This study showed that very few patients who were given a referral advice from a CMV completed their referral. This is an important operational finding, questioning the quality of care of community-based malaria programmes relying on referral of cases with RDT-negative fever.
The RDT result was found to be a significant determinant of referral completion. Less than one percent of patients with a negative RDT completed their referral, while nearly 90 percent of patients who tested positive for malaria, but could not be treated by the CMV, arrived at a health structure. This finding suggests that patients with signs of severe malaria usually complete their referral.
A large variation of referral completion across the 16 networks was notable. It could possibly be linked to variations in distance, CMV performance or differences in registration and follow-up by the health personnel in the various health structures. Due to the small number of completed referrals, no conclusions could be drawn regarding the outcome or timeliness of referrals.
Limited access to healthcare and lack of trained health workers has increased attention to task-shifting. Community Health Workers (CHWs), trained to recognize and treat simple malaria at village level, are becoming an integral part of many health systems. Research has shown that community-based malaria control programmes can reduce paediatric hospitalisations , improve care-seeking behaviour  and reduce the workload in health facilities . Still, only a few studies give evidence for reduced mortality [14, 15] and additional research is definitely needed in order to guide policy decision-making and implementation .
An important factor in the use of CHWs to reduce mortality is completion of their referrals. Referral compliance varies in different contexts and at various levels of health care. In Uganda, 87% of patients referred to health centres by CHWs complied with the referral , while compliance among referrals from health centres to hospital was only 28% . Several factors can influence adherence to referrals including access (e.g. transportation, overnight stay) and family dynamics (mother not being the decision maker) . Signs of severity, such as altered consciousness and convulsions, can increase adherence up to three times, while costs (e.g. for lab services) led to four to five times lower adherence .
It is difficult to understand why the referral completion rate in MSF's CMV programme was so low. There could be several reasons, as described above, but adequate supervision and training could be one way of improving the referral compliance. Several studies have shown that training is essential in order to ensure a successful community-based malaria programme. For example, training of caretakers, as well as of private drug vendors in Nigeria has lead to improved recognition, treatment and referrals of malaria cases [19, 20]. A training programme for CHWs in Mali increased adherence to referrals to an encouraging 87% . Health worker actions, such as giving a referral slip and advising caretakers to go to the hospital immediately, can also improve referral completion. It is, however, important to bear in mind that CHWs require not only training, but also adequate supervision and remuneration . In MSF's programme supervision is a challenge, with only two CMV supervisors overseeing and coordinating the CMV activities.
One possible strategy to reduce the poor referral compliance rate is to narrow the diagnostic criteria for referral so that patients could avoid unnecessary costs and time in attending a health structure. CMVs could receive additional training regarding respiratory infections to distinguish between serious signs of pneumonia (e.g. increased respiratory rate) and those of common cold viruses. This strategy has support in the literature where CHWs have been able to diagnose and treat pneumonia in Nepal  and in Uganda . To offer appropriate treatment to RDT-negative patients with fever, adapting the treatment criteria and training CMVs in diagnosing and treating for example pneumonia and diarrhoea, should also be considered. In Zambia, a cluster-randomized controlled trial among CHWs showed great potential for CHWs to manage both malaria and pneumonia at community level. Over 68% of the children treated for simple pneumonia by CHWs received early and appropriate treatment, compared to 13.3% of the children who were referred to the health centre . This strategy would, however, put heavier responsibilities on CMVs that would require additional training, supervision and support.
There are some encouraging results from this study. Previous research shows that CHWs are capable of using and interpreting RDTs [25, 26]. The high positivity rate in this study suggests that the CMVs are following the correct indications for testing, treating patients with simple malaria. The study also shows that the CMVs are, at least, suggesting referral for patients with signs of severe malaria and with negative RDTs. There is a possibility that the high RDT-positivity rate is due to over-diagnosing of malaria, and that patients might not have received a referral advice, though noted as referred. With increased supervision this could be prevented.
This retrospective study had several limitations. The short study duration may have been influenced by disease patterns, disease variations and seasonal activities such as farming, affecting caretakers' behaviour. The low number of referral completions did not allow for a proper analysis of outcome and timeliness of referrals, as well as for impact of distance from villages to referral structures. As analysis of data was done through record reviews it could not be verified that all referred RDT-positive patients had signs of severe malaria. Analysis of signs of severe disease among RDT-negative patients could also not be carried out. The record review was done in two steps: comparison of the two records (CMV recording forms and CMV referral follow up register) and encoding of this data was done in the field. The data analysis was done at Karolinska Institutet, and there is a possibility of misinterpretation of data. With only two CMV supervisors responsible for data collection and direct CMV supervision, it is possible that referred patients were not properly entered in the CMV recording forms. There is also a risk that patients were missed due to variations in their own or their caretaker's name. Using three different data-clerks in the field for encoding might have increased data errors. Finally, a qualitative, in-depth study, interviewing CMVs, health personnel and patients could give a better picture of the underlying reasons for non-compliance of referrals.