Tropical Medicine and International Health volume 14 no 12 pp 15341541 december 2009

doi:10.1111/j.1365-3156.2009.02404.x

Determinants of handwashing practices in Kenya: the role of media exposure, poverty and infrastructure
Wolf-Peter Schmidt1, Robert Aunger1, Yolande Coombes2, Peninnah Mukiri Maina3, Carol Nkatha Matiko3, Adam Biran1 and Val Curtis1
1 Department of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, UK 2 Water and Sanitation Program, World Bank, Nairobi, Kenya 3 Steadman Group Kenya, Riverside Drive Riverside, Nairobi, Kenya

Summary

background To explore how structural constraints such as lack of reliable water supply, sanitation, educational and other socio-economic factors limit the adoption of better hygiene. methods In preparation for the Kenya National Handwashing Campaign, we conducted a nationwide cross sectional survey in 800 households with two components: (i) direct structured observation of hygiene practices at key junctures (food handling, cleaning a child after defaecation, toilet use), followed by (ii) a structured interview addressing potential socio-economic, water access and behavioural determinants of handwashing. results We observed a total of 5182 critical opportunities for handwashing, and handwashing with soap at 25% of these. Handwashing with soap was more often practised after faecal contact (32%) than in connection with food handling (15%). In univariate and multivariate analysis, water access, level of education, media exposure and media ownership were associated with handwashing with soap. Only households with very poor access to water and sanitation, and with the lowest levels of education and media exposure, washed their hands markedly less than the majority of the households. conclusion The results underscore that structural constraints can limit hygiene practices in the very disadvantaged sections of a population, thus jeopardizing the potential success of hygiene promotion campaigns in those most at risk of disease. Nevertheless, the strong association of handwashing with media ownership and exposure supports the view that mass media can play a role in hygiene promotion. keywords hygiene, diarrhoea, sanitation, socio-economic status, handwashing approaches, led by the World Banks Water and Sanitation Programme (WSP) and the Ministry of Health in Kenya. The study was conceived as a baseline study for the campaign with the overall aim of guiding the programme content and allowing an impact evaluation after the campaign. Previous large-scale hand hygiene surveys have mostly relied on questionnaires and spot checks of handwashing facilities (Scott et al. 2005; Luby & Halder 2008) but rarely observed actual hygiene behaviour. Small-scale studies on handwashing behaviour in sub-Saharan Africa observed handwashing prevalences between 3% and 29% (Curtis et al. 2009). In this large-scale national survey we applied directly observed hygiene behaviour as the main outcome measure in combination with a questionnaire survey, allowing a unique insight into the socio-economic, psychosocial, and behavioural determinants of handwashing across Kenya. The aim of this analysis was to determine to what extent

Introduction Simple hygiene behaviours, especially handwashing with soap, have been suggested to reduce the occurrence of gastro-intestinal infections, respiratory infections, trachoma, helminths and skin infections in poor settings (Cairncross & Feachem 1991; Curtis & Cairncross 2003; Luby et al. 2005; Rabie & Curtis 2006; Ejemot et al. 2008; Fung & Cairncross 2008). Even if the true effect of handwashing on these conditions were much lower than evidence (which almost certainly contains an element of bias) suggests, large-scale promotion of handwashing with soap may be a very cost-effective intervention to reduce the burden of disease in low income settings (Laxminarayan et al. 2006). This paper reports ndings from a cross-sectional survey to evaluate handwashing practices in Kenya prior to a nationwide programme of handwashing promotion through mass media and community-based 1534

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

structural constraints such as water access, media access and other poverty-associated factors limit the adoption of better handwashing practices.

Household interviews A questionnaire interview was conducted with the primary caregiver by the eld worker immediately after the observation period. The interview covered basic demographic and socio-economic characteristics, as well as psychosocial questions on attitudes and beliefs about hygiene, based on previous questionnaires used in evaluation of national handwashing campaigns (Scott et al. 2005). The questionnaire also covered media ownership, media exposure and social activities, as well as type of water supply and sanitation facilities. Variables and statistical analysis To facilitate analysis, we combined some questionnaire and observation items into summary variables. Media ownership (TV, radio, postal address, e-mail address) and media exposure in the last month (newspaper, radio, TV, movie) were summarized in two variables indicating the number of different items that households owned or the number of types of media households were exposed to. Social activities (parents association meetings, road shows, church mosque, public meetings, going to a restaurant bar) were summarized in a similar way. Education was classied into four levels: primary school not completed, primary school completed, secondary school completed, and specialized training. The unit of analysis was the potential handwashing occasion as observed by the eld workers. The main outcome was the proportion of occasions followed by washing both hands with soap. We pooled all occasions within a household regardless of who was observed. For the univariate and multivariate analysis of the association between explanatory variables and handwashing with soap, we used an additive binary regression model (distribution: binomial; link: identity) to explore the differences between different exposure groups (Stata 10). Clustering at household level was accounted for by the use of generalized estimating equations (GEE). There was no statistical evidence that the analysis was affected by clustering at village level (all variables were either individual or household level characteristics). Adjusting for clustering at village level produced nearly identical P values as the household cluster models in the univariate analysis. However, since the multivariate models did not converge when adjusting for village clustering, we only adjusted for household level clustering throughout. Results Socio-demographic characteristics of the study population are shown in Table 1. A small minority of study 1535

Methods Study population and survey procedures We recruited households from seven of the eight provinces of Kenya; the North Eastern province was left out because of logistic and security problems. One district was randomly chosen from each province, except for Nyanza province, where two districts were randomly chosen. Within a district, two sub-locations (the smallest administrative unit in Kenya) were selected according to probability proportional to size. A listing of villages was then obtained for each of these sub-locations. The interviews were spread across villages in the same sublocation by choosing every fourth household from the list of families with children under the age of ve until 100 households per district province had been selected, aiming at an overall sample size of 800 households. On average, ve households per village were selected (range 129). Twenty households dropped out on the day of observation, after conrming their availability the previous day, so the sample was extended to replace these households by following the same sampling approach. The overall sample size was 802. The survey was conducted in March 2007. Structured observation Observers interviewers with extensive experience in conducting household surveys were provided by a market research company (Steadman Group, Kenya). They received extensive training in structured observation techniques for this survey. The observer interviewer visited eligible households to seek consent from head of household and primary caregiver and to book an appointment for the following day. Specic study objectives were kept discrete to avoid biasing behaviour, although the household was told the study concerned family life. On the following day the observer interviewer arrived at the household a few minutes before 6 am and strategically positioned himself herself to observe handwashing behaviour at key junctures: after faecal contact (e.g. after toilet or cleaning a child following defaecation), before preparing food, and before eating. All observations started at 6 am unless the household woke up late. The observers collected data about the type of key junctures occurring, and whether hands were washed with water or soap. The observation continued for 3 h.

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

Table 1 Socio-demographic characteristics of caregivers and households n Overall Carers characteristics Age of primary care giver (years) <18 1924 2530 3135 3640 4145 >45 Marital status Married Widowed Single Cohabiting Divorced separated Education Did not complete primary school Completed primary school Completed secondary school Specialized training Literacy Unable to read Able to read Household characteristics Water source Tap piped water in the house Tap in the yard compound Communal tap public bore hole Well springs river dam A well in the yard Rain water Water truck Vendor Water scarcity None 12 months 34 months 56 months >6 months Sanitation facility Open defaecation Public toilet Uncovered pit latrine Covered pit latrine Covered pit latrine private Flush toilet shared Flush toilet private 802 % 100

12 241 236 182 66 35 16 643 15 102 13 10 156 338 187 105 73 702

2 31 30 23 8 4 2 82 2 13 2 1 20 43 24 13 9 91

63 118 210 68 237 26 20 1 44 508 102 112 25 21 15 32 358 129 89 85 78

8 15 27 9 30 3 3 0 6 66 13 15 3 3 2 4 46 16 11 11 10

The eldworkers observed a total of 5182 events (key junctures) at which handwashing with soap should have been practised from a public health perspective (Table 2). Handwashing with soap was observed on 24% of such occasions and with water alone on a further 25%. Handwashing with one hand only was observed at 9% of occasions. Most events involved the primary care giver and were due to potential faecal contamination. Handwashing with soap was more common after faecal contamination (32%) than at other key junctures (15%). Table 3 shows the univariate analysis of determinants of handwashing with soap (i) all events combined and (ii) restricted to faecal contamination events. There was no association with age of caregiver, but marked regional differences in handwashing practices, with the highest rates being found in and around Mombasa and Transzoia, and among native Kiswahili speakers. As expected, higher levels of education and literacy were associated with handwashing with soap. Having a water source in the house rather than in or outside the compound was associated with markedly higher handwashing rates. Handwashing rates among those without a tap in the house were quite similar, regardless of whether the source was a nearby tap or borehole, or a street vendor. Water scarcity hardly inuenced handwashing except in those experiencing very prolonged periods of water scarcity. The univariate analysis indicates that the number of media a respondent was exposed to and the number of media owned were strongly correlated with handwashing. There did not appear to be a threshold effect: handwashing increased in a roughly linear way with every additional

Table 2 Handwashing occasions observed n Total number of occasions 5182 Person observed Primary caregiver 3387 Secondary caregiver 623 Schoolboy 566 Schoolgirl 606 Occasions observed After defaecation 1442 After cleaning child 672 Other faecal contact 653 Before feeding child 1547 Before handling food 868 Handwashing practice following occasion None 2039 One hand 437 Both hands with water 1250 Both hands with soap 1219 % 100 65 12 11 12 28 13 13 30 17 42 9 25 24

participants reported being illiterate, relied on a street vendor for water or practised open defaecation, factors that indicate very low socio-economic status. 1536

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

Table 3 Univariate analysis of factors associated with handwashing with soap Handwash with soap (%) 24 Handwash with soap after faecal contact (%) 32

n Overall Carers characteristics Age (years) <25 2535 >35 Marital status Married Widowed Single Cohabiting Divorced separated Education Did not complete primary school Completed primary school Completed secondary school Specialized training Literacy Unable to read Able to read First language Kisii Kiswahili English Kikuyu Kamba Luhya Luo Other Number of languages spoken One Two Three More than four District Kisii Nyeri Mombasa Kisumu Transzoia Machakos Nairobi Bungoma Media used in past month None One Two Three More than four Number of media owned One Two Three Four More than ve 802

% 100

253 418 117 643 15 102 13 10 156 338 187 105 73 702 91 189 29 129 101 125 80 42 19 285 362 128 97 99 100 102 101 99 103 101 67 226 213 215 81 108 263 198 171 62

32 53 15 82 2 13 2 1 20 43 24 13 9 91 11 24 4 16 13 16 10 5 2 36 46 16 12 12 12 13 13 12 13 13 8 28 27 27 10 13 33 25 21 8

23 24 21 24 21 25 20 5 18 24 25 25 18 24 12 31 24 25 18 25 21 23 24 23 24 22 12 30 30 24 33 18 21 28 15 19 25 25 31 14 24 25 26 27

0.38

30 32 30 32 21 33 28 11 23 32 33 36 23 32 16 42 39 34 25 32 29 34 34 29 32 30 16 37 39 34 43 25 29 34 21 27 30 35 41 18 30 31 37 39

0.80

<0.001

0.05

<0.001

<0.001

0.01

0.02

<0.001

0.71

0.57

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

2009 Blackwell Publishing Ltd

1537

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

Table 3 (Continued) Handwash with soap (%) 20 20 25 26 30 Handwash with soap after faecal contact (%) 28 28 32 33 36

n Social activities in past month None One Two Three More than four Household characteristics Water source Tap piped water in the house Tap in the yard compound Public tap bore hole Well or surface water Vendor Water scarcity None 12 months 34 months 56 months >6 months Sanitation facility Open defaecation Public toilet Uncovered pit latrine Covered pit latrine Covered pit latrine private Flush toilet shared Flush toilet private 77 295 227 121 82

% 10 37 28 15 10

P <0.001

P 0.12

63 118 278 284 44 508 102 112 25 21 15 32 358 129 89 85 78

8 15 35 36 6 66 13 15 3 3 2 4 46 16 11 11 10

33 23 22 22 19 24 25 23 20 13 6 22 23 24 25 23 27

0.02

46 27 31 30 29 32 29 31 27 17 8 29 30 33 35 31 37

0.003

0.06

0.07

<0.001

<0.001

Binomial regression (additive model, family: binomial, link: identity), adjusted for clustering at the household level (GEE). P values indicate overall evidence for differences between categories.

item exposed to or owned. To a lesser extent the number of social activities participated in within the last month was also positively correlated with handwashing. The nal multivariate models are shown in Table 4 (water scarcity was omitted due to colinearity). The two models largely conrm the univariate analysis. The association between education and handwashing was reduced, which indicates that the effect of education on handwashing is partly mediated through other variables, e.g. media ownership and media use. The other factors associated with handwashing in the univariate analysis largely remained after adjusting for confounders. The within-household intra-cluster correlation coefcient (ICC) of handwashing with soap was 0.13, the within village ICC was 0.06. Discussion In the analysis, media ownership and media exposure were strongly associated with observed handwashing with soap. 1538

Both variables appear to contribute to handwashing independently. Structural constraints such as water access, sanitation or lack of school education were associated with lower handwashing rates predominantly in the lowest category of each factor. There are several potential limitations to our analysis. Households were selected from an existing census list of households, which could under-represent less stable and potentially less wealthy households. Evaluating national campaigns is a challenge since hygiene behaviour is difcult to measure. Proxy indicators of handwashing, like self-report, presence of soap or beliefs regarding the health benets of hand washing, have been shown to correlate poorly with observed behaviour (Biran et al. 2008). Directly observed handwashing by structured observation is not frequently used as an outcome measure in large studies because of assumed logistical constraints. It has also been suggested that direct observation may change the behaviour of those observed. However, earlier studies have found little evidence for a

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

Table 4 Multivariate analysis of factors associated with handwashing with soap All occasions % change Media exposure Increase per item exposed to Media ownership Increase per item owned Education Did not complete primary school Completed primary school Completed secondary school Specialized training Water source Tap piped water in the house Tap in the yard compound Public tap bore hole Well or surface water Vendor Sanitation facility Open defaecation Public toilet Uncovered pit latrine Covered pit latrine Covered pit latrine private Flush toilet shared Flush toilet private +2.6 +1.6 Ref +4.7 +2.8 +1.8 Ref )9.1 )9.7 )7.9 )16.0 Ref +11.4 +12.3 +12.5 +12.7 +9.6 +6.9 P 0.001 0.057 0.019 0.256 0.563 0.027 0.010 0.050 0.000 0.020 0.000 0.000 0.001 0.020 0.131 Faecal contact % change +2.1 +4.2 Ref +6.5 +4.2 +6.5 Ref )14.6 )10.9 )10.8 )18.3 Ref +11.7 +15.6 +17.7 +16.9 +9.3 +6.8 P 0.043 0.000 0.021 0.223 0.143 0.009 0.036 0.050 0.004 0.068 0.000 0.000 0.001 0.085 0.272

Binomial regression (additive model, family: binomial, link: identity), adjusted for clustering at the household level (GEE).

marked effect (Curtis et al. 2001; Scott et al. 2007). The present study which included 800 households across Kenya demonstrates that the logistical and staff constraints of using structured observation to measure handwashing are not insurmountable, even for large-scale national surveys to evaluate hygiene promotion campaigns. This was largely a post-hoc analysis aiming at hypothesis generation. A problematic issue for this analysis lies in the interpretation of the variables. For example, number of months with water scarcity and number of media exposed to do not fully capture complex issues like water scarcity and media exposure. The survey was conducted at the end of the dry season, so it can be assumed that water availability may have been compromised in some participating households. We did not obtain further information on these topics because the questionnaire (which included a large number of psychosocial questions) was already very long. For the same reason we were unable to calculate a more direct scale for socio-economic status, which could have included items on profession and income. Future studies may benet from putting more emphasis on these topics, which are, as our analysis clearly suggests, important determinants of hygiene practices.

The multivariate analysis suggests that media ownership and exposure were important determinants of handwashing behaviour. There are several ways to interpret this nding. First, there is little doubt that media ownership and use are markers for socio-economic status, which is likely to correlate with hygiene behaviour. The multivariate analysis (adjusted for a range of other socio-economic proxies) suggested that education and wealth alone do not explain the association between media exposure and handwashing, although residual confounding by unmeasured wealth indicators may have contributed to this nding. Possibly, media ownership and use stand for a certain positive attitude towards the adoption of new things, be it in terms of behaviour or consumer products, to some extent independent of socio-economic class. The tendency to adopt a modern lifestyle has previously been identied as a predictor for hygiene behaviour (Curtis et al. 1995). While the association between media exposure and handwashing as suggested by this analysis needs to be treated with caution, our ndings are consistent with the idea that exposure to media may play a role in the formation of hygiene behaviours. Prior to the survey, cholera outbreaks had prompted the nationwide 1539

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

promotion of handwashing via mass media, which might partly explain the association between media exposure and handwashing in this analysis, as well as the relatively high handwashing prevalence observed compared to other surveys (Curtis et al. 2009). Our earlier ndings from a study in Ghana showed that mass media hygiene promotion campaigns reach large parts of the population and lead to pronounced changes in self-reported attitudes towards hygiene and self-reported practice (Scott et al. 2005). However, in the same analysis we also found that different communication channels complement each other, each reaching somewhat different sections of a population. Thus, the association between media and handwashing in this analysis does not mean that promotion campaigns should only focus on mass media, especially if these fail to reach the poorest sections of a population. Perhaps one of the most obvious results of our analysis is that a large proportion of handwashing behaviour was left unexplained by the variables included in the present analysis. Handwashing may be a simple behaviour to execute, but the formation of hygiene behaviours is complex and not yet well understood. Our analysis of the psychosocial variables included in the questionnaire revealed that participants who reported handwashing as a habit or automatic reex in certain situations rather than a means to improve health had the highest handwashing rates regardless of socio-economic status (Bob Aunger, personal communication). Religious afliation may also inuence handwashing behaviour. Areas with a high proportion of Muslims (like Mombasa) and native Kiswahili speakers (who are mostly Muslims) showed high handwashing rates. While factors such as water access, sanitation and education played a role, the differences between the categories were small in absolute terms. The only groups with considerably lower handwashing rates were those with very limited media ownership, with experience of prolonged water scarcity and those practising open defaecation. Most likely, these households represent those with the highest risk of disease who would benet most from improved hygiene. A link between socio-economic factors, water access and education is a common nding in studies using direct observation of handwashing as outcome (Gorter et al. 1998; Sakisaka et al. 2002). This raises an important question that has long been in the focus of thinking in public health in general, i.e. whether intervention efforts should target the whole population or rather those at highest risk of disease (Rose 1985). Many poor or middle income countries in which parts of the population are at high risk of diarrhoea and child mortality show a very large disparity between rich and poor in terms of wealth and 1540

mortality much larger than for example cardiovascular risk in wealthier countries which makes the targeted approach potentially more effective. On the other hand, in the case of hygiene promotion, the population approach through mass media is the less cost intensive method compared to a community approach, so one could argue that mass media may always be a worthwhile addition to efforts on the ground, especially those targeting the water and sanitation infrastructure. Mass media may also have other effects, e.g. in changing social norms and the attitudes of health workers and teachers which can, in turn, inuence the practices of the poorest section of the population. In conclusion, this analysis conrms the link between education, socio-economic status, water access and handwashing practices. The results also suggest that media exposure may play a role in the formation of hand hygiene practices independent of structural constraints, and therefore supports the concept of nationwide hygiene promotion campaigns. Such campaigns however, are at risk of neglecting the poorest section of a population with the highest disease risk, who may need to be targeted by community based approaches and infrastructure measures, especially improving access to water and sanitation. Acknowledgements We thank the Kenyan families who participated in the study. We thank the Water and Sanitation Program Africa for commissioning the original study on behalf of the Ministry of Health, Republic of Kenya; Rufus Eshuchi, National Coordinator for the Kenya Handwash Campaign, and Jason Cardosi of the World Bank for access to data. The work was funded by the Water & Sanitation Program (WSP) Africa on behalf of the Ministry of Health, Republic of Kenya. References
Biran A, Rabie T, Schmidt W et al. (2008) Comparing the performance of indicators of hand-washing practices in rural Indian households. Tropical Medicine and International Health 13, 278285. Cairncross S & Feachem R (1991) Environmental Health Engineering in the Tropics, 2nd edn. John Wiley and Sons, Chichester. Curtis V & Cairncross S (2003) Effect of washing hands with soap on diarrhoea risk in the community: a systematic review. The Lancet Infectious Diseases 3, 275281. Curtis V, Kanki B, Mertens T et al. (1995) Potties, pits and pipes: explaining hygiene behaviour in Burkina Faso. Social Science and Medicine 41, 383393.

2009 Blackwell Publishing Ltd

Tropical Medicine and International Health W.-P. Schmidt et al. Determinants of handwashing in Kenya

volume 14 no 12 pp 15341541 december 2009

Curtis V, Kanki B, Cousens S et al. (2001) Evidence of behaviour change following a hygiene promotion programme in Burkina Faso. Bulletin of World Health Organisation 79, 518527. Curtis VA, Danquah LO & Aunger RV (2009) Planned, motivated and habitual hygiene behaviour: an eleven country review. Health Education and Research 24, 655673. Ejemot RI, Ehiri JE, Meremikwu MM & Critchley JA (2008) Hand washing for preventing diarrhoea. Cochrane Database Systematic Reviews 23, CD004265. Fung IC & Cairncross S (2008) Ascariasis and handwashing. Transactions of the Royal Society of Tropical Medicine and Hygiene 103,215222. Gorter AC, Sandiford P, Pauw J et al. (1998) Hygiene behaviour in rural Nicaragua in relation to diarrhoea. International Journal of Epidemiology 27, 10901100. Laxminarayan R, Chow J & Shahid-Salles SA (2006) Intervention cost-effectiveness: overview of main messages. In: Disease Control Priorities in Developing Countries, 2nd edn (eds DT Jamison, JG Breman & AR Measham) Oxford University Press and The World Bank, New York, pp. 3558. Luby SP & Halder AK (2008) Associations among handwashing indicators, wealth, and symptoms of childhood respiratory ill-

ness in urban Bangladesh. Tropical Medicine and International Health 13, 835844. Luby SP, Agboatwalla M, Feikin DR et al. (2005) Effect of handwashing on child health: a randomised controlled trial. Lancet 366, 225233. Rabie T & Curtis V (2006) Handwashing and risk of respiratory infections: a quantitative systematic review. Tropical Medicine and International Health 11, 258267. Rose G (1985) Sick individuals and sick populations. International Journal of Epidemiology 14, 3238. Sakisaka K, Wakai S & Wongkhomthong SA (2002) Domestic hygiene behaviour of mothers with children aged 05 years old in Tayabo village, Nueva Ecija, The Philippines. Asia Pacic Journal of Public Health 14, 9198. Scott BE, Schmidt W, Aunger R et al. (2005) Marketing hygiene behaviours: the impact of different communications channels on reported handwashing behaviour of women in Ghana. Health Education and Research 23, 392401. Scott BE, Lawson DW & Curtis V (2007) Hard to handle: understanding mothers handwashing behaviour in Ghana. Health Policy and Planning 22, 216224.

Corresponding Author Wolf-Peter Schmidt, Department of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, Keppel Street, WC1E 7HT London, UK. Tel.: +44 020 7927 2461; Fax: +44 020 7636 7843; E-mail: wolf-peter.schmidt@lshtm.ac.uk

2009 Blackwell Publishing Ltd

1541