In this large population-based prospective study of rural Bangladeshi children, we observed increased childhood mortality associated with increased As exposure via contaminated well water.
Few previous studies have focused on childhood cancer in low exposure settings but those studies used ecologic exposure assessment, and there was possibility of misclassification of outcome data 
. The issue of As exposure and childhood mortality has not been studied in any arsenic exposed areas as such Bangladesh. The results available from previous studies cannot be used to infer any causal relationship due to their ecologic design, rather they may be considered as hypothesis generating, and therefore merits further study to seek more definitive answer to confirm or refute. Thus, this is a first prospective cohort study with individual level As exposure exploring causal association. Earlier we reported excess adult deaths in Matlab population 
, whereby Matlab provides a unique opportunity to investigate these arsenic health effects, considering individual exposure assessments. Considering the magnitude of HR we identified, along with the consistency of findings in all exposure categories suggest that these findings are true, not due to chance or bias, although the number of overall childhood death cases was relatively very small contrary to the large exposed population. Furthermore, we also demonstrated that arsenic-induced epigenetic modifications in utero
may potentially influence disease outcomes in later life 
. Thus this is a biologically significant risk that merits public concern.
More importantly, we observed a gender difference on the effect of all-cause childhood deaths. Girls had higher risk of deaths compared to boys in all exposure categories. Nearly all epidemiological studies concluded that men have higher risk of developing all arsenic-induced negative effects 
including our previous skin lesion study 
, except one 
. Earlier we demonstrated in the Matlab population that arsenic readily crosses the placenta, therefore a positive correlation occurred between maternal and cord blood arsenic 
Secondly, the magnitude of the effect estimates we observed were large in relation to baseline exposure categories comparing relatively small number of childhood cancers and cardiovascular deaths that have been identified in this cohort. There has been some interest on the association of As exposure and childhood cancer mortalities in recent years using group-level aggregated exposure data, but we found a sharp increased risk of liver and leukemia cancer mortalities in this study, which could also be due to the increased fatality of cancers related to arsenic based on the evidence that arsenic causes increased mortality from many cancers including lung, liver, kidney, and bladder cancers and not just due to increased incidence alone 
. This is also true in case of mortality from pulmonary TB as reported by Smith et al, 2011 
. Higher multivariable adjusted HRs were found for adolescents (HR
2.46, 95% CI, 1.03–5.88) than children (HR
1.31, 95% CI 0.42–4.09) (data not shown). The plausibility of a causal association between exposure and disease is enhanced if experimental treatment can produce a similar condition.
Strictly speaking, there have been only five childhood cancer studies 
to compare with our results. All are ecologic in nature, and no study showed childhood cardiovascular or all-cause death risks. A non-significant relative risk 1.39 (CI: 0.7–2.76) for lymphoblastic leukemia was reported from a Canadian case-control study 
. Another non-significant relative risk was reported for all cancer combined 1.25 (CI 0.91–1.69), and for leukemia 1.37 (CI: 0.92–1.83) in Nevada 
. Similar results were observed in Chile for all cancers combined 
, except liver cancer mortality was markedly increased (RR
10.6, 95% CI 2.9–39.2, P
<0.001). Ecologic bias may have occurred in those studies based on aggregated group level exposure data causing measurement error. As such those studies may not be used as inferring a causal relationship rather used as hypothesis generating. Truly this is the first study that enabling us to analyze childhood mortalities. Major studies focused on adult carcinogenic risks, evidencing multisite carcinogenetic role and therefore its carcinogenicity among children remained uninvestigated. But the overall carcinogenicity process is still undiscovered, and theoretically arsenic causes aberrant cell proliferation including alteration in apoptosis. In combination of disrupt cell proliferation, genetic mutation, chromosomal breakage and genetic damage may enhance carcinogenesis in children 
. However, children and adults may have differences in carcinogenic risks reflecting differences in their tissue dosages and thereby differences in sensitivity.
The major strength of our study included larger sample size, individual exposure data, assessment of different exposures by outcomes, and independent outcome data were collected prospectively from the HDSS databases. The mortality data set included relatively larger childhood mortality data over an extended follow-up time (over 185 deaths during more than 0.4 million person-years at risks). Without a death registry, proper case ascertainment may be uncertain in a developing country, but, standardized approaches from widely validated verbal autopsy methods attributed to the strength of the study. VA is a well-known instrument ascertaining the cause of death based on information obtained from close relatives through systematic retrospective questioning 
. Moreover, it is stemmed from findings of earlier ecologic studies 
that indicated increased childhood deaths due to arsenic exposure. Another strength of our study is that any biases were minimized by combining individual outcome data from the regular HDSS monthly surveillance at the household level and exposure data at the individual level.
Despite major strengths, limitations can be attributed to some unmeasured or imprecisely measured potential confounding factors. A major weakness of this study is lack of exposure history between 2003 to 2010. Change of drinking water sources of the cohort during this period is quite natural. Well water arsenic exposure is the only exposure parameter considered no other sources of exposure have been estimate.The unaccounted changes of As sources across the observational time is a weakness. Therefore, findings are to be interpreted under the assumption of constant source of As as measure in 2003 and that the impact of possible changes is unknown. Secondly 39% individuals were lost during the study period. Majority moved out (taking new job 35.4%, married 17.7%, education 12.3%, children follow parents 12.7%, others 37.4%). Moreover, lost to follow-up is a problem regardless of any possible assuring comparisons. We simply do not know how they could have affected HRs if they were not lost.
However, there were no significant difference in exposure distribution between included or excluded persons of this cohort (Table S3
). Thirdly, drinking water was the reported primary source of exposure estimates. Arsenic exposure may also be contributed by food and other water sources 
. A study from Bangladesh has shown that rice often contains more than 100 µg/kg of arsenic in arsenic affected areas 
and rice alone may contribute about 50 µg arsenic per day. However, recent study from Bangladesh demonstrated that it is often adequate to estimate an individual’s past arsenic exposure based on the reported main sources of drinking water and the influence of neighbouring water sources was limited 
. Fourthly, we did not have pathological reports for case ascertainment, and we have no information on SES among 1039 individuals (3.9% of total cohort). We do not believe our study findings would differ much with exclusion of these factors. Moreover, there are no reasons to believe that misclassification of cause of death is associated with arsenic exposure levels. Not knowing the level of arsenic exposure indicates no bias while assigning cause of childhood deaths.
We also found marked increase in all cause childhood mortality (HRs
1.53, 95% 1.03–2.28) among adolescents (age range 12–18) as well as liver cancer and leukemia deaths having relatively short latency. In addition, girls were at increased risk compared to boys. There have been many studies of effects of arsenic exposure in adults, but very little attention has been given to potential effects resulting from exposure to infants and children. It has become clear that the long-term effects of toxic substances in children need to be investigated and understood to protect children’s health, and later their health as adults.
Despite some limitation, this study was feasible because of the uniqueness of the Matlab population, which includes individual level As exposure data and the independent prospective demographic surveillance system covering 0.2 million population for about five decades. Further studies are needed with a longer follow up period to investigate if the risk is further enhanced.