Study design

Our systematic review is reported according to PRISMA 2020 guidelines [12] (S1 Checklist), and was pre-registered with the International Prospective Register of Systematic Reviews (PROSPERO)—CRD42021288727. We followed best practice for systematic reviews of economic evaluation evidence [13].

Search strategy

We searched Medline, Embase, Global Health, EconLit, Web of Science, Cochrane library, International Bibliography of the Social Sciences, Global Health Cost-Effectiveness Analysis Registry, and the NHS Economic Evaluation Database, for literature published from January 1980 to September 3, 2025. We also searched for grey literature studies using databases from National Bureau of Economic Research, International Initiative for Impact Evaluation, Research Papers in Economics, WHO Index Medicus (all regions), and Copenhagen Consensus Centre.

Our search strategy (S1 Text) combines terms for economic evaluation and terms for promotion of handwashing with soap used in recent reviews [3,4]. We used Endnote (Clarivate, Philadelphia, USA) for de-duplication, Rayyan for managing blinded abstract screening [14], and Microsoft Excel for data extraction. We also screened reference lists of included studies and previous reviews. Two reviewers (IR with DB or JW) independently screened titles, abstracts, and full texts. Differences between reviewers about inclusion were reconciled by discussion, with recourse to a third reviewer (OC), if necessary.

Selection criteria

Eligible settings were domestic (households), educational, or childcare, in any population worldwide. We excluded other settings, notably healthcare facilities, because they have fundamentally different disease transmission risk and target audience of interventions (usually trained staff rather than the public).

Eligible interventions promoted handwashing with soap through communication and/or provision of facilities or products. Examples of communication include mass media campaigns, door-to-door visits, or group activities. Examples of provision include distribution or marketing of handwashing stations (fixed or mobile) or soap. Accordingly, we excluded studies without promotion, for example, studies which assumed that populations spontaneously changed behaviour without external intervention. Following [4], we included interventions combining handwashing with other interventions (e.g., water treatment, face masks) if handwashing was a “major” behavioural target of the intervention (S1 Table).

We only included interventions promoting handwashing with soap and excluded those promoting alcohol-based hand rub (ABHR), antimicrobial towels, or other soap alternatives. Reasons for this decision were, first, that ABHR has historically been unavailable (or very expensive relative to soap) in many low- and middle-income country (L&MIC) settings [15]. Second, sustainable development goal indicator 6.2.1 focuses only on handwashing facilities with soap and water [16]. Third, a focus on soap reflects the effectiveness evidence. A recent Cochrane review of physical interventions to reduce the spread of respiratory viruses identified 16 randomised trials in domestic, school, or childcare settings with the outcome of acute respiratory infection or influenza-like-illness [17]. None was an of an intervention promoting ABHR in a domestic setting in a L&MIC.

Eligible study designs were full economic evaluations, i.e., comparisons of two or more options (one of which can be existing practice or doing nothing) in terms of their relative costs and outcomes [18]. Specifically, we included benefit-cost analyses, which are studies that value health and other outcomes in monetary units (e.g., US$). We also included cost-effectiveness analyses, which are studies that value outcomes in natural units (e.g., death averted) or utility-weighted units (e.g., disability-adjusted life-year [DALY] averted). We excluded partial economic evaluations, such as those reporting cost analysis only (e.g., cost per person targeted/reached). We also excluded prospective economic appraisals, e.g., project planning documentation.

Eligible outcomes were summary measures of efficiency, such as a ratio (e.g., incremental cost-effectiveness ratio [ICER] or benefit-cost ratio [BCR]), a probability (e.g., that an intervention is cost-effective given a threshold), a difference (e.g., incremental net benefit), or net present value [19]. We also included studies that did not report such a measure but demonstrated dominance, e.g., an intervention being both less costly and more effective than the comparator.

Data extraction

We extracted and reported data related to interventions, methods, and results, with multiple data points for studies reporting multiple interventions meeting inclusion criteria. Two reviewers (IR and DB) independently extracted data and assessed quality using a structured Excel spreadsheet. Differences between reviewers about extraction were reconciled by discussion, with recourse to a third reviewer (OC), if necessary.

Study quality

We assessed quality of reporting using the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) [20], which is widely used in systematic reviews of economic evaluations. We followed a common scoring approach [21,22] by awarding studies 1 point for each CHEERS item “fully met”, 0.5 for “partially met”, and 0 for “not met” or when insufficient information was reported (S2 Table). The item was coded as missing if not applicable. We calculated a percentage score, giving all criteria equal weight, and excluding “not applicable” items from the denominator for that study. While recognising the arbitrary nature of cut-offs, we assigned illustrative labels to studies based on their percentage score, with studies scoring ≥75% labelled as “high” quality, those scoring 60%–74% as “medium”, 45%–59% as “low”, and <45% as “very low”.

Data synthesis

To compare study results, we converted “cost per outcome” metrics to US$ in 2024 prices [23]. We first converted to local currency using World Bank [24] rates for the study year, then adjusted for inflation in the study country using World Bank [25] deflators, then converted to 2024 US$. Studies not specific to a country were adjusted using US inflation. Guidance recommends against meta-analysis in most systematic reviews of economic evaluations because costs are highly specific to settings and time [13,26]. We therefore compare metrics in a narrative synthesis to qualitatively assess the overall strength of the evidence and of methods used. We draw conclusions based only on studies judged medium-or high-quality, though we describe methods and results of all included studies.

Country-specific cost-effectiveness thresholds should reflect national resource availability and capture health benefits forgone if resources were withdrawn from interventions already funded [27,28]. For studies that reported cost per DALY averted, we compare to the supply-side (opportunity cost) thresholds for the respective country estimated by Ochalek and colleagues [29]. Supply-side thresholds indicate how much health is “lost” elsewhere if we add or replace an intervention and the budget is fixed. If interventions above the threshold for a given country are funded, they may displace more health benefits than they generate. Supply-side threshold estimates are obtained from econometric analysis of variation in health spending and health outcomes to estimate elasticities, with outcomes captured as DALYs averted [29] or quality-adjusted life years (QALYs) gained [30]. Demand-side thresholds, by contrast, indicate an aspirational judgment about how much the health system should be willing to pay (e.g., 1 × gross domestic product [GDP] per capita) and are discouraged because they fail to reflect opportunity cost of health resources [31]. Supply-side threshold estimates carry uncertainty, but are generally substantially lower than 1 × GDP per capita, particularly in the poorest countries.

For adjustment to 2024 prices, we took the highest and lowest of the percentages of GDP per capita estimated by Ochalek and colleagues [29] using four different methods, considering these to provide a “plausible” range [32]. We then applied those two percentages to GDP per capita for 2024 using World Bank [33] data to generate a range for the 2024 supply-side threshold. We compare studies’ results from the provider perspective, where available, to that range. Since recent cuts to official development assistance have likely tightened governments’ budget constraints in the short run [34], the lower end of the range is likely to be more appropriate. As a sensitivity analysis, we also compare results to a range for cost per QALY gained estimated by Pichon-Riviere and colleagues [30], further explained in S1 Fig.

We also categorised studies in several ways. First, we distinguish between empirical and modelled studies [13]. Empirical studies collect primary data alongside and/or following actual implementation of an intervention, sometimes supplemented by secondary data. Modelled studies evaluate scenarios constructed exclusively from secondary data, rather than from a single occurrence of implementation that actually happened. Second, we distinguish between studies that incorporated modelling of adherence to handwashing over time after the intervention (using primary or secondary data) and those that did not.

Identification, screening, eligibility, and inclusion

Searches of electronic databases and websites yielded 2,610 results, with a further 10 identified from previous reviews and the study team (Fig 1). After removing duplicates, we screened titles and abstracts of 1,369 unique publications and reviewed 45 full-texts. Following full-text review, we included 15 studies. For the 30 studies excluded in full text review, the most common reasons were an ineligible intervention or only reporting a partial (costs only) economic evaluation (S3 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flow diagram.

https://doi.org/10.1371/journal.pmed.1004982.g001

Study settings and interventions

Of the 15 included studies, 8 were in specific L&MICs, 4 in generic L&MIC settings, and 3 in specific high-income countries (HICs) (Table 1). Most interventions assessed behaviour change in domestic settings (n = 13), though 1 assessed a childcare setting [35] and 1 university accommodation [36]. No interventions were in humanitarian settings.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Study characteristics.

https://doi.org/10.1371/journal.pmed.1004982.t001

Five studies promoted handwashing only, and 4 promoted broad or unspecific “hygiene” including handwashing (Table 1). The other 6 studies promoted handwashing alongside other specific behaviours, namely 2 with respiratory hygiene, 1 with food hygiene, 1 with child faeces disposal, 1 with water chlorination, and 1 with water supply. For 13 studies, the comparator was “no intervention” or “existing practice”, and 2 trial-based studies had active controls, namely thermometer provision [36] and vegetable gardening [48] (S4 Table).

No studies evaluated more than one handwashing with soap intervention. However, Varley and colleagues [37] modelled the same handwashing with soap intervention in two different contexts, one with unimproved water (e.g., unprotected wells) and one with public taps. One study compared two hand hygiene interventions to a control, but one intervention used sanitiser so did not meet inclusion criteria [35].

Study design

Of the 15 studies, 4 used benefit-cost analysis, 11 used cost-effectiveness analysis, and none used both (Table 1). Only 5 studies were empirical (some primary data alongside an actual intervention), and most (n = 10) were modelled (wholly secondary data evaluating a hypothetical intervention). All 10 modelled studies assumed health benefits from the literature, as did 2 of the empirical studies, and only 3 empirical studies estimated health effects directly (Table 1). Most studies (n = 11) valued reductions in diarrhoea risk only (all in L&MICs), 3 valued respiratory infections only (all in HICs), and none valued both. The remaining study valued helminth infections only. Studies assuming impact on diarrhoea from the literature employed a wide range of risk reductions (10%–48%).

Cost per DALY averted was reported in 8 studies, cost per QALY gained in 1, and a BCR in 4 (Table 2). Two studies reported only cost per case averted, and 2 reported only intervention-specific measures that cannot easily be compared across studies (e.g., cost per weighted percentage point reduction in three helminths).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Study results.

https://doi.org/10.1371/journal.pmed.1004982.t002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Incremental cost per DALY averted for handwashing interventions in high- or medium-quality studies.

Data are presented in US$ 2024, and cost-effectiveness thresholds are from Ochalek. For visualisation of results compared to alternative thresholds from Pichon-Riviere and colleagues, see S1 Fig. The Borghi and colleagues [39] result for DALYs was estimated by Horton [8], see S1 Fig. Studies did not present confidence intervals for incremental cost-effectiveness ratios.

https://doi.org/10.1371/journal.pmed.1004982.g002

A provider perspective only was taken in 8 studies (explicitly or implicitly), while 5 took a societal perspective only, with 2 reporting results from both perspectives [35,39] (S4 Table). The time horizon of models was unclear in five studies and ranged from 8 months to 80 years in the remaining 10 studies (Table 2), indicating a wide variety of approaches to modelling disease and behaviour. Amongst the 10 studies with clear horizons, the median was 1.5 years. Simple deterministic sensitivity analysis (varying one parameter at a time) was reported in 7 studies, 3 included a probabilistic sensitivity analysis (varying many parameters simultaneously) and 5 reported no sensitivity analysis (S4 Table).

Of the 5 empirical studies, only 2 measured handwashing behaviour change, both using structured observations. Both saw modest improvements in observed handwashing with soap at critical times, from 13% to 31% in Borghi and colleagues [39] and 11% to 19% in Siu and colleagues [48] (Table 2), though these were not explicitly part of the analysis and health benefits were estimated directly. The remaining 3 empirical studies did not measure handwashing behaviour.

Of the 10 modelled studies, only 1 incorporated estimates of behavioural adoption into the model [50]. Specifically, the authors modelled the proportion of targeted households who adopt handwashing with soap (median 40% based on literature review). They also modelled the proportion of adopting households who adhere to the behaviour for the rest of the 1.5 year time horizon (median 50%). The achievement of benefits is then determined by these parameters. Amongst the remaining 9 modelled studies which but did not model behavioural adoption, 6 assumed that behaviour change was sustained for more than 3 years without attenuation.

Study quality

Five studies were scored as high quality, 1 as medium, 6 as low, and 3 as very low.

Percentage scores ranged 36%−90%, with median 55%. Empirical studies tended to score more highly (median 82%) than modelled studies (median 52%). Distributions of scores per CHEERS item are presented in S2 Fig, a heat map for item scores in S3 Fig, and a plot of study quality over time in S4 Fig (which does not appear to show an improvement).

Across CHEERS items, studies were most likely to adequately describe which outcomes were used as the measure(s) of benefit (93% fully met, 0% not met) and least likely to describe the effects of uncertainty (13% fully met, 60% not met). Authors tended to give more attention to effects than to costs, with only 27% fully meeting the CHEERS items for resources/costs. Only 5 studies collected primary cost data from clearly-reported sources, that is, all 5 empirical studies. Amongst the 10 studies using secondary cost data, only 3 scored “fully met” for the costing CHEERS item (S3 Fig). Reasons for lower scores included limited information being provided on cost data sources and assumptions, being unclear on what activities/interventions were assumed to be costed, and using secondary cost data without explanation.

Cost-effectiveness results

Of the 6 medium- or high-quality cost-effectiveness studies, 3 reported cost per DALY averted (Fig 2) as an ICER. Borghi and colleagues [39] report an empirical study of an intervention promoting handwashing and child faeces disposal to mothers in urban Burkina Faso, using secondary data for health impact (Table 1). The intervention was relatively large-scale, being delivered across all administrative areas of a town of 300,000 people over 3 years. Horton [8] uses Borghi and colleagues’s [39] result for cost per death averted (provider perspective) to estimate US$ 96 per DALY averted (2024 prices), by assuming that a death in early childhood is equivalent to 32 discounted DALYs [56]. Both Borghi and Horton conclude that the intervention was cost-effective. This appears justified in comparison to the range of plausible supply-side thresholds for Burkina Faso in 2024 (Fig 2). In a sensitivity analysis using the Pichon-Riviere and colleagues thresholds (S1 Fig), the intervention would still be cost-effective for most of the range.

Siu and colleagues [48] report an empirical study alongside a randomised controlled trial of an intervention promoting handwashing and food hygiene to mothers in rural The Gambia (Table 1). The intervention was relatively small-scale, being delivered across 30 villages (population around 10,000). The authors estimate US$ 937 per DALY averted (societal perspective, 2024 prices) and conclude that the intervention was cost-effective in relation to a threshold of 3 × GDP per capita but not cost-effective in relation to a supply-side threshold. Comparisons to supply-side thresholds would ideally be taken from the provider perspective (rather than societal). Nonetheless, this intervention seems quite unlikely to be cost-effective in comparison to plausible supply-side thresholds for The Gambia in 2024 (Fig 2), and very unlikely to be cost-effective using the Pichon-Riviere and colleagues thresholds (S1 Fig). Both the Siu and colleagues [48] and Borghi and colleagues [39] studies were undertaken after formative research and with researcher-supported delivery, rather than being implemented by the health system in general.

Varley and colleagues [37] report a modelled study of the same broad-based hygiene intervention in two hypothetical water supply contexts, one costing US$ 37 per DALY averted and the other $81 per DALY averted in 2024 prices (Fig 2). No country is modelled but rather a large city in a generic setting characterised by “slums” with few formal public services. The authors conclude that the intervention is cost-effective. Given the lack of a specific country, we present results (Fig 2) against some plausible thresholds for three African countries at low, medium and high points in the distribution of GDP per capita. In all three cases, the intervention would be likely to be cost-effective.

There were two medium- or high-quality studies in HICs. The first evaluated a handwashing intervention in a Spanish childcare setting [35]. It was cost-saving compared to control from the societal perspective, and therefore highly cost-effective. The second evaluated a handwashing and respiratory hygiene intervention in USA student accommodation, and cost $82,967 per QALY gained from societal perspective. This is cost-effective compared to the $100,000 per QALY threshold predominantly applied in the USA [57].

Of the 9 low- and very-low-quality studies, 4 reported a cost per DALY averted, with ICERs ranging from US$ 5 to 101 per DALY averted. This would be cost-effective in most L&MICs—see the threshold for the Democratic Republic of the Congo in Fig 2 which is the lowest amongst Ochalek and colleagues estimates, and about 10 times lower than the Kenya threshold.

Benefit-cost results

The only medium- or high-quality benefit-cost analysis was a modelled study of a generic community-based intervention promoting handwashing with soap [50]. Assuming “medium” levels of handwashing adoption and adherence, the study estimated a mean BCR of 2.1, meaning each US$ 1 invested in the intervention returned outcomes for society valued at $2.1. One-way sensitivity analyses indicated that the intervention would achieve BCRs greater than 1 with different proportions of the population adopting the behaviour (BCR range 1.7–2.3) and adhering to it over time (1.2–2.6)(Fig 3). However, when adoption and adherence were both low (20%), the mean BCR was 0.9, indicating costs higher than benefits. Amongst the 2 low- and very-low-quality benefit-cost analyses, BCRs ranged from 1 to 92.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Benefit-cost ratios (BCR) under one-way sensitivity analyses for initial handwashing adoption and later adherence (Whittington and colleagues [50]).

Length of the horizontal bars is the BCR (2.1) when all parameters are at median values. Error bars show deterministic sensitivity analyses with other parameters remaining at median values. In the upper bar adherence is set at 50% when adoption is being varied, and the lower bar adoption set at 40% when adherence is being varied. The blue line shows break-even where benefits equal costs.

https://doi.org/10.1371/journal.pmed.1004982.g003