The results of this study showed that there was poor performance of routine microscopy in health facilities at all levels of care. Although hospital two, with a positivity rate of only 13.3% by routine microscopy, seemed to perform better in terms of specificity, it is difficult to comment on individual health facilities due to the small numbers of slides collected. However, as group the hospitals did not seem to perform better than the health centres or dispensaries, despite being considered as referral centres providing more specialized and higher quality services . The low sensitivity (although not measured precisely due to the very small number of true malaria cases) and specificity of routine microscopy we observed in this study are consistent with earlier studies done in Tanzania [11, 38, 45] and elsewhere in Africa across different malaria transmission areas [12, 29, 46]. This poor performance has profound clinical implications for two main reasons. Firstly, the low sensitivity leads to malaria cases being missed and therefore patients being denied a safe treatment for a potentially lethal disease. Fortunately, the prevalence of malaria in Dar es Salaam is very low, hence this was less an issue in this setting. On the other hand, the low specificity with an associated very low positive predictive value of 2.8% leads to a massive over-diagnosis and subsequent over-treatment with antimalarials, a practice commonly experienced in low to moderate transmission settings in Africa [21, 29, 47]. The low specificity of routine microscopy has many negative consequences such as a waste of limited resources, increased costs for drugs, unnecessary exposure of patients to the adverse effects of drugs, and chiefly the fact that clinicians tend to overlook other causes of febrile illness [24, 25]. Finally, an analysis of these data will give a completely incorrect picture of the epidemiology of malaria, with serious implications for the management of patients and the planning of control measures .
Further evidence for the reporting of many false positive results by the microscopists was the fact that 93% of all positive findings were reported as very low parasitaemia (1-5 malaria parasites per 200 white blood cells), which contrasted with the results of expert microscopists reporting a median parasitaemia of more than 1000 malaria parasites per 200 WBCs. Hence, it is very likely that because they knew the poor quality of their microscope, slide preparation and staining (and thus the difficulty in differentiating real parasites from artifacts), the microscopists simply wrote down having seen a few parasites to avoid taking any risk.
In the present study we observed a reduction of more than 80% in the proportion of malaria positive tests in the intervention facilities following the introduction and routine implementation of RDTs. This reduction was observed in all health facility levels and type of facility, i.e. hospitals, health centres and dispensaries. The effect was so marked and coincided so well with the introduction of RDTs that it is almost certain to be causally related to the improved diagnostic performance of the malaria test. This conclusion is reinforced by the fact that the RDT results were confirmed by expert microscopy. As a result of this reduction, the consumption of AL decreased by 68% , and this approach could be shown to be safe for the patients .
The poor performance of routine microscopy may be attributed to several factors, including lack of skilled and competent laboratory technicians, and also an inadequate number of laboratory personnel, who are overwhelmed by the large volume of blood slides requested by clinicians. At the hospitals in Dar es Salaam, the number of slides performed per technician ranges from 100 to 200 per day, while at health center level it ranges from 70 to 100 per technician per day. Other factors that contributed to this situation are the poor condition of laboratory equipment (especially microscopes), insufficient or substandard reagents, frequent power interruptions, ambiguous guidelines, inadequate supervision and virtually non-existent quality control systems [6, 30, 32, 34, 37, 49]. An in-depth assessment of these factors was not within the scope of our study.
In some places in Africa, efforts to improve microscopy have been successful, such as in Ghana  and Uganda . However, it is uncertain how such isolated efforts can be sustained in the long-term when implemented routinely and on a large scale [31, 34]. This is illustrated by a study in a rural district of Tanzania which documented that there had been no substantial improvement in the performance of microscopy despite intensive refresher training for laboratory technicians and supervision . Training needs to be properly planned, regular and considering the high turn-over of health staff. Moreover, other important structural factors need to be addressed, such as levels of laboratory staffing, ensuring adequate supportive supervision, provision and maintenance of essential laboratory supplies and improving the basic health facility infrastructure. In view of this, achieving substantial improvements in malaria diagnosis by microscopy requires an enormous investment that has to be weighted against the benefits of introducing RDTs. Finally, the recognized suboptimal quality of microscopy in health facilities appears to have led clinicians throughout Africa to mistrust the results, and to prescribe anti-malarials to patients regardless of the test result [2, 6, 11, 12, 29, 32].
The large reduction in the malaria positive rates seen in the intervention facilities after RDT implementation were not observed in the control facilities, despite a slight decrease in the positivity rates observed during the second half of RDT implementation. This late trend may have been due to progressive 'contamination' of the control facilities due to transfers of staff from intervention to control health facilities or to informal inter-facility communication (no formal training on microscopy took place during the study period).
The true prevalence rate of malaria in feverish patients in our study was found to be below 10% in all health facilities. The facilities were fairly well distributed across the three municipalities of Dar es Salaam and since they served a significant proportion of the population we can consider our results to be representative for Dar es Salaam. This clearly indicates that the prevalence of malaria in Dar es Salaam is much lower than has been previously documented . The low malaria positivity rates are consistent with results from an earlier study done by Wang et al in Dar es Salaam during the dry season, which found less than 7% of all fever to be due to malaria . These rates also compare well with a recent Malaria Indicator Survey which found a prevalence rate in children less than five years in Dar es Salaam Region of 1.2% .
This assessment was based on data collected from routine health service statistics recorded in health facility register books (MTUHA), which may have been inaccurate or incomplete. When in doubt, the data from other health facility sources (laboratory registries) were counterchecked for consistency. In any case, the considerable reduction in the proportion of positive tests for malaria following the introduction of RDTs is very unlikely to have been due simply to erroneous records.
The implementation of the study focused on public health facilities, so these findings may not be generalized to private health facilities, which see the smaller part of health consultations in Dar es Salaam . Anecdotal evidence suggests that the malaria positivity rates for malaria microscopy are even higher in private health facilities, but we could not investigate this in the present study. This is certainly a priority for future research.