Defining hypoxemia from pulse oximeter measurements of oxygen saturation in well children at low altitude in Bangladesh: an observational study

Background: The World Health Organization defines hypoxemia, a low peripheral oxyhemoglobin saturation (SpO2), as <90%. Although hypoxemia is an important risk factor for mortality of children with respiratory infections, the optimal SpO2 threshold for defining hypoxemia is uncertain in low-income and middle-income countries (LMICs). We derived a SpO2 threshold for hypoxemia from well children in Bangladesh residing at low altitude. Methods: We prospectively enrolled well, 3-35 month old children participating in a pneumococcal vaccine evaluation in Sylhet district, Bangladesh between June and August 2017. Trained health workers conducting community surveillance measured the SpO2 of children using a Masimo Rad-5 pulse oximeter with a wrap sensor. We used standard summary statistics to evaluate the SpO2 distribution, including whether the distribution differed by age or sex. We considered the 2.5th, 5th, and 10th percentiles of SpO2 as possible lower thresholds for hypoxemia. Results: Our primary analytical sample included 1,470 children (mean age 18.6 +/- 9.5 months). Median SpO2 was 98% (interquartile range, 96-99%), and the 2.5th, 5th, and 10th percentile SpO2 was 91%, 92%, and 94%. No child had a SpO2 <90%. Children 3-11 months old had a lower median SpO2 (97%) than 12-23 month olds (98%) and 24-35 month olds (98%) (p=0.039). The SpO2 distribution did not differ by sex (p=0.959). Conclusion: A SpO2 threshold for hypoxemia derived from the 2.5th, 5th, or 10th percentile of well children is higher than <90%. If a higher threshold than <90% is adopted into LMIC care algorithms then decision-making using SpO2 must also consider the childs clinical status to minimize misclassification of well children as hypoxemic. Younger children in lower altitude LMICs may require a different threshold for hypoxemia than older children. Evaluating the mortality risk of sick children using higher SpO2 thresholds for hypoxemia is a key next step.


Introduction 84
Lower respiratory infections (LRIs) kill more young children than any other infectious 85 disease in the world. 1 The most recent 2017 global estimates report more than 800,000 86 LRI deaths annually among children below five years of age, 1 equating to 1-2 deaths 87 every minute. The vast majority of pediatric LRI deaths occur in low-income and middle-88 income countries (LMICs). 1 Approximately 30% of all global LRI deaths take place in 89 South Asia each year, and Bangladesh has the 3 rd highest annual pediatric LRI 90 incidence and mortality burden among all South Asian countries. 1 91 92 LRIs may be complicated by pulmonary inflammation and areas of ventilation-perfusion 93 mismatch that cause acute hypoxemia, or a low peripheral arterial oxyhemoglobin 94 saturation (SpO2) as measured non-invasively by a pulse oximeter. 2 Acute hypoxemia is 95 an important risk factor for mortality among children with LRIs in LMICs. 3 For LMICs at 96 lower altitude (i.e., <2,500 meters) the SpO2 hypoxemia threshold endorsed by the 97 World Health Organization (WHO) is <90%, a threshold associated with elevated 98 mortality risk among children with LRIs like pneumonia. 3-5 Per WHO guidelines children 99 with caregiver reported cough and/or difficult breathing accompanied by a SpO2 <90% 100 are recommended for hospitalization, parenteral antibiotics, and oxygen 101 administration. 4, 5 Recent observational studies from Malawi reveal that SpO2 thresholds 102 higher than 90% may also be associated with elevated mortality risk among children 103 under five years with clinically diagnosed pneumonia. 6-8 This evidence suggests the 104 current WHO SpO2 hypoxemia threshold of <90% may be suboptimal for identifying 105 higher risk pediatric pneumonia cases for hospitalization in some LMICs. 106 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 23, 2021. ; https://doi.org/10.1101/2021.06.15.21258979 doi: medRxiv preprint Despite both the importance and uncertainty around the optimal SpO2 threshold for 108 defining hypoxemia few studies from lower altitude settings in LMICs address this issue. 109 One approach commonly used for deriving thresholds for diagnostic tests is to produce 110 a reference range from a healthy population representative of the test's intended target 111 population. 9, 10 This approach has been applied to SpO2 measurements in children, with 112 most research to date focused on children residing at higher altitudes. oximeter with a LNCS® Y-I wrap sensor as a part of enhanced respiratory surveillance 143 activities for children during the parent PCV study. The initial training was one day and 144 included theoretical sessions on pulse oximetry supplemented by practice using pulse 145 oximeters to measure the SpO2 of volunteer adults and children. During the study period 146 CHWs participated in refresher sessions at least every six months and were routinely 147 supervised by study physicians during household participant screening with the device. 148 Remediation was provided when needed. CHWs were trained to apply the wrap sensor 149 to the big toe of children and gently hold the foot to mitigate movement artifact. SpO2 150 values were considered adequate quality measurements when the CHW achieved the 151 following three metrics; (1) the SpO2 value remained stable and non-drifting for no less 152 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. week. CHWs observed children for cough, counted the child's respiratory rate for one 160 minute, measured an axillary temperature with a thermometer, and observed children 161 for any sign of respiratory distress (i.e., head nodding, nasal flaring, audible wheezing, 162 grunting, stridor, tracheal tugging, or lower chest wall indrawing). Children were 163 excluded and referred to the study clinic if aged 3-11 months and had a respiratory rate 164 of >50 breaths/minute, or 12-35 months old with a respiratory rate of >40 165 breaths/minute, an axillary temperature >101º F, any vomiting or diarrhea, any WHO-166 defined general danger sign (lethargy, convulsions, not eating or drinking, severe acute 167 malnutrition), or any sign of respiratory distress as specified above. Children with 168 isolated nasal congestion and/or rhinorrhea were not considered acutely ill and were 169 enrolled. 170

171
In order to further filter potentially unwell children from our sample, post-hoc we created 172 three analytic samples from children with a recorded SpO2 measurement using different 173 reference heart rate ranges, since an abnormal heart rate may suggest unrecognized 174 illness. Analytic sample 1 is our primary analytic sample, and applies the most 175 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Statistical analysis 184
Normally distributed continuous variables were described using means and standard 185 deviations, non-normally distributed continuous variables were characterized by 186 medians and interquartile ranges, and bivariate or categorical variables were described 187 using proportions. We considered the 2.5 th , 5 th , and 10 th percentile of SpO2 as possible 188 thresholds for defining hypoxemia. We used the Wilcoxon-Mann-Whitney test for 189 comparisons including a dependent variable without a normal distribution. The Kruskal 190 Wallis test was used for comparisons between a multi-level independent variable and a 191 dependent variable lacking a normal distribution. We fit a linear regression model, 192 adjusted for sex, to explore the association between SpO2 and age. Using a power of 193 80%, significance level of 0.05, and that 25% of children will either be ill, unavailable, or 194 fail measurement, we needed to screen 700 households for each of the three child age 195 strata of 3-11 months, 12-23 months, and 24-35 months (total 2,100) to estimate a 196 mean SpO2 of 96% +/-0.2%. Stata version 16.0 (College Station, TX) was used for all 197 analyses. 198 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. SpO2 measurements were excluded at the analysis stage. For primary analytic sample 217 1 a total of 552 children were additionally omitted due to abnormal heart rates. For 218 analytic sample 2, 157 children were excluded based on reference heart ranges. Among 219 the 1,470 children analyzed for analytic sample 1, the mean age was 18.6 months (SD, 220 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 23, 2021. similar across the three analytic samples. 222 223

Effects of age and sex on SpO2 231
After stratifying measurements into three age strata, 3-11 months, 12-23 months, and 232 24-35 months, we found children 3-11 months old in primary analytic sample 1 to have 233 a median SpO2 of 97%, compared to 98% for each of the two older age strata (p=0.038; 234 Table 1 and Figure 3). We observed similar findings in analytic samples 2 and 3 (Table  235 1 and Supplemental Figures 1 and 2). When regressing SpO2 on age in months, 236 adjusted for sex, we found that for every one month increase in age the SpO2 increased 237 by 0.01% (95% CI, 0.001%, 0.02%, p=0.030) in analytic sample 1 (Supplemental Figure  238 3). We did not observe any difference in the SpO2 distribution after stratifying by child 239 sex (p=0.959). 240

241
Health system implications of varying SpO2 thresholds 242 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. There are two key findings from this research. First, cutoffs for hypoxemia from the 261 2.5 th , 5 th , and 10 th percentile are all higher than the current WHO-defined <90% 262 threshold and we did not find any well children below the SpO2 <90% threshold. Thus, if 263 any of these cutoffs for hypoxemia are adopted then measuring the SpO2 earlier in 264 clinical care pathways when healthier children may be over-represented could increase 265 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 23, 2021. ; https://doi.org/10.1101/2021.06.15.21258979 doi: medRxiv preprint limited resources and potential challenges coping with a higher volume of patient 267 referrals. These results, coupled with findings from Malawi that children with LRI and a 268 SpO2 between 90% and 92% are at elevated mortality risk, suggest that the current 269 referral threshold of SpO2 <90% minimizes false positives at the expense of false 270 negatives. 6-8 In order to ensure minimal misclassification of well children as hypoxemic, 271 we recommend care algorithms incorporating a hypoxemia threshold at SpO2 levels 272 higher than <90% also consider the child's clinical status when deciding whether to refer 273 and hospitalize. Second, we found that the SpO2 distribution differs by age. Age may 274 therefore need to be considered when establishing a SpO2 threshold for hypoxemia. 275 at lower altitudes, understanding the SpO2 distribution among this population has broad 281 relevance for identifying the optimal SpO2 threshold for hypoxemia. In one study from 282 Chennai, India (altitude 7 meters) the authors measured the SpO2 of 626 healthy 283 children aged one month to five years. 19 In contrast to our findings, the authors found no 284 difference in the SpO2 distribution by age. The inclusive 5 th percentile of participants in 285 Chennai was a SpO2 <96%, which was 4% higher than in our study. Other studies that 286 also included healthy children from lower altitudes reported the mean and standard 287 deviation of the SpO2 distribution. 13, 14, 20 Given the SpO2 distribution of healthy children 288 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 23, 2021. ; https://doi.org/10.1101/2021.06.15.21258979 doi: medRxiv preprint is negatively skewed we described these data using median and percentiles and are 289 therefore unable to make meaningful comparisons. in India was notably consistent with the Chennai study at 96%. Unlike the Chennai 299 study, however, HAPIN investigators found lower 5 th percentile thresholds for age in 300 Rwanda (92%) and Guatemala (93%), and observed a correlation between younger age 301 and lower SpO2. None of these studies reported the 2.5 th percentile cutoff. Overall, it is 302 somewhat surprising that our data from Bangladesh aligns closer with Rwanda and 303 Guatemala than India, another South Asian setting closer in altitude. 304 305 Methodology may largely be responsible for the variation in results across these 306 studies. Variation may be due to a combination of device accuracy, including 307 differences in accuracy between devices, variation inherent to measurements on 308 children, measurement variation between healthcare workers and healthcare worker 309 cadres with different training backgrounds, and possible varying degrees of 310 misclassification bias of sick children in each of the three studies. Specifically, the 311 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 23, 2021.   interventions that have affected the burden of lower respiratory infections among 403 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 23, 2021.