Malaria Atlas Project: Data visualization or special effect?

A map from the Malaria Atlas Project, modified and superimposed on a photograph of Maarten Vanden Eynde’s “IKEA Vase

The Malaria Atlas Project (MAP) found that human interventions this century averted fully 663 million cases of the disease. “Malaria in Africa,” according to MAP, “has halved since the turn of the millennium.”

MAP’s interactive application visually depicts human triumph over disease, malaria driven back, year after year. But is the triumph real or a special effect? More broadly, is malariology accurately representing reality or is it giving malaria a makeover?

Both the visual aspects and the science of MAP invite scrutiny and raise questions. What the maps show sometimes diverges from what the data actually say, for example. And MAP's data sometimes contradict the World Malaria Report when they ought to be nearly the same. It is doubtful that MAP accounted for age shifting while it is certain that MAP did not model the impact of an epidemic of insecticide resistance on the effectiveness of insecticide-treated bed nets. Both decisions might lead to an overestimate of human progress against malaria. Indeed, a different set of choices might show malaria is now resurgent rather than falling.

Images and science are being tweaked elsewhere in the malaria world. A paper in the Lancet on insecticide resistance presents a map that may have been improperly manipulated. In a separate study of insecticide resistance, a senior author "muted" the finding that resistance substantially reduced the protective benefit of bed nets. In addition, estimates from malaria researchers of the economic benefits of malaria have jumped implausibly from $0 in 2010 to $4 trillion today. Malariaologists are also going as far as saying that artemisinin-resistant malaria is spreading in Southeast Asia and threatens a leap to Africa when current published evidence does not support this contention. 

A Lancet ombudsman fended off criticism of one publication saying: “the paper conveys information that suffices for the message,” a philosophy that mis-informs too much malaria research.

These dissimulations may be well-intentioned, but they are not science.

Malaria Atlas Project (MAP)

Modelers make choices that shape the model. A few shards from an IKEA coffee mug became an amphora (pictured above) by the hands of artist Maarten Vanden Eynde. Similarly, the actual shape of malaria’s burden is ambiguous. Shards of malaria incidence data are so scarce that the World Health Organization (WHO) said it can't tell if cases are rising or falling in 32 of the 45 countries in the Africa region.

The MAP visualization mostly displays modeled estimates, not data. Importantly, MAP relies not on reports of malaria cases (which tend to be few and dubious) but parasite prevalence surveys. These surveys test for malaria parasites in blood samples taken from people in numerous different locales over time. MAP combines geo-located survey information with many other factors, like satellite weather data, all processed by minutely engineered statistical methods. Along with the visualization, MAP and other malaria researchers (Bhatt et al.) produced a numerical summary, published in Nature last September: “The effect of malaria control on Plasmodium falciparum in Africa between 2000 and 2015.”

But for countries like Madagascar, the map and the numbers used in the paper disagree. MAP displays malaria cases estimated from parasite prevalence surveys while, behind the scenes, the aggregate statistics for malaria cases in Bhatt et al. are based on country data.

Maps of Madagascar that should show about the same amount of malaria, based on the bar graph (top), but do not

According to the bar graph, average malaria incidence in Madagascar was roughly 69 cases per thousand people in both 2005 and 2014. The map for 2014, however, clearly shows far less malaria than in 2005. Peter Gething, corresponding author for Bhatt et al., confirmed the disparity, saying it is “entirely correct that there is a mis-match between the time series and the map.”

More than mis-matched, the map and data tell contradictory stories. Visually, malaria in Madagascar appears to be getting crushed. But the data—considered by the researchers as more reliable—say malaria has been rising since 2008, approaching the same level seen in 2000. Malaria in Madagascar looks much better than scientists believe it really is.

Gething, who leads the MAP effort, explained that for some countries “it makes far more sense to base estimates of cases on the country-reported data.” He added: “The list of countries for which the second approach was used is listed in the paper if you are interested.” But the list is not in the paper. Gething did not reply to multiple email requests for the list. However, after the editors of Nature intervened, the MAP tool was changed to list the 11 countries handled like Madagascar where the country data “may not correspond to the parasite-rate derived maps.”

The text in red is a post-publication clarification for MAP (Source: Malaria Atlas Project)

Gething said the 11 countries had smaller malaria burdens and better health systems, leading to more reliable reporting. “Many were unambiguous but inevitably some were borderline situations where arguments could be made for either approach.” But the position of the borderline might have decided the conclusions of the paper. Based on WHO-published country data, reported malaria cases appear to be rising in up to 28 African countries. If some or all of those 28 countries had been chosen, Bhatt et al. might have found a malaria resurgence. Gething would not further detail the criteria used to select the 11 countries.

In concussion research, the NFL stands accused of cherry picking data to produce a milder picture of head trauma. As one critic put it, in excluding unflattering data, “You’re not doing science here; you are putting forth some idea that you already have,” like Maarten Vanden Eynde choosing to make an amphora from fragments of a coffee mug.

Of concern, MAP used country reporting for Gambia, Mauritania and Senegal, three countries which WHO categorized as not having assessable country data. Also puzzling, Senegal has a relative abundance of parasite prevalence surveys. (See figure 2 in the Bhatt et al. supplement.) Senegal even contributed to the much more rarefied data used to transform parasite prevalence into malaria case estimates.

There are presumably good reasons to use country data for Senegal, but Gething would not say what they were. Also, it is not clear if MAP used all the available parasite prevalence surveys or, NFL-style, an unspecified subset. Again, Gething would not say. (He answered 2 of 13 emails which I sent over a three-month period.)

More concerning, what Gething described as “official country-reported data” used by MAP differs radically at times from similar data published by WHO in the 2015 World Malaria Report (WMR). Gething said “some adjustments for known under-reporting or missing data” were applied to the country data. But for Rwanda in 2014, MAP and the WMR present very different pictures of what is happening, although both are based on some form of national reporting.

WHO shows malaria surging in Rwanda (top graph, orange line) whereas MAP (bottom graph) shows malaria tailing off in recent years. (Note: The time axis for the MAP chart runs from 2000 to 2015, one more year than the WHO chart.)

A press account corroborates a sizable malaria resurgence in Rwanda: “Malaria cases in Rwanda rose at 68.6% last year [2014] to reach 1,598,076, against 947,689 cases last year; According to figures released by the Rwandan Ministry of Health.”

It’s not just Rwanda. For 2014, of the 11 nations for which MAP used country reporting, MAP figures undershoot WHO confirmed cases in five: Botswana, Namibia, South Africa, Swaziland and Rwanda.

(Swaziland, as a side note, is not mapped but shows as gray for all years, indicating either intermittent malaria transmission or none. Intentional or not, a gray Swaziland slyly promotes the strategy of “shrinking the malaria map.”)

MAP does not use country data for Burundi. MAP’s survey-based algorithms, however, produce estimates that directly contradict WHO-reported country data.

Malaria incidence in Burundi: Rising sharply according to WHO-reported country data (top, orange line) but falling steadily to its lowest point this century according to the Malaria Atlas Project (bottom).

“We are not sure why the estimates exceed the reported number of cases,” said WHO’s Richard Cibulskis who is also a co-author of Bhatt et al. Cibulskis was uncertain “whether this reflects some double counting of cases or the estimates are just off.” Double counting can be excluded, unless it also afflicts previous versions of the WMR which show much the same chart for Burundi. WHO has not corrected the 2015 edition, so the MAP estimates are “just off.” While data and estimates must be expected to differ in a modeling exercise, the degree of divergence in Burundi might raise proportional concern regarding the model’s validity.

Gaussian process models and reconstruction paste

Maarten Vanden Eynde’s amphora mostly took its shape from reconstruction paste, with just a few pieces of the original blue coffee mug. Similarly, the malaria map for Chad is almost all model. Over the 2000-2015 period, MAP had only a single 2004 study of 960 people.

Red arrow points to the single data fragment, during the 2000-2015 period, to map malaria for all of Chad. (Adapted from Bhatt et al. supplement, Figure 2.)

MAP fills in this data void with exquisite math, computing power and data borrowed from elsewhere to find a steady decline of malaria in Chad, from a peak in 2006 to a low in 2015.

But an amphora is not a coffee mug and malaria in Chad is differently shaped in the eyes of other academics. According to Foster et al., “616,722 malaria cases were reported in 2012, an increase of over 200,000 cases since 2006.” The World Malaria Report also sees malaria in Chad very differently. (Graph not shown.)

Richard Cibulskis suggested that greater use of rapid diagnostic technology possibly increased detection of cases, although “this does not necessarily reflect a true increase in malaria incidence, just an increase in diagnostic effort.”

But Cibulskis acknowledged there were true increases: “Some countries such as Uganda have experienced a resurgence in cases.” To distinguish signal from noise, researchers consider malaria hospital admissions, deaths, diagnostic practices and test positivity rates. I asked Cibulskis if, after taking those factors into account, “can an increase in cases be ruled out for any of the [28] countries which are showing increasing confirmed cases?” In other words, can the possibility that malaria is actually on the rise across most of Africa be excluded? Cibulskis did not reply.

Paying it forward: shifting malaria to older age groups

Malaria interventions frequently target very young children who lack immune protection which develops over time—and by becoming infected with malaria. Averting malaria in the very young reduces cases and overall malaria transmission, but it also prevents acquisition of immunity. As children get older, even where malaria transmission has been pushed down, some will become clinically ill with malaria because of reduced immunity. Overall, cases are greatly diminished but some are “shifted” to older age groups.

Bhatt et al. report parasite prevalence estimates for a cohort aged from 2 to 10. But it is unclear if they adjusted their estimates for the age shifting effects of malaria interventions. If not, their estimates might overstate progress against malaria by leaving the effects of age shifting off the books.

Best scientific practice seems to require accounting for age shifting. According to Briët & Penny (2013): “Many malariological studies limit themselves to examining malaria in children under ten or under five years of age...” However, “analyses for the whole population are preferred as the analyses for children under five do not capture the shifts of morbidity and mortality to older age groups...” I asked Melissa Penny, a co-author of Bhatt et al., whether that paper did as she recommended.  Penny deputized Peter Pemberton-Ross to answer my question, but he didn’t. He said the software used “certainly includes the possibility for age-shifting through its immunity submodel.” He also said that inferring incidence data from prevalence data “may implicitly assume some age-shifting.” But Pemberton-Ross would not say, yes or no, if the Bhatt et al. estimate of 663 million cases averted accounted for age shifting. Peter Gething did not reply to my inquiry about age shifting.

Bed Nets

Spatial only Gaussian Markov random field

Bed nets were “by far the largest contributor,” to averting those estimated 663 million cases, blocking 68% or 450 million malaria episodes, according to Bhatt et al. However, although the MAP interactive application shows the distribution of insecticide-treated nets (ITNs) changing in both space and time, the Bhatt et al. paper used a spatial only model for bed nets. The paper’s supplement states that a spatial only model for bed nets “was preferred over the spatio-temporal model.” Researchers made do with “national means estimated previously” for nets, published in the 2013 World Malaria Report.

Conceivably this creates a mismatch between the maps of bed nets shown by the interactive application and the data used to estimate cases averted, perhaps like the mismatch  of map and data for Madagascar. But a spatial only model for nets might mean a mismatch for all countries.

According to Pemberton-Ross, the spatial only model is just “a technical issue… This choice will have affected the results, but not necessarily by making them less accurate.”

The issue might be fundamental rather than technical, but Peter Gething did not reply when I asked if the conclusion that nets averted 450 million cases of malaria since 2000 rested on a comparatively crude, spatial only model. I also asked Gething whether the interactive mapping tool was displaying spatio-temporal bed net data when, behind the scenes in the Bhatt et al. paper, calculations for cases averted were actually based on a static, spatial only model. Gething did not reply.

I raised these issues to Nature. The editors were responsive to the Madagascar map discrepancy, and appear to have occasioned MAP’s listing of the 11 countries treated like Madagascar “in the interests of transparency,” said Rebecca Walton, Nature’s Senior Press Manager. But the other issues were met with pro forma dismissal: “The paper was rigorously peer reviewed as part of our usual editorial procedures.”

Insecticide resistance—and intransigence

Although bed nets are thought to have stopped 450 million cases of malaria, Bhatt et al. urged that maintaining their effectiveness in the face of insecticide resistance “should form a cornerstone” of future control strategies. But this grave threat to bed nets is entirely absent from the model, as if it’s solely a future concern. (Peer reviewers presumably agreed.)

However, the very distribution of hundreds of millions of nets sparked a proportionally vast rise of resistant mosquitoes, “a worsening situation that needs urgent action to maintain malaria control,” as the subtitle of a recent paper put it. Only a single class of insecticide, pyrethroids, is used to treat nets. Unsurprisingly, mosquitos have developed multiple genetic escape mechanisms, very much as they did when faced with DDT, the primary weapon used in earlier, mid-20th century efforts to eradicate malaria.

Mosquito resistance to DDT increased gradually, ultimately rendering it ineffective and leading to the failure of eradication efforts. With pyrethroids, we seem to be watching a brutal remake of the DDT story. But researchers raise more questions than they answer about pyrethroid resistance: "Is it a problem? How do you know?" asked David Smith, a member of MAP and co-author of Bhatt et al.

In addition to doubting if insecticide resistance is a problem, Smith suggested that dispelling such doubts is nearly impossible: “The experimental unit is the population,” he contended, and “we would need to start collecting data from across the continent,” meaning Africa. A second study, also at continent scale, would be needed to measure the “attributable effect of resistance.” Even more remarkably, Smith said he “would expect the effect size of ITNs to go up overall,” if these two massive studies were somehow completed.

Smith is not alone in denial and casuistry. In Strode et al., researchers set out to investigate “the evidence that resistance is attenuating the effect of ITNs.” But instead, the scientists declared “ITNs are more effective than [untreated nets] regardless of resistance,” which is tautological. Until 100% of mosquitoes are 100% resistant to pyrethroids, an insecticide-treated net will always be more effective than an untreated one.

“Agreed with respect to the tautology,” acknowledged first author Clare Strode. “The ability of ITNs to kill insecticide resistant mosquitoes was significantly reduced when faced with resistance mosquitoes,” but, this message was “muted” in the paper according to Strode. “I originally included a much stronger statement of fact that ITNs kill fewer resistant mosquitoes than susceptible counterparts,” she continued, “but the statistician and Cochrane expert recommended a less bold statement.”

The Cochrane expert, Paul Garner, did not dispute Strode’s account or explain why he recommended a less bold statement. (Also at Garner’s suggestion, the Strode et al. review excluded 914 studies without explaining why.) In 2004, Garner was involved in the Cochrane Review of bed nets that provided much of the basis for the massive scale-up that intervention. Muting the conclusions of Strode et al. might serve to protect the conclusions of the earlier review and the subsequent, massive, bed net intervention that seems to have gone awry.

Many malariaologists demand proof that pyrethroid resistance reduces the impact of nets, a stance akin to the tobacco lobby’s denial that cigarettes cause cancer. According to Clare Strode, “I cannot see how increasing resistance would NOT impact ITN efficacy.” It’s undeniable: “There is no biological basis to argue otherwise,” said Strode.

Nick Hamon agrees: “Yes, resistance compromises efficacy. That is no longer in question.” Hamon runs IVCC, a consortium tasked with developing new insecticides.

Nonetheless recent peer-reviewed papers still ignore resistance. A paper in the Lancet estimating how much malaria might be reduced by further expanding interventions “assumed no loss of effect due to drug or insecticide resistance.” In fact, the authors almost doubled the killing effect of nets in their model, adapted from Menach et al. (2007). Senior and corresponding author, Azra Ghani, did not reply to emails asking how to reconcile today’s stronger insecticide resistance with an assumption of greater killing power than nine years ago.

Insecticide resistance has been modeled. Brady et al. found that resistance cuts the effectiveness of nets by half or as much as two thirds, depending on how swiftly resistance develops. (See Figure 3D.)

In 2013, Penny & Briët determined that introducing nets in high transmission areas with insecticide resistance only reduced transmission by 75% instead of 90%. Penny went on to co-author Bhatt et al., but that paper ignored insecticide resistance even though one third of the population of Africa lived in high transmission settings in 2000.

I am not aware of any papers estimating increased malaria cases and/or deaths resulting from insecticide resistance. IVCC’s Nick Hamon made recourse to “two respected, independent malaria scientists” for a crude estimate of the number of deaths that would be averted if there were a new insecticide to replace failing pyrethroids. According to Hamon: “One scientist gave me a range of 141,000 – 228,000 and another 125,000.” Adding even part of 125,000 to WHO’s estimated 395,000 malaria deaths in 2015 would make for a very sizeable increase in mortality.

Hamon cautioned that “these are ‘back of envelope calculations’ and should be treated as such,” adding that “nobody wanted to be quoted, and for good reasons.” He would not say what the good reasons were. Nonetheless, among themselves, malaria experts countenance disturbingly large increases in malaria deaths resulting from insecticide resistance.

Another mostly insider conversation is the effect of resistance on the infectivity of mosquitos. Against hope and expectation, early indications are that the genetics of resistance also increase mosquito susceptibility to infection by malaria parasites. (See Ndiaith et al. and Alout et al.) Conceivably, not only has the scale-up of bed nets sparked a massive wave of resistant mosquitos, those mosquitos are also more likely to become infected with malaria and thus are potentially more likely to transmit the disease.

Averting the appearance of a malaria disaster: Possible image manipulation in Hemingway et al.

Another paper this year in the Lancet, “Averting a malaria disaster,” drew attention to mounting insecticide resistance and the need to develop new chemicals to replace those that are failing.

However, the map (Figure 2B) accompanying the paper might have understated the extent of the resistance problem. According to the caption, Figure 2B was “reproduced” from an online tool called IR Mapper. However, Figure 2B includes a number of green dots, indicating no resistance, that are not found on IR Mapper, in Sudan, for example. Some yellow dots in IR Mapper, showing possible insecticide resistance in Angola, appear as green dots in Figure 2B, indicating no resistance.

Figure 2B (left) from Hemingway et al. shows a green dot in Sudan but IR Mapper (right) does not although Figure 2B was “reproduced” from IR Mapper. Also, dots in Angola show as green in Figure 2B but are yellow in IR Mapper. (Purple triangles added to identify discrepancies)

I found at least seven such discrepancies between Figure 2B and IR Mapper. In addition, Figure 2B does not display any yellow dots. IR Mapper has been displaying red, green and yellow dots since its inception in 2012, according to Duncan Kobia Athinya of Vestergaard Frandsen, which oversees IR Mapper. Said Athinya: “I cannot speak as to why the Lancet figure does not feature possible resistance (yellow) points, but IR Mapper has followed WHO criteria since its launch in 2012.”

Also unexplained is the green dot in Sudan in Figure 2B. Said Vestergaard’s Melinda Hadi: “I still do not have an answer regarding the green susceptible point in Sudan.” Figure 2B was created in October of 2014 but not published until April of 2016. Studies, and thus dots, have been added subsequently. Also, some might have been removed. But regarding the green dot in Sudan, Hadi said:  “I can confirm a publication was not removed from IR Mapper.”

Perhaps explaining these and other discrepancies, according to Hadi: “the maps in the Lancet article were reproduced. The IR Mapper database was provided to the Liverpool School.” Of possible relevance, IR Mapper includes a “View own data” facility that allows users to create a map from a database.

Hadi explained that yellow dots were left out: “Data points that were classified as possible resistance (90-97% mortality) were not presented in Figure 2.” In addition, Figure 2B “only included data from peer-reviewed publications, so you will note other data points available on the platform (e.g., PMI data) were excluded.” PMI, the President’s Malaria Initiative, collected data in 18 countries.

Asked about issues regarding Figure 2B, first author Janet Hemingway said: “the figure is a screen shot downloaded back in 2014…” Hemingway is Director of the Liverpool School of Tropical Medicine. The discrepancies resulted from the passing of time, according to Hemingway: “Lancet have sat on the paper for almost a year since submission and acceptance so I guess it is possible over this period that IR mapper has been updated for historical data, but we made no alterations to the download.”

Hemingway declined to answer any more questions:  “I have no intention of responding further on this, as there is no further explanation…”

However, the passing of time did not explain the discrepancies such as the missing yellow dots and the presence of a green dot in Sudan and half a dozen other anomalies that surfaced in a non-exhaustive analysis.

I raised these issues with Figure 2B to the attention of Lancet editor-in-chief, Richard Horton. Horton did not reply.

Prompted by the intervention of the Committee on Publication Ethics (COPE), Lancet editor Zoë Mullan relayed an explanation from Hemingway (who had earlier declined to answer questions). Hemingway said: “The single green dot in Sudan he refers to, I suspect is a point that was subsequently corrected if GPS co-ordinates had been incorrectly allocated for example.” Hemingway did not say where the dot could now be found.

Regarding the absence of yellow dots, Mullan explained, perhaps implausibly: “IR Mapper was clearly not showing any yellow dots on the day the authors downloaded the screenshot that became their figure.” Besides the unfortunate timing, Mullan’s explanation would also seem to require that the authors, who are experts on insecticide resistance, didn’t notice the absence of yellow dots that indicate possible resistance, nor did peer reviewers.

In addition to being an editor at the Lancet, Mullan is a trustee of COPE, perhaps creating a conflict of interest in responding to COPE-initiated inquiries. (Mullan’s role at COPE was not disclosed to me; I happened upon it later by chance.)

I also asked Richard Horton directly: “Is Figure 2B a screen capture or reproduced from the IR Mapper database? Why does Figure 2B not show any yellow dots?”

However, Horton only replied that the editors “feel confident that the data reported” in the paper “are accurate and reliable.” He vouched, to some degree, for the data but not for the accuracy of Figure 2B. He did not address the absence of yellow dots.

COPE declined to press the Lancet further. Wrote COPE’s Iratxe Puebla: “Given that the issues relate to a specific figure, we do not feel this falls within COPE’s remit to evaluate...” Furthermore, “COPE considers it beyond its remit to comment on… how facts are presented in individual publications.”

Richard Horton recently criticized COPE for not intervening sufficiently in a controversy regarding statins, bemoaning “the lack of a central institution where scientists who wish to question the actions or ethics of other scientists or scientific institutions can go.”

COPE suggested that I contact “the authors' institution so that they can review and consider what follow up may be appropriate.” Via transatlantic mail, I contacted LSTM board secretary R.E. Holland inquiring about a possible institutional investigation. Holland replied, also by letter: “Having reviewed the issue thoroughly, and having spoken to several experts in respect of resistance incident imaging, I have concluded there is no case for the authors to answer in respect of your complaint.”

Holland’s review assumed that Figure 2B was a screen shot. His letter said: “it is impossible to compare a snapshot of image data from one period to that of another and therefor there is no case to answer.” Holland’s answer was essentially the same as LSTM’s Director, Janet Hemingway. Holland did not explain the absence of yellow dots or the green dot in Sudan. He touted LSTM’s “rigorous research misconduct policy” which had been used in this case.

“[T]he paper conveys information that suffices for the message” 

Figure 2B and the statements by the authors and editors of the Lancet also passed muster with Lancet ombudsman, Malcolm Molyneux. However, Molyneux acknowledged the possibility that Figure 2B was not a screen capture. 

The word ‘reproduced,’ according to Molyneux, “can mean either of the possibilities - a screen-shot or a figure re-drawn from data.” He added: “I really do not think it matters.” In his view, if the authors changed the figure to suit their purposes—including adding dots—they were within their rights: “the paper conveys information that suffices for the message - removing or adding yellow dots or (a few) other dots would make no difference at all to that message.”

Molyneux's statement, “the paper conveys information that suffices for the message,” nicely captures the philosophy that is mis-informing too many papers in malaria.

Placing a green dot in Sudan or anywhere appears to be legitimate in the eyes of the Lancet ombudsman, as long as the number of dots added does not exceed "a few.” Leaving dots out is no infraction, according to Molyneux because “the legend to Fig 2 says ‘reproduced from...’ - it does not say ‘with no subtractions’.” 

He distinguished “falsification of data” from “simplification for purposes of clarity.” However, adding green dots that do not actually represent studies of insecticide resistance means those dots are fake, while changing the color of dots misrepresents the findings of actual studies. It is hard to see how that would not be falsification of data.

Continued Molyneux, “if the authors had been trying to manipulate the figure in order to make their case more compelling, we would expect them to err towards the red in the later time-period (2b)… In every single case you mention of a difference between the IRMapper and Fig 2b, the difference is from red to green, not the other way round.” 

Image manipulation requires no explanation. However, as I wrote to Molyneux, “the title of the paper is ‘Averting a malaria disaster.’ Unless the map shows that there is a disaster to avert then the title doesn't fit… A sea of red and yellow dots might lead readers to conclude that mosquitoes had already won.”

As it stands, figures “reproduced” in the Lancet may differ in unspecified, undisclosed ways from the source in order to convey the authors' message, which might differ from their scientific findings.

Tale of two resistances

If the malaria research community is downplaying insecticide resistance, it is exaggerating the threat of drug resistance in Southeast Asia spreading to Africa. Resistance to artemisinin is not spreading even in Southeast Asia and faces scientifically demonstrated obstacles to overtaking Africa. It’s not happening, but researchers are saying it is.

Arjen Dondorp heads malaria research at the Mahidol-Oxford Tropical Medicine Research Unit in Bangkok. He laid out an accurate chronology of the discovery of resistance to artemisinin-based malaria treatments. Resistance was first found in Western Cambodia, then at the Thai-Myanmar border, in Myanmar, Northern Cambodia, Northeastern Thailand, Eastern Cambodia, Southwestern Vietnam, and Southern Laos.

“Out of the ‘epicentre’ of Western Cambodia,” said Dondorp, “over time the resistant parasite has spread westward, northward, and eastward.” He concluded: “This is spread.”

However, Dondorp’s statement, if not simply false, is not scientifically supported. He described the spread of surveillance, a trick that could equally demonstrate that broken arms or bad breath are “spreading” in Southeast Asia just by conducting surveys in the same places and order he described for artemisinin resistance.

Genetic sequencing has, against expectation, found that nearly every artemisinin resistance hot spot emerged independently, not as a result of spread. Future research might change the current understanding of the epidemiology of artemisinin resistance, but the most comprehensive survey found only three instances of spread out of 112 samples from across the region.

More word play and dissembling are on view in a Lancet paper on resistance in Myanmar that included the word “spread” in its title but adduced little evidence and no claims for it. I wrote two of the authors, saying “the title of your paper, ‘Spread of artemisinin-resistant Plasmodium falciparum in Myanmar’ seems belied by the evidence actually in the paper (and other papers).” I asked them if they would correct “any misperceptions on my part,” but neither Mallika Imwong nor Charles Woodrow replied.

François Nosten, who runs a clinic in Mae Sot near the Thai-Myanmar border, also claims resistance is spreading. “Resistance to artemisinin,” according to Nosten, “has emerged in different places in SEA [Southeast Asia] but then it has spread.” I asked: “Can you describe unpublished data or point to papers where spread is documented?” Nosten, regarded by many as a public health hero, replied not with science but anecdote and sophistry: “We find that over 80% of our patients with malaria have parasites that are resistant to artemisinin. It did not emerge in each and every one independently, did it?” Nosten is correct it did not emerge in each patient independently but that is not at all the same as spread. To establish spread requires DNA sequencing from at least two places; Nosten claims spread based on a single cohort and no sequencing data.

A malaria press tour to Southeast Asia, funded by Malaria No More, featured journalist visits and interviews with Nosten and Dondorp. Stories in Slate and other outlets told readers artemisinin resistance was spreading and threatened a malaria apocalypse in Africa.

Distorted science is creating distorted journalism. An AFP story suggested that the reason artemisinin resistance hadn’t spread to Africa was that “international efforts to contain the spread of resistant parasites have been effective.” It is more the case that biologically it is difficult or impossible to install the multiple genetic changes required to create artemisinin resistance. However, the international containment efforts, by increasing drug pressure, might be forcing malaria down evolutionary pathways which could result in a more compact genetic form of resistance that could be more easily exported to Africa.

Meanwhile, the actual spread of insecticide resistance in Africa is ignored. Another journalist field trip funded by Malaria No More featured Tanzania as the destination. Insecticide resistance might have been among the briefing topics, but it did not appear once in an article for Vice written by one of the journalists on the trip.

No one has died from drug-resistant malaria. “As far as I know,” said Nosten, “there has been no confirmed fatal case.” Meanwhile, according to Nick Hamon’s sources, some part of 125,000 people (or more) have died from malaria because insecticide resistance has reduced the effectiveness of bed nets.

Debasing the currency: $4 trillion drawn on the account of science

Worth less than the paper it’s printed on (Source: Wikimedia)

In 2010, researchers concluded that malaria eradication was unlikely to break even and advised that “financial savings should not be a primary rationale for elimination.” But a few years later, an overlapping constellation of researchers discovered that eradication would quickly generate $4.1 trillion in economic benefits, in just 15 years.

The paper touting a $4.1 trillion windfall “is an advocacy document rather than an academic analysis,” according to Rima Shretta at the University of California, San Francisco (UCSF). Shretta is part of the UCSF group which led development of the 2010 Lancet series which found no cost savings from eradication. Shretta also served on the Action and Investment for Malaria task force that developed the advocacy document projecting $4.1 trillion in benefits.

To reconcile the academic analysis of the 2010 Lancet paper with the later discovery of trillions of dollars in benefits, Shretta seems to suggest scientists are free from the standards of science if they are engaged in advocacy. And functionally, it appears malaria advocacy has detached from science, although much of the advocacy comes from scientists.

The End: Malaria Goes Hollywood

Images, which can surreptitiously mislead, end up exposing a conscious mis-shaping of malaria research. Authors are making undisclosed and perhaps improper choices regarding the visual elements in their papers. However, within the papers, the same authors are free to make any number of choices that can decisively influence findings and there is little or no possibility of suggesting impropriety. The sources of data, how they are processed, type of model and parameters partly or entirely decide the research results. Authors can “mute” statements about the loss in bed net effectiveness caused by pyrethroid resistance. Editors can entitle a paper “spread of resistance” when there is none mentioned in the paper and the evidence contradicts the spread hypothesis. But when the philosophy of managing reader beliefs extends to choices about visual elements, the curtain is drawn aside and we see not a scientist but the Wizard of Oz.