| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
 |
ABSTRACT |
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
17% (plausible values range from 12% to 25%). Second, if malaria interventions are scaled-up from very low (2%) to reasonably high coverage levels (70%), epidemiologic models predict that ACCM would decrease by 17% or 18% (precision estimate: 15–19%). These results suggest that if malaria interventions are scaled-up to reach or exceed 70% coverage, it is plausible that a goal of reducing malaria mortality by 50% could be achieved. It is also likely that a pair of "typical"-size mortality surveys could detect this ACCM change as being statistically significant. Although existing models have important limitations, they could be improved by incorporating empirical results during scale-up of multiple interventions and by adding precision estimates and sensitivity analyses. |
INTRODUCTION |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
One evaluation method recommended repeatedly over the past decade is to monitor trends in all-cause childhood mortality (ACCM) for ages 0–4 years, along with trends in malaria intervention coverage and indicators of malaria morbidity and transmission.5,8–15 The main justifications are that malaria mortality is overwhelmingly concentrated in young children in SSA, and thus ACCM is sensitive to changes in malaria-associated mortality16–18; that ACCM trends capture changes in direct and indirect malaria mortality14; ACCM estimates are available for all countries in SSA14,19; and ACCM can be measured reliably (i.e., it does not suffer from limitations of methods, such as verbal autopsies for identifying malaria deaths20,21).
ACCM, however, is not a malaria-specific measure. Therefore, a fundamental question is: What is the quantitative relationship between ACCM and malaria mortality? In the context of RBM scale-up efforts, this question can be reformulated to address two practical concerns. First, how much does ACCM decrease when a goal to reduce malaria mortality (e.g., by 50%) is achieved? Second, as it is not known exactly what coverage of malaria interventions is needed to halve malaria mortality, how much would ACCM decrease if intervention coverage increased from low to reasonably high levels? Our aim is to explore these two questions for populations in SSA with a high malaria burden. Our predictions could be useful for interpreting ACCM trends as malaria interventions are scaled-up (and knowing whether it is plausible that a malaria mortality reduction target has been reached), and for calculating sample size for household surveys that measure ACCM. In the course of addressing our second question, we also explore 2 published epidemiologic models that predict mortality reductions as a function of increases in the coverage of malaria interventions.22,23 We discuss the strengths and limitations of these models and consider how they could be improved.
|
METHODS |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Estimates of direct malaria mortality were based on a recent analysis for SSA for the year 2000 (Box 1
).24 As scale-up efforts have been generally slow over the past 7 years, estimates from this analysis still might be reasonably valid in 2007 for the proportion of deaths attributable to malaria, and the proportion of the population in various malaria risk groups. Moreover, 2000 is the baseline year for RBM’s goal of halving malaria’s burden.
Box 1 Summary of methods for estimating deaths directly attributable to malaria among children 0–4 years old in middle Africa for the year 200024
|
|
footnotes). These countries were categorized into two groups: Group 1 countries, in which a large majority of the population (average of 91%) lives in areas with high-intensity malaria transmission, and Group 2 countries, in which a much lower proportion (average of 28%) live in high-transmission areas (Table 1). This categorization was done because results for Group 1 countries best reflect the high-transmission areas where a large majority (94%24) of direct malaria deaths are thought to occur. For Group 1 countries, we assumed 22% (precision estimate: 20–25%) of deaths were directly attributable to malaria. Note that Table 1 shows numbers to the tenths place, so our calculations can be followed more easily; however, to avoid having predictions appear more precise than they truly are, we present main results rounded to the ones place.
Assumptions about the burden of indirect malaria mortality in children < 5 years old.
The burden of indirect malaria mortality is far more difficult to estimate. Some evidence suggests the burden could be as high as that of direct malaria mortality. First, in the 1950s to the 1970s, studies in high-transmission areas of Kenya, Tanzania, and Nigeria found that when malaria transmission was drastically reduced with IRS, which would have had little impact on nonmalaria mortality, ACCM decreased 40–50%.28–30 If malaria directly caused 22% of deaths (Table 1, Group 1, row 7), then malaria’s indirect effects would have caused the remaining 18–28% of deaths. Thus, direct and indirect malaria mortality might be roughly similar. Second, an ITN trial in The Gambia from the early 1990s found that for children 1–9 years old, ITNs reduced ACCM by 2.2 deaths per 1000 children per year, direct malaria mortality by 0.5 death/1000/year, and nonmalaria-nontrauma mortality by 1.2 deaths/1000/year.31 At least 20% of deaths had unknown causes, so it is difficult to make a firm estimate of the indirect malaria mortality burden; however, these results do suggest that the indirect burden might be as large as (or greater than) the direct burden. Third, a meta-analysis of the relationship between the prevalence of Plasmodium falciparum parasitemia and ACCM in African demographic surveillance systems found that, as parasite prevalence decreased from 95% to 0%, the predicted ACCM decreased from 44 to 8 deaths/1000/year, a reduction of 82%.32 If malaria directly caused 22% of deaths, then the indirect burden would be substantially greater than the direct burden. The authors, however, recognize that associations from their model might not be causal. Additionally, the model did not include other factors that influence child mortality (immunization coverage, nutritional status, etc.), and thus the association between parasite prevalence and mortality might have been overestimated. Finally, another meta-analysis of infant mortality and malaria transmission intensity in 12 African sites estimated that for children < 5 years old, indirect malaria mortality rates were similar or somewhat greater than direct malaria mortality rates. At entomological inoculation rates of 5 and 100 infective mosquito bites per person per year, the indirect/direct mortality rate ratios were 1.1 and 1.4, respectively (Ross and others33; personal communication from T. Smith, Swiss Tropical Institute, September 28, 2006).
In contrast, results of 2 trials suggest the burden of indirect malaria mortality might be more modest. A study from The Gambia found that when ITNs and chemoprophylaxis were used to prevent malaria in children, ACCM decreased by 42%.18 In the study setting, the percentage of deaths directly attributable to malaria was 25%; so, a minimum estimate of the burden of indirect malaria mortality (assuming ITNs plus chemoprophylaxis prevented all direct malaria mortality, an assumption that leads to a conservative estimate of the indirect burden) is that indirect malaria mortality might be about two-thirds of direct malaria mortality. In Ghana, a trial found that ITNs reduced direct malaria mortality by 2.1 deaths/1000/year and nonmalaria mortality by 1.0/1000/year,34 which suggests that the indirect burden might be about half the direct burden.
A fundamental problem with using the above evidence to estimate how much indirect malaria mortality would be prevented if the coverage of malaria interventions were increased is that indirect malaria deaths can be prevented in at least 2 ways: prevent malaria or prevent the other comorbid disease(s). As there has been renewed interest in preventing child deaths from all causes, it is likely that public-health activities affecting nonmalaria mortality will prevent some of the indirect malaria deaths. Therefore, to be conservative, we assumed the burden of indirect malaria mortality was half the direct burden (i.e., for every 2 direct malaria deaths, there is 1 indirect malaria death). For a sensitivity analysis, the lower limit of the indirect burden was somewhat arbitrarily chosen as one-quarter of the direct burden (i.e., half-way between no effect, which we thought was implausible, and our base assumption of 0.5), and the upper limit of the indirect burden was chosen to be equal to the direct burden.
Assumptions about trends in nonmalaria mortality. Clearly, when predicting reductions in ACCM for a given decrease in malaria mortality, one must consider mortality trends for conditions unrelated to malaria. For simplicity, we assume no net change, which could reflect either static trends or dynamic counter-balancing trends in all nonmalaria causes.
Assumptions about the impact of malaria interventions on nonmalaria mortality. We assumed that malaria interventions have no impact on nonmalaria mortality. In fact, malaria control implementation efforts that strengthen health systems might reduce nonmalaria mortality. For example, improving drug supplies and health worker performance for the case management of malaria might lead to improvements in the care-seeking for and management of other childhood illnesses. For simplicity, however, no impact was assumed.
Predicted reductions in ACCM for a given decrease in malaria mortality. Given an estimate of the total malaria-associated mortality burden (i.e., direct plus indirect malaria mortality) as a proportion of all deaths and the assumptions that malaria control efforts have no impact on nonmalaria mortality and that nonmalaria mortality does not change over the time period of interest, the predicted reduction in ACCM equals the total malaria mortality burden times the malaria mortality reduction. For example, if one-third of deaths were caused by malaria at baseline and malaria mortality decreases by 50%, then ACCM would decrease by one-sixth (i.e., 1/3 x 50%).
Predicted reductions in ACCM as malaria interventions are scaled-up.
Predictions of ACCM reductions that would occur if ITNs, PECM, and IPTp were scaled-up from 2% to 70% were obtained for published models by Jones and others22 (Box 2) and Morel and others.23 The models are conceptually similar, and we illustrate the method with details from the Jones model. First, assumptions are made about the protective efficacy (Ef) of each malaria intervention and the proportion of child deaths attributed to the direct effects of malaria. For example, in the Jones model, among children 1–59 months old, the Ef values of ITNs and PECM are 0.75 and 0.67, respectively (personal communication, G. Jones, UNICEF, February 13, 2006). Second, the pre-scale-up coverage (pc) and post-scale-up coverage (pt) of each intervention is measured or assumed (with ITN "coverage" meaning ITN use). For our analysis, pc = 2% and pt = 70% for both interventions. Third, the proportion of malaria deaths averted is calculated for each intervention in isolation (i.e., ignoring all other interventions) with the expression: [(pt – pc) x Ef]/[1 – (pc x Ef)]. For example, the proportion of malaria deaths averted by ITNs equals [(0.70 – 0.02) x 0.75]/[1 – (0.02 x 0.75)], or 0.52. Similarly, the proportion for PECM is 0.46. Fourth, interventions are arbitrarily ordered (e.g., ITNs first, PECM second). The specific order has no bearing on the final result but is needed to implement the method. Fifth, the number of malaria deaths averted for each intervention is calculated according to its place in the ordered list. For the first intervention, the number of averted malaria deaths equals the total number of direct malaria deaths multiplied by the proportion of malaria deaths averted by the intervention. For example, if there were a total of 100,000 direct malaria deaths, the number of malaria deaths prevented by scaling-up ITNs would be 100,000 x 0.52, or 52,000. For the second ordered intervention, deaths prevented by the first intervention must be subtracted from the total malaria deaths, and the proportional reduction of the second intervention is applied to the remaining deaths. To continue the example, if ITNs prevented 52,000 deaths, then there are only 48,000 deaths (i.e., 100,000 – 52,000) that can be prevented by PECM; and the number of deaths prevented by PECM equals 48,000 x 0.46, or 22,080. This subtraction avoids double-counting averted deaths. Thus, after scaling-up ITNs and PECM from 2% to 70%, the model predicts 74,080 deaths (52,000 + 22,080) averted, or 74% of the original 100,000 deaths. The last steps are repeated for all other ordered interventions. Finally, the proportional reduction in ACCM equals the proportional reduction in malaria mortality times the proportion of all child deaths directly attributable to malaria.
| Box 2 Predicted reductions in all-cause mortality for children 0–4 years old that would occur if ITNs, IPTp, and PCEM were scaled-up from 2% to 70%, based on the model by Jones and others22 Assumptions
Results
|
|
of Ref. 22); 2) regarding PECM, use of artemisinin-based combination therapy was rare in 2000 and other commonly used antimalarials (e.g., chloroquine) were rapidly losing efficacy due to resistance; 3) scale-up of malaria interventions has been slow in most countries over the past 7 years; and 4) year 2000 estimates reflect coverage for the reference year for RBM’s burden reduction goal. We chose a value of 70% for pt because it seemed a target that could be realistically achieved by 2010 (the year for reaching RBM’s goal). Additionally, as 70% corresponds to the approximate level of ITN use in the 5 rigorous trials of ITN efficacy in Africa,16 we could then use trial results to examine the validity of the prediction models.
Regarding the assumed burden of direct malaria mortality, each model uses a slightly different estimate. Jones and others use the estimate of 23.7%,35 and Morel and others use estimates of 23.4%.36 For better comparability with other results we present, and to make a simplistic calculation of uncertainty, we repeated our predictions using the Jones model with the small modification of assuming the malaria mortality burden of Group 1 countries (Box 2).
Assumptions about measurement of malaria intervention coverage. As previously mentioned, a key input for the prediction models is coverage of malaria interventions. Perfectly valid and precise measurements of coverage are assumed; however, it important to note that standard measurement tools have limitations.37 For example, standard questions on ITN usage do not account for holes in nets, whether insecticide has been washed away, or insecticide resistance; standard questions on PECM do not account for correct dosing and drug quality and only partially capture information on drug adherence.
|
RESULTS |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
, row 8) are indirect malaria deaths, then 34% (22.4% + 11.2%, rounded) of all deaths are malaria-associated (Table 1, row 9). If interventions are implemented and malaria-associated mortality is reduced by 50% and nonmalaria mortality is unchanged, then in the post-scale-up period, relative to baseline, ACCM would decrease by 17% (i.e., 50% of 33.6%, rounded) (analysis A of Figure 1). Limits of a simple precision estimate for this ACCM reduction are 15–19% (i.e., 50% reduction of precision estimate limits of the percentage of all child deaths that are malaria associated [29.7–37.4%], rounded). For the sensitivity analysis to predict minimum plausible ACCM reductions, we assumed indirect malaria mortality was one-quarter of direct malaria mortality. If malaria-associated mortality decreases by 50%, the predicted ACCM reduction is 14% (precision estimate: 12–16%). For maximum reductions, assuming indirect mortality equals direct mortality, the predicted ACCM reduction is 22% (precision estimate: 20–25%).
Thus far, we have assumed that malaria mortality decreases by 50%; however, it is easy to predict ACCM reductions for other levels of malaria mortality reduction.
Predicted reductions in ACCM as malaria interventions are scaled-up.
If ITNs, IPTp, and PECM coverage increased from 2% to 70% and malaria mortality estimates by Morris and others were used, the Jones model predicted that ACCM would decrease by 18% (i.e., 17.5% from the last row in Table 2, rounded [no precision estimate]) (Box 2; analysis D of Figure 1). For better comparability with our previous results, when malaria mortality estimates for Group 1 countries were used, the Jones model predicted that ACCM would decrease by 17% (our precision estimate: 15–19%) (analysis E of Figure 1). When the same scale-up is assumed, the Morel model predicted that ACCM would decrease by 17% (no precision estimate) (Table 2; analysis F of Figure 1).
|
|
DISCUSSION |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
17% (left side of Figure 1). Depending on the burden of indirect malaria mortality, however, plausible values range between 12% and 25%. For the scenario in which malaria mortality is reduced by 50%, most of the range of the predicted ACCM decrease exceeds 15%. Therefore, a pair of mortality surveys (e.g., Demographic and Health Surveys [http://www.measuredhs.com]) with a "typical" sample size of 4000–8000 women is likely to have the statistical power to detect such a decrease.39 We recognize that assumptions in these calculations are simplistic and that the limited evidence does not permit accurate predictions for a specific place and time. However, our results do reflect how much ACCM could decrease when malaria mortality is reduced by 50% for a plausible set of assumptions. Furthermore, the assumptions illustrate critical knowledge gaps and potential pitfalls in the interpretation of changes in ACCM. For example, in the absence of valid information on trends in cause-specific mortality (unfortunately, the norm in most of SSA), increases in diarrhea or pneumonia mortality could offset reductions in malaria-associated mortality, leaving ACCM unchanged despite the successful scale-up of efficacious malaria interventions.
To answer our second question (how much would ACCM decrease if intervention coverage were increased?), if malaria interventions are scaled-up from 2% to 70%, ACCM is predicted to decrease by 17% or 18% (right side of Figure 1). Although the predictions in Figure 1 vary substantially, the similar ranges on the left and right side suggest that malaria mortality could be reduced by 50% if malaria interventions were scaled-up to 70%.
Some evidence suggests that the models might somewhat underestimate the impact of ITNs and IPTp (Table 2, rows 11 and 12). For ITNs, one can demonstrate the underestimation empirically. A meta-analysis of 5 high-quality trials found that if ITN use is scaled-up to 70% (in settings where ITNs are distributed to cover sleeping spaces for all ages), ACCM in 1-to 59-month-old children decreased by 17% (95% confidence interval: 10–24%).16 Assuming that 26% of deaths among children 0–59 months of age occur in the first month of life (Table 1, last row) and that ITN use by neonates does not reduce neonatal mortality, then scaling up ITN use to 70% should decrease ACCM in children 0–59 months of age by 13% (i.e., 17% x [100 – 26%]). For this age group, the Jones model predicts a slightly lower reduction of 12%, and the Morel model predicts an even lower reduction of 10%.
Regarding IPTp, although it is more difficult to estimate its impact, a review by Guyatt and Snow predicted that scaling-up to 100% coverage would prevent 80,000 child deaths.40 At 70% coverage in Group 1 and Group 2 countries, which includes most of SSA with stable malaria transmission, IPTp would reduce ACCM by 1.3% (i.e., [80,000 deaths averted x 70% coverage]/4.3 million all-cause deaths in children 0–59 months old). Both the Jones model and the Morel model predicted slightly lower ACCM reductions (0.3% and 0.2%, respectively).
When comparing models, however, it should be noted that all these predictions are based on imprecise estimates and numerous assumptions of unknown validity. Therefore, small differences among models are probably unimportant. Indeed the large degree of uncertainty with these models underscores the importance of presenting (even simplistic) precision estimates and sensitivity analyses to avoid conveying a false sense of validity and precision.
These models have other limitations, the greatest being that the validity of predictions for the scale-up of multiple interventions cannot be evaluated at present. The reason is that scale-up of multiple malaria interventions has not yet been achieved in a high-burden country in SSA. Thus no gold standard exists against which model-based predictions can be compared. The "real" answer about impact will come from measurements in countries as scale-up progresses, and it will be instructive to see how close (or far) model-based predictions compare with such results.
Other limitations are mentioned in Table 2. Intervention efficacy assumptions are not validated, and efficacy is assumed constant (i.e., no consideration that efficacy might vary with coverage level, a particular concern for ITNs). The prevalences of interventions are assumed to be independent, which does not allow for the possibility that people exposed to one malaria intervention might be more likely to be exposed to other malaria interventions and which could lead to an overestimation of impact. The models also generally ignore intervention impact on indirect malaria mortality, variation in malaria transmission, community-wide effects (e.g., for ITNs41), and potential interactions among interventions.
Finally, the models do not account for measured changes in malaria morbidity and mortality. The issue here is that many countries are measuring trends in anemia among children (an indicator of malaria morbidity38) and ACCM before and after scaling-up malaria interventions. Thus, an "ideal" model might be one in which the inputs are pre- and post-scale-up measurements of intervention coverage, anemia, and ACCM, and the output is the estimated proportional reduction in malaria mortality. Better still would be a model that could also account for variation in non-malaria-program factors such as seasonal rainfall fluctuations. Perhaps expansion of the recently published dynamic framework on the impact of malaria vaccines could lead to such an ideal model.33,42
After describing their limitations, however, we hasten to add that the models we have discussed are considerable methodological accomplishments, and given the complexity and cost of measuring changes in malaria mortality, they fill an important need. Moreover, as monitoring and evaluation tools, they have several practical advantages. They are rapid and inexpensive, they can be used at the subnational level and in areas where surveys cannot be conducted, and they can be used for any of the leading causes of child deaths.
|
CONCLUSIONS |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Received August 21, 2006. Accepted for publication April 13, 2007.
Acknowledgments: The authors thank members of the Roll Back Malaria Monitoring and Evaluation Reference Group and staff from the Malaria Branch of the Centers for Disease Control and Prevention for their critical review of the manuscript. We also thank Gareth Jones, Jeremy Lauer, and Chantal Morel for their help in clarifying methods for the existing published models.
* Address correspondence to Alexander K. Rowe, Centers for Disease Control and Prevention, Mailstop F22, 4770 Buford Highway, Atlanta, GA 30341-3724. E-mail: axr9{at}cdc.gov 
Authors’ addresses: Alexander K. Rowe, Centers for Disease Control and Prevention, Mailstop F22, 4770 Buford Highway, Atlanta, GA 30341-3724, Telephone: +1 (770) 488-3588, Fax: +1 (770) 488-7761, E-mail: axr9{at}cdc.gov. Richard W. Steketee, Program for Appropriate Technology for Health (PATH), Ferney-Voltaire, France.
Reprint requests: Alexander K. Rowe, Centers for Disease Control and Prevention, Mailstop F22, 4770 Buford Highway, Atlanta, GA 30341-3724, Telephone: +1 (770) 488-3588; Fax: +1 (770) 488-7761, E-mail: axr9{at}cdc.gov.
|
REFERENCES |
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  I. Kleinschmidt, C. Schwabe, L. Benavente, M. Torrez, F. C. Ridl, J. L. Segura, P. Ehmer, and G. N. Nchama Marked Increase in Child Survival after Four Years of Intensive Malaria Control Am J Trop Med Hyg, June 1, 2009; 80(6): 882 - 888. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. K. Rowe Potential of Integrated Continuous Surveys and Quality Management to Support Monitoring, Evaluation, and the Scale-Up of Health Interventions in Developing Countries Am J Trop Med Hyg, June 1, 2009; 80(6): 971 - 979. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. G. Breman, M. S. Alilio, and N. J. White Defining and Defeating the Intolerable Burden of Malaria III. Progress and Perspectives Am J Trop Med Hyg, December 1, 2007; 77(6_Suppl): vi - xi. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
0.75; areas with low transmission defined by MARA index > 0 and < 0.75.