Background
As of 2010, fewer than one-third of the 68 priority countries with the highest levels of child mortality were on track to meet the Fourth Millennium Development Goal (MDG) of reducing mortality in children aged under five by two-thirds between 1990 and 2015. In many countries where overall progress has been made, that progress has been inequitable, with higher mortality among children in marginalised regions, especially remote areas and those affected by armed conflict [
1,
2]. In such settings, the leading causes of childhood deaths are pneumonia, diarrhoea, neonatal causes, and (where present) malaria [
3‐
5].
Humanitarian relief agencies often provide services through mobile clinics to serve populations living beyond the reach of existing facilities. These mobile clinics often operate in chronic crises, where security is poor, logistics are difficult, and mortality is high. An increasing number of people affected by complex emergencies now live spread out across large and remote geographical areas, in places such as Somalia, Afghanistan, the Central African Republic, and Southern Sudan [
3]. They are consequently harder to reach with humanitarian interventions and suffer prolonged excess mortality [
3]. There is strong evidence that these mobile clinics, which usually provide relatively skilled services on an intermittent basis, can make a positive health impact with preventive activities such as vaccinations, hygiene promotion, and antenatal counselling, as well as through certain curative services for chronic or slow-onset diseases such as intestinal parasites, guinea worm, leishmaniasis, onchocerciasis, and trypanosomiasis [
6‐
9]. In many cases, however, mobile clinics also focus on curative care for rapid-onset illnesses, such as malaria, pneumonia, and diarrhoea [
10].
There is also growing consensus around the effectiveness and safety of community-based management of pneumonia by supporting low-skilled health workers to provide oral antibiotics in low-income and remote areas [
11‐
14]. A meta-analysis of nine trials in seven countries by Sazawal
et al found that community antibiotic treatment strategies resulted in an average 24% reduction in total under-five mortality and a 36% reduction in pneumonia mortality in under-fives in high mortality settings [
11]. In Nepal, one of the priority countries on track to meet MDG4 [
2], female community health volunteers (FCHV) save an estimated 6000 child lives each year by providing oral antibiotics for non-severe pneumonia in remote mountain communities. Despite being only semi-literate, Nepal's FCHVs have shown "minimal" evidence of antibiotics misuse, and 30,007 case reviews in 2005-6 revealed "no evidence of emerging co-trimoxizole resistance" [
12].
Despite this evidence, community-based management of pneumonia is not standard practice amongst relief agencies in less stable conflict-affected situations, and the most commonly used guidelines for humanitarian settings are designed primarily for physicians and nurses [
15,
16]. Relief agencies train low-skilled community health workers (CHWs) largely as health promoters and instruct them to recommend referral for children needing antibiotic treatment [
15]. However, CHW referral recommendations often do not lead to action even in stable settings, and CHWs unable to provide medication often lack credibility [
12]. Indeed, even trained health workers often fail to convince families to take a child with severe complications to hospital in a timely manner [
17,
18]. Insecurity reinforces those factors that prevent many families from seeking care in stable environments; financial resources shrink, security concerns increase, travel is restricted, and heightened social tensions may change perceptions of the acceptability of seeking care from particular facilities or individuals. Relief agencies, donors, ministries of health, and other health actors should therefore consider how community case management of childhood pneumonia may compare with mobile clinics and other strategies they employ to reduce child mortality in hard-to-access populations.
This paper aims to generate quantitative evidence for health policy regarding the relative effectiveness of two alternative strategies, mobile clinics and CHWs in fixed posts, for antibiotic treatment of childhood pneumonia in hard-to-access populations. It develops a mathematical decision model that reflects both clinical aspects of pneumonia case management and practical aspects of implementing each intervention in order to provide recommendations to humanitarian agencies, donors, and other key decision-makers.
Discussion
We find that a CHW in a fixed health post would achieve greater mortality reductions than a weekly mobile clinic strategy in the majority of sensitivity analyses conducted. In univariate sensitivity analyses, mobile clinics achieved greater mortality reductions where the probability of seeking care from a mobile clinic was significantly higher than from a CHW, and where mobile clinics visited communities daily. Mobile clinics would also achieve greater mortality reductions than the CHW strategy if the mobile clinic had 100% sensitivity to diagnose and treat children while the CHW strategy had approximately 30% or lower sensitivity, or if a number of variables simultaneously differed substantially from our baseline assumptions. For a number of important parameters, only limited evidence is available and parameter values may differ between contexts. These findings therefore do not conclusively show one strategy to be more effective than the other. Rather, current evidence indicates that a CHW in a fixed post is a viable alternative to mobile clinics, and therefore warrants consideration in high-mortality, hard-to-access areas.
In line with humanitarian decision-making, this model focussed on the short term impact of the alternative strategies; however, the longer term impact on health and health systems should also be taken into account. Mobile clinics may be quicker to implement in response to a sudden-onset crisis, but they may also risk creating dependence and undermining community coping strategies [
33]. Supporting CHWs may foster sustainable local development and be more readily integrated into the public health system, but identifying potential CHWs is not always straightforward [
34], especially when social tensions are high and different individuals are acceptable to different groups. Both strategies rely upon effective and reliable drugs supply and CHWs especially require regular supervision and support to be effective [
12,
34].
Agencies with the resources to respond quickly to an emerging crisis by deploying mobile clinics should consider initiating a transition to community-based treatment from their very first visit by identifying existing or potential CHWs to train. The role of the mobile clinic can then evolve into providing regular high-quality supplies and supervision to CHWs. With preparation, this supportive approach can also ensure that treatment remains available even if insecurity or other obstacles prevent the mobile team from visiting for a period. By supporting CHWs to treat pneumonia and other acute diseases, the mobile team could potentially also use its visits to engage in preventive interventions, such as immunizations, and to provide specialized treatment for more chronic conditions.
By providing a direct quantitative comparison of the health impact of these two strategies for reaching hard-to-access populations, this paper responds to calls from leaders in epidemiology and public health for research into implementation strategies in high-mortality settings [
1,
35,
36]. Modelling can be a useful tool to foster rational decision-making based on explicit assumptions and scientific evidence for hard-to-access populations in low-income areas. While computationally intensive, the model is transparent and intuitive, and the decision analytic approach is adapted to the needs of decision-makers. As new data comes to light, parameter values can be changed and aspects of the model can be refined.
In quantifying the uncertainty in the decisions, the model also highlights the need for humanitarian and development programmes to undertake research in hard-to-access areas to reduce decision uncertainty. Improving the evidence base would require focussed data collection on parameter values from a given setting where both strategies are implemented or formal comparisons of the two approaches, although the latter could potentially have untenable ethical implications. Cost-effectiveness analyses would offer a more complete evidence base for deciding between the two strategies, as even in emergency interventions, efficiency is required to maximize the impact of the finite human, logistic, and financial resources that can be mobilized. Although the relative costs of the interventions are likely to vary considerably according to the specificities of each context, in general mobile clinics have been a resource-intensive intervention [
10]; this suggests that a CHW strategy will usually be the more cost-effective option of the two if local contextual parameters indicate that the two strategies would have similar curative effectiveness.
This model can and should also be extended to the treatment of other key childhood diseases. In addition to diarrhoea and malaria, neonatal deaths constitute a significant proportion of under five mortality and many are attributable to invasive bacterial infections, including pneumonia, sepsis, and meningitis, which can be treated with antibiotics [
2,
35,
37,
38]. Sazawal
et al provide some evidence for community management of neonatal pneumonia [
11], and a number of countries have begun to include protocols for neonatal care within IMCI guidelines [
39], but further research into community case management of neonates, especially in hard-to-access areas, is likely to offer a significant contribution to public health.
As many programmes have turned towards CHWs to increase coverage of their interventions, it has become clear that CHWs cannot do everything, especially if they are unpaid [
34,
40]. Detailed national and sub-national analysis is required to understand how integrated community-based approaches can be implemented in high mortality hard-to-access areas [
38], including those affected by instability.
Limitations
By its very nature, modelling entails simplifying reality to make useful predictions. Some important aspects of childhood pneumonia were excluded because they were not expected to have a significant impact on the relative effectiveness of the two strategies being compared. Certain unexplored issues may, however, have a more significant influence on the results. Sensitivity analysis was not conducted to explore the responsiveness of the outcome to changes in the median duration of disease, or how treatment adherence may differ between mobile clinics and fixed health posts. Analysis of alternative models of care-seeking behaviour revealed the possibility that mobile clinics may be the dominant strategy when attendance is very high and the mobile clinic is able to consult all patients who attend. In practice, mobile clinics may not be able to see all patients and so the degree to which these findings hold true will depend on the effectiveness of triage. Insecurity or other constraints may also cause a mobile team to miss planned visits, which, in addition to reducing this strategy's impact, may potentially also have untoward consequences for care-seeking practices, although evidence on this latter point is not currently available. Similarly, we assumed that CHWs would be available 100% of the time, and did not model the potential repercussions of the CHW running out of antibiotics or being unavailable. Estimating the true effect of either intervention would also require considering simultaneous diagnosis (or misdiagnosis) and treatment of co-morbidities, such as malaria.
Standard errors used for sensitivity analysis were estimated somewhat crudely and may therefore have underestimated the uncertainty in the results. All children began the model in the healthy state and a burn-in period was not used, which likely biased the expected number of child deaths downwards. While sensitivity analysis was conducted to examine the impact on results of different pneumonia incidences, outcomes were not examined in contexts of heightened vulnerability, such as would accompany high malnutrition or HIV rates, where the rate of disease progression and the CFR would increase [
41]. Perhaps most influentially, the model does not measure the negative impact of unnecessary antibiotic treatment, incorrect prescriptions, and poor adherence on the individual child or on the wider community, particularly with respect to the development of resistant bacterial strains. Lim provides an example of how such negative outcomes can be included in a decision tree [
42].
Like most studies, this analysis did not formally address methodological and model uncertainty, which Brisson and Edmunds have shown can play an even greater role than parameter uncertainty [
43]. Parameter estimates themselves were based on a non-systematic review of the literature and the Delphi method was not used to elicit expert opinion on parameter values, although this method also has drawbacks [
44]. The number of relevant studies with quantitative data on certain topics, such as care-seeking behaviour and treatment adherence in crises, was limited. While some grey literature was obtained, studies and reports, particularly from within the humanitarian community may have been missed. A broader limitation of this project is the lack of rigorous data collected by humanitarian agencies and ministries of health working in hard-to-reach areas.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
CP designed the study, developed the model, carried out analyses and co-wrote the paper. BR supervised the study and co-wrote the paper. FC designed and supervised the study and co-wrote the paper. All authors read and approved the final manuscript.