Statistics

Data were entered into EpiData (Odense, Denmark) and analyzed using Stata version 12 (StataCorp LP, College station, Texas, USA). Normally distributed variables were expressed as means ± standard deviations (SD), and variables that did not follow a normal distribution were expressed at medians and interquartile ranges (IQR). Two-sample t-tests were used to evaluate differences in means except in case of non-normally distributed outcomes where Mann-Whitney rank-sum- tests were used. Chi-square tests were used to compare proportions, except when the expected numbers were less than five, in which case Fisher’s exact test was used.

To identify correlates of thymus area on admission, while including children with an invisibly small thymus, we used an analysis of covariance (ANCOVA) allowing for left-censored measurements from children with an invisibly small thymus. Since area measurements <1 cm² may be less accurate, we assigned all children with an undetectable thymus and children with measured area <1 cm² to an unknown low value <1 cm², and such left—censored measurements received less weight in the analysis as compared to accurately observed measurements, using the command “tobit” in Stata. Thymus area did not follow a normal distribution, and hence was logarithm-transformed (base 10); estimates were subsequently back-transformed. The analysis was adjusted for age and sex.

ANCOVA was also used to evaluate predictors of growth in thymus area at discharge and follow-up, defined as the change in thymus area (Δ thymus area = thymus area at follow-up – thymus area at admission). These analyses were adjusted for age, sex, number of days since first scan and thymus area on admission. Children with an undetectable thymus on admission were assigned a thymus area of 1 cm² in order to calculate the change in thymus area over time.

As sensitivity analyses, we firstly analyzed correlates of thymus area and of predictors of growth in thymus area while only including children with a visible thymus, and without left-censoring of low values. Secondly, to assess if the associations were explained by body size, we analyzed correlates of thymus size while adjusting for body weight after loss of edema.