Given continued population growth, utilities in many cities in sub-Saharan Africa struggle to provide sufficient domestic water to meet residents’ needs [1
]. The gap between supply capacity and demand manifests itself through water rationing, with supply interruptions common in many urban neighbourhoods [2
]. Faced with water rationing, residents not only have to find alternative sources of water when piped supplies are unavailable and store water to prepare for interruptions, but they may also be exposed to water contamination events associated with pressure drops within the supply system [3
]. Furthermore, stored water may often become contaminated [4
], and recent systematic review evidence suggests the extent of stored water contamination is greater among those using piped supplies [5
A recent nationally representative household survey in Ghana, the Living Standards Survey Round 6 (GLSS6), was conducted in 2012–2013 and included an additional module on drinking water quality. Although some household surveys have previously collected water quality parameters [6
], this was the first nationally representative survey published which tested for a microbiological water quality parameter. The GLSS6 [7
] suggested that rates of detectable E. coli
, routinely used to indicate the presence of faecal contamination, were lower in sachet water than in piped water, consistent with low microbial contamination in packaged water relative to many other sources in a recent systematic review [8
]. However, more in-depth analyses of the water quality data from this survey, examining patterns of microbial contamination while controlling for important confounding factors (e.g., water storage practices), have not yet been conducted.
In many sub-Saharan African cities, packaged waters have emerged as an alternative source of drinking water, both for those with piped connections and those without. In urban Ghana, consumption of “sachet” water (water sold in 500 mL sealed plastic bags) has grown rapidly [9
]. Bottled water is also available, but less commonly used as a main source of drinking water. Larger corporate sachet producers and smaller producers registered with the regulatory bodies, the Ghana Standards Authority and Food Standards Authority, typically use pre-filtration, ultra-violet and reverse osmosis treatment as part of their production processes. However, unregistered producers also exist whose production processes have not been subject to regulatory scrutiny [9
], and who often produce cheaper sachets to undercut more established sachet brands. It has been suggested that those living in poorer neighbourhoods may be particularly exposed to these lower quality brands [10
], with associated concerns around sachet water safety.
This study therefore draws on GLSS6 data to examine factors associated with microbial contamination of point-of-consumption drinking water. In doing so, it seeks to examine the plausibility of microbial contamination patterns from the new household survey water quality module. As a secondary objective, the study examines risk factors for packaged (sachet and bottled) water contamination, both at the point of consumption and point of sale, while controlling for confounders. It aims to assess whether poorer households are differentially exposed to microbially contaminated packaged water. It also aims to quantify the protective effect, if any, of packaged water against point-of-consumption microbial contamination.
This analysis suggests that E. coli
contamination patterns in the GLSS6 are plausible and broadly consistent with evidence from elsewhere. For example, systematic review evidence suggests household stored water is often more contaminated than water taken directly from the source [4
], and in the GLSS6, uncovered stored water was associated with higher contamination (Table 2
and Table 3
). The relative extent of E. coli
contamination across the different source types is also broadly consistent with recent systematic review evidence [8
], in that there was some contamination even among “improved” sources such as piped supplies and boreholes.
The findings also broadly support the WHO/UNICEF Joint Monitoring Programme (JMP) “ladder,” which differentiates surface water from other unimproved sources such as unprotected wells, and piped water to the premises from other forms of improved supply. This “ladder” forms the basis for proposed post-2015 monitoring of differing levels of service access, over and above “improved” versus
“unimproved” water sources [19
]. In the GLSS6 data, surface water was associated with the greatest risk of high levels of E. coli
contamination at the point of consumption, whilst contamination risks were higher for standpipes than for water piped to the premises (Table 3
). There was no evidence of borehole contamination differences between community-managed, NGO-managed and other arrangements.
The analysis suggests that among measured behaviours, packaged water use (which is predominantly sachet water) afforded the greatest protective effect against high levels of E. coli
contamination of water at the point of consumption (Table 2
and Table 3
). Use of packaged water reduced risk of E. coli
contamination, even relative to use of piped water onto premises. Sachet samples taken directly from packaging were less contaminated than those taken from drinking cups, supporting recent work suggesting sachet contamination increases between point of manufacture and point of sale [20
]. This is consistent with some evidence that packaged water has a protective effect against child diarrhoea [2
]. Furthermore, since many households stored water in sachets (Table 1
), it is plausible that the sachets protected against recontamination from handling and since, unlike bottles, sachet packaging cannot be reused, this may offer further protection against recontamination. This apparent protective effect of packaged water against contamination with E. coli
suggests that restricting use of sachets (for example via a ban proposed in Ghana in 2007 as reported in [9
]) could potentially have consequences for public health.
However, in absolute terms, over 30% of sachet samples tested positive for E. coli
, a proportion greater than that reported in any of the sachet water studies included in a recent systematic review [8
Packaged water is often drunk directly from the packaging, although the water can also be decanted into a glass or bottle (e.g., for consumption by a child): thus, the method of sampling from a glass may not be as representative of the conditions at the point of consumption as the other water source categories which would mostly be consumed from a glass. This is unlikely to impact the overall results, although there is the potential for overestimation of point-of-consumption contamination of packaged water due to contamination of the drinking vessel used to provide the sample.
There is no evidence here that poorer households are differentially exposed to contaminated packaged water (Table 1
) as hypothesised previously [10
], though there is evidence that households with low expenditure using surface waters are more exposed to contaminated water at the point of consumption (Table 3
). This latter finding supports the suggestion that socio-economic inequality in safe water access may be even more pronounced if water quality is taken into account in addition to water source type [6
The membrane filtration results presented here show evidence of digit preference (Figure 1
and Figure 2
), which may reflect the field-based rather than laboratory-based testing undertaken. Whilst there is a long history of examination of digit preference to assess the quality of demographic data [21
] and clinical data such as blood pressure readings [22
], there are few if any published studies examining digit preference in microbiological data from water samples. In the absence of other studies of digit preference in water microbiology data, we cannot ascertain whether the problem is particularly pronounced in this dataset. This apparent digit preference in cfu counts has a number of implications. Many studies (e.g., [23
] report water quality data in risk bandings of 1–10 or 1–9 cfu/100 mL, 11−100 or 10–99 cfu/100 mL, and greater than 99 or 100 cfu/100 mL. Despite this banding, systematic reviews have found no evidence for increased risk of diarrhoeal disease as cfu/100 mL values increase above 1 cfu/100 mL [25
], so these categories are essentially arbitrary. Where there is digit preference because those interpreting membrane filtration results round cfu counts to pleasing numbers, these arbitrarily chosen risk interval boundaries may be inappropriate. This is because the interval boundaries fall at the “heaped” values of 10 and 100 and these rounded values will all be assigned to either the higher or lower class. Choosing alternative risk band boundaries that do not end in zero would avoid this issue, as would using smoothing or related techniques [28
] to redistribute “heaped” values prior to reporting. As in demography and clinical medicine, it may be that analysis of end digits could be used more widely as a quality control measure in examining membrane filtration results.
These findings are subject to several sources of uncertainty, most notably with the field-based testing for E. coli
. Problems have been noted with the field implementation of the GLSS6 [11
]. For example, some of the quality control measures were inconclusive because of inconsistent recording of “blank” sample results and a lack of capacity in laboratories scheduled to undertake duplicate testing. In addition, consistency checks between the 100 mL and 1 mL samples resulted in the exclusion of approximately 10% of the overall sample. This may be due to errors in the testing procedure or recording of results, although assumed inconsistencies between the two samples may also arise due to chance, particularly where the level of contamination is low. The regression results did not change substantially after removing these potentially erroneous results.
More generally, this analysis focused only on E. coli
as faecal indicator bacteria, but the relationship between indicator bacteria and pathogen presence is complex [29
]. One review [30
] found that E. coli
counts were correlated with intestinal pathogens in only 11 out of 40 studies of recreational or drinking waters. Moreover, there is evidence from tropical environments that E. coli
can originate from non-faecal sources [31
], and that its regrowth in such environments can be affected by parameters such as soil moisture [31
]. On the other hand, E. coli
may be attenuated or inactivated in the environment more rapidly than some pathogens, so the absence of E. coli
does not guarantee the absence of pathogens. In part for these reasons, epidemiological evidence linking diarrhoeal disease risk to E. coli
in drinking water is mixed. Despite some studies showing no apparent relationship, a meta-analysis found a pooled association between diarrhoea and E. coli
presence from 14 studies [27
]. Thus, whilst for logistical and budgetary reasons a household survey module necessarily has to concentrate on a very limited set of water quality parameters, E. coli
remains the principle recommended water quality parameter [13
]. Because of the need to conduct tests in remote locations lacking laboratory infrastructure, Compact Dry EC media were used to enumerate E. coli
rather than a standard method. Although this is one of 20 products identified as suitable for E. coli
enumeration in remote or resource-poor settings [32
], there remain few comparative evaluations of such field methods and none of the Compact Dry EC media.
The small number of “source” samples taken directly from sachets (i.e.
without transferring to a drinking vessel) limits any ability to detect differences in exposure among rich versus
poor sachet users. Similarly, the small proportions of households practicing behaviours such as home water treatment also limit our ability to detect their water quality impacts. Confounding may also affect this cross-sectional survey. Sachet use is predominantly an urban phenomenon, particularly in Greater Accra, and may be associated with other unmeasured factors that protect against point-of-consumption water contamination in urban areas. Several potential predictors of contamination, such as piped supply interruptions [3
], residual-free chlorine [5
], the characteristics of the cup or glass used to serve water [33
] and sachet water brands [10
], were either not recorded in the GLSS6 or recorded too inconsistently to be usable. Although sachet use is less common in rural areas, univariate analysis indicated that the quality of sachet water in rural areas was comparable to, or better than, the quality of sachet water in urban areas (67% of sachet samples in urban areas and 81% of sachet samples in rural areas were uncontaminated by E. coli
). There are some additional influences on water contamination not considered in this manuscript, notably rainfall patterns [34
] and seasonality. There would be some potential to expand the set of risk factors for contamination considered in this analysis, though this would require spatial linkage to gridded rainfall and other datasets and the lack of detailed spatial representation in the GLSS6 somewhat restricts such analysis.
Given these issues, we propose several revisions to any future water quality module implemented alongside a household survey. Firstly, field-based quality control measures should be consistently implemented and recorded. Field teams should routinely analyse field blanks (water known to be free of E. coli contamination), with at least one blank per ten actual tests. Secondly, where non-technical staff are conducting the water quality test, they should be supported by local water quality laboratory workers. These laboratory workers should at a minimum participate in training sessions, and if possible also visit field teams during data collection to ensure that field staff are correctly and consistently following standard operating procedures. Ideally, a subset of duplicate samples should be sent on ice to laboratories for cross-checking analysis, within 24 h of collection. If this is logistically infeasible, a smaller number of samples collected from sites close to the laboratories could be cross-checked. Finally, data quality checks should be implemented alongside data collection (including comparison of 1 mL and 100 mL sample results, and examination of digit preference patterns), to identify any substantial variation between different testing teams and to take appropriate corrective measures while field work is still underway. A further challenge to be addressed is the identification of an appropriate statistical distribution for a given bacterial density dataset, where digit preference exists. This would enable the subsequent identification of statistically significant differences in replicate sample results as an additional quality control measure.