Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Antibiotic overuse in the primary health care setting: a secondary data analysis of standardised patient studies from India, China and Kenya

Antibiotic overuse in the primary health care setting: a secondary data analysis of standardised... Original research Antibiotic overuse in the primary health care setting: a secondary data analysis of standardised patient studies from India, China and Kenya 1,2 3 4 5 Giorgia Sulis , Benjamin Daniels , Ada Kwan , Sumanth Gandra, 6,7 3,8 1,2,9 Amrita Daftary , Jishnu Das , Madhukar Pai To cite: Sulis G, Daniels B, ABSTRACT INTRODUCTION Kwan A, et al. Antibiotic Introduction Determining whether antibiotic prescriptions Antibiotic stewardship is critical for tackling overuse in the primary health are inappropriate requires knowledge of patients’ antimicrobial resistance (AMR), especially care setting: a secondary data underlying conditions. In low- income and middle- income in the context of the ongoing COVID-19 analysis of standardised patient countries (LMICs), where misdiagnoses are frequent, this is studies from India, China and pandemic. In a recent systematic review on challenging. Additionally, such details are often unavailable Kenya. BMJ Global Health antibiotic prescription practices in primary for prescription audits. Recent studies using standardised 2020;5:e003393. doi:10.1136/ care settings across low-income and middle- patients (SPs) offer a unique opportunity to generate bmjgh-2020-003393 income countries (LMICs), we showed that unbiased prevalence estimates of antibiotic overuse, as approximately 50% of patients of any age the research design involves patients with predefined Handling editor Seye Abimbola conditions. seeking care for any reason received at least JD and MP contributed equally. Methods Secondary analyses of data from nine SP one antibiotic. studies were performed to estimate the proportion of SP– However, determining inappropriate JD and MP are joint senior provider interactions resulting in inappropriate antibiotic prescription in LMICs is a challenge, and authors. prescribing across primary care settings in three LMICs a standardised tool for its assessment is (China, India and Kenya). In all studies, SPs portrayed Received 8 July 2020 currently unavailable. Inappropriate antibi- conditions for which antibiotics are unnecessary (watery Revised 1 August 2020 otic prescribing can derive from a range of diarrhoea, presumptive tuberculosis (TB), angina and Accepted 3 August 2020 failings: (1) prescription in the absence of asthma). We conducted descriptive analyses reporting overall prevalence of antibiotic overprescribing by clinical indication (ie, ‘overprescription’), healthcare sector, location, provider qualification and case. which not only produces zero benefit to the The WHO Access–Watch–Reserve framework was used to patient but can also be harmful (eg, drug categorise antibiotics based on their potential for selecting toxicities or costs for patients); (2) failure resistance. As richer data were available from India, we to prescribe antibiotics when necessary examined factors associated with antibiotic overuse in that (ie, ‘underprescription’); (3) suboptimal country through hierarchical Poisson models. antibiotic choice with respect to aetiology Results Across health facilities, antibiotics were given (confirmed or presumptive), site, severity inappropriately in 2392/4798 (49.9%, 95% CI 40.8% of infection and patient characteristics (eg, to 54.5%) interactions in India, 83/166 (50.0%, 95% CI 42.2% to 57.8%) in Kenya and 259/899 (28.8%, 95% CI age, comorbidities and pregnancy status); (4) 17.8% to 50.8%) in China. Prevalence ratios of antibiotic prescription of wrong dosage and/or dura- overuse in India were significantly lower in urban versus tion of antibiotic treatment as compared with rural areas (adjusted prevalence ratio (aPR) 0.70, 95% 3 4 national and international guidelines. CI 0.52 to 0.96) and higher for qualified versus non- Methods used to assess inappropriateness, qualified providers (aPR 1.55, 95% CI 1.42 to 1.70), and such as prescription audits, medical records for presumptive TB cases versus other conditions (aPR and patient exit interviews, have multiple 1.19, 95% CI 1.07 to 1.33). Access antibiotics were © Author(s) (or their 3 5 limitations. Electronic records are seldom predominantly used in Kenya (85%), but Watch antibiotics employer(s)) 2020. Re- use available in LMICs, particularly in primary (mainly quinolones and cephalosporins) were highly permitted under CC BY. Published by BMJ. prescribed in India (47.6%) and China (32.9%). care, thus making accurate prescription audit Conclusion Good- quality SP data indicate alarmingly For numbered affiliations see tools difficult to implement. Also, the paucity high levels of antibiotic overprescription for key conditions end of article. and variation of clinical details that can be across primary care settings in India, China and Kenya, captured through medical records (paper- Correspondence to with broad- spectrum agents being excessively used in based or not), if they even exist, makes it even Dr Madhukar Pai; India and China. madhukar. pai@ mcgill. ca harder to determine the appropriateness Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 1 BMJ Global Health opportunity to overcome the methodological limitations Key questions typical of other studies, thus making the assessment of inappropriateness of antibiotic use less biased and more What is already known? accurate. Because the underlying illness is prespecified, ► A recent systematic review and meta- analysis showed that, across the SP methodology allows accurate assessment if an anti- 48 studies from 27 low-income and middle- income countries biotic is inappropriately prescribed. The SP approach is including China, India and Kenya, approximately half of all pa- tients evaluated in outpatient primary care received an antibiotic not affected by poor recall, recall bias or the Hawthorne prescription. effect, which is commonly observed in patient exit ► Methods used to assess inappropriateness of antibiotic prescrip- interviews and direct observations of patient–provider tion, such as prescription audits, medical records and patient exit encounters. interviews, have multiple limitations. Considering the aforementioned advantages, we ► Standardised patients (SPs) offer a unique opportunity to explore performed a secondary analysis of prescription data prescribing practices and accurately estimate overprescription be- from previously conducted SP studies in three LMICs cause case presentations are fixed by design, thus allowing com- (India, China and Kenya) with two objectives: (1) to esti- parisons across settings and providers. mate the overall proportion of SP–provider interactions What are the new findings? (separately for pharmacy-based and health facility- based ► In this secondary analysis of data from nine SP studies carried out studies) that resulted in prescription or dispensing of at in India, Kenya and China, we provide a more unbiased prevalence least one antibiotic in the absence of clinical indication estimate of antibiotic overprescription for selected clinical condi- (ie, overprescription) and (2) to identify factors associ- tions (asthma, angina, watery diarrhoea, presumptive or confirmed ated with antibiotic overprescribing in health facilities. tuberculosis (TB)) across a range of primary healthcare providers. ► About 30% of SP–provider interactions in China and 50% of those performed in India and Kenya resulted in inappropriate antibiotic METHODS prescription. Study design and data sources ► Watch antibiotics (ie, broad-spectrum agents with higher potential for selecting resistance) were very commonly prescribed in India Data on SP–provider interactions (ie, completed SP visits (about 50%) and China (over 32%), and some patients (0.8%) even with a provider at a health facility or a pharmacy) from received last- resort antibiotics belonging to the ‘Reserve’ group. studies conducted by members of our team (India and ► In India, the average prevalence of antibiotic prescribing was 30% Kenya) or had used SP cases developed by our team or lower in urban versus rural areas, 55% higher among qualified obtained from publicly accessible sources (China) were providers compared with non- qualified ones and 19% higher for gathered to compile a pooled dataset for secondary anal- patients presenting with presumptive TB versus other conditions. 6–15 yses. The methods used are described in our published What do the new findings imply? manual and toolkit on how to conduct SP studies. ► Our findings indicate alarming levels of antibiotic overprescription Among studies carried out in India, four involved for conditions that are frequently encountered in primary care, po- primary health facilities across five sites (Delhi, Mumbai, tentially leading to toxic effects and diagnostic delays. Patna, three districts in the State of Madhya Pradesh, ► The choice of antibiotics given to patients is concerning, as several 6–9 and Birbhum district in the State of West Bengal), agents with high potential for resistance selection are often inap- while two were performed in pharmacies located in four propriately prescribed. different areas (Mumbai, Patna, Delhi and Udupi district ► The SP methodology could prove useful to further investigate anti- 10 11 of Karnataka). We also examined data from a pilot biotic prescribing practices and its underlying determinants, using study carried out in Nairobi (Kenya) and two studies other case presentations across a range of different contexts. completed in rural areas of China (Shaanxi, Sichuan and Anhui provinces), all involving only primary healthcare 12–15 providers. of prescription. Patient exit interviews are commonly Information regarding medications prescribed by used alternatives but come with several major drawbacks healthcare providers were collected in these published that can result in poor and inaccurate estimates that are SP studies but were not analysed in depth, especially with incomparable. Data collected in this manner are subject regard to inappropriate use. This is because, in most to recall bias, poor recall and limited clinical expertise instances, the primary publications focused on overall among patients. Further, not only are clinical presenta- quality of care, rather than the specific components of tions highly heterogeneous but also the difficulty in actu- care. ally determining what patients have makes comparisons very challenging for research. Provider selection in original studies A less biased method is the use of standardised patients Sampling approaches adopted in each primary study from (SPs), also known as ‘simulated’ or ‘mystery’ patients, that which our data were drawn are summarised in table 1. is, healthy individuals recruited from local communities For the two pharmacy- based studies, a random sample of and extensively trained to portray a standardised clinical pharmacies was selected from a comprehensive list of all 5 10 11 condition to a healthcare provider. Since their clinical those eligible obtained from relevant authorities. In presentations are fixed by design, SPs offer an important six of the other eight studies, healthcare providers were 2 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 3 Table 1 Main features of SP studies included in our analyses SP–provider Healthcare Facility Provider Provider Study site (year) interactions Tracer conditions sector location Provider selection approach consent participation* China (2013) 600 Angina, child Public Rural Census of all clinics designated under the New Cooperative Yes 100% diarrhoea Medical Scheme (ie, the major public health insurance programme in rural areas), followed by random selection of providers China (2015) 299 Presumptive TB Public Rural Census of all public providers followed by random sampling Yes 274/274 (100%) from one prefecture in each of 3 provinces out of a total of 47 prefectures, chosen to be representative of rural health systems Kenya (2014) 166 Angina, asthma, Public and Urban Non- random convenience sample designed to include low- Yes 46/49 (93.9%) child diarrhoea, private income, middle- income and high- income neighbourhoods in presumptive TB various Nairobi areas Madhya Pradesh, 1123 Angina, asthma, Public and Rural Census of all medical care providers working in 60 villages No Not applicable India (2010–2011) child diarrhoea private randomly sampled in three districts in Madhya Pradesh; all public providers and qualified private providers were automatically sampled; for each public provider, the closest private practitioner was also sampled Delhi, India (2014) 250 Presumptive and Private Urban Convenience sample (pilot study) Yes Not available confirmed TB, presumptive MDR- TB Mumbai and Patna, 2602 Presumptive and Private Urban Street- by- street mapping of private providers who were known No Not applicable India (2014–2015) confirmed TB, to see adult outpatients with respiratory symptoms, followed by presumptive MDR- random sampling stratified by provider qualification and private TB provider interface agency registration status Birbhum district, 823 Angina, respiratory Private Rural Census of private health providers who had been practising for at Yes 304/360 (84.4%) West Bengal, India distress, child least 3 years in 203 villages across Birbhum district (2012–2014) diarrhoea Mumbai, Patna and 1200 Presumptive TB, Pharmacies Urban Convenience sample of 54 pharmacies from 28 low- income No Not applicable Delhi, India confirmed TB localities in Delhi (pilot phase), random sampling of pharmacies in (2014–2015) Mumbai and Patna from a list of all pharmacies registered in the two cities Udupi district, 1522 For both adults Pharmacies Urban Of the 350 pharmacies registered in the district as per the local No Not applicable Karnataka, India and children: upper and rural pharmacy association, 279 were considered eligible for the (2018) respiratory tract study after excluding those operating inside hospitals (47), those infection, diarrhoea, permanently closed or under renovations (10), those that could presumptive malaria not be identified by the field team (4), those for veterinarian purposes only (1) and those used for SP training (10). *For studies in which provider consent was required. MDR- TB, multidrug resistant tuberculosis; SP, standardised patient; TB, tuberculosis. BMJ Global Health randomly sampled after performing a census or street- by- proportion of SP–provider interactions that resulted 7–9 13–15 street mapping in the study areas. A convenience in antibiotic prescription or dispensing. The overall sample of practitioners was selected in two pilot studies proportion of prescribed or dispensed antibiotics, along 6 12 respectively performed in Delhi and Nairobi. A waiver with ATC- class and AWaRe group- specific proportions, of provider consent was obtained in four out of nine was calculated across strata defined by key variables of studies, all carried out in India, two of which involved interest, such as healthcare sector (public/private), 7 9–11 pharmacies. In all the others, verbal or written facility location (urban/rural), provider qualification informed consent was sought at least 6 weeks prior to the (qualified/non- qualified, defined based on whether they commencement of SP–provider interactions in order to had at least a bachelor’s degree in medicine) and tracer reduce the risk of SP detection. Yet, participation rates conditions. For all prevalence proportions, we computed were very high (85%–100%) among eligible health 95% CIs using bootstrapping in order to account for clus- practitioners, and non- participation was usually due to tering at the study level. logistical issues on the day of the visits rather than active In order to examine the factors associated with anti- refusal to be involved in the project. Hence, it is reason- biotic prescribing in health facilities in India, we fit a able to expect negligible differences between participants hierarchical Poisson regression model that allows direct and non-participants, making non- response bias a minor estimation of adjusted prevalence ratios (aPRs) even if the concern. In all studies, SPs were randomly assigned to outcome is common as in this case. Our model included providers, and completion rates of SP–provider interac- a random intercept for studies and dummy variables for tions were always very high. facility location, healthcare sector, provider qualification and tracer conditions as predictors. As we anticipated a Tracer conditions fair amount of between- study heterogeneity, we decided Tracer conditions (ie, SP case presentations) were to opt for a mixed model that could better account for defined similarly across SP studies, thus allowing compar- it as compared with including the study or study site as a isons across settings. Cases ranged from presumptive or covariate. Among tracer conditions, only angina, asthma confirmed tuberculosis (TB) (which requires specific and presumptive TB could be included in order to avoid anti-TB treatment as per WHO recommendations) to sparse data problems (ie, violations of the positivity self- limiting infections, such as watery diarrhoea or upper assumption). The effect of all predictors was expected to respiratory tract illness (which only need support treat- be similar across studies, and therefore only fixed slopes ment, eg, rehydration therapy for diarrhoea), to non- were considered. These analyses were restricted to India communicable diseases like asthma or chest pain indica- because we had diverse and more data. We also consid- tive of angina (these should be referred to a higher level ered alternative models and examined the pros and cons of care). Importantly, none of such conditions requires of each. A full description of our analyses is provided in antibiotics, which means that any antibiotic prescribed to online supplementary file 1. SPs is deemed inappropriate by indication (ie, overpre- Data from pharmacies were not pooled because contexts scription). and tracer conditions were highly heterogeneous in the two available studies. Therefore, we only calculated prev- Outcome assessment alence proportions and 95% CIs of dispensed antibiotics, Raw data from original studies were harmonised and both overall and in stratified analyses. recoded as needed. We used the available information All analyses were performed using Stata 16. on medications that were prescribed or dispensed during each SP–provider interaction to categorise individual Patient and public involvement drugs. Antibacterial agents were further classified using It was not possible to involve patients or the public in the both the ATC (Anatomical–Therapeutic–Chemical) design, conduct, reporting or dissemination plans of our Index and the WHO Access–Watch–Reserve (AWaRe) research because this is a secondary analysis of previously 16 17 framework. Fixed- dose combinations (FDCs) of anti- conducted studies. biotics (eg, ciprofloxacin/ornidazole) were classified as ‘discouraged’ antibiotics as per WHO recommendations. The primary outcome measure was expressed as the proportion of SP–provider interactions that resulted RESULTS in antibiotic prescription or dispensing. Secondary The main features of SP studies that were included in outcomes were proportions of specific groups of antibi- our analyses are summarised in table 1. A total of 4798 otics that were prescribed or dispensed both overall and SP–provider interactions were completed in health across strata of key variables of interest. These propor- facilities across urban and rural India, predominantly tions provide a direct measure of antibiotic overuse. in the private sector. Both private and public health- care providers were involved in the pilot study carried Statistical analyses out in Nairobi (166 interactions), whereas studies from For studies carried out in health facilities, we conducted rural China only targeted the public sector (899 inter- country- level descriptive analyses and reported the crude actions). For these health facility-based studies, we first 4 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health present summary statistics and then report results from Discouraged antibiotics, that is, FDCs other than anti- our models. mycobacterial drugs (such as norfloxacin+tinidazole or ofloxacin+ornidazole) accounted for 12.1%, of which all but one were given for child diarrhoea. Anti- TB medi- Antibiotic overuse across settings cations represented 8.3% of antibiotics in India; almost In India, 2392 of 4798 (49.9%, 95% CI 40.8% to 54.5%) all of them were given by healthcare providers in urban SP–provider interactions resulted in at least one anti- areas; and none could be considered appropriate based biotic prescription (table 2). Similar proportions were on the expected correct management of such cases. observed in Nairobi (83 of 166; 50.0%, 95% CI 42.2% to About one- quarter of drugs prescribed in studies from 57.8%), while a lower percentage was found in the China China could not be categorised based on the AWaRe studies (259 of 899; 28.8%, 95% CI 17.8% to 50.8%). framework because only the drug class was reported. However, in the latter case, the CI was substantially wide, These were mainly cephalosporins, most likely second reflecting the considerable between-study variance due or higher generation, and therefore the overall propor- to differences in tracer conditions evaluated. tion of Watch-group antibiotics is expected to be greater In most instances, only one antibiotic was given than 32.9% (table 3). Undefined cephalosporins were during an individual SP–provider interaction; less than by far the most prescribed antibiotics in China (76/301, 5% of interactions across all settings resulted in two or 25.2%), followed by gentamicin (45/301, 15.0%), amox- more antibiotics prescriptions. Crude analyses of data icillin (37/301, 12.3%), erythromycin (26/301, 8.6%) from India indicate that antibiotic overprescription was and levofloxacin (18/301, 6.0%). more common among healthcare providers in urban Subgroup analyses of antibiotic prescription patterns areas, among those working in the private sector and among SP–provider interactions that took place in among qualified professionals. Furthermore, antibiotics Nairobi were limited by the small sample size. However, were largely overprescribed to patients presenting with 85.4% (76/89) of all antibiotics prescribed were first-line a diverse range of clinical conditions in all countries and narrow- spectrum agents from the ‘Access’ group, (figure 1). In India, the percentage of subjects receiving while the remaining belonged to the ‘Watch’ group. antibiotics was close to 50% for most case types, with a peak of 59.4% (95% CI 50.5% to 75.0%) among child Factors associated with antibiotic overuse in India diarrhoea cases. However, for angina cases, it was 19.2% Prevalence ratios of antibiotic overuse and their 95% (95% CI 16.8% to 21.1%). About half of the visits for CIs estimated through mixed-effects Poisson regression presumptive TB in China received antibiotics inappro- analysis are reported in figure 2. The adjusted preva- priately, as opposed to 9.2% (95% CI 5.9% to 12.4%) of lence of antibiotic prescribing was lower in urban versus visits for suspicious angina and 27.4% (95% CI 21.8% to rural areas (aPR=0.70; 95% CI: 0.52 to 0.96), for subjects 32.5%) for child diarrhoea. Case- specific estimates from presenting with suspicious angina (aPR=0.33; 95% CI: Nairobi are highly imprecise due to the small sample size. 0.27 to 0.40), and asthma (aPR=0.77; 95% CI: 0.66 to 0.89). Patients with presumptive TB were more likely to Type of antibiotics used receive inappropriate antibiotics (aPR=1.19; 95% CI: 1.07 Across studies performed in India, 2768 antibiotics were to 1.33) as compared with individuals with other clinical given to 2392 patients. The top 10 most prescribed antibi- conditions. Qualified practitioners were more likely to otics across SP–provider interactions in India were azith- prescribe antibiotics than non-qualified ones (aPR 1.55; romycin (381, 13.8%), amoxicillin+beta-lactamase inhib - 95% CI: 1.42 to 1.70). itor (344, 12.4%), amoxicillin (264, 9.5%), levofloxacin The hierarchical Poisson model did not show any signif- (202, 7.3%), cefixime (198, 7.2%), ofloxacin (165, 6.0%), icant difference between public and private providers, ofloxacin+ornidazole (150, 5.4%), norfloxacin+tinida- but this is in contrast with what emerged from alternative zole (136, 4.9%), ciprofloxacin (102, 3.7%) and cefpo- models as described in online supplementary file 1. doxime (88, 3.2%). Broad-spectrum agents with higher potential for selecting resistance (Watch antibiotics) were Antibiotic dispensing in pharmacies disproportionately represented (47.6%, 95% CI 26.8% to Our secondary analysis of data from two pharmacy- 54.0%), and even more so in urban areas (54.9%, 95% CI based SP studies showed that over-the- counter antibiotic 54.9% to 55.4%) (table 3). This reflects the heavy use of dispensing is also a common problem in various parts of quinolones, cephalosporins and macrolides that respec- India (table 4). tively accounted for 18.8% (95% CI 16.6% to 24.2%), In Udupi district (Karnataka state) the proportion of 13.0% (95% CI 8.2% to 14.6%) and 15.4% (95% CI 4.1% SP—pharmacist interactions that resulted in antibiotic to 19.3%) of all antibiotics prescribed in India. Nearly dispensing was 3.6% (95% CI: 2.6 to 4.6), with a similar 80% of Watch antibiotics were given to SPs portraying pattern in both urban and rural areas. In contrast, at a TB case (1086/1362). Three different last-resort or least one antibiotic was dispensed in 319/1,200 inter- ‘Reserve’ antibiotics (colistin, linezolid and faropenem) actions performed across Delhi, Mumbai and Patna, were prescribed in a total of 23 SP–provider interactions corresponding to 26.6% (95% CI: 24.2 to 29.2) of the in India, mainly for child diarrhoea (14/23). total. However, a direct comparison between these two Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 5 BMJ Global Health 6 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 Table 2 Number, proportion and bootstrapped 95% CIs (based on study- level clusters) of standardised patient–provider interactions in health facilities that resulted in prescription or dispensing of antibiotics across strata of key variables Country All India China Kenya Proportion Proportion Proportion Proportion Variable n/N (95% CI) n/N (95% CI) n/N (95% CI) n/N (95% CI) At least one antibiotic 2734/5863 46.6 (33.4 to 53.9) 2392/4798 49.9 (40.8 to 54.5) 259/899 28.8 (17.8 to 50.8) 83/166 50.0 (42.2 to 57.8) Antibiotics, n 0 3129/5863 53.4 (46.1 to 66.6) 2406/4798 50.1 (45.4 to 57.9) 640/899 71.2 (49.2 to 71.2) 83/166 50.0 (42.2 to 57.8) 1 2465/5863 42.0 (31.4 to 47.4) 2159/4798 45.0 (39.8 to 48.2) 229/899 25.5 (25.5 to 42.8) 77/166 46.4 (39.2 to 54.2) 2 260/5863 4.4 (1.6 to 6.5) 225/4798 4.7 (1.4 to 6.6) 29/899 3.2 (3.2 to 7.7) 6/166 3.6 (1.2 to 6.6) 3 9/5863 0.2 (0.02 to 0.3) 8/4798 0.2 (0.03 to 0.3) 1/899 0.1 (0.1 to 0.3) 0/166 0 Health facility location Urban 1653/3018 54.8 (50.0 to 55.2) 1570/2852 55.0 (53.0 to 55.2) – – 83/166 50.0 (42.8 to 57.8) Rural 1081/2845 38.0 (26.6 to 48.1) 822/1946 42.2 (39.0 to 46.7) 259/899 28.8 (17.8 to 50.8) – – Healthcare sector Public 443/1321 33.5 (20.6 to 50.8) 156/367 42.5 (37.6 to 47.7) 259/899 28.8 (17.8 to 50.8) 28/55 50.9 (38.2 to 63.6) Private 2291/4542 50.4 (40.8 to 54.5) 2236/4431 50.5 (50.2 to 54.5) – – 55/111 49.5 (40.1 to 51.6) Provider qualification Qualified 1186/1906 62.2 (45.4 to 71.3) 1115/1768 63.1 (44.6 to 71.8) 71/138 51.4 (42.8 to 59.4) NA NA Non- qualified 1358/3191 42.6 (38.7 to 48.6) 1277/3030 42.1 (37.8 to 47.9) 81/161 50.3 (42.9 to 57.8) NA NA Clinical presentation Angina 169/955 17.7 (12.2 to 28.3) 115/598 19.2 (16.8 to 21.1) 29/315 9.2 (5.9 to 12.4) 25/42 59.5 (45.2 to 73.8) Asthma 330/718 46.0 (44.0 to 50.2) 308/676 45.6 (43.5 to 49.0) – – 22/42 52.4 (38.1 to 66.7) Child diarrhoea 490/997 49.1 (33.4 to 67.9) 399/672 59.4 (50.5 to 75.0) 78/285 27.4 (21.8 to 32.5) 13/40 32.5 (17.5 to 45.5) Presumptive TB 1293/2253 57.4 (51.3 to 58.6) 1118/1912 58.5 (58.4 to 59.3) 152/299 50.8 (44.8 to 56.2) 23/42 54.8 (39.3 to 69.0) Confirmed TB 194/404 48.0 (47.7 to 50.0) 194/404 48.0 (47.7 to 50.0) – – – – Presumptive MDR- TB 258/536 48.1 (48.0 to 48.1) 258/536 48.1 (48.0 to 48.1) – – – – Patient referred for further evaluation* Yes 101/767 13.2 (9.4 to 20.4) 65/498 13.1 (9.7 to 17.4) 33/263 12.5 (7.3 to 31.6) 3/6 50.0 (16.7 to 83.3) No 2163/4384 50.7 (35.6 to 57.5) 1928/3628 53.1 (38.4 to 58.0) 226/636 35.5 (23.3 to 55.4) 67/120 55.8 (47.5 to 64.2) *All child diarrhoea cases from India and Kenya (n=712) were excluded from this analysis because children were not directly assessed by the provider. MDR- TB, multidrug resistant tuberculosis; NA, not available; TB, tuberculosis. BMJ Global Health Figure 1 Crude percentage of SP—provider interactions resulting in antibiotic prescription/dispensing, by country and selected conditions (pharmacy- based studies are not included). SP, standardised patient; TB, tuberculosis. studies is not possible owing to the very different contexts result in antibiotic prescription as compared with other involved and particularly to the different types of cases clinical conditions. Among the two pharmacy- based SP 10 11 that were examined. As observed in studies from health- studies done in India, the proportion of antibiotic care facilities, subjects presenting to pharmacies with dispensing was 26.6% and 3.6%, respectively. symptoms suggestive of TB were generally more likely to Although our focus was on LMICs, the overuse of anti- receive an antibiotic as compared with other conditions. biotics is not confined to LMICs. Large population-based The average proportion of Watch- antibiotics (predom- cohort data have shown that antibiotic overuse in ambu- inantly quinolones and cephalosporins) dispensed across latory settings across the United States was 30% among the three cities was 49.4% (95% CI: 43.9 to 54.4), ranging children and 17% among adults with certain respiratory from 24.0% (95% CI: 15.0 to 32.0) in Mumbai to 60.9% tract illnesses for which antibiotics are not indicated (95% CI: 55.1 to 67.1) in Patna. A deeper evaluation of (eg, asthma, allergies, acute bronchitis or bronchiol- antibiotic dispensing in Udupi district is limited by the itis). An analysis of antibiotic prescription practices small sample size. Only 55 antibiotics were dispensed based on administrative data from Ontario, Canada, across 1522 interactions, thus making subgroup anal- recently reported an overall rate of unnecessary antibi- yses less meaningful. Yet, it is worth highlighting that otic prescribing in primary care of 15.4%, though much nearly half of these antibiotics were discouraged FDCs higher percentages were observed for some respiratory of two antibiotics, whereas the remaining were almost conditions such as acute bronchitis (52.6%). However, equally distributed among Access- and Watch- groups. a direct comparison with higher income countries cannot More details regarding the types of antibiotics dispensed be done due to differences in study methodologies and across pharmacies in both studies are presented in online local epidemiology. supplementary file 2. Nearly 50% of all antibiotics prescribed in the context of India SP studies belonged to the ‘Watch’ list, with a peak of 80% among patients presenting with symptoms DISCUSSION suggestive of TB, which is consistent with national antibi- Our analysis of past SP studies involving 4798 SP–provider otic sales. Watch- antibiotics accounted for almost 33% interactions in India showed that healthcare providers of all antibiotics across China SP studies, but this is likely in primary care settings prescribed antibiotics to about underestimated because nearly one quarter of all antibi- half (49.9%) of patients presenting with clinical condi- otics could not be classified due to insufficient informa- tions that do not require antibiotics. Antibiotic overpre- tion. Of note, we observed a large use of cephalosporins scribing was found to be similar (50% of SP–provider (presumably second or third generation ones), which is interactions) in a small SP study carried out in Nairobi, in line with previous findings from drug sales analyses Kenya. Pooled data from two studies conducted in China and prescription audits conducted in various parts of showed lower levels of antibiotic overuse (28.8%), but it 2 23 24 China. In contrast, the small SP study conducted should be noted that percentages differed substantially in Nairobi revealed that over 85% of prescribed antibi- across individual studies, likely reflecting the different type of cases being involved. In fact, SP–provider interac- otics were from the ‘Access’ group, and half of these were tions involving presumptive TB cases were more likely to either trimethoprim/sulfamethoxazole or amoxicillin. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 7 BMJ Global Health 8 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 Table 3 Frequency of antibiotics prescribed/dispensed in health facilities across study countries, overall and according to both the AWaRe and ATC classifications India All settings Urban India Rural India China Drug type N Proportion (95% CI) N Proportion (95% CI) N Proportion (95% CI) N Proportion (95% CI) Any antibiotic 2768 – 1896 – 872 – 301 – AWaRe classification Access 876 31.6 (30.0 to 38.9) 584 30.8 (29.8 to 30.8) 292 33.5 (29.9 to 37.1) 126 41.9 (36.2 to 47.2) Watch 1317 47.6 (26.8 to 54.0) 1041 54.9 (54.9 to 55.4) 276 31.7 (21.2 to 40.3) 99 32.9 (27.6 to 37.9) Reserve 23 0.8 (0.5 to 1.8) 8 0.4 (0.4 to 0.5) 15 1.7 (1.0 to 2.1) 1 0.3 (0.3 to 1.3) Discouraged 334 12.1 (4.3 to 36.3) 50 2.6 (2.6 to 2.8) 284 32.6 (25.1 to 44.8) 1 0.3 (0.3 to 1.3) Not available* 218 7.9 (5.4 to 10.8) 213 11.2 (11.2 to 11.5) 5 0.57 (0.3 to 1.0) 74 24.6 (19.9 to 29.2) ATC classification Penicillin 711 25.7 (18.8 to 27.0) 535 28.2 (27.7 to 28.2) 176 20.2 (17.6 to 21.7) 68 22.6 (17.6 to 27.2) Cephalosporin 361 13.0 (8.2 to 14.6) 294 15.0 (14.9 to 15.0) 76 8.7 (7.8 to 10.7) 75 24.9 (20.9 to 29.2) First generation 21 0.8 (0.6 to 1.8) 9 0.5 (0.47 to 0.51) 12 1.4 (1.1 to 2.1) 0 0 Second generation 22 0.8 (0.2 to 1.1) 20 1.1 (1.1 to 1.2) 2 0.2 (0.2 to 0.4) 7 2.3 (0.7 to 4.0) Third generation 318 11.5 (7.1 to 12.9) 256 13.5 (13.3 to 13.5) 62 7.1 (6.4 to 8.1) 1 0.3 (0.3 to 1.0) Not available* 0 0 0 0 0 0 67 22.3 (18.3 to 26.6) Macrolide 425 15.4 (4.1 to 19.3) 389 20.5 (20.4 to 21.3) 36 4.1 (4.1 to 4.3) 60 19.9 (15.6 to 24.3) Quinolone 520 18.8 (16.6 to 24.2) 354 18.7 (18.5 to 18.7) 166 19.0 (18.5 to 26.8) 37 12.3 (9.0 to 15.9) Tetracycline 67 2.4 (1.7 to 4.6) 34 1.8 (1.4 to 1.8) 33 3.8 (3.0 to 4.1) 0 0 Imidazole† 61 2.2 (0.8 to 7.1) 1 0.05 (0.05 to 0.06) 60 6.9 (6.3 to 7.5) 1 0.3 (0.3 to 1.3) Sulfonamide‡ 18 0.7 (0.2 to 1.9) 3 0.16 (0.16 to 0.17) 15 1.7 (0.9 to 2.1) 9 3.0 (1.3 to 5.0) Aminoglycoside 6 0.2 (0.1 to 1.0) 0 0 6 0.7 (0.7 to 1.3) 45 15.0 (11.3 to 18.6) Combinations§ 289 12.1 (5.1 to 34.2) 50 2.6 (2.6 to 2.8) 284 32.6 (25.1 to 34.2) 1 0.3 (0.3 to 1.3) Antimycobacterial 229 8.3 (0.3 to 10.9) 226 11.9 (11.9 to 12.2) 3 0.3 (0.2 to 0.5) 1 0.3 (0.3 to 1.3) Other antibiotics 36 1.3 (1.0 to 2.4) 19 1.0 (0.1 to 1.0) 17 1.9 (1.8 to 2.6) 4 1.3 (0.3 to 2.7) The unit of analysis is the individual drug, not the standardised patient–provider interaction. *For these drugs, only the antibiotic class (eg, cephalosporin) was available. †Only metronidazole was prescribed/dispensed. ‡Only trimethoprim–sulfamethoxazole was prescribed/dispensed. §This category does not include combinations of antimycobacterial drugs. ATC, Anatomical–Therapeutic–Chemical; AWaRe, Access–Watch–Reserve. BMJ Global Health Figure 2 Factors associated with antibiotic prescribing/dispensing in health facilities in India. Covariate-adjusted pr evalence ratios and their 95% CIs estimated from a hierarchical Poisson model are reported. SP, standardised patient; TB, tuberculosis. This is in line with that observed in another SP study draw meaningful conclusions on antibiotic prescribing carried out in urban public primary healthcare facilities patterns in the area. in South Africa, where 10/119 (8.4%) interactions for Discouraged FDCs of antibiotics were commonly given presumptive TB resulted in antibiotic prescriptions, all of in India but not in other settings, accounting for 10.4% of which belonged to the access group. As with the Nairobi the total. FDCs were finally banned in India in September study, however, the small sample size does not allow to 2018, thus leaving hope for a change in the near future. Table 4 Antibiotic dispensing in Indian pharmacies Study setting Udupi district, Karnataka Mumbai, Delhi and Patna (n=1522) (n=1200) Variable n/N Proportion (95% CI) n/N Proportion (95% CI) Number of antibiotics 1 55/1522 3.6 (2.6 to 4.6) 294/1,00 24.5 (22.2 to 27.0) 2 0 0 25/1200 2.1 (1.3 to 2.9) Pharmacy location Urban 25/744 3.3 (2.2 to 4.7) 319/1200 26.6 (24.2 to 29.2) Rural 30/778 3.9 (2.7 to 5.2) – – Clinical presentation Adult with URI 11/250 4.4 (2.0 to 7.2) – – Adult with diarrhoea 12/259 4.6 (2.3 to 7.1) – – Adult with fever (malaria suspect) 10/252 4.0 (1.6 to 6.3) – – Child with URI 0/252 0 – – Child with diarrhoea 20/250 8.0 (4.8 to 11.2) – – Child with fever (malaria suspect) 2/259 0.8 (0.4 to 1.9) – – Adult with presumptive TB – – 221/599 36.9 (33.1 to 40.7) Adult with confirmed TB – – 98/601 16.3 (13.5 to 19.3) Patient referred to health provider Yes 15/710 2.1 (1.1; 3.1) 41/497 8.2 (5.8; 10.9) No 40/812 4.9 (3.6; 6.4) 278/703 39.5 (36.1; 43.2) TB, tuberculosis; URI, upper respiratory illness. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 9 BMJ Global Health Alarmingly, we observed the use of some ‘Reserve’ anti- Of note, available data originated from a range of biotics in primary care settings. In India, oral colistin was geographical areas with different sociocultural and prescribed for paediatric diarrhoea, and faropenem was economic profiles and could be generalisable to similar given to one patient with presumptive TB. This is very contexts in India. For all these reasons, the representa- concerning as parenteral colistin is the last resort drug tiveness of our findings is very good, and selection bias is for treatment of extremely drug-resistant Gram- negative likely negligible due to the robust mapping and sampling infections, and using the oral formulation could drive approach used across all SP studies. resistance in the community. Similarly, faropenem is an There are limitations in our study. First, the SP study data oral penem antibiotic which has been shown to cause from China and Kenya were limited and lacked general- cross- resistance to intravenous carbapenems. In China isability. Second, our analyses were restricted to overpre- SP studies, one presumptive TB case received aztreonam, scription and to a limited number of clinical scenarios. indicated for treatment of serious infections due to drug- Third, we could not investigate other important forms resistant Gram- negative bacteria. of inappropriate antibiotic use, such as the choice of the According to our findings from India, antibiotic overuse incorrect drug and dosage to treat a given infection. This was particularly common in rural areas, among qualified is an intrinsic limitation that arises from the type of tracer providers and for patients presenting with presumptive conditions used across SP studies so far. Although the SP TB. Besides leading to potentially dangerous diagnostic methodology was initially implemented to assess overall 28 29 delays, the unnecessary use of antibiotics causes quality of care in LMICs and to evaluate educational/ harms to the patient in terms of drug- associated adverse behavioural programmes in high-income countries, this events and increased out- of- pocket costs. approach is being increasingly adopted to gain insight While normative boundaries may partly explain why into medication use, and especially drug dispensing prac- qualified providers prescribed more antibiotics than tices among pharmacists. Data recording systems in SP non- qualified ones as observed in our analyses for studies are therefore improving in order to facilitate the India, the widespread overuse of antibiotics suggests collection of key details regarding medications that were that important training gaps likely exist. However, harder to capture from studies whose main objective was prescribing behaviours among healthcare providers not related to drug use. also depend on a number of other factors, including In conclusion, the prevalence of antibiotic overpre- financial incentives from pharmaceutical companies, scribing estimated from SP studies ranged from 29% patient expectations and requests, or just old habits that in China to 50% in India and Kenya, and Watch anti- 8 30 31 are hard to die. biotics accounted for a large proportion of antibiotics The biggest strength of our study lies in the nature prescribed in both India and China. Combining the SP and quality of the data used to investigate the extent and methodology with new tracer conditions would allow patterns of antibiotic overprescribing. Although previous overcoming many of the typical limitations of most research had already highlighted that Watch group anti- studies aimed at evaluating inappropriate antibiotic use biotics are highly prescribed across India and China, in greater detail. SPs represent a unique opportunity to such studies could not provide a clear picture of inap- further explore prescription practices among healthcare propriate antibiotic use owing to the limited amount of providers, including the management of common infec- clinical information available from prescription audits tious diseases, such as pneumonia or urinary tract infec- 32–34 and evaluations of drug sales data. Among the main tions, that contribute substantially to the overall antibiotic advantages of using SPs to evaluate prescription practices use in primary care. Future studies also need to focus on is the fact that tracer conditions are standardised. In all untangling the channels for antibiotic overprescription studies included in our analyses, such conditions were and better understand the determinants of such practice very common illnesses that are frequently encountered among public and private healthcare providers in various in primary care and that require a well- defined diagnostic contexts. and therapeutic management that does not involve anti- The extent of antibiotic overuse in primary care across biotic use. LMICs is a serious concern and requires targeted anti- Furthermore, representative samples of healthcare microbial stewardship interventions aimed at improving providers from public and/or private sectors were rational and locally adapted prescribing practices. An selected in all SP studies conducted in India, with the active involvement of private providers in all such inter- only exception of one relatively small pilot study in ventions would be essential to ensure uptake, particularly Delhi. In this pooled dataset, private practitioners were in countries where the private sector plays a major role in much more represented than public providers, but we healthcare. Greater efforts are also necessary to develop lacked statistical power to make appropriate compari- and scale up accurate point- of- care tests that could guide sons between the two groups. Yet, this distribution well therapeutic choices where resources are scarce. Addi- reflects the fact that about 75% of outpatient visits in tional research is also required to evaluate whether anti- India take place in the private sector, with nearly 70% of biotic use (especially use of drugs such as azithromycin primary care in the country being delivered by informal and hydroxychloroquine) will dramatically increase as a 35 36 providers. consequence of the COVID-19 pandemic, and concerns 10 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health 3 Spivak ES, Cosgrove SE, Srinivasan A. Measuring appropriate have already been raised about the implications for antimicrobial use: attempts at opening the black box. Clin Infect Dis AMR. 2016;63:1639–44. 4 Smieszek T, Pouwels KB, Dolk FCK, et al. Potential for reducing inappropriate antibiotic prescribing in English primary care. J Author affiliations 1 Antimicrob Chemother 2018;73:ii36–43. Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, 5 Kwan A, Daniels B, Bergkvist S, et al. Use of standardised patients Québec, Canada for healthcare quality research in low- and middle- income countries. McGill International TB Centre, McGill University, Montreal, Québec, Canada BMJ Glob Health 2019;4:e001669. McCourt School of Public Policy, Georgetown University, Washington, District of 6 Das J, Kwan A, Daniels B, et al. Use of standardised patients to Columbia, USA assess quality of tuberculosis care: a pilot, cross- sectional study. Lancet Infect Dis 2015;15:1305–13. School of Public Health, University of California Berkeley, Berkeley, California, USA 5 7 Kwan A, Daniels B, Saria V, et al. Variations in the quality of Division of Infectious Diseases, Department of Medicine, Washington University in tuberculosis care in urban India: a cross- sectional, standardized Saint Louis, Saint Louis, Missouri, USA patient study in two cities. PLoS Med 2018;15:e1002653. Dahdaleh Institute of Global Health Research, York University, Toronto, Ontario, 8 Das J, Chowdhury A, Hussam R, et al. The impact of training Canada informal health care providers in India: a randomized controlled trial. Centre for the AIDS Programme of Research in South Africa (CAPRISA), University Science 2016;354:aaf7384. 9 Das J, Holla A, Das V, et al. In urban and rural India, a standardized of KwaZulu- Natal, Durban, KwaZulu- Natal, South Africa 8 patient study showed low levels of provider training and huge quality Centre for Policy Research, New Delhi, Delhi, India gaps. Health Aff 2012;31:2774–84. Manipal McGill Program for Infectious Diseases, Manipal Centre for Infectious 10 Satyanarayana S, Kwan A, Daniels B, et al. Use of standardised Diseases, Manipal Academy of Higher Education, Manipal, Karnataka, India patients to assess antibiotic dispensing for tuberculosis by pharmacies in urban India: a cross- sectional study. Lancet Infect Dis Twitter Giorgia Sulis @giorgiasulis, Benjamin Daniels @_bbdaniels, Ada Kwan 2016;16:1261–8. @kwantada, Amrita Daftary @DaftaryAmrita and Madhukar Pai @paimadhu 11 Nafade V, Huddart S, Sulis G, et al. Over- The- Counter antibiotic dispensing by pharmacies: a standardised patient study in Udupi Contributors GS, MP, JD and BD designed the study. GS and BD performed the district, India. BMJ Glob Health 2019;4:e001869. data cleaning and coding. GS analysed the data and prepared the first draft of the 12 Daniels B, Dolinger A, Bedoya G, et al. Use of standardised patients paper. All authors critically revised the manuscript until final approval. to assess quality of healthcare in Nairobi, Kenya: a pilot, cross- sectional study with international comparisons. BMJ Glob Health Funding Most studies included in our analyses were funded by the Bill and 2017;2:e000333. Melinda Gates Foundation (OPP1091843). GS is a recipient of a Richard H. 13 Sylvia S, Shi Y, Xue H, et al. Survey using incognito standardized Tomlinson Doctoral Fellowship (McGill University), and MP holds a Tier 1 Canada patients shows poor quality care in China's rural clinics. Health Research Chair from Canadian Institutes of Health Research. Policy Plan 2015;30:322–33. 14 Sylvia S, Xue H, Zhou C, et al. Tuberculosis detection and the Competing interests MP is on the editorial boards of BMJ Global Health. All other challenges of integrated care in rural China: a cross- sectional authors declare that they have no conflicts of interest. standardized patient study. PLoS Med 2017;14:e1002405. Patient and public involvement Patients and/or the public were not involved in 15 Xue H, Shi Y, Huang L, et al. Diagnostic ability and inappropriate antibiotic prescriptions: a quasi- experimental study of primary the design, conduct, reporting or dissemination plans of this research. care providers in rural China. J Antimicrob Chemother Patient consent for publication Not required. 2019;74:256–63. 16 Sharland M, Gandra S, Huttner B, et al. Encouraging AWaRe- Ethics approval Each study had been conducted after ethics approval. Our study ness and Discouraging inappropriate antibiotic use- the new 2019 was approved by the McGill Faculty of Medicine Institutional Review Board (IRB essential medicines list becomes a global antibiotic stewardship review number: A04-B19- 20B (20-04-053)), and all primary studies we included tool. Lancet Infect Dis 2019;19:1278–80. had their own independent ethics approvals. 17 WHO. Anatomical therapeutic chemical (ATC) classification system. Oslo, Norway: WHO Collaborating Centre for Drug Statistics Provenance and peer review Not commissioned; externally peer reviewed. Methodology, 2020. https://www. whocc. no/ atc_ ddd_ index/ 2020 Data availability statement Data are available upon request. All data relevant 18 DiCiccio TJ, Efron B. Bootstrap confidence intervals. Stat Sci to the study are included in the article or uploaded as supplementary information. 1996;11:189–228. Parts of the data used for our analyses are available through original publications 19 Stewart GB, Altman DG, Askie LM, et al. Statistical analysis of individual participant data meta- analyses: a comparison of methods as reported in the reference list. Additional data included in this study are available and recommendations for practice. PLoS One 2012;7:e46042. upon request. 20 Fleming- Dutra KE, Hersh AL, Shapiro DJ, et al. Prevalence of Open access This is an open access article distributed in accordance with the inappropriate antibiotic prescriptions among US ambulatory care Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits visits, 2010-2011. JAMA 2016;315:1864–73. 21 Schwartz KL, Langford BJ, Daneman N, et al. Unnecessary antibiotic others to copy, redistribute, remix, transform and build upon this work for any prescribing in a Canadian primary care setting: a descriptive analysis purpose, provided the original work is properly cited, a link to the licence is given, using routinely collected electronic medical record data. CMAJ Open and indication of whether changes were made. See: https:// creativecommons. org/ 2020;8:E360–9. licenses/ by/ 4. 0/. 22 Klein EY, Milkowska- Shibata M, Tseng KK, et al. Assessment of who antibiotic consumption and access targets: 2000-2015. Lancet ORCID iDs Infectious Diseases 2020:S1473-3099(20)30332-7. Giorgia Sulis http:// orcid. org/ 0000- 0001- 6641- 0094 23 Wushouer H, Tian Y, Guan X- D, et al. Trends and patterns of Benjamin Daniels http:// orcid. org/ 0000- 0001- 9652- 6653 antibiotic consumption in China's tertiary hospitals: based on Ada Kwan http:// orcid. org/ 0000- 0003- 4889- 9433 a 5 year surveillance with sales records, 2011-2015. PLoS One Amrita Daftary http:// orcid. org/ 0000- 0003- 2275- 3540 2017;12:e0190314. 24 Lin H, Dyar OJ, Rosales- Klintz S, et al. Trends and patterns of Jishnu Das http:// orcid. org/ 0000- 0002- 5909- 3585 antibiotic consumption in Shanghai municipality, China: a 6 year Madhukar Pai http:// orcid. org/ 0000- 0003- 3667- 4536 surveillance with sales records, 2009-14. J Antimicrob Chemother 2016;71:1723–9. 25 Christian CS, Gerdtham U- G, Hompashe D, et al. Measuring quality gaps in TB screening in South Africa using standardised patient analysis. Int J Environ Res Public Health 2018;15:ijerph15040729. REFERENCES 26 Bergen PJ, Smith NM, Bedard TB, et al. Rational combinations 1 Huttner BD, Catho G, Pano- Pardo JR, et al. COVID-19: don't of polymyxins with other antibiotics. Adv Exp Med Biol neglect antimicrobial stewardship principles! Clin Microbiol Infect 2019;1145:251–88. 2020;26:808–10. 27 Gandra S, Choi J, McElvania E, et al. Faropenem resistance causes 2 Sulis G, Adam P, Nafade V, et al. Antibiotic prescription practices in vitro cross- resistance to carbapenems in ESBL- producing in primary care in low- and middle- income countries: a systematic Escherichia coli. Int J Antimicrob Agents 2020;55:105902. review and meta- analysis. PLoS Med 2020;17:e1003139. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 11 BMJ Global Health 28 Getnet F, Demissie M, Assefa N, et al. Delay in diagnosis of middle- income country setting: a repeated cross- sectional study in pulmonary tuberculosis in low- and middle- income settings: Vellore, South India. J Antimicrob Chemother 2013;68:229–36. systematic review and meta- analysis. BMC Pulm Med 2017;17:202. 34 Farooqui HH, Mehta A, Selvaraj S. Outpatient antibiotic prescription 29 Robson J, Ayerbe L, Mathur R, et al. Clinical value of chest pain rate and pattern in the private sector in India: evidence from medical presentation and prodromes on the assessment of cardiovascular audit data. PLoS One 2019;14:e0224848. disease: a cohort study. BMJ Open 2015;5:e007251. 35 Mackintosh M, Channon A, Karan A, et al. What is the 30 Currie J, Lin W, Meng J. Addressing antibiotic abuse in China: an private sector? understanding private provision in the health experimental audit study. J Dev Econ 2014;110:39–51. systems of low- income and middle- income countries. Lancet 31 Fitzpatrick A. Do Informed Consumers Reduce the Price and 2016;388:596–605. Prevalence of Counterfeit Drugs ?Evidence from the Antimalarial 36 Gopalkrishnan R, Walia K, Ohri VC. Treatment guidelines for Market, 2015. antimicrobial use in common syndromes. 2nd ed. New Delhi, India: 32 Kotwani A, Holloway K. Antibiotic prescribing practice for acute, Indian Council of medical research (ICMR), 2019. uncomplicated respiratory tract infections in primary care settings in 37 WHO. Record number of countries contribute data revealing New Delhi, India. Trop Med Int Health 2014;19:761–8. disturbing rates of antimicrobial resistance. Geneva, Switzerland: 33 Chandy SJ, Thomas K, Mathai E, et al. Patterns of antibiotic use in the community and challenges of antibiotic surveillance in a lower- World Health Organization (WHO), 2020. 12 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png BMJ Global Health British Medical Journal

Antibiotic overuse in the primary health care setting: a secondary data analysis of standardised patient studies from India, China and Kenya

Loading next page...
 
/lp/british-medical-journal/antibiotic-overuse-in-the-primary-health-care-setting-a-secondary-data-R9AkABCydN

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
British Medical Journal
Copyright
© Author(s) (or their employer(s)) 2020. Re-use permitted under CC BY. Published by BMJ.
eISSN
2059-7908
DOI
10.1136/bmjgh-2020-003393
Publisher site
See Article on Publisher Site

Abstract

Original research Antibiotic overuse in the primary health care setting: a secondary data analysis of standardised patient studies from India, China and Kenya 1,2 3 4 5 Giorgia Sulis , Benjamin Daniels , Ada Kwan , Sumanth Gandra, 6,7 3,8 1,2,9 Amrita Daftary , Jishnu Das , Madhukar Pai To cite: Sulis G, Daniels B, ABSTRACT INTRODUCTION Kwan A, et al. Antibiotic Introduction Determining whether antibiotic prescriptions Antibiotic stewardship is critical for tackling overuse in the primary health are inappropriate requires knowledge of patients’ antimicrobial resistance (AMR), especially care setting: a secondary data underlying conditions. In low- income and middle- income in the context of the ongoing COVID-19 analysis of standardised patient countries (LMICs), where misdiagnoses are frequent, this is studies from India, China and pandemic. In a recent systematic review on challenging. Additionally, such details are often unavailable Kenya. BMJ Global Health antibiotic prescription practices in primary for prescription audits. Recent studies using standardised 2020;5:e003393. doi:10.1136/ care settings across low-income and middle- patients (SPs) offer a unique opportunity to generate bmjgh-2020-003393 income countries (LMICs), we showed that unbiased prevalence estimates of antibiotic overuse, as approximately 50% of patients of any age the research design involves patients with predefined Handling editor Seye Abimbola conditions. seeking care for any reason received at least JD and MP contributed equally. Methods Secondary analyses of data from nine SP one antibiotic. studies were performed to estimate the proportion of SP– However, determining inappropriate JD and MP are joint senior provider interactions resulting in inappropriate antibiotic prescription in LMICs is a challenge, and authors. prescribing across primary care settings in three LMICs a standardised tool for its assessment is (China, India and Kenya). In all studies, SPs portrayed Received 8 July 2020 currently unavailable. Inappropriate antibi- conditions for which antibiotics are unnecessary (watery Revised 1 August 2020 otic prescribing can derive from a range of diarrhoea, presumptive tuberculosis (TB), angina and Accepted 3 August 2020 failings: (1) prescription in the absence of asthma). We conducted descriptive analyses reporting overall prevalence of antibiotic overprescribing by clinical indication (ie, ‘overprescription’), healthcare sector, location, provider qualification and case. which not only produces zero benefit to the The WHO Access–Watch–Reserve framework was used to patient but can also be harmful (eg, drug categorise antibiotics based on their potential for selecting toxicities or costs for patients); (2) failure resistance. As richer data were available from India, we to prescribe antibiotics when necessary examined factors associated with antibiotic overuse in that (ie, ‘underprescription’); (3) suboptimal country through hierarchical Poisson models. antibiotic choice with respect to aetiology Results Across health facilities, antibiotics were given (confirmed or presumptive), site, severity inappropriately in 2392/4798 (49.9%, 95% CI 40.8% of infection and patient characteristics (eg, to 54.5%) interactions in India, 83/166 (50.0%, 95% CI 42.2% to 57.8%) in Kenya and 259/899 (28.8%, 95% CI age, comorbidities and pregnancy status); (4) 17.8% to 50.8%) in China. Prevalence ratios of antibiotic prescription of wrong dosage and/or dura- overuse in India were significantly lower in urban versus tion of antibiotic treatment as compared with rural areas (adjusted prevalence ratio (aPR) 0.70, 95% 3 4 national and international guidelines. CI 0.52 to 0.96) and higher for qualified versus non- Methods used to assess inappropriateness, qualified providers (aPR 1.55, 95% CI 1.42 to 1.70), and such as prescription audits, medical records for presumptive TB cases versus other conditions (aPR and patient exit interviews, have multiple 1.19, 95% CI 1.07 to 1.33). Access antibiotics were © Author(s) (or their 3 5 limitations. Electronic records are seldom predominantly used in Kenya (85%), but Watch antibiotics employer(s)) 2020. Re- use available in LMICs, particularly in primary (mainly quinolones and cephalosporins) were highly permitted under CC BY. Published by BMJ. prescribed in India (47.6%) and China (32.9%). care, thus making accurate prescription audit Conclusion Good- quality SP data indicate alarmingly For numbered affiliations see tools difficult to implement. Also, the paucity high levels of antibiotic overprescription for key conditions end of article. and variation of clinical details that can be across primary care settings in India, China and Kenya, captured through medical records (paper- Correspondence to with broad- spectrum agents being excessively used in based or not), if they even exist, makes it even Dr Madhukar Pai; India and China. madhukar. pai@ mcgill. ca harder to determine the appropriateness Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 1 BMJ Global Health opportunity to overcome the methodological limitations Key questions typical of other studies, thus making the assessment of inappropriateness of antibiotic use less biased and more What is already known? accurate. Because the underlying illness is prespecified, ► A recent systematic review and meta- analysis showed that, across the SP methodology allows accurate assessment if an anti- 48 studies from 27 low-income and middle- income countries biotic is inappropriately prescribed. The SP approach is including China, India and Kenya, approximately half of all pa- tients evaluated in outpatient primary care received an antibiotic not affected by poor recall, recall bias or the Hawthorne prescription. effect, which is commonly observed in patient exit ► Methods used to assess inappropriateness of antibiotic prescrip- interviews and direct observations of patient–provider tion, such as prescription audits, medical records and patient exit encounters. interviews, have multiple limitations. Considering the aforementioned advantages, we ► Standardised patients (SPs) offer a unique opportunity to explore performed a secondary analysis of prescription data prescribing practices and accurately estimate overprescription be- from previously conducted SP studies in three LMICs cause case presentations are fixed by design, thus allowing com- (India, China and Kenya) with two objectives: (1) to esti- parisons across settings and providers. mate the overall proportion of SP–provider interactions What are the new findings? (separately for pharmacy-based and health facility- based ► In this secondary analysis of data from nine SP studies carried out studies) that resulted in prescription or dispensing of at in India, Kenya and China, we provide a more unbiased prevalence least one antibiotic in the absence of clinical indication estimate of antibiotic overprescription for selected clinical condi- (ie, overprescription) and (2) to identify factors associ- tions (asthma, angina, watery diarrhoea, presumptive or confirmed ated with antibiotic overprescribing in health facilities. tuberculosis (TB)) across a range of primary healthcare providers. ► About 30% of SP–provider interactions in China and 50% of those performed in India and Kenya resulted in inappropriate antibiotic METHODS prescription. Study design and data sources ► Watch antibiotics (ie, broad-spectrum agents with higher potential for selecting resistance) were very commonly prescribed in India Data on SP–provider interactions (ie, completed SP visits (about 50%) and China (over 32%), and some patients (0.8%) even with a provider at a health facility or a pharmacy) from received last- resort antibiotics belonging to the ‘Reserve’ group. studies conducted by members of our team (India and ► In India, the average prevalence of antibiotic prescribing was 30% Kenya) or had used SP cases developed by our team or lower in urban versus rural areas, 55% higher among qualified obtained from publicly accessible sources (China) were providers compared with non- qualified ones and 19% higher for gathered to compile a pooled dataset for secondary anal- patients presenting with presumptive TB versus other conditions. 6–15 yses. The methods used are described in our published What do the new findings imply? manual and toolkit on how to conduct SP studies. ► Our findings indicate alarming levels of antibiotic overprescription Among studies carried out in India, four involved for conditions that are frequently encountered in primary care, po- primary health facilities across five sites (Delhi, Mumbai, tentially leading to toxic effects and diagnostic delays. Patna, three districts in the State of Madhya Pradesh, ► The choice of antibiotics given to patients is concerning, as several 6–9 and Birbhum district in the State of West Bengal), agents with high potential for resistance selection are often inap- while two were performed in pharmacies located in four propriately prescribed. different areas (Mumbai, Patna, Delhi and Udupi district ► The SP methodology could prove useful to further investigate anti- 10 11 of Karnataka). We also examined data from a pilot biotic prescribing practices and its underlying determinants, using study carried out in Nairobi (Kenya) and two studies other case presentations across a range of different contexts. completed in rural areas of China (Shaanxi, Sichuan and Anhui provinces), all involving only primary healthcare 12–15 providers. of prescription. Patient exit interviews are commonly Information regarding medications prescribed by used alternatives but come with several major drawbacks healthcare providers were collected in these published that can result in poor and inaccurate estimates that are SP studies but were not analysed in depth, especially with incomparable. Data collected in this manner are subject regard to inappropriate use. This is because, in most to recall bias, poor recall and limited clinical expertise instances, the primary publications focused on overall among patients. Further, not only are clinical presenta- quality of care, rather than the specific components of tions highly heterogeneous but also the difficulty in actu- care. ally determining what patients have makes comparisons very challenging for research. Provider selection in original studies A less biased method is the use of standardised patients Sampling approaches adopted in each primary study from (SPs), also known as ‘simulated’ or ‘mystery’ patients, that which our data were drawn are summarised in table 1. is, healthy individuals recruited from local communities For the two pharmacy- based studies, a random sample of and extensively trained to portray a standardised clinical pharmacies was selected from a comprehensive list of all 5 10 11 condition to a healthcare provider. Since their clinical those eligible obtained from relevant authorities. In presentations are fixed by design, SPs offer an important six of the other eight studies, healthcare providers were 2 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 3 Table 1 Main features of SP studies included in our analyses SP–provider Healthcare Facility Provider Provider Study site (year) interactions Tracer conditions sector location Provider selection approach consent participation* China (2013) 600 Angina, child Public Rural Census of all clinics designated under the New Cooperative Yes 100% diarrhoea Medical Scheme (ie, the major public health insurance programme in rural areas), followed by random selection of providers China (2015) 299 Presumptive TB Public Rural Census of all public providers followed by random sampling Yes 274/274 (100%) from one prefecture in each of 3 provinces out of a total of 47 prefectures, chosen to be representative of rural health systems Kenya (2014) 166 Angina, asthma, Public and Urban Non- random convenience sample designed to include low- Yes 46/49 (93.9%) child diarrhoea, private income, middle- income and high- income neighbourhoods in presumptive TB various Nairobi areas Madhya Pradesh, 1123 Angina, asthma, Public and Rural Census of all medical care providers working in 60 villages No Not applicable India (2010–2011) child diarrhoea private randomly sampled in three districts in Madhya Pradesh; all public providers and qualified private providers were automatically sampled; for each public provider, the closest private practitioner was also sampled Delhi, India (2014) 250 Presumptive and Private Urban Convenience sample (pilot study) Yes Not available confirmed TB, presumptive MDR- TB Mumbai and Patna, 2602 Presumptive and Private Urban Street- by- street mapping of private providers who were known No Not applicable India (2014–2015) confirmed TB, to see adult outpatients with respiratory symptoms, followed by presumptive MDR- random sampling stratified by provider qualification and private TB provider interface agency registration status Birbhum district, 823 Angina, respiratory Private Rural Census of private health providers who had been practising for at Yes 304/360 (84.4%) West Bengal, India distress, child least 3 years in 203 villages across Birbhum district (2012–2014) diarrhoea Mumbai, Patna and 1200 Presumptive TB, Pharmacies Urban Convenience sample of 54 pharmacies from 28 low- income No Not applicable Delhi, India confirmed TB localities in Delhi (pilot phase), random sampling of pharmacies in (2014–2015) Mumbai and Patna from a list of all pharmacies registered in the two cities Udupi district, 1522 For both adults Pharmacies Urban Of the 350 pharmacies registered in the district as per the local No Not applicable Karnataka, India and children: upper and rural pharmacy association, 279 were considered eligible for the (2018) respiratory tract study after excluding those operating inside hospitals (47), those infection, diarrhoea, permanently closed or under renovations (10), those that could presumptive malaria not be identified by the field team (4), those for veterinarian purposes only (1) and those used for SP training (10). *For studies in which provider consent was required. MDR- TB, multidrug resistant tuberculosis; SP, standardised patient; TB, tuberculosis. BMJ Global Health randomly sampled after performing a census or street- by- proportion of SP–provider interactions that resulted 7–9 13–15 street mapping in the study areas. A convenience in antibiotic prescription or dispensing. The overall sample of practitioners was selected in two pilot studies proportion of prescribed or dispensed antibiotics, along 6 12 respectively performed in Delhi and Nairobi. A waiver with ATC- class and AWaRe group- specific proportions, of provider consent was obtained in four out of nine was calculated across strata defined by key variables of studies, all carried out in India, two of which involved interest, such as healthcare sector (public/private), 7 9–11 pharmacies. In all the others, verbal or written facility location (urban/rural), provider qualification informed consent was sought at least 6 weeks prior to the (qualified/non- qualified, defined based on whether they commencement of SP–provider interactions in order to had at least a bachelor’s degree in medicine) and tracer reduce the risk of SP detection. Yet, participation rates conditions. For all prevalence proportions, we computed were very high (85%–100%) among eligible health 95% CIs using bootstrapping in order to account for clus- practitioners, and non- participation was usually due to tering at the study level. logistical issues on the day of the visits rather than active In order to examine the factors associated with anti- refusal to be involved in the project. Hence, it is reason- biotic prescribing in health facilities in India, we fit a able to expect negligible differences between participants hierarchical Poisson regression model that allows direct and non-participants, making non- response bias a minor estimation of adjusted prevalence ratios (aPRs) even if the concern. In all studies, SPs were randomly assigned to outcome is common as in this case. Our model included providers, and completion rates of SP–provider interac- a random intercept for studies and dummy variables for tions were always very high. facility location, healthcare sector, provider qualification and tracer conditions as predictors. As we anticipated a Tracer conditions fair amount of between- study heterogeneity, we decided Tracer conditions (ie, SP case presentations) were to opt for a mixed model that could better account for defined similarly across SP studies, thus allowing compar- it as compared with including the study or study site as a isons across settings. Cases ranged from presumptive or covariate. Among tracer conditions, only angina, asthma confirmed tuberculosis (TB) (which requires specific and presumptive TB could be included in order to avoid anti-TB treatment as per WHO recommendations) to sparse data problems (ie, violations of the positivity self- limiting infections, such as watery diarrhoea or upper assumption). The effect of all predictors was expected to respiratory tract illness (which only need support treat- be similar across studies, and therefore only fixed slopes ment, eg, rehydration therapy for diarrhoea), to non- were considered. These analyses were restricted to India communicable diseases like asthma or chest pain indica- because we had diverse and more data. We also consid- tive of angina (these should be referred to a higher level ered alternative models and examined the pros and cons of care). Importantly, none of such conditions requires of each. A full description of our analyses is provided in antibiotics, which means that any antibiotic prescribed to online supplementary file 1. SPs is deemed inappropriate by indication (ie, overpre- Data from pharmacies were not pooled because contexts scription). and tracer conditions were highly heterogeneous in the two available studies. Therefore, we only calculated prev- Outcome assessment alence proportions and 95% CIs of dispensed antibiotics, Raw data from original studies were harmonised and both overall and in stratified analyses. recoded as needed. We used the available information All analyses were performed using Stata 16. on medications that were prescribed or dispensed during each SP–provider interaction to categorise individual Patient and public involvement drugs. Antibacterial agents were further classified using It was not possible to involve patients or the public in the both the ATC (Anatomical–Therapeutic–Chemical) design, conduct, reporting or dissemination plans of our Index and the WHO Access–Watch–Reserve (AWaRe) research because this is a secondary analysis of previously 16 17 framework. Fixed- dose combinations (FDCs) of anti- conducted studies. biotics (eg, ciprofloxacin/ornidazole) were classified as ‘discouraged’ antibiotics as per WHO recommendations. The primary outcome measure was expressed as the proportion of SP–provider interactions that resulted RESULTS in antibiotic prescription or dispensing. Secondary The main features of SP studies that were included in outcomes were proportions of specific groups of antibi- our analyses are summarised in table 1. A total of 4798 otics that were prescribed or dispensed both overall and SP–provider interactions were completed in health across strata of key variables of interest. These propor- facilities across urban and rural India, predominantly tions provide a direct measure of antibiotic overuse. in the private sector. Both private and public health- care providers were involved in the pilot study carried Statistical analyses out in Nairobi (166 interactions), whereas studies from For studies carried out in health facilities, we conducted rural China only targeted the public sector (899 inter- country- level descriptive analyses and reported the crude actions). For these health facility-based studies, we first 4 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health present summary statistics and then report results from Discouraged antibiotics, that is, FDCs other than anti- our models. mycobacterial drugs (such as norfloxacin+tinidazole or ofloxacin+ornidazole) accounted for 12.1%, of which all but one were given for child diarrhoea. Anti- TB medi- Antibiotic overuse across settings cations represented 8.3% of antibiotics in India; almost In India, 2392 of 4798 (49.9%, 95% CI 40.8% to 54.5%) all of them were given by healthcare providers in urban SP–provider interactions resulted in at least one anti- areas; and none could be considered appropriate based biotic prescription (table 2). Similar proportions were on the expected correct management of such cases. observed in Nairobi (83 of 166; 50.0%, 95% CI 42.2% to About one- quarter of drugs prescribed in studies from 57.8%), while a lower percentage was found in the China China could not be categorised based on the AWaRe studies (259 of 899; 28.8%, 95% CI 17.8% to 50.8%). framework because only the drug class was reported. However, in the latter case, the CI was substantially wide, These were mainly cephalosporins, most likely second reflecting the considerable between-study variance due or higher generation, and therefore the overall propor- to differences in tracer conditions evaluated. tion of Watch-group antibiotics is expected to be greater In most instances, only one antibiotic was given than 32.9% (table 3). Undefined cephalosporins were during an individual SP–provider interaction; less than by far the most prescribed antibiotics in China (76/301, 5% of interactions across all settings resulted in two or 25.2%), followed by gentamicin (45/301, 15.0%), amox- more antibiotics prescriptions. Crude analyses of data icillin (37/301, 12.3%), erythromycin (26/301, 8.6%) from India indicate that antibiotic overprescription was and levofloxacin (18/301, 6.0%). more common among healthcare providers in urban Subgroup analyses of antibiotic prescription patterns areas, among those working in the private sector and among SP–provider interactions that took place in among qualified professionals. Furthermore, antibiotics Nairobi were limited by the small sample size. However, were largely overprescribed to patients presenting with 85.4% (76/89) of all antibiotics prescribed were first-line a diverse range of clinical conditions in all countries and narrow- spectrum agents from the ‘Access’ group, (figure 1). In India, the percentage of subjects receiving while the remaining belonged to the ‘Watch’ group. antibiotics was close to 50% for most case types, with a peak of 59.4% (95% CI 50.5% to 75.0%) among child Factors associated with antibiotic overuse in India diarrhoea cases. However, for angina cases, it was 19.2% Prevalence ratios of antibiotic overuse and their 95% (95% CI 16.8% to 21.1%). About half of the visits for CIs estimated through mixed-effects Poisson regression presumptive TB in China received antibiotics inappro- analysis are reported in figure 2. The adjusted preva- priately, as opposed to 9.2% (95% CI 5.9% to 12.4%) of lence of antibiotic prescribing was lower in urban versus visits for suspicious angina and 27.4% (95% CI 21.8% to rural areas (aPR=0.70; 95% CI: 0.52 to 0.96), for subjects 32.5%) for child diarrhoea. Case- specific estimates from presenting with suspicious angina (aPR=0.33; 95% CI: Nairobi are highly imprecise due to the small sample size. 0.27 to 0.40), and asthma (aPR=0.77; 95% CI: 0.66 to 0.89). Patients with presumptive TB were more likely to Type of antibiotics used receive inappropriate antibiotics (aPR=1.19; 95% CI: 1.07 Across studies performed in India, 2768 antibiotics were to 1.33) as compared with individuals with other clinical given to 2392 patients. The top 10 most prescribed antibi- conditions. Qualified practitioners were more likely to otics across SP–provider interactions in India were azith- prescribe antibiotics than non-qualified ones (aPR 1.55; romycin (381, 13.8%), amoxicillin+beta-lactamase inhib - 95% CI: 1.42 to 1.70). itor (344, 12.4%), amoxicillin (264, 9.5%), levofloxacin The hierarchical Poisson model did not show any signif- (202, 7.3%), cefixime (198, 7.2%), ofloxacin (165, 6.0%), icant difference between public and private providers, ofloxacin+ornidazole (150, 5.4%), norfloxacin+tinida- but this is in contrast with what emerged from alternative zole (136, 4.9%), ciprofloxacin (102, 3.7%) and cefpo- models as described in online supplementary file 1. doxime (88, 3.2%). Broad-spectrum agents with higher potential for selecting resistance (Watch antibiotics) were Antibiotic dispensing in pharmacies disproportionately represented (47.6%, 95% CI 26.8% to Our secondary analysis of data from two pharmacy- 54.0%), and even more so in urban areas (54.9%, 95% CI based SP studies showed that over-the- counter antibiotic 54.9% to 55.4%) (table 3). This reflects the heavy use of dispensing is also a common problem in various parts of quinolones, cephalosporins and macrolides that respec- India (table 4). tively accounted for 18.8% (95% CI 16.6% to 24.2%), In Udupi district (Karnataka state) the proportion of 13.0% (95% CI 8.2% to 14.6%) and 15.4% (95% CI 4.1% SP—pharmacist interactions that resulted in antibiotic to 19.3%) of all antibiotics prescribed in India. Nearly dispensing was 3.6% (95% CI: 2.6 to 4.6), with a similar 80% of Watch antibiotics were given to SPs portraying pattern in both urban and rural areas. In contrast, at a TB case (1086/1362). Three different last-resort or least one antibiotic was dispensed in 319/1,200 inter- ‘Reserve’ antibiotics (colistin, linezolid and faropenem) actions performed across Delhi, Mumbai and Patna, were prescribed in a total of 23 SP–provider interactions corresponding to 26.6% (95% CI: 24.2 to 29.2) of the in India, mainly for child diarrhoea (14/23). total. However, a direct comparison between these two Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 5 BMJ Global Health 6 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 Table 2 Number, proportion and bootstrapped 95% CIs (based on study- level clusters) of standardised patient–provider interactions in health facilities that resulted in prescription or dispensing of antibiotics across strata of key variables Country All India China Kenya Proportion Proportion Proportion Proportion Variable n/N (95% CI) n/N (95% CI) n/N (95% CI) n/N (95% CI) At least one antibiotic 2734/5863 46.6 (33.4 to 53.9) 2392/4798 49.9 (40.8 to 54.5) 259/899 28.8 (17.8 to 50.8) 83/166 50.0 (42.2 to 57.8) Antibiotics, n 0 3129/5863 53.4 (46.1 to 66.6) 2406/4798 50.1 (45.4 to 57.9) 640/899 71.2 (49.2 to 71.2) 83/166 50.0 (42.2 to 57.8) 1 2465/5863 42.0 (31.4 to 47.4) 2159/4798 45.0 (39.8 to 48.2) 229/899 25.5 (25.5 to 42.8) 77/166 46.4 (39.2 to 54.2) 2 260/5863 4.4 (1.6 to 6.5) 225/4798 4.7 (1.4 to 6.6) 29/899 3.2 (3.2 to 7.7) 6/166 3.6 (1.2 to 6.6) 3 9/5863 0.2 (0.02 to 0.3) 8/4798 0.2 (0.03 to 0.3) 1/899 0.1 (0.1 to 0.3) 0/166 0 Health facility location Urban 1653/3018 54.8 (50.0 to 55.2) 1570/2852 55.0 (53.0 to 55.2) – – 83/166 50.0 (42.8 to 57.8) Rural 1081/2845 38.0 (26.6 to 48.1) 822/1946 42.2 (39.0 to 46.7) 259/899 28.8 (17.8 to 50.8) – – Healthcare sector Public 443/1321 33.5 (20.6 to 50.8) 156/367 42.5 (37.6 to 47.7) 259/899 28.8 (17.8 to 50.8) 28/55 50.9 (38.2 to 63.6) Private 2291/4542 50.4 (40.8 to 54.5) 2236/4431 50.5 (50.2 to 54.5) – – 55/111 49.5 (40.1 to 51.6) Provider qualification Qualified 1186/1906 62.2 (45.4 to 71.3) 1115/1768 63.1 (44.6 to 71.8) 71/138 51.4 (42.8 to 59.4) NA NA Non- qualified 1358/3191 42.6 (38.7 to 48.6) 1277/3030 42.1 (37.8 to 47.9) 81/161 50.3 (42.9 to 57.8) NA NA Clinical presentation Angina 169/955 17.7 (12.2 to 28.3) 115/598 19.2 (16.8 to 21.1) 29/315 9.2 (5.9 to 12.4) 25/42 59.5 (45.2 to 73.8) Asthma 330/718 46.0 (44.0 to 50.2) 308/676 45.6 (43.5 to 49.0) – – 22/42 52.4 (38.1 to 66.7) Child diarrhoea 490/997 49.1 (33.4 to 67.9) 399/672 59.4 (50.5 to 75.0) 78/285 27.4 (21.8 to 32.5) 13/40 32.5 (17.5 to 45.5) Presumptive TB 1293/2253 57.4 (51.3 to 58.6) 1118/1912 58.5 (58.4 to 59.3) 152/299 50.8 (44.8 to 56.2) 23/42 54.8 (39.3 to 69.0) Confirmed TB 194/404 48.0 (47.7 to 50.0) 194/404 48.0 (47.7 to 50.0) – – – – Presumptive MDR- TB 258/536 48.1 (48.0 to 48.1) 258/536 48.1 (48.0 to 48.1) – – – – Patient referred for further evaluation* Yes 101/767 13.2 (9.4 to 20.4) 65/498 13.1 (9.7 to 17.4) 33/263 12.5 (7.3 to 31.6) 3/6 50.0 (16.7 to 83.3) No 2163/4384 50.7 (35.6 to 57.5) 1928/3628 53.1 (38.4 to 58.0) 226/636 35.5 (23.3 to 55.4) 67/120 55.8 (47.5 to 64.2) *All child diarrhoea cases from India and Kenya (n=712) were excluded from this analysis because children were not directly assessed by the provider. MDR- TB, multidrug resistant tuberculosis; NA, not available; TB, tuberculosis. BMJ Global Health Figure 1 Crude percentage of SP—provider interactions resulting in antibiotic prescription/dispensing, by country and selected conditions (pharmacy- based studies are not included). SP, standardised patient; TB, tuberculosis. studies is not possible owing to the very different contexts result in antibiotic prescription as compared with other involved and particularly to the different types of cases clinical conditions. Among the two pharmacy- based SP 10 11 that were examined. As observed in studies from health- studies done in India, the proportion of antibiotic care facilities, subjects presenting to pharmacies with dispensing was 26.6% and 3.6%, respectively. symptoms suggestive of TB were generally more likely to Although our focus was on LMICs, the overuse of anti- receive an antibiotic as compared with other conditions. biotics is not confined to LMICs. Large population-based The average proportion of Watch- antibiotics (predom- cohort data have shown that antibiotic overuse in ambu- inantly quinolones and cephalosporins) dispensed across latory settings across the United States was 30% among the three cities was 49.4% (95% CI: 43.9 to 54.4), ranging children and 17% among adults with certain respiratory from 24.0% (95% CI: 15.0 to 32.0) in Mumbai to 60.9% tract illnesses for which antibiotics are not indicated (95% CI: 55.1 to 67.1) in Patna. A deeper evaluation of (eg, asthma, allergies, acute bronchitis or bronchiol- antibiotic dispensing in Udupi district is limited by the itis). An analysis of antibiotic prescription practices small sample size. Only 55 antibiotics were dispensed based on administrative data from Ontario, Canada, across 1522 interactions, thus making subgroup anal- recently reported an overall rate of unnecessary antibi- yses less meaningful. Yet, it is worth highlighting that otic prescribing in primary care of 15.4%, though much nearly half of these antibiotics were discouraged FDCs higher percentages were observed for some respiratory of two antibiotics, whereas the remaining were almost conditions such as acute bronchitis (52.6%). However, equally distributed among Access- and Watch- groups. a direct comparison with higher income countries cannot More details regarding the types of antibiotics dispensed be done due to differences in study methodologies and across pharmacies in both studies are presented in online local epidemiology. supplementary file 2. Nearly 50% of all antibiotics prescribed in the context of India SP studies belonged to the ‘Watch’ list, with a peak of 80% among patients presenting with symptoms DISCUSSION suggestive of TB, which is consistent with national antibi- Our analysis of past SP studies involving 4798 SP–provider otic sales. Watch- antibiotics accounted for almost 33% interactions in India showed that healthcare providers of all antibiotics across China SP studies, but this is likely in primary care settings prescribed antibiotics to about underestimated because nearly one quarter of all antibi- half (49.9%) of patients presenting with clinical condi- otics could not be classified due to insufficient informa- tions that do not require antibiotics. Antibiotic overpre- tion. Of note, we observed a large use of cephalosporins scribing was found to be similar (50% of SP–provider (presumably second or third generation ones), which is interactions) in a small SP study carried out in Nairobi, in line with previous findings from drug sales analyses Kenya. Pooled data from two studies conducted in China and prescription audits conducted in various parts of showed lower levels of antibiotic overuse (28.8%), but it 2 23 24 China. In contrast, the small SP study conducted should be noted that percentages differed substantially in Nairobi revealed that over 85% of prescribed antibi- across individual studies, likely reflecting the different type of cases being involved. In fact, SP–provider interac- otics were from the ‘Access’ group, and half of these were tions involving presumptive TB cases were more likely to either trimethoprim/sulfamethoxazole or amoxicillin. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 7 BMJ Global Health 8 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 Table 3 Frequency of antibiotics prescribed/dispensed in health facilities across study countries, overall and according to both the AWaRe and ATC classifications India All settings Urban India Rural India China Drug type N Proportion (95% CI) N Proportion (95% CI) N Proportion (95% CI) N Proportion (95% CI) Any antibiotic 2768 – 1896 – 872 – 301 – AWaRe classification Access 876 31.6 (30.0 to 38.9) 584 30.8 (29.8 to 30.8) 292 33.5 (29.9 to 37.1) 126 41.9 (36.2 to 47.2) Watch 1317 47.6 (26.8 to 54.0) 1041 54.9 (54.9 to 55.4) 276 31.7 (21.2 to 40.3) 99 32.9 (27.6 to 37.9) Reserve 23 0.8 (0.5 to 1.8) 8 0.4 (0.4 to 0.5) 15 1.7 (1.0 to 2.1) 1 0.3 (0.3 to 1.3) Discouraged 334 12.1 (4.3 to 36.3) 50 2.6 (2.6 to 2.8) 284 32.6 (25.1 to 44.8) 1 0.3 (0.3 to 1.3) Not available* 218 7.9 (5.4 to 10.8) 213 11.2 (11.2 to 11.5) 5 0.57 (0.3 to 1.0) 74 24.6 (19.9 to 29.2) ATC classification Penicillin 711 25.7 (18.8 to 27.0) 535 28.2 (27.7 to 28.2) 176 20.2 (17.6 to 21.7) 68 22.6 (17.6 to 27.2) Cephalosporin 361 13.0 (8.2 to 14.6) 294 15.0 (14.9 to 15.0) 76 8.7 (7.8 to 10.7) 75 24.9 (20.9 to 29.2) First generation 21 0.8 (0.6 to 1.8) 9 0.5 (0.47 to 0.51) 12 1.4 (1.1 to 2.1) 0 0 Second generation 22 0.8 (0.2 to 1.1) 20 1.1 (1.1 to 1.2) 2 0.2 (0.2 to 0.4) 7 2.3 (0.7 to 4.0) Third generation 318 11.5 (7.1 to 12.9) 256 13.5 (13.3 to 13.5) 62 7.1 (6.4 to 8.1) 1 0.3 (0.3 to 1.0) Not available* 0 0 0 0 0 0 67 22.3 (18.3 to 26.6) Macrolide 425 15.4 (4.1 to 19.3) 389 20.5 (20.4 to 21.3) 36 4.1 (4.1 to 4.3) 60 19.9 (15.6 to 24.3) Quinolone 520 18.8 (16.6 to 24.2) 354 18.7 (18.5 to 18.7) 166 19.0 (18.5 to 26.8) 37 12.3 (9.0 to 15.9) Tetracycline 67 2.4 (1.7 to 4.6) 34 1.8 (1.4 to 1.8) 33 3.8 (3.0 to 4.1) 0 0 Imidazole† 61 2.2 (0.8 to 7.1) 1 0.05 (0.05 to 0.06) 60 6.9 (6.3 to 7.5) 1 0.3 (0.3 to 1.3) Sulfonamide‡ 18 0.7 (0.2 to 1.9) 3 0.16 (0.16 to 0.17) 15 1.7 (0.9 to 2.1) 9 3.0 (1.3 to 5.0) Aminoglycoside 6 0.2 (0.1 to 1.0) 0 0 6 0.7 (0.7 to 1.3) 45 15.0 (11.3 to 18.6) Combinations§ 289 12.1 (5.1 to 34.2) 50 2.6 (2.6 to 2.8) 284 32.6 (25.1 to 34.2) 1 0.3 (0.3 to 1.3) Antimycobacterial 229 8.3 (0.3 to 10.9) 226 11.9 (11.9 to 12.2) 3 0.3 (0.2 to 0.5) 1 0.3 (0.3 to 1.3) Other antibiotics 36 1.3 (1.0 to 2.4) 19 1.0 (0.1 to 1.0) 17 1.9 (1.8 to 2.6) 4 1.3 (0.3 to 2.7) The unit of analysis is the individual drug, not the standardised patient–provider interaction. *For these drugs, only the antibiotic class (eg, cephalosporin) was available. †Only metronidazole was prescribed/dispensed. ‡Only trimethoprim–sulfamethoxazole was prescribed/dispensed. §This category does not include combinations of antimycobacterial drugs. ATC, Anatomical–Therapeutic–Chemical; AWaRe, Access–Watch–Reserve. BMJ Global Health Figure 2 Factors associated with antibiotic prescribing/dispensing in health facilities in India. Covariate-adjusted pr evalence ratios and their 95% CIs estimated from a hierarchical Poisson model are reported. SP, standardised patient; TB, tuberculosis. This is in line with that observed in another SP study draw meaningful conclusions on antibiotic prescribing carried out in urban public primary healthcare facilities patterns in the area. in South Africa, where 10/119 (8.4%) interactions for Discouraged FDCs of antibiotics were commonly given presumptive TB resulted in antibiotic prescriptions, all of in India but not in other settings, accounting for 10.4% of which belonged to the access group. As with the Nairobi the total. FDCs were finally banned in India in September study, however, the small sample size does not allow to 2018, thus leaving hope for a change in the near future. Table 4 Antibiotic dispensing in Indian pharmacies Study setting Udupi district, Karnataka Mumbai, Delhi and Patna (n=1522) (n=1200) Variable n/N Proportion (95% CI) n/N Proportion (95% CI) Number of antibiotics 1 55/1522 3.6 (2.6 to 4.6) 294/1,00 24.5 (22.2 to 27.0) 2 0 0 25/1200 2.1 (1.3 to 2.9) Pharmacy location Urban 25/744 3.3 (2.2 to 4.7) 319/1200 26.6 (24.2 to 29.2) Rural 30/778 3.9 (2.7 to 5.2) – – Clinical presentation Adult with URI 11/250 4.4 (2.0 to 7.2) – – Adult with diarrhoea 12/259 4.6 (2.3 to 7.1) – – Adult with fever (malaria suspect) 10/252 4.0 (1.6 to 6.3) – – Child with URI 0/252 0 – – Child with diarrhoea 20/250 8.0 (4.8 to 11.2) – – Child with fever (malaria suspect) 2/259 0.8 (0.4 to 1.9) – – Adult with presumptive TB – – 221/599 36.9 (33.1 to 40.7) Adult with confirmed TB – – 98/601 16.3 (13.5 to 19.3) Patient referred to health provider Yes 15/710 2.1 (1.1; 3.1) 41/497 8.2 (5.8; 10.9) No 40/812 4.9 (3.6; 6.4) 278/703 39.5 (36.1; 43.2) TB, tuberculosis; URI, upper respiratory illness. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 9 BMJ Global Health Alarmingly, we observed the use of some ‘Reserve’ anti- Of note, available data originated from a range of biotics in primary care settings. In India, oral colistin was geographical areas with different sociocultural and prescribed for paediatric diarrhoea, and faropenem was economic profiles and could be generalisable to similar given to one patient with presumptive TB. This is very contexts in India. For all these reasons, the representa- concerning as parenteral colistin is the last resort drug tiveness of our findings is very good, and selection bias is for treatment of extremely drug-resistant Gram- negative likely negligible due to the robust mapping and sampling infections, and using the oral formulation could drive approach used across all SP studies. resistance in the community. Similarly, faropenem is an There are limitations in our study. First, the SP study data oral penem antibiotic which has been shown to cause from China and Kenya were limited and lacked general- cross- resistance to intravenous carbapenems. In China isability. Second, our analyses were restricted to overpre- SP studies, one presumptive TB case received aztreonam, scription and to a limited number of clinical scenarios. indicated for treatment of serious infections due to drug- Third, we could not investigate other important forms resistant Gram- negative bacteria. of inappropriate antibiotic use, such as the choice of the According to our findings from India, antibiotic overuse incorrect drug and dosage to treat a given infection. This was particularly common in rural areas, among qualified is an intrinsic limitation that arises from the type of tracer providers and for patients presenting with presumptive conditions used across SP studies so far. Although the SP TB. Besides leading to potentially dangerous diagnostic methodology was initially implemented to assess overall 28 29 delays, the unnecessary use of antibiotics causes quality of care in LMICs and to evaluate educational/ harms to the patient in terms of drug- associated adverse behavioural programmes in high-income countries, this events and increased out- of- pocket costs. approach is being increasingly adopted to gain insight While normative boundaries may partly explain why into medication use, and especially drug dispensing prac- qualified providers prescribed more antibiotics than tices among pharmacists. Data recording systems in SP non- qualified ones as observed in our analyses for studies are therefore improving in order to facilitate the India, the widespread overuse of antibiotics suggests collection of key details regarding medications that were that important training gaps likely exist. However, harder to capture from studies whose main objective was prescribing behaviours among healthcare providers not related to drug use. also depend on a number of other factors, including In conclusion, the prevalence of antibiotic overpre- financial incentives from pharmaceutical companies, scribing estimated from SP studies ranged from 29% patient expectations and requests, or just old habits that in China to 50% in India and Kenya, and Watch anti- 8 30 31 are hard to die. biotics accounted for a large proportion of antibiotics The biggest strength of our study lies in the nature prescribed in both India and China. Combining the SP and quality of the data used to investigate the extent and methodology with new tracer conditions would allow patterns of antibiotic overprescribing. Although previous overcoming many of the typical limitations of most research had already highlighted that Watch group anti- studies aimed at evaluating inappropriate antibiotic use biotics are highly prescribed across India and China, in greater detail. SPs represent a unique opportunity to such studies could not provide a clear picture of inap- further explore prescription practices among healthcare propriate antibiotic use owing to the limited amount of providers, including the management of common infec- clinical information available from prescription audits tious diseases, such as pneumonia or urinary tract infec- 32–34 and evaluations of drug sales data. Among the main tions, that contribute substantially to the overall antibiotic advantages of using SPs to evaluate prescription practices use in primary care. Future studies also need to focus on is the fact that tracer conditions are standardised. In all untangling the channels for antibiotic overprescription studies included in our analyses, such conditions were and better understand the determinants of such practice very common illnesses that are frequently encountered among public and private healthcare providers in various in primary care and that require a well- defined diagnostic contexts. and therapeutic management that does not involve anti- The extent of antibiotic overuse in primary care across biotic use. LMICs is a serious concern and requires targeted anti- Furthermore, representative samples of healthcare microbial stewardship interventions aimed at improving providers from public and/or private sectors were rational and locally adapted prescribing practices. An selected in all SP studies conducted in India, with the active involvement of private providers in all such inter- only exception of one relatively small pilot study in ventions would be essential to ensure uptake, particularly Delhi. In this pooled dataset, private practitioners were in countries where the private sector plays a major role in much more represented than public providers, but we healthcare. Greater efforts are also necessary to develop lacked statistical power to make appropriate compari- and scale up accurate point- of- care tests that could guide sons between the two groups. Yet, this distribution well therapeutic choices where resources are scarce. Addi- reflects the fact that about 75% of outpatient visits in tional research is also required to evaluate whether anti- India take place in the private sector, with nearly 70% of biotic use (especially use of drugs such as azithromycin primary care in the country being delivered by informal and hydroxychloroquine) will dramatically increase as a 35 36 providers. consequence of the COVID-19 pandemic, and concerns 10 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 BMJ Global Health 3 Spivak ES, Cosgrove SE, Srinivasan A. Measuring appropriate have already been raised about the implications for antimicrobial use: attempts at opening the black box. Clin Infect Dis AMR. 2016;63:1639–44. 4 Smieszek T, Pouwels KB, Dolk FCK, et al. Potential for reducing inappropriate antibiotic prescribing in English primary care. J Author affiliations 1 Antimicrob Chemother 2018;73:ii36–43. Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, 5 Kwan A, Daniels B, Bergkvist S, et al. Use of standardised patients Québec, Canada for healthcare quality research in low- and middle- income countries. McGill International TB Centre, McGill University, Montreal, Québec, Canada BMJ Glob Health 2019;4:e001669. McCourt School of Public Policy, Georgetown University, Washington, District of 6 Das J, Kwan A, Daniels B, et al. Use of standardised patients to Columbia, USA assess quality of tuberculosis care: a pilot, cross- sectional study. Lancet Infect Dis 2015;15:1305–13. School of Public Health, University of California Berkeley, Berkeley, California, USA 5 7 Kwan A, Daniels B, Saria V, et al. Variations in the quality of Division of Infectious Diseases, Department of Medicine, Washington University in tuberculosis care in urban India: a cross- sectional, standardized Saint Louis, Saint Louis, Missouri, USA patient study in two cities. PLoS Med 2018;15:e1002653. Dahdaleh Institute of Global Health Research, York University, Toronto, Ontario, 8 Das J, Chowdhury A, Hussam R, et al. The impact of training Canada informal health care providers in India: a randomized controlled trial. Centre for the AIDS Programme of Research in South Africa (CAPRISA), University Science 2016;354:aaf7384. 9 Das J, Holla A, Das V, et al. In urban and rural India, a standardized of KwaZulu- Natal, Durban, KwaZulu- Natal, South Africa 8 patient study showed low levels of provider training and huge quality Centre for Policy Research, New Delhi, Delhi, India gaps. Health Aff 2012;31:2774–84. Manipal McGill Program for Infectious Diseases, Manipal Centre for Infectious 10 Satyanarayana S, Kwan A, Daniels B, et al. Use of standardised Diseases, Manipal Academy of Higher Education, Manipal, Karnataka, India patients to assess antibiotic dispensing for tuberculosis by pharmacies in urban India: a cross- sectional study. Lancet Infect Dis Twitter Giorgia Sulis @giorgiasulis, Benjamin Daniels @_bbdaniels, Ada Kwan 2016;16:1261–8. @kwantada, Amrita Daftary @DaftaryAmrita and Madhukar Pai @paimadhu 11 Nafade V, Huddart S, Sulis G, et al. Over- The- Counter antibiotic dispensing by pharmacies: a standardised patient study in Udupi Contributors GS, MP, JD and BD designed the study. GS and BD performed the district, India. BMJ Glob Health 2019;4:e001869. data cleaning and coding. GS analysed the data and prepared the first draft of the 12 Daniels B, Dolinger A, Bedoya G, et al. Use of standardised patients paper. All authors critically revised the manuscript until final approval. to assess quality of healthcare in Nairobi, Kenya: a pilot, cross- sectional study with international comparisons. BMJ Glob Health Funding Most studies included in our analyses were funded by the Bill and 2017;2:e000333. Melinda Gates Foundation (OPP1091843). GS is a recipient of a Richard H. 13 Sylvia S, Shi Y, Xue H, et al. Survey using incognito standardized Tomlinson Doctoral Fellowship (McGill University), and MP holds a Tier 1 Canada patients shows poor quality care in China's rural clinics. Health Research Chair from Canadian Institutes of Health Research. Policy Plan 2015;30:322–33. 14 Sylvia S, Xue H, Zhou C, et al. Tuberculosis detection and the Competing interests MP is on the editorial boards of BMJ Global Health. All other challenges of integrated care in rural China: a cross- sectional authors declare that they have no conflicts of interest. standardized patient study. PLoS Med 2017;14:e1002405. Patient and public involvement Patients and/or the public were not involved in 15 Xue H, Shi Y, Huang L, et al. Diagnostic ability and inappropriate antibiotic prescriptions: a quasi- experimental study of primary the design, conduct, reporting or dissemination plans of this research. care providers in rural China. J Antimicrob Chemother Patient consent for publication Not required. 2019;74:256–63. 16 Sharland M, Gandra S, Huttner B, et al. Encouraging AWaRe- Ethics approval Each study had been conducted after ethics approval. Our study ness and Discouraging inappropriate antibiotic use- the new 2019 was approved by the McGill Faculty of Medicine Institutional Review Board (IRB essential medicines list becomes a global antibiotic stewardship review number: A04-B19- 20B (20-04-053)), and all primary studies we included tool. Lancet Infect Dis 2019;19:1278–80. had their own independent ethics approvals. 17 WHO. Anatomical therapeutic chemical (ATC) classification system. Oslo, Norway: WHO Collaborating Centre for Drug Statistics Provenance and peer review Not commissioned; externally peer reviewed. Methodology, 2020. https://www. whocc. no/ atc_ ddd_ index/ 2020 Data availability statement Data are available upon request. All data relevant 18 DiCiccio TJ, Efron B. Bootstrap confidence intervals. Stat Sci to the study are included in the article or uploaded as supplementary information. 1996;11:189–228. Parts of the data used for our analyses are available through original publications 19 Stewart GB, Altman DG, Askie LM, et al. Statistical analysis of individual participant data meta- analyses: a comparison of methods as reported in the reference list. Additional data included in this study are available and recommendations for practice. PLoS One 2012;7:e46042. upon request. 20 Fleming- Dutra KE, Hersh AL, Shapiro DJ, et al. Prevalence of Open access This is an open access article distributed in accordance with the inappropriate antibiotic prescriptions among US ambulatory care Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits visits, 2010-2011. JAMA 2016;315:1864–73. 21 Schwartz KL, Langford BJ, Daneman N, et al. Unnecessary antibiotic others to copy, redistribute, remix, transform and build upon this work for any prescribing in a Canadian primary care setting: a descriptive analysis purpose, provided the original work is properly cited, a link to the licence is given, using routinely collected electronic medical record data. CMAJ Open and indication of whether changes were made. See: https:// creativecommons. org/ 2020;8:E360–9. licenses/ by/ 4. 0/. 22 Klein EY, Milkowska- Shibata M, Tseng KK, et al. Assessment of who antibiotic consumption and access targets: 2000-2015. Lancet ORCID iDs Infectious Diseases 2020:S1473-3099(20)30332-7. Giorgia Sulis http:// orcid. org/ 0000- 0001- 6641- 0094 23 Wushouer H, Tian Y, Guan X- D, et al. Trends and patterns of Benjamin Daniels http:// orcid. org/ 0000- 0001- 9652- 6653 antibiotic consumption in China's tertiary hospitals: based on Ada Kwan http:// orcid. org/ 0000- 0003- 4889- 9433 a 5 year surveillance with sales records, 2011-2015. PLoS One Amrita Daftary http:// orcid. org/ 0000- 0003- 2275- 3540 2017;12:e0190314. 24 Lin H, Dyar OJ, Rosales- Klintz S, et al. Trends and patterns of Jishnu Das http:// orcid. org/ 0000- 0002- 5909- 3585 antibiotic consumption in Shanghai municipality, China: a 6 year Madhukar Pai http:// orcid. org/ 0000- 0003- 3667- 4536 surveillance with sales records, 2009-14. J Antimicrob Chemother 2016;71:1723–9. 25 Christian CS, Gerdtham U- G, Hompashe D, et al. Measuring quality gaps in TB screening in South Africa using standardised patient analysis. Int J Environ Res Public Health 2018;15:ijerph15040729. REFERENCES 26 Bergen PJ, Smith NM, Bedard TB, et al. Rational combinations 1 Huttner BD, Catho G, Pano- Pardo JR, et al. COVID-19: don't of polymyxins with other antibiotics. Adv Exp Med Biol neglect antimicrobial stewardship principles! Clin Microbiol Infect 2019;1145:251–88. 2020;26:808–10. 27 Gandra S, Choi J, McElvania E, et al. Faropenem resistance causes 2 Sulis G, Adam P, Nafade V, et al. Antibiotic prescription practices in vitro cross- resistance to carbapenems in ESBL- producing in primary care in low- and middle- income countries: a systematic Escherichia coli. Int J Antimicrob Agents 2020;55:105902. review and meta- analysis. PLoS Med 2020;17:e1003139. Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393 11 BMJ Global Health 28 Getnet F, Demissie M, Assefa N, et al. Delay in diagnosis of middle- income country setting: a repeated cross- sectional study in pulmonary tuberculosis in low- and middle- income settings: Vellore, South India. J Antimicrob Chemother 2013;68:229–36. systematic review and meta- analysis. BMC Pulm Med 2017;17:202. 34 Farooqui HH, Mehta A, Selvaraj S. Outpatient antibiotic prescription 29 Robson J, Ayerbe L, Mathur R, et al. Clinical value of chest pain rate and pattern in the private sector in India: evidence from medical presentation and prodromes on the assessment of cardiovascular audit data. PLoS One 2019;14:e0224848. disease: a cohort study. BMJ Open 2015;5:e007251. 35 Mackintosh M, Channon A, Karan A, et al. What is the 30 Currie J, Lin W, Meng J. Addressing antibiotic abuse in China: an private sector? understanding private provision in the health experimental audit study. J Dev Econ 2014;110:39–51. systems of low- income and middle- income countries. Lancet 31 Fitzpatrick A. Do Informed Consumers Reduce the Price and 2016;388:596–605. Prevalence of Counterfeit Drugs ?Evidence from the Antimalarial 36 Gopalkrishnan R, Walia K, Ohri VC. Treatment guidelines for Market, 2015. antimicrobial use in common syndromes. 2nd ed. New Delhi, India: 32 Kotwani A, Holloway K. Antibiotic prescribing practice for acute, Indian Council of medical research (ICMR), 2019. uncomplicated respiratory tract infections in primary care settings in 37 WHO. Record number of countries contribute data revealing New Delhi, India. Trop Med Int Health 2014;19:761–8. disturbing rates of antimicrobial resistance. Geneva, Switzerland: 33 Chandy SJ, Thomas K, Mathai E, et al. Patterns of antibiotic use in the community and challenges of antibiotic surveillance in a lower- World Health Organization (WHO), 2020. 12 Sulis G, et al. BMJ Global Health 2020;5:e003393. doi:10.1136/bmjgh-2020-003393

Journal

BMJ Global HealthBritish Medical Journal

Published: Sep 15, 2020

Keywords: other study designtreatmentepidemiology

References