MEASURING MALARIA

Accurate measurement of malaria incidence is of great importance to malaria control, but it is very hard to achieve in most circumstances. Robert Snow and colleagues recently reported the results of a method to determine the case incidences of Plasmodium falciparum malaria around the world.¹ For non-African P. falciparum malaria, they estimated three times more cases than in recent WHO figures.² They suggested that the WHO estimates² were lower due to the use of passively reported national malaria records. We, who prepared the WHO estimates² of non-African malaria cases and published the method used to derive them,³ discuss here the suggestion by Snow and colleagues¹ that, because of the data and methods used, the WHO estimates² must be an under-representation of the true incidence of malaria cases.

It is stated by Snow and colleagues¹ that “The Global Burden of Diseases programme of the WHO has attempted to enumerate the health consequences of malaria infection . for regions outside Africa . from ‘passive’ national disease and mortality notifications to WHO regions without precisely defining the population exposed to varied malaria infection risks.” (For these statements, they cite Refs. 2–5 in the current article.) Snow and colleagues¹ continue, “This use of national disease registration systems to provide accurate reflections of disease rests on three assumptions: that there is complete temporal coverage (every month is reported by a facility), that there is a complete spatial coverage (every health facility reports nationwide), and that all disease events present to, and are reported by, health facilities.” There follows the presentation of their own method (based on malaria reports of active malaria case detection from around the world obtained from published literature from 1985 onwards, combined with spatially linked data on human populations and geographic limits of malaria risk from a 1968 source) and its results from which arises an estimate of 150 million cases of non-African P. falciparum malaria worldwide projected to the year 2002. This is compared with the figure of 51 million cases of P. falciparum malaria per year from our own estimates,^2,3 which covered the period 1993–1997, but was mainly from 1995 and 1996. Snow and colleagues¹ note that our figure is about one third of theirs and propose that ours is an underestimation. They then state, “The most obvious explanation [for our “underestimation”] is the dependence on national statistics derived from passive detection of cases for the WHO’s present global disease estimates outside Africa.”

We now quote from our own published description of how we approached the problem of estimating malaria incidence outside of Africa: “Unfortunately, because of logistic and other difficulties involved, the malaria cases reported in national government statistics usually vastly underestimate the true numbers.”³ We continue, “Because of this [the problems arising from the use of passively collected case data], it was necessary to devise an approach that might yield numbers [for malaria case incidences] closer to the real ones.” Our method³ is summarized as follows.

Briefly, deaths (not malaria cases) directly attributed to malarial infection were used as the raw starting data. This is an independent statistic from the numbers of cases of malaria reported, only a tiny proportion of which will normally be fatal. For each country, we extrapolated from the reported malaria mortality data to malaria case incidences using 1) an adjustment for the likely under-reporting of malaria mortality for each set of national data, 2) an estimate of the likely P. falciparum case fatality rate in the circumstances applying to the particular data, and 3) the proportion (not the absolute numbers) of all cases of malaria in each country’s reported records that were due to P. falciparum (to which virtually all direct malaria mortality today is attributed). Values for the factors in 1) and 2) were obtained by a combination of consultation with other experts for a particular country or region, literature and other data where available, and our own scientific judgment. Values for the factor in 3) was derived directly from the relevant country data itself. By applying the resulting values for each of these factors to each national set of reported malaria mortality data, a figure was reached for the number of P. falciparum malaria cases from which the reported malaria mortality figures for each country would have been expected. Each estimate of national P. falciparum case incidence was tested for correspondence with a plausible entomological malaria inoculation rate for the country concerned.³

Of our approach we state, “This method clearly lays itself open to potentially large inaccuracies, certainly in the case of specific estimates for individual countries. Unlike the unprocessed data, however, there is no reason to suppose that these inaccuracies will be strongly biased in any particular direction, either up or down.”³

In other words, we expected our approach to be inaccurate but not biased in any particular direction.

We agree with Snow and colleagues¹ that unmodified passively collected malaria case records, and to a lesser degree those for malaria-associated mortality, almost always vastly under-represent the true case incidences, as was upheld by our own estimates. Thus, our unpublished spreadsheets show that for the 51 individual countries outside Africa whose records we examined, our estimates for the numbers of malaria cases were from 2.1 to 60.7 times greater than the numbers of cases reported in the individual national statistics. Our global mean estimate for malaria cases outside Africa was 19.4 times the global mean for the national statistics. The numbers of P. falciparum cases provided by individual non-African countries was estimated by Snow and colleagues¹ to be under-reported by a still greater margin of 3- to 1,000-fold, while, as already noted, their overall estimate for the P. falciparum malaria case rate for outside Africa was about three times our own.

Does this mean that ours is a threefold underestimate of the “true” value of P. falciparum case incidences outside Africa? Before addressing this question directly, it can be pointed out that the time periods covered by the two studies are different (1985 to 2002, or thereabouts, for the study by Snow and colleagues¹ and mainly 1995 and 1996 for our own^2,3). As a result, the estimates produced by these two studies are not comparable.

Leaving this consideration aside, however, a case that one figure was an underestimate would require either 1) that it could be logically argued that the specific method used for its estimation leads to such a bias—this has not been done—or 2) that the accuracy of one figure could be validated and that the other could be shown to be statistically significantly lower than the validated figure. This also has not been done and could be difficult or impossible to achieve given the methods used. Unless these conditions can be met, therefore, the view¹ that the difference between our two estimates^1–3 for the numbers of non-African P. falciparum malaria cases is significant, or reflects a particular methodological bias, is a speculation and is scientifically unfounded.