Towards a precise test for malaria diagnosis in the Brazilian Amazon: comparison among field microscopy, a rapid diagnostic test, nested PCR, and a computational expert system based on artificial neural networks

Despite global efforts, the malaria burden is increasing worldwide, with almost two million estimated deaths annually [1]. The lack of precise malaria diagnosis remains an important obstacle to the treatment adherence and effectiveness and clinical management of severe cases. Additionally, the invasiveness and expense of tests limit their utilization in asymptomatic individuals. Within the Brazilian Amazon, the microscopic detection and identification of Plasmodium spp. in Giemsa-stained blood smears from individuals presenting with malaria-like symptoms persists as the gold standard for the diagnosis of malaria and is mandatory to obtain access to the anti-parasitic treatment. Microscopic parasitological diagnosis requires continued personnel training and supervision of users in addition to a minimum laboratory structure, which is difficult to maintain in remote areas of the rainforest. Additionally, such a test is prone to large observer-related variation [2, 3] and lacks sensitivity when performed by non-expert laboratory microscopists [4].

Other diagnostic methodologies have arisen to overcome the inefficient malaria diagnosis, such as PCR-based genetic tests. Nested PCR and real time PCR present higher sensitivity and specificity to malaria diagnosis compared to light microscopy [5, 6]. Nevertheless, these molecular assays are costly and require even more laboratory support and personnel than microscopy, making it difficult to use routinely in the endemic areas. Rapid immunochromatographic tests (rapid diagnostic test, RDT) do not require laboratory support, are easily read and can reach a sensitivity similar to that commonly achieved by well-performed microscopy [7]. Nevertheless, most field evaluations of malaria RDTs were performed in countries with very high malaria endemicity [8, 9], and validation studies in the Amazonian region are still scarce.

Within the Rondônia State in the Brazilian Amazon, the incidence of malaria and the occurrence of drug resistant cases are increasing [10]. In Buritis, a recently urbanized municipality, this situation is worsened by the lack of infrastructure of the health care system and the malaria control program. In addition, many other infectious diseases with similar clinical presentations, such as yellow fever, dengue and leptospirosis, are also common in this area, and the correct malaria diagnosis is of utmost importance to the adequate management of the patients. Certainly, one of the determining factors for morbidity and mortality is the delayed access to the health care. Moreover, the incidence of asymptomatic Plasmodium infection is very high in the Brazilian Amazon [11], further compounding the problem of malaria diagnosis. These individuals are not screened by the health care system, but they can transmit Plasmodium to uninfected Anopheles mosquitoes [12] and may represent important reservoirs. Therefore, the development of simple and noninvasive diagnostic tools is critical to hamper the spread of this infection.

Herein, the diagnostic effectiveness between an RDT (Optimal-IT), field microscopy and nested PCR was compared in individuals with malaria-related symptoms from an Amazonian region, which presents an increasing incidence of malaria [10]. Furthermore, a computational expert system based on artificial neural networks using epidemiological and clinical information was developed in an attempt to diagnose asymptomatic Plasmodium infection, and it was compared to field microscopy and nested PCR.

The present study adds some relevant issues for the diagnosis of malaria in the Amazonian region. Firstly, only 55.63% of the individuals who sought care in a malaria diagnosis center presenting with malaria-related symptoms were diagnosed by the most sensitive diagnostic method applied in this study. The individuals with other diseases looked for exclusion of malaria before seeking care in a regular health clinic, possibly due to the high prevalence of malaria in the municipality. This custom can lead to a delay of the correct diagnosis and in severe cases, could compromise an adequate early management, directly impacting the prognosis and the cost of the health care.

The use of nested PCR as the gold standard was done when it was noted that it presented the highest sensitivity. Considering the principle of this molecular assay, in which small fragments of Plasmodium DNA can be detected, the results were not surprising. Nevertheless, it is worthy to evidence that until today the nested PCR is available only as a research tool, and the cost and technical complexity of this technique hamper the its use in quotidian screening and survey works. The rationale for choosing PCR as gold standard was the necessity of testing the diagnosis accuracy of the field microscopy and the RDT. In this study, 10.3% of the symptomatic individuals with a positive nested PCR (nine P. falciparum and 23 P. vivax cases) were negative by light field microscopy. Although these individuals probably had low parasitaemia, they were symptomatic and did not receive anti-malarial treatment because of the negative microscopy exam. This finding reinforces that the field microscopists from this area need continued refinement, and investments are necessary to improve the quality of the malaria screening. It also indicates that the large demand for microscopic tests in this area, including those from patients with other infections, could contribute to the reduced quality of the tests. Recent evidence indicates that there is a large inter-rater reliability of the parasite counts for the malaria diagnosis [3]. The thin film method is not feasible at a parasitaemia below 500 parasites per microlitre, while the thick film method gives slightly better inter- rater agreements [3]. Moreover, it is well known that most routine malaria microscopists require constant retraining, and that their ability to detect a high proportion of malaria cases is suspect [20].

In addition, the RDT presented higher effectiveness in the identification of malaria cases with low parasitaemia than the light microscopic test. Many other studies worldwide have indicated diverse findings [5, 21, 22]. This result reinforces the idea that assays for rapid diagnosis have the potential to enhance diagnostic capabilities in those instances in which skilled microscopy is not readily available [23]. In order to identify individuals with low parasitaemia neglected by the light microscopy screening in this endemic area, the use of a RDT is advisable in symptomatic individuals who presented a negative thick blood smear exam. This method should be tested in field conditions but it will likely expand the detection of infected individuals and may favour the early clinical intervention and adequate case management. In the sample of 311 individuals, this approach would have resulted in the use of 32 RDT tests, with a minor impact of the health care cost compared to the possible outcomes resulting from a delayed diagnosis.

Optimal-IT, the RDT used in the present study, cannot discriminate mixed infections. The discrimination between P. vivax and P. falciparum infections is critical because the drug therapies and the treatment durations are different. Actually, there are other RDT that can discriminate mixed infections [24], and these should be validated in this endemic area because the occurrence of mixed infections found here was 4.2%. The choice of using the Optimal-IT in this study was made because FUNASA was validating its use in the field during the study period. This work took advantage of this occasion and decided to compare the power of this RDT with other diagnostic tools.

The asymptomatic Plasmodium infection is a major problem in many regions worldwide [25, 26], including the Brazilian Amazon [11]. Symptomless individuals probably develop clinical immunity to Plasmodium parasites after repeated infections [27], which lead to modifications on the host physiology that minimize the intensity of the symptoms, maintaining a very low parasitaemia for long periods [28]. While under this occult infection, these individuals have no reason to seek care in the malaria diagnosis centers. On the other hand, the quotidian primary care activities do not include active detection of asymptomatic malaria. Consequently, these symptomless individuals remain parasitaemic and can serve as a parasite source for uninfected mosquitoes [12], which in turn favours the spread of the infection. Knowledge ofthe prevalence of asymptomatic malaria cases in certain regions could assist in the implementation of control strategies, which may include treatment of asymptomatic Plasmodium-infected individuals.

Asymptomatic individuals frequently refuse to give blood for tests, which hampers the detection of symptomless plasmodial infections. Additionally, the routinely used thick blood smear exam displays a low performance in individuals with low parasite burdens, as is the case in asymptomatic Plasmodium infection [29]. Herein, a pilot investigation was performed addressing whether a computational system could discriminate asymptomatic malaria cases. For this purpose, major epidemiological determinants of asymptomatic malaria, such as the age, time of residence in the endemic area, number of previous malaria infections, gender and use of bed nets, were used. The technique of ANN was chosen because it offers good robustness against noise and typically works very well when no previous knowledge is available in order to facilitate the classification [18]. Thus, the network can be trained to recognize the pattern of the disease to be diagnosed from the medical database used. With a very intuitive interface that could be used by primary care professionals in the endemic areas, the software used an expert system based on neural networks [16]. The disappointing results obtained by the exclusive use of epidemiological data indicate that other complex factors may be more influential for the development of the asymptomatic Plasmodium infection. This idea was confirmed when the software performance significantly improved after the addition of information regarding IL10 - and IFN-gamma plasma measurements. In addition, other routine biochemical laboratory exams did not improved the MalDANN performance. These cytokines were chosen in the light of evidences that individually or as ratios they are associated with the malaria severity [15, 30]. Besides the known epidemiological factors, the genetic background and/or the common occurrence of co-infections within the population may play a fundamental role on the occurrence of asymptomatic malaria.

This study shows that the chosen RDT (Optimal-IT) performed superiorly in discriminating symptomatic malaria cases with low parasitaemia than field microscopy, although it did not discriminate mixed infections. An RDT for malaria diagnosis may find a promising use in the Brazilian Amazon integrating a rational diagnostic approach. Despite the low performance of the MalDANN test using solely epidemiological data, an approach basedon neural networks may be feasible in cases where simpler methods for discriminating individuals below and above threshold cytokine levels are available.

BBA, ARF, SMSN and LLN received fellowships from the Brazilian National Research Council (CNPq). EC, AB and MBN are senior investigators from CNPq. This work was supported by FINEP (010409605)/FNDCT-CT Amazônia. The funders had no role in the study design, data collection and analysis, decision to publish or preparation of the manuscript.