Prediction models for post-discharge mortality among under-five children with suspected sepsis in Uganda: A multicohort analysis.

In many low-income countries, over five percent of hospitalized children die following hospital discharge. The lack of available tools to identify those at risk of post-discharge mortality has limited the ability to make progress towards improving outcomes. We aimed to develop algorithms designed to predict post-discharge mortality among children admitted with suspected sepsis. Four prospective cohort studies of children in two age groups (0-6 and 6-60 months) were conducted between 2012-2021 in six Ugandan hospitals. Prediction models were derived for six-months post-discharge mortality, based on candidate predictors collected at admission, each with a maximum of eight variables, and internally validated using 10-fold cross-validation. 8,810 children were enrolled: 470 (5.3%) died in hospital; 257 (7.7%) and 233 (4.8%) post-discharge deaths occurred in the 0-6-month and 6-60-month age groups, respectively. The primary models had an area under the receiver operating characteristic curve (AUROC) of 0.77 (95%CI 0.74-0.80) for 0-6-month-olds and 0.75 (95%CI 0.72-0.79) for 6-60-month-olds; mean AUROCs among the 10 cross-validation folds were 0.75 and 0.73, respectively. Calibration across risk strata was good: Brier scores were 0.07 and 0.04, respectively. The most important variables included anthropometry and oxygen saturation. Additional variables included: illness duration, jaundice-age interaction, and a bulging fontanelle among 0-6-month-olds; and prior admissions, coma score, temperature, age-respiratory rate interaction, and HIV status among 6-60-month-olds. Simple prediction models at admission with suspected sepsis can identify children at risk of post-discharge mortality. Further external validation is recommended for different contexts. Models can be digitally integrated into existing processes to improve peri-discharge care as children transition from the hospital to the community.

Repeatability of Pulse Oximetry Measurements in Children During Triage in 2 Ugandan Hospitals.

BACKGROUND: In low- and middle-income countries, health workers use pulse oximeters for intermittent spot measurements of oxygen saturation (SpO2). However, the accuracy and reliability of pulse oximeters for spot measurements have not been determined. We evaluated the repeatability of spot measurements and the ideal observation time to guide recommendations during spot check measurements. METHODS: Two 1-minute measurements were taken for the 3,903 subjects enrolled in the study conducted April 2020-January 2022 in Uganda, collecting 1 Hz SpO2 and signal quality index (SQI) data. The repeatability between the 2 measurements was assessed using an intraclass correlation coefficient (ICC), calculated using a median of all seconds of non-zero SpO2 values for each recording (any quality, Q1) and again with a quality filter only using seconds with SQI 90% or higher (good quality, Q2). The ICC was also recalculated for both conditions of Q1 and Q2 using the initial 5 seconds, then the initial 10 seconds, and continuing with 5-second increments up to the full 60 seconds. Lastly, the whole minute ICC was calculated with good quality (Q2), including only records where both measurements had a mean SQI of more than 70% (Q3). RESULTS: The repeatability ICC with condition Q1 was 0.591 (95% confidence interval [CI]=0.570, 0.611). Using only the first 5 seconds of each measurement reduced the repeatability to 0.200 (95% CI=0.169, 0.230). Filtering with Q2, the whole-minute ICC was 0.855 (95% CI=0.847, 0.864). The ICC did not improve beyond the first 35 seconds. For Q3, the repeatability rose to 0.908 (95% CI=0.901, 0.914). CONCLUSIONS: Training guidelines must emphasize the importance of signal quality and duration of measurement, targeting a minimum of 35 seconds of adequate-quality, stable data. In addition, the design of new devices should incorporate user prompts and force quality checks to encourage more accurate pulse oximetry measurements.

Mortality after hospital discharge among children younger than 5 years admitted with suspected sepsis in Uganda: a prospective, multisite, observational cohort study.

BACKGROUND: Substantial mortality occurs after hospital discharge in children younger than 5 years with suspected sepsis, especially in low-income countries. A better understanding of its epidemiology is needed for effective interventions to reduce child mortality in these countries. We evaluated risk factors for death after discharge in children admitted to hospital for suspected sepsis in Uganda, and assessed how these differed by age, time of death, and location of death. METHODS: In this prospective, multisite, observational cohort study, we recruited and consecutively enrolled children aged 0-60 months admitted with suspected sepsis from the community to the paediatric wards of six Ugandan hospitals. Suspected sepsis was defined as the need for admission due to a suspected or proven infectious illness. At admission, trained study nurses systematically collected data on clinical variables, sociodemographic variables, and baseline characteristics with encrypted study tablets. Participants were followed up for 6 months after discharge by field officers who contacted caregivers at 2 months and 4 months after discharge by telephone and at 6 months after discharge in person to measure vital status, health-care seeking after discharge, and readmission details. We assessed 6-month mortality after hospital discharge among those discharged alive, with verbal autopsies conducted for children who had died after hospital discharge. FINDINGS: Between July 13, 2017, and March 30, 2020, 16 991 children were screened for eligibility. 6545 children (2927 [44·72%] female children and 3618 [55·28%] male children) were enrolled and 6191 were discharged from hospital alive. 6073 children (2687 [44·2%] female children and 3386 [55·8%] male children) completed follow-up. 366 children died in the 6-month period after discharge (weighted mortality rate 5·5%). Median time from discharge to death was 28 days (IQR 9-74). For the 360 children for whom location of death was documented, deaths occurred at home (162 [45·0%]), in transit to care (66 [18·3%]), or in hospital (132 [36·7%]) during a subsequent readmission. Death after hospital discharge was strongly associated with weight-for-age Z scores less than -3 (adjusted risk ratio [aRR] 4·7, 95% CI 3·7-5·8 vs a Z score of >-2), discharge or referral to a higher level of care (7·3, 5·6-9·5), and unplanned discharge (3·2, 2·5-4·0). Hazard ratios (HRs) for severe anaemia (<7g/dL) increased with time since discharge, from 1·7 (95% CI 0·9-3·0) for death occurring in the first time tertile to 5·2 (3·1-8·5) in the third time tertile. HRs for some discharge vulnerabilities decreased significantly with increasing time since discharge, including unplanned discharge (from 4.5 [2·9-6·9] in the first tertile to 2·0 [1·3-3·2] in the third tertile) and poor feeding status (from 7·7 [5·4-11·0] to 1·84 [1·0-3·3]). Age interacted with several variables, including reduced weight-for-age Z score, severe anaemia, and reduced admission temperature. INTERPRETATION: Paediatric mortality following hospital discharge after suspected sepsis is common, with diminishing, although persistent, risk during the first 6 months after discharge. Efforts to improve outcomes after hospital discharge are crucial to achieving Sustainable Development Goal 3.2 (ending preventable childhood deaths under age 5 years). FUNDING: Grand Challenges Canada, Thrasher Research Fund, BC Children's Hospital Foundation, and Mining4Life.

Choosing blood pressure thresholds to inform pregnancy care in the community: An analysis of cluster trials.

OBJECTIVE: To inform digital health design by evaluating diagnostic test properties of antenatal blood pressure (BP) outputs and levels to identify women at risk of adverse outcomes. DESIGN: Planned secondary analysis of cluster randomised trials. SETTING: India, Pakistan, Mozambique. POPULATION: Women with in-community BP measurements and known pregnancy outcomes. METHODS: Blood pressure was defined by its outputs (systolic and/or diastolic, systolic only, diastolic only or mean arterial pressure [calculated]) and level: normotension-1 (<135/85 mmHg), normotension-2 (135-139/85-89 mmHg), non-severe hypertension (140-149/90-99 mmHg; 150-154/100-104 mmHg; 155-159/105-109 mmHg) and severe hypertension (≥160/110 mmHg). Dose-response (adjusted risk ratio [aRR]) and diagnostic test properties (negative [-LR] and positive [+LR] likelihood ratios) were estimated. MAIN OUTCOME MEASURES: Maternal/perinatal composites of mortality/morbidity. RESULTS: Among 21 069 pregnancies, different BP outputs had similar aRR, -LR, and +LR for adverse outcomes. No BP level (even normotension-1) was associated with low risk (all -LR ≥0.20). Across outcomes, risks rose progressively with higher BP levels above normotension-1. For each of maternal central nervous system events and stillbirth, BP ≥155/105 mmHg showed at least good diagnostic test performance (+LR ≥5.0) and BP ≥135/85 mmHg at least fair performance, similar to BP ≥140/90 mmHg (+LR 2.0-4.99). CONCLUSIONS: In the community, normal BP values do not provide reassurance about subsequent adverse outcomes. Given the similar performance of BP cut-offs of 135/85 and 140/90 mmHg for hypertension, and 155/105 and 160/110 mmHg for severe hypertension, digital decision support for women in the community should consider using these lower thresholds.

Smart triage: Development of a rapid pediatric triage algorithm for use in low-and-middle income countries.

Introduction: Early and accurate recognition of children at risk of progressing to critical illness could contribute to improved patient outcomes and resource allocation. In resource limited settings digital triage tools can support decision making and improve healthcare delivery. We developed a model for rapid identification of critically ill children at triage. Methods: This was a prospective cohort study of acutely ill children presenting at Jinja Regional Referral Hospital in Eastern Uganda. Variables collected in the emergency department informed the development of a logistic model based on hospital admission using bootstrap stepwise regression. Low and high-risk thresholds for 90% minimum sensitivity and specificity, respectively generated three risk level categories. Performance was assessed using receiver operating characteristic curve analysis on a held-out test set generated by an 80:20 split with 10-fold cross validation. A risk stratification table informed clinical interpretation. Results: The model derivation cohort included 1,612 participants, with an admission rate of approximately 23%. The majority of admitted patients were under five years old and presenting with sepsis, malaria, or pneumonia. A 9-predictor triage model was derived: logit (p) = -32.888 + (0.252, square root of age) + (0.016, heart rate) + (0.819, temperature) + (-0.022, mid-upper arm circumference) + (0.048 transformed oxygen saturation) + (1.793, parent concern) + (1.012, difficulty breathing) + (1.814, oedema) + (1.506, pallor). The model afforded good discrimination, calibration, and risk stratification at the selected thresholds of 8% and 40%. Conclusion: In a low income, pediatric population, we developed a nine variable triage model with high sensitivity and specificity to predict who should be admitted. The triage model can be integrated into any digital platform and used with minimal training to guide rapid identification of critically ill children at first contact. External validation and clinical implementation are in progress.

Health worker perspectives of Smart Triage, a digital triaging platform for quality improvement at a referral hospital in Uganda: a qualitative analysis.

BACKGROUND: Effective triage at hospitals can improve outcomes for children globally by helping identify and prioritize care for those most at-risk of death. Paper-based pediatric triage guidelines have been developed to support frontline health workers in low-resource settings, but these guidelines can be challenging to implement. Smart Triage is a digital triaging platform for quality improvement (QI) that aims to address this challenge. Smart Triage represents a major cultural and behavioural shift in terms of managing patients at health facilities in low-and middle-income countries. The purpose of this study is to understand user perspectives on the usability, feasibility, and acceptability of Smart Triage to inform ongoing and future implementation. METHODS: This was a descriptive qualitative study comprising of face-to-face interviews with health workers (n = 15) at a regional referral hospital in Eastern Uganda, conducted as a sub-study of a larger clinical trial to evaluate Smart Triage (NCT04304235). Thematic analysis was used to assess the usability, feasibility, and acceptability of the platform, focusing on its use in stratifying and prioritizing patients according to their risk and informing QI initiatives implemented by health workers. RESULTS: With appropriate training and experience, health workers found most features of Smart Triage usable and feasible to implement, and reported the platform was acceptable due to its positive impact on reducing the time to treatment for emergency pediatric cases and its use in informing QI initiatives within the pediatric ward. Several factors that reduced the feasibility and acceptability were identified, including high staff turnover, a lack of medical supplies at the hospital, and challenges with staff attitudes. CONCLUSION: Health workers can use the Smart Triage digital triaging platform to identify and prioritize care for severely ill children and improve quality of care at health facilities in low-resource settings. Future innovation is needed to address identified feasibility and acceptability challenges; however, this platform could potentially address some of the challenges to implementing current paper-based systems.

A 2-Phase Survey to Assess a Facility's Readiness for Pediatric Essential Emergency and Critical Care in Resource-Limited Settings: A Literature Review and Survey Development.

OBJECTIVE: Infectious diseases, including pneumonia, malaria, and diarrheal diseases, are the leading causes of death in children younger than 5 years worldwide. The vast majority of these deaths occur in resource-limited settings where there is significant variation in the availability and type of human, physical, and infrastructural resources. The ability to identity gaps in healthcare systems that may hinder their ability to deliver care is an important step to determining specific interventions for quality improvement. Our study objective was to develop a comprehensive, digital, open-access health facility survey to assess facility readiness to provide pediatric critical care in resource-limited settings (eg, low- and lower middle-income countries). METHODS: A literature review of existing facility assessment tools and global guidelines was conducted to generate a database of survey questions. These were then mapped to one of the following 8 domains: hospital statistics, services offered, operational flow, facility infrastructure, staff and training, medicines and equipment, diagnostic capacity, and quality of clinical care. A 2-phase survey was developed and an iterative review process of the survey was undertaken with 12 experts based in low- and middle-income countries. This was built into the REDCap Mobile Application for electronic data capture. RESULTS: The literature review process yielded 7 facility assessment tools and 7 global guidelines for inclusion. After the iterative review process, the final survey consisted of 11 sections with 457 unique questions in the first phase, "environmental scan," focusing on the infrastructure, availability, and functionality of resources, and 3 sections with 131 unique questions in the second phase, "observation scan," focusing on the level of clinical competency. CONCLUSIONS: A comprehensive 2-phase survey was created to evaluate facility readiness for pediatric critical care. Results will assist hospital administrators and policymakers to determine priority areas for quality improvement, enabling them to implement a Plan-Do-Study-Act cycle to improve care for the critically ill child.

Clinical feasibility of an advanced neonatal epidermal multiparameter continuous monitoring technology in a large public maternity hospital in Nairobi, Kenya.

Clinically feasible multiparameter continuous physiological monitoring technologies are needed for use in resource-constrained African healthcare facilities to allow for early detection of critical events and timely intervention for major morbidities in high-risk neonates. We conducted a prospective clinical feasibility study of a novel multiparameter continuous physiological monitoring technology in neonates at Pumwani Maternity Hospital in Nairobi, Kenya. To assess feasibility, we compared the performance of Sibel's Advanced Neonatal Epidermal (ANNE) technology to reference technologies, including Masimo's Rad-97 pulse CO-oximeter with capnography technology for heart rate (HR), respiratory rate (RR), and oxygen saturation (SpO2) measurements and Spengler's Tempo Easy non-contact infrared thermometer for temperature measurements. We evaluated key performance criteria such as up-time, clinical event detection performance, and the agreement of measurements compared to those from the reference technologies in an uncontrolled, real-world setting. Between September 15 and December 15, 2020, we collected and analyzed 503 h of ANNE data from 109 enrolled neonates. ANNE's up-time was 42 (11%) h more for HR, 77 (25%) h more for RR, and 6 (2%) h less for SpO2 compared to the Rad-97. However, ANNE's ratio of up-time to total attached time was less than Rad-97's for HR (0.79 vs 0.86), RR (0.68 vs. 0.79), and SpO2 (0.69 vs 0.86). ANNE demonstrated adequate performance in identifying high and low HR and RR and high temperature events; however, showed relatively poor performance for low SpO2 events. The normalized spread of limits of agreement were 8.4% for HR and 52.2% for RR and the normalized root-mean-square deviation was 4.4% for SpO2. Temperature agreement showed a spread of limits of agreement of 2.8 °C. The a priori-identified optimal limits were met for HR and temperature but not for RR or SpO2. ANNE was clinically feasible for HR and temperature but not RR and SpO2 as demonstrated by the technology's up-time, clinical event detection performance, and the agreement of measurements compared to those from the reference technologies.

Evaluation of Sibel's Advanced Neonatal Epidermal (ANNE) wireless continuous physiological monitor in Nairobi, Kenya.

BACKGROUND: Neonatal multiparameter continuous physiological monitoring (MCPM) technologies assist with early detection of preventable and treatable causes of neonatal mortality. Evaluating accuracy of novel MCPM technologies is critical for their appropriate use and adoption. METHODS: We prospectively compared the accuracy of Sibel's Advanced Neonatal Epidermal (ANNE) technology with Masimo's Rad-97 pulse CO-oximeter with capnography and Spengler's Tempo Easy reference technologies during four evaluation rounds. We compared accuracy of heart rate (HR), respiratory rate (RR), oxygen saturation (SpO2), and skin temperature using Bland-Altman plots and root-mean-square deviation analyses (RMSD). Sibel's ANNE algorithms were optimized between each round. We created Clarke error grids with zones of 20% to aid with clinical interpretation of HR and RR results. RESULTS: Between November 2019 and August 2020 we collected 320 hours of data from 84 neonates. In the final round, Sibel's ANNE technology demonstrated a normalized bias of 0% for HR and 3.1% for RR, and a non-normalized bias of -0.3% for SpO2 and 0.2°C for temperature. The normalized spread between 95% upper and lower limits-of-agreement (LOA) was 4.7% for HR and 29.3% for RR. RMSD for SpO2 was 1.9% and 1.5°C for temperature. Agreement between Sibel's ANNE technology and the reference technologies met the a priori-defined thresholds for 95% spread of LOA and RMSD. Clarke error grids showed that all HR and RR observations were within a 20% difference. CONCLUSION: Our findings suggest acceptable agreement between Sibel's ANNE and reference technologies. Clinical effectiveness, feasibility, usability, acceptability, and cost-effectiveness investigations are necessary for large-scale implementation.

Assessment of neonatal respiratory rate variability.

Accurate measurement of respiratory rate (RR) in neonates is challenging due to high neonatal RR variability (RRV). There is growing evidence that RRV measurement could inform and guide neonatal care. We sought to quantify neonatal RRV during a clinical study in which we compared multiparameter continuous physiological monitoring (MCPM) devices. Measurements of capnography-recorded exhaled carbon dioxide across 60-s epochs were collected from neonates admitted to the neonatal unit at Aga Khan University-Nairobi hospital. Breaths were manually counted from capnograms and using an automated signal detection algorithm which also calculated mean and median RR for each epoch. Outcome measures were between- and within-neonate RRV, between- and within-epoch RRV, and 95% limits of agreement, bias, and root-mean-square deviation. Twenty-seven neonates were included, with 130 epochs analysed. Mean manual breath count (MBC) was 48 breaths per minute. Median RRV ranged from 11.5% (interquartile range (IQR) 6.8-18.9%) to 28.1% (IQR 23.5-36.7%). Bias and limits of agreement for MBC vs algorithm-derived breath count, MBC vs algorithm-derived median breath rate, MBC vs algorithm-derived mean breath rate were - 0.5 (- 2.7, 1.66), - 3.16 (- 12.12, 5.8), and - 3.99 (- 11.3, 3.32), respectively. The marked RRV highlights the challenge of performing accurate RR measurements in neonates. More research is required to optimize the use of RRV to improve care. When evaluating MCPM devices, accuracy thresholds should be less stringent in newborns due to increased RRV. Lastly, median RR, which discounts the impact of extreme outliers, may be more reflective of the underlying physiological control of breathing.

Clinical feasibility of a contactless multiparameter continuous monitoring technology for neonates in a large public maternity hospital in Nairobi, Kenya.

Multiparameter continuous physiological monitoring (MCPM) technologies are critical in the clinical management of high-risk neonates; yet, these technologies are frequently unavailable in many African healthcare facilities. We conducted a prospective clinical feasibility study of EarlySense's novel under-mattress MCPM technology in neonates at Pumwani Maternity Hospital in Nairobi, Kenya. To assess feasibility, we compared the performance of EarlySense's technology to Masimo's Rad-97 pulse CO-oximeter with capnography technology for heart rate (HR) and respiratory rate (RR) measurements using up-time, clinical event detection performance, and accuracy. Between September 15 and December 15, 2020, we collected and analyzed 470 hours of EarlySense data from 109 enrolled neonates. EarlySense's technology's up-time per neonate was 2.9 (range 0.8, 5.3) hours for HR and 2.1 (range 0.9, 4.0) hours for RR. The difference compared to the reference was a median of 0.6 (range 0.1, 3.1) hours for HR and 0.8 (range 0.1, 2.9) hours for RR. EarlySense's technology identified high HR and RR events with high sensitivity (HR 81%; RR 83%) and specificity (HR 99%; RR 83%), but was less sensitive for low HR and RR (HR 0%; RR 14%) although maintained specificity (HR 100%; RR 95%). There was a greater number of false negative and false positive RR events than false negative and false positive HR events. The normalized spread of limits of agreement was 9.6% for HR and 28.6% for RR, which met the a priori-identified limit of 30%. EarlySense's MCPM technology was clinically feasible as demonstrated by high percentage of up-time, strong clinical event detection performance, and agreement of HR and RR measurements compared to the reference technology. Studies in critically ill neonates, assessing barriers and facilitators to adoption, and costing analyses will be key to the technology's development and potential uptake and scale-up.

Variability of respiratory rate measurements in neonates- every minute counts.

BACKGROUND: Respiratory rate is difficult to measure, especially in neonates who have an irregular breathing pattern. The World Health Organisation recommends a one-minute count, but there is limited data to support this length of observation. We sought to evaluate agreement between the respiratory rate (RR) derived from capnography in neonates, over 15 s, 30 s, 120 s and 300 s, against the recommended 60 s. METHODS: Neonates at two hospitals in Nairobi were recruited and had capnograph waveforms recorded using the Masimo Rad 97. A single high quality 5 min epoch was randomly chosen from each subject. For each selected epoch, the mean RR was calculated using a breath-detection algorithm applied to the waveform. The RR in the first 60 s was compared to the mean RR measured over the first 15 s, 30 s, 120 s, full 300 s, and last 60 s. We calculated bias and limits of agreement for each comparison and used Bland-Altman plots for visual comparisons. RESULTS: A total of 306 capnographs were analysed from individual subjects. The subjects had a median gestation age of 39 weeks with slightly more females (52.3%) than males (47.7%). The majority of the population were term neonates (70.1%) with 39 (12.8%) having a primary respiratory pathology. There was poor agreement between all the comparisons based on the limits of agreement [confidence interval], ranging between 11.9 [- 6.79 to 6.23] breaths per minute in the one versus 2 min comparison, and 34.7 [- 17.59 to 20.53] breaths per minute in the first versus last minute comparison. Worsening agreement was observed in plots with higher RRs. CONCLUSIONS: Neonates have high variability of RR, even over a short period of time. A slight degradation in the agreement is noted over periods shorter than 1 min. However, this is smaller than observations done 3 min apart in the same subject. Longer periods of observation also reduce agreement. For device developers, precise synchronization is needed when comparing devices to reduce the impact of RR variation. For clinicians, where possible, continuous or repeated monitoring of neonates would be preferable to one time RR measurements.

A proposed de-identification framework for a cohort of children presenting at a health facility in Uganda.

Data sharing has enormous potential to accelerate and improve the accuracy of research, strengthen collaborations, and restore trust in the clinical research enterprise. Nevertheless, there remains reluctancy to openly share raw data sets, in part due to concerns regarding research participant confidentiality and privacy. Statistical data de-identification is an approach that can be used to preserve privacy and facilitate open data sharing. We have proposed a standardized framework for the de-identification of data generated from cohort studies in children in a low-and-middle income country. We applied a standardized de-identification framework to a data sets comprised of 241 health related variables collected from a cohort of 1750 children with acute infections from Jinja Regional Referral Hospital in Eastern Uganda. Variables were labeled as direct and quasi-identifiers based on conditions of replicability, distinguishability, and knowability with consensus from two independent evaluators. Direct identifiers were removed from the data sets, while a statistical risk-based de-identification approach using the k-anonymity model was applied to quasi-identifiers. Qualitative assessment of the level of privacy invasion associated with data set disclosure was used to determine an acceptable re-identification risk threshold, and corresponding k-anonymity requirement. A de-identification model using generalization, followed by suppression was applied using a logical stepwise approach to achieve k-anonymity. The utility of the de-identified data was demonstrated using a typical clinical regression example. The de-identified data sets was published on the Pediatric Sepsis Data CoLaboratory Dataverse which provides moderated data access. Researchers are faced with many challenges when providing access to clinical data. We provide a standardized de-identification framework that can be adapted and refined based on specific context and risks. This process will be combined with moderated access to foster coordination and collaboration in the clinical research community.

Evaluation of a contactless neonatal physiological monitor in Nairobi, Kenya.

BACKGROUND: Globally, 2.5 million neonates died in 2018, accounting for 46% of under-5 deaths. Multiparameter continuous physiological monitoring (MCPM) of neonates allows for early detection and treatment of life-threatening health problems. However, neonatal monitoring technology is largely unavailable in low-resource settings. METHODS: In four evaluation rounds, we prospectively compared the accuracy of the EarlySense under-mattress device to the Masimo Rad-97 pulse CO-oximeter with capnography reference device for heart rate (HR) and respiratory rate (RR) measurements in neonates in Kenya. EarlySense algorithm optimisations were made between evaluation rounds. In each evaluation round, we compared 200 randomly selected epochs of data using Bland-Altman plots and generated Clarke error grids with zones of 20% to aid in clinical interpretation. RESULTS: Between 9 July 2019 and 8 January 2020, we collected 280 hours of MCPM data from 76 enrolled neonates. At the final evaluation round, the EarlySense MCPM device demonstrated a bias of -0.8 beats/minute for HR and 1.6 breaths/minute for RR, and normalised spread between the 95% upper and lower limits of agreement of 6.2% for HR and 27.3% for RR. Agreement between the two MCPM devices met the a priori-defined threshold of 30%. The Clarke error grids showed that all observations for HR and 197/200 for RR were within a 20% difference. CONCLUSION: Our research indicates that there is acceptable agreement between the EarlySense and Masimo MCPM devices in the context of large within-subject variability; however, further studies establishing cost-effectiveness and clinical effectiveness are needed before large-scale implementation of the EarlySense MCPM device in neonates. TRIAL REGISTRATION NUMBER: NCT03920761.

Identification of thresholds for accuracy comparisons of heart rate and respiratory rate in neonates.

Background: Heart rate (HR) and respiratory rate (RR) can be challenging to measure accurately and reliably in neonates. The introduction of innovative, non-invasive measurement technologies suitable for resource-constrained settings is limited by the lack of appropriate clinical thresholds for accuracy comparison studies. Methods: We collected measurements of photoplethysmography-recorded HR and capnography-recorded exhaled carbon dioxide across multiple 60-second epochs (observations) in enrolled neonates admitted to the neonatal care unit at Aga Khan University Hospital in Nairobi, Kenya. Trained study nurses manually recorded HR, and the study team manually counted individual breaths from capnograms. For comparison, HR and RR also were measured using an automated signal detection algorithm. Clinical measurements were analyzed for repeatability. Results: A total of 297 epochs across 35 neonates were recorded. Manual HR showed a bias of -2.4 (-1.8%) and a spread between the 95% limits of agreement (LOA) of 40.3 (29.6%) compared to the algorithm-derived median HR. Manual RR showed a bias of -3.2 (-6.6%) and a spread between the 95% LOA of 17.9 (37.3%) compared to the algorithm-derived median RR, and a bias of -0.5 (1.1%) and a spread between the 95% LOA of 4.4 (9.1%) compared to the algorithm-derived RR count. Manual HR and RR showed repeatability of 0.6 (interquartile range (IQR) 0.5-0.7), and 0.7 (IQR 0.5-0.8), respectively. Conclusions: Appropriate clinical thresholds should be selected a priori when performing accuracy comparisons for HR and RR. Automated measurement technologies typically use median values rather than counts, which significantly impacts accuracy. A wider spread between the LOA, as much as 30%, should be considered to account for the observed physiological nuances and within- and between-neonate variability and different averaging methods. Wider adoption of thresholds by data standards organizations and technology developers and manufacturers will increase the robustness of clinical comparison studies.

Night to night variability of pulse oximetry features in children at home and at the hospital.

Objective. Investigation of the night-to-night (NtN) variability of pulse oximetry features in children with suspicion of Sleep Apnea.Approach. Following ethics approval and informed consent, 75 children referred to British Columbia Children's Hospital for overnight PSG were recorded on three consecutive nights, including one at the hospital simultaneously with polysomnography and 2 nights at home. During all three nights, a smartphone-based pulse oximeter sensor was used to record overnight pulse oximetry (SpO2 and photoplethysmogram). Features characterizing SpO2 dynamics and heart rate were derived. The NtN variability of these features over the three different nights was investigated using linear mixed models.Main results. Overall most pulse oximetry features (e.g. the oxygen desaturation index) showed no NtN variability. One of the exceptions is for the signal quality, which was significantly lower during at home measurements compared to measurements in the hospital.Significance. At home pulse oximetry screening shows an increasing predictive value to investigate obstructive sleep apnea (OSA) severity. Hospital recordings affect subjects normal sleep and OSA severity and recordings may vary between nights at home. Before establishing the role of home monitoring as a diagnostic test for OSA, we must first determine their NtN variability. Most pulse oximetry features showed no significant NtN variability and could therefore be used in future at-home testing to create a reliable and consistent OSA screening tool. A single night recording at home should be able to characterize pulse oximetry features in children.

Using self-supervised feature learning to improve the use of pulse oximeter signals to predict paediatric hospitalization.

Background: The success of many machine learning applications depends on knowledge about the relationship between the input data and the task of interest (output), hindering the application of machine learning to novel tasks. End-to-end deep learning, which does not require intermediate feature engineering, has been recommended to overcome this challenge but end-to-end deep learning models require large labelled training data sets often unavailable in many medical applications. In this study, we trained self-supervised learning (SSL) models for automatic feature extraction from raw photoplethysmography (PPG) obtained using a pulse oximeter, with the aim of predicting paediatric hospitalization.  Methods: We compared logistic regression models fitted using features extracted using SSL with models trained using both clinical and SSL features. In addition, we compared end-to-end deep learning models initialized randomly or using weights from the SSL models. We also compared the performance of SSL models trained on labelled data alone (n=1,031) with SSL trained using both labelled and unlabelled signals (n=7,578). Results: Logistic regression models were more predictive of hospitalization when trained on features extracted using labelled PPG signals only compared to SSL models trained on both labelled and unlabelled signals (AUC 0.83 vs 0.80). However, features extracted using SSL model trained on both labelled and unlabelled PPG signals were more predictive of hospitalization when concatenated with clinical features (AUC 0.89 vs 0.87). The end-to-end deep learning model had an AUC of 0.80 when initialized using the SSL model trained on all PPG signals, 0.77 when initialized using SSL trained on labelled data only, and 0.73 when initialized randomly. Conclusions: This study shows that SSL can extract features from PPG signals that are predictive of hospitalization or initialize end-to-end deep learning models. Furthermore, SSL can leverage larger unlabelled data sets to improve performance of models fitted using small labelled data sets.

Evaluation of a digital triage platform in Uganda: A quality improvement initiative to reduce the time to antibiotic administration.

BACKGROUND: Sepsis is the leading cause of death in children under five in low- and middle-income countries. The rapid identification of the sickest children and timely antibiotic administration may improve outcomes. We developed and implemented a digital triage platform to rapidly identify critically ill children to facilitate timely intravenous antibiotic administration. OBJECTIVE: This quality improvement initiative sought to reduce the time to antibiotic administration at a dedicated children's hospital outpatient department in Mbarara, Uganda. INTERVENTION AND STUDY DESIGN: The digital platform consisted of a mobile application that collects clinical signs, symptoms, and vital signs to prioritize children through a combination of emergency triggers and predictive risk algorithms. A computer-based dashboard enabled the prioritization of children by displaying an overview of all children and their triage categories. We evaluated the impact of the digital triage platform over an 11-week pre-implementation phase and an 11-week post-implementation phase. The time from the end of triage to antibiotic administration was compared to evaluate the quality improvement initiative. RESULTS: There was a difference of -11 minutes (95% CI, -16.0 to -6.0; p < 0.001; Mann-Whitney U test) in time to antibiotics, from 51 minutes (IQR, 27.0-94.0) pre-implementation to 44 minutes (IQR, 19.0-74.0) post-implementation. Children prioritized as emergency received the greatest time benefit (-34 minutes; 95% CI, -9.0 to -58.0; p < 0.001; Mann-Whitney U test). The proportion of children who waited more than an hour until antibiotics decreased by 21.4% (p = 0.007). CONCLUSION: A data-driven patient prioritization and continuous feedback for healthcare workers enabled by a digital triage platform led to expedited antibiotic therapy for critically ill children with sepsis. This platform may have a more significant impact in facilities without existing triage processes and prioritization of treatments, as is commonly encountered in low resource settings.

Community-level interventions for pre-eclampsia (CLIP) in Pakistan: A cluster randomised controlled trial.

OBJECTIVES: To reduce all-cause maternal and perinatal mortality and major morbidity through Lady Health Worker (LHW)-facilitated community engagement and early diagnosis, stabilization and referral of women with preeclampsia, an important contributor to adverse maternal and perinatal outcomes given delays in early detection and initial management. STUDY DESIGN: In the Pakistan Community-Level Interventions for Pre-eclampsia (CLIP) cluster randomized controlled trial (NCT01911494), LHWs engaged the community, recruited pregnant women from 20 union councils (clusters), undertook mobile health-guided clinical assessment for preeclampsia, and referral to facilities after stabilization. MAIN OUTCOME MEASURES: The primary outcome was a composite of maternal, fetal and newborn mortality and major morbidity. FINDINGS: We recruited 39,446 women in intervention (N = 20,264) and control clusters (N = 19,182) with minimal loss to follow-up (3∙7% vs. 4∙5%, respectively). The primary outcome did not differ between intervention (26·6%) and control (21·9%) clusters (adjusted odds ratio, aOR, 1∙20 [95% confidence interval 0∙84-1∙72]; p = 0∙31). There was reduction in stillbirths (0·89 [0·81-0·99]; p = 0·03), but no impact on maternal death (1·08 [0·69, 1·71]; p = 0·74) or morbidity (1·12 [0·57, 2·16]; p = 0·77); early (0·95 [0·82-1·09]; p = 0·46) or late neonatal deaths (1·23 [0·97-1·55]; p = 0·09); or neonatal morbidity (1·22 [0·77, 1·96]; p = 0·40). Improvements in outcome rates were observed with 4-7 (p = 0·015) and ≥8 (p < 0·001) (vs. 0) CLIP contacts. INTERPRETATION: The CLIP intervention was well accepted by the community and implemented by LHWs. Lack of effects on adverse outcomes could relate to quality care for mothers with pre-eclampsia in health facilities. Future strategies for community outreach must also be accompanied by health facility strengthening. FUNDING: The University of British Columbia (PRE-EMPT), a grantee of the Bill & Melinda Gates Foundation (OPP1017337).