When should oxygen be given to children at high altitude? A systematic review to define altitudespecific hypoxaemia
 ^{1}Centre for International Child Health, Department of Paediatrics, University of Melbourne, Royal Children’s Hospital, Parkville, Victoria, Australia
 ^{2}Clinical Epidemiology and Biostatistics Unit, Murdoch Childrens Research Institute, Royal Children’s Hospital, Parkville, Victoria, Australia
 ^{3}Centre for International Child Health, Department of Paediatrics, University of Melbourne, Royal Children’s Hospital, Parkville, Victoria, Australia
 Rami Subhi, Centre for International Child Health, Department of Paediatrics, University of Melbourne, Royal Children’s Hospital, Parkville, 3052, Victoria, Australia; rami.subhi{at}rch.org.au
 Accepted 16 September 2008
 Published Online First 1 October 2008
Abstract
Background: Acute respiratory infections (ARI) cause 3 million deaths in children worldwide each year. Most of these deaths occur from pneumonia in developing countries, and hypoxaemia is the most common fatal complication. Simple and adaptable indications for oxygen therapy are important in the management of ARI. The current WHO definition of hypoxaemia as any arterial oxygen saturation (SpO_{2}) <90% does not take into account the variation in normal oxygen saturation with altitude. This study aimed to define normal oxygen saturation and to estimate the threshold of hypoxaemia for children permanently living at different altitudes.
Methods: We carried out a systematic review of the literature addressing normal values of oxygen saturation in children aged 1 week to 12 years. Hypoxaemia was defined as any SpO_{2} at or below the 2.5th centile for a population of healthy children at a given altitude. Metaregression analysis was performed to estimate the change in mean SpO_{2} and the hypoxaemia threshold with increasing altitude.
Results: 14 studies were reviewed and analysed to produce prediction equations for estimating the expected mean SpO_{2} in normal children, and the threshold SpO_{2} indicating hypoxaemia at various altitudes. An SpO_{2} of 90% is the 2.5th centile for a population of healthy children living at an altitude of ∼2500 m above sea level. This decreases to 85% at an altitude of ∼3200 m.
Conclusions: For health facilities at very high altitudes, giving oxygen to all children with an SpO_{2} <90% may be too liberal if oxygen supplies are limited. In such settings, Spo_{2} <85% may be more appropriate to identify children most in need of oxygen supplementation.
Each year pneumonia causes more than 3 million deaths in children under 5 years of age (including neonates) and results in 11–20 million hospitalisations.1 2 Some 98% of deaths from pneumonia in children occur in developing countries.3 Hypoxaemia is a significant risk factor for mortality4^{–}6 and a strong predictor of radiographic pneumonia.5 7 8 The recognition of hypoxaemia among children with pneumonia is crucial in management, may assist in diagnosis and helps determine prognosis.
Currently the World Health Organization (WHO) sets out the clinical signs which should guide the starting and stopping of oxygen therapy. Where pulse oximetry is available, WHO recommends an arterial oxygen saturation measured by pulse oximetry (Spo_{2}) of 90% as the threshold for oxygen administration.9 However, such an approach may only be possible in settings where oxygen supplies are not limited. At high altitudes, where normal oxygen saturation levels are lower than at sea level, a threshold of 90% may be less relevant. A clearer definition of the lower limit of normal Spo_{2} among children at different altitudes would enable protocols for oxygen therapy to be adapted to local conditions, and enable resources to be appropriately applied.
To assist in the identification and management of hypoxaemia, this study aimed to use the published literature to define normal values of Spo_{2} in children and to propose a model for defining hypoxaemia at varying altitudes.
METHODOLOGY
Search strategy and inclusion criteria
A review was conducted using OVID Medline (1950–August week 2 2007) and Embase (1980–2007 week 33), according to the search strategy described in box 1. The abstracts of potentially relevant publications were reviewed, and the complete texts of studies addressing the inclusion criteria (outlined below) were obtained. References cited by these articles, as well as the link option “Find similar” in OVID and Embase were then used to widen the search.
Studies were included if they reported the Spo_{2} of healthy, awake children aged 1 week to 12 years and permanently residing in the study location. In order to ensure that study data used in the analysis provide a reasonable estimate of the normal Spo_{2} of the populations they represent, we set the cutoff for sample size as greater than 30 children. Studies including Spo_{2} measurements of preterm, hospitalised, anaesthetised or chronically ill children were excluded. Only papers in English were reviewed, and no attempts were made to obtain unpublished data.
The inclusion criteria were designed to include the largest sample of children for whom such data were available, but it was considered important to exclude studies that described normal episodes of desaturation during the foetal–neonatal transition throughout the first hours of life. Although recovery to adult Spo_{2} levels has been previously demonstrated within 15 min after birth,10 a minimum age of 1 week was set to ensure that such normal fluctuations were not represented.
Data extracted from studies were age, sample size, mean (standard deviation) Spo_{2}, oximeter type, and altitude above sea level. Where the standard deviation was not reported, either the 95% confidence interval or the standard error was extracted, depending on the information reported. Missing standard errors and standard deviations were then derived from available statistics using standard methods.11 All altitudes less than 100 m above sea level were considered to be equivalent to sea level (0 m). If a study reported Spo_{2} for different age categories without providing overall summary data, the subgroup with the greatest number of subjects not including neonates was chosen to be included in the analysis.
We defined hypoxaemia as any Spo_{2} value at or below the 2.5th centile for a population of healthy children at a given altitude. If the values of Spo_{2} are normally distributed, this definition corresponds to a Spo_{2} reading of more than 2 SD below the mean.
Box 1 Search strategy
OVID Medline (771 results)

(exp Anoxia/ and exp Oximetry/) or (exp Reference Values/ and exp Oximetry/)


Limited to (“newborn infant (birth to 1 month)” OR “infant (1 to 23 months)” OR “preschool child (2 to 5 years)” OR “child (6 to 12 years)”).

Embase (11 results)

(exp Anoxia/ and exp Oximetry/) or (exp Reference Values/ and exp Oximetry/)


Limited to (infant <to one year> OR child <unspecified age> OR preschool child <1 to 6 years> OR school child <7 to 12 years>)

Statistical analysis
Data were entered into Microsoft Excel, and analysed using Stata v 10. The 2.5th centile of the distribution of Spo_{2} values could not be estimated directly, as only one study in the literature reported this statistic.6 For this reason, the hypoxaemia threshold was estimated by predicting mean Spo_{2} based on altitude, and subtracting 2 SD from the prediction. The prediction equation was obtained using a linear random effects metaregression, which was implemented using the usercontributed Stata command “metareg”.12 The betweenstudies variance was estimated using restricted maximum likelihood.
Inspection of the summary statistics reported in table 1 indicates that the distribution of Spo_{2} values for many studies is negatively skewed rather than normally distributed; adding 2 SD to the mean produces a value exceeding the theoretical maximum of 100% for the six studies carried out at an altitude of 1600 m or less. We required the outcome to be normally distributed, both to satisfy the assumptions of the analysis method, and because we wished for the mean minus 2 SD to correspond to the 2.5th centile. For these reasons, metaregression was performed on the logtransformed scale. A prediction equation for hypoxaemia threshold was derived by subtracting 2 SD from the predicted mean Spo_{2} on the transformed scale, and backtransforming the resulting equation to the natural scale.
RESULTS
Twenty four studies fulfilled the selection criteria.5^{–}7 13^{–}33 Eleven studies were subsequently excluded including a study in Peru of a 6 to 18yearold population which did not meet the age criteria,21 and one in Nepal of children presenting with cough and coryza.14 An additional six studies did not report a mean Spo_{2} value and appropriate measure of variation,6 15 19 25 26 34 and three studies were excluded due to having a sample size of less than 30.27 30 31
Mean Spo_{2}, altitude, sample size and an appropriate measure of variation were reported in the remaining 13 studies (table 1). Of these 13 studies, nine reported a standard deviation, two a standard error and the remaining two a 95% confidence interval. One study examined two populations living at different altitudes,7 reporting summary statistics separately. Each population was treated as a separate population for the purpose of the metaregression. In total therefore, the analyses of Spo_{2} values included 14 data points from 13 studies.
Very strong evidence was found for an association between study altitude and the study’s mean Spo_{2} value (p<0.001). The equation obtained for predicting mean Spo_{2} at a given altitude was:
Mean Spo_{2 }(%) = 100.5−1.374e^{0.590×altitude(km)}
The equation obtained for predicting the hypoxaemia threshold at a given altitude was:
Spo_{2 }(%) = 100.5−1.374e^{(0.590×altitude(km)+0.563)}
Both prediction equations are graphically depicted in fig 1. At an altitude of approximately 2500 m above sea level, the 2.5th centile of the distribution of Spo_{2} in normal, healthy children is 90%. This decreases to 85% at approximately 3200 m above sea level.
DISCUSSION
Oxygen therapy has been shown to improve outcomes in pneumonia.4 35 Ensuring a reliable and efficient system for detecting hypoxaemia and supplying oxygen and having clear and simple guidelines for its use are therefore vital to good quality paediatric care. This is particularly so in developing countries given the magnitude of the burden of pneumonia. And yet, such systems are often of poor quality or nonexistent where they are most needed, where oxygen administration is often dictated more by availability than by need.
There have been many hurdles to implementing effective systems for oxygen therapy in developing countries, including the cost and scarcity of cylinder oxygen and the unreliability of clinical signs in predicting hypoxaemia. In recent times, the introduction of pulse oximetry into the health facilities of developing countries has been advocated for its accuracy in detecting hypoxaemia and costeffectiveness in limiting oxygen wastage.36 37 If pulse oximetry is to be widely used and understood by health workers, consensus as to the Spo_{2} level below which oxygen supplementation should be given is needed.
There are physiological and safe practice arguments for maintaining Spo_{2} above 90% at sea level, to ensure that the relationship between Sao_{2} and Po_{2} is on the flat segment of the oxygen dissociation curve. While Spo_{2} levels below 90% at sea level indicate severe respiratory pathology, in high altitude locations such levels are found in normal local children owing to the low barometric pressure. Despite this, the oxygen balance in these children remains adequate in health due to several physiological adaptations, including hyperventilation, polycythaemia, enhanced alveolar growth and increased capillary proliferation.38 39 For ill children living at such high altitudes, the aim of oxygen supplementation should not therefore be based on normal levels at sea level, but should aim to achieve an adjusted expected normal Spo_{2} for that setting. Our study quantifies the expected reduction in Spo_{2} levels with increasing altitudes. The strong association found between study altitude and a study’s average Spo_{2} value suggests that a lower threshold for giving oxygen may be appropriate at very high altitude if supplies are limited.
Figure 1 suggests that at altitudes above 2500 m using an Spo_{2} threshold of 90% to administer oxygen is conservative and may result in oxygen supplementation to some children with Spo_{2} in the normal range. In facilities with limited oxygen supplies located at these altitudes, therefore, a lower level of Spo_{2} can be used, and an approximation could be an Spo_{2}<85%. Above 3200 m, even this lower definition is conservative, and may become too sensitive as a screening test for oxygen need. Such guidelines are based on statistical definitions and their clinical relevance needs to be further studied. The use of thresholds of hypoxaemia below 90% to indicate oxygen therapy at altitude has been reported in the literature,7 18 19 but clearly, in addition to locallyrelevant thresholds of Spo_{2}, oxygen therapy should be based on the overall clinical picture of each child, as well as a careful consideration of resource availability.
In many health facilities, the scarcity of oxygen means that health workers are faced not with the question of which children need oxygen, but which children need it most. It is optimal practice to gather children requiring oxygen together in a high dependency area of the ward, allowing for regular observation and monitoring. This necessitates accurate identification of the sick children requiring such attention. Measuring oxygen saturation by pulse oximetry can be used to facilitate this process and to indicate the severity of respiratory disease.4 5 40 In high altitude settings, consideration of the normally lower levels of Spo_{2} and being able to approximate these quantitatively, will be useful in informing clinical decisions.
It should be emphasised that ultimately what is important in maintaining adequate tissue oxygenation is the oxygen content in the blood not the oxygen saturation. This is given by the equation (1.30×[Hb]×Sao_{2})+0.003 Pao_{2}. Children with severe anaemia or infants in their physiological nadir of haemoglobin concentration may be hypoxic despite having an Spo_{2} level within the normal range. Ideally, oxygen therapy needs to be guided by pulse oximetry as well as measurement of haemoglobin concentration. However, in the absence of an immediate measure of haemoglobin concentration, emphasis may be placed on identifying the clinical signs of anaemia or identifying children in their nadir of haemoglobin concentration, in whom more liberal thresholds for giving oxygen would be appropriate.
Previous studies have used the mean Spo_{2} minus 2 SD to define hypoxaemia in children, and assumed normally distributed saturation levels.5 7 17 18 Our study has taken into account the skewed distribution of normal Spo_{2}, and has weighted each study estimate by its precision. We have also attempted to include the largest number of relevant studies that reported results in a format which would allow a pooled analysis of the data. Our findings suggest that the definition of hypoxaemia as mean Spo_{2} minus 2 SD may not accurately capture the lowest 2.5% of values at low and moderate altitudes. We recommend that future studies check distributional assumptions and explicitly report centile values such as the 2.5th centile and/or median.
There are a number of limitations to the study design we used. Where children were selected on grounds of ethnic background, the normal Spo_{2} reported may not be representative of unstudied populations of children living at comparable altitudes. In addition, variations between study findings could be due to differences in sampling, ethnicity and age ranges studied, as well as the use of different oximeters and/or protocols for measuring Spo_{2}. Also, there was no way of controlling for either the precision or the accuracy of the oximeters used in each study. The statistical model employed assumes reliable estimates of study means and standard deviations, as well as assuming that the transformation applied would have resulted in an approximately normal distribution for study populations at all altitudes. Ideally, had raw data been available, summary statistics would have been calculated on an appropriate scale rather than estimated.
What is already known on this topic

Hypoxaemia is a fatal complication of acute lower respiratory infections (ALRI) and many nonALRI conditions.

Pulse oximetry is currently available in the health facilities of many developing country to guide the allocation of limited oxygen supplies.

WHO guidelines recommend oxygen for SpO_{2} <90%, but this may not be relevant to all settings, particularly those at high altitude.
What this study adds

Using SpO_{2} <90% to dispense oxygen may be too liberal if oxygen supplies are limited.

SpO_{2} <85% can be used to indicate the need for oxygen at altitudes greater than 2500 m above sea level.
CONCLUSION
Above altitudes of 2500 m, giving oxygen for Spo_{2} less than 90% may be too liberal for facilities with limited oxygen supplies. There is evidence that for altitudes greater than 2500 m a threshold of Spo_{2} of 85% can be used to identify children most in need of oxygen. A balance needs to be achieved between using accurate altitudespecific definitions of hypoxaemia, and ensuring simple and safe indications for oxygen that can be taught to and used by health workers. We hope the model proposed for predicting normal Spo_{2} and the definition of hypoxaemia at different altitudes will be useful for clinicians in determining when to give oxygen.
Acknowledgments
We would like to thank Professor John Carlin for his statistical advice.
Statistical appendix
Consider a positivelyskewed distribution Y with log(Y)≅Normal(μ, σ^{2}). We can estimate μ and σ^{2} from Y̅ and SD(Y) using the following relations42:
We can obtain a positivelyskewed distribution Y from a negativelyskewed distribution of Spo_{2} values by applying the transformation Y = 100.5−Spo_{2}. The 0.5 is added to avoid the need to take the log of zero for values with Spo_{2} = 100. Since Y is obtained by a linear transformation with coefficient −1 and intercept 100.5, we have:
Putting this all together, we can estimate μ and σ^{2} from the sample mean and standard deviation of a distribution of Spo_{2} values using the following relations:
Footnotes

See Perspective, p 1

Competing interests: None.

This review was part of the International Child Health Review Collaboration (www.ichrc.org).