The study by Rahman et al. evaluated the effect of consumption of arsenic contaminated tube well water on infant death and natal loss in Bangladesh.
There are a number of bias that may be present in any research study. These include recall bias (where there is difference between what a subject and a control recalls), misclassification bias (in simplest terms, measurement error) and selection bias (introduced when sample is not representative enough or when there is difference between exposed and control group). Cohort studies least suffer from these biases (as compared to other study designs), since subjects are chosen before hand (in a prospective study). It seems that selection bias may be present in the current case. For example, the cases of infant deaths and natal loss could have been followed more closely than non-fatal cases; although the authors state that selection and information bias have been minimized. Recall bias also has been minimized by asking the subjects to relate drinking water history to significant life events. Given the authors' efforts to minimize the bias, it is difficult to imagine any bias in this cohort study, without further information.
I would imagine that certain factors that were not included in study design, such as, history of previous natal loss or infant death, malnutrition, presence of water treatment measures, availability of medical care, etc., could have biased the results in favor of positive correlation between exposure and outcomes. Thus the systematic bias has been introduced due to study design.
Hope this helps.