An interactive explainer
Routine surveillance records a malaria case when a person with a fever tests positive. But where malaria is endemic, many infections are asymptomatic, and a fever from another cause can coincide with an asymptomatic malaria infection. Those fevers test positive and are recorded as malaria, even though malaria did not cause them. Because non-malaria fevers occur year-round while true malaria can be seasonal, this over-counting is not uniform through the year.1
Following the shift to test-based case management, a person presenting at a health facility with a fever is given a diagnostic test, usually a rapid diagnostic test (RDT), before antimalarials are prescribed.1 Routine surveillance then records a confirmed malaria case for each febrile individual who tests positive. The recorded count is therefore a count of test-positive fevers presenting for care, not a direct count of illness caused by malaria.
A positive test establishes that the person carries a detectable infection. It does not establish that the infection is the cause of their fever. Two conditions have to hold for a recorded case to be a true clinical malaria case: the person must be carrying parasites, and those parasites must be the reason they are febrile. Where the two are not causally linked, the case is recorded all the same. The routine count is thus the sum of true clinical malaria and a second, coincidental component.
In endemic settings a large share of the population carries a bloodstream infection at any given time without being ill from it.2 Acquired immunity allows people to tolerate parasites, and untreated infections can persist for weeks to months, so this asymptomatic reservoir is both substantial and slow to turn over. As transmission intensity rises, clinical illness begins to level off while the prevalence of detectable infection keeps climbing, so the infection reservoir grows relative to disease: immunity increasingly limits illness without clearing infection.
Because carriage builds up over the preceding weeks of transmission and clears only slowly, the prevalence of detectable infection is a broadened, lagged and floored version of the transmission signal: it rises after the season is under way, falls gradually once transmission recedes, and does not collapse in the low season the way new clinical illness does. Anyone tested during the low season still has an appreciable chance of a positive result, whether or not malaria is making them ill.
Fevers have many causes other than malaria (respiratory and other infections chief among them) and across a year these are far more common than malaria-caused fevers.1 Their rate is higher in young children. Some of these causes carry their own seasonality, for example respiratory and diarrhoeal illnesses. But that seasonality generally differs in timing and duration from the malaria transmission season. We assume here that the non-malaria fever rate is constant through the year.
A non-malaria fever that is tested returns a positive result whenever the person also carries a detectable infection. The coincidental positives are therefore the non-malaria fever rate times the share tested times the prevalence of infection. With the fever rate steady, that count follows the asymptomatic reservoir rather than the seasonal course of disease.
A toy model of true versus routine-recorded malaria cases through a season
Set the transmission intensity (which fixes both the level of clinical malaria and the prevalence of infection) and how seasonal transmission is, then set the non-malaria fever rate and the share of fevers that seek care and are tested. The chart shows true malaria cases recorded from those who seek care (purple) and the total routine-recorded count (dark line); the shaded band is the falsely attributed cases: non-malaria fevers that coincided with a detectable infection.
Two things to try: raise the non-malaria fever rate and watch the recorded curve lift and its seasonality flatten; and move treatment seeking and notice the vertical scale shifts while the peak- and low-season figures do not. Better access records more cases but does not fix the misattribution.
True versus routine-recorded malaria cases through the year
The recorded line sits above the truth all year; the gap is the coincidental positives. In the peak the gap is small relative to a large true signal, but in the low season the true signal has collapsed while the reservoir persists, so the same non-malaria fevers make up most of what is recorded.
The recorded count is always at or above the truth. Every non-malaria fever that coincides with a detectable infection adds a case that malaria did not cause.
The impact of non-malaria fevers is differential across the season. Compare the first two readouts. Because the recorded count is set against true malaria, the over-count tends to be smaller in the peak season, when the true signal is large, and larger in the low season, when true clinical malaria has fallen towards its floor while the asymptomatic reservoir, and the steady non-malaria fever rate that samples it, has not. In a seasonal setting this can make the recorded season look flatter and longer than the real one.
The over-count leans to the months after the peak. The falsely attributed band is not symmetric around the season; it is heavier on the falling side. Infections take weeks to clear, so the pool of detectable infection lags the transmission that produced it and wanes only slowly: it is still large in the months after cases have peaked, when it keeps turning non-malaria fevers into recorded positives. Prevalence therefore peaks later than clinical incidence and trails a long tail into the following dry season, and the coincidental count follows it.
The more concentrated the season, the larger the low-season over-count. Raise the seasonality slider so the true peak is narrow and its off-season floor is low. The low-season over-count climbs, because the true signal it is measured against has almost vanished while the coincidental component holds up.
Higher transmission enlarges the reservoir faster than the disease. Raise the transmission intensity: clinical illness levels off while the prevalence of infection keeps climbing, as immunity limits disease without clearing infection. So a fixed non-malaria fever rate yields more coincidental positives per true case, and the over-count grows. Where infection is common, a positive test carries less information about the cause of a fever.2
Better access records more cases but does not fix the bias. Move treatment seeking: it multiplies both the true malaria cases and the coincidental ones, so the recorded volume rises or falls (the vertical scale moves with it), but the over-count ratios and the share of recorded cases not caused by malaria do not move. In this model the misattribution is a property of the case definition and the reservoir, not of how many people present for care.3
Reading routine data as clinical burden may overstate malaria. The share of recorded cases not caused by malaria is substantial at plausible fever rates: largest in the low season, and, as a share of positive tests, greater the higher the transmission intensity, because a common infection reservoir turns more non-malaria fevers into coincidental positives.3 Trends and seasonal profiles built from confirmed-case counts inherit this bias, most heavily off-peak.
Methods. A deliberately simple, illustrative toy model showing the shape of how non-malaria fevers inflate routine malaria counts, not a specific setting; not to be used for decision making. The vertical axis is a relative case scale, not a count in real units. It rescales to fill the frame at every slider setting, so as transmission, the fever rate and treatment-seeking move it is the scale that shifts, not the height of the curve. Its absolute level is not calibrated to any setting, so only the shape and the relative over-counts are meaningful; the readouts, which are ratios and shares, are unaffected by the scale. It models a single focal group (young children), with the case peak fixed at mid-year and a symmetric rise and fall.
References.