Fever without a localized cause is one of the most common presenting complaints among persons seeking healthcare in many low- and middle-income countries [1], [2]. However, unlike the syndromes of pneumonia and diarrhea that feature in global disease burden estimates and have well coordinated programs integrating efforts across the range of responsible pathogens to avert morbidity and mortality, there has been a lack of a coordinated approach for febrile illness. While illness and death due to some specific infections causing fever, such as malaria [3] and increasingly bacterial sepsis are well quantified [4]–[6], others such as a range of zoonoses and viral infections are uncounted and consequently may be underappreciated.

The various etiologies of febrile illnesses are difficult to distinguish from one another clinically [7], [8]. As clinical laboratory services are often limited in areas where febrile conditions are particularly common [9], [10], clinicians may have few diagnostic tools to establish an etiologic diagnosis. Therefore, clinical management is often driven by syndrome-based guidelines employing empiric treatment [11]–[13]. In the absence of systematically collected data on fever etiology, considerable mismatch between clinical diagnosis, clinical management, and actual etiology may occur resulting in poor patient outcomes [14]. It is increasingly recognized that malaria is over-diagnosed in many areas [14], [15]. To address this problem, the World Health Organization (WHO) malaria treatment guidelines moved away from clinical diagnosis of malaria to treatment based on the results of a malaria diagnostic test such as a blood smear or a malaria rapid diagnostic test. With more widespread availability of diagnostic tests to exclude malaria and apparent declines in malaria worldwide [3], clinicians in resource-limited areas are faced with a growing proportion of febrile patients who do not have malaria and few tools to guide subsequent management.

We sought to describe comprehensively the causes of febrile illness in northern Tanzania among patients sufficiently ill to require hospitalization. Febrile patients admitted to two hospitals were evaluated for a wide range of infectious etiologies using conventional standard diagnostic techniques.

We demonstrate among hospitalized febrile patients in northern Tanzania that malaria is uncommon and over-diagnosed, while invasive bacterial, mycobacterial, and fungal infections are underappreciated. At the same time, the bacterial zoonoses leptospirosis, Q fever, and spotted fever rickettsioses, and to a lesser extent brucellosis, and the arbovirus infection chikungunya are common yet unrecognized causes of fever. Our findings point to important mismatches between clinical diagnosis and management with actual diagnoses that have major implications for patient care, disease control and prevention, and for judicious use of antimalarial medications.

While the problem of malaria over-diagnosis has been appreciated for some time [14], [15], studies that comprehensively describe the causes of severe non-malaria fever requiring hospital admission beyond bloodstream infections have been lacking. The over-diagnosis of malaria results in inappropriate use of antimalarial medications and may be associated with higher case fatality rates among patients treated for malaria who do not have the infection [14], [15], [34]. While the underlying causes of the over-diagnosis of malaria are complex [35], the lack of epidemiologic information about the importance of alternative infections and guidance on their management is likely to play a role. Our findings confirm the potential benefits of making reliable malaria diagnostic tests available at healthcare facilities and using the results as the basis for prescription of antimalarial medications [36]. When adopted, such an approach to malaria treatment would support the judicious use of antimalarials and would define the population of patients with nonmalaria fever.

We found that the bacterial zoonoses, leptospirosis, Q fever, and spotted fever group rickettsioses, and to a lesser extent brucellosis, are major causes of febrile illness among patients sufficiently unwell to require hospitalization. That a group of neglected bacterial zoonoses are of major clinical and public health importance in sub-Saharan Africa is a new and paradigm-changing finding. For clinical practice, with the exception of leptospirosis that may be effectively treated with commonly prescribed antibacterials, patients with brucellosis, Q fever, and the rickettsioses are likely to leave hospital without specific treatment. In northern Tanzania where many rely on livestock for their health and economic wellbeing, Leptospira, Brucella, and Coxiella spp. also indirectly affect human health through their impact on animal fertility, growth, and survival. The control and prevention of the neglected bacterial zoonoses is likely to involve interventions that require the collaboration of human health experts with the animal and environmental health disciplines, an approach that is underdeveloped in many parts of the world.

Clinical guidelines for management of febrile patients in low resource areas focus on the identification and treatment of malaria and bacterial sepsis [11]–[13]. Our findings suggest that there is a need to identify and incorporate guidance on when to use a tetracycline for treatment of Q fever or rickettsial infection and when to consider treatment for brucellosis. We have previously demonstrated that features of the clinical history and physical examination do not perform well for identifying fever etiology [7], [8], [21], [26], [30]. Therefore, improvements to treatment algorithms for febrile patients are likely to require the development and incorporation of reliable diagnostic tests that provide timely diagnostic information to clinicians [37]. Unfortunately, many rapid diagnostic tests for infections related to fever management other than malaria and HIV suffer from poor performance characteristics [38], [39].

Lack of coordination among groups working on the various etiologies of febrile illness in low-resource areas has meant that sentinel studies that could provide much more comprehensive information on a wide range of responsible organisms instead have focused on only one or a small handful of etiologies. For example, a clinical trial evaluating the impact of pneumococcal conjugate vaccine on rates of Streptococcus pneumoniae bacteremia in a community has the potential to identify and report all bloodstream infections. Similarly, a study designed to estimate the incidence of typhoid fever to inform vaccine policy could collect acute serum along with the blood culture and, with subsequent collection of convalescent serum, would have the ability to estimate the incidence of leptospirosis and a range of other etiologic agents using conventional serologic methods [40]. However, resources for research have tended to be targeted to specific pathogens and researchers have struggled to leverage additional resources to address a broader range of organisms. Sentinel site studies seeking to understand the infectious causes of febrile illness in low-resource settings have utilized blood culture to highlight the importance of invasive bacterial and fungal infections [4], [41]. Expanding laboratory evaluations to include serologic and molecular approaches to diagnosing infections requiring specific antimicrobial management such as the bacterial zoonoses brucellosis, leptospirosis, Q fever, and the rickettsioses adds considerable value [40]. Detection of infections of public health importance such as those caused by the arboviruses dengue, Rift Valley fever, and yellow fever can inform national control programs. Since considerable etiologic overlap exists between the syndromes of fever, acute respiratory tract infection, and diarrhea [42], [43], addressing these simultaneously in integrated sentinel studies would inform enhancements in empiric treatment guidelines and improvements in the accuracy of syndrome-based disease burden estimates.

Our study had a number of limitations. While we examined a wide range of etiologies of fever, a large proportion of patients were undiagnosed suggesting that we failed to identify potentially important infections. The undiagnosed group is being investigated further using pathogen discovery approaches. Some of the diagnostic tests used in our study are less than 100% sensitive and specific and we did not test for every known pathogen. As a consequence, we probably underestimated the prevalence of some infections while misclassifying others that were falsely positive. Because a number of our diagnostic tests relied on the demonstration of a four-fold rise in antibody titer between the acute and convalescent serum sample, not all enrolled patients returned for collection of convalescent serum to have diagnoses confirmed. It follows that calculation and comparison of case fatality rate was not possible since those who died before the convalescent visit could not be confirmed cases. Incomplete diagnostic information meant that we had to extrapolate prevalence from the tested population to the untested population, potentially introducing bias. Similarly, instances of apparent infection with multiple agents were not accounted for in presentation of pie graphs. Inclusion of a well control group would have allowed the calculation of attributable fractions for individual pathogens, something that should be considered for future febrile illness research, especially in areas where malaria is endemic. Since considerable geographic variation in fever etiology is known to occur, the generalizability of our findings is uncertain.

What is needed to support an integrated approach to the syndrome of fever in resource-limited areas? First, fever should be recognized alongside pneumonia and diarrhea as a major clinical syndrome of public health importance. Achieving this is likely to require leadership from international institutions of public health and reappraisal of the way that the febrile illnesses are approached in burden of disease estimates. This could include estimating total morbidity and mortality from the syndrome of fever as a first step before attributing the associated illnesses and deaths to specific etiologies, much as is done for the other major syndromes [44], [45]. Second, efforts are needed to bring together the diverse groups and disciplines currently working on the febrile illnesses to quantify the morbidity and mortality attributable to each major etiologic agent. Such integration could be facilitated by support for research efforts that study the syndrome of fever comprehensively as well as its etiologies individually, an approach that has been modeled by studies addressing the syndromes of pediatric pneumonia and diarrhea in developing countries [46], [47]. Third, improved diagnostic services are urgently needed to establish disease burden estimates and patient management for the febrile illnesses in resource-limited areas [10]. Conventional diagnostic tests for some infections, such as leptospirosis, are complex. For example, the collection of both acute and convalescent serum samples may be required, and testing services may be available at only a few national or supra-national reference laboratories. Assays relying on convalescent samples cannot be used to estimate case fatality rates [21], [26]. Conversely, simple, rapid tests applied to acute samples may have poor performance characteristics [38]. Finally, clinical studies, including clinical trials, are needed to test and improve clinical management algorithms for febrile patients. The goal should be to target antimicrobial therapy to those who need it and to avoid inappropriate use among patients who will not benefit. In this way, patient outcomes can be improved, health resources can be conserved, and disease prevention and control efforts for febrile conditions can be rationally resourced.