Josephine van Dongen

2 Acute gastroenteritis disease burden among infants with medical risk conditions 41 Supplementary material Laboratory analysis AGE stool samples were send to the central laboratory at the University Medical Center Utrecht and tested by multiplex realtime PCR for presence of rota-, noro-, adeno(type 40/41)- and astrovirus. Briefly, RNA and DNA were isolated from the samples using the MagnaPura96® (Roche Diagnostics, Pleasanton, CA USA) and amplified by ABI75000 realtime PCR system® (Thermofisher Scientific, Foster City, CA, USA). Prior to extraction, a non-human internal control was spiked into the lysis buffer of the samples to monitor for sample inhibition. Positive and negative controls for each pathogen were tested in every run. Samples tested negative for all four viruses are defined as pan negative. Fecal samples collected within four teen days of symptom onset were defined as AGE samples. Primary outcome definitions 1) To calculate IRs we used the total number of AGE episodes identified. IRs were calculated for the total cohor t, per age-category (< 6 months, 6-11 months and 12-17 months), by subgroup of medical risk condition and by virus type. Given the strong seasonality of viral AGE, we also calculated IRs by month of the year. 2) Healthcare attendance was defined as the propor tion of episodes for which healthcare was required and was stratified into doctor visit or hospitalization. Healthcare usage was also assessed per virus-type. 3) The AGE severity based on the MVS 1,2 was calculated from the daily symptom- score completed for the seven days following AGE onset and classified into mild (0-8 points), moderate (9-10 points) or severe ( ≥ 11 points). Severity was assessed for all-cause and virus specific AGE. Missing data and multiple imputation Because of a moderate to large propor tion of missing information, we applied multiple imputation by chained equation.The imputation model included variables with infant, household and AGE characteristics which contains auxiliary information, the 22 imputed datasets were combined to obtain IR estimates and 95% CI accounting for uncer tainty and the imputation procedure. By using auxiliary variables the imputation model can make use of more information than used for analyses 3 . In this case, seven variables, not included in the analysis, containing information related to missing covariates were added to the multiple imputation dataset. For the multiple imputation procedure, we used R package mice, with m=22 and maxit=10. The propor tion of missing in this observational data was up to 50% for some of the variables.The fraction of missing information (FMI) is a quantification of the loss of information in a dataset