Primary question

Are interventions to reduce the impact of arsenic contamination of groundwater on human health in developing countries effective?

Secondary question

What factors enable or constrain the effectiveness of these interventions in developing countries?

Searches

Search terms and languages

The search syntax was formulated to use arsenic (the term was truncated arsen* to capture arsenates or arsenates or other species of arsenic prevalent in groundwater [7] and water, or groundwater. This syntax was split into search clusters which were joined with the Boolean connector AND to give items which made reference in the title and/or abstract and/or keyword to (arsen*) AND (water or groundwater). The search was conducted in the English language. Please see Additional file 1 for further details.

In resources with no controlled syntax, or where the specificity of the search was lost given the broad remit of some of the search resources (i.e. Web of Science), we used a population filter to increase the specificity of search returns to developing countries. This filter named not only countries but also populations, and included a cross-check of developing countries, or third world terminology (i.e. under-developed), to maintain sensitivity. The list of named countries was developed from the funder’s list of includable countries (http://www.ausaid.gov.au/ngos/Pages/devel_list.aspx) – which was, in turn, cross-checked with the International Human Development Indicators (http://hdr.undp.org/en/statistics/). The IMF list of developing countries was also consulted. This filter was extensively tested in scoping against known includable studies.

All searching was limited as follows:

Databases

The following databases were searched:

Searches for grey literature: contacts, internet searches, use of specific search terms or strings, filtering or limitations

The following web-resources, web-sites and organisations of note were searched, alongside conference abstracts (as above), library catalogues and Google, for grey, or difficult to locate, literature [8, 9].

Citation chasing

Citation chasing was conducted on items included after full-text screening. Backwards citation chasing was conducted manually through reference scanning of bibliographies. Forwards citation chasing was conducted in Web of Knowledge (ISI Thompson Reuters).

The results from the citation chase were aggregated and de-duplicated against each other and the main review libraries (which contains the items from the main search and which have already been screened). This gave an additional 143 references to be screened. The results from the forwards citation chasing can be found in Additional file 1.

Study inclusion criteria

After the searches had been completed relevant studies were identified through two stages of screening. All titles and abstracts returned by the search strategy were screened by one of three reviewers (with a random sample of 10% being checked by a second reviewer). Where it was not clear from the title and abstract alone if the study met the inclusion criteria, the full-text of the study was retrieved to enable the second stage of screening also conducted by one of three reviewers (with a random sample of 25% being checked by a second reviewer). All disagreements about study inclusion and exclusion were resolved by discussion, and where necessary the involvement of a third reviewer. A kappa analysis was not conducted.

In order to be included, the study had to fulfil the following criteria:

Population (or ‘subject’)

The outcomes of interest involve both groundwater and people, therefore the subjects are either people living in, or groundwater sources in, developing countries (as defined by AusAID; see Additional file 2).

Intervention

A study of any technology (see Table 1) designed to remove or lower arsenic concentration in water intended for human consumption and tested in the community/household (field setting). This does NOT include national or local policy or specific educational interventions.

Comparators

Studies had to include a form of comparison e.g. pre/post comparison of filtered and unfiltered water, comparison with another technology intended to remove or lower arsenic concentration in groundwater, or comparison with no intervention.

Outcomes

Since this review was concerned with the viability of interventions during ‘real-life’ scenarios, a distinction was made to include only field tests which involved members of the community, as opposed to being operated solely by researchers.

Study design

Only comparative study designs were included (including, but not limited to, randomised controlled trials, before and after studies, and cross-sectional studies).

Exclusion criteria

Any studies which looked at the following topics were excluded from this review:

Interventions such as dugwells, pond sand filters, rain water harvesting mechanisms, well-switching and GIS modelling were also excluded from the review since they do not reduce or remove arsenic from groundwater.

Study quality assessment

The studies included at full text were assessed using the McMaster University Effective Public Health Practice Project quality appraisal tool (http://www.ephpp.ca/tools.html; see Additional file 3). This tool is suitable for use across a wide range of quantitative study designs and has explicit criteria for rating selection bias, study design, confounders, blinding, data collection methods, withdrawal, intervention integrity and analyses as ‘strong’, ‘moderate’, or ‘weak’. Study quality appraisal was conducted by one of three reviewers, with a random sample of 25% checked by a second reviewer. Disagreements were resolved by discussion, with the involvement of a third reviewer if necessary.

Data extraction strategy

Data was extracted from included studies by one of three reviewers using a modified version of a data extraction form designed for use with complex public health studies [10]. Extracted data followed the outcomes described above and included details of the population, type of intervention, and context in which implemented – see Additional file 4. A random sample of 10% of the completed data extraction forms were checked by a second reviewer. Discrepancies were resolved by discussion, with the involvement of a third reviewer if necessary.

Data synthesis and presentation

All measures of arsenic concentration in groundwater are reported as mg/L in this review. All measures of arsenic concentration in human tissue are reported as μg/g creatinine and μg/L urine.

Primary outcomes

The primary outcome is the percentage of effluent water samples, i.e., water samples treated by the intervention, with arsenic concentrations below WHO guidelines (≤0.01 mg/L) [4]. The WHO guidelines are based on ‘safe/acceptable’ levels of exposure for humans, therefore those interventions meeting WHO guidelines are providing safe water for populations. Where data were available, exact 95% confidence intervals (CIs) for the percentage of samples meeting WHO guidelines were calculated using the binomial distribution [11]. The 95% CIs provide a measure of the uncertainty associated with the proportion of samples meeting WHO guidelines. This reflects the number of samples measured in the individual studies.

The results are presented by intervention, with interventions categorised into groupings depending on the primary processes involved.

Secondary outcomes

There are three secondary outcomes: (a) the proportion (and 95% CIs) of effluent water samples with arsenic concentrations below the guideline limit for the country in which the study was conducted (e.g. < 0.05 mg/L in Bangladesh), (b) arsenic concentrations measured in human tissue, and (c) aspects related to acceptance of the interventions.

The primary focus of this review was to establish the effectiveness of interventions to remove arsenic from groundwater. However, the qualitative information also identified in the articles was considered important in providing a context for the effectiveness results and was therefore compiled as a summary.

Overall summary of effectiveness

Arsenic concentrations measured in human tissue are also presented, where such outcomes are measured and reported in the individual studies.

Variations of effectiveness

Thus, where reported in the individual studies, we have presented and discussed evidence on variation in arsenic removal effectiveness due to these factors. Details on maintenance and flow rate (in terms of L/hr), whether from manufacturer guidance or from evaluation are presented. Costs associated with the technology are reported by the currency and year of the individual study. However, to allow comparison of costs across technologies, all costs have also been converted to Aus$ for the year 2011 using Shemilt (2010) [12].

Synthesis

Formal synthesis of quantitative outcomes across studies, i.e. meta-analysis, was not undertaken due to poor study quality, poor reporting quality and issues of heterogeneity between studies.

The main inhibitor of meta-analysis for this review was heterogeneity in the design and conduct of studies, such as differences in the intervention being assessed and the unit of analysis (samples from different sites, repeat samples from just one site, repeat samples from a number of sites, etc.). The poor quality of study conduct also influenced our decision, as potential biases may be masked in a meta-analysis. In terms of reporting quality, the outcome data required for synthesis could not be extracted for some studies, therefore this information would be lacking from the synthesis.

Overall, conducting a meta-analysis across these studies would have produced results which would have been difficult, if not impossible, to interpret. As many informative texts state, meta-analysis of study results, when studies are diverse or of poor quality, may be meaningless, with any real differences concealed [13–16].

Included studies

The databases were searched from 1980 to August 2011 and forward citation chasing was completed by November 2011. The searches identified 11 457 potentially relevant articles (after de-duplication). Screening by title and abstract revealed a large proportion of the studies were laboratory based and therefore excluded. Three hundred and twenty one papers were requested for examination of full text, of which 44 were included in the review. A further seven articles were identified for inclusion via web searching and forward chasing of bibliographies (Figure 1).

Included study characteristics

Additional file 5 provides a summary of each study detailing the country in which the study was conducted, the size and design of the study, the interventions and outcomes assessed, and an overview of study quality. Details on interventions such as household/community, costs and flow rates are in Tables 3, 4, 5, 6, 7 and 8. Further description of study quality is given in 'Included study characteristics' and Additional file 6. Fifty of the fifty-one included studies were appraised as ‘weak’ using the EPHPP quality appraisal tool, with only one study (a RCT; Milton et al. (2007)) appraised as ‘strong’ [17]. Whilst we recognise the difficulties for decision-makers of using a synthesis of methodologically-weaker studies, we did not exclude studies on the basis of quality as:

○ allows the direct comparison of water samples before and after the intervention.

A key for the ‘Outcomes’ and ‘Quality’ headings in Additional file 5 is provided below. A key for the study design elements of Additional file 5 can be found in Additional file 6, but in particular, under the ‘Study design’ heading in Additional file 5, a straight single line indicates that a survey of one group of participants was undertaken (i.e. as in Akman & Higano, (2002)) [18]. Two parallel straight lines indicates that a survey of two different groups of participants was undertaken (i.e. Ahmad et al. (2005) and Ahmad et al. (2006)) [19, 20]. A before/after study is indicated by the elongated S-shaped line as in Ashraf Ali et al. (2001) [21]. A randomised control trial design is denoted by a straight line that splits into two as for Milton et al. (2007) in Additional file 5 [17]. The width of the lines indicate the relative number of samples collected and analysed in each study, along with the blocks under the ‘No. samples’ heading.

Note that in Figure 2 (the key for Additional file 5) there is no marker for moderate overall quality, since all studies were considered to be of weak or strong quality. The validity and reliability items in Additional file 5 correspond to appraisal of data collection methods from the EPHPP quality appraisal tool (Additional file 7).

• For validity: “Were data collection tools shown to be valid?”

• For reliability: “Were data collection tools shown to be reliable?”

These two items are shown in Additional file 5 as they were deemed to be important for appraising study quality in the context of interventions to remove or reduce arsenic in groundwater.

The majority of studies (n = 34, 67%) are based in Bangladesh, with the remaining set in India (n = 10, 20%), Nepal (n = 3, 6%), Vietnam (n = 2, 4%), Cambodia (n = 1, 2%) and China (n = 1, 2%). Forty-four studies (86%) report an evaluation of the effectiveness of interventions (as indicated in column 3 of Additional file 5, under the “Outcome/Effectiveness” heading). Across these effectiveness studies there is great variation in the number of water samples measured, ranging from two samples reported in Oh et al. (2000) to 1074 samples reported in Ngai et al. (2006) [22, 23]. In addition, there are differences in the sample units across studies. For example, some studies report repeated samples over a period of days at one site [24], while other studies report on repeated samples at a number of wells at a number of sites [25]. For eight studies it is unclear whether a reported value is a single measurement or an average of an unknown number of measurements.

Six studies (12%) report on the effectiveness of interventions and provide detail on behavioural aspects related to the use of these interventions. A further seven studies (14%) provide evidence on behavioural aspects only. Studies assessing behavioural aspects generally have a large number of participants, e.g. Ahmad et al. (2006) reports on a study of 2700 households and Jakariya et al. (2007) who surveyed 2600 families [20, 26].

Of the 44 studies reporting on the effectiveness of interventions, all but one of these are before and after studies where influent and effluent water samples are analysed and compared for their arsenic content. The remaining study is a randomised control trial, where households are randomised to receive the intervention or to receive no intervention.

The most common design for the studies reporting on behavioural aspects of arsenic removal/reduction interventions is that of a survey where users of different interventions are surveyed at a particular time-point.

Across the 51 included studies, 50 different interventions are described. The wide range of technologies appears to be due to attempts to meet the varying needs of villagers in terms of availability of resources, cost, maintenance, and ease of use. In order to appraise the effectiveness of the interventions, they were classified into eight groups based on the primary process involved in the intervention as described by Ravenscroft et al. (2009) [1]:

A summary of all interventions evaluated by each study can be found in Additional file 5 under the ‘Technologies’ heading. A breakdown of the number of studies and interventions included in the review is given in Figure 3, with interventions classified into the eight groupings. There are more interventions in the adsorption grouping because a number of studies have assessed a greater number of adsorption interventions than other types, in particular Amiri et al. (2010), BCSIR (2003) and BRAC (2000) (see Additional file 5) [27–29]. On the other hand, for zero valent iron technologies many studies have assessed the same type of intervention: the sono, three-pitcher/kolshi/gagri filter.

The final column of Additional file 5 details the funding source for each study. For 36 of the studies (71%), the funding source could not be determined from the article and a further study reported no funding for their research.

Study quality assessment

The overall quality of the studies was poor, with all but one ranked as weak (Additional file 5). The one RCT was appraised as strong [17]. Fifteen studies were at least somewhat likely to have a representative population but in almost as many studies it was not possible to tell from the information available. Eighteen out of 23 relevant studies did not have enough information to be able to say what percentage of the sample population agreed to participate and most studies did not detail how the sample size was determined. Of the four that did, they were convenience samples.

Overall, the general issues with quality included:

Synthesis of evidence

The effectiveness evidence for each of the intervention groupings is presented in this section. The primary outcome is the percentage of effluent water samples with arsenic concentrations below WHO guidelines (≤0.01 mg/L), based on ‘safe/acceptable’ levels of exposure for humans. There are three secondary outcomes: (a) the proportion of effluent water samples with arsenic concentrations below the guideline limit for the country in which the study was conducted (e.g. < 0.05 mg/L in Bangladesh), (b) arsenic concentrations measured in human tissue, and (c) behavioural aspects related to the use and acceptance of the interventions.

Figure 4 shows the number of studies reporting for which a quantitative effectiveness outcome is available within each intervention group. This figure also shows the number of unique interventions from these studies where effectiveness evidence is available. For instance, within the ‘oxidation and filtration’ group 16 studies provide effectiveness evidence for 12 different interventions, while for the ‘zero valent iron’ group, 13 studies provide effectiveness data for just two interventions.

Given that some studies evaluate a number of interventions and some interventions are evaluated across a number of studies, there are 90 potential study-by-intervention outcomes available to inform the effectiveness of arsenic removal technologies for groundwater. For 77 of these outcomes, data on the percentage of samples meeting WHO and/or national arsenic guidelines could be extracted from the study articles. For 42 of these study-by-intervention outcomes, data on the percentage of samples meeting both WHO and national guidelines for arsenic concentrations are available, while for nine and 26 of the study-by-intervention outcomes data are only available to inform the percentage of samples meeting WHO guidelines or national guidelines, respectively. These numbers are broken down by the intervention groupings in Figure 5.

For 13 study-by-intervention outcomes (= 90–77), data on the percentage of samples meeting WHO or national arsenic guidelines could not be extracted. This was mainly due to a lack of detail reported in the results section of the article. A list of the reasons why these data are not available for these 13 study-by-intervention outcomes is given in Additional file 8.

The available effectiveness data are presented in the following sections, by intervention grouping, for the primary outcome: the percentage of samples meeting WHO guidelines for As concentrations. First we provide detail on the interpretation of the figures in the following section.

Interpretation of figures

The solid dots in Figure 6 represent the percentage of samples meeting WHO guidelines for As concentrations for each intervention evaluated. The vertical lines above and below the solid dots represent the 95% CI for the percentage of samples meeting WHO guidelines, as defined to be excellent evidence of effectiveness (see Table 2). The solid horizontal line represents the cut-off where 95% of samples meet WHO guidelines. Thus, for the sand filter, 40% of samples meet the WHO guideline (solid dot), with a 95% CI ranging from 25% to 56%. The diamonds represent the approximate arsenic concentration in the influent water samples for each intervention (see right-hand-side y-axis of Figure 6). The dashed horizontal line represents the WHO guideline As concentration of 0.01 mg/L. The approximate influent As concentration for the sand filter is 0.2 mg/L. Note that where data are available, arsenic concentrations in influent water samples, except those for iron removal ceramic filters 1 and 2 (IRCF1 and IRCF2), are all well above the WHO guideline level (i.e. the dashed horizontal line). The arsenic concentrations in the influent water samples presented in Figures 6, 7, 8, 9, 10, 11 and 12 should be considered approximate as the reporting of this information is not consistent across, or even within, studies. For example, the approximate influent As concentration may reflect the median, mean or mid-point from a range of As concentrations reported in a study. For clarity, where more than one author has investigated the same technology, the abbreviation is suffixed with the author’s initials. The aim is to present the reader with an understanding of the magnitude of the influent As concentration, rather than report the exact As concentration. The interpretation of Figure 6, applies equally to Figures 7, 8, 9, 10, 11 and 12.

Primary outcome: Samples meeting WHO guidelines for As concentrations

Arsenic removal in situ, membrane and electrolytic technologies

As displayed in Figure 13, evidence from the two studies providing data on subterranean arsenic removal (SAR) ((1)SAR(Gupta)) and (1)SAR(V.Halem)) is inconsistent, with the SAR reported in Gupta having 72% of samples meeting WHO guidelines, while the percentage of samples meeting WHO guidelines in the Van Halem et al. (2010) study is just 6% [51].

Evidence for the effectiveness of nanofiltration is poor, as it is only based on two samples at two sites, hence the great width of the confidence intervals. However, the evidence from the only study evaluating electrocoagulation as a technology suggest this performs reasonably well with 82% of samples meeting WHO guidelines, but it does not reach our criteria for excellent evidence of effectiveness [53].

Secondary outcome: Percentage of samples meeting national guidelines for As concentrations

As Figure 5 shows, much more data are available from the included studies to calculate the percentage of samples meeting national guidelines than for WHO guidelines. This is perhaps due to the national guideline being seen as more achievable than the WHO guideline for arsenic concentrations in groundwater, but also field testing kits may have difficulty measuring at the WHO guidelines for arsenic concentrations in groundwater [1]. For all countries in which these studies were conducted (Bangladesh, India, Nepal, Cambodia, Vietnam and China), the national guideline for arsenic concentrations in groundwater is0.05 mg/L. This therefore helps comparison across studies and interventions for this outcome.

It should be noted that although there is evidence that clinical symptoms of arsenicosis may occur at < 0.05 mg/L [54], it has been argued that attempting to adopt a lower standard in a developing country, when the existing standard has not yet been achieved, may lead to scarce resources being misdirected [1].

Oxidation and filtration

The evidence on these technologies to reduce or remove arsenic to ≤ 0.5 mg/L is mixed, with most interventions having >60% of samples meeting national guidelines (see Additional file 9, Figures A and B). In particular, three interventions have 100% of samples meeting national guidelines: the Space ARP evaluated by Brennan (2011) [37], the garnet filter evaluated by Rahman et al. (2005) [38] and the IOB evaluated by Hassan et al. (2008) [40] in location 1. In contrast, the ardasha filter and passive sedimentation technologies have particularly poor performance with <10% of samples meeting national guidelines.

Coagulation, co-precipitation and filtration

Of the three different coagulation, co-precipitation and filtration technologies providing data for this outcome, only the Stevens Institute Technology, as implemented by BAMWSP, Cheng et al. (2004) and Meng et al. (2001), have any outcomes with ≥95% (see Additional file 9, Figures C and D) [25, 43, 44]. The P&G powder evaluated by Norton et al. (2009) has 86% of samples meeting national guidelines [33]. The DPHE/Danida technologies do not perform well across sites and studies, ranging from 3% to 80% of samples meeting national guidelines.

Adsorption

All but one of the adsorption interventions having available data have >95% of samples meeting national guidelines, therefore defined in this report as good evidence of effectiveness (see Additional file 9, Figures E and F). The intervention with < 95% of samples meeting national guidelines is the ARP as investigated by Hossain et al. (2006) [48].

Zero valent

The zero valent iron technologies are considered to have good evidence of effectiveness with 12 out of the 14 providing available data, having ≥95% of samples meeting national guidelines (see Additional file 9, Figure G). The sono intervention assessed by Milton et al. (2007) [17], the only RCT, has the lowest percentage of all zero valent iron technologies with 65% of samples meeting national guidelines.

Ion exchange

There is mixed evidence for the effectiveness of ion exchange technologies in terms of the percentage of samples meeting national guidelines for arsenic concentrations in groundwater (see Additional file 9, Figure H). The effectiveness of tetrahedron technologies is inconsistent across sites and within studies, ranging from 57% to 100% for different sites within the same study [25]. The ArsenX technologies have poor evidence of effectiveness [50].

Arsenic removal in situ, membrane and electrolytic technologies

Due to the low number of samples assessed in Oh et al. (2000) where membrane technology was evaluated [22], it is difficult to comment on the effectiveness of this technology (see Additional file 9, Figure I). Data on the subterranean arsenic removal in situ interventions is inconsistent ranging from 34% [51] to 85% [52]. However, the electrocoagulant technology evaluated by Wan et al. (2011) has 100% of samples meeting national guidelines (see Additional file 9, Figure I) [53].

Groundwater effectiveness summary

In summary, the effectiveness of the oxidation and filtration interventions is poor, while the evidence for coagulation, co-precipitation and filtration, subterranean and membrane and electrolytic methods is mixed with results providing either poor or good evidence for effectiveness. The evidence on adsorption and zero valent iron interventions is more promising with most results suggesting good effectiveness evidence. In particular, the activated alumina and sono three-kolshi/gagri/pitcher filters have ≥95% of samples meeting national guidelines. Disappointingly, only one study reports excellent evidence of effectiveness: BRAC (2000) for activated alumina [27].

Synthesis results – human tissue samples

Only three of the included studies report on measures of arsenic concentrations in human tissue or body fluids: Milton et al. (2007), Norton et al. (2009) and Hossain et al. (2005) [17, 33, 45]. Milton et al. (2007) report from their RCT urinary arsenic concentrations from individuals before and after (at one and 12 months post-intervention) use of the three-pitcher filter compared to individuals receiving no intervention [17]. They found that for controls, post-intervention urinary arsenic concentrations were statistically significantly greater than the pre-intervention concentrations. Regardless of whether the analysis was restricted to those individuals who self-reported compliance, the post-intervention urinary arsenic concentrations for those receiving the three-pitcher filter were not statistically significantly different to the pre-intervention concentrations for one and 12 month total urinary inorganic arsenic metabolites, and one month urinary inorganic arsenic. At 12 months, measures of urinary inorganic arsenic were statistically significantly higher than pre-intervention levels for the three-pitcher filter (mean 18.3 vs 24.4 μg/g creatinine). Milton et al. (2007) provide three possible explanations for why post-intervention urinary arsenic concentrations were not found to be as low as expected: a) incomplete compliance, b) individuals receiving sources of arsenic through routes other than household consumption, and c) the intervention not being as effective as expected [17].

Hossain et al. (2005) report on the effectiveness of ARPs based on adsorption, ion exchange or oxidation and filtration primary processes [45]. As a secondary analysis to arsenic concentrations in water, they analysed the urine samples of individuals. Hossain et al. (2005) briefly describe their results stating that 82% of the 150 individuals had urinary arsenic concentrations above the normal limit, two years after the ARPs had first been installed, with a mean of 104 μg/1.5 L (SD 116 μg/1.5 L). This was significantly higher than the urinary arsenic concentrations measured in 78 individuals in unaffected areas. Hossain et al. (2005) discuss technical reasons for the lack of effectiveness of the ARPs: maintenance, clogging and lack of user friendliness [45].

Norton et al. (2009) measured total urine arsenic concentration before and after the initiation of the Proctor and Gamble (flocculant-disinfectant) powder in 103 individuals [33]. They report a 42% reduction over 12 weeks in total urinary arsenic levels from a median of 385 μg/g to 225 μg/g. But they note that for 24 of 103 participants, total urinary arsenic levels were actually higher after the intervention was initiated than the arsenic levels before.

Norton et al. (2009) provide three possible explanations for what they consider a modest reduction in total urinary arsenic levels: a) the participants consumed water outside of the home where interventions to reduce arsenic were not in place, b) the consumption of food containing high levels of arsenic due, for example, to irrigation, and c) that the body stores arsenic and will release these stores when arsenic consumption decreases [33].

Reasons for variation in effectiveness

The effectiveness of each technology needs to be understood in context, for example, perception of the arsenic problem by the users, cost, flow rate and maintenance. Therefore, the supporting information from the included studies, has been summarised where available. Please note, it was not within the remit of this review to synthesise or interpret the extracted qualitative data.

Perception of the arsenic problem

The perception of people using the arsenic mitigation technologies has been shown to be integral to successful implementation of technologies, for example, it took 25 to 30 years to convert up to 97% of the rural population of Bangladesh to using tubewell water, as opposed to contaminated surface water, and even then it was easier than today’s alternative safe water options in terms of acceptance, technical viability, financial and maintenance aspects [55]. Furthermore, there have been instances of confusion due to previous experiences which have delayed uptake of arsenic removal technologies; for example, people using passive sedimentation as originally promoted by some organisations were approached to change their practise again despite having already made an effort to treat arsenic and change their behaviour [56].

General awareness of the arsenic problem is varied. Often, older people did not consider arsenic to be such a problem, particularly if they had been drinking water from the present sources for 25–30 years without noticing any issues. There was a general perception that arsenic was less likely to be considered a problem among illiterate communities [55]. In contrast, Jakariya (2003) also found that in a village with better socioeconomic conditions and higher levels of literacy, people were not willing to accept arsenic may be a problem in the long run [55]. It should be noted they had no individuals displaying the symptoms of arsenicosis. They were aware of the long incubation period of the disease and believed that their better nutritional status helped prevent it; this made the villagers reluctant to use arsenic free water.

Culture can be an issue when attempting to introduce new technologies. There is a ‘traditional’ practice in some areas that men never collect drinking water. Due to religious perceptions and cultural traditions, rural women avoid any sort of contact, including visual contact, with unknown male persons and non-relatives. This discourages men from fetching water from places where women usually do. If men do collect water they are often taunted for doing women’s work [55]. Hence, having to walk further to bring safe water can mean extra work for women. Lack of privacy has also been identified as a barrier to collecting water from the improved water options and some women considered having to go out of their house for water a prestige issue [57].

It would seem that whether a new technology was household- or community-based influenced its success, yet, the issue is complex. Hoque et al. (2004) found that people discontinued the use of all household-based water options after several weeks [57]. One third of these users switched to piped or manual deep tubewell water as it became available. People shared the costs for these options and claimed that walking to a safe, reliable and easily manageable water source once a day was more convenient than operating and maintaining household-based options [57].

An earlier study by Hoque et al., (2000) investigated various methods of implementation for the arsenic removal technologies were employed [56]. Proper education, training and regular sharing of information with the stakeholders were identified as the main driving factors behind the spontaneous participation of villagers. However, lack of information, education and communication materials for the promoting/motivating of stakeholders as well as for people was found to hamper the results.

Overall, understanding how new technologies will be received is highly contextual, not only for the users, but also with regard to the agencies installing them. Hossain et al. (2006) found indiscriminate installation of ARPs without proper preparatory steps, coupled with ill-maintenance, gross mismanagement and a sense of disowning on the parts of the users have resulted in misuse of these costly plants and the treated water [48]. Ngai et al. (2006) also found the Kanchan Arsenic Filter as an open source technology in some cases was not implemented with fidelity and coupled with lack of government approval slowed dissemination and acceptability of the project [23]. This was successfully overcome by training local entrepreneurs to solve problems. Similarly, Jakariya et al. (2007) found that options still in operation were those having highly motivated caretakers using personal initiatives to run the options [26].

Costs and flow rate

In developing countries cost will clearly be an issue. Many investigators were mindful of this; ensuring interventions used local materials and resources.

Interestingly, Hossain et al. (2006) found that technologies which were free were not seen as an asset by the community and therefore not maintained [48]. It seemed because the villagers had not invested, they felt no sense of ownership. A similar issue was found by BRAC (2000) [27]. None of the village water committees set up by them were taking responsibility for the free demonstration units provided, which subsequently fell into disrepair. BRAC (2000) considered this was either because of minor faults that cost little money to fix or routine maintenance of the options that would need physical labour to replace or regenerate. BRAC (2000) considered it may be that because the villagers did not have to pay, they lost interest, therefore they intend to ensure some community contribution to future projects to create a sense of ownership.

Other studies contradict the hypothesis of ownership. For example, Hoque et al., (2000) found community-based options were used by many families who initially showed no interest in the options at the planning phase and so did not share the costs [56]. A survey by Jakariya (2003) found 54% of respondents wanted free options from the government or some kind of government help to alleviate the problem, with 64% saying that they did not want to pay for any type of alternative safe water options [55].

The ability of the technology to produce sufficient water is also key to the success of a technology and should be balanced against cost. As a guide, approximately 50 L/day is considered an acceptable volume for a family’s needs [58]. The following tables summarise the cost and flow rate of technologies, where that information is given. Costs are given as Australian dollars for 2011. Where possible, flow rates have been converted to L/hr.

Household oxidation and filtration technologies listed in Table 3 display flow rates from 1–60 L/hr, with the majority of technologies achieving 1–5 L/hr. The sand filter, however, reached 60 L/hr. Community interventions achieved 140-200 L/hr. The costs within this group, where given, appear relatively low, presumably due to the use of local materials, such as sand.

With regard to the coagulation, co-precipitation and filtration group, flow rate is not necessarily a practical measure since some interventions involve only adding a powder, followed by stirring and settling. However, an estimate can be made of the production rate of arsenic free water (see Table 4) e.g. Star/CIWPL, 16–20 L after mixing and standing from 15 min to an hour. Otherwise the range for household technologies was 4.3-120 L/hr. The one community option produced 1000 L/hr.

Comparisons with regard to cost are difficult due to the lack of information, however, there are likely to be on-going costs due to the regular need for chemicals.

The range of flow rates for household adsorption technologies is broad (Table 5), from <1 to 120 L/hr. However, the community options appear capable of producing flow rates of 42 to 1000 L/hr. Unfortunately, information on operational costs was sparse. Capital costs for communities were generally high, in comparison to other technologies.

The zero valent iron technologies generally employ local materials, such as nails, brick chips and sand, therefore they are a comparatively cheap technology. However, as shown in Table 6 flow rates appear to be low, at around 2 L/hr for a household.

Table 7 shows ion exchange technology to be comparatively expensive and generally a community, as opposed to a household, option. Again, there is little information on operational costs, although start-up costs range from Aus$ 1300 to 5672.

The subterranean method of arsenic removal in Table 8 shows impressive flow rates of 167–1200 L/hr. Electrocoagulation and nanofiltration, which are more likely to be household options display respectable flow rates of 50 L/hr and 17 L/hr, respectively. Again, information on cost is minimal.

Maintenance

The level of maintenance required appears to be a key factor in the success or failure of arsenic removal technologies in developing countries. In some cases, it was not only the user who was unclear about the need for maintenance, for example, Hossain et al. (2006) found that authorities responsible for installation and upkeep of ARPS had no clear idea regarding frequency of backwashing [48].

It should also be noted, some of the technologies generate waste, which may or may not be contaminated with arsenic. Often, since there were no specific regulations in regard to the disposal of arsenic-laden spent regenerant on the Indian subcontinent, this waste is disposed of in the open field [46, 48].

Review limitations

This review is limited primarily by the reporting and conduct of research in the included papers. All but one of the studies was assessed as weak due to poor reporting standards. Our primary outcome (percentage of samples meeting WHO guidelines) could not be determined from many of the study reports. It is unclear if this introduces outcome reporting bias into the review. For example, we have been unable to determine whether outcomes were not reported because none of the samples met the WHO guideline, or simply because older testing kits were not capable of measuring at this level.

Reviewers’ conclusions

Establishing the effectiveness of interventions to reduce the impact of arsenic contamination of groundwater on human health in developing countries is complex. Using systematic review methods, we were able to summarise evidence for the effectiveness of interventions, with the caveat that there were significant methodological weaknesses in much of the underlying research:

○ oxidation and filtration

○ ion exchange

○ subterranean (As removal in-situ)

○ membrane technologies

○ electrolytic methods

○ adsorption (with ≥95% of activated alumina samples meeting national guidelines)

○ zero valent iron (with ≥95% of sono three-pitcher filters samples meeting national guidelines)

We did not locate any evaluations of the effectiveness of lime softening or phytofiltration technologies.

Through extracting data about the implementation of interventions, we were also able to summarise some of the complexity of the contexts into which these interventions were introduced. These contexts may begin to explain why the effectiveness of interventions to reduce the impact of arsenic contamination of groundwater can vary greatly, although our findings here did not enable us to rank these issues in order of importance. Consideration of the following issues in decision-making about funding of interventions may be pivotal in whether or not they are effective and have the desired positive impact on human health in developing countries:

We identified significant shortcomings in the rigour of the studies eligible for inclusion in this review. To improve the evidence-base for decision-making about the effectiveness of interventions about arsenic mitigation technologies at household- and community-level will require commissioning of primary research that:

Research methods in the field of international development are developing in order to address the complexity of the situation and produce a more informative evidence base for decision makers. As highlighted by the challenges in this review, commissioners should pay close attention to these methods in both primary and secondary research.

The importance of understanding ‘how, when and why’ interventions are effective, as opposed to simply ‘are’ they effective, has been highlighted with the perspective of international development by Svoronos and Mate (2011) [59]. Context-specific evaluation designs such as randomised controlled trials with greater protocol flexibility, process evaluations, realist evaluation, and multiple case study designs may provide more nuanced understanding of the effectiveness of complex interventions [59]. It follows from this that systematic reviews of research in international development may also need to accommodate similar context-specificity and apply the principles of conventional systematic reviews rather than the methods in their entirety [60]. Context-specific systematic review methods developed in the health and social policy fields of high-income countries, which integrate quantitative and qualitative data [61–63], may prove fruitful methodological avenues to pursue in synthesising complex evidence for international development.