The design of the study has been reported previously [12, 13]. In brief, the primary aim of the study was to investigate the natural history of rotavirus infections in children. The study was conducted in three neighbouring urban slums in Vellore measuring 2.2 sq. km with a population density of approximately 17,000 per sq. km. A common occupation in the area is 'beedi work', the manual production of cigarettes for a daily wage. Pregnant women were identified during a survey conducted in 2002. Infants were recruited from birth between March 2002 and August 2003 following written informed consent from the mother. Those living in brick-built houses with more than four rooms (n = 46) or who had birth weight less than 1.5 kg (n = 2) were excluded. Socio-economic status within this relatively homogenous population was stratified as previously described . Field workers visited infants at home soon after birth and twice weekly thereafter collecting details of morbidity and referring children to a physician-run study clinic as necessary. Follow-up continued until the child's third birthday. The last child was followed up in August 2006. Breastfeeding uptake was high in this population with 445 (98%) of the 452 enrolled children known to start.
Height and weight at birth, where available, were obtained from delivery records available at the first home visit. Subsequently, height and weight were obtained by field workers at the study clinic using single measurements. Weight was measured using a Salter weighing scale to the nearest 100 grams. Recumbent length was measured using a standard infantometer up to the child's first birthday or until the child was able to stand, and subsequently using a stadiometer, both to the nearest millimetre. The instruments were calibrated at least once a week. Field workers were retrained and procedures standardised once every three months.
We calculated total duration of illness; we defined major illness as diarrhoea, lower respiratory infection, tuberculosis, jaundice, central nervous system infection, seizures or convulsions, neonatal sepsis, neonatal jaundice, congenital diseases, burns, scalds, fracture, crush injuries, dengue, and physician diagnosed malnutrition; and defined minor illness as fever, cough or cold, asthma, wheezing, bronchiolitis, exanthematous fever (except measles), skin morbidities, eye morbidities, ear, nose and throat morbidities, incessant crying, abrasions and bites, pica, aphthous ulcer, localised infections, anaemia and loss of appetite. These comprised caregiver diagnosis of minor illness during field worker visits and physician diagnosis of minor and major illness from visits to the study clinic and from hospital admissions .
Respiratory illness was common, occurring at a rate of 7.4 episodes per child year (and a median of 48 days of illness) in the first year of life , 7.1 episodes (and a median of 67.5 days of illness) in the second year of life and 6.6 episodes (and a median of 50 days of illness) in the third year of life (Gladstone et al. unpublished). Gastro-intestinal illness was more common in the first year of life than in the second and third years, occurring at a rate of 3.6 episodes per child year (and a median of 8 days of illness) in the first year of life , 1.6 episodes (and a median of 3 days of illness) in the second year of life and 1.2 episodes (and a median of 1 day of illness) in the third year of life (Gladstone et al. unpublished). Peak rates of illness were experienced between 3 and 5 months of age, around the time of weaning . The rate of hospitalisation was low, 0.28 per child year in the first year, 0.11 in the second year and 0.07 in the third (Gladstone et al. unpublished).
Poor growth was quantified as the percentage of children stunted (low height-for-age), wasted (low weight-for-height) and underweight (low weight-for-age). 'Low' was characterised as less than -2 standard deviations from the growth reference. We used the 2006 WHO child growth standards  as the reference population to calculate height-for-age (HAZ), weight-for-height (WHZ) and weight-for-age (WAZ) z-scores. Plots of individual's growth curves revealed some discrepancies in height measurements. Therefore measurements which were more than three standard deviations different from the nearest four measurements were set to missing height a priori. All measurements relate to attained month of age, for example a measurement taken on a day of life between six times 30.4375 (183 days) and seven times 30.4375 (213 days) relates to six months of age. The first measurement in an attained month of age was used in analysis.
As stunting is considered to be a long-term deficit in growth and it was the dominant growth faltering seen at three years of age, we investigated the probability of being stunted at three years using logistic regression with potential risk factors defined a priori. We investigated whether temporal changes in both duration of major illness or duration of all illness were associated with stunting at 36 months, but found no patterns. We therefore present the association with total duration of major illness over three years. We investigated causes of missing values, in both stunting at 36 months and in potential risk factors, using observed data. To be conservative we included all potential risk factors in a multivariate model, whether associated with the outcome or with missing values . We investigated whether risk factors were gender-dependent by testing interaction terms between the risk factor and gender including interactions with likelihood ratio p < 0.05. We conducted both a complete case analysis (including only children who had no missing values in the outcome or potential risk factors) and for comparison an analysis where all missing values were imputed using chained equations [15–17] [see additional file 1]. We then carried out a sensitivity analysis to establish the stability of the estimated odds ratios [see additional file 2]. Finally, we also investigated risk factors for early growth faltering (stunted, wasted or underweight) at six months of age using the same methodology and investigating the relationship between growth faltering and all illness in the first six months of life [see additional file 3]. All analyses were undertaken using STATA version 10.1 . The study was approved by the Institutional Review Boards of the London School of Hygiene and Tropical Medicine, London, UK, Baylor College of Medicine, Houston, USA, and Christian Medical College, Vellore, India.