Reconsidering ‘appropriate technology’: the effects of operating conditions on the bacterial removal performance of two household drinking-water filter systems

We examined the performance of two household water treatment and safe storage (HWTS) systems, the Danvor plastic biosand filter and the Potters for Peace Filtron ceramic filter, under ideal as well as modified operating conditions using systematic and comparable measurements. The operating variables for the biosand filter were (i) pause times between filtration runs, (ii) water-dosing volumes and (iii) the effluent volume at which a filtered water sample was collected. For the ceramic filter we examined overflow filtration versus standard filtration. We used the bacterial indicators of total coliforms and Escherichia coli to quantify microbiological removal. With the biosand filter, a 12 h pause time had significantly higher total coliform removal than a 36 h pause time at the 20 l collection point (79.1% versus 73.7%; p < 0.01) and borderline significance at the 10 l collection point (81.0% versus 78.3%; p = 0.07). High-volume filtration (20 l) had significantly lower total coliform removal efficacy than low-volume (10 l) filtration at the 10 l collection point (81.0% versus 84.2%; p = 0.03). We observed a decreasing trend in total coliform removal by sample collection volume with the highest removal efficacy at the 5 l sample collection point (versus at the 10 and 20 l collection points). Using the ceramic filter, mean total coliform and E. coli removal were significantly lower (p < 0.01) in overflow filtration than in standard filtration. The findings indicate that operating conditions can reduce the effectiveness of the systems in a field-based setting and increase environmental risk exposure.


Introduction
Worldwide 1.1 billion people lack access to sufficient quantities of safe drinking water [1]. Unsafe drinking water, combined with insufficient water supply for sanitation and hygiene, is responsible for an estimated 4 billion cases of diarrhoeal disease, 1.79 million diarrhoea deaths and 3.7% of the global burden of disease [1]. Improving access to clean water, an important policy imperative in international development for decades, is receiving renewed and increased attention; for example, goal 7 (target 10) of the Millennium Development Goals (MDGs) aims to 'reduce by half the proportion of people without sustainable access to safe drinking water'.
Access to safe water can in principle be achieved through centralized water treatment and/or point-of-use treatment. Widespread expansion of centralized water treatment (also referred to as 'scaling up') has been slow, particularly in urban fringes, poor rural areas and indigenous communities, largely due to the high cost and technical complexities of the required infrastructure [2]. Further, even when safe water is available through a communal source, which is a common form of water provision in many low-income communities, there remains a high risk of contamination between the time of collection and consumption in the home [3]. Household water treatment and safe storage (HWTS) interventions have therefore been considered low-cost and effective alternatives for reduction of diarrhoeal disease that can be implemented at the point of use [1,4].
The effectiveness of an intervention in conditions of actual use and operation, referred to as community-based effectiveness, can be dramatically different from its efficacy in controlled laboratory or trial settings due to differences in the physical environment (e.g. water source characteristics) and user behaviour. Much of the previous HWTS literature has addressed the technical and economic feasibility of these systems, and how these factors impact household adoption. A separate body of research has established the role of hygiene behaviour in achieving significant reductions in the incidence or severity of diarrhoeal diseases [5][6][7]. However, there have been few studies on the role of behavioural parameters that determine operating conditions on the performance of HWTS systems in reducing exposure to unsafe water [8,9], with just two studies identifying potential non-ideal behaviours, one of which quantified its effects on performance [10,11].
We examined the performance of two HWTS systems, the Danvor plastic biosand filter and the Potters for Peace Filtron ceramic filter, under ideal as well as modified operating conditions using systematic and comparable measurements. Laboratory studies conducted in controlled environments have established faecal coliform or Escherichia coli (E. coli) removal rates of 93-99% [12][13][14] for the biosand filter and 99.9% for the Filtron ceramic filter [15]. However, the bacterial removal of the biosand filter in developing country settings has been less than ideal, varying from less than 60% to 99.9% removal [14,[16][17][18][19][20][21]. Similar variation in bacterial removal is reported for field studies on the Filtron ceramic filter [22,23]. The variable performance of the systems in household settings mirrors the situation for other household environmental technologies, such as alternative cooking stoves [24,25].
This study establishes systematic evidence for the robustness of the two HWTS technologies to specific environmental and behavioural conditions, which can be used to inform the choice and delivery of interventions, for example in contributing to the MDGs' safe water target. In addition to its application to HWTS interventions, this study provides an example of an analysis that helps design, evaluate and deliver environmental health technologies to resource-poor settings based on systematic evaluation of actual performance of the technology in reducing environmental risk exposure, and therefore contributes to our definition of 'appropriate technology', which has thus far primarily utilized design and operation, and not outcome-based indicators of technology performance.

Study design
We conducted a series of experiments on the biosand and ceramic filters that included ideal operating procedures as well as selected operating conditions that deviated from ideal use, yet still fell within each system's current recommended usage guidelines. The parameters examined were based on field experiences observing how each unit may be used, and are shown in figures 1(a) and (b).
To reflect these operating conditions, experiments with the plastic biosand filter included the following (figure 1(a)). Access to water supply services is defined as the availability of at least 20 l per capita per day (lpcd) from an improved water source within 1 km of the user's dwelling, a portion of which is used as drinking water [26]. The water dosing volumes used for experiments were based on estimates for the minimum drinking water requirements; 3.0 lpcd to maintain adult hydration and 5.5 lpcd for lactating women engaging in moderate physical activity in above-average temperatures [27,28]. Comparison of the bacterial reduction between scenarios 1 and 2 addresses the question of how filtration frequency affects bacterial removal, and comparison of scenarios 2 and 3 addresses the question of how water dosing volume affects bacterial removal. Comparing bacterial reduction across the sample collection points in each scenario addresses the question of the variation of the filter's removal effectiveness throughout a filter run.
The experiments for the Filtron ceramic filter included (figure 1(b)): • standard filtration: with influent volume less than the filter's capacity; • overflow filtration.
Filter overflow is an important potential limitation to the effectiveness of the ceramic filter because the place where the filter lip fits onto the rim of the collection receptacle is not tightly sealed, allowing for water to seep through, hence potentially increasing the bacterial load of the filtered water.
We did not perform tests on water dosing volume or filtration frequency for the ceramic filter because the bacterial removal mechanisms of the ceramic filter, specifically small pore size and its colloidal silver coating, are relatively insensitive to these factors in the short term. We also did not test the removal efficacy of the ceramic filter at various sample collection points because bacterial removal is not very sensitive to this parameter with a relatively constant flow rate.
Using one unit of each filter type, we implemented each biosand filter experiment in a sequence of 16 identical trials and each ceramic filter experiment in a sequence of 10 identical trials (see supplementary online material for experimental procedures; available at stacks.iop.org/ERL/2/024003). We used membrane filtration to detect E. coli and total coliforms because it allows the measurement of bacterial indicators as a continuous variable, hence quantifying bacterial removal in each trial (see supplementary online material for detailed methods and discussion of implications). E. coli serves as an indicator of faecal pollution; total coliforms, though not useful as an index of faecal pathogens, can be used as an indicator of the effectiveness of treatment and performance of the biofilm [29].

Statistical analysis
Bacterial removal for each trial was the ratio of total coliform and E. coli counts from samples of filtered water to those of unfiltered water.
We examined normal probability plots, histograms and skewness and kurtosis measures for all trials related to each experiment to test if the data were normally distributed. We used Bartlett's test for equal variance. Neither the normality nor the equal variance assumptions were violated.
For the biosand filter, we used unpaired two-sided means tests with equal variance to compare the mean bacterial removal of the biosand filter under different scenarios. We used one-way analysis of variance (ANOVA) to test the null hypothesis that bacterial removal efficacy at various 5, 10 and 20 l collection points was equal. We used paired t-tests for comparison of bacterial removal in standard versus overflow filtration with the ceramic filter because we compared bacterial removal before and after overflow filtration on the same filter unit.
Previous research has indicated a negative association between bacterial removal efficacy of the biosand filter and the bacterial content of the source water [12]. We examined whether the proportional reduction in the bacterial indicators was affected by the bacterial content of the unfiltered source water using Pearson correlation coefficients between bacterial content of source water and bacterial removal efficacy of the filter. We also regressed the removal efficacy of the filter in the first set of experiments (scenario 3) against the order of trial to examine whether the biofilm, or schmutzdecke, was fully developed when experiments began. If the biofilm was not yet mature, bacterial removal efficacies would have increased over time.

Biosand filter
The Pearson correlation coefficient between bacterial removal and the bacterial content of the unfiltered source water was 0.12 ( p = 0.18) and 0.18 ( p = 0.35) for total coliform and E. coli, respectively. This implies that the proportional reduction in bacteria was not affected by the bacterial concentration of the source water. The regression coefficient for the relationship The water dosing volume was held constant at 20 l. b The pause time was held constant at 12 h. c p-value refers to the statistical significance of the difference between bacterial removal of each pair of operating conditions; for the sample collection point experiments, it refers to the significance of ANOVA. d We removed two outliers from the biosand filter experiment using frequent filtration and low water dosing volume. Laboratory notes indicate that influent <5 • C was poured through the filter unit on the day these outliers were recorded. This low influent temperature increased the probability that bacteria passed through the biofilm during this trial [32].
between bacterial removal and the order of trial was also statistically non-significant ( p = 0.13 at 5 l; p = 0.33 at 10 l). In other words, the bacterial removal efficacy did not increase as the trials progressed, indicating that the biofilm was fully developed prior to beginning the experiments. Table 1 and figure 2 show the results for total coliform removal in the biosand filter experiments, organized by the hypotheses in figure 1(a). Results obtained for E. coli removal follow a pattern that mirrors the one shown for total coliform removal on average (see supplementary online material). E. coli removal results had larger variance because the E. coli count in each sample was, by definition, smaller than that of total coliform, leading to a larger stochastic error in the former data. As a result none of the differences in mean E. coli removal efficacy under different operating conditions were statistically significant. p = 0.07). This finding may be a consequence of decreased activity of microorganisms that live in the biofilm and consume the pathogens in the water, when the pause period becomes too long. Previous research, using a 24 h pause time, suggests the survival (and hence the density) of microorganisms in the biofilm relies on inflow of contaminated source water [13]. Therefore less frequent inflow of contaminated source water (i.e. a longer pause period) may reduce the viability of the biofilm's removal function relative to more frequent filtration. This mechanistic hypothesis is supported by results from the 5 l collection points. Samples collected at the 5 l collection point in scenario 1 (36 h) had longer contact with the sand held in the filter bed than those at the same collection point in scenario 2 (12 h). Typically, this longer contact would increase pathogen removal through either adsorption to the sand grains or natural death from nutrient starvation [30]. However, results in table 1 show that total coliform removal at 5 l sample collection point was almost identical for the two scenarios (95.7% versus 95.9%; p = 0.84). This is probably because the decreased efficacy attributable to fewer biofilm organisms may have offset any benefits gained from increased exposure to the sand grains at 5 l in scenario 1.

3.1.2.
Scenario 2 (20 l water dosing volume) versus scenario 3 (10 l water dosing volume). We observed an inverse relationship between water dosing volume and bacterial removal. Scenario 2 (20 l) had a lower total coliform removal than scenario 3 (10 l) that was statistically significant at the 10 l collection point (81.0% versus 84.2%; p = 0.03), but not at the 5 l collection point (95.9% versus 96.5%; p = 0.52). The difference at 10 l may be an effect of the difference in hydraulic loading between the two experiments: using a high water dosing volume, water at 10 l is pushed through the biofilm and sand at a faster rate, resulting in less exposure to both of these bacterial removal mechanisms; at 5 l this is not observed because there is no difference in hydraulic loading or exposure time to the sand media between the two experiments.

Sample collection point.
There was a decreasing trend in total coliform removal by sample collection point, with the highest removal efficacy at the 5 l sample collection point for all three scenarios. In scenario 3, there was a statistically significant decrease in total coliform removal at 10 l compared to 5 l (84.2% versus 96.5%; p < 0.01). In experiments using a 20 l water dosing volume, there was a decreasing trend in total coliform removal with increasing sample collection point (5, 10 and 20 l) in both scenario 1 ( p < 0.01) and scenario 2 ( p < 0.01).
Results from flow hydraulics tests using tinted water showed that filtrate at the 5 l collection point mostly, if not entirely, comprises influent from the previous trial held in the sand media during the pause period. Filtrate at the 10 l collection point largely comprises new influent whereas filtrate at 20 l entirely consists of new influent. These results support previous research indicating that a portion of the biological removal happens during the pause times when the source water has a longer contact time with the biofilm and pore media [12,30]; as the sample is collected later in the filtration process, it has less contact with the pore media where some level of biological removal takes place.

Filtron ceramic filter
Mean total coliform and E. coli removal were significantly lower ( p < 0.01) in overflow filtration than in standard filtration (table 2). In overflow filtration, filtrate contamination per 100 ml sample varies according to (a) the ratio of overflow to filtrate water in the collection receptacle and (b) the bacterial content of the filtered and unfiltered source water. Although removal efficacy differs according to these two factors, the quality of the filtrate water will always be compromised unless the unfiltered water is free of microbes. Further, while parasites and most bacteria are removed, one study suggests that viruses are able to pass through the pores of the ceramic filter version used in this study [15].

Conclusions
The results for the biosand filter illustrate the importance of achieving an optimal balance between the water dosing volume, pause time and the quantity of safe water required by households to meet daily needs. More research is required to determine the ideal combination of water dosing volume and pause time in order to maximize the filter performance without requiring users to adhere to a complex water dosing schedule. In the short term, filter implementers might consider advising households to either filter just the water volume held in the influent reservoir (approximately 5 l) or separately collect the initial 5 l of filtrate for consumption and use the remaining filtered water for other household needs. Alternatively, households could use a combination of HWTS technologies such as combining filtration with disinfection, for example with chlorination or solar disinfection in polyethylene terephthalate plastic bottles. Finally, the user instructions and training manual for the ceramic filter currently fail to address how overflowing the unit may contaminate the filtrate. Both of these documents should inform users to cease filling the filter 3-5 cm below its lip as well as to avoid disturbing the unit during filtration to prevent unintentional overflow. Additionally, small changes to the design of the ceramic filter such as an extension of the filter lip would reduce the chance of overflow contamination without compromising the unit's durability or dramatically increasing its cost.
This research establishes how operating conditions that are affected by user behaviour can alter the bacterial removal efficacy of two HWTS systems, thereby reducing their community-based effectiveness and increasing environmental risk exposure. Studies conducted in developing countries suggest that increased microbial contamination in water poses increased health risks [31]. Using scenario 3 (10 l water dosing volume every 12 h) as a reference, our results show that operating parameters such as pause time, water dosing volume and sample collection point for the biosand filter can double the level of bacterial contamination remaining in the filtered water, and thereby increase the risk of water-borne disease. Therefore, evaluation of an HWTS system should be accompanied by routine monitoring of operating conditions in the field and testing the robustness of technology performance to these conditions, to better inform the choice and delivery of environmental health technologies for specific environmental and behavioural settings.