Diarrhoea and pneumonia are the leading causes of death in children < 5 years old worldwide, accounting for 11% and 18% of the 7.6 million deaths that occurred in 2010, respectively (Liu et al. 2012). In Kenya in 2010, 9% of the deaths in children < 5 years old were attributed to diarrhoea and 17% to pneumonia (World Health Organization 2010).

Interventions promoting handwashing with soap can reduce diarrhoea and pneumonia in resource-poor settings (Curtis et al. 2001; Stanton & Clemens 1987; Luby et al. 2004, 2005; Peterson et al. 1998). A meta-analysis assessing the health impact of promoting handwashing with soap reported a 31% reduction in risk of acute gastrointestinal disease and 21% reduction in risk of acute respiratory illness (Aiello et al. 2008). Of the ten studies conducted in resource-poor settings included in the meta-analysis, only one was conducted in sub-Saharan Africa. In that study, in Zaire, the prevalence of diarrhoea during the 3 months after the handwashing intervention was 11% lower among children in the intervention arm than children in the control arm (Haggerty et al. 1994).

Large-scale efforts to promote handwashing are ongoing in many low- and middle-income countries. Robust evaluations of handwashing promotion programmes are necessary to determine their impact on behaviour. Handwashing behaviour can be directly observed by placing trained enumerators in households who record key transmission events, such as before eating or after toileting, and associated hand cleansing, but this method requires extensive field staff time and the respondent may alter behaviour due to the presence of the observer (Manun'ebo et al. 1997; Ram et al. 2010). Because of these limitations, surrogate measures that reliably characterise handwashing behaviour are needed. Such measures should be objective and relatively immune to social desirability, which adversely affects both self-report and directly observed behaviour (Manun'ebo et al. 1997; Curtis et al. 1993; Ram et al. 2010).

A potential surrogate measure of handwashing behaviour is the presence of soap for handwashing in the home. Studies in several resource-poor settings have found an association between the presence of handwashing materials and lower rates of diarrhoea or respiratory disease (Luby & Halder 2008; Peterson et al. 1998; Dubois et al. 2006). The presence of any kind of soap in the home for handwashing, and soap at specific handwashing places, can be objectively measured and feasibly incorporated into large studies. Community-based studies assessing an association between these surrogate household measures and health outcomes in children living in sub-Saharan Africa have not been published.

Our objective was to evaluate whether the presence of soap in the home or a fixed, designated location for handwashing with soap and water was associated with lower prevalence of diarrhoea and acute respiratory illness in children < 5 years old in rural western Kenya.

The protocol for PBIDS and the handwashing survey were approved by the Ethical Review Committees of CDC and KEMRI. Caretakers of participating children provided signed informed consent.

The sample population included all children < 5 years old in April 2009 living in a household that had participated in at least one PBIDS home visit between January 1 and April 30 2009 and completed a handwashing survey in April 2009.

We measured two potential surrogate measures of handwashing behaviour: presence of soap in the home and the presence of a handwashing station. These measures were collected once during the handwashing survey in April 2009. Soap presence was defined as observation of any of the following types of soap: beauty bar soap (marketed for personal cleansing), multipurpose bar soap (marketed for both household and personal cleansing), powdered soap (laundry detergent) and liquid soap. We defined a handwashing station as a specific location in the home or courtyard identified by the respondent as the place where hands are most often washed with soap and/or water present. The presence of water included running water and water in any type of storage container or basin. A fully stocked handwashing station was defined as a handwashing station where both soap and water were observed. A partially stocked handwashing station had either soap or water observed, but not both. Households that identified a specific location for handwashing where neither soap nor water was present were excluded from the handwashing station analyses to reduce likelihood of misclassifying the exposure.

Outcomes of this study were prevalence of diarrhoea and acute respiratory illness (ARI) in children < 5 years old reported by mothers or caregivers during PBIDS visits between January 1 and April 30 2009. We conducted these analyses using prevalence as an indicator of the burden of disease, which has been shown to be more strongly associated with mortality than incidence in a study of childhood diarrhoea in Ghana (Morris et al. 1996). Diarrhoea was defined as the report of ≥ 3 looser than normal stools in a 24-h period (Feikin et al. 2011). ARI was defined as the report of acute onset of cough or difficulty breathing (Feikin et al. 2011). Although interviewers recorded symptom data for each day of the previous 2 weeks, we used symptoms reported for the 3 days prior to each visit based on substantial reductions in recall for periods longer than 3 days (Feikin et al. 2010; Alam et al. 1989). We defined prevalence for each syndrome as the number of days with syndrome-specific symptoms divided by the number of days of observation. The independent association between each handwashing indicator and the prevalence of diarrhoea or ARI was estimated using mixed-effect regression models with proc mixed (SAS v 9.2, Cary, NC). Illnesses among children living in the same household may be correlated; thus, we included clustering at the household level in all regression models. Potential confounders were selected based on previously published literature and retained in the multivariate models if statistically significantly associated with the outcome (P < 0.05) (Lopez et al. 2006; Luby & Halder 2008; Pickering et al. 2010; Masangwi et al. 2010). All multivariate models were adjusted for household wealth quintile, gender, and age. Both handwashing indicators were included in the multivariate models to determine their independent association with each health outcome. We calculated the difference in adjusted square means of the prevalence between the groups with and without each indicator and estimated the 95% confidence interval around the differences.

Children < 5 years old living in homes with soap for handwashing experienced a 41% lower prevalence of diarrhoea than children living in homes without soap. The presence of soap in the home may indicate caregivers of young children are washing their hands during key events that disrupt transmission of diarrhoea pathogens to the children, such as after faecal contact and before feeding the child. In Bangladesh, observed handwashing with soap by caregivers at key events was associated with reduced diarrhoea in children < 5 years old (Luby et al. 2011a; Alam & Wai 1991). Although the presence of soap does not necessarily result in handwashing with soap, its absence minimises opportunities for the behaviour.

Given the nearly universal presence of soap in our sample, soap presence is likely an insensitive predictor of handwashing behaviour. Using an insensitive measure will limit the ability to detect a moderate effect of handwashing on child health. The small number of households without soap also increases the possibility that the association might be explained by other factors. Rather than predicting handwashing behaviour, the absence of soap may only discriminate the small fraction who are least likely to wash. Because soap presence was widespread in our sample, a community-wide intervention to promote maintenance of soap in the home is unlikely to have a large impact on reducing child morbidity. However, diarrhoea prevalence was higher in children in households without soap, and half of the households without soap were in the poorest two quintiles. An intervention that identifies households without soap and provides soap to the households that have financial constraints could reach children at higher risk of illness.

We did not find reduced prevalence of ARI in children living in households with soap. Handwashing with soap after contact with respiratory secretions may be rare; a study in Bangladesh reported that no participants were observed to wash hands after coughing or sneezing into their hands (Nasreen et al. 2010). Furthermore, repeated handwashing throughout the day may be necessary to interrupt transmission from hands to nasal mucosa because of frequent hand-to-face contact (Nicas & Best 2008; Jefferson et al. 2011).

Our study is consistent with previous research showing an inverse association between soap presence in the home and child diarrhoea but no association with ARI. During epidemic cholera in Zambia, healthy controls had significantly higher odds of having hand soap in the home compared with cholera cases (Dubois et al. 2006). In a refugee camp in Malawi, diarrhoea incidence was lower when soap was present in the household compared with when soap was absent. (Peterson et al. 1998). A study in urban Bangladesh found no association between observation of soap in the home and reduced cough or difficulty breathing in children after adjustment for household wealth (Luby & Halder 2008).

We did not find an association between the presence of a fixed handwashing station and prevalence of diarrhoea or ARI in children < 5 years old. Evidence for a beneficial relationship between a designated place for handwashing and diarrhoea or ARI in children is mixed. In the previously mentioned investigation of a cholera epidemic in Zambia and one study of diarrhoea in Bangladesh, a designated handwashing station was not significantly associated with reduced illness (Dubois et al. 2006; Luby et al. 2011b). A randomised controlled trial in Bangladesh that placed and maintained a handwashing station with soap and water in homes did not reduce secondary transmission of influenza (DiVita et al. 2011). In contrast, two observational studies in Bangladesh found that a convenient place to wash hands with water inside the home was associated with a lower prevalence of respiratory illness in children (Luby & Halder 2008; Luby et al. 2011b). In rural western Kenya, it is common practice to bring a pitcher and basin to the persons to wash hands before meals; by limiting our observation to fixed locations for handwashing, we may have underestimated the proportion of households among whom regular handwashing occurred.

Unexpectedly, the presence of a handwashing station with both soap and water was associated with a significantly higher prevalence of ARI in children. Given the very limited data available for this specific analysis, it is possible an unidentified bias influenced the estimate. Only six households had soap and water present at a designated handwashing station at the time of the interview. The association may have been due to confounding that was not measured; respondents with a handwashing station may have been more health conscious and reported more symptoms.

Although we did not see an association between a handwashing station and reduced child morbidity, placing soap and water near latrines, food preparation, or eating areas may promote handwashing with soap by acting as a visual reminder. A handwashing station placed near the kitchen or latrine may create an environment favourable to handwashing with soap by reducing the effort needed to wash hands (Aunger et al. 2010; Schmidt et al. 2009; Biran et al. 2012).

Our study had several limitations. Both indicator measures of handwashing showed a lack of heterogeneity in this sample; 97% of the households showed soap to the field worker, and 1% had a handwashing station. Thus, these measures have limited usefulness to discriminate handwashing behaviour and may only serve to identify the small groups who are either the most or least likely to wash hands in this sample. The lack of heterogeneity also limits the precision of the point estimates and increases the possibility that the associations are confounded by characteristics we did not measure or residually confounded by characteristics we included. Furthermore, we did not observe handwashing behaviour and cannot confirm the measured indicators are associated with behaviour, although other studies have reported an association (Luby et al. 2009; Biran et al. 2005).

The handwashing surrogate measures were collected in a single visit at the end of the study period after the health data were collected. The presence of handwashing materials during this visit may not indicate presence prior to the measure of health outcomes or their usual presence in the household. In Bangladesh, a study that measured the presence of soap repeatedly in the same households over one month showed that a majority had soap present at any one visit, but only half had soap present during every visit, reflecting the inconsistency in soap availability in the home (Gadgil et al. 2011).

Handwashing behaviour and the presence of materials for washing hands may change seasonally in response to changes in availability or perceived risk of infection. The presence of a handwashing station was associated with a water source on the premises during the dry season. Many more households may have an accessible handwashing station in the wet season when four times as many households reported a water source on the premises, compared with the dry season. The incidence of diarrhoea and ARI also varies seasonally, and handwashing may increase when perceived risk of disease is high. We used health data collected during the same season as the handwashing measures survey to reduce the effect of seasonality, but the association may differ during the wet season.

A study in Kenya observed that handwashing with water was more common than washing with soap, and in Bangladesh, washing with water was protective against diarrhoea in children (Aunger et al. 2010; Luby et al. 2011a). The presence of soap in the home or at a handwashing station would not identify those who wash with water and lead to misclassification of a handwashing practice that could decrease illness.

Several factors regarding handwashing behaviour in sub-Saharan Africa warrant further exploration. There are no published data describing handwashing behaviour after contact with respiratory secretions or knowledge about these events as critical times for handwashing in Africa. This information could inform large-scale handwashing programmes that aim to reduce the high burden of ARI among young children in Kenya. A fixed, designated location for handwashing may not represent a handwashing station in this region. Additional research could help to elucidate the times at which soap and water are brought out for handwashing, how often this is done during the day, whether or not this type of handwashing station is associated with childhood illness, and how these materials serve as a cue to wash hands when recommended.

The majority of households in our sample had soap, but few maintained the soap at a specific handwashing station. The presence of observed soap in the home was protective against diarrhoea, but not ARI, among children < 5 years old. Despite being low-cost, objective measures, the homogeneity in both surrogate measures suggests the limited utility of these data to determine whether the presence of soap or a handwashing station can differentiate people who wash their hands at key times for transmitting pathogens to and from hands. Soap should be universally available to facilitate the handwashing behaviour that can reduce the burden of diarrhoea among young children in western Kenya. The behavioural and health effects of efforts to promote maintaining soap in an accessible location where it may act as a visual cue to washing hands at times of pathogen transmission should be further investigated.