Use of quantitative molecular diagnostic methods to investigate the effect of enteropathogen infections on linear growth in children in low-resource settings: longitudinal analysis of results from the MAL-ED cohort study

Background: Enteropathogen infections in early childhood not only cause diarrhoea but contribute to poor growth. We used molecular diagnostics to assess whether particular enteropathogens were associated with linear growth across seven low-resource settings. Methods: We used quantitative PCR to detect 29 enteropathogens in diarrhoeal and non-diarrhoeal stools collected from children in the first 2 years of life obtained during the Etiology, Risk Factors, and Interactions of Enteric Infections and Malnutrition and the Consequences for Child Health and Development (MAL-ED) multisite cohort study. Length was measured monthly. We estimated associations between aetiology-specific diarrhoea and subclinical enteropathogen infection and quantity and attained length in 3 month intervals, at age 2 and 5 years, and used a longitudinal model to account for temporality and time-dependent confounding. Findings: Among 1469 children who completed 2 year follow-up, 35 622 stool samples were tested and yielded valid results. Diarrhoeal episodes attributed to bacteria and parasites, but not viruses, were associated with small decreases in length after 3 months and at age 2 years. Substantial decrements in length at 2 years were associated with subclinical, non-diarrhoeal, infection with Shigella (length-for-age Z score [LAZ] reduction −0·14, 95% CI −0·27 to −0·01), enteroaggregative Escherichia coli (−0·21, −0·37 to −0·05), Campylobacter (−0·17, −0·32 to −0·01), and Giardia (−0·17, −0·30 to −0·05). Norovirus, Cryptosporidium, typical enteropathogenic E coli, and Enterocytozoon bieneusi were also associated with small decrements in LAZ. Shigella and E bieneusi were associated with the largest decreases in LAZ per log increase in quantity per g of stool (−0·13 LAZ, 95% CI −0·22 to −0·03 for Shigella; −0·14, −0·26 to −0·02 for E bieneusi). Based on these models, interventions that successfully decrease exposure to Shigella, enteroaggregative E coli, Campylobacter, and Giardia could increase mean length of children by 0·12–0·37 LAZ (0·4–1·2 cm) at the MAL-ED sites. Interpretation: Subclinical infection and quantity of pathogens, particularly Shigella, enteroaggregative E coli, Campylobacter, and Giardia, had a substantial negative association with linear growth, which was sustained during the first 2 years of life, and in some cases, to 5 years. Successfully reducing exposure to certain pathogens might reduce global stunting. Funding: Bill & Melinda Gates Foundation. The MAL-ED study design has been described previously.10 Children were enrolled within 17 days of birth at eight locations between November, 2009, and February, 2012. Linear anthropometric measurements were available from seven locations: Dhaka, Bangladesh; Vellore, India; Bhaktapur, Nepal; Fortaleza, Brazil; Loreto, Peru; Venda, South Africa; and Haydom, Tanzania. Children were included if their mother was aged 16 years or older, their family intended to remain in the study area for at least 6 months from enrolment, they were from a singleton pregnancy, they had no other siblings enrolled in the study, and had a birthweight or enrolment weight of more than 1500 g. Children diagnosed with congenital disease or severe neonatal disease were excluded. All sites received ethical approval from their respective governmental, local institutional, and collaborating institutional ethics review boards. Written informed consent was obtained from the parent or guardian of every child. Fieldworkers visited children twice weekly until age 2 years for active surveillance of child illnesses, antibiotic use, breastfeeding, and food intake. Sociodemographic information was collected every 6 months. Linear anthropometric measurements were obtained by fieldworkers monthly to age 2 years (length) and once at age 5 years (±6 months; height).11 Diarrhoeal stools were defined by maternal report of three or more loose stools in 24 h or one stool with visible blood. Non-diarrhoeal stool samples were collected monthly (at least 3 days before or after a diarrhoea episode) from birth to age 2 years. We tested all stool specimens using custom-designed TaqMan Array Cards (ThermoFisher, Carlsbad, CA, USA) that compartmentalised probe-based quantitative PCR assays for 29 enteropathogens (appendix). Assay validation, nucleic acid extraction, quantitative PCR conditions, and quality control have been previously described.13, 14 Both Shigella and enteroinvasive E coli are detected using the ipaH target; however, on the basis of previous findings that Shigella flexneri and Shigella sonnei account for the majority of ipaH detections,13 and Shigella positive stool cultures are metagenomically similar to ipaH positive stools,15 for simplicity the presence of ipaH was considered diagnostic of Shigella. Pathogen-specific aetiology of diarrhoea was determined using attributable fractions (AFe) to adjust for subclinical pathogen infections, as previously described.13, 16, 17 We defined pathogen-attributable episodes when the pathogen quantity-derived AFe was 0·5 or higher (ie, majority attribution). Episodes with a sum of all pathogen-specific AFes of less than 0·5 (ie, the majority of the episode was not attributed to pathogens) were considered non-attributable. We assessed the associations between diarrhoeal aetiologies and growth for diarrhoea episodes attributable to any infection, and to viral, parasitic, and bacterial pathogen groups (appendix). We also assessed individual pathogens, specifically the ten enteropathogens with the highest attributable diarrhoeal incidence in the MAL-ED study (identified in the companion Article16): Shigella, typical enteropathogenic E coli, Campylobacter jejuni or Campylobacter coli, enterotoxigenic E coli, Cryptosporidium, astrovirus, sapovirus, norovirus, rotavirus, and adenovirus 40/41. For the subclinical infection and growth analysis, we assessed all 29 pathogens (appendix) and included the 13 most prevalent pathogens as covariates in the models (including all pathogens with significant associations with growth in the height attainment model): enteroaggregative E coli, enterotoxigenic E coli, Giardia, Campylobacter, atypical enteropathogenic E coli, adenovirus 40/41, sapovirus, typical enteropathogenic E coli, norovirus, Shigella, astrovirus, Enterocytozoon bieneusi and Cryptosporidium. Length measurements were converted into length-for-age Z scores (LAZ) using 2006 WHO child growth standards.18 Socioeconomic status was summarised using a construct of water, assets, maternal education, and income11, 15 and was averaged over four biannual surveys. Exclusive breastfeeding was defined as the proportion of days in a specified time period in which children were breastfed and received no liquids or solids. Potential confounders were included on the basis of previous associations with enteropathogen exposure19, 20, 21 and linear growth.11 To estimate the associations between aetiology-specific diarrhoea and linear growth after 3 months, we used repeated measures linear regression with general estimating equations to account for correlation between children’s outcomes over time. Models were adjusted for age, site, sex, socioeconomic status, maternal height, LAZ at the beginning of the interval, exclusive breastfeeding, and number of non-attributable diarrhoea episodes in the same period. We also estimated the associations of diarrhoea with fever, dehydration, vomiting, blood, prolonged duration (diarrhoea for 7 days or longer), and high severity (modified Vesikari score >6)22 with LAZ after 3 months. We used linear regression to estimate associations between aetiology-specific diarrhoea episodes and LAZ measured at 2 years (acceptable window for measurement was age 731 days ±15 days) in a height attainment model. Models were adjusted for enrolment LAZ, sex, socioeconomic status, exclusive breastfeeding in the first 6 months, maternal height, number of non-attributable diarrhoea episodes, and number of episodes treated with any antibiotics. Effects were estimated for the difference in LAZ at 2 years and scaled to compare a high burden of attributable diarrhoea episodes with a low burden (ie, the difference between the 90th and 10th percentile). We used linear regression to estimate associations between subclinical enteropathogen infections and LAZ measured at 2 years in a height attainment model, adjusting for site, enrolment LAZ, sex, socioeconomic status, exclusive breastfeeding in the first 6 months, and maternal height. Exposure to each enteropathogen was summarised as the proportion of non-diarrhoeal stools obtained between age 1 and 24 months that were positive for that enteropathogen. The summative effect of pathogen groups was assessed by calculating the mean number of detections between age 1 and 24 months. The difference in LAZ at 2 years associated with each pathogen was scaled to compare the 90th percentile with the 10th percentile for stool positivity (appendix). In a second analysis, we specified pathogen quantity in non-diarrhoeal stools as the exposure, defined by mean log-copy number per g of stool (appendix) and scaled effects per one log increase in pathogen quantity. In a sensitivity analysis, we specified enteropathogen exposure as the proportion of all positive stools (non-diarrhoeal and diarrhoeal) obtained between age 1 and 24 months. We also estimated the associations of enteropathogen exposures with weight-for-age and weight-for-length Z scores in models with the same structure, additionally adjusting for enrolment weight-for-age Z scores. We estimated the associations with height-for-age Z score at 5 years of age (±6 months) in a model with the same structure. To ensure temporality and include potential lag periods between exposures and growth, we investigated enteropathogens in longitudinal models. We defined exposures in 6 month intervals from birth to 2 years and used the parametric g-formula23 to model interim effects on LAZ at the end of the intervals and overall effect on LAZ at 2 years. The parametric g-formula was fitted for each pathogen individually, specified first as the proportion of positive non-diarrhoeal stools and second as the mean quantity in non-diarrhoeal stools during the 6 month interval. Pathogen exposures were assessed with a flexible lag structure including the exposure in the current and previous intervals (appendix). All models were adjusted for the five pathogens with the strongest associations with LAZ at 2 years. We first used the observed data to estimate β-coefficients in longitudinal repeated measures models for each time-dependent covariate in the 6 month intervals (appendix). We used Monte Carlo simulations with the estimated coefficients to predict the time-dependent covariates, pathogen exposures, and LAZ outcomes for each interval to age 2 years in a random sample of 50 000 replicates from the study population at baseline. Simulations were run for high (90th percentile in each interval) and low (10th percentile in each interval) pathogen exposure conditions and the difference due to a one log increase in pathogen quantity. We estimated the cumulative effect of pathogens on LAZ by calculating the difference of the mean predicted outcomes at 2 years between the high and low exposure conditions (population-standardised LAZ difference). 95% CIs were constructed by bootstrapping at the individual level to account for correlation between observations over time with 1000 replicates. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication.