Malaria remains a major public health problem. Global estimates of the malaria disease burden for 2000 indicated that there were at least 300 to 500 million clinical cases annually, of which 90% occurred in sub-Saharan Africa. Moreover, around one million deaths are related to malaria every year, of which an overwhelming proportion occurs in Africa (WHO 1997; WHO 2003). In Africa, malaria accounts for an estimated 25% of all childhood mortality below age five, excluding neonatal mortality (WHO 2003). Recent studies suggest that this percentage might even be higher because of the contribution of malaria as indirect cause of death (Alonso 1991; Molineaux 1997). In addition, it might be more of a problem in adults than thought previously, as suggested by the high proportion of adults dying of "acute febrile illness" in Tanzania (Kitange 1996). In Africa, malaria is the primary cause of disease burden measured by disability-adjusted life years (WHO 2003; World Bank 1993). In countries outside the African continent, malaria appears to be an increasing problem; for example, in India malaria is making a comeback after decades of effective control. Malaria places an enormous economic burden on affected countries and has a highly detrimental effect on economic and social development.
In 1992, the World Health Organization convened a ministerial conference in Amsterdam to give a new impetus to control activities. While the consensus at this meeting was that prompt access to diagnosis and treatment remained the mainstay of malaria control, there was a renewed emphasis on preventive measures, both at the community and at the individual level (WHO 1993). The most promising preventive measures mentioned were insecticide-treated bed nets and curtains, collectively known as insecticide-treated nets (ITNs). In 1998, the main international health agencies launched an ambitious partnership, Roll Back Malaria, to tackle the global malaria issue. The wide-scale implementation of ITNs is now one of the four main strategies to reduce morbidity and mortality from malaria (WHO 2003), with a target set by African Heads of State to protect 60% of all pregnant women and children by 2005. As a result, many large-scale programmes have taken off during the last few years.
Insecticide-treated nets (ITNs)
Using mosquito nets as a protection against nuisance insects was practiced in historical times (Lindsay 1988). During World War II, Russian, German, and US armies treated bed nets and combat fatigues with residual insecticide to protect soldiers against vector-borne diseases (mainly malaria and leishmaniasis) (Curtis 1991). In the late 1970s, entomologists started using synthetic pyrethroids: their high insecticidal activity and low mammalian toxicity made them ideal for this purpose.
In the 1980s, studies of ITNs showed that pyrethroids were safe and that ITNs had an impact on various measures of mosquito biting (such as the proportion of mosquitoes successfully feeding on humans and the number of times a mosquito bit humans in one night). These studies showed that pyrethroids worked by both repelling and killing mosquitoes. In addition, researchers determined optimal doses of various insecticides with different materials (Curtis 1991; Curtis 1992a; Curtis 1996; Lines 1996; Rozendaal 1989a). The cost-effectiveness of ITNs has also been demonstrated (Goodman 1999; Hanson 2003).
Given the part played by Plasmodium falciparum malaria as a direct and indirect cause of death in African children, the main public health question for ITNs is whether they reduce mortality in children. One observational study of impregnated bed nets in The Gambia reported a 42% reduction in all mortality in children aged 1 to 59 months in 1991 (Alonso 1991). This dramatic result from the first mortality trial prompted the UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases (TDR) to collaborate with around 20 agencies to launch four additional large-scale trials to measure the impact of ITNs on overall child mortality in different endemic areas of Africa (Burkina Faso, The Gambia, Ghana, and Kenya). Since this time, several trials have been conducted including a large-scale trial completed in 2000 in Western Kenya in an area of high perennial transmission.
To assess the impact of insecticide-treated bed nets or curtains on mortality, malarial illness (life-threatening and mild), malaria parasitaemia, anaemia, and spleen rates.
Any effect of ITNs compared to routine antimalarial control measures in reducing malaria-specific and all-cause morbidity and mortality will be:
- less in areas with high entomological inoculation rates (ie stable malarious areas with > 1 infective bite per year) compared to areas with low inoculation rates (unstable malaria with < 1 infective bite per year);
- less when the population under study already uses untreated bed nets regularly before the start of the trial (coverage of untreated nets by household at least 40%).
The original protocol aimed to explore whether the impact of ITNs on all-cause mortality is greater in areas where access to treatment for malarial illness is limited. However, I could not investigate this because the relevant measures of treatment access were not available.
Criteria for considering studies for this review
Types of studies
Individual and cluster randomized controlled trials.
Types of participants
- Children and adults living in rural and urban malarious areas.
- Excluded: trials dealing only with pregnant women, because they are reviewed elsewhere (seeGamble 2006); and trials examining the impact of ITNs among soldiers or travellers, because they are not representative of the general population.
Types of interventions
Bed nets or curtains treated with a synthetic pyrethroid insecticide at a minimum target impregnation dose of:
- 200 mg/m
2permethrin or etofenprox;
- 30 mg/m
- 20 mg/m
- 10 mg/m
No distinction was made between insecticide-treated bed nets and door/window/eave/wall curtains, which were assumed to have approximately the same impact.
Recently, other types of materials such as wall curtains, blankets, sheets, and veils have also been treated and assessed. However, these are excluded from the review because they are difficult to compare to treated mosquito nets and curtains for which many more studies are available; they are listed in the 'Characteristic of excluded trials'.
Types of outcome measures
- Child mortality from all causes: measured using protective efficacy and rate difference.
- Severe disease: measured using site-specific definitions, which were based on the World Health Organization guidelines (WHO 1990) and on Marsh 1995. The definition included P. falciparum parasitaemia. Cerebral malaria was defined as coma or prostration and/or multiple seizures. The cut-off for severe, life-threatening anaemia was set at 5.1g/litre (WHO 1990).
- Uncomplicated clinical episodes: measured using site-specific definitions, including measured or reported fever, with or without parasitological confirmation. Measurements were usually done in the frame of prospective longitudinal studies, but I also considered trials using validated retrospective assessments in the frame of cross-sectional surveys. In areas with entomological inoculation rates below 1 (unstable malaria), I considered P. falciparum and P. vivax episodes separately.
- Parasite prevalence: parasite prevalence due to P. falciparum and P. vivax was obtained using the site-specific method for estimating parasitaemia − usually thick and/or thin blood smears. When more than one survey was done, the reported prevalence result is the average prevalence of all the surveys.
- High parasitaemia: measured using site-specific definitions of high parasitaemia, provided the cut-off value between high and low was determined prior to data analysis.
- Anaemia: expressed in mean packed cell volume (PCV); it is equivalent to the percentage haematocrit. Results given in g/decilitre were converted with a standard factor of 3:1, that is, 1 g/decilitre equals 3% PCV (Wallach 1986).
- Splenomegaly: measured in all trials using the Hackett scale.
- Anthropometric measures: standard anthropological measures (weight-for-age, height-for-age, weight-for-height, skinfold thickness, or mid-upper arm circumference) and the impact of ITNs on them.
Search methods for identification of studies
I attempted to identify all relevant trials regardless of language or publication status (published, unpublished, in press, and in progress).
I searched the following databases using the search terms and strategy described in Appendix 1: Cochrane Infectious Diseases Group's trials register (January 2003); Cochrane Central Register of Controlled Trials (CENTRAL), published in The Cochrane Library (Issue 1, 2003); MEDLINE (1966 to October 2003); EMBASE (1974 to November 2002); and LILACS (1982 to January 2003).
I handsearched some foreign language tropical medicine journals (Bulletin OCEAC, Bulletin de la Société de Pathologie Exotique, Médecine Tropicale, Revista do Instituto de Medicina Tropical de Sao Paulo) for the period 1980 to 1997.
Researchers, organizations, and pharmaceutical companies
I contacted many researchers actively involved in the field of ITNs and asked about unpublished past or ongoing work.
I contacted the following agencies, which have funded ITN trials, for unpublished and ongoing trials: UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases (TDR); International Development Research Center (IDRC), Canada; The Department for International Development, UK; and The European Union Directorate-General XII.
I contacted the following manufacturers of pyrethroids used for treating netting for unpublished and ongoing trials: AgrEvo (now part of Bayer); Bayer; Cyanamid; Mitsui; Sumitomo; and Zeneca (now part of Syngenta).
I consulted the following reviews: Abdulla 1995; Bermejo 1992; Carnevale 1991; Cattani 1997; Choi 1995; Curtis 1992b; Molineaux 1994; Rozendaal 1989a; Sexton 1994; Voorham 1997; WHO 1989; Xu 1988; Yadav 1997; and Zimmerman 1997.
I consulted the following books dealing with ITNs: 'Control of disease vectors in the community' (Curtis 1991); 'Malaria: waiting for the vaccine' (Targett 1991); and 'Net Gain, a new method for preventing malaria deaths' (Lengeler 1996a; Lengeler 1997a).
I also checked the reference lists of all trials identified by the above methods.
Data collection and analysis
Selection of studies
The reviewer and two independent assessors experienced in trial epidemiology (Dr Gerd Antes and Dr Daniel Galandi, German Cochrane Centre) applied the inclusion criteria to all identified trials and reached agreement by consensus.
Data extraction and management
I used standard forms to extract the following descriptive data.
- Trial location.
- Duration and type of intervention.
- Randomization procedure.
- Type of control group.
- Age and gender of participants.
- Percentage of target group protected by ITNs and untreated nets.
- Malarial endemicity (as defined by the entomological inoculation rate: the number of times on average a person living in the area receives an infected mosquito bite per year).
- Species and proportion of Plasmodium parasites.
- Main vectors.
When these data were not given in the primary trial reference, I used secondary sources and included the references.
Assessment of risk of bias in included studies
I assessed the risk of bias in the included trials using generation of allocation sequence, allocation concealment, inclusion of all randomized participants, and blinding, as described in Appendix 2.
I entered data as numerators and denominators for all dichotomous outcomes. For continuous variables, I entered data as the number of participants, mean, and standard deviation.
I used EasyMA 2001 and Review Manager 4.2.2 to calculate the risk ratio, relative rate, rate difference, summary risk ratio/relative rate, summary mean difference, and for testing the homogeneity between trials (using a chi-squared (chi
I considered only crude rate or risk ratios, that is, not adjusted for any co-variates. If only adjusted rates were given in a reference, I attempted to contact the authors to provide the crude rates/risks.
Many trials in the area of vector control interventions are randomized by cluster. While the actual rate/risk ratio is not affected by cluster allocation, the confidence interval (CI) has to be adjusted (made wider) to take into account the inter-cluster variability. This problem has been reviewed by several authors (Bennett 2002; Donner 1993; Donner 1994; Hayes 2000; Klar 1995). This presented me with the problem of interpreting the statistical significance of trials that had not corrected for design effects in their calculations of confidence intervals, and how to obtain accurate confidence intervals when combining data between such trials. For the child mortality from all causes outcome, corrected confidence intervals were available, and I used the generic inverse variance method available in Review Manager 4.2 to combine cluster randomized controlled trials and obtain corrected confidence intervals. Unfortunately, corrected confidence intervals or standard errors were not available for all trials for the other outcomes. Because of this, I have presented summary risk ratios without confidence intervals and in tables, rather than with meta-analysis figures.
For parasite prevalence, I calculated an average denominator from all the surveys and chose the appropriate numerator to fit the average prevalence and average denominator. I selected this procedure in order not to inflate the denominator artificially by adding up the participants from repeated surveys. This procedure gives more weight to larger trials doing only one survey rather than smaller trials doing multiple surveys.
I performed a limited number of additional analyses with the mortality data. I used Epi Info 2002 to perform linear regressions in order to test for trends in the mortality outcomes as a result of transmission intensity. Mortality was measured using protective efficacy and rate difference. Protective efficacy is based on the risk ratio or relative rate. The protective efficacy (PE) is calculated as PE = (1 - risk ratio or relative rate) x 100. Rate difference estimates directly how many child deaths can be avoided through the use of the intervention (in this case deaths per 1000 children protected per year). I only calculated rate difference for mortality from all causes since it was the only measure for which similar incidence measures were used in all trials.
I pre-specified two comparisons: trials in which the control group did not have a net at all; and trials in which the control group had untreated bed nets or curtain; and pre-specified one stratified analysis: entomological inoculation rate above or below one (stable versus unstable malaria).
Description of studies
I identified 113 potentially relevant studies. Of these I excluded 32 published studies without further analysis (and did not include in the 'Characteristics of excluded studies') for the following reasons:
- 15 were only descriptive in nature with no defined control groups, used a before-after evaluation design or a comparison of users versus non-users, and mainly concerned untreated nets (Bradley 1986; Burkot 1990; Campbell 1987; Cattani 1986; Clarke 2001; Dulay 1992; Dutta 1989; Fernandez 1991; Genton 1994; Millen 1986; Rozendaal 1989b; Samarawickrema 1992; Sandy 1992; van der Hoek 1998; Voorham 1997).
I identified the remaining 81 trials (including 10 (12%) unpublished trials) through the following sources.
- Electronic sources and manual search of references: 57 (70%).
- Handsearch of journals in non-English language journals: 4 (5%).
- Books and reviews: 2 (3%).
- Insecticide manufacturers: 12 (15%).
- Personal contacts with authors and internet search: 6 (7%).
Of these, 59 trials were excluded: 55 because they were not randomized (in two, allocation was achieved "by chance"); 2 because they used materials other than bed nets or curtains (such as wall curtains or blankets); and 2 because they were not adequately controlled (before and after assessments). I have provided the reasons for excluding them in the 'Characteristics of excluded studies'.
The remaining 22 trials, including 1 trial that is currently unpublished, met the inclusion criteria for this review. These trials are described below (see the 'Characteristics of included studies' for details).
Trial design and location
Fourteen of the included trials were cluster RCTs (by villages, blocks of villages, zones within one village), and 8 were individual RCTs (6 by household and 2 by individuals) (Appendix 2). The eight individual randomized controlled trials were analysed on an intention-to-treat basis.
Thirteen trials were conducted in sub-Saharan Africa, 5 in Latin America, 2 in Thailand, 1 in Pakistan, and 1 in Iran. Thus 13 trials were carried out in areas of stable endemicity areas, and 9 in areas of unstable endemicity.
Trials included either the whole population of selected areas (typically in low endemicity areas) or specific age groups (typically children in high endemicity areas), and gender ratios were well balanced (range of male:female ratio: 0.8 to 1.2).
Nineteen trials examined the impact of treated bed nets, while two examined the impact of treated curtains. One trial compared treated nets, treated curtains, and no bed nets or curtains (Kenya (Sexton)). In some trials the intervention consisted of treating existing nets with an insecticide ('treatment of nets') while in other trials the investigators provided treated mosquito nets or curtains to the population ('treated nets' and 'treated curtains'). Most nets or curtains were treated with permethrin (200 (n = 3), 500 mg (n = 10), or 1000 mg/m
Half of the trials did not use bed nets or curtains as the control group, and other 11 trials used untreated nets or curtains. The usage rate of the untreated nets was high (> 80%), except in Gambia (D'Alessandro), in which it varied between 50% and 90% (according to the area) in both the intervention and control groups, and in Peru Coast (Kroeger) in which it was 63%; no usage rate provided for Madagascar (Rabarison).
The five trials that examined child mortality from all causes as an outcome were conducted in highly malaria endemic areas in sub-Saharan Africa. No trial presented results for all the possible outcomes, and the majority of trials presented two to five different outcomes (see Appendix 2).
Risk of bias in included studies
See Figure 1 for a summary of the risk of bias in the included trials.
|Figure 1. Methodological quality (risk of bias) in included trials.|
Generation of allocation sequence
Generation of the allocation sequences used random number tables or an equivalent method in 9 trials (graded 'A'); randomization was mentioned without details in 13 trials (graded 'B').
Allocation was concealed in 16 trials (graded 'A') and was not reported on in the remaining 6 (graded 'B').
Inclusion of all randomized participants
In 16 trials losses to follow up were less than 10%, and in 6 trials they were not reported but likely to be below 10%.
Four trials blinded the investigator and the trial participants to impregnation, and they did this by using dummy preparations for dipping the nets.
Effects of interventions
Child mortality from all causes
Five cluster randomized controlled trials examined child mortality from all causes (Appendix 3). They were all conducted in areas with stable malaria in sub-Saharan Africa: (Burkina Faso (Habluetzel); Gambia (D'Alessandro); Ghana (Binka); Kenya (Nevill); Kenya (Phillips-Howard)). Four of the trials did not use any nets as the control group, and one trial used untreated nets. Both the relative and the absolute impact were analysed.
When the five trials were pooled regardless of the type of control group, the summary relative rate was 0.82 (95% CI: 0.76 to 0.89; Analysis 1.1), giving a summary protective efficacy of 18%. The chi
A regression analysis of the protective efficacy (ln) on the transmission intensity (as measured by the entomological inoculation rate: 10 Gambia (D'Alessandro), 30 Kenya (Nevill), 300 Ghana (Binka), 300 Kenya (Phillips-Howard), 500 Burkina Faso (Habluetzel)) was statistically significant at the 5% level (r
It was possible to summarize the rate difference because the trials used similar methods and a similar denominator for their rate calculations (person-years at risk). Each trial corrected the confidence limits in their analysis to take into account cluster allocation (see Appendix 3). Four trials showed a statistically significant effect, and the direction of effect in the fifth trial favoured treated nets.
The summary rate difference, which expresses how many lives can be saved for every 1000 children protected, was 5.53 deaths averted per 1000 children protected per year (95% CI 3.39 to 7.67; Analysis 1.2). I performed a regression analysis of the natural logarithm of the rate difference on the entomological inoculation rate and could not find a trend (r
Stratified by type of control group
There was a small non-statistically significant difference in the summary results of protective efficacy in the two comparisons − controls with no nets versus controls with untreated nets: 17% versus 23% reduction in mortality. The summary rate differences in the two comparison groups were virtually identical (5.5 versus 5.6 averted deaths per 1000 per year).
Controls without nets (4 trials)
The summary rate ratio was 0.83 (95% CI 0.76 to 0.90; Analysis 1.1), or a protective efficacy of 17%. In other words, overall mortality was reduced by 17% among children aged 1 to 59 months. The chi
The risk difference was 5.52 per 1000 protected children per year (95% CI 3.16 to 7.88; Analysis 1.2).
Controls with untreated nets (1 trial)
The summary rate ratio was 0.77 (95% CI 0.63 to 0.95; Analysis 1.1), or a protective efficacy of 23%. The risk difference was 5.60 deaths per 1000 protected children per year (95% CI 0.50 to 10.70; Analysis 1.2).
Malaria-specific child mortality
The impact of ITNs on malaria-specific death rates was looked at only briefly because of the problems using verbal autopsies in determining malaria deaths. In the two trials for which the data were available, the percentage reduction in malaria-specific mortality was similar or smaller than the percentage reduction in all-cause mortality: 14% (versus 23%) for Gambia (D'Alessandro), and 22% (versus 18%) for Ghana (Binka). One interpretation is that malaria-specific death rates were not reflecting the true impact of ITNs on mortality (since a much higher specific impact would have been expected).
Only one trial examined severe malarial disease as an outcome Kenya (Nevill). The trial used passive and hospital-based case ascertainment, and observed a 45% (cluster-adjusted 95% CI 20 to 63) reduction in the frequency of severe malaria episodes following the introduction of ITNs (Appendix 4).
Uncomplicated clinical episodes
The trial results are available in Appendix 5 for no nets controls and in Appendix 6 for untreated nets controls. A summary of the main findings for protective efficacies is available in Appendix 7; confidence intervals were not calculated as this analysis includes both cluster and individually randomized controlled trials. No risk or rate differences were calculated because the denominators were not uniform and the sensitivity of the reporting systems of the different trials is likely to have varied considerably. Three findings can be highlighted.
- The effect of ITNs on uncomplicated clinical episodes of malaria is shown by large effect estimates in all trials. Overall, the reduction in clinical episodes was around 50% for all subgroups (stable and unstable malaria; no nets and untreated nets) and for both P. falciparum and P. vivax.
- The protective efficacy is higher (at least 11% for P. falciparum) when the control group had no nets. This was expected and it was the reason to create two separate comparisons. In areas with stable malaria (entomological inoculation rate > 1) the differences in protective efficacies against uncomplicated malaria was 11% (50% no nets versus 39% untreated nets). In areas with unstable malaria (entomological inoculation rate < 1), the differences were bigger: 23% (62% no nets versus 39% untreated nets) for P. falciparum, and 41% (52% no nets versus 11% untreated nets) for P. vivax.
- In areas of unstable malaria (entomological inoculation rate < 1), the impact against P. falciparum episodes seemed to be higher than the impact against P. vivax episodes.
The results are available in Appendix 8 for no nets and in Appendix 9 for untreated nets controls. The results for both groups are summarized in Appendix 10; confidence intervals were not calculated as this analysis includes both cluster and individually randomized controlled trials. Two points can be highlighted from these results.
- In areas of stable malaria, impact on prevalence of infection (measured through cross-sectional surveys) was small: 13% reduction when the control group did not have any nets and 10% reduction when the control group had untreated nets.
- In areas with unstable malaria, the results are of limited value because there was only a single trial in each subgroup (treated versus no nets; and treated versus untreated nets).
The results are shown in Appendix 11 for no nets and Appendix 12 for untreated nets controls. This outcome was only assessed for trials in areas of stable malaria, where parasitaemia does not necessarily lead to a clinical episode, and where parasitaemia cut-offs are useful to define disease episodes. Five trials measured this outcome: four used 5000 trophozoites/ml as the cut-off, while the fifth trial used an age-specific cut-off (Kenya (Phillips-Howard)). The protective efficacy was 29% for the two trials in which the control group did not have nets, and was 20% for the three trials in which controls had untreated nets.
The nine trials that measured anaemia were conducted in areas of stable malaria; six trials compared treated to untreated nets (Appendix 13), and three trials compared treated nets to untreated nets (Appendix 14).
Overall, the packed cell volume of children in the ITN group was higher by 1.7 absolute packed cell volume per cent compared to children not using nets. When the control group used untreated nets, the difference was 0.4 absolute packed cell volume per cent.
Prevalence of splenomegaly was defined as the prevalence rate of children with at least a degree '1' of spleen enlargement on the Hackett's scale. Together with overall mortality it was the only outcome to be properly standardized between the sites (although inter-observer variability can be substantial).
Four out of the five trials that measured splenomegaly were carried out in areas with stable malaria (Appendix 15 and Appendix 16). Because the exception was one trial carried out in Thailand whose weight is very small (only 2.6% in the relevant comparison) (Thailand (Luxemberger)), I did not carry out a subgroup analysis.
Splenomegaly was significantly reduced for both types of controls: there is a 30% protective efficacy when controls were not using nets, and a 23% protective efficacy when the control group used untreated nets.
Three trials carried out with ITNs have demonstrated a positive impact on anthropological measurements in children sleeping under treated nets.
In The Gambia (Gambia (D'Alessandro)), mean z-scores of weight-for-age and weight-for-height were higher in children from treated villages (-1.36 and -0.98, respectively) than in those from untreated villages (-1.46 and -1.13, respectively). The differences were statistically significant after adjustment for area, age, differential bed net use, and gender (P = 0.008 and P = 0.001, respectively). There was no statistically significant difference in mean z-scores for height-for-age.
In the trial carried out in Kenya (Kenya (Nevill)), infants sleeping under ITNs in the intervention areas had statistically significantly higher z-scores for weight-for-age than control infants not under treated nets (analysis of variance allowing for season, gender, and age: F = 21.63, P = 0.03). Mean mid-upper arm circumference z-scores were also statistically significantly higher among infants in the intervention communities (analysis of variance allowing for survey, gender, and age: F = 19.0, P = 0.005) (Snow 1997).
In Kenya (Kenya (Phillips-Howard)), protected children under two years of age had a statistically significantly better weight-for-age z-score than unprotected children (P < 0.04). No other statistically significant differences were measured for other parameters or other age groups, although all z-score differences between intervention and control groups were in favour of the protected group.
A large number of trials with insecticide-treated bed nets or curtains has been carried out all over the world. We identified 81 trials investigating insecticide-treated mosquito nets or curtains. The 22 trials meeting this review's inclusion criteria span 17 countries. Five of these trials measured mortality, and they showed that the use of ITNs reduces under five mortality in malaria-endemic areas in sub-Saharan Africa by about a fifth.
More trials examined morbidity, and showed an impact of ITNs nets on illness, and on both P. falciparum and P. vivax infections.
The impact on overall mortality
The relative decrease in mortality (as given by the protective efficacy) afforded by ITNs seemed to be lower in areas with high malaria transmission (entomological inoculation rate > 100) than in areas with a lower transmission rate. However, this was not reflected in terms of absolute risk reduction: the estimated numbers of lives saved per 1000 protected children were similar in all the areas (5.5 lives saved per 1000 children protected per year). With a high coverage of treated nets over two-year period, the benefit of ITNs in terms of lives saved per unit of investment was high in the five trial areas in which overall mortality was measured as outcome.
An approximate extrapolation to the current population of children under five years of age at risk for malaria in sub-Saharan Africa (14% of approximately 480 million population at risk, or 67 million children) indicates that approximately 370,000 child deaths could be avoided if every child could be protected by an ITN.
A cost-effectiveness assessment has shown that ITN programmes compare well in terms of cost-effectiveness with other child survival interventions such as the Expanded Programme on Immunization (EPI) (Goodman 1999).
The impact on morbidity
The impact of ITNs on uncomplicated episodes of malaria is also marked with a halving of episodes under most transmission conditions (stable and unstable malaria). If these results are sustained in large-scale implementation, then ITN programmes could lead to substantial savings both at the healthcare level and at the household level, where the cost of disease episodes is considerable (Sauerborn 1995).
The one trial that demonstrated a substantial impact on severe malaria disease provided evidence that ITNs can have an impact on preventing severe illness and the associated high costs to both patients and healthcare providers (Kenya (Nevill)).
The finding that ITNs improve the haemoglobin level in African children by 1.7% packed cell volume also has important public health implications.
ITNs have a benefit on growth in children too, although these effects appear to be modest.
ITN impact in trials versus programmes
The results presented in this review are from randomized controlled trials where the intervention was deployed under highly controlled conditions, leading to high coverage and use rates. The one exception is Gambia (D'Alessandro), which was a randomized evaluation of a national ITN programme in which the intervention deployment was not as good as in the other trials. Therefore, the bulk of data in this review describe impact under ideal trial conditions (efficacy) rather than impact under large-scale programme conditions (effectiveness). While the difference between efficacy and effectiveness is likely to be small for certain medical interventions (such as vaccination or surgery), it can potentially be large for preventive interventions such as ITNs.
Some of the consequences of moving from a scientific trial towards a large-scale programme is illustrated by the results of the two mortality trials carried out in The Gambia. The first trial was carried out under well-controlled implementation conditions, with a high coverage rate in the target population (Gambia (Alonso)). Unfortunately it was not randomized and hence not included in the present analysis. The second one was the evaluation of a national impregnation programme carried out by primary health care personnel and which faced some operational problems (leading, for example, to a lower than expected insecticide dosage) and a lower coverage rate (around 60%) of the target population (Gambia (D'Alessandro)). The difference of impact between the two studies is important: the first trial achieved a total reduction in mortality of 42%, while the protective efficacy in the second trial was 23%. It is not clear whether the difference in the baseline mortality rate (42.1 versus 24.3 deaths per 1000 in the control group) played a role in this difference of impact.
Unfortunately, randomization is unlikely to be a feasible option for evaluating most programmes. Impact assessment methodology is not optimal and research is still needed in this area (Lengeler 1996b). Recently, a number of evaluations of small-scale and large-scale programmes have documented good impact on different health parameters (Abdulla 2001; D'Alessandro 1997b; McClean 2002; Rowland 1997; Schellenberg 2001). Most notably, the evaluation of a large social marketing programme in Tanzania showed a 27% improvement in survival in ITN users compared to non-users (Schellenberg 2001) and a substantial (63%) impact on anaemia in children (Abdulla 2001).
A related aspect of programme monitoring is the question of how impact varies with the coverage rate. Especially under high transmission conditions, maximum impact might well be obtained only if a certain level of coverage is achieved and if a substantial part of the mosquito population is killed as a result. Such a "mass effect" has been detected in some trials and not in others, but it is likely that if it is present the impact of ITNs will be enhanced (Lines 1992). Recently, a series of studies have clearly documented a "mass effect" on malaria morbidity (Howard 2000) and especially on child mortality (Binka 1998; Hawley 2003). In Ghana and western Kenya, children living in control areas but within a few hundred meters of an intervention cluster experienced the same reductions in mortality as children in the intervention areas. Since such a "mass effect" is very likely to occur before 100% coverage is achieved, this has potentially important consequences for equity: poorer segments of the population unable to afford an ITN might well benefit from the ITNs used by their better-off neighbours.
Short-term versus long-term benefits
The results from the large-scale ITN trials have re-activated a discussion that has been central in malaria control since the 1950s: does reducing exposure to malaria in areas of very high transmission intensity lead to a long-term gain in mortality or merely to a delay in the time of death? For this review, the relevant question is whether the short-term benefits of ITNs, as seen in trials lasting one to two years, will result in a long-term survival benefit of the protected children.
Different researchers have hypothesized that where malaria transmission is particularly high, the benefits of ITNs will be transitory, and that morbidity and mortality may only be postponed to an older age as a result of preventing the natural development of immunity to malaria that occurs through repeated exposure (Lines 1992; Snow 1994; Snow 1995; Snow 1997; Trape 1996). This does have obvious serious implications for decision-making, and this view has been discussed and sometimes challenged by a number of other authors (D'Alessandro 1997a; Greenwood 1997; Lengeler 1995; Lengeler 1997b; Lines 1997; Molineaux 1997; Shiff 1997; Smith 2001). Despite ongoing disagreements on this question among researchers, there is at least one point on which there is consensus: if such a delay in mortality exists it will only occur in very high transmission areas (a commonly quoted cut-off entomological inoculation rate is 100, although this is at present based on little evidence).
Unfortunately, there is little evidence for or against such a delayed mortality effect following interventions that potentially interfere with the development of natural immunity. The best information comes from two five-year follow-up studies of large ITN trials in Burkina Faso (Diallo 2004) and Ghana (Binka 2002). In both trials the overall survival of children who had slept since birth under an ITN was significantly better than for children who had only received ITNs at the end of the trial. The major implication of these findings is that such a "delayed mortality effect" does not seem to exist, but more studies are needed before this can be proven beyond doubt.
Certainly, stopping or delaying ITN programme implementation because of this fear is not warranted and should even be considered unethical in the light of good evidence of benefit. However, it is important that ITN programmes carried out in areas of high transmission have a well-designed mortality monitoring component alongside implementation.
Comparisons of insecticide-treated nets and indoor residual spraying for malaria control
A number of studies in recent years have compared the implementation of ITNs with the application of indoor residual spraying, the other large-scale vector control intervention. While there have been some arguments about which method is the most efficacious, effective, and cost-effective, the views vary, and some people consider that they are equivalent (Lengeler 2003).
People in malaria endemic areas primarily use bed nets and curtains as a protection against nuisance biting, rather than as a malaria control measure (Zimicki 1996). Since most malarious areas also have a perceived mosquito nuisance problem, treated nets have proved very popular and large-scale trials had few problems in achieving rapidly high coverage rates and maintain high usage rates for up to three years. Unfortunately, re-treatment of existing nets has proved a much bigger challenge. It is expected that the development of nets with a long-lasting insecticide treatment will offer a solution to this problem.
With the inclusion of ITNs as one of the main strategies for preventing malaria by the Roll Back Malaria partnership, large-scale programmes have started to be implemented in a number of countries. Recently, Roll Back Malaria has developed a global strategy for the up scaling of ITN programmes (RBM 2002), which included a focus on developing of a commercial market for ITNs, as well as additional mechanisms to protect those at highest risk, essentially children and pregnant women. One book chapter has dealt with some of the key operational issues to consider (Feilden 1996), and at least two manuals aimed at national and district level personnel involved in malaria control have been produced (Chavasse 1999; RBM 2003).
The high proportion of trials that could not be included in the primary review (59 out of the 81 identified trials) is a cause for concern. The main reasons for exclusion were because the studies were not randomized, were not adequately controlled (before and after assessments), and used materials other than bed nets or curtains (such as wall curtains or blankets).
Randomization is important in any intervention study to avoid the investigator's preferences from biasing the results. However, randomization is not always possible, especially if the intervention is considered to be very beneficial. An alternative design can then be required by the ethical review committee, as was the case for the first Gambian trial (Gambia (Alonso)).
Equally important is the fact that potential investigators wanting to test preventive measures that are applied at a group level (for example, at the village level) choose a sufficient number of units to make comparisons meaningful. It is clear that a 1:1 design (one intervention village versus one control village) should not be done because it is highly likely that the two groups will not be comparable at baseline. An absolute minimum of randomization units is six (that is, 3:3), but 10 units would be much better.
Finally, some of the cluster randomized controlled trials presented confidence intervals as if allocation had been on an individual level, described by Cornfield as "an exercise in self-deception" (Cornfield 1978). Trialists, statisticians, and journal editors need to get together to address this widespread problem in trial analysis and publication; and statisticians working in meta-analysis could also help to tackle this problem.
Implications for practice
Five randomized controlled trials have provided strong evidence that the widespread use of ITNs can reduce overall mortality by about a fifth in Africa. For every 1000 children protected, on average about 5.5 lives can be saved in children aged 1 to 59 months every year. In Africa, full ITN coverage could prevent 370,000 child deaths per year.
The impact of ITN use on clinical episodes of uncomplicated malaria is also considerable, halving clinical attacks in areas of stable malaria transmission in Africa. One trial in Kenya further documented a substantial impact of ITN use on cases of severe malaria disease seen in hospital. In Asia and Latin America (areas with low malaria transmission, entomological inoculation rate < 1), the use of ITNs also significantly reduced the number of clinical episodes due to both P. falciparum and P. vivax.
Given the strength of this evidence there is a need to promote the large-scale application of this control tool in the frame of malaria control programmes in endemic areas. The Roll Back Malaria partnership and major international health donors have endorsed this view (WHO 2003).
Because of the lack of data on the long-term impact of ITNs in areas with very high malaria transmission (entomological inoculation rate > 100), a careful monitoring of impact on child survival should be conducted in at least a few sites to provide more data. This consideration is currently not a reason to halt the implementation of ITN programmes.
Implications for research
The beneficial impact of ITNs has been largely demonstrated under trial conditions. Given the consistency of the impact results for different outcomes and different areas of the world, it is unlikely that many more trial data are required. However, four major issues regarding impact assessment remain.
In relation to trial reports, researchers and editors need to ensure confidence limits are correctly calculated for cluster randomized controlled trials and that adjusted standard errors are always reported; and meta-analysis specialists could usefully examine how data from cluster randomized controlled trials can be combined.
Christian Lengeler acknowledges and thanks the following organizations and people who have helped make this review possible:
The Swiss National Science Foundation (via a PROSPER grant to the reviewer), the UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases (TDR), and the Swiss Tropical Institute, Basel, Switzerland, for financial support.
Steve Bennett, Simon Cousens, and Linda Williams for statistical support and for providing cluster-corrected results for the mortality trials.
Gerd Antes and Daniel Galandi, University of Freiburg, Germany for independently applying the inclusion criteria.
Pedro Alonso and Simon Cousens for constructive comments on earlier drafts of this review.
Paul Garner, Harriet G MacLehose, and staff at the Cochrane Infectious Diseases Group at the Liverpool School of Tropical Medicine for help in editing the review at various stages.
Many thanks to Fred Binka, Chris Curtis, Umberto D'Alessandro, Fulvio Esposito, Pierre Guillet, Annette Habluetzel, Marie-Claire Henry, Feiko terKuile, Axel Kroeger, Jo Lines, Chris Nevill, Marbiah Nuahn, Patrick Rabarison, Indra Vythilingam, Jaco Voorham, Morteza Zaim, and Robert Zimmerman for supplying additional data.
Bayer, Mitsui, Sumitomo, and AgrEvo, insecticide manufacturers, for providing additional trial data.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Index terms
Appendix 1. Search methods: search strategies for databases
Appendix 2. Randomization and outcomes
Appendix 3. Child mortality from all causes
Appendix 4. Severe disease
Appendix 5. Treated nets versus no nets: Prevention of uncomplicated clinical episodes
Appendix 6. Treated versus untreated nets: Prevention of uncomplicated clinical episodes
Appendix 7. Summary: Prevention of uncomplicated clinical episodes
Appendix 8. Treated nets versus no nets: Parasite prevalence (any infection)
Appendix 9. Treated versus untreated nets: Parasite prevalence (any infection)
Appendix 10. Summary: Parasite prevalence
Appendix 11. Treated nets versus no nets: High parasitaemia
Appendix 12. Treated versus untreated nets: High parasitaemia
Appendix 13. Treated nets versus no nets: Anaemia
Appendix 14. Treated versus untreated nets: Anaemia
Appendix 15. Treated versus no nets: Splenomegaly (Hackett's scale 1 to 5)
Appendix 16. Treated versus untreated nets: Splenomegaly (Hackett's scale 1 to 5)
Last assessed as up-to-date: 18 January 2004.
Protocol first published: Issue 1, 1995
Review first published: Issue 3, 1998
Contributions of authors
Christian Lengeler is the sole contributor.
Declarations of interest
Sources of support
- Swiss Tropical Institute, Basel, Switzerland.
- Department for International Development, UK.
- Swiss National Science Foundation, Bern, Switzerland.
- UNDP/WB/WHO Special Programme for Research and Training in Tropical Diseases, Switzerland.
Medical Subject Headings (MeSH)
*Bedding and Linens; Insecticides [*administration & dosage]; Malaria [*prevention & control]; Malaria, Falciparum [prevention & control]; Malaria, Vivax [prevention & control]; Mosquito Control [*methods]; Randomized Controlled Trials as Topic
MeSH check words
Female; Humans; Male; Pregnancy
* Indicates the major publication for the study