In 50 villages across rural Bangladesh where fecal environmental contamination, undernutrition, and diarrhea are common, in those households where fieldworkers observed food preparers washing their hands before handling food, children under the age of 5 y experienced less diarrhea over the next 2 y compared with children living in households where food preparers did not wash their hands before preparing food. This observation suggests that before preparing food may be a particularly important time to promote handwashing 
Tomatoes, cucumbers, carrots, and various seasonal vegetables and greens are common components of meals in rural Bangladesh. Some of these vegetables are served raw but most are boiled and made into a curry that is commonly served with rice, the primary staple of the Bangladeshi diet. Many foods that are not further cooked, for example boiled root vegetables, fruits including bananas, and dried fish are often mashed and mixed by hand with spices and other ingredients during food preparation. Raw vegetables are commonly contaminated with pathogens and are a common vehicle for gastrointestinal pathogen transmission. Numerous outbreaks of gastroenteritis from a variety of pathogens have been traced to raw vegetables 
. The surface of raw cut lettuce and tomatoes is a hospitable environment for the growth of Shigella
and Salmonella 
. Similarly, there is considerable microbiological and epidemiological evidence that implicates cross-contamination of food as an important pathway for gastrointestinal pathogen transmission 
. Food that is inoculated with bacterial pathogens from contaminated hands may provide a nutrient rich environment that permits exponential growth for numerous pathogens 
. The risk of diarrhea for many bacterial pathogens is proportional to the dose of the pathogens ingested 
. In several outbreaks of bacterial gastroenteritis, food that was contaminated several hours before serving was associated with high attack rates of gastroenteritis among persons who consumed it 
If persons preparing food did not wash their pathogen-contaminated hands before touching raw vegetables and rice, these foods may have become contaminated with gastrointestinal pathogens, which could subsequently multiply in a conducive growth environment before consumption. However, if the vegetables were cooked at a high enough temperature for a long enough time the pathogens would not survive and would not be transmitted. In future research, it would be useful to have fieldworkers specifically code the context of the handwashing opportunity around food preparation, so that the association between handwashing before handling raw vegetables and other foods that were subsequently cooked, handwashing before handling foods that were eaten raw, and handwashing before cross contaminating food that was not further cooked could be separately assessed.
In contrast to standard recommendations for handwashing that stress the central importance of using soap and specify detailed techniques for washing underneath fingernails, continuing lathering for over 20 s, and using either a clean towel or air drying to ensure effective handwashing 
, in this observational study, children who lived in households where food preparers practiced suboptimal handwashing (including briefly washing their hands with water alone) experienced significantly less diarrhea than children living in households where the food preparer did not wash hands at all. Fieldworkers did not directly measure the duration of handwashing with soap in this study, but in another study that used structured observation in urban Bangladesh to assess handwashing behavior and timed the duration of handwashing with soap with a stopwatch, the baseline mean duration of handwashing with soap was 5 s before preparing food and 11 s after defecation 
Although the benefits of handwashing with water alone observed in this evaluation conflict with standard recommendations, they are consistent with an older randomized controlled intervention study from urban Bangladesh. Stanton and Clemens used structured observation to observe handwashing behavior and noted an association between washing hands with or without soap and reported childhood diarrhea in a case control study in low-income urban communities in Dhaka Bangladesh 
. In a subsequent intervention study in households that received the intervention, food preparers were significantly more likely to wash their hands with or without soap compared with food preparers in nonintervention households (49% versus 33%) 
. Intervention households reported 26% less diarrhea than nonintervention households.
Microbiological studies demonstrate that washing hands with water alone reduces the concentration of various bacteria on hands 
. The reduction in these bacteria is generally less than the reduction in hand contamination following handwashing with soap 
. Field workers did not record the source of water used to wash hands, but the most common source of household water in rural Bangladesh is shallow tube wells. In other studies approximately 40% of water samples directly collected from shallow tube wells in Bangladesh were contaminated with fecal bacteria, though generally at a low-level of contamination 
. The present evaluation suggests that even the modest reduction in hand contamination achieved by washing with water alone reduces the risk of pathogen transmission at least during food preparation, albeit to a lesser degree than handwashing with soap.
The low proportion of households that followed recommended hand drying procedures suggests that substantial efforts would be required to change community habits to conform with hand drying recommendations. Since children living in households that practiced recommended hand drying behavior did not have significantly less diarrhea than other households, these data suggest that efforts to promote improved hand hygiene would be better focused on behaviors more strongly associated with child health, for example on handwashing before preparing food and after defecation, than on prescribing specific hand drying behavior.
People wash their hands more frequently when they know they are being observed 
. In a previous study in rural Bangladesh that placed accelerometers within bars of soap to detect soap motion, the presence of an observer increased the frequency of soap motions consistent with handwashing by 35% 
. Since Bangladeshi culture views adult feces as impure 
, social desirability bias may have increased observed handwashing with soap, especially after defecation. In the Bangladesh motion sensor study, residents of households with more education and who owned a mobile phone or watch were more likely to increase handwashing in the presence of an observer, and in the present study households with more education and those that owned mobile phones or televisions had less diarrhea 
. Thus, an alternative interpretation of these observations is that the association between washing hands and subsequent childhood diarrhea is not causal. Rather, the observed reactive handwashing behavior might be an indicator of broader hygiene awareness that identified a subset of households that practiced a number of behaviors that contributed to less childhood diarrhea.
But there are two difficulties with this alternative interpretation. First, the strong association of handwashing with water alone before preparing food with diarrhea is less likely to result from social desirability bias, because there is no strong cultural norm for handwashing before preparing food. Indeed, only 1% of households washed hands with soap before preparing food. If household residents washed hands before preparing food because of social desirability bias that was then linked to other behaviors associated with less diarrhea, then we would also expect to find a significant association of diarrhea with handwashing before eating, before feeding a child, and after cleaning a child who defecated, associations that were not significant in this analysis. A second difficulty with attributing the observed associations to a theoretical unknown, unnamed, and unmeasured confounder is that the analysis implies that such a causal pathway for reduced diarrhea was independent of education, wealth, exclusive breastfeeding, and other evaluated household characteristics. An unmeasured personal or household characteristic that is so powerful that it dominates the relationship between handwashing and diarrhea, but is so elusive that we cannot even name it, seems a less likely explanation than pathogen contaminated hands and food, a biologically plausible explanation that invokes a pathway of gastrointestinal pathogen transmission repeatedly demonstrated in other contexts.
The observation in this evaluation that children living in households where residents washed their hands with soap after defecation had less diarrhea compared with children living in households where handwashing after defecation was less thorough is consistent with findings of previous intervention studies 
and with handwashing interrupting the transmission of pathogens from the gastrointestinal tracts of household members to a susceptible child. The lack of a significant association of diarrhea with handwashing after cleaning a child's anus who defecated or handwashing before feeding a child or before eating also have plausible microbiological explanations. A child's gastrointestinal tract and immune system has already been exposed to the organisms in his/her own feces. Further exposure to these organisms is unlikely to cause clinical illness in the child. Unwashed hands can transmit pathogens to food, but when contaminated hands contact food at the time of eating or feeding, the dose of ingested pathogen is limited to the number of organisms that are passively transferred from hand to food. In contrast, when pathogens are transmitted to food items that are stored and not further cooked, bacterial pathogen populations may reproduce exponentially, resulting in a much higher dose of pathogen and a greater risk of diarrhea.
An important limitation of this study is that measuring handwashing on a single day risks misclassifying exposure. Among mothers in Burkina Faso, observed handwashing behavior after cleaning a child who had defecated was concordant with observations on a different day between 57% and 73% of the time 
. This imperfect repeatability of handwashing assessments risks misclassifying exposure, which reduces the statistical power to identify associations. Such misclassification could explain why handwashing at some key times was not associated with less child diarrhea in this evaluation. However, handwashing in this evaluation was not classified on the basis of a single observation, but on the basis of the best behavior observed among multiple observations within the household (). Handwashing is a habitual behavior 
. For example, in the Burkina Faso study, not washing hands on one occasion was significantly associated with subsequent behavior 
. Importantly, even with reduced power from misclassification, the Bangladesh evaluation presented in this article identified associations between handwashing at two biologically plausible occasions with reduced prevalence of subsequent diarrhea.
A second limitation is that fieldworkers observed an opportunity to wash hands after defecation in only 29% of households. The resulting limited statistical power precluded a thorough assessment of the utility of washing hands after defecation with water only or with ash/mud, the contribution of other determinants of diarrhea, or a combined model that included both handwashing before preparing food and handwashing after defecation. However, even with limited power there was a strong association between handwashing with soap and less subsequent diarrhea, and the point estimates of the odds ratios are suggestive of less diarrhea for handwashing with water alone.
A third limitation is that different gastrointestinal pathogens have different routes of transmission within different contexts, which might limit the generalizability of this study. It is possible that transmission of gastrointestinal pathogens from hands to food during preparation is a less important route of transmission in other settings. Additionally, in settings where water to wash hands is more heavily contaminated with feces than available water in rural Bangladesh, washing hands with water alone may be less protective. However this evaluation was conducted in 50 rural villages in 26 districts across Bangladesh and Bangladesh is the eighth most populous country in the world, so the analysis is not identifying a highly isolated phenomenon. In an assessment of hygiene indicators in rural Nicaragua, washing hands before preparing food was the single hygiene indicator most strongly associated with child diarrhea 
. Nevertheless, it would be useful to conduct similar evaluations in other contexts.
A fourth limitation is that the program evaluation was not designed to evaluate the hypothesis that observed handwashing behavior was associated with a change in the prevalence of subsequent diarrhea. Because this is a secondary analysis of the data, there is some risk of data mining to identify an interesting but ultimately not robust finding. However, we planned these data analyses at the time we designed the program evaluation. There was a dose effect between thoroughness of handwashing before preparing food and subsequent observed diarrhea and the associated p-values were <0.005.
An important flaw in this evaluation was that we did not have the protocol reviewed by an independent human study participant committee. The amount of time we asked from participants, the intensity of the interaction with the field team, and the use of these data to draw generalizable insights to improve global scientific understanding mean that the activity had substantial research components and should have been reviewed by an independent human study participant committee. The study team did implement standard procedures to minimize risks and harms to evaluation participants, but similar future evaluations should be reviewed by human study participant committees. Rigorous evaluations of large public health programs provide insights that can translate into improved programs that save lives and improve community health. However large public health programs in low-income countries often have extremely tight implementation schedules. Human study participant committees in low-resource settings may need to develop additional capacities to provide appropriate independent review more promptly for these type of evaluations.
Most people living in low-income settings have apparently concluded that following recommendations that require them to wash hands with soap ten, 20, or more times per day is not feasible 
. The observations from this program evaluation suggest that to prevent childhood diarrhea the most important occasions for handwashing and the technique for effective handwashing differ from standard recommendations. Specifically, handwashing promotion programs in rural Bangladesh should not attempt to modify handwashing behavior at all five key times, but should focus primarily on handwashing after defecation and before food preparation. Because handwashing before food preparation is such a different context than after defecation, developing and evaluating strategies to promote handwashing before food preparation is an important area for future research.
The lower prevalence of childhood diarrhea seen in this evaluation among children living in households where residents washed hands with soap are consistent with the many intervention trials that demonstrate less childhood diarrhea in households where residents are encouraged to wash hands with soap 
. The findings from this study, that children living in households where field workers observed food preparers washing their hands with water alone before preparing food had less diarrhea compared with children living in households where fieldworkers observed that food preparers did not wash their hands, suggest that promoting handwashing exclusively with soap may be unwarranted. Handwashing with water alone might be seen as a step on the handwashing ladder: handwashing with water is good; handwashing with soap is better. Additional controlled trials evaluating the effect on child health of interventions that include encouraging handwashing either with water alone or with soap and water would be particularly helpful to guide public health programs. More generally, research to develop and evaluate handwashing messages that account for the limited time and soap supplies available for low-income families, and are focused on those behaviors where there is the strongest evidence for a health benefit could help identify more effective strategies.