This rapid impact evaluation shows that scale-up of ACT as first-line treatment of malaria, combined with vector control using ITNs/LLINs and IRS resulted a dramatic decline in the malaria burden. Within four years of intervention scale-up, malaria deaths, hospitalizations, laboratory-confirmed outpatient cases and slide positivity rates fell by 76% or more, both in children under-5 years and older age groups.
Unconfirmed (suspected) outpatient cases and numbers of slides examined also fell, but to a lesser extent (two- to three-fold), and starting later than laboratory-confirmed indicators. This suggests that the decline in confirmed malaria cases reflect a real reduction in malaria-attributable burden, and not an artefact related to changing laboratory testing patterns. The lesser decline in suspected malaria cases compared to parasitologically-confirmed cases or malaria hospitalizations and deaths confirms the notion that in endemic settings a large proportion of suspected cases is not due to malaria. This supports the WHO's 2010 recommendation for parasitological confirmation using either microscopy or RDT of all cases as a condition for ACT-based treatment, including in children under-five years in high-endemic Africa for whom presumptive treatment was recommended until 2010 - in an effort to contain costs of ACT-based treatment and to preserve ACT efficacy .
Hospitalizations and deaths due to anaemia fell in parallel with the malaria-attributed events, confirming the importance of malaria as an underlying cause, especially in children under-five years. These observations are consistent with previous studies that showed steady reductions in childhood anaemia in response to malaria control , and with observed correlation between rates of malaria and of blood transfusions in young children . Anaemia represented a much larger proportion of hospital deaths than hospital admissions, e.g. among under-fives in 1999, 26% and 8.5%, respectively. This difference illustrates the poor prognosis of severe anaemia cases in Zanzibar, which is possibly due to varying availability and quality of blood transfusion services and late presentation of patients [10, 11].
The decline in malaria-related burden started around 2003, when ACT was introduced, although for in-patient cases and deaths in children under five years it started slightly earlier, around 2002, possibly reflecting the increasing use, since 2001, of sulphadoxine-pyrimethamine as second-line treatment. In addition, ongoing socio-economic development and urbanization may have led to a better health over time in including malaria burden, before and during the intensified malaria control . However, impact estimates using segmented log-linear regression (Table 1) were adjusted for such trends during pre-intervention period and the decrease in malaria was observed against an increase in non-malaria attendances - which may be the result of improved health services and access in recent years.
Across all 147 out-patient peripheral health facilities in Zanzibar (primary health care units), confirmed malaria cases fell by 73% between 1999 and 2008 while the slide positivity rate fell from 36% to 1.5% (Zanzibar Malaria Control Programme, unpublished, 2010). Over the same period, the slide testing rate of all outpatient consultancies in the peripheral facilities increased from 6% to 30% as a result of RDT roll-out.
In the weekly case-based surveillance system, which covers approximately one-third of health facilities, slide positivity rate was 3% in peak malaria months (April-June) during 2008 . Reductions in parasitologically-confirmed malaria out-patient cases furthermore fit with results from a nation-wide population survey in May 2007, showing parasite prevalence of 0.4% in children <5 years old and 0.9% in all ages .
A 2007 study in North A district of Unguja using surveillance records in 13 public health facilities found a decline in under-five mortality by 52% in 2006 compared to 2003. Similarly, malaria-related admissions, blood transfusions, and malaria-attributed mortality decreased significantly by 77%, 67% and 75%, respectively, between 2002 and 2005 in children under five. While climatic conditions favourable for malaria transmission persisted throughout the observational period, additional distribution of LLINs in early 2006 resulted in a 10-fold reduction of malaria parasite prevalence .
In response to its current low endemicity, Zanzibar has since 2008 shifted its malaria surveillance system to weekly reporting of laboratory-based confirmed malaria cases. A Malaria Early Epidemic Detection System (MEEDS) is now operational in 52 health facilities, with the plan to expand it to all facilities by 2011. The next step for enhancing disease surveillance should be reporting and investigation of individual in-patient cases and deaths, which at low transmission levels represent a failure of the health system in adequately treating both uncomplicated and severe malaria cases. Future reporting of individual case records will also enable more precise geographical tracking of remaining transmission foci or "hotspots" and resurgences, as well as to identify risk groups and factors.
Despite Zanzibar's enormous success in reducing malaria, the risk of an explosive resurgence is still very real. This is not the first time that the islands have achieved such dramatic decline in malaria burden. In the 1970s malaria had been reduced to very low levels through IRS with dichlorodiphenyltrichloroethane (DDT), only to resurge again once partner funds decreased and IRS was stopped . Aggressive malaria control activities and adequate funding therefore need to be maintained to keep the risk of malaria resurgence near to zero.
Substantial decreases in malaria burden have also been reported by other high-endemic sub-Saharan African countries that achieved high coverage of ACT, LLINs and/or IRS. In Zambia, as artemether-lumefantrine was introduced as first-line treatment, and LLINs were distributed over 2002-2008, hospital admissions and deaths, and to a lesser extent outpatient cases from malaria and anaemia, decreased substantially. Declines were more marked in children under five than among older ages, and time trends were consistent across the indicators. Repeated household surveys demonstrated parallel decreases in parasite prevalence and anaemia among children under five years. In addition, among children under five years, both all-cause mortality from household surveys and hospital-recorded deaths fell by half, in a markedly similar time pattern . Importantly, these downward trends were followed by levelling off in 2009 in malaria admissions and deaths - including a major resurgence in two provinces, where parasite prevalence among children again rose according to a 2010 survey [17, 18]. This rebound, possibly related to decay of insecticide and physical deterioration of ITNs distributed several years before, underscores the importance also for Zanzibar to maintain malaria control, surveillance and funding to prevent similar resurgence.
In Bioko Island, Equatorial Guinea, four years after achieving high intervention coverage, repeated household surveys found a decrease in all-cause under-five mortality of 60% from 2004 to 2008, and reductions of 70% in parasite prevalence, 90% in anaemia and 60% in reported fevers among children under-5 . In Rwanda, within two years of nationwide implementation of LLIN distribution and ACT as first-line treatment, in-patient malaria cases and death in nine hospitals and 10 health centers sampled throughout 10 districts fell by 55% and 67% in children under-five, respectively. Non-malaria cases and deaths remained stable or increased . Similarly, in Ethiopia, in a convenience sample of public facilities with relatively complete data, the in-patient case and death burden decreased by 73% and 62%, respectively, after the first two years of scaled-up LLIN and ACT usage . In Sao Tome and Principe, after three years of intensified interventions with IRS, LLINs, ACT and SP-IPTp, malaria-attributed outpatient consultations, hospitalizations, and deaths decreased by more than 85%, 80%, and 95%, respectively, in all age groups . In The Gambia, a similar retrospective analysis at four sites, showed between 2003 and 2007 a significant decline in malaria slide positivity rate, malaria admissions and malaria-related deaths. The same study also demonstrated a significant increase in mean haemoglobin concentrations for all-cause admissions (12 g/l) and age of paediatric malaria admissions (from 3.9 to 5.6 years) .
When interpreting the results presented, the limitations of health facility-based studies should not be underestimated . Importantly, the use of a five-year period (1999-2003) as reference was not necessarily long enough to provide a stable pre-intervention baseline, given the historical resurgence of malaria in Zanzibar in the past . In addition, a longer period should ideally be used as the post-intervention. Therefore, point estimates of impact based on a five-year baseline and one-year post-intervention must, therefore, be regarded as indicative, rather than precise effect sizes.
It should be also considered that time trends observed in health facility statistics may not reflect trends in malaria burden at the population level, if the completeness of case or death notifications and/or access to health facilities changes over the evaluation period, and in particular if notification fraction changed differently for malaria versus other causes of attendance. As a result, the decrease in malaria burden may be of lesser magnitude than that observed in the hospitals sampled. Of note, the reduction in all-cause under-five mortality may be lower at a population level than shown here from hospitals data, because typically in sub-Saharan African countries with stable malaria around 15-30% of deaths, not 53% as in the Zanzibar hospitals between 1999 and 2003, are due to malaria .
The child survival Millennium Development Goal is to reduce all-cause under-5 mortality by two-thirds by 2015 compared to 1990 baseline levels [25, 26]. The new Roll Back Malaria Partnership goal is the near elimination of malaria deaths by 2015 . Based on mortality estimates from 1990 to 2009, the under-5 mortality rates are now declining in all regions of the world, with declines in sub-Saharan Africa having accelerated in the 2000-2010 decade. However, a further acceleration in these declines will be needed to meet the MDG child survival goal .