American Journal of Respiratory and Critical Care Medicine

Pneumonia is the leading infectious cause of death in children worldwide, with most deaths occurring in developing countries. Measuring respiratory rate is critical to the World Health Organization’s guidelines for diagnosing childhood pneumonia in low-resource settings, yet it is difficult to accurately measure. We conducted a systematic review to landscape existing respiratory rate measurement technologies. We searched PubMed, Embase, and Compendex for studies published through September 2017 assessing the accuracy of respiratory rate measurement technologies in children. We identified 16 studies: 2 describing manual devices and 14 describing automated devices. Although both studies describing manual devices took place in low-resource settings, all studies describing automated devices were conducted in well-resourced settings. Direct comparison between studies was complicated by small sample size, absence of a consistent reference standard, and variations in comparison methodology. There is an urgent need for affordable and appropriate innovations that can reliably measure a child’s respiratory rate in low-resource settings. Accelerating development or scale-up of these technologies could have the potential to advance childhood pneumonia diagnosis worldwide.

Pneumonia remains the leading infectious cause of death among children younger than 5 years of age. In 2015, pneumonia accounted for 16% of child deaths globally (1). Although childhood pneumonia deaths can be prevented with simple interventions and appropriate treatment, pneumonia often goes undiagnosed and untreated in the community until the child is severely ill (2, 3). In low-resource settings, pneumonia is diagnosed using the World Health Organization Integrated Management of Childhood Illness and Integrated Community Case Management (iCCM) guidelines, which rely on the appreciation of subjective clinical signs and symptoms. An important component of the Integrated Management of Childhood Illness and iCCM criteria is the accurate classification of fast breathing, defined as 60 or more breaths per minute in infants younger than 2 months, 50 or more breaths per minute in infants aged 2 to 11 months, and 40 or more breaths per minute in children aged 12 to 59 months (4). In low-resource settings, counting the number of breaths typically is performed manually with the aid of watches, timers, and counting beads (5, 6). However, even with these counting aids, measuring a child’s respiratory rate (RR) through visual observation requires focused concentration and can be challenging in a child who may be moving, crying, or breathing rapidly. Inaccurate or imprecise measurements can stem from factors including poor visibility of the start or end of a breath, an irritable or moving child, or difficulty counting or remembering the count (7).

Accurate assessment of RR is critical in low-resource settings where other diagnostic tools, such as pulse oximetry or chest radiography, are typically not available and pneumonia is diagnosed based on the child’s clinical signs alone. Given the high burden and mortality of childhood pneumonia, there is growing demand for better ways to measure RR accurately and reliably. Although there are numerous existing and potential approaches for measuring RR (Table 1), it is important to assess whether these have been rigorously evaluated in a way that facilitates comparisons of accuracy and performance. This systematic review provides an overview of the RR measurement tools that have undergone a clinical evaluation of accuracy against a reference standard among spontaneously breathing children younger than 5 years of age. Some of the results of these studies have been previously reported in the form of an abstract (8).

Table 1. Respiratory Rate Method Categories

Manual MethodsDescriptionExamples*
Manual count  
 Timers onlyAnalog devices used to inform the observer when to start and stop counting breaths.ARI timer, wristwatch
Assisted count  
 Counters onlyColor-coded string of beads used in combination with a timer to eliminate the need for an observer to remember breath count and age-designated cutoff rates.Breath Abacus, International Rescue Committee counting beads
 Combined timer and counterStand-alone digital devices or software-based mobile applications with a built-in 1-min timer to eliminate the need for an observer to remember breath count by having the user press a button or tap the screen to register each breath.Mobile software applications
   
Automatic MethodsDescriptionExamples*
Exhaled breath  
 HumidityRR derived from oronasal moisture sensors measuring fluctuations in humidity with respiration.Interferometry sensors, relative humidity sensors, absolute humidity (moisture) sensors, hygrometers
 TemperatureRR derived from oronasal temperature sensors measuring fluctuations in temperature with respiration.Thermistors, infrared thermography, thermocouple sensors, nasal prongs, face masks
 Air pressureRR derived from oronasal sensors measuring fluctuations in air pressure with respiration.Barometric pressure sensors, pressure transducers, airflow velocity sensors, spirometers, pneumotachometers
 ETco2RR derived from oronasal capnography sensors measuring fluctuations in carbon dioxide concentrations with respiration.Capnometers, mid-infrared LED detectors
 ETo2RR derived from oronasal oxygen sensors measuring fluctuations in oxygen concentrations with respiration.Differential paramagnetic sensors, fiber optic fluorescence-based oxygen sensors, gas analysis systems
Thoracic effort  
 Thoracic circumferenceRR derived from sensors measuring fluctuations in thoracic circumference with respiration.Inductance plethysmography sensors, piezoelectric sensors, rubber dilation sensors, stretch sensors, chest straps, or belts
 Thoracic motionRR derived from sensors measuring fluctuations in thoracic motion with respiration.Accelerometers, gyroscopes, ballistocardiography sensors, bioradiolocation sensors, noncontact microwave and wireless networks, ferroelectric sensors, mattress sensors, electromagnetic generator, small movement motion amplification programs
 VtRR derived from electrodes measuring fluctuations in lung volume with respiration.Bioimpedance electrode sensors varying in presentation (disposable skin adhesive, portable body-worn devices, wearable garments)
Respiratory sounds  
 OronasalRR derived from acoustic respiratory signals measured near the oronasal area.Noncontact microphones or wearable headsets
 ThoracicRR derived from acoustic respiratory signals measured near the chest, back, or armpit.Electronic auscultation of signals collected from modified digital stethoscopes
 TrachealRR derived from acoustic respiratory signals measured near the throat or neck.Sensor adhesives placed near the throat or neck
Indirect effects on cardiovascular physiology/blood flow  
 ECGRR derived indirectly from the ECG measured by a configuration of electrodes.Numerous portable ECG electrode systems integrating patented signal extraction techniques
 PPGRR derived indirectly from the PPG measured by a pulse oximeter.Numerous portable pulse oximetry systems integrating patented signal extraction techniques, including finger clip sensors and wrist-worn monitors, among others
 PtcCO2RR derived indirectly from the PtcCO2 measured by a wet Ag/AgCl electrode heated above fevered temperatures.CO2 measured potentiometrically by determining the pH of an electrolyte layer and used to calculate RR
 PATRR derived indirectly from the PAT waveform measured by an apparatus capable of sensing finger pulsatile arterial volume changes.Fusion algorithms process PAT waveforms to identify respiratory events used to calculate RR
 ABPRR derived indirectly from the ABP measured by a sphygmomanometer.Complete health monitoring systems displaying vital signs (e.g., ABP, pulse, and temperature) in addition to RR

Definition of abbreviations: ABP = arterial blood pressure; ARI = acute respiratory infection; ETco2 = end-tidal carbon dioxide concentration; ETo2 = end-tidal oxygen concentration; LED = light-emitting diode; PAT = peripheral arterial tonometry; PPG = photoplethysmogram; RR = respriatory rate.

*Examples provided are for illustrative purposes and are not claimed to be complete.

Search Strategy and Selection Criteria

PubMed, Embase, and Compendex/Engineering Village were searched through September 2017 for English-language publications reporting on measuring or monitoring RR in infants and children younger than 5 years of age. The search strategy used preferred indexing terms from each database and was built around a set of controlled vocabulary terms and relevant text words in the title, abstract, or subject fields (or a combination of those fields).

Articles were screened by two team members for relevance to the analysis on the basis of sequential review of study titles, abstracts, and full text. The inclusion criteria were specified as English publications describing accuracy of assessment of breath count or RR in spontaneously breathing human infants and/or children younger than 5 years of age compared with a reference standard. The exclusion criteria included mechanically ventilated subjects, nonhuman animal subjects, failure to report RR values, lack of a reference standard, no assessment of accuracy, non-English publication, subjects older than 5 years of age, inability to disaggregate data on children younger than 5 years from data on older children and/or adults, inability to disaggregate data on mechanically ventilated children from data on non–mechanically ventilated children, review article reporting secondary data, and insufficiently detailed description of methods. If not explicitly stated, study authors were contacted for confirmation that subjects were spontaneously breathing and not mechanically ventilated. For this review, accuracy is defined as the difference between the values measured by the experimental device and the “true value” as quantified by the reference standard.

Data Extraction and Analysis

Included articles were critically appraised for eligibility criteria and clearly described methods following Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (9). The following data were extracted into a database with designated fields: author, title, publication date, journal, country/countries of study, population(s) studied, study setting (e.g., laboratory, hospital, field), sample size, time period, study type, primary objective, categorization of device as manual or automated, reference standard, measurement interval, who took the measurement (e.g., research team, nurse), outcome measure, and statistical method of comparison. Included studies were divided into two groups on the basis of how the devices measured RR: manually by an observer or automatically recording from a physical measurement. Manual methods were further categorized as manual count (timers only) or assisted count (combined timer and counter) (Table 1). Automated technologies were categorized based on the physiological parameter from which the measurement was obtained: exhaled breath (humidity, temperature, air pressure, carbon dioxide, oxygen), thoracic effort (circumference, motion, volume), respiratory sounds (oronasal, thoracic, tracheal), and indirect effects on cardiovascular physiology (electrocardiogram, photoplethysmogram [PPG], transcutaneous partial pressure of carbon dioxide, peripheral arterial tonometry, arterial blood pressure).

All data were analyzed qualitatively by assessing the device category, method of evaluation, and accuracy in comparison to a reference standard. An objective of this review was to pool outcome measures across studies through meta-analytic methods. However, the dearth of studies reporting like outcomes precluded this type of quantitative analysis.

Studies Included through Systematic Review

From a total of 7,669 unique citations, 89 publications were identified as sufficiently relevant for full text review, and 16 were ultimately included (Figure 1). Fourteen of those papers reported on automated devices, and two reported on manual devices (Table 2). All included studies on automated devices reported data from well-resourced settings, including the United Kingdom (n = 4), Sweden (n = 3), the Netherlands (n = 3), the United States (n = 1), Israel (n = 1), Switzerland (n = 1), and Australia (n = 1) (1023). Both papers on manual devices reported data from low-resource settings, with one conducted in India and the other consolidating evidence from studies in Ghana, South Sudan, and Uganda (Table 2) (6, 24).

Table 2. Respiratory Rate Measuring Methods Evaluated for Accuracy in Children Younger Than 5 Years of Age

Author, YearDeviceCategoryLocation, SettingUsers, SubjectsReference StandardResults
Manual
 Bang and Bang, 1992 (24)Breath Counter counting beadsAssisted count, counter onlyIndia, householdUsers: TBAs (n = 10); subjects: children <5 yr (n = 5)Physician count, 1 min82% TBAs accurately identified fast breathing using Breath Counter, compared with 60% accuracy without using Breath Counter (P < 0.05).
 Noordam et al., 2015 (6)Study 1: ARI timer vs. ARI timer and counting beadsAssisted count, combined timer and counterStudy 1: South Sudan and Uganda, householdStudy 1: Users: CHWs (n = 65); subjects: children 2–59 mo (n = 65)Physician count, 1 min (accuracy window ±3 breaths for Study 1 and Study 2, ±2 breaths for Study 3)Study 1: Illiterate CHWs 5.7 times as likely to accurately classify fast breathing when using ARI timer and counting beads together compared with using ARI timer alone (OR, 5.7; P < 0.005).
Study 2: Counting beadsStudy 2: South Sudan, hospitalStudy 2: Users: CHWs (n = 27); subjects: children 2–59 mo (n = 69)Study 2: Among illiterate CHWs, 60% accurately classified fast breathing using counting beads.
Study 3: ARI timer vs. ARI timer and counting beads; ARI timer vs. mobile phone application in which button pressed for every breath observed, and which beeped after 1 minStudy 3: Uganda, virtual (video)Study 3: Users: CHWs (n = 94); subjects: video case series (each CHW shown two children with fast breathing and one without)Study 3: Literate CHWs 5.6 times as likely to accurately classify fast breathing when using ARI timer alone, compared with using ARI timer and counting beads together (OR, 5.6; P < 0.001). No significant difference in accuracy between ARI timer and mobile phone application (OR, 1.1; P = 0.08).
Automatic
 Johansson et al., 1999 (10)PPGIndirect effects on cardiovascular physiology/blood flowSweden, NICUUsers: research team; subjects: neonates (n = 6)TTI, 30 sPPG signal 2.7% (SD, ±1.1%) false-negative breaths and 1.5% (SD, ±0.4%) false-positive breaths
 Olsson et al., 2000 (11)PPG at three monitoring sites (leg, buttock, and IR)Indirect effects on cardiovascular physiology/blood flowSweden, NICU and neonatal intermediate care unitUsers: research team; subjects: infants (n = 10)TTI, 30 sStrong association between RR recorded by PPG and RR recorded by TTI at all three sites, with correlation coefficients as follows:
Leg: 0.995 (95% CI, 0.980–0.999)
IR: 0.997 (95% CI, 0.988–0.999)
Buttock: 0.995 (95% CI, 0.980–0.999)
 Neuman et al., 2001 (12)CHIME device, which measures RR using Respitrace Plus RIPIndirect effects on cardiovascular physiology/blood flowUnited States, home (data collection) and CHIME base station (interpretation)Users: research team; subjects: infants (n = 20)Human observer using simultaneously recorded CO2 and oronasal air temperature (thermistor probe) to identify breaths, 5 minSensitivity: 96%
Specificity: 65%
 Roback et al., 2005 (13)FOREFORE: exhaled breath, humiditySweden, NICUUsers: research team; subjects: neonates (n = 17)Manual count (1 min, repeated every 6 min until at least 8 recordings were obtained)FORE: Mean bias, −3.4%; deviation from reference of more than 20% in 22.7% observations
TTITTI: thoracic effort, VtTTI: Mean bias, −1.6%; deviation from reference of more than 20% in 23.8% observations
 Wertheim et al., 2009 (14)Pleth traceIndirect effects on cardiovascular physiology/blood flowUnited Kingdom, postnatal hospital wardUsers: research team; subjects: infants (n = 14)Respiratory airflow measured using low–dead space ultrasonic flowmeter connected to face mask, 2 minMedian difference (pleth − flow), −0.01 breaths/min (range, −5.84 to 0.76; P = 0.802). Mean difference (pleth − flow) using Bland-Altman analysis, −0.82 breaths/min. Using 1 Hz LPF to generate a respiratory-like trace, median difference, −0.89 breaths/min (range, −8.41 to 0.38; P = 0.038).
 Wertheim et al., 2013 (15)Pleth recordingsIndirect effects on cardiovascular physiology/blood flowUnited Kingdom, inpatient hospital unitUsers: research team; subjects: acutely wheezy children aged 1–4 yr (n = 18)Manual count by direct observation of chest wall movement, 30 sAnalysis of pleth within 10 breaths/min of manual count in 15 of 18 children during acute wheezing and 15 of 16 children at follow-up. Using paired t test, no significant difference between pleth and manual count during acute wheezing or follow-up.
 Petrus et al., 2015 (16)Tidal breathing flow–volume loop measurements using vest-based inductive plethysmograph system (FloRight)Thoracic effort, VtSwitzerland, hospitalUsers: research team; subjects: healthy infants (n = 19); infants with lung disease (n = 18)Ultrasonic flowmeter with face mask (Spiroson)Mean difference (vest − mask), 0.71 breaths/min (95% CI, 0.24–1.17; P = 0.031).
 Kohn et al., 2016 (17)1. Transthoracic impedance continuously monitored via standard pediatric chest electrodesThoracic effort, VtIsrael, NICUUsers: research team; subjects: premature infants (n = 9; 11 sessions)Manual count using video recording of chest and abdomen (1-min span counted every 5 min)Tight correlation between motion sensor modality and visual count (r2 = 0.83), with slope of 0.96, very close to line of equality. Impedance-based RR determinations had lower correlation with manual count (r2 = 0.65) and larger deviation from expected line of equality.
2. Continuous monitoring of respiratory dynamics and measurement of local tidal displacement using three motion sensors attached to both sides of chest wall and upper abdomen
 Shah et al., 2015 (18)PPG using modified autoregressive modeling; RR measured over 60-s sliding window with 50-s overlapIndirect effects on cardiovascular physiology/blood flowUnited Kingdom, hospital emergency departmentUsers: research team; subjects: children aged 0–5 yr (n = 126)Manual count by nurse (15 or 30 s)Mean absolute error in breaths per minute between intervention and reference, 7.6.
Median absolute error, 6.0.
 Kraaijenga et al., 2015 (19)dEMG; 1-h recording taken on Days 1, 3, and 7; calculated mean RR from 1-min time intervals at six fixed time points (5, 15, 25, 35, 45, and 55 min)Thoracic effort, thoracic motionNetherlands, NICUUsers: research team; subjects: preterm infants (n = 31)Chest impedance, measured at same time points as dEMGRR measured by dEMG significantly correlated to chest impedance (r = 0.85; P < 0.001). Bland-Altman plot between techniques with mean difference of −2.3 breaths/min.
 van Gastel et al., 2016 (20)Noncontact camera detecting respiratory-induced color differences of the skin; two remote PPG algorithms used: CHROM and PBVIndirect effects on cardiovascular physiology/blood flowNetherlands, NICUUsers: research team; subjects: neonates (n = 2; 20 videos analyzed)Overall performance compared with ECG-derived respiratory signal. CHROM and PBV algorithms compared with two benchmark PPG-based algorithmsCHROM strongest performer of four algorithms, with correlation coefficient (r) of 0.87 and mean absolute error of 4.67 breaths, compared with ECG-derived respiratory signal.
 Seddon et al., 2016 (21)Pleth traceIndirect effects on cardiovascular physiology/blood flowUnited Kingdom, homeUsers: parents; subjects: healthy preterm infants with no respiratory disease (n = 12); preterm infants who subsequently developed chronic lung disease of prematurity (n = 9)RIP bandsMedian difference (RIP − pleth), 0 (range, −1.75 to 6.5 breaths/min). Bland-Altman plot indicated no difference in accuracy at higher rates.
 Janssen et al., 2016 (22)VRM systemThoracic effort, thoracic motionNetherlands, NICUUsers: research team; subjects: neonates (n = 2; 20 videos)Contact-based thoracic impedance plethysmography (ECG)Average accuracy of VRM in close-view and wide-view videos among neonates, 88.7 and 92.6%, respectively.
 Al-Naji and Chahl, 2016 (23)Thermal camera using a motion magnification technique to magnify breathing movement (15-s video clips)Thoracic effort, thoracic motionAustralia, setting not statedUsers: research team; subject: 8-mo-old infant (n = 1)Visual count using input videoCorrelation analysis between experimental device and reference carried out using PCC and SRC (PCC, 0.966; SRC, 0.9566). Reproducibility coefficient, 0.67 breaths/min (2.8%) and mean difference (bias) 0.21, with limits of agreement +0.88 and −0.46.

Definition of abbreviations: ARI = acute respiratory infection; CI = confidence interval; CHIME = Collaborative Home Infant Monitor Evaluation; CHROM = chrominance-based; CHW = community health worker; dEMG = transcutaneous electromyography; FORE = fiberoptic respirometry; IR = interscapular region; LPF = low-pass filtering; NICU = neonatal ICU; OR = odds ratio; PBV = pulse blood volume; PCC = Pearson correlation coefficient; pleth = pulse oximeter plethysmogram; PPG = photoplethysmogram; RIP = respiratory inductance plethysmograph; RR = respiratory rate; SRC = Spearman Rho coefficient; TBA = traditional birth attendant; TTI = transthoracic impedance; VRM = video respiration monitoring.

Manual Devices

The two publications describing the accuracy of fast breathing assessment using manual devices included primarily low literacy, community-based, frontline providers in low-resource settings and used a reference standard of clinician count over 1 minute (Table 2). In a study conducted in India in the early 1990s, the accuracy of fast breathing assessment by 10 traditional birth attendants was higher when assisted by age-specific color-coded counting beads in comparison to no device (82 vs. 60%) (24). In more recent years, studies assessing the relative benefit of counting beads in conjunction with an acute respiratory infection timer have been conducted to inform iCCM guidelines on pneumonia (6). Formative research in Ghana suggested that color-coded beads could facilitate classification accuracy by assisting community health workers (CHWs) in identifying fast breathing by age without the need to remember age-specific cutoff rates. Pooled data from studies among primarily illiterate CHWs in South Sudan and Uganda indicate a 41% absolute increase in the accuracy of fast breathing classification when assisted by counting beads in conjunction with a timer compared with an acute respiratory infection timer alone (odds ratio [OR], 5.7; P < 0.005) (Table 2). In South Sudan, the use of counting beads enabled 60% of illiterate CHWs to accurately classify fast breathing. In Uganda, findings differed based on the literacy level of the CHWs. Among illiterate CHWs, the ability to classify fast breathing increased from 37% using the timer alone to 73% using the timer and counting beads combined (OR, 4.4; P < 0.005). However, literate CHWs were 5.6 times as likely to report a breath count within plus or minus two breaths of the reference standard when assisted with a timer alone compared with a timer in conjunction with counting beads (OR, 5.6; P < 0.001). A mobile phone application was also assessed in this study, and no significant difference in accuracy was noted between the timer and the mobile phone application (OR, 1.1; P = 0.08) (6). Among literate CHWs using any method, breath count was typically more accurate at slower breathing rates rather than faster rates, and CHWs tended to overestimate RR in the slow-rate scenario and underestimate RR in the fast-rate scenario.

Automated Devices

We identified 14 studies assessing the accuracy of automated breath counters (Table 2). One study assessed multiple devices, for a total of 15 technologies assessed. Automated breath counters fell into four categories: indirect effects on cardiovascular physiology/blood flow (n = 8); thoracic effort, Vt (n = 3); thoracic effort, thoracic motion (n = 3); and exhaled breath humidity (n = 1) (Table 2).

Indirect effects on cardiovascular physiology/blood flow

Two studies describing PPG took place in a neonatal ICU (NICU) in Sweden, using transthoracic impedance (TTI) plethysmography as the reference standard (10, 11). In one study, a PPG probe, designed to cover a small skin surface area while still having a sufficiently large detector area, was attached to neonates’ skin on the lateral left thigh, and RR was extracted using a band-pass filter. When compared with the reference, the PPG signal included 2.7% (±1.1%) false-negative breaths and 1.5% (±0.4%) false-positive breaths (mean ± SD) (10). In another study, PPG was used to assess RR at three monitoring sites (i.e., leg, buttock, and interscapular region). A strong association between RR recorded by PPG and RR recorded by TTI was noted at all three sites, with the highest correlation in the interscapular region (correlation coefficient, 0.997; 95% confidence interval, 0.988–0.999) (11).

A device developed for the Collaborative Home Infant Monitor Evaluation (CHIME) project in the United States involving multiple components including a pulse oximeter, a thoracic impedance monitor, and a Respitrace Plus respiratory inductance plethysmography (RIP) device collected data from infants at their homes, and data was analyzed at the CHIME base station. Compared with a human observer using simultaneously recorded CO2 and oronasal air temperature to identify breaths, the CHIME monitor was found to have 96% sensitivity and 65% specificity (12).

Three additional studies assessed the accuracy of RR detection using a pulse oximeter plethysmographic (pleth) trace (14, 15, 21). In a study analyzing pleth traces among infants in a postnatal hospital ward in the United Kingdom and comparing to the reference standard of respiratory airflow measured using an ultrasonic flowmeter connected to a face mask, the median difference was −0.01 breaths/min and the mean difference (using Bland-Altman analysis) was −0.82 breaths/min (14). In a study analyzing pleth traces among acutely wheezy children in an inpatient hospital unit in the United Kingdom, analysis of pleth was within 10 breaths/min of the manual count by direct observation reference standard in 15 of 18 children during acute wheezing and in 15 of 16 children at follow-up. Using a paired t test, there was no significant difference between pleth and manual count during acute wheezing or follow-up (15). In another study from the United Kingdom assessing the accuracy of pleth traces using home recordings of both healthy preterm infants and preterm infants with chronic lung disease of prematurity and using RIP bands as the reference standard, the Bland-Altman plot indicated no difference in accuracy at higher rates, and the median difference was 0 (21).

In a study evaluating PPG recordings using modified autoregressive modeling among children aged 0 to 5 years of age presenting at a hospital emergency department in the United Kingdom, and using manual count as the reference standard, the mean absolute error was 7.6 breaths/min and the median absolute error was 6.0 breaths/min (18). Finally, a study using a noncontact camera to detect respiratory-induced color differences of the skin among a sample of neonates in a NICU in the Netherlands, and using an ECG-derived respiratory signal as the reference standard, the camera method demonstrated a mean absolute error of 4.67 breaths/min and a correlation coefficient of 0.87 (20).

Thoracic effort, Vt

To measure Vt to assess RR, TTI was used among a sample of neonates hospitalized in a NICU in Sweden, using manual count as the reference standard. TTI demonstrated a mean bias of −1.6%, and a deviation from the reference of more than 20% was found in 23.8% of observations. The accuracy of TTI was subject dependent, with decreased accuracy during body movement, and TTI tended to overestimate because of motion artifacts (13). In another study among a sample of premature infants in a NICU in Israel, two methods to assess RR were used 1) TTI continuously monitored via standard pediatric chest electrodes, and 2) continuous monitoring of respiratory dynamics and measurement of local tidal displacement using motion sensors attached to the chest wall and upper abdomen. Using manual count via video recording as the reference standard, a tight correlation was obtained between the motion sensor modality and the visual count (r2 = 0.83), with a slope of 0.96. The impedance-based measurements had a lower correlation with the manual count (r2 = 0.65) and a larger deviation from the expected line of equality (17). In a study among both healthy infants and infants with lung disease in a hospital setting in Switzerland, a FloRight vest–based inductive plethysmograph system was used to collect tidal breathing flow–volume loop measurements to assess RR. Using an ultrasonic flowmeter with a face mask as the reference standard, the mean difference was 0.71 breaths/min (95% confidence interval, 0.24–1.17; P = 0.031) (16).

Thoracic effort, thoracic motion

Thoracic motion was captured using transcutaneous electromyography to calculate RR among preterm infants in a NICU in the Netherlands. Using chest impedance as the reference standard, RR measured by transcutaneous electromyography was significantly correlated to chest impedance (r = 0.85; P < 0.001). The Bland-Altman plot between techniques showed a mean difference of −2.3 breaths/min (19). In another NICU-based study in the Netherlands, a video respiration monitoring system was used to assess RR among neonates. With contact-based thoracic impedance plethysmography as a reference, the average accuracy of the video system in close-view and wide-view among neonates was, respectively, 88.7 and 92.6% (22). In a study from Australia, a thermal camera using a motion magnification technique was used to identify respirations in 15-second video clips taken from a single 8-month-old infant. Using visual count from the same input video as the reference standard, correlation analysis between the experimental device and the reference was performed using Pearson correlation coefficient and Spearman Rho coefficient (Pearson correlation coefficient, 0.966; Spearman Rho coefficient, 0.9566). The reproducibility coefficient was 0.67 breaths/min (2.8%) and the mean difference was 0.21 (23).

Exhaled breath, humidity

To assess the humidity of exhaled breath to measure RR among neonates in a NICU in Sweden and using manual count as the reference, fiber optic respirometry found a mean bias of −3.4% and a deviation from the reference of more than 20% in 22.7% of observations. The accuracy of fiber optic respirometry was found to be subject dependent, with decreased accuracy during body movement, and tended to underestimate RR because of probe displacement (13).

Given the immense burden of childhood pneumonia and the fact that RR is the primary method for diagnosing pneumonia in low-resource settings, it is critical to understand the landscape of RR measurement technologies. This systematic review identified 3 manual and 15 automated RR counting devices evaluated for accuracy among spontaneously breathing children younger than 5 years of age. Although automated technologies were divided a priori into 4 categories and 16 subcategories (Table 1), only 4 of the subcategories were identified in this review: PPG (n = 8), Vt (n = 3), thoracic motion (n = 3), and exhaled breath humidity (n = 1) (Table 2). Although devices in other categories may be in the development or evaluation stage, these four categories represent devices whose accuracy has been assessed against a reference standard. No devices in the respiratory sounds category were identified in this review. Acoustically derived respiratory devices may be subject to signal artifact, leading to difficulties in obtaining an accurate count. On the basis of the devices identified in this review, promising RR technologies include noncontact devices, those that can detect changes in motion or color among children of a range of ages and skin tones, and those that may be integrated into an existing device like a pulse oximeter or a multiuse device such as a tablet or smart phone.

A rigorous evaluation of accuracy is necessary to validate any technology before widespread use as a diagnostic tool. Although all studies in this review included an assessment of accuracy, there was wide variation in reference standard selection and statistical methods of comparison. This variation precluded the ability to compare devices head-to-head and complicated attempts to develop standard criteria or a cutoff for determining accuracy. Among the studies evaluating automated technologies, reference standards included manual count (n = 5), TTI (n = 2), ECG (n = 2), ultrasonic flowmeter connected to face mask (n = 2), RIP bands (n = 1), chest impedance (n = 1), and human observer using simultaneously recorded CO2 and oronasal air temperature to identify breaths (n = 1). This variation in reference standards made interpreting the accuracy of the experimental devices challenging, because the accuracy of the reference standards themselves was not clear. Consensus regarding an appropriate reference standard would improve the generalizability of future RR device evaluations.

Although manual breath count was the most common reference standard in this review, it is known to have issues with accuracy and reliability (13, 18). Challenges to manual count include distraction, an agitated or crying child, and mistaking nonrespiratory movements and sounds as breaths (13, 18, 25). Furthermore, if the premise of evaluating RR measurement technologies is to develop a device that improves on the accuracy of manual count, one would need a reference standard accepted as more accurate than manual count. In the absence of a gold standard for RR measurement, we would recommend that future RR accuracy assessments include a clear indication of the start of the breath count and that the repeatability of the reference standard, whether it be manual count, auscultation, video recordings, and/or capnography, be evaluated with simultaneous observations and methods. This methodology will allow for the estimation of the measurement error relative to the device accuracy.

The appropriate methods and the challenges involved in measuring agreement between a clinical standard and a new device have previously been described in detail (26, 27). With regard to statistical methods, we would suggest that the accuracy of RR measurement be reported as the root-mean-square deviation between measured values and reference values over a range of RRs (as is done with devices such as pulse oximeters). It should be clearly stated whether observations were independent (more than one observation per subject). Although there is no currently accepted reference standard for RR, an indication of test–retest reliability of the standard (the variation in measurements taken by a single person or instrument on the same subject) would give an indication of precision.

Another factor to consider is the definition of an accurate RR count. The current UNICEF target product profile for acute respiratory infection diagnostic aids requires that the “accuracy of obtained RR should be at least ±2 breaths per minute when compared with the number of respiratory cycles measured over a period of 60 seconds” (28). If a child is breathing at 20 breaths/min, a single missed breath at the start or end of counting (due to the variation in the count’s start/end time) would account for this 2% error. This was confirmed in a study that measured RR twice over a 10-minute period with the same observer and found a mean difference of 1.8 breaths/min (29). Although a root-mean-square deviation of less than 2.5 breaths/min at 20 breaths/min appears technically achievable, the current cutoff of ±2 breaths/min may be too strict, given the aforementioned challenges in starting and stopping times (30).

In well-resourced settings, measuring RR and other vital signs can be achieved in multiple ways, sometimes involving sophisticated and expensive sensors and signal processing. However, the greatest need for accurate RR technologies is in low-resource settings, where such high-cost technologies may not be appropriate or feasible. Although all of the manual devices assessed were evaluated in low-resource settings, all of the automated devices were tested in well-resourced settings, often in tightly controlled environments. In the absence of validation in low-resource settings, it is difficult to assess the strengths and weaknesses of the devices included in this review. However, low-cost devices with multiple uses (including the CHIME monitor and those that assess RR using a pulse oximeter) may have an advantage over single-purpose devices, because they would allow providers to assess multiple vital signs with a single device, potentially reducing the complexity and training involved in incorporating a new technology into clinical care.

It is critical that RR devices are developed with the needs of low-resource settings in mind and should be validated in the settings where they will ultimately be used. In-country stakeholders and CHWs should be consulted during the planning and development phases, as they may provide different perspectives (31). Characteristics of an RR counter most promising for advancement include portability, durability, and low cost (Table 3). The RR device should be simple to use and should provide the result in a way that a CHW or the child’s caregiver can understand (31, 32).

Table 3. Performance Needs for Respiratory Rate Counter Technologies

ReliabilityAutomatically and accurately measures respiratory rate
NoninvasivenessEasily placed on and tolerated by an infant or child for the shortest time possible
AffordabilityLow cost
SimplicitySimple and intuitive to use
Cultural appropriatenessResult can be understood by caregiver
PortabilityCompact, lightweight, and portable
DurabilityRobust, dust- and water-resistant
SustainabilityIndependent and long-lasting power source
CustomizationCredible and culturally acceptable
MinimalismLeast essential hardware and appropriate for low-skilled community health workers
ValidationPrototype refined and demonstrates significant progress throughout product development cycle
IntegrationLeverages a multipurpose-built platform to provide an integrated diagnostic toolkit

Based on information in Reference 32.

The manual count assessments identified in this review targeted lower-level providers like CHWs and traditional birth attendants as the users of the device. In contrast, in all but one of the automated device evaluations, research team members were the users; the exception was a study of pulse oximeter pleth trace, in which parents were the users. As RR technologies move further along in development and validation, it will be important to determine the level of training required for accurate use as well as the appropriate amount of information to present. Whether the RR device should present only the breath count or whether it should identify tachypnea depends on the training of the end user and the needs of the setting. If the device does identify tachypnea, the device should be agile enough to have settings that may be adapted if guidelines change globally or regionally.

In performing a landscape analysis of available RR technologies, it is also important to address the value of RR as a diagnostic tool for pneumonia. Many factors can influence RR, including temperature, agitation, and whether the child is awake. A 2017 systematic review assessing the accuracy of clinical symptoms in identifying children with radiographic pneumonia found that among children with cough, fever, or both, tachypnea did not increase the likelihood of radiographic pneumonia (33). Although tachypnea may be a tool in the diagnosis of pneumonia, tachypnea should be assessed in tandem with other clinical signs and symptoms (such as work of breathing) and should not act as a stand-alone diagnostic criterion.

There are clear limitations of current RR devices. Manual counter technologies do not eliminate reliance on the observer or susceptibility to observer error and distractions. Automated RR technologies can be expensive and may not be feasible for low-resource settings. Contact-based monitors may not be suitable for neonates because of the fragility of their skin and may distress the child, which could impact the child’s breathing pattern (13, 22, 34). Movement-based RR counter technologies can be vulnerable to motion artifacts. The challenge is to separate respiratory from nonrespiratory signals, including discriminating a moving child from a breath movement or a cry from a breath sound (13, 19, 20). Technologies that calculate RR on the basis of a subset of breaths may be biased by respiratory pauses or brief periods of tachypnea in healthy newborns (35). In younger children, periods of respiratory distress and fast breathing can be irregular and intermittent. With this in mind, RR measurement tools should allow for an assessment over a period of time rather than at a single time point. The challenge in this is finding a middle ground between a rapid device appropriate for health workers with limited time and a tool with a long enough window to capture variation and identify the peak RR.

There were also limitations to our landscape analysis and literature review, most significantly that our review may not include all RR counters in development. Work on RR counting and monitoring technologies spans many disciplines that use disparate methods and have different aims. Studies that did not include an assessment of accuracy against a reference standard and those that did not provide results among spontaneously breathing children younger than 5 years of age were excluded from this analysis. We also did not review the non-English literature, and we were not able to verify whether the RR counters highlighted here have moved beyond the concept or development phase. Although some of the included RR counters did list costs, cost estimates were not always available and were excluded from this analysis. Because this review focuses on accuracy, it does not address reliability, feasibility, usability, and acceptability. However, these factors are important to evaluate and consider when introducing a new technology.

Finally, although this systematic review was not prospectively registered, several steps were taken to minimize selection bias. The inclusion and exclusion criteria were developed a priori, and an effort was made to minimize ambiguity within these criteria. Two individuals screened the citations for inclusion in this review to reduce the risk of biased study decisions. All decisions regarding inclusion or exclusion of each citation were thoroughly documented (36).

Conclusions

Assessment of RR is integral to the pneumonia diagnostic pathway in low-resource settings. Accelerating the development of innovations and spurring the adaptation of current devices could significantly improve the process of pneumonia diagnosis. New devices not only must be evaluated for their effectiveness in measuring RR in young children and infants but also should be designed specifically for use in low-resource settings. This should involve prioritizing the needs of healthcare providers and understanding the constraints of using this technology in low-resource settings. Furthermore, the lack of a gold standard or even a common reference standard may impede the process of developing and evaluating future RR devices. By identifying existing RR counters and highlighting the gaps in current evaluation protocols, we hope to encourage researchers, developers, manufacturers, and innovators to address the need for better ways to diagnose pneumonia and reduce childhood mortality in low-resource settings. A single accurate measurement of RR alone is unlikely to provide a clear diagnosis of pneumonia, but, rather, measurement will need to be repeated over time and combined with other clinical signs and symptoms interpreted by a trained frontline health worker.

The authors thank Jaclyn Delarosa for her assistance in developing the device categories and Laura Lamberti for her support on a previous draft of this manuscript.

1. World Health Organization. Fact sheet: pneumonia. 2016 [accessed 2017 Aug 18]. Available from: http://www.who.int/mediacentre/factsheets/fs331/en/.
2. Rambaud-Althaus C, Althaus F, Genton B, D’Acremont V. Clinical features for diagnosis of pneumonia in children younger than 5 years: a systematic review and meta-analysis. Lancet Infect Dis 2015;15:439450.
3. Nair H, Simões EA, Rudan I, Gessner BD, Azziz-Baumgartner E, Zhang JSF, et al.; Severe Acute Lower Respiratory Infections Working Group. Global and regional burden of hospital admissions for severe acute lower respiratory infections in young children in 2010: a systematic analysis. Lancet 2013;381:13801390.
4. World Health Organization. Integrated management of childhood illness chart booklet. Geneva: World Health Organization; 2014.
5. Gadomski AM, Khallaf N, el Ansary S, Black RE. Assessment of respiratory rate and chest indrawing in children with ARI by primary care physicians in Egypt. Bull World Health Organ 1993;71:523527.
6. Noordam AC, Barberá Laínez Y, Sadruddin S, van Heck PM, Chono AO, Acaye GL, et al. The use of counting beads to improve the classification of fast breathing in low-resource settings: a multi-country review. Health Policy Plan 2015;30:696704.
7. Synovate Uganda. ARI timer research report for UNICEF: technical report. Kampala, Uganda: Synovate Uganda; 2011.
8. Lenahan JL, Ginsburg A. Towards a better diagnosis of childhood pneumonia: a systematic review of tools to measure respiratory rate. Presented at the 48th Union World Conference on Lung Health. October 11–14, 2017, Guadalajara, Mexico. Abstract A-911-0011-00681.
9. Moher D, Liberati A, Tetzlaff J, Altman DG; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med 2009;6:e1000097.
10. Johansson A, Oberg PA, Sedin G. Monitoring of heart and respiratory rates in newborn infants using a new photoplethysmographic technique. J Clin Monit Comput 1999;15:461467.
11. Olsson E, Ugnell H, Oberg PA, Sedin G. Photoplethysmography for simultaneous recording of heart and respiratory rates in newborn infants. Acta Paediatr 2000;89:853861.
12. Neuman MR, Watson H, Mendenhall RS, Zoldak JT, Di Fiore JM, Peucker M, et al.; Collaborative Home Infant Monitor Evaluation (CHIME) Study Group. Cardiopulmonary monitoring at home: the CHIME monitor. Physiol Meas 2001;22:267286.
13. Roback K, Nelson N, Johansson A, Hass U, Strömberg T. A new fiberoptical respiratory rate monitor for the neonatal intensive care unit. Pediatr Pulmonol 2005;39:120126.
14. Wertheim D, Olden C, Savage E, Seddon P. Extracting respiratory data from pulse oximeter plethysmogram traces in newborn infants. Arch Dis Child Fetal Neonatal Ed 2009;94:F301F303.
15. Wertheim D, Olden C, Symes L, Rabe H, Seddon P. Monitoring respiration in wheezy preschool children by pulse oximetry plethysmogram analysis. Med Biol Eng Comput 2013;51:965970.
16. Petrus NC, Thamrin C, Fuchs O, Frey U. Accuracy of tidal breathing measurement of FloRight compared to an ultrasonic flowmeter in infants. Pediatr Pulmonol 2015;50:380388.
17. Kohn S, Waisman D, Pesin J, Faingersh A, Klotzman IC, Levy C, et al. Monitoring the respiratory rate by miniature motion sensors in premature infants: a comparative study. J Perinatol 2016;36:116120.
18. Shah SA, Fleming S, Thompson M, Tarassenko L. Respiratory rate estimation during triage of children in hospitals. J Med Eng Technol 2015;39:514524.
19. Kraaijenga JV, Hutten GJ, de Jongh FH, van Kaam AH. Transcutaneous electromyography of the diaphragm: a cardio-respiratory monitor for preterm infants. Pediatr Pulmonol 2015;50:889895.
20. van Gastel M, Stuijk S, de Haan G. Robust respiration detection from remote photoplethysmography. Biomed Opt Express 2016;7:49414957.
21. Seddon P, Sobowiec-Kouman S, Wertheim D. Infant home respiratory monitoring using pulse oximetry. Arch Dis Child 2016; [online ahead of print] 7 Jul 2016; DOI: 10.1136/archdischild-2016-310712.
22. Janssen R, Wang W, Moço A, de Haan G. Video-based respiration monitoring with automatic region of interest detection. Physiol Meas 2016;37:100114.
23. Al-Naji A, Chahl J. Remote respiratory monitoring system based on developing motion magnification technique. Biomed Signal Proc Control 2016;29:110.
24. Bang AT, Bang RA. Breath Counter: a new device for household diagnosis of childhood pneumonia. Indian J Pediatr 1992;59:7984.
25. Simoes EAF, Roark R, Berman S, Esler LL, Murphy J. Respiratory rate: measurement of variability over time and accuracy at different counting periods. Arch Dis Child 1991;66:11991203.
26. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307310.
27. Severinghaus JW, Naifeh KH, Koh SO. Errors in 14 pulse oximeters during profound hypoxia. J Clin Monit 1989;5:7281.
28. UNICEF. Target product profile: Acute Respiratory Infection Diagnostic Aid (ARIDA). Copenhagen: UNICEF Supply Division; 2014.
29. Muro F, Mosha N, Hildenwall H, Mtei F, Harrison N, Schellenberg D, et al. Variability of respiratory rate measurements in children suspected with non-severe pneumonia in north-east Tanzania. Trop Med Int Health 2017;22:139147.
30. Karlen W, Gan H, Chiu M, Dunsmuir D, Zhou G, Dumont GA, et al. Improving the accuracy and efficiency of respiratory rate measurements in children using mobile devices. PLoS One 2014;9:e99266.
31. Spence H, Baker K, Wharton-Smith A, Mucunguzi A, Matata L, Habte T, et al. Childhood pneumonia diagnostics: community health workers’ and national stakeholders’ differing perspectives of new and existing aids. Glob Health Action 2017;10:1290340.
32. UNICEF. Target product profile: Pneumonia acute respiratory infection diagnostic aid - introduction. Copenhagen: UNICEF Supply Division; 2014.
33. Shah SN, Bachur RG, Simel DL, Neuman MI. Does this child have pneumonia?: The rational clinical examination systematic review. JAMA 2017;318:462471.
34. Al-Khalidi F, Saatchi R, Elphick H, Burke D. An evaluation of thermal imaging based respiration rate monitoring in children. Am J Eng Appl Sci 2011;4:586597.
35. Brooks LJ, DiFiore JM, Martin RJ; Collaborative Home Infant Monitoring Evaluation. Assessment of tidal volume over time in preterm infants using respiratory inductance plethysmography, the CHIME Study Group. Pediatr Pulmonol 1997;23:429433.
36. McDonagh M, Peterson K, Parminder R, Chang S, Shekelle P. Avoiding bias in selecting studies. In: Methods guide for comparative effectiveness reviews. Rockville, MD: Agency for Healthcare Research and Quality; 2012.
Correspondence and requests for reprints should be addressed to Jennifer L. Lenahan, M.P.H., 501 Kings Highway East, Suite 400, Fairfield, CT 06825. E-mail: .

Author Contributions: A.S.G.: study design, search term development, device category development, literature search, article review and analysis, table and figure design, and manuscript writing and editing; J.L.L.: literature search, article review and analysis, table and figure design, and manuscript writing and editing; R.I.: manuscript review and editing; and J.M.A.: manuscript writing, review, and editing.

CME will be available for this article at www.atsjournals.org.

Originally Published in Press as DOI: 10.1164/rccm.201711-2233CI on February 23, 2018

Author disclosures are available with the text of this article at www.atsjournals.org.

Related

No related items
Comments Post a Comment




New User Registration

Not Yet Registered?
Benefits of Registration Include:
 •  A Unique User Profile that will allow you to manage your current subscriptions (including online access)
 •  The ability to create favorites lists down to the article level
 •  The ability to customize email alerts to receive specific notifications about the topics you care most about and special offers
American Journal of Respiratory and Critical Care Medicine
197
9

Click to see any corrections or updates and to confirm this is the authentic version of record