We analyzed the doseeresponse relationship using fractional
polynomial regression of the inverse variance-weighted data. Bestfit
curves were selected using decreased deviance compared with
the reference model [39]. Comparisons of curves to determine best
fit were done using a chi-square distribution