Team Healing Code

Team Members

Nkalubo Jonathan - Medical Doctor, Accident and Emergency Department, Mulago National Referral Hospital, Uganda

Buhle Siphelele Makongwana - Paediatric Infectious Specialist, Nelson Mandela Academic Hospital, South Africa

Vidhubala Balasegar - Clinical Scientist (Microbiology), Department of Pathology, Segamat General Hospital, Johor, Malaysia 

Ouissal Aissaoui, MD - Professor of Pediatric Anesthesiology and Intensive Care, Ibn Rochd University Hospital, Morocco

About Your Team

The team is a multinational, multidisciplinary team from Uganda, South Africa, Malaysia and Morocco. Members work in accident and emergency medicine, paediatric infectious diseases, clinical science and paediatric anaesthesia.

Summary of Approach

The aim was to create a simple, lightweight model to predict in-hospital mortality from sepsis. Our approach tried to provide a simple yet practical tool to predict in-hospital mortality. 

Variables which did not have too many missing values were prioritised as this showed ease of capturing in the setting. Further narrowing of the selection was aided by correlations on heatmaps.

Variables that made it into the final model included heart rate on admission, mid-upper arm circumference on admission (which was transformed to severe wasting, moderate wasting and normal), height on admission (reflecting stunting presence or absence), oxygen saturation on admission (hypoxia), temperature on admission (fever, hypothermia, normal temperature) and Blantyre score (compiled by adding-Blantyre eyes score, Blantyre motor score and Blantyre verbal scores). Google Colab and Python were used to generate the model and finally exported to GitHub.

Data Preprocessing

Categorical values were transformed with a label encoder, and missing variables with a simple imputer (median). Oversampling with imblearn was used and the final model used a random forest classifier (split into 80% training and 20% testing datasets).