Development of machine learning models predicting mortality using routinely collected observational health data from 0-59 months old children admitted to an intensive care unit in Bangladesh: critical role of biochemistry and haematology data.

INTRODUCTION: Treatment in the intensive care unit (ICU) generates complex data where machine learning (ML) modelling could be beneficial. Using routine hospital data, we evaluated the ability of multiple ML models to predict inpatient mortality in a paediatric population in a low/middle-income country. METHOD: We retrospectively analysed hospital record data from 0-59 months old children admitted to the ICU of Dhaka hospital of International Centre for Diarrhoeal Disease Research, Bangladesh. Five commonly used ML models- logistic regression, least absolute shrinkage and selection operator, elastic net, gradient boosting trees (GBT) and random forest (RF), were evaluated using the area under the receiver operating characteristic curve (AUROC). Top predictors were selected using RF mean decrease Gini scores as the feature importance values. RESULTS: Data from 5669 children was used and was reduced to 3505 patients (10% death, 90% survived) following missing data removal. The mean patient age was 10.8 months (SD=10.5). The top performing models based on the validation performance measured by mean 10-fold cross-validation AUROC on the training data set were RF and GBT. Hyperparameters were selected using cross-validation and then tested in an unseen test set. The models developed used demographic, anthropometric, clinical, biochemistry and haematological data for mortality prediction. We found RF consistently outperformed GBT and predicted the mortality with AUROC of ≥0.87 in the test set when three or more laboratory measurements were included. However, after the inclusion of a fourth laboratory measurement, very minor predictive gains (AUROC 0.87 vs 0.88) resulted. The best predictors were the biochemistry and haematological measurements, with the top predictors being total CO2, potassium, creatinine and total calcium. CONCLUSIONS: Mortality in children admitted to ICU can be predicted with high accuracy using RF ML models in a real-life data set using multiple laboratory measurements with the most important features primarily coming from patient biochemistry and haematology.

Dermatological changes in a prospective cohort of acutely ill, hospitalised Malawian children, stratified according to nutritional status.

RATIONALE: Since the first documentation of skin changes in malnutrition in the early 18th century, various hair and skin changes have been reported in severely malnourished children globally. We aimed to describe the frequency and types of skin conditions in children admitted with acute illness to Queen Elizabeth Central Hospital, Blantyre, Malawi across a spectrum of nutritional status and validate an existing skin assessment tool. METHODS: Children between 1 week and 23 months of age with acute illness were enrolled and stratified by anthropometry. Standardised photographs were taken, and three dermatologists assessed skin changes and scored each child according to the SCORDoK tool. RESULTS: Among 103 children, median age of 12 months, 31 (30%) had severe wasting, 11 (11%) kwashiorkor (nutritional oedema), 20 (19%) had moderate wasting, 41 (40%) had no nutritional wasting and 18 (17%) a positive HIV antibody test. Six (5.8%) of the included patients died. 51 (50%) of children presented with at least one skin change. Pigmentary changes were the most common, observed in 35 (34%), with hair loss and bullae, erosions and desquamation the second most prevalent skin condition. Common diagnoses were congenital dermal melanocytosis, diaper dermatitis, eczema and postinflammatory hyperpigmentation. Severe skin changes like flaky paint dermatosis were rarely identified. Inter-rater variability calculations showed only fair agreement (overall Fleiss' kappa 0.25) while intrarater variability had a fair-moderate agreement (Cohen's kappa score of 0.47-0.58). DISCUSSION: Skin changes in hospitalised children with an acute illness and stratified according to nutritional status were not as prevalent as historically reported. Dermatological assessment by means of the SKORDoK tool using photographs is less reliable than expected.

Childhood growth during recovery from acute illness in Africa and South Asia: a secondary analysis of the childhood acute illness and nutrition (CHAIN) prospective cohort.

Background: Growth faltering is well-recognized during acute childhood illness and growth acceleration during convalescence, with or without nutritional therapy, may occur. However, there are limited recent data on growth after hospitalization in low- and middle-income countries. Methods: We evaluated growth following hospitalization among children aged 2-23 months in sub-Saharan Africa and South Asia. Between November 2016 and January 2019, children were recruited at hospital admission and classified as: not-wasted (NW), moderately-wasted (MW), severely-wasted (SW), or having nutritional oedema (NO). We describe earlier (discharge to 45-days) and later (45- to 180-days) changes in length-for-age [LAZ], weight-for-age [WAZ], mid-upper arm circumference [MUACZ], weight-for-length [WLZ] z-scores, and clinical, nutritional, and socioeconomic correlates. Findings: We included 2472 children who survived to 180-days post-discharge: NW, 960 (39%); MW, 572 (23%); SW, 682 (28%); and NO, 258 (10%). During 180-days, LAZ decreased in NW (-0.27 [-0.36, -0.19]) and MW (-0.23 [-0.34, -0.11]). However, all groups increased WAZ (NW, 0.21 [95% CI: 0.11, 0.32]; MW, 0.57 [0.44, 0.71]; SW, 1.0 [0.88, 1.1] and NO, 1.3 [1.1, 1.5]) with greatest gains in the first 45-days. Of children underweight (<-2 WAZ) at discharge, 66% remained underweight at 180-days. Lower WAZ post-discharge was associated with age-inappropriate nutrition, adverse caregiver characteristics, small size at birth, severe or moderate anaemia, and chronic conditions, while lower LAZ was additionally associated with household-level exposures but not with chronic medical conditions. Interpretation: Underweight and poor linear growth mostly persisted after an acute illness. Beyond short-term nutritional supplementation, improving linear growth post-discharge may require broader individual and family support. Funding: Bill & Melinda Gates FoundationOPP1131320; National Institute for Health ResearchNIHR201813.

Stunting in the first year of life: Pathway analysis of a birth cohort.

Malnutrition among infants aged below 6 months has been largely overlooked creating gaps in our understanding of factors underlying stunting in early infancy. Recent evidence suggests that pre-natal and early childhood factors may contribute more to driving childhood stunting than previously appreciated. The study was set up to examine pathways including parental and household characteristics, birth size and gestation, and illness in infancy with stunting at birth and months 3, 6 and 12 using an a priori hypothesized framework. It was a secondary analysis of a birth cohort of 1017 infants recruited from four health facilities in Burkina Faso and followed up for one year. Structural equation models (SEM) were generated to explore pathways to stunting at birth and months 3, 6 and 12. The prevalence of being stunted at birth and months 3, 6 and 12 was 7.4%, 23%, 20% and 18% respectively. The fractions of month 12 stunting attributable to being stunted at birth, months 3 and 6 were 11% (95%CI 5.0‒16%), 32% (95%CI 22‒41%) and 40% (95%CI 31‒49%) respectively. In the structural equation model, male sex and maternal characteristics had direct effects on stunting at birth and at 3 months, but not subsequently. Premature birth, twin birth and being stunted at a previous time point were directly associated with stunting at months 3, 6 and 12. Both maternal and paternal characteristics were directly associated with preterm birth. Non-exclusive breastfeeding had borderline positive direct effect on stunting at month 6 but not at month 12. The direct and indirect pathways identified in this study highlight the complex interlinks between child, maternal, paternal and household characteristics. Interventions tackling preterm birth, in utero growth, exclusive breastfeeding and maternal wellbeing may reduce stunting in the first year of life.