Bulletin of the World Health Organization

Emergency triage assessment for hypoxaemia in neonates and young children in a Kenyan hospital: an observational study

Michael K Mwaniki a, D James Nokes a, James Ignas a, Patrick Munywoki a, Mwanajuma Ngama a, Charles RJC Newton a, Kathryn Maitland a & James A Berkley a

a. Centre for Geographic Medicine Research, Kenya Medical Research Institute, PO Box 230, Kilifi, 80108, Kenya.

Correspondence to Michael Mwaniki (e-mail: mmwaniki@kilifi.kemri-wellcome.org).

(Submitted: 01 November 2007 – Revised version received: 24 June 2008 – Accepted: 04 August 2008 – Published online: 13 February 2009.)

Bulletin of the World Health Organization 2009;87:263-270. doi: 10.2471/BLT.07.049148


Recognizing and correcting poor oxygenation early is an essential aspect of paediatric emergency care. The management of hypoxaemia includes establishing an open airway and alveolar ventilation, providing supplemental oxygen, restoring circulation and treating the underlying cause. Hypoxaemia correlates with mortality,1 and early assessment and prompt oxygen therapy probably improve survival.2,3

In developing countries, bottled oxygen is expensive and commonly in short supply. However, most hospitals providing secondary care cannot perform pulse oximetry or measure arterial oxygen concentration to properly target oxygen delivery. Furthermore, clinical assessment is often carried out by relatively inexperienced health workers, as a result of which simple algorithms for identifying hypoxaemia have been developed and studied. However, the studies have been conducted almost entirely among children with lower respiratory tract infection (LRTI),49 and little prospective validation has been performed.

Certain clinical signs have been identified as predictive of hypoxaemia: a fast respiratory rate for age, lower chest wall indrawing, grunting, head nodding, cyanosis, the absence of crying during the examination, and the inability to breastfeed or drink. No single sign is a reliable predictor, and sensitivity is generally low for any single sign.10 WHO currently recommends giving oxygen urgently to children with cyanosis, signs of shock, severe respiratory distress, the syndrome of “very severe pneumonia” or a respiratory rate of 70 breaths per minute (min) or more.11

Given the paucity of data outside the context of LRTI, we first aimed to describe the prevalence of hypoxaemia in relation to the WHO clinical syndromes of “pneumonia” at admission and the final diagnoses made at discharge. We then tested the hypothesis that hypoxaemia could be identified on the basis of clinical signs by studying a large population of children consecutively admitted to a rural, Kenyan district hospital, irrespective of their diagnosis. We report findings from 15 289 children and neonates admitted during a 3-year period and validate our findings using data from 4695 admissions over a fourth year.


Study setting

Since 1998 we have conducted continuous inpatient surveillance to describe the causes and clinical features of common illnesses among children admitted to Kilifi District Hospital in Kenya.12,13 The hospital is located at sea level in an area with endemic malaria. Government-employed clinical officers not involved in research refer children to the paediatric ward (40 beds) or the high-dependency unit (6 beds). On admission, discharge or death, standardized clinical and laboratory data are collected by clinical officers, who in Kenya receive 3 years of basic medical training, or by fully trained medical officers with less than 5 years of paediatric experience.12,13 For this analysis, we used data collected at admission from January 2002 to December 2005 and final discharge diagnosis for each consecutive admission over the same period of time. The Kenyan National Scientific and Ethical Review Boards and the Coventry Research Ethics Committee approved the study.

Clinical definitions

WHO defines a set of clinical syndromes of “pneumonia” on the basis of clinical history and clinical signs at presentation for the purpose of determining the need for admission and the type of antibiotic treatment required.11,14 The definitions have high sensitivity with respect to LRTI, but they lack specificity. The “syndrome of pneumonia” is defined as a history of cough or difficulty breathing plus an elevated respiratory rate for age (if < 2 months, ≥ 60 breaths per min; if 2–11 months, ≥ 50 breaths per min; and if 1–5 years, ≥ 40 breaths per min), and no signs of severe pneumonia syndrome. “Severe pneumonia syndrome” is defined as a history of cough or difficulty breathing plus lower chest wall indrawing or nasal flaring and no signs of very severe pneumonia syndrome. “Very severe pneumonia syndrome” is defined as cough or difficulty breathing plus any of the following signs: cyanosis, inability to drink or breastfeed, convulsions, lethargy or unconsciousness.

Regarding signs not included in the above definitions, prostration was defined as the inability to sit unsupported for a child ≥ 9 months of age, or the inability to breastfeed or drink for a child < 9 months of age. Level of consciousness was assessed using the Blantyre coma scale.15 Shock was defined as the presence of any one of the following signs: capillary refill delayed for > 3 seconds (s), a noticeable temperature gradient or a weak pulse volume. For this analysis, we defined severe anaemia as a haemoglobin concentration ≤ 4 g/dl. Malaria screening was performed on every admission with thick and thin blood films stained and examined by standard techniques.

Oxygen saturation (SaO2) was measured at admission by trained clinical assistants using fingertip pulse oximetry (Nellcor Puritan Bennett NPB-40, United States of America). We defined hypoxaemia as an SaO2 < 90%. Bottled oxygen was always available for hypoxaemic children. LRTI and other conditions were managed according to current WHO guidelines.11 All final diagnoses, including LRTI, were made prospectively by the discharging clinician after a review of admission history, inpatient management notes and all available laboratory and radiologic tests.

For this analysis, we first used the WHO pneumonia syndromes to classify children according to the clinical findings observed at admission. We then examined up to two final diagnoses recorded at discharge or death using all available clinical, laboratory and radiological information. Thus, the sum of all final diagnoses may exceed 100%.

Statistical methods

At the point of contact, clinical data were entered into a FileMaker Pro database 5.5v1 (FileMaker Inc., USA). Laboratory data were double entered and verified using FoxPro 2.5b for Windows (Microsoft Corporation, Seattle, WA, USA). Stata version 9.2 (Stata Corp, College Station, TX, USA) was used for the final analysis.

We split the data into a 3-year derivation set and a 1-year validation set. We divided admissions into three age groups because we hypothesized that the prevalence, causes and predictors of hypoxaemia could differ among neonates, young infants and older children. We first determined the prevalence of hypoxaemia in all age groups. We then examined hypoxaemia in relation to the WHO clinical syndromes of pneumonia at admission and final diagnosis at discharge. After excluding children in whom oximetry was not performed or with a failed oximeter reading or who had known cardiac disease or asthma, we developed predictive models of hypoxaemia in three stages by means of likelihood ratios (LRs). The LR indicates the degree to which a positive or negative result increases or lowers, respectively, the odds of having the disease in question (hypoxaemia in this case). LRs change less under the influence of disease prevalence than sensitivity and specificity, and they may be calculated for multiple test results.16 For each putative clinical sign used to predict hypoxaemia, we first calculated the crude LR for a positive result and for a negative result. Signs with crude LR ≥ 2.0 or ≤ 0.5 were considered clinically useful. We then used the method of Speigelhalter and Knill-Jones to adjust for the confounding effects of related variables, as determined by multivariate analyses, in four separate groups: respiratory, cardiovascular, neurological and general signs.13,17,18 In the third step, we constructed practical prediction rules for hypoxaemia using signs with adjusted likelihood ratios of ≥ 2.0 or ≤ 0.5, prioritized by strength of prediction and clinical practicability. We finally evaluated the final rules using the area under the receiver operating characteristic (ROC) curves using the validation data set. Distributions of age were compared by means of the Wilcoxon rank sum test.


Between 2002 and 2004, 15 401 children were admitted (derivation set) to the hospital. Of these children, 13 183 were aged ≥ 60 days; 991 were aged 7–59 days, and 1115 were aged < 7 days. Excluded from the study were 112 children: 35 without pulse oximetry results; 59 with known cardiac disease and 18 with known asthma.


Admissions aged  60 days

Among the 13 183 children admitted, the proportion of males was 56% and the median age was 32 months (interquartile range: 11–42). The most common final diagnoses were malaria (4982, or 38%), LRTI (2869, or 22%), gastroenteritis (2026, or 15%), malnutrition (1261, or 10%) and severe anaemia (717, or 5%).

Hypoxaemia was found in 693 (5.3%) of the children admitted. The median age of the children with and without hypoxaemia was 18.9 and 22.9 months, respectively (P = 0.036). The most frequent final diagnoses among hypoxaemic children were malaria (244, or 35%), LRTI (221, or 32%), malnutrition (68, or 10%) and gastroenteritis (49, or 7%). Severe anaemia was found in 30 children (< 1%). Overall, 753 (6%) children died, including 150 of the 693 (22%) with hypoxaemia. Irrespective of the final clinical diagnosis, hypoxaemia was strongly associated with inpatient death (age-adjusted risk ratio, RR: 4.5; 95% confidence interval, CI: 3.8–5.3).

Very severe pneumonia syndrome

Of the children admitted, 2525 (19%) satisfied the WHO criteria for very severe pneumonia: 291 (12%) of these children were hypoxaemic and 287 (11%) died. The final diagnoses included severe malaria (1268, or 50%), LRTI (514, or 20%), gastroenteritis (185, 7%), acute febrile convulsion of unknown cause (without malaria, meningitis or epilepsy) (181, or 7%), malnutrition (173, or 7%), meningitis (75, or 3%) and severe anaemia (56, or 2%). In 86 (3%) children, both malaria and LRTI were present. Of the 291 hypoxaemic children, 119 (41%) had a final diagnosis of malaria and 82 (28%), of LRTI. The case fatality rate among hypoxaemic children was 30% (89/291).

Severe pneumonia syndrome

Of the children admitted, 2267 (17%) satisfied the WHO criteria for severe pneumonia, 156 (7%) were hypoxaemic and 107 (5%) died. The final diagnoses included LRTI (1654, or 73%), malaria (349, or 15%), malnutrition (180, or 8%), gastroenteritis (49, or 2%) and severe anaemia (43, or 2%). A further 155 (7%) children had both malaria and LRTI. Of the 156 hypoxaemic children, 113 (72%) had a final diagnosis of LRTI and 18 (12%), of malaria. The case fatality rate was 16% (25/156) among children who were hypoxaemic.

Pneumonia syndrome

Of the children admitted, 697 (5%) satisfied the WHO criteria for pneumonia, and of this group, 14 (2%) were hypoxaemic and 33 (5%) died. Frequent final diagnoses among these children were malaria (245, or 35%), LRTI (197, or 28%), malnutrition (93, or 13%), gastroenteritis (76, or 11%) and severe anaemia (26, or 4%). Of the 14 hypoxaemic children, 8 (57%) had a final diagnosis of malaria and 4 (29%), of LRTI. The case fatality rate was 21% (3/14) among children who were hypoxaemic.

Overall, 461 of 693 (67%) hypoxaemic children who were admitted met the clinical criteria for one of the WHO pneumonia syndromes, and 215 of 693 (31%) of the hypoxaemic children admitted had a final discharge diagnosis of LRTI (Fig. 1).

Fig. 1. Distribution of LRTI and hypoxaemic cases across the WHO “pneumonia” syndromes
Fig. 1. Distribution of LRTI and hypoxaemic cases across the WHO “pneumonia” syndromes
LRTI, lower respiratory tract infection.
Admissions aged 7–59 days

Of the 991 children aged 7–59 days who were admitted (57% males) 78 (8%) were hypoxaemic and 82 (8%) died. The main final diagnoses were neonatal sepsis (497, or 50%); LRTI (301, or 30%) and neonatal jaundice (37, or 4%). Of the hypoxaemic children, 37 (47%) had a final diagnosis of LRTI and 25 (32%), of neonatal sepsis. Among 78 hypoxaemic children, 22 (28%) died. Hypoxaemic children had higher mortality (age-adjusted RR: 4.3; 95% CI: 2.8–6.6).

Admissions in the first week of life

Of 1105 children (58% males) admitted during the first week of life, 206 (19%) were hypoxaemic on admission, and 336 (30%) died. Common final diagnoses were neonatal sepsis (600, or 54%); prematurity (216, or 20%); birth asphyxia (153, or 14%) and uncomplicated neonatal jaundice 98 (9%). Among those with hypoxaemia, neonatal sepsis (80, or 39%), birth asphyxia (62, or 30%), prematurity (49, or 24%) and neonatal jaundice (7, or 3%) were common. Among 206 children admitted with hypoxaemia, 118 (57%) died. This represented 35% of all deaths. Hypoxaemia was associated with inpatient death (age-adjusted RR: 2.4; 95% CI: 2.0–2.8).

Clinical indicators of hypoxaemia

Admissions aged  60 days

Clinical signs that appeared predictive on univariate analysis (LR ≥ 2.0 or ≤ 0.5) were lower chest wall indrawing, nasal flaring, central cyanosis, irregular breathing, deep breathing, stridor, auscultatory crackles, dullness on percussion, respiratory rate ≥ 60 breaths per min, weak pulse volume, delayed capillary refill (≥ 3 s), heart rate < 80 beats per min, impaired consciousness with a Blantyre coma scale < 3, prostration, convulsion on admission, restlessness and hypothermia (Table A1 in Appendix A, available at: http://www.kemri-wellcome.org/images/appendixa.pdf). On multivariate analysis within each group, central cyanosis, heart rate < 80 beats per min, irregular breathing, impaired consciousness with a Blantyre coma scale < 3, respiratory rate ≥ 60 breaths per min, delayed capillary refill (≥ 3 s) and weak pulse volume remained predictive of hypoxaemia (Table A2 in Appendix A).

The overall sensitivity and specificity of these signs in the validation data set are shown in (Table 1). Although lower chest wall indrawing did not meet our criteria as a useful predictor when included with other signs in the respiratory group, we re-tested it with the final list of predictive signs because it is easy to recognize clinically and WHO recommends it as an indication for oxygen therapy when severe. Although including lower chest wall indrawing improved sensitivity, significant specificity was lost (Table 1). The area under the ROC curve of these signs in the validation set was 0.80 (95% CI: 0.77–0.84) overall, 0.77 (95% CI: 0.73–0.82) among children who fulfilled the clinical criteria for a pneumonia syndrome, and 0.81 (95% CI: 0.75–0.89) among children who did not fulfil them.

Admissions aged 7–59 days

The clinical signs that appeared predictive on univariate analysis were cyanosis, irregular breathing, nasal flaring, crackles on auscultation, respiratory rate > 80 per min, weak pulse, delayed capillary refill (≥ 3 s), heart rate < 100 beats per min, absent cry, unresponsiveness to painful stimuli, convulsions present on admission, restlessness, a bulging fontanelle and axillary temperature < 36°C (Table A3 in Appendix A). Cyanosis, heart rate < 100 beats per min, respiratory rate > 80 breaths per min, irregular breathing, unresponsiveness to painful stimuli, restlessness, inability to breastfeed and delayed capillary refill (≥ 3 s) remained predictive after controlling for other signs (Table A4 in Appendix A). For the reasons given above, we also re-tested lower chest wall indrawing in the final list of signs. The sensitivity and specificity of the signs are shown in Table 2. The area under the ROC curve for the prospective validation data set was 0.77 (95% CI: 0.71–0.84).

Admissions the first week of life

In this group of children, predictive signs on univariate analysis were cyanosis, irregular breathing, deep breathing, respiratory rate > 80 breaths per min, auscultatory crackles, indrawing, nasal flaring, heart rate < 100 beats per min, weak pulse, delayed capillary refill (≥ 3 s), absent cry, unresponsiveness to painful stimuli, history of birth asphyxia and birth trauma (Table A5 in Appendix A). Signs that remained significantly associated on multivariate analysis were cyanosis, heart rate < 100 beats per min, irregular breathing, absent cry, unresponsiveness to painful stimuli, respiratory rate > 80 breaths per min, history of birth asphyxia, delayed capillary refill (≥ 3 s) and inability to breastfeed (Table A6 in Appendix A). The sensitivity and specificity of the signs are shown in (Table 3). For these signs, the area under the ROC in the validation set was 0.76 (95% CI: 0.72–0.81).


Most studies of the prevalence and clinical signs of hypoxaemia have focused on children with LRTI. We therefore chose to examine unselected paediatric admissions. Hypoxaemia was present in 5%–19% of all children admitted to a Kenyan district hospital, depending on age, and this concurs with recent data from the Gambia.19 Although two-thirds of hypoxaemic children aged ≥ 60 days presented with one of the clinical syndromes of pneumonia, only one-third had a final diagnosis of LRTI. Most children meeting the WHO criteria for very severe pneumonia were finally diagnosed with severe illnesses other than LRTI, most commonly severe malaria or shock. Children who have metabolic acidosis in severe malaria may present with respiratory distress (Kussmaul’s respiration) and could thus overlap with the syndrome of very severe pneumonia.11,20,21 The situation may differ in geographical regions where malaria is not endemic.

Clinical signs were relatively poor predictors of hypoxaemia. The most sensitive rules we developed still missed 5–15% of hypoxaemic children, depending on age. As in other studies, specificity posed a significant problem, as it was markedly reduced when we included lower chest wall indrawing (82–67% among children ≥ 60 days of age and 77–50% among infants aged 1 week to 59 days) and the inability to breastfeed among infants in the first week of life (66–45%). Had we followed these rules, even after excluding lower chest wall indrawing at all ages and an inability to breastfeed in the first week of life, we would have given bottled oxygen to a total of 862 of 4695 (18%) children in the validation set who were not hypoxaemic by pulse oximetry. Although the false positive and false negative assessment results were fewer than reported recently from Papua New Guinea, they still represent a significant financial burden.22 At an estimated daily cost of US$ 6–14 per child on oxygen at 1 litre per min (depending on the size of cylinder available), in our setting this would have translated into US$ 5172 to US$ 12 068 in 1 year had each child not requiring oxygen (862) received it for 1 day.2 A hand-held pulse oxymeter costs less than US$ 1000.

Where supplies of oxygen are limited and pulse oximetry is unavailable, as in most health facilities that treat severely ill children in resource-limited settings, prioritization depends on clinical signs. The signs we found to be most strongly predictive of hypoxaemia among admissions aged ≥ 60 days correspond with those given in the combined chapters of “triage and emergency conditions” and the “cough and difficult breathing” in the current WHO guidelines.11 The signs of shock (delayed capillary refill and a weak pulse volume) included in the triage guidelines have not been evaluated in previous studies of hypoxaemia in this setting, and including them in the WHO emergency care guidelines appears to be justified as part of the assessment of the airway, breathing and circulation. Importantly, hypoxaemia detected by transcutaneous measurement may be due in part to poor tissue perfusion (hence poor peripheral oxygenation), and oxygen may therefore be needed only in the initial stabilization phase. Although a slow heart rate and irregular breathing are not included in the currently recommended triage assessment, they were among the strongest predictors of hypoxaemia in our study. These signs, together with the signs of shock, led to identification of 50% of the cases of hypoxaemia and had reasonable specificity (85%). Importantly, although cyanosis alone was a very strong positive predictor (LR: 24), few of the children admitted had this sign, and the absence of cyanosis was not predictive of normal oxygenation (LR: 0.94).

Lower chest wall indrawing by itself was a weak predictor of hypoxaemia, and signs such as pallor, dullness on chest percussion, wheeze or crackles were not predictive. We did not include laboratory results in the final models because they are seldom available when an initial decision regarding oxygen therapy has to be made.

We performed separate analyses for different age groups on the hypothesis that predictive signs may differ in each. However, we found the signs to be remarkably similar. Few studies have looked at hypoxaemia in neonates and young infants in this setting. One study of 132 neonates in Papua New Guinea showed that a model of cyanosis, respiratory rate > 70 breaths per min, or respiratory rate < 30 breaths per min and reduced activity could predict hypoxaemia with a sensitivity of 84% and a specificity of 55%.23 A recent study from the Gambia showed that the prevalence of hypoxaemia in neonates was about 16.5%.19 However, in that study 40% of all neonatal admissions were missed. Importantly, no neonates were recruited in the first week of life.

Our study has several weaknesses. We did not collect data on head nodding and grunting, which have yielded variable findings in previous studies.4,9 Pulse oximetry was performed routinely during admission, and the admitting clinicians were not blinded to the SaO2 results during their initial clinical assessment. However, the low sensitivity of signs such as cyanosis, commonly regarded as a cardinal sign of hypoxaemia, and the fact that the predictive value of the clinical signs was similar to that reported in other studies, suggest that clinicians were not unduly influenced when recording their findings.1,8 A generalized problem is that among children admitted with overt shock and signs of poor circulation, peripheral pulse oximetry results may not always accurately reflect core hypoxaemia.24

We defined hypoxaemia as an SaO2 < 90% by pulse oximetry.11 However, children aged ≥ 60 days whose SaO2 was 90–94% had a higher case fatality proportion (102/1339, or 7.6%) than those whose SaO2 was ≥ 95% (503/11 203, or 4.5%) (P < 0.001). Some children with an SaO2 of 90–95% may benefit from oxygen supplementation; further studies are needed to elucidate if oxygen affects the outcome at this level.

Our findings strongly support current WHO recommendations surrounding triage for the detection of hypoxaemia. This is important because almost all prior studies have only included children thought to have pneumonia, and none has evaluated the triage signs of shock. Based on our findings, we advise against excessive emphasis on LRTI, which may potentially result in the underdiagnosis of other conditions associated with respiratory distress and hypoxaemia. Rather, we recommend a structured approach to assessment and triage that will target children likely to have hypoxaemia irrespective of the working diagnosis. Finally, pulse oximetry is essential for detecting hypoxaemia and is likely to save costs. ■


We thank the Kenya Medical Research institute (KEMRI) and The Wellcome Trust (Grants codes 061584, 076278, 081186) for their support. We also thank the Medical Officer of Health, the Medical Superintendent and the paediatric staff of Kilifi District Hospital. This article is submitted for publication with the permission of the Director of the KEMRI.

Competing interests: None declared.