Anticipating the future: prognostic tools as a complementary strategy to improve care for patients with febrile illnesses in resource-limited settings

Introduction

Globally, febrile illnesses are among the most common reasons to seek healthcare.1 While most can be managed at the community level, a small proportion (~1%–2%) progress to life-threatening disease.2 This burden is carried disproportionately by individuals in low-income and middle-income countries (LMICs), where febrile illnesses remain a leading cause of morbidity.3

Understanding the underlying causes of the main febrile syndromes is critical to successful treatment of febrile illnesses. Several recent initiatives have addressed this topic.4–6 Nevertheless, approaches that focus solely on diagnosis struggle to reconcile the fact that patients with the same infection or syndrome can have markedly different illness trajectories,7 perhaps reflecting differing host nutritional and other susceptibility states.

Most febrile patients in LMICs are managed by community health workers and healthcare providers working at primary or district level. These practitioners often have limited training and inadequate access to the necessary supervision and diagnostic testing to support their clinical decision-making. In such contexts, in addition to assessing the cause of a patient’s illness, an equally pertinent question is: is my patient’s condition likely to progress and require a higher level of care? A prognostic tool that could reliably risk stratify patients would have immense potential for benefit, through timely identification of patients at risk of deterioration and guiding appropriate use of scarce resources.

In contrast to a diagnostic test which determines whether a specific disease or health state is present at the moment the test is performed, a prognostic test provides information on the likelihood of a particular outcome occurring in the future.8 Used appropriately, prognosis can complement diagnosis to improve precision and efficiency of management algorithms for febrile illnesses. This could be particularly impactful in resource-constrained settings where diagnosis remains most challenging, triaging practices predominantly rely on clinical evaluation, and decisions to refer must be made early due to complex context-related referral mechanisms.

Common pathophysiological pathways to severe febrile illness exist across a range of microbial aetiologies.9 10 Biochemical markers of these pathways, reflecting endothelial injury, immune activation and coagulation, appear to add value to simple bedside assessments to improve identification of patients with a poor prognosis.11–13 Reliable and practicable tests for these markers could help risk stratify febrile patients and inform management decisions at critical junctures in the patient care pathway. While a standalone test for a biochemical biomarker could provide useful prognostic information, these tests might be more effective as part of an algorithm, combining measurement of a biomarker(s) with other clinical parameters (signs and symptoms, demographic information, comorbidities, etc) to more accurately assess risk and guide rational management.

Unlike diagnosis, prognosis is inherently context-dependent: a patient’s eventual outcome is inextricably influenced by the available resources and quality of care. Hence, in order to advance the conversation around prognostic testing in febrile illnesses, specific use-cases must be defined. Each use-case should detail the clinical problem and consider the resources available to treat febrile illnesses in that setting (eg, health worker and laboratory capabilities, referral capacity, and availability of essential resources such as oxygen, fluids, antimicrobials and provision of vital organ support), in order to contextualise the outcomes against which a candidate prognostic test or algorithm is to be assessed.

In this paper, we first review the concepts of prognosis and diagnosis, with a focus on assessment of the severity of febrile illness. We then apply these concepts to define three potential use-cases for prognostic tools in the management of febrile illnesses in resource-limited settings: (1) guiding referrals from the community to higher-level care; (2) informing resource allocation for patients admitted to hospital and (3) identifying patients who may benefit from closer follow-up post-hospital discharge. For each use-case, we explore practical implications for new technologies, with an emphasis on the requirements for putative tests to measure biochemical biomarkers within various healthcare settings in LMICs. We conclude by reflecting on the challenges and knowledge gaps that must be addressed before prognostic tools could be incorporated into routine care settings, drawing on the findings from multiple recent stakeholder consultations.14

Prognosis and diagnosis in the assessment of disease severity

Healthcare providers regularly integrate multiple sources of data (eg, patient demographics, comorbidities, clinical signs and symptoms, and results of radiological and laboratory investigations) to determine the ‘true’ underlying disease or health state of their patient. Depending on the temporal relationship between these baseline data (predictors) and the disease or health state in question (the outcome), these predictions are either diagnostic or prognostic.8

Often the distinction between diagnosis and prognosis is clear: integration of clinical, laboratory and radiological information to predict whether a patient may have infective endocarditis (Duke criteria)15 is a diagnostic process, whereas predicting the probability that an individual may develop active tuberculosis (TB) within the next 2 years based on their demographics, medical history and latent TB infection test result (PERISKOPE-TB)16 is easily recognisable as prognosis (figure 1).

Request permission

For predictors of severity these concepts can become blurred. A patient’s severity reflects their likelihood of a poor outcome and hence predictors of severity are inherently prognostic. However, certain predictors also indicate a patient’s ‘level of severity’ at the time of measurement (eg, peripheral oxygen saturation in a patient with pneumonia) and in this sense, as well as providing prognostic information, can also be considered diagnostic of a patient’s severity at that moment. This is in contrast to other predictors that are primarily harbingers of future deterioration in patients who appear otherwise well (eg, various clinical and laboratory parameters measured during the febrile phase of a dengue or COVID-19 infection).7 17 18

Most predictors used for the assessment of severity fall into the first group, serving both diagnostic and prognostic purposes. For example, many guidelines and tools devised to inform the management of febrile illnesses in resource-limited settings use ‘Danger Signs’ to identify patients who are severely ill at the time of assessment and at high risk of mortality.19 20 Hence, these ‘Danger Signs’ can be considered both diagnostic (of the severity of illness at the time of assessment) and prognostic (for future risk of death). However, their lack of sensitivity and specificity, and high interobserver variability, make their performance poor for these particular purposes.21 22 Improving identification of impending serious illness in patients that lack clinical signs of severity as determined by existing management algorithms is a global public health priority.14

It is often poorly reported whether (and what proportion of) patients had met the predefined severity endpoint (eg, hospital admission, vital organ dysfunction or disease-specific severity scores) at the time the baseline predictors were measured (ie, whether the predictors are serving predominantly diagnostic or prognostic functions).23–25 This is particularly important in community settings where mortality is rare and may not be a feasible or relevant endpoint. Failure to identify a study as prognostic or diagnostic is a common shortcoming in the reporting of clinical prediction research.26 Recent guidance on the design, reporting and assessment of prognostic studies aims to improve this.27 To best leverage this, clearly defined use-cases for prognostic tools in the management of febrile illnesses (table 1; figure 2) are required to standardise data collection, encourage consistency of reporting, contextualise interpretation of results and maximise comparability of findings from disparate studies.

Request permission

Use-cases for prognostic tools in the management of febrile illnesses in resource-limited settings

Technical considerations for prognostic tests of biochemical biomarkers for use in resource-limited settings

Markers of common pathophysiological pathways can improve identification of febrile patients with a poor prognosis.41 However, the potential for prognostic biomarker tests to contribute to febrile illness management in LMICs is inextricably linked to the human and technical capacity in the settings in which they would be deployed, which in turn defines the technology requirements for test design.

For tests performed at the community level, simplicity is key. Tests will require the most robust features to address stability and transport stress across wide temperature and humidity ranges. Additionally, these tests need to be energy-independent and results must be straightforward to interpret. For prognostic biomarker tests this poses a specific design challenge as many emerging biomarkers are concentration dependent, requiring either quantitative or semiquantitative results.11 For improved quantitative assessment, a rapid test reader that can be reliably and affordably used at the community level in LMICs will be required. Reader requirements for use in these settings have been defined but require careful cost–benefit assessment to avoid adding to the biomedical graveyard that already exists in many LMIC primary care contexts.42

For tests employed at the district hospital level or above more advanced infrastructure exists (reliable power supply, ambient temperature control, etc) to enable the use of robustly designed instrumented technologies that account for the varied conditions in laboratory settings common in LMICs. Instrumentation permits technologies capable of high sensitivity measurement and quantification, as well as more versatile throughput, sample processing and analyte detection methodologies.

Future directions

Results of a recent multinational stakeholder consultation indicate that better tools to guide community healthcare providers in their referral and admission decisions are urgently required, in particular to prevent under-referral (figure 3).14 Crucially, future work evaluating candidate prognostic tests or algorithms for use in community contexts must include individuals managed as outpatients.43 Attention must be focused on patients in whom there is clinical uncertainty, rather than on those in whom severe disease can readily be ruled in or out using existing guidelines and algorithms.

Request permission

The ideal prognostic test or algorithm would have sufficient accuracy and also be generalisable across similar settings. Determining generalisability necessitates comparisons between different studies. Although clearly defined use-cases can harmonise study designs and promote consistency of reporting, standardised assessment methods are essential. Tools to support standardised measurement of baseline predictors have been proposed,44 however as the outcome of a febrile illness is influenced by the resources available (a severe outcome may be averted by appropriate treatment) outcome assessment requires a more contextualised approach.

One possibility is to adjust for the differing treatments and interventions a patient receives when data from disparate studies are synthesised. However, treatments and interventions are often applied imprecisely and adjustment for mediating variables is difficult and can introduce selection bias. An alternate solution is to use a standardised tool to assess the overall ‘level of care’ available in different healthcare settings (eg, different treatments, adjunctive procedures and equipment, health worker capacity and provider–patient ratios).45 This is already commonplace in critical care medicine, facilitating comparisons across settings with similar ‘levels of care’.46 This harmonised yet contextualised approach could be particularly advantageous given the heterogeneous settings in which care is provided for patients living in LMICs.

Additional measures of disease severity, such as need for hospital-level care, must be included alongside mortality. Although surrogate outcomes have important limitations (eg, hospital admission may be influenced by patient and provider preferences), a comprehensive outcome set that includes, but is not limited to these surrogates, is important to consider. Mortality, although a ‘hard’ outcome, occurs infrequently in community settings and predicting death may be of limited utility, compared with predicting severe (and in many instances treatable) illnesses.

Many studies evaluate clinical and biochemical prognostic factors in febrile patients.47 Fewer examine the added value of combining biochemical biomarker tests with clinical features. Any new proposed test should add value to the current standard of care. This is particularly important given the added costs and logistical challenges of implementing new tests in decentralised healthcare settings. Many factors determine the optimal sequence for combining different tests, including cost, patient–provider workflow, pretest probability and whether a high NPV or PPV is desired in a particular setting.48 Summarising an algorithm’s prognostic capacity using a single metric (eg, the area under the receiver-operating characteristic curve) is of limited use to a health worker confronted with an individual patient.49

Few studies consider the additional information that trends in different parameters may provide.13 50 Serial measurements of clinical and biochemical parameters may enable more personalised risk prediction than static assessments at a single point in time. To fully realise the benefits of this approach a better understanding of the temporal kinetics of different markers is required.

Finally, once promising prognostic markers are identified, simple, affordable and reliable tests to measure them must be manufactured and supported by robust supply chains to ensure equitable access. Ideally, tests and algorithms would be quantitative, providing a mechanism to adjust cut-offs to achieve the desired NPV or PPV for a particular setting. Candidate predictors already routinely collected for other purposes should be prioritised for evaluation. The growing use of electronic health records may make this more feasible. For clinical features this could include increasing access to technologies such as pulse oximetry and other vital sign devices.51 For biochemical biomarker measurements, simple lateral flow tests (with or without quantitative readers), analogous to those in widespread use for the diagnosis of malaria could be envisioned (table 2).

Prognostic tools that improve risk stratification of patients with febrile illness would have enormous potential to improve patient outcomes and allocation of scarce resources. Each proposed technology requires careful cost-benefit assessment and must be developed in partnership with the healthcare providers working within the targeted contexts. Defining essential product design requirements in consultation with users is essential to ensure usability and promote understanding, acceptance and trust of these technologies. Importantly, donors and implementers must embrace integrated community care and move away from vertical disease-specific models, as the settings where prognostic tools could have greatest impact are precisely the contexts in which diagnosis remains most challenging.