical review, I wonder on the value of the pooled estimates.

Minor comments:

* Authors' use of 'community' level in the paper has to be clarified or removed.

* Clarify and correct use of antimicrobial vs antibiotic in the text

* Ref 68 is obsolete and need to be removed

Reviewer #4: I confine my remarks to statistical aspects of this paper.

These were well done and I recommend publication.

Peter Flom

Any attachments provided with reviews can be seen via the following link:

[LINK]

26 Mar 2020

Submitted filename: Response to reviewers.docx

Click here for additional data file.

22 Apr 2020

Dear Dr. Pai,

Thank you very much for re-submitting your manuscript "Antibiotic prescription practices in primary care in low- and middle-income countries: a systematic review and meta-analysis" (PMEDICINE-D-20-00248R2) for review by PLOS Medicine.

I have discussed the paper with my colleagues and the academic editor and it was also seen again by three reviewers. I am pleased to say that provided the remaining editorial and production issues are dealt with we are planning to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

Our publications team (plosmedicine@plos.org) will be in touch shortly about the production requirements for your paper, and the link and deadline for resubmission. DO NOT RESUBMIT BEFORE YOU'VE RECEIVED THE PRODUCTION REQUIREMENTS.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by Apr 29 2020 11:59PM.

Sincerely,

Adya Misra, PhD

Senior Editor

PLOS Medicine

plosmedicine.org

------------------------------------------------------------

Requests from Editors:

Abstract

My apologies for the ambiguous request regarding abstract structure. We ask that you include the relevant information provided here within the PLOS Medicine abstract subheadings of background, methods and findings, conclusions. Therefore, the background section should contain the aim of your study whereas the remainder of the information should be provided within methods and findings, followed by conclusions. The last sentence of the methods and findings section should include the limitations of your study.

Author summary

Lines 101-102 require some clarification since the access group is not a widely used term. Could you please simplify or introduce the AWaRe classification in the author summary?

Line 347 could you use an alternative term to “all-comers”

Please update the bibliography in Vancouver style

PRISMA checklist- please use sections and paragraphs instead of page numbers as these are likely to change during publication

Comments from Reviewers:

Reviewer #1: Virtually all comments are addressed.

However, the sentence 'Only nine studies assessed rationality, and the proportion of inappropriate prescription ranged from 8% to 100%.' from the abstract is still a bit misleading.

Could the authors please add that this is for various specific conditions? There is unlikely a country in the world where antibiotics are only prescribed inappropriately, for specific symptoms/conditions yet, but not for all conditions/symptoms.

Reviewer #3: The points I raised are adequately addressed or explained. I have no further comment

Reviewer #4: I confine my remarks to statistical aspects of this paper.

I had already approved it, so I recommend publication

Peter

Any attachments provided with reviews can be seen via the following link:

[LINK]

8 May 2020

Dear Dr. Pai,

On behalf of my colleagues and the academic editor, Dr. Margaret Kruk, I am delighted to inform you that your manuscript entitled "Antibiotic prescription practices in primary care in low- and middle-income countries: a systematic review and meta-analysis" (PMEDICINE-D-20-00248R3) has been accepted for publication in PLOS Medicine.

PRODUCTION PROCESS

Before publication you will see the copyedited word document (in around 1-2 weeks from now) and a PDF galley proof shortly after that. The copyeditor will be in touch shortly before sending you the copyedited Word document. We will make some revisions at the copyediting stage to conform to our general style, and for clarification. When you receive this version you should check and revise it very carefully, including figures, tables, references, and supporting information, because corrections at the next stage (proofs) will be strictly limited to (1) errors in author names or affiliations, (2) errors of scientific fact that would cause misunderstandings to readers, and (3) printer's (introduced) errors.

If you are likely to be away when either this document or the proof is sent, please ensure we have contact information of a second person, as we will need you to respond quickly at each point.

PRESS

A selection of our articles each week are press released by the journal. You will be contacted nearer the time if we are press releasing your article in order to approve the content and check the contact information for journalists is correct. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact.

PROFILE INFORMATION

Now that your manuscript has been accepted, please log into EM and update your profile. Go to https://www.editorialmanager.com/pmedicine, log in, and click on the "Update My Information" link at the top of the page. Please update your user information to ensure an efficient production and billing process.

Thank you again for submitting the manuscript to PLOS Medicine. We look forward to publishing it.

Best wishes,

Adya Misra, PhD

Senior Editor

PLOS Medicine

plosmedicine.org