The article by Rahman et al., determined the effect of arsenic concentration in contaminated tube well water on 5 outcomes (of infant death and fetal loss) in Bangladesh. This was a cohort study.
Any scientific study can suffer from bias, whether they are in planning stage, data collection stage, data analysis stage or reporting (or publication) stage. In general, there are two major classes of bias: random and systematic. If we assume that no bias was introduced during the publication/reporting stage and that the very large sample size (29,134 pregnancies) will remove any potential bias in sample selection and also that no bias was introduced during analysis stage; we are left with bias during the planning and data collection stages. There are a number of factors that, in my opinion have been left out, could have strongly affected the results. These include, any past history of fetal loss and infant death for a given subject, nutritional practices (availability of adequate nutritional diet), availability of healthcare at appropriate time, changes in arsenic concentration over the duration of study (pregnancy to first 12 months of child birth), availability of water treatment methods (i.e., was the pregnant woman given treated water or regular tube-well water). There could be a number of other factors that could have made the study more robust, by improving the study design.
Hope this helps.