Note: this page was initially compiled by GiveWell volunteer Jonah Sinick for primarily internal communication. After some editing by GiveWell staff, we've decided to make it public. First person singular statements are made by Jonah.

A model for the cost-effectiveness of deworming interventions

School deworming targets two types of parasitic worms; schistosoma and soil transmitted helminths (STH). These two types of worms cause diseases in humans called schistosomiasis and helminthiasis respectively. There are two species of schistosoma in Sub-Saharan Africa: schistosoma mansoni and schistosoma haematobium. Each of these can be treated with a drug called praziquantel. There are three species of STH in Sub-Saharan Africa: ascaris lumbricoides, trichuris trichiura and hookworm. These can be treated simultaneously with either a drug called albendazole or a drug called mebendazole. It’s useful to analyze the treatment of schistosomiasis and the treatment of helminthiasis separately.

While schistosoma mansoni and schistosoma haematobium are sufficiently similar so that disabilities associated to each of the two are about the same and so that the treatment effectiveness rate of praziquantel is the same on each of them, the analogous statement is not true for the STH.

The model that I use here is as follows.

• Praziquantel treatment of school children for schistosomiasis.

  Let:

  • (i) w = Cost per child treated per year
  • (ii) x = The portion of treated children who are infected (baseline infection rate)
  • (iii) y = The portion of infected children who are cured by the intervention (treatment effectiveness rate)
  • (iv) z = The disability weight of the infection

  Then we compute the $/DALY ratio for the intervention as

  $/DALY = w/(xyz)

• Albendazole/mendabenzole treatment of school children for STH.

  Let:

  • (i) s = Cost per child treated per year
  • (ii)
    a = Baseline infection rate for A. lumbricoides
    b = Baseline infection rate for T. Trichuira
    c = Baseline infection rate for hookworm
  • (iii)
    d = Treatment effectiveness rate for A. lumbricoides
    e = Treatment effectiveness rate for T. Trichuira
    f = Treatment effectiveness rate for hookworm
  • (iv)
    g = disability weight for A. lumbricoides
    h = disability weight for T. Trichuira
    i = disability weight for hookworm

  Then we compute the $/DALY ratio for the intervention as:

  $/DALY = s/(adg + beh + cfi)

Implicit assumptions

The model mentioned above carries many implicit assumptions. As such, the ranges computed using the model need to be broadened to account for model uncertainty. Among assumptions implicit in the model, the most controvertible ones are:

1. Each child cured experiences exactly one additional year free of worms.
2. The benefit of the treatment is entirely concentrated in those children who are completely cured.
3. The conditions underlying the large-scale national deworming programs which have been studied and reported on in the past are representative of the conditions underlying future such programs.
4. The disability weights attached to infestation by each of the worms are known.

These assumptions are idealized and do not hold in practice. Nevertheless, I’ve used them because:

• The assumptions appear to be identical with those used by the authors of the DCP-2 cost-effectiveness calculations for deworming interventions.
• The assumptions are simple enough so that they’ve allowed us to do an analysis of manageable length. We hope that those who wish to create more refined and accurate models of the cost-effectiveness of deworming interventions can utilize the information and analysis which we present here.

We will revisit these assumptions at the end of the present article.

Findings

• The revised disability weight for schistosomiasis given by King et. al. seem to be well grounded relative to the original disability weight attributed to schistosomiasis.
• The existing research on parasitic worms does not suffice to give a precise $/DALY figure for the cost-effectiveness of deworming interventions. This is in line with a quotation from tropical epidemiology and disease control expert Simon Brooker:
  

  The precise amount of morbidity and mortality caused by intestinal nematodes will never be known. This elusiveness is due to the non-specificity of clinical signs, difficulties in parasitological diagnosis and a paucity of reliable and accurate epidemiological data, compounded by the fact that much of the burden is concentrated among countries with weak disease surveillance systems. (Brooker 2010)

• Plausible estimates from the literature for each of (i), (ii), (iii) and (iv) in the above model vary across large ranges.
• Using the arithmetic means of the best available estimates for the quantities in each of (i), (ii) and (iii) and then allowing (iv) to vary across the range of best available disability weight estimates for each worm disease gives cost-effectiveness estimates ranges of
  

  $28.19-$70.48/DALY for schistosomiasis treatment

  $82.54/DALY for STH treatment.

• In view of the above ranges it appears that while school-based deworming is among the more cost-effective health interventions (in $/DALY), they are not the most cost-effective health interventions. As such, the $/DALY figures for deworming should not be used as justification for favoring deworming charities over all other charities working in international health. The present article deals only with the $/DALY issue; as mentioned at the beginning of the article, the DALY system has limitations. Considering factors outside of the DALY construct may work either for or against deworming relative to other health interventions.

Data concerning schistosomiasis

Here I’ll give the ranges that I used for each of the quantities (i) - (iv) in the cost-effectiveness calculation for schistosomiasis treatment.

1. Cost per child treated: $0.27/year - $0.47/year; midpoint $0.37/year.

   A standard treatment program consists of one treatment per year, so it suffices to give to give a range for the cost of a single treatment. School based deworming programs carry a very low cost of medicine delivery and distribution because the programs utilize existing school infrastructure to administer the treatment.

   The lowest delivery cost comes from Olds et. al. (1999), but in Guyatt (2003) one of the authors explicitly stated in that there were contingencies in this study that made the delivery costs unrealistically low. The next lowest delivery cost was $0.084 in the 2006 nation-wide program in Burkina Faso reported on in Gabrielli et. al. (2006). The highest delivery cost that I found is $0.17 per child; calculated from a nationwide school-based program in Uganda reported on in Brooker et. al. (2008). So I used the range $0.084-$0.17 for the delivery cost.

   The bulk of the cost per child treated is the cost of the drug. The standard treatment for schistosomiasis infestation is ingesting several 600mg praziquantel tablets, with the appropriate number of tablets depending on the child's weight. The paper Guyatt (2003) gives 2.5 as the average number of tablets administered and this number is corroborated by Gabrielli et. al. (2006) which reads

   On average, praziquantel costs US$ 0.075 per 600 mg tablet and each school-age child requires 2.5 tablets.

   Brand name praziquantel is much more expensive than generic praziquantel, but with the single exception of Olds et. al. (1999) all of the studies that I've looked at use generic praziquantel. Moreover a number of such studies are conducted by staff from the deworming charity Schistosomiasis Control Initiative. From this, I've inferred that it's standard practice to use the cheaper generic praziquantel. Thus I’ve assumed the usage of generic praziquantel. According to Guyatt (2003) “The cost for generic praziquantel ranges from $0.075-$0.12 per pill.” I'll note that Guyatt's lower bound of $0.075 was attained by the 2006 Burkina Faso program reported on in Gabrielli et. al. (2006). Because the average number of tablets is reported to be 2.5, I multiplied Guyatt's range by a factor of 2.5 to obtain a range for the average medication cost.

   Putting these things together I obtained a range of $0.27 - $0.47 for the cost of treating a child once a year.

   One might wonder about whether the treatment reaches the intended recipients. I don't have high quality information about this, but I would refer to the fact (discussed below) that there's strong evidence that the vast majority of intended recipients of STH treatment do in fact receive treatment.

   The only complicating factor with schistosomiasis treatment is the fact that the dose is not predetermined (as it is for STH) but rather is intended to be determined by measuring the students' height by a standardized device and using their heights as a proxy for the students' weight. I was unable to find data on how frequently students receive dosage based on their height as intended. I would guess that student heights are sufficiently close to one another so that even when treated students do not receive the exact intended dose, the dose that they receive is very close to the intended dose so that the effects of the treatment are very close to those of the intended treatment. However, the potential for misdosing does drive the expected cost-effectiveness of deworming down.

2. The portion of those who are treated who are infected: 30% - 80%; midpoint 50%.

   The WHO recommends that schistosomiasis be treated in schools with prevalence of schistosomiasis greater than 50%. The study reported on in Miguel and Kremer (2004) used a lower threshold of 30%. These points can be taken to be weak evidence that the portion of those who are treated who are also infected is ≥ 30%. This is corroborated by the fact that the lowest figure for preexisting prevalence that I found in the literature is 33.4% in the Albertine Nile area of Uganda from Zhang et. al. (2007).

   As for the upper bound, there are figures up to 90% given in the literature but most of these are taken from small studies involving areas with unrepresentatively high prevalence of schistosomiasis. The fraction of these figures which are 80% or above is very small and there were several data points around 80% so I used 80% as an upper bound. That being said, I have a hunch that an initial prevalence rate of 70% is the highest that one sees for national treatment programs that involve schools sprawled over a diverse collection of regions.

3. The portion of infected children who are cured by the intervention: 45% - 60%; midpoint 52.5%

   I encountered four studies which bear on this point: Olds et. al. (1999), Tohon et. al. (2008), Toure et. al. (2008) and Zhang et. al. (2008). The last two of these had very high (> 55%) drop-out rates which could have led to selection effects which badly skewed the study results so I discarded them. Of the two remaining studies, Olds et. al. (1999) had no drop-outs, small samples (~100 person) and reported cure rates of 47.9% and 57.8% (depending on the species of schistosomiasis). The study Tohon et. al. (2008) had a 10% drop out rate but a much larger (~1500 person) sample; this one reported a cure rate of 49.6%. Rounding off the endpoints of the implied interval for simplicity gives the quoted range.

4. Disability weight: 2% - 5%

   The official disability weight given by the WHO Global Burden for Disease is 0.5% or 0.6%. However, according to a 2002 WHO report:

   For schistosomiasis, the disease burden was originally calculated by estimating the prevalence of infection and associating a low disability weight (0.005 - 0.006) with infection. The calculation took into account no clincial sequelae and only directly attributable mortality (then assumed to be 7000 per year (2)). It is widely believed that the calculated figure for DALYs lost to schistosomiasis represents a significant underestimate and should be revised.

   King et. al. (2005) reevaluates the disability weight attached to schistosomiasis using a meta-analysis of 135 papers about schistosomiasis symptoms and concludes that the disability weight should be in the range 2%-15%. Here the upper bound of 15% is based on a single unreplicated study and is not calculated within the DALY paradigm, so I discarded it. The lower bound of 2% comes from a conservative analysis of the frequency with which various symptoms of schistosomiasis occur as a result of the parasite and from the WHO disability weights of the symptoms. In King (2010), King mentions a less conservative figure of 5% which I've used as my upper bound.

   The data that goes into the analysis of the frequency with which schistosomiasis causes a given symptom does not come from randomized controlled trials but rather from more indirect (and less solid) methods. My intuition is that though there's a lot of uncertainty as to the extent to which a given factor influences the disability weight, these factors as sufficiently unrelated so that it's unlikely that a large majority of these factors would point in one direction or the other. I concluded that a range of 2% - 5% for the DALY based disability weight schistosomiasis is reasonable.

Data concerning soil-transmitted helminths

Here I give the data that I used to obtain ranges for each of the quantities (i) - (iv) in estimating the cost-effectiveness of STH treatment.

1. Cost per child treated: $0.03/year - $0.14/year; midpoint $0.085/year.

   Standard treatment is either annual or biannual depending on the prevalence of STH in the region treated.

   The lower limit comes from Montressor et. al. (2007) which reports a cost of $0.03/child for annual treatment.

   The upper limit comes from Phommosack et. al. (2008) adding the cost of $0.12/child for biannual treatment to a fifth of the capital cost (spent once every five years) of $0.11

   As I mentioned in the discussion of schistosomiasis treatment, the distribution rates for STH medications in school-based deworming programs are consistently very high. In the papers that I read between 95%-100% of the children targeted received treatment. These figures come from nationwide teacher surveys together with student surveys at randomly chosen schools which corroborated the results of the teacher surveys.

2. STH Baseline Prevalence Rates:

   I was able to find five triples of baseline prevalence rates. I list them here:

   • From a national survey in Cameroon reported on in Ratard (1991):

     Ascaris lumbricoides: 42.3%
     Trichuris trichiura: 54.7%
     Hookworm: 16.6%

   • From a cross-sectional school survey in Busia, Kenya reported on in Brooker (2000):

     Ascaris lumbricoides: 41.9%
     Trichuris trichiura: 55.2%
     Hookworm: 77.5%

   • From a cross-sectional school survey in southern Uganda reported on in Kabaterine (2001):

     Ascaris lumbricoides: 17.5%
     Trichuris trichiura: 7.3%
     Hookworm: 44.5%

   • From a nationwide study in Uganda reported on in Kabaterine (2007):

     Ascaris lumbricoides: 2.8%
     Trichuris trichiura: 2.2%
     Hookworm: 50.9%

   • From a nationwide study in Lao PDR reported on in Phommosack et. al. (2008):

     Ascaris lumbricoides: 60.4%
     Trichuris trichiura: 42.5%
     Hookworm: 19.7%

   Taking taking the arithmetic means of the above percentages for each worm gives

   Ascaris lumbricoides: 33.0%
   Trichuris trichiura: 32.4%
   Hookworm: 41.8%

   All of these triples are consistent with the hypothesized distribution of different STH across different countries in sub-Saharan Africa given in de Silva et. al. (2003).

3. STH Treatment Effectiveness Rates:

   The literature review Bennett, Guyatt (2000) collected the treatment effectiveness rates for STH treatment from many studies. For each STH the authors then discarded outliers to arrive at the treatment effectiveness ranges listed below:

   Ascaris lumbricoides: 90% - 100%
   Trichuris trichiura: 22% - 72%
   Hookworm: 22% - 90%

   The midpoints of these ranges are given by

   Ascaris lumbricoides: 95%
   Trichuris trichiura: 47%
   Hookworm: 56%

   According to the data in the review, mebendazole is slightly more effective at treating T. Trichiura than albendazole is but much less effective at treating hookworm than albendazole is. If one assumes that the treatment uses albendazole then treatment effectiveness rate for range hookworm becomes 70%-90%; raising the midpoint treatment effectiveness to 80%.

4. STH Disability Weights:

   I obtained two different triples of disability weights using information from the literature. I computed the first triple using the standard DALY burdens attached to STH provided by the WHO and computed the second triple using the data in Chan (1997). I give these triples in turn:

   • According to WHO (2004), the DALY burdens attached to the three STH in 2004 were
     

     A. lumbricoides: 1851000 DALYs
     T. trichiura: 1012000 DALYs
     Hookworm 1092000 DALYs

     According to de Silva et. al. (2003)

     ...ascariasis remains exceedingly common with over 1.2 billion infections globally. Almost half these infections are in China, which still has the highest prevalence. Trichuriasis and hookworm amount to about 700 – 800 million infections each.

     I took the number of Trichuriasis and hookworm infections to be 750 million and for each worm I divided the DALY burden in WHO (2004) by the global number of infections in de Silva et. al. (2003) to obtain the following average disability weights:

     Ascariasis: 0.154%
     Trichuriasis: 0.135%
     Hookworm: 0.1456%

   • Chan (1997) gives DALY burden figures of

     A. lumbricoides: 10.5 million DALYs
     T. trichiura: 6.4 million DALYs
     Hookworm 22.1 million DALYs

     The prevalence estimates that Chan gives are

     1273 million A. lumbricoides infections
     902 million T. trichiura infections
     1277 million hookworm infections

     Dividing through, I obtained average disability weights of

     0.8% for A. lumbricoides
     0.7% for T. trichiura
     1.7% for hookworm

     These disability weights, combined with the midpoint cost, prevalence, and treatment effectiveness would imply a cost-effectiveness for STH deworming of $11.25 per DALY.

     Note: we do not regard Chan's DALY burden estimates as credible at this point. Chan's analysis was based on a lower worm threshold for experiencing morbidity than further research has found appropriate (Brooker 2010). Furthermore, the cited source of the relevant data is a working paper, the published version of which does not contain the data cited. We therefore do not think that estimates were ever thoroughly peer-reviewed; even if they were, they no longer represent scholarly consensus.

Revisiting implicit assumptions

Earlier I highlighted some controvertible assumptions. Here I’ll comment on ways in which they break down in practice.

1. If reinfection typically occurs between treatments, the children treated will not experience a full year free of worms due to treatment. According to Olds et. al. (1999), "After 45 days [beyond treatment]...prevalence rose slowly for hookworm and schistosomiasis but rapidly for Ascariasis due to reinfection." In all studies that I’ve encountered, acariasis is treated at least once every 180 days. The above passage suggests that the average time that it takes for those who are treated for ascariasis to be infected may be less than 180 days. As such, children who are treated for ascariasis may not receive a full year’s benefit.

   On the flip side, according to Toure (2008):

   In previous small-scale studies on S. haematobium control in eastern Africa, an annual treatment strategy was predominantly used, with varying results. [21], [22]. However, in western Africa, one study in the Niger showed that, 3 years after a single PZQ treatment, prevalence and intensity of S. haematobium infection remained significantly lower than at baseline [23], [24]. In another study in Ghana, with one single PZQ treatment, intensity of S. haematobium infection was reduced by 80–99% 12 months after treatment and remained very low in two of three study areas 24 months after treatment. [25]

   This suggests that treating a child a single time with praziquantel allows the child to experience more than one year free of schistosoma and symptoms. (Of course, during the period when the child is free of reinfection, further treatments will not confer additional benefit onto the child).

   Another factor which has bearing on whether a child is free of a worm due to treatment is that the percentage of children seems to drop dramatically on its own accord when they mature (see Bundy et. al. (2004), tables 9.9, 9.10 & 9.11 for STH).

2. Infestation by parasitic worms is not an “either-or” quality; there are “ligher” and “heavier” cases of worm infestation. Treating an infected child can have the effect of shifting him or her from a “heavy” case to a “light” case even if it doesn’t completely cure the child (see e.g. Phommasack et. al. (2008)). In this respect the model understates the cost-effectiveness of deworming interventions.

   On a different note, because humans are part of the life cycle of parasitic worms, curing individuals carries the positive externality of making it less likely that other members of the community will be infected with worms. It’s unclear how large this externality is. Miguel and Kremer (2004) attempt to address this point quantitatively but there remain questions of whether the conditions described in the paper are representative and whether the analysis used to quantify the externality is correct.

3. There’s the usual issue of publication bias: all else being equal the deworming programs that have been reported on in published studies are more likely to be highly effectiveness of deworming than typical deworming program. My subjective sense from researching the subject is that the published studies are pretty representative but I may be wrong.

   Even assuming that the published studies are representative of deworming interventions up to present date there’s the question of whether expanding deworming programs will carry the same cost-effectiveness. Intuitively, the areas that deworming researchers studied first and the areas that charities worked in first interventions are likely to carry unusually high baseline prevalence of parasitic worms relative to the collection of all areas that could benefit from deworming interventions. So one would expect to see the baseline prevalence of worms infection to be lower at the margin than on average. Very little information is available about this topic: the effect could be negligible or large.

   There is also a question of whether the drugs used to treat worms will remain effective in the future. According to OneWorld Health:

   The health benefits of deworming, the anticipated increase in use of a limited number of drugs, and concerns over the reduced efficacy of existing drugs have intensified the need to develop new drug(s) against STH.

   suggesting that there's some reason to question this assumption for STH treatment. In principle there might be a similar issue of praziquantel's efficacy against schistosomiasis dropping over time. This seems to be temporally far off but still possible (as discussed in sections 3.7 and 3.8 of a 2002 WHO report).

4. The disability weights attributed to deworming interventions seem to be on very shaky ground. The disability weights that the literature attached to infection by a given parasitic worm are computed by taking a weighted average of the GBD disability weights of several symptoms of a given worm. However:
   • It’s not clear that the GBD disability weights attached to the symptoms are themselves well grounded. The proper disability weight attached to a symptom plausibly varies with its severity and cultural/environmental context, and the GBD disability weights may not be representative of the “average case” disability weight of the symptom.
   • Even if the GBD disability weights attached to the symptoms are well grounded, taking a weighted average of them need not give the proper disability weight for schistosomiasis. The proper disability weight attached to simultaneously possessing two health conditions may be less than or greater than the sum of the proper disability weights attached to each health problem. For a case in point, it’s plausible that the proper disability weight of losing both arms is different from the sum of the disability weight of losing one’s left arm and of losing one’s right arm.
   • The lists of symptoms taken into account in the calculations are probably not exhaustive.

Sources

• Bennett and Guyatt. Reducing intestinal nematode infection: efficacy of albendazole and mebendazole. (2000)
• Brooker et. al. Epidemiology of single and multiple species of helminth infections among school children in Busia District, Kenya. (2000)
• Brooker. Estimating the global distribution and disease burden of intestinal nematode infections: Adding up the numbers – A review. (2010)
• Bundy et. al. Global Epidemiology of Infectious Diseases, Chapter 9: Intestinal Nematode Infections. (2004)
• Chan. The global burden of intestinal nematode infections--fifty years on. (1997)
• de Silva et. al. Soil-transmitted helminth infections: updating the global picture. (2003)
• Hotez et. al. Disease Control Priorities in Developing Countries. 2nd edition. Chapter 24. (2006)
• Gabrielli et. al. A combined school- and community-based campaign targeting all school-age children of Burkina Faso against schistosomiasis and soil-transmitted helminthiasis: performance, financial costs and implications for sustainability. (2006)
• Guyatt. The cost of delivering and sustaining a control programme for schistosomiasis and soil-transmitted helminthiasis. (2003)
• Kabatereine et. al. Epidemiology of intestinal helminth infestations among schoolchildren in southern Uganda. (2001)
• Kabatereine et. al. Impact of a national helminth control programme on infection and morbidity in Ugandan schoolchildren. (2007)
• King. Reassessment of the cost of chronic helmintic infection: a meta-analysis of disability-related outcomes in endemic schistosomiasis. (2005)
• King and Bertino. Asymmetries of Poverty: Why Global Burden of Disease Valuations Underestimate the Burden of Neglected Tropical Diseases. (2008)
• King and Dangerfield-Cha. The unacknowledged impact of chronic schistosomiasis. (2008)
• King. Parasites and Poverty: The Case of Schistosomiasis. (2010)
• Miguel and Kremer. Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities. (2004)
• Montresor et. al. Soil-transmitted helminthiasis in Myanmar and approximate costs for countrywide control. (2004)
• Montresor et. al. Cost containment in a school deworming programme targeting over 2.7 million children in Vietnam. (2007)
• Olds et. al. Double-blind placebo-controlled study of concurrent administration of albendazole and praziquantel in schoolchildren with schistosomiasis and geohelminths. (1999)
• OneWorld Health. Other Initiatives. (2011) Available online at: http://www.oneworldhealth.org/other-initiatives
• PCD - The Partnership for Child Development. The cost of large-scale school health programmes which deliver antihelmintics to children in Ghana and Tanzania. (1999)
• Phommasack et. al. Coverage and costs of a school deworming programme in 2007 targeting all primary schools in Lao PDR. (2008)
• Ratard et. al. Ascariasis and trichuriasis in Cameroon. (1991)
• Sinuon et. al. Financial costs of deworming children in all primary schools in Cambodia. (2005)
• Tohon. Controlling Schistosomiasis: Significant Decrease of Anaemia Prevalence One Year after a Single Dose of Praziquantel in Nigerien Schoolchildren. (2008)
• Toure et. al. Two-year impact of single praziquantel treatment on infection in the national control programme on schistosomiasis in Burkina Faso. (2008)
• World Health Organization. WHO Technical Report Series 912: Prevention and Control of Schistosomiasis and Soil-Transmitted Helminthiasis. (2002)
• World Health Organization. The global burden of disease: 2004 update. (2008)
• Zhang et. al. Parasitological impact of 2-year preventive chemotherapy on schistosomiasis and soil-transmitted helminthiasis in Uganda. (2007)