Annually over 800,000 deaths of children under 5 years old are attributable to acute lower respiratory infections worldwide ¹ . Due to its high prevalence, pneumonia is one of the primary diseases addressed by the WHO Integrated Management of Childhood Illness (IMCI) guidelines, the standard of care at many frontline, outpatient healthcare facilities in low-income and middle-income countries (LMICs). These guidelines rely on health worker recognition of clinical ‘danger signs’ such as inability to feed, lethargy, intractable vomiting, or convulsions to determine if a child should be referred for further evaluation by a trained medical provider at a hospital and receive more intensive treatment. Two key pneumonia case classification policy changes were made in the 2014 IMCI revision. First, when pulse oximetry is available, children aged 2–59 months old with a peripheral oxyhemoglobin saturation (SpO ₂) <90% (i.e., hypoxemia) should be referred for hospital care. Second, children with chest indrawing, but no general danger signs, were reclassified as appropriate for outpatient management with oral antibiotics, even when SpO ₂ is not measured.

There have been several attempts to determine if chest indrawing pneumonia, without the presence of general danger signs, can be appropriately managed as an outpatient ^{2– 6} . One effort by Agweyu et al. retrospectively examined the implication of shifting hospitalized Kenyan children with chest indrawing but without general danger signs into outpatient care. The authors reported that children previously categorized as severe pneumonia (i.e., chest-indrawing but no general danger signs) had an in-hospital case fatality ratio (CFR) of 2.7% ² . Similar to many LMIC healthcare settings, pulse oximetry was unavailable, so the authors could not comment on the impact of identifying hypoxemia.

The Malawi Ministry of Health (MOH) adopted the 2014 IMCI pneumonia guidelines in 2018. However, based on anecdotal evidence [SH], some clinicians are hesitant to adopt the revised guidelines into practice due to concerns that the change may result in increased out of hospital complications and/or community deaths. This may in part stem from pulse oximetry’s limited availability in outpatient settings throughout Malawi, and a high proportion of children dying from pneumonia at home after seeking hospital care ^{7, 8} . A non-inferiority trial conducted in Malawi showed HIV-uninfected children with chest indrawing pneumonia could be safely treated as outpatients with 3 versus 5 days of amoxicillin. However, all enrolled children were screened with pulse oximetry and none had a SpO ₂<90% ⁴ . In a study linking community health worker (CHW) and outpatient pneumonia episodes with 30-day mortality, augmentation of existing health centre guidelines with pulse oximetry identified high risk cases that would otherwise be missed by clinical guidelines alone ⁸ . The current CHW guidelines, unlike those for outpatient settings, recommend hospital referral of children with chest indrawing. The added value of pulse oximetry for identifying referral eligible children was less when chest indrawing triggered referral ⁹ .

To some degree chest indrawing may be a clinical marker of respiratory insufficiency or impending failure. Chest indrawing is the paradoxical inward pulling of the soft tissue below the ribs during inspiration. It often occurs due to increasing negative intrapleural pressure generated by patients attempting to expand stiff, diseased lungs, although in younger children and infants with more compliant, cartilaginous chest walls it may also occur when the lungs are healthy. Chest indrawing increases lung tidal volumes and minute ventilation to improve overall ventilation ¹⁰ . Aside from respiratory rate and SpO ₂, chest indrawing is the only indicator of lung mediated respiratory distress in the 2005 IMCI guidelines. The 2014 IMCI guidelines do not contain any lower airway sign that triggers hospital referral in settings where pulse oximetry is not used.

We aimed to determine the impact of implementing the 2005 or 2014 IMCI pneumonia guidelines on inpatient case fatality rates with either universal or no pulse oximetry availability in routine care settings.

This analysis includes 13,413 cases with an overall CFR of 2.8% ( Figure 1). Clinical characteristics of the cohort are described in Table 1. A smaller proportion of the hospitalized cases were girls compared with boys (P<0.001), with 3.7% without a recorded sex.

In Table 2 we compared, using the 2005 and 2014 guidelines, the proportion of cases diagnosed with pneumonia meeting hospital referral and/or admission criteria versus outpatient management and their respective CFRs. The CFRs of children meeting outpatient care criteria per the 2005 and 2014 guidelines were not statistically different (0.8 vs 0.7%). Due to missing data for parameters such as respiratory rate, chest indrawing and danger signs (independent of pulse oximetry reading), 720 cases (5.4%) and 16 deaths (4.3%) with an overall CFR of 2.2% could not be classified as having a specific pneumonia severity. Applying the 2005 guidelines without pulse oximetry, 11,484 cases and 99.1% (347/357) of the pneumonia severity classifiable deaths met hospital referral criteria. Notably, an additional 1,089 cases were unable to be categorized by the 2014 guidelines due to unrecorded pulse oximetry data (deaths=54, CFR 5.0%). In comparison, when using the 2014 guidelines with pulse oximetry, 3,106 cases and 79.2% (240/303) of categorizable deaths met referral criteria. This is a 73.0% reduction in hospital referral cases, however the number of deaths categorized as non-severe increased from 10 (CFR 0.7%) to 63 (CFR 0.8%). When SpO ₂ is treated as an optional criteria, consistent with how the 2014 guidelines are written, the majority of these cases were re-classified as non-severe, outpatient appropriate (N=1,378, deaths = 60, CFR 4.4%). Of these cases 96.9% had chest indrawing (N = 1,335, deaths =59, CFR 4.4%).

Pulse oximetry is not universally available in outpatient clinics or hospitals in Malawi. Under this assumption we calculated CFRs for differing pneumonia severities without considering SpO ₂. The CFR for patients with 2014 non-severe pneumonia, without pulse oximetry, was twice that of those categorized as having 2005 IMCI guideline non-severe pneumonia (1.5% vs 0.8%, P=0.052). Under the scenario of implementing the 2014 WHO guidelines with pulse oximetry, in comparison to its implementation without pulse oximetry, the CFR of non-severe pneumonia decreased by over 50% (1.5% without pulse oximetry vs 0.7% with pulse oximetry, P<0.001). However it is crucial to emphasize that many of the pre-2014 WHO guideline non-severe pneumonia cases and associated deaths were unable to be categorized due to missing pulse oximetry data. We were unable to categorize the pneumonia severity of 14.5% (N= 54) of the total deaths due to unrecorded SpO ₂ data. Our analysis, presented in Table 3, suggests that oxygen saturation was MNAR as an unrecorded SpO ₂ was an independent mortality risk factor.

We explored the impact of applying the 2014 WHO IMCI guidelines, using a large existing dataset of children admitted to hospital with clinically defined pneumonia. If the 2014 WHO IMCI guidelines are implemented without compulsory pulse oximetry measurement of children evaluated for pneumonia in Malawian outpatient settings, some children that could benefit from inpatient oxygen treatment will likely be missed. This could result in increased child pneumonia deaths. A direct comparison of non-severe, outpatient pneumonia CFRs showed a two-fold higher CFR when using the 2014 rather than the 2005 guidelines without pulse oximetry, from 1.5% vs 0.8% (P=0.052). However, implementation of the 2014 IMCI guidelines with compulsory pulse oximetry decreased non-severe pneumonia CFR by 50%, to 0.7% for this group.

We found that the inability to obtain a pulse oximetry reading was MNAR, suggestive as an independent mortality risk factor. We found a similar finding amongst hospitalized Malawian infants under 2 months old ¹⁴ . In a separate study evaluating pulse oximetry use in outpatient settings, we found that 1 out of the 32 cases that were not clinically eligible for referral and had an unrecorded pulse oximetry level died within 30 days ⁹ . It is challenging to interpret this finding because although we witnessed study staff struggle with obtaining a pulse oximetry reading in some children, it is plausible that these data may be missing not at random but for other reasons. For instance, clinicians may not have paused to measure a pulse oximetry reading in children who presented to care when obviously critically ill. Biologically, the inability to measure pulse oximetry could reflect poor peripheral perfusion, which occurs in high systemic vascular resistance states such as shock. There are ongoing studies evaluating if a peripheral perfusion index, derived from the plethysmography signal of pulse oximetry, could be used as an early marker of microcirculatory dysfunction during states such as sepsis and compensated hypovolemic shock before changes in mean arterial blood pressure, which is currently used to determine macrovascular collapse ^{17– 19} .

In contrast with Agweyu’s findings from Kenya, we were able to explore the potential implications of the 2014 guidelines in the context of pulse oximetry. Our study has similar findings, using retrospective data from hospitalized cases, in that there was an unacceptably high CFR amongst children diagnosed with non-severe pneumonia using the 2014 guidelines without pulse oximetry. There are some notable differences between the study populations, with the Kenyan study reporting a higher overall CFR (5.2% vs 2.8%), a higher CFR in children classified as non-severe pneumonia (2.7% vs 1.5%), and fewer cases from malaria-endemic regions (33% vs 100%) ² . These are important distinctions as the case fatality difference could be attributed to a variety of factors, including an improved survival benefit in Malawi with inpatient use of pulse oximetry and subsequent oxygen treatment. While this is conjecture, the approximate 50% reduction in mortality between the settings is in line with findings from Nigeria, showing a similar reduction in inpatient pneumonia mortality following pulse oximetry introduction ²⁰ . Another explanation could be the difference in malaria prevalence amongst these populations, as the IMCI guidelines for pneumonia diagnosis overlap with the clinical diagnosis of malaria ²¹ . Given that they did not have pulse oximetry data and our dataset did not include information about dehydration or pallor, direct comparisons of independent risk factors for pneumonia-associated mortality between the two studies is problematic. Though we do report mortality risk factors in this study, the intention was solely to explore if pulse oximetry was MNAR. We have previously published our findings, using this same dataset, on pneumonia related mortality risk factors and developed a clinical prediction rule, the Respiratory Index of Severity in Children – Malawi (RISC-Malawi), though excluding patients without a recorded pulse oximetry reading ²² . An important finding was that girls had a higher adjusted odds for in-hospital pneumonia related mortality.

Notably both the Kenyan and our studies report fewer girls were hospitalized with pneumonia, though female sex was an independent mortality risk factor. We observed the same pattern when examining pneumonia related mortality risk factors in Malawian infants under two months old, and this is consistent with a large metanalysis of acute lower respiratory illness mortality risk factors ^{14, 23} . However, in our previous Malawian studies, utilizing data collected concurrently as this study, in outpatient settings (community health workers and health centres), we found that similar proportions of boys and girls, aged 0–59 months old, sought evaluation for pneumonia ^{9, 14} . Our incidental findings that an equal proportion of girls and boys presented for outpatient pneumonia care, fewer girls were hospitalized, yet being a girl was an independent pneumonia-related mortality risk factor cannot go ignored. These findings are consistent with those of a retrospective qualitative study assessing care seeking patterns in the same region of Malawi (Mchinji) that suggested that girls presented to the hospital more often than boys but were admitted less often ⁸ . Differences in the type of care girls receive and their outcomes must be examined.

Pulse oximetry implementation, while crucial to health systems strengthening, is not without challenges. First, more work must be done to identify context appropriate SpO ₂ thresholds at which to refer patients for hospital evaluation, hospitalize patients, and initiate or wean supplemental oxygen. Previous findings suggest that moderate hypoxemia of SpO ₂ 90–92% is a mortality risk factor in Malawian children under 5 with acute lower respiratory infections ^{22, 24} . One hypothesis for this finding is that pulse oximeters may be inaccurate in children with dark skin tones. A research report from the United States suggested that pulse oximetry may over-estimate SpO ₂ in adults with dark skin, potentially leading to missed hypoxemic cases and further exacerbating racial health disparities ^{25, 26} . While this study had some methodological limitations such as using paired pulse oximetry and arterial blood gas readings that were time-delayed by up to 10 minutes and only adjusting for poor peripheral perfusion in one single center cohort, this concept warrants investigation in children. While there are clinical trials in low- and high-income countries exploring the optimum SpO ₂ threshold for initiation of supplemental oxygen, they have prompted debate on the ethics of such studies and, at the time of writing, have not reported findings ^{27, 28} .

Second, pulse oximetry programs must be implemented with a clearly defined clinician mentoring and longitudinal implementation plan. Even when available in health facilities, barriers exist to pulse oximeter utilization ^{29, 30} . Enoch et al. conducted a comprehensive, multi-site, mixed methods study in Kenya to assess pulse oximetry use and limitations to its uptake by different types of healthcare workers ³¹ . They found an interdependence of pulse oximetry usage and oxygen therapy availability. This suggests that, in order to ensure usage, pulse oximetry must be implemented in concert with supplemental oxygen and other types of reparatory support, rather than as a separate health system intervention.

Cost savings from reduced hospitalizations at both the household and health system levels, compounded with the reduced clinical burden on already limited hospital staff, may have significant spillover effects that we cannot capture in this analysis. It is plausible that, in certain settings, pulse oximetry implementation may in effect pay for itself by reducing downstream costs. As health systems move towards universal integration of pulse oximetry, these potential savings should be accounted for in order to make accurate comparisons with other high priority interventions.

The main limitation of this analysis is that we extrapolate potential outpatient outcomes from hospital inpatient data. We may be underestimating, to varying degrees, the true CFR of these groups. Hospitalized patients may have received more effective interventions, such as intravenous antibiotics, that they would not have received if discharged home. Alternatively, hospitalization may place these children at risk of contracting hospital-acquired infections and other communicable diseases. We found that pulse oximetry recordings were MNAR. As we have previously reported, it is unclear if these data are missing because patients were critically ill and clinicians were unable to obtain an appropriate pulse oximeter read due to poor perfusion or if they simply did not record the measurement ¹² . Our anecdotal experience suggests the former scenario is not uncommon. In real world clinical settings, not all children with severe disease present with the limited criteria in the IMCI guidelines. Clinician judgement of illness severity was not recorded in this data set, so we are unable to determine how many cases classified as non-severe would have been hospitalized regardless of the guideline recommendation.

These data strongly support compulsory pulse oximetry use if Malawi is to fully adopt the 2014 IMCI pneumonia guidelines in outpatient facilities. Our findings suggest that the implementation of the 2014 IMCI guidelines would likely result in a marked decrease in under-five pneumonia related hospitalizations. We would recommend that children under 5 years old with chest indrawing in whom pulse oximetry is unable to be obtained, still warrant referral for further evaluation. It is imperative for country level ministries of health in other LMICs to assess the potential effects of the 2014 policy changes in their own settings before widespread implementation. In Malawi, implementation of the 2014 WHO IMCI guidelines, without standard pulse oximetry measurement in children evaluated for pneumonia in outpatient settings may result in missed hospital referral opportunities of high-risk cases.