Use of antibiotics in children younger than two years in eight countries: a prospective cohort study

1Antibiotics can be a lifesaving treatment for children with bacterial infections and are the most commonly prescribed therapy among all medications given to children. However, antibiotics can also result in adverse events, drug toxicity and detrimental effects on the gut microbiota, and enteric immune system., Furthermore, both at the individual and population levels, antibiotic overuse drives the development and transmission of antimicrobial resistance., International guidelines for the treatment of childhood illnesses recommend antibiotic treatment for diarrhoea with bloody stools and for acute lower respiratory tract infections, but not for non-bloody diarrhoea and for upper respiratory infections., Interventions to promote rational antibiotic use are critical for preserving the effectiveness of available drugs.– Conversely, in low-resource settings, the high burden of bacterial causes of diarrhoea in children, has led to proposals for antibiotics to be used more widely for the treatment of diarrhoea even in the absence of dysentery.– Antibiotics may also be a potential intervention for malnutrition and environmental enteropathy.

2Differences in antibiotic use practices around the world reflect differences in local medication policies, in barriers to access to care and in the preferences of health-care providers and mothers. The availability of antibiotics without a doctor’s prescription varies,, and laws to limit access to antibiotics are often poorly enforced.,– In some settings, drug shortages may be a major limiter of antibiotic use., Cultural preferences, such as high demand by mothers, also influence patterns of antibiotic use.,,, Even when health-care providers are aware of the appropriate indications for antibiotics, there can be differences between knowledge and practice.,

3Many studies of antibiotic use have been conducted in various health-care settings,– and in cross-sectional community-based surveys.– Nevertheless, high-resolution, systematic assessments of antibiotic use in prospective, observational cohort studies have not been reported. MAL-ED (Etiology, Risk Factors, and Interactions of Enteric Infections and Malnutrition and the Consequences for Child Health and Development Project) was a multisite birth cohort study conducted in eight sites in different countries of South America, sub-Saharan Africa and Asia. This study included community-based surveillance for antibiotic use and provides an opportunity to compare antibiotic use patterns across diverse low-resource sites. We aimed to describe the frequency of antibiotic use by children in the first two years of life; determine the characteristics associated with antibiotic use; and estimate the proportions of diarrhoea and respiratory illness episodes treated with antibiotics, as reported by mothers in the MAL-ED study.

4The MAL-ED study design and cohort characteristics have been previously described. Briefly, the study was conducted at sites in eight different countries: Dhaka (Bangladesh), Fortaleza (Brazil), Vellore (India), Bhaktapur (Nepal), Naushahro Feroze (Pakistan), Loreto (Peru), Venda (South Africa) and Haydom (United Republic of Tanzania). Healthy children were enrolled between November 2009 and February 2012 within 17 days of birth. Two-year follow-up for all enrolled children was completed in February 2014. The criteria for enrolment were children without severe or chronic conditions, enteropathy or hospitalization, and enrolment weight ≥ 1500 g.

5Surveillance for illnesses and antibiotic use was conducted twice per week by fieldworkers at home visits until the child was two years of age or was lost to follow-up. Children were referred to locally available care, generally a local clinic, when ill. Fieldworkers asked the mother (or other caregiver) to report all oral or injected antibiotics given to their child on each day since the previous visit and to show the medication packaging to confirm the antibiotic and class. If packaging were not available, fieldworkers documented antibiotic use from any paperwork provided by health-care providers. When a mother reported seeking medical care or medications for their child, fieldworkers recorded separately prescribed medicines in medical care report forms. Socioeconomic characteristics were assessed through twice-yearly questionnaires.

6To validate mothers’ reports of antibiotic use, we randomly selected 4409 of the fieldworkers’ medical care report forms (including at least 200 records of antibiotics per site) and extracted all antibiotic information. We assessed the concordance between mother-reported antibiotic use and antibiotic use as documented on the medical care report forms.

7All sites received ethical approval from their respective government, local institution and collaborating institution ethical review boards. We obtained informed consent from the mother of each child.

8We counted distinct antibiotic courses when separated by at least two antibiotic-free days. The results were insensitive to an alternative definition using three antibiotic-free days; only 580 (3.4%) of 16 913 courses occurred within three days. The duration of antibiotic courses was defined as the total number of days on which antibiotics were received, assuming antibiotics were not received on missed surveillance days (2.0% of all surveillance days). A child was classified as exposed to high antibiotic use if he or she received more than or equal to the median number of courses received by children at his or her study site in the first two years of life.

9We based illness definitions on Integrated Management of Childhood Illness guidelines. Non-bloody diarrhoea was defined as mother’s reports of three or more loose stools in 24 hours. Bloody diarrhoea was defined as mother’s report of at least one loose stool with visible blood. Respiratory illness was defined as cough or shortness of breath. Acute lower respiratory tract illness was defined as cough or shortness of breath with a rapid respiratory rate determined by fieldworkers (defined by the average of two measurements per day that were: > 60 breaths per minute when the child was < 2 months old; > 50 breaths per minute at age 2 months to 1 year; and > 40 breaths per minute at age ≥ 1 year). If antibiotics were taken during any day of the illness episode, the episode was classified as treated with antibiotics.

10Socioeconomic status was described using the child’s average score on the WAMI index based on: household access to improved water and sanitation; wealth measured by eight household assets; mother’s education; and monthly household income. Crowding was defined as the mean number of people per room. Improved water and sanitation were defined following World Health Organization (WHO) guidelines.

11We calculated the incidence of antibiotic use as the number of courses divided by the number of at-risk surveillance days. The incidence over the first two years of life was estimated using a pooled logistic regression model with a restricted quadratic spline for age with seven knots. Cumulative incidence curves were constructed non-parametrically as the inverse of Kaplan–Meier estimates.

12We adjusted for the following: study site; the proportion of days ill with diarrhoea, cough, fever, vomiting and fieldworker-confirmed acute lower respiratory infection in the first two years of life; and the interaction between this proportion and study site. To estimate the associations between overall antibiotic use and child and household characteristics we used linear regression for the proportion of days on antibiotics and log-binomial regression for risk of high antibiotic use.

13We then described the frequency of treatment for diarrhoea and respiratory illnesses. We estimated the associations between the characteristics of those episodes and antibiotic treatment using log-binomial regression, adjusting for study site and other episode characteristics. We also accounted for correlations between episodes in the same child using generalized estimating equations with a robust variance estimator. Among treated episodes, we used these log-binomial models to estimate the associations between antibiotic class chosen and episode and child characteristics.

14We included 2134 children in the MAL-ED cohort who were surveyed for antibiotic use for any illness on at least one day in the first two years of life. Over a mean of 631 days of observation per child (1 346 388 total days of observation), 16 913 courses (100 342 total days) of antibiotics were recorded. This corresponded to an overall average antibiotic use of 4.9 courses per child per year. The median duration of antibiotic courses was 5 days (interquartile range: 3 to 7). Extended courses of antibiotics were rare; only 53 (0.3%) courses had durations longer than 1 month.

15The magnitude of use differed across the eight sites (Fig. 1). Frequency of use was highest at the site in Naushahro Feroze (an average of 11.9 courses per child-year) and in Dhaka (10.3 courses per child-year). In contrast, the use was ≤ 1.0 course per child-year in Fortaleza and Venda, respectively. Antibiotic use peaked between 6 and 12 months of age in all sites, and peaked again in the second year of life in Loreto, Vellore and Fortaleza.

16Early antibiotic use was common even in the first 6 months of life. In Dhaka and Naushahro Feroze, more than 98.0% of children followed until at least age 6 months had received antibiotics by that age (Fig. 2). More than half of children had received antibiotics by age 6 months in Bhaktapur, Haydom, Loreto and Vellore. Children in Naushahro Feroze were exposed to antibiotics on 32 345 (17.5%) of 152 176 observed child-days in the first two years of life, which corresponds to more than 4 months of cumulative antibiotic treatment. This proportion was even higher in the first 6 months of life (9383 child-days; 19.5%). The days of treatment in the first two years of life was lower in other sites, ranging from 1823 (1.3%) of 138 060 observed child-days in Fortaleza to 25 663 (15.5%) of 140 237 child-days in Dhaka.

17A total of 1741 children (81.6%) remained under surveillance until at least two years of age. More boys (543/888; 61.2%) than girls (436/853; 51.1%) received at or above the site-specific median number of antibiotic courses (Table 1). Adjusting for the proportion of days ill, the risk of high antibiotic use was 13% greater among boys compared with girls (risk ratio, RR: 1.13; 95% confidence interval, CI: 0.99 to 1.28). The association between antibiotic use and sex was driven by Naushahro Feroze (RR: 1.44; 95% CI: 1.02 to 2.03) and Bhaktapur (RR: 1.37; 95% CI: 0.9 to 1.97), with no associations at the other country sites. Socioeconomic status and income were associated with small increases in overall antibiotic use (Table 1).

18A total of 10 161 diarrhoea episodes were recorded among 1201 of the children; 4649 (45.8%) episodes were treated with antibiotics (Table 2). The use of antibiotics for the treatment of diarrhoea varied across sites from 10.6% of 180 episodes in Fortaleza to 59.1% of 3212 episodes in Naushahro Feroze.

19Mothers reported bloody stools in 499 (4.9%) diarrhoea episodes. A higher proportion of episodes of bloody diarrhoea (375; 75.2%) were treated with antibiotics than those without bloody stools (4274/9661; 44.2%; Table 3). Adjusting for study site and other characteristics of illness episodes, the risk ratio of antibiotic treatment was 1.50 (95% CI: 1.40 to 1.64) for episodes with bloody stools. Greater age at episode, duration, number of loose stools, and presence of fever, dehydration and vomiting were all independently associated with an increased risk of antibiotic treatment.

20Slightly fewer diarrhoea episodes in girls (2178/4897; 44.5%) were treated with antibiotics than those in boys (2471/5264; 46.9%; adjusted RR: 0.95; 95% CI: 0.90 to 1.00; Table 3). Higher socioeconomic status was associated with an increase in treatment (RR: 1.19; 95% CI: 1.10 to 1.29).

21The antibiotic class chosen for diarrhoea treatment varied across and within sites (Fig. 3). Diarrhoea episodes in the sites in Dhaka and Loreto were most often treated with macrolides, while metronidazole was the most common class for diarrhoea treatment in Bhaktapur, Haydom and Naushahro Feroze. Episodes in Fortaleza and Venda were mainly treated with sulfonamides and penicillins. Fluoroquinolones were rarely used for diarrhoea treatment in most sites; their use was most frequent in the South Asian sites of Dhaka and Vellore.

22Bloody diarrhoea episodes (75/499) were twice as likely to be treated with fluoroquinolones compared with non-bloody episodes (462/9661) (RR adjusted for study site: 2.01; 95% CI: 1.63 to 2.48). Bloody episodes were also 20 to 40% more likely to be treated with macrolides, cephalosporins and metronidazole, and were less likely to be treated with penicillins (51/499 bloody episodes and 870/9661 non-bloody episodes; RR adjusted for study site: 0.57; 95% CI: 0.44 to 0.75).

23Diarrhoea episodes among children from higher socioeconomic status were more likely than those occurring among children of lower socioeconomic status to be treated with metronidazole (RR per 0.5 difference in WAMI score: 1.17; 95% CI: 1.06 to 1.30) and macrolides (RR: 1.14; 95% CI: 0.93 to 1.39) and less likely to be treated with penicillins (RR: 0.79; 95% CI: 0.62 to 1.00). The child’s sex did not affect choice of antibiotic class for diarrhoea.

24Of 20 685 respiratory illness episodes among 1672 children, 8992 (43.5%) episodes were treated with antibiotics. Use of antibiotics was lowest in Fortaleza (35.5% of 431 episodes) and highest in Haydom (62.8% of 1399 episodes; Table 2).

25Fieldworkers confirmed 3943 (19.1%) episodes of respiratory illnesses had signs of acute lower respiratory tract illness. A higher proportion of these episodes (2384; 60.5%) were treated with antibiotics than were episodes of upper respiratory illness (6608/16742; 39.5%; Table 3). Adjusting for site, the risk ratio of antibiotic treatment for acute lower respiratory tract illness compared to upper respiratory illness was 1.53 (95% CI: 1.47 to 1.59).

26Respiratory illnesses were significantly more likely to be treated if the episode was of longer duration, and if independently there was fever, shortness of breath or rapid respiratory rate reported (Table 3). Treatment did not vary by age. Similar to diarrhoea treatment, respiratory illness episodes in girls were slightly less likely to be treated with antibiotics (4199/10 019; 41.9%) than those in boys (4793/10 666; 44.9%); after adjusting for study site and episode characteristics, the risk ratio was 0.94 (95% CI: 0.91 to 0.98). Higher socioeconomic status was also associated with a significant but small increase in treatment (adjusted RR: 1.10; 95% CI: 1.03 to 1.17).

27The antibiotic class used for respiratory illness treatment was fairly consistent across sites, with penicillins the most frequently used drug in all sites except Naushahro Feroze (Fig. 4). Cephalosporins were also often chosen in the South Asian sites of Naushahro Feroze, Vellore and Dhaka, while macrolides were also highly used in Dhaka. Because penicillins were almost exclusively used at several sites, a cross-site analysis of antibiotic classes by type of respiratory illness was not possible.

28Among antibiotic-treated respiratory illnesses, higher socioeconomic status was significantly associated with more use of macrolides (RR per 0.5 difference in WAMI score: 1.30; 95% CI: 1.09 to 1.56) and cephalosporins (RR: 1.48; 95% CI: 1.30 to 1.69) and correspondingly less use of penicillins (RR: 0.93; 95% CI: 0.85 to 1.01). There was no association between sex and antibiotic class used for respiratory illness.

29Overall concordance between antibiotics reported in the medical care report forms and caregiver-reported antibiotic use was high 1737 (85.8%) of 2042 forms analysed (Box 1).

30Medical care report forms were identified for 13 393 (79.2%) of all 16 913 antibiotic courses, and the validation sample yielded 2024 antibiotic prescriptions from 4409 forms. Concordance between use of antibiotics reported in the medical care report forms and mother-reported antibiotic use was high; on 1737 (85.8%) of forms the antibiotics corresponded with mothers’ reports on the same day. The concordance between antibiotic classes was also high: 95.1% (849/893) for penicillins, 94.8% (329/247) for cephalosporins, 86.6% (265/306) for macrolides, 85.2% (213/250) for metronidazole, 73.4% (177/241) for sulphonamides and 70.0% (42/60) for fluoroquinolones.

31Despite substantial heterogeneity, the frequent and early use of antibiotics in these low-resource settings is striking. In the most extreme case, children at the site in Pakistan were exposed to antibiotics on approximately one-fifth of days in their first 6 months of life. Antibiotic usage rates were higher in most sites than that reported for children aged 3 to 24 months in the United States of America in 2010 (0.9 to 1.7 courses per child-year). Higher antibiotic use in the South Asian sites compared with the African and South American sites is explained by more episodes of diarrhoea and respiratory illnesses as well as a higher proportion of illness episodes treated in this region. Differences in the proportion of episodes treated may be explained by site-specific treatment guidelines and availability of antibiotics. For example, access to antibiotics is less restricted in the South Asian sites,, while drug shortages are common in South Africa.

32Illness symptoms were strong drivers of antibiotic treatment for both diarrhoea and respiratory illnesses, demonstrating that treatment decisions were made rationally according to illness severity. However, many episodes of non-bloody diarrhoea (44.2%) and non-acute-lower respiratory-tract illness (39.5%) were treated with antibiotics, contrary to international recommendations against routine use of antibiotics for non-bloody diarrhoea and upper respiratory tract infections., These percentages were higher than the overall antibiotic treatment frequency of 37% reported for 17 693 paediatric inpatients from 226 hospitals in 41 countries. Because only 4.9% of diarrhoea episodes in our study were bloody, almost all antibiotic treatment of diarrhoea (4274/4649 episodes; 91.9%) was for non-bloody episodes, which is inconsistent with treatment guidelines. Similarly, only one-fifth of respiratory infections were fieldworker-confirmed acute lower respiratory tract infection and therefore 73.5% (6608/8992) treated episodes of respiratory illnesses were inconsistent with treatment guidelines.

33Conversely, while antibiotic treatment is recommended for dysentery and acute lower respiratory tract infection, only 75.2% of diarrhoea episodes with bloody stools and 60.5% of fieldworker-confirmed acute lower respiratory illness episodes were treated with antibiotics. Choice of antibiotic class for diarrhoea treatment was also inconsistent, suggesting diarrhoea treatment guidelines were not clearly followed. Fluoroquinolones and macrolides are recommended by WHO for the treatment of dysentery, but metronidazole was given most frequently, in more than one-third of dysentery cases. Presence of bloody stools was associated with a higher probability of appropriate treatment with fluoroquinolones and macrolides, but these drugs were still underused.

34The lower frequency of treatment among girls compared with boys after adjusting for illness burden and severity indicates that social factors also likely played a role in treatment decisions. Socioeconomic status was associated with frequency of antibiotic treatment as well as the antibiotic classes chosen for both diarrhoea and respiratory illnesses. Increased macrolides and cephalosporins use compared to less penicillins use among families with higher socioeconomic status corresponds to higher prices for these drugs, which may be a barrier to access for low-income families.

35This analysis provides a comprehensive description of antibiotic use across eight low-resource country settings, using data reported on every day of the first two years of life, a method which is superior to that of retrospective surveys. Using data on diarrhoea and respiratory illness symptoms, we were able to document treatment frequency and to comment on compliance with international guidelines. Mothers’ reports of antibiotic use ensured that we counted antibiotics taken (not only prescribed) and captured antibiotic use from all sources, including those that would not be included in clinic or prescription records, e.g. those from alternative health-care providers. We found high concordance between mothers’ reports and medical care report forms, as has been previously described, suggesting mothers’ reporting was reliable.

36The study was limited by incomplete details of antibiotic use, including specific drugs given and their formulations, prophylactic versus treatment use, how and where antibiotics were obtained, and the antibiotic class for the courses classified as unknown or other. We also inferred indication for treatment by concurrent illnesses and symptoms reported, without direct reports of the cause of treatment, which limits our ability to determine conclusively the appropriateness of treatment.

37Overall, antibiotic use early in life was common, and we found evidence of both overuse for the treatment of non-bloody diarrhoea and upper respiratory tract illnesses, and underuse for the treatment of bloody diarrhoea and acute lower respiratory tract infection. We also found evidence for sex and class differences in access to medicines. Rational antibiotic use programmes and promotion of illness-specific treatment guidelines may have the greatest impact in South Asia, where antibiotic use was highest. Planning of intervention studies involving antibiotic treatment needs to address complex, site-specific variations in use and consider the potentially high baseline frequency of antibiotic use. Further inquiry into the consequences of this highly prevalent exposure among children will be an important contribution to our understanding of child development in low-resource settings.