Grant Mackenzie, Gambia
Medical Research Council Unit The Gambia at the London School of Hygiene & Tropical Medicine Disease Control and EliminationPoster Author Of 1 e-Poster
DEVELOPMENT AND VALIDATION OF A MACHINE LEARNING PREDICTION MODEL FOR PNEUMONIA MORTALITY IN CHILDREN AGED UNDER FIVE YEARS IN RURAL GAMBIA
Author Of 5 Presentations
FOLLOWING A DECADE OF PCV IN THE GAMBIA SHOULD A DECLINE IN RESISTANCE BE ANTICIPATED? (ID 1205)
- Muhammed Arafat Cham, Gambia
- Brenda Kwambana-Adams, United Kingdom
- Madikay Senghore, United States of America
- Effua Usuf, Gambia
- Archibald Worwui,
- Rasheed Salaudeen, Gambia
- Lesley McGee, United States of America
- Stephen D. Bentley, United Kingdom
- Robert F. Breiman, United States of America
- Anna Roca, Gambia
- Grant Mackenzie, Gambia
- Martin Antonio, Gambia
DEVELOPMENT AND VALIDATION OF A MACHINE LEARNING PREDICTION MODEL FOR PNEUMONIA MORTALITY IN CHILDREN AGED UNDER FIVE YEARS IN RURAL GAMBIA (ID 126)
Abstract
Background
Pneumonia accounts for many deaths in children aged under 5 years in developing countries. A reliable and generalizable tool to predict mortality and thus assess the severity of pneumonia would aid patient management.
Methods
We used a dataset of 11,012 children admitted with clinical pneumonia to develop a model to predict mortality. Using a High Performance Computing platform, we generated multiple models for all possible feature combinations, applying support vector machine, neural networks, random forests and logistic regression to 2/3 of the dataset with repeated cross-validation (5 repetitions, 10 folds). We chose the final model based on its performance and on the number of and measurement reliability of the included features to increase generalizability. In the validation stage, we applied the selected model to the held-out dataset to test its performance on unseen cases.
Results
Not only did the selected model have good sensitivity and specificity (both >80%) on the training set, but more importantly, it had promising performance when applied to the test set.
Conclusions
Our predictive model performed well not only in cross-validated data, but also in our test dataset, increasing our confidence in its generalizability.
INSIGHTS INTO PNEUMOCOCCAL PNEUMONIA USING LUNG ASPIRATES AND NASOPHARYNGEAL SWABS COLLECTED FROM PNEUMONIA PATIENTS IN THE GAMBIA (ID 668)
NASOPHARYNGEAL PNEUMOCOCCAL SEROTYPE DETECTION COMBINING LATEX AGGLUTINATION SWEEP SEROTYPING WITH QUELLUNG IN THE GAMBIA. (ID 714)
Abstract
Background
Detection of multiple pneumococcal serotypes in nasopharyngeal swabs (NPS) by latex agglutination, either by the WHO protocol or sweep serotyping may yield incomplete serum factor reactions. We investigated the impact of combining latex sweep serotyping with Quellung to improve the specificity of reactions while detecting multiple pneumococcal serotypes.
Methods
We randomly selected 103 STGG-NPS samples cultured and serotyped by latex agglutination of different morphological colonies (WHO protocol) in two cross-sectional carriage studies. We performed Quellung capsular serotyping on 1, 2 or 3 morphologically distinct pneumococcal colonies from the culture plates and then swept the same plates.
Results
Sweep serotyping was more likely to detect multiple serotypes (Table 1) than the WHO protocol (p=0.004). A common serotype was identified in 95/103 (92.2%) NPS samples by the WHO, latex sweep and Quellung. Quellung gave more conclusive antisera reactions (Table 2) than the WHO protocol (p=0.007). No significant difference in conclusiveness of antisera reaction between Quellung and latex sweep (p=0.119).
Conclusions
Combining sweep serotyping and Quellung is a cost-effective way of detecting multiple pneumococcal serotype co-colonization and resolving issues related to antisera reactivity in latex agglutination. Inconclusive Quellung results occurred because we detected more serotypes by latex sweep than were selected according to different morphological colonies.
INNOVATIVE DATA MANAGEMENT IN THE PNEUMOCOCCAL VACCINE SCHEDULES IN RURAL GAMBIA (ID 726)
Abstract
Background
Research data management for field studies in rural Africa poses unique challenges. Data management in the Pneumococcal Vaccine Schedules (PVS) trial in The Gambia uses the following innovative procedures to maintain data integrity and validity with large numbers of observations where there is no electricity or internet.
Methods
•Confirming residential status and identity of infants in the study area in a continuously updated live sampling frame of over 10,000 births per year.
•User prompts and decision rules for group allocation for over 10,000 births per year and correct vaccine administration according to village of residence for over 50,000 PCV doses per year.
•Weekly synchronisation of data to a central server (up to 400 trial enrolments, 2500 PCV doses and 800 clinical presentations per week monitoring residence of 300,000 population).
•Weekly update of a back-end relational database combining variables from five different applications.
•Linkage of demographic identity details to patients presenting to health facilities using family geneology and photographs.
•Remote data viewing for off-site external trial monitoring, query and resolution audit trail, as well as trial staff real-time viewing of data.
Results
Interactive experience of the data systems will be available at the conference.
Conclusions
Those systems are working properly for PVS data collection.