Incidence and etiology of clinically-attended, antibiotic-treated diarrhea among children under five years of age in low- and middle-income countries: Evidence from the Global Enteric Multicenter Study

Diarrhea is a leading cause of antibiotic consumption among children in low- and middle-income countries. While vaccines may prevent diarrhea infections for which children often receive antibiotics, the contribution of individual enteropathogens to antibiotic use is minimally understood. We used data from the Global Enteric Multicenter Study (GEMS) to estimate pathogen-specific incidence of antibiotic-treated diarrhea among children under five years old residing in six countries of sub-Saharan Africa and South Asia before rotavirus vaccine implementation. GEMS was an age-stratified, individually-matched case-control study. Stool specimens were obtained from children presenting to sentinel health clinics with newly-onset, acute diarrhea (including moderate-to-severe and less-severe diarrhea) as well as matched community controls without diarrhea. We used data from conventional and quantitative molecular diagnostic assays applied to stool specimens to estimate the proportion of antibiotic-treated diarrhea cases attributable to each pathogen. Antibiotics were administered or prescribed to 9,606 of 12,109 moderate-to-severe cases and 1,844 of 3,174 less-severe cases. Across all sites, incidence rates of clinically-attended, antibiotic-treated diarrhea were 12.2 (95% confidence interval: 9.0–17.8), 10.2 (7.4–13.9) and 1.9 (1.3–3.0) episodes per 100 child-years at risk at ages 6 weeks to 11 months, 12–23 months, and 24–59 months, respectively. Based on the recommendation for antibiotic treatment to be reserved for cases with dysentery, we estimated a ratio of 12.6 (8.6–20.8) inappropriately-treated diarrhea cases for each appropriately-treated case. Rotavirus, adenovirus serotypes 40/41, Shigella, sapovirus, Shiga toxin-producing Escherichia coli, and Cryptosporidium were the leading antibiotic-treated diarrhea etiologies. Rotavirus caused 29.2% (24.5–35.2%) of antibiotic-treated cases, including the largest share in both the first and second years of life. Shigella caused 14.9% (11.4–18.9%) of antibiotic-treated cases, and was the leading etiology at ages 24–59 months. Our findings should inform the prioritization of vaccines with the greatest potential to reduce antibiotic exposure among children.

as a multiplier indicating the ratio of adjusted attributable fraction estimates based on qPCR data and conventional diagnostic data. We used adjusted attributable fraction estimates aggregated across sites (excluding Kenya) to generate 6 ℓ (0|MSD) estimates (Table S4). Further stratification by site led to unstable estimates, in particular where analyses were underpowered to supply statistically-significant odds ratio estimates for individual pathogens within each site and age group ( Figure S1).
We defined '( ℓ ' ( (0|LSD) as the estimate of the fraction of clinically-attended, antibiotic-treated LSD attributable to pathogen ℓ in age group k that we would expect to obtain if quantitative molecular diagnostic data were available, given our estimates from conventional diagnostic data, '( ℓ & (0|LSD). We estimated this quantity as assuming the ratio 6 ℓ (0|MSD) of estimates based on differing diagnostic assays in clinically-attended, antibiotic-treated MSD cases would provide the best approximation of the same ratio in clinicallyattended, antibiotic-treated LSD cases. We present adjusted attributable fraction estimates for clinicallyattended, antibiotic-treated MSD and LSD based on conventional diagnostic data in Tables S7-S8. We present pooled, adjusted attributable fraction estimates for all clinically-attended, antibiotic-treated diarrhea based on conventional diagnostic data in Table S9.
Though we interpret '( ℓ & , '( ℓ ' , and '( ℓ ' ( as proportions, these estimators were not strictly bounded on [0,1] due to the continuous nature of the odds ratio estimate. We used the full distribution of our '( ℓ & and '( ℓ ' estimates (potentially below zero or greater than one) to define 6 ℓ (0) multipliers. Because proportion estimates below zero or above one do not have a natural interpretation, we report values of '( ℓ & , '( ℓ ' , and '( ℓ ' ( below zero or above one as zero or one, respectively. Thus, inclusion of 0% in the 95% confidence interval can be interpreted to indicate a pathogen is not significantly associated with clinicallyattended, antibiotic-treated MSD or LSD at the two-sided p<0.05 threshold; inclusion of 100% in the confidence interval can be interpreted to indicate that the results do not exclude the possibility that all cases are attributable to a given pathogen, at the two-sided p<0.05 threshold.

S3 INCIDENCE ESTIMATION FOR CLINICALLY-ATTENDED, ANTIBIOTIC-TREATED MSD AND LSD
Data from serial healthcare attitudes and utilization surveys provide an estimate of : )*+ ,)and : )*+ .)-, the number of acute MSD or LSD cases taken to sentinel hospital or health center within seven days of onset during the study period across N child-years at risk in the demographic surveillance site population of each setting, as well as ; ,)and ; .)-, the proportion of all incident MSD or LSD cases taken to a sentinel hospital or health center within seven days of disease onset. 3 Whereas primary analyses of the GEMS datasets used the formulations : )*+ ,)-/(; ,)-× <) and : )*+ .)-/(; .)-× <) to estimate total MSD and LSD incidence, respectively, we could not assume the likelihood of antibiotic treatment was equal among cases that did and did not seek care.

S4 INCIDENCE AND ADJUSTED ATTRIBUTABLE FRACTION ESTIMATION FOR ALL CLINICALLY-ATTENDED, ANTIBIOTIC-TREATED DIARRHEA
We aimed to estimate total incidence and pathogen-specific adjusted attributable fractions accounting for antibiotic use associated with both MSD and LSD cases. We defined the incidence of all clinicallyattended, antibiotic-treated diarrhea as Λ /00 (0) = Λ ,)-(0) + Λ .)-(0). We estimated the incidence all clinically-attended, antibiotic-treated attributable to each pathogen i, within each age group k, as We estimated the fraction of all clinically-attended, antibiotic-treated diarrhea attributable to each pathogen i, within each age group k, as M " /00 (0)/Λ /00 (0).
We repeated the analyses described in sections S2-S4 for both dysenteric and non-dysenteric MSD endpoints in order to estimate the incidence of clinically-attended, antibiotic-treated associated with dysenteric and non-dysenteric Shigella-attributable MSD. We summed non-dysenteric Shigellaattributable clinically-attended, antibiotic-treated incidence as Shigella-attributable non-dysenteric clinically-attended, antibiotic-treated MSD plus Shigella-attributable clinically-attended, antibiotic-treated LSD.

S5 PATHOGEN ASSOCIATIONS WITH ANTIBIOTIC TREATMENT AND OTHER CLINICAL FACTORS
We also sought to determine the association of each pathogen with the likelihood of receiving of antibiotics (given MSD or LSD presentation), becoming hospitalized (for MSD cases), and experiencing dysentery (for MSD cases). Using conditional logistic regression models, we estimated adjusted ORs of detection of each pathogen among antibiotic-treated cases versus those not treated with antibiotics (for each of MSD and LSD cases), hospitalized MSD cases versus those not hospitalized, and dysentery cases versus non-dysenteric MSD cases. We stratified cases by site, fortnight, and age (by month of life). For analyses of MSD data, we defined pathogen detection as Cq<35 and absence as Cq≥35. We excluded children diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, and controlled for the occurrence of any other non-diarrhea diagnosis in regression models. S1   Percentages indicate the proportion of cases administered antibiotics or receiving an antibiotic prescription (for which antibiotic administration could not be verified) for each of dysenteric MSD, nondysenteric MSD, and LSD. We considered pneumonia/lower respiratory infection, meningitis or other invasive infection, and typhoid diagnoses to warrant antibiotic treatment irrespective of diarrhea symptoms; thus, cases receiving antibiotics but diagnosed with any of these conditions were excluded from the definition of clinically-attended, antibiotic-treated diarrhea. Incidence estimates (per 100 child-years at risk) listed in this table are plotted in Figure 3 for each setting. We present overall pathogen-specific incidence rates for clinically-attended, antibiotic-treated diarrhea, across all settings, in Figure 4. Estimated incidence rates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. ----Estimates represent the age-specific clinically-attended, antibiotic-treated MSD adjusted attributable fraction based on qPCR data, divided by the estimate of the same quantity based conventional diagnostic data. Estimates are based on pooled data across all settings, with each setting weighted equally. We present estimates as median (95% CI). We present setting-specific estimates in S1 Figure. S5 Table: Adjusted attributable fractions of clinically-attended, antibiotic-treated moderate-to-severe diarrhea associated with individual pathogens, by age stratum and site, estimated from quantitative molecular diagnostic data.     Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for quantity of the other pathogens, together with 95% confidence intervals. Empty cells indicate where point estimates of the model-estimated adjusted attributable fraction were zero or below. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. 1. Aggregated estimates across sites weight individual cases so that each site receives equal weight within each age stratum.  Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for presence of the other pathogens, together with 95% confidence intervals. Empty cells indicate where point estimates of the model-estimated adjusted attributable fraction were zero or below. We did not identify associations of either C. jejuni or non-typhoidal Salmonella with clinically-attended, antibiotic-treated diarrhea in any age group. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. 1. Aggregated estimates across sites weight individual cases so that each site receives equal weight within each age stratum.  Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for presence of the other pathogens, together with 95% confidence intervals. Empty cells indicate where point estimates of the model-estimated adjusted attributable fraction were zero or below. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. 1. Aggregated estimates across sites weight individual cases so that each site receives equal weight within each age stratum.  Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for presence of the other pathogens. Models are fitted with data from all MSD cases as well as their matched controls. Empty cells indicate where point estimates of the model-estimated adjusted attributable fraction were zero or below. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. 1. Aggregated estimates across sites weight individual cases so that each site receives equal weight within each age stratum. Table: Adjusted attributable fractions of all clinically-attended, antibiotic-treated diarrhea associated with individual pathogens, by age stratum and site, estimated from conventional diagnostic data for both moderate-to-severe diarrhea and less-severe diarrhea.    Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for presence of the other pathogens, together with 95% confidence interavls.. Empty cells indicate where point estimates of the model-estimated adjusted attributable fraction were zero or below. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. 1. Aggregated estimates across sites weight individual cases so that each site receives equal weight within each age stratum S10  Estimates are presented as adjusted attributable fractions (%) for each pathogen, controlling for quantity of the other pathogens, together with 95% confidence intervals. Estimates are aggregated across sites and weight individual cases so that each site receives equal weight within each age stratum. Estimates exclude diarrhea cases receiving antibiotics who were also diagnosed with pneumonia/lower respiratory tract infection, meningitis or other invasive disease, or typhoid, as these conditions warrant antibiotic treatment irrespective of diarrhea symptoms. S11 The diarrhea incidence rate in the population of each Demographic Surveillance Site population is estimated from the proportion of children who experienced acute, new-onset diarrhea episodes in a 7-day recall period before the HUAS/HUAS-lite interview as detailed by Blackwelder and colleagues 3 , and the proportion of these children who received care at sentinel clinics Adjusted attributable fraction in moderate-to-severe diarrhea TaqMan array card re-analyses of specimens from 300 moderate-tosevere diarrhea cases within each site and age stratum, and first available matched controls 4 We fitted conditional logistic regression models to estimate the adjusted association of case status with abundance (Cq value) of each pathogen Uncorrected adjusted attributable fraction in moderate-to-severe diarrhea (conventional diagnostic data)

S9
Qualitative detection of each pathogen among moderate-tosevere diarrhea cases and matched controls based on conventional diagnostic assays We fitted conditional logistic regression models to estimate the adjusted association of case status with detection (binary) of each pathogen or pairwise combination of pathogens based on stepwise model selection Uncorrected adjusted attributable fraction in less-severe diarrhea (conventional diagnostic data) Qualitative detection of each pathogen among less-severe diarrhea cases and matched controls based on conventional diagnostic assays We fitted conditional logistic regression models to estimate the adjusted association of case status with detection (binary) of each pathogen or pairwise combination of pathogens based on stepwise model selection Corrected adjusted pathogenattributable fraction in less-severe diarrhea We multiplied estimates of the uncorrected adjusted attributable fraction in less-severe diarrhea by the ratio of estimates of the corrected-to-uncorrected adjusted attributable fraction in moderateto-severe diarrhea S1 Figure. Multipliers for etiologic fractions based on site-specific estimates. We illustrate estimated multipliers for the relative proportion of clinically-attended, antibiotic-treated MSD attributed to each pathogen, by age and setting, on the basis of quantitative diagnostics (relative to conventional diagnostics). We use the same multipliers to estimate the fraction of clinically-attended, antibiotic-treated LSD attributable to each pathogen, on the basis of conventional microbiological data obtained in the primary study.