Development and validation of a diagnostic aid for convulsive epilepsy in sub-Saharan Africa: a retrospective case-control study

Jones, Gabriel Davis; Mwesige, Angelina Kakooza; Cross, J. Helen; Newton, Charles R.; Abban, Hanna; Adjei, Patrick; Ae-Ngibise, Ken; Agbokey, Francis; Aissaoui, Lisa; Akpalu, Albert; Akpalu, Bright; Asiamah, Sabina; Asiki, Gershim; Atieno, Mercy; Bauni, Evasius; Bhwana, Dan; Bitta, Mary A.; Bottomley, Christian; Chabi, Martin; Chengo, Eddie; Chowdhary, Neerja; Connor, Myles; Cross, Helen; Collinson, Mark; Darkwa, Emmanuel; Denison, Timothy; Doku, V.; Dua, Tarun; Egesa, Isaac J.; Godi, Tony; Gómez‐Olivé, F. Xavier; Grassi, Simone; Iddi, Samuel; Abankwah, Daniel Nana Yaw; Kahn, Kathleen; Kakooza, Angelina; Kariuki, Symon M.; Kamuyu, Gathoni; Khalayi, Clarah; Kimambo, Henrika; Kleinschmidt, Immo; Kwasa, Thomas; Mahone, Sloan; Manolova, Gergana; Masanja, Honorati; Alexander, Mathew; Matuja, William; McDaid, David; Mmbando, Bruno P.; Mwanga, Daniel Mtai; Muli, Dorcas; Odera, Victor Mung'ala; Wekesah, Frederick; Mushi, Vivian; Ngugi, Anthony; Odermatt, Peter; Mageto, James; Otieno, Peter; Owusu‐Agyei, Seth; Pariyo, George; Peterson, Stefan; Sander, Josemir W.; Sen, Arjune; Sottie, Cynthia Akorfa; Sylvester, Isolide; Tollman, Stephen; Thoya, Yvonne; Twine, Rhian; Vallentin, Sonia; Wagner, Ryan G.; Walker, Richard; Waruingi, Stella

doi:10.1016/s2589-7500(22)00255-2

Cited by 9 publications

(6 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Information on sociodemographic variables, historical risk factors and clinical history was collected for each participant. The resulting dataset comprised sociodemographic information and approximately 170 unique variables for each individual in five domains: clinical history, clinical examination, seizure description and EEG interpretation (6).…”

Section: Data Acquisition Study Design and Pre-processingmentioning

confidence: 99%

“…We used predictors of epilepsy formatted as questions for the person with suspected epilepsy. These predictors have been reported and validated in previous studies from this cohort and were chosen for their maximally discriminative predictive ability to diagnose ACE (6). The predictors were:…”

Section: Data Pre-processingmentioning

confidence: 99%

“…Clinical machine learning (ML) models therefore offer a practical solution for epilepsy diagnosis in low-resource settings. Such models have demonstrated promising outcomes for epilepsy diagnosis (6) and treatment (7). Thus, their application in LMICs could reduce the diagnostic and treatment gaps (8).…”

Section: Introductionmentioning

confidence: 99%

“…We have previously developed a predictive model to support epilepsy diagnosis in LMICs (6), trained on data from five sub-Saharan African regions. This study aims to describe the necessity of comprehensive cross-regional validation.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Evaluating the generalisability of region-naïve machine learning algorithms for the identification of epilepsy in low-resource settings

Duta,

Kariuki,

Ngugi

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

ObjectivesApproximately 80% of people with epilepsy live in low- and middle-income countries (LMICs), where limited resources and stigma hinder accurate diagnosis and treatment. Clinical machine learning models have demonstrated substantial promise in supporting the diagnostic process in LMICs without relying on specialised or trained personnel. How well these models generalise to naïve regions is, however, underexplored. Here, we use a novel approach to assess the suitability and applicability of such clinical tools for diagnosing active convulsive epilepsy in settings beyond their original training contexts.MethodsWe sourced data from the Study of Epidemiology of Epilepsy in Demographic Sites dataset, which includes demographic information and clinical variables related to diagnosing epilepsy across five sub-Saharan African sites. For each site, we developed a region-specific (single-site) predictive model for epilepsy and evaluated its performance on other sites. We then iteratively added sites to a multi-site model and evaluated its performance on the omitted regions. Model performances and parameters were then compared across every permutation of sites. We used a leave-one-site-out cross-validation analysis to assess the impact of incorporating individual site data in the model.ResultsSingle-site clinical models performed well within their own regions, but worse in general when evaluated on other regions (p<0.05). Model weights and optimal thresholds varied markedly across sites. When the models were trained using data from an increasing number of sites, mean internal performance decreased while external performance improved.ConclusionsClinical models for epilepsy diagnosis in LMICs demonstrate characteristic traits of ML models, such as limited generalisability and a trade-off between internal and external performance. The relationship between predictors and model outcomes also varies across sites, suggesting the need to update specific aspects of the model with local data before broader implementation. Variations are likely to be specific to the cultural context of diagnosis. We recommend developing models adapted to the cultures and contexts of their intended deployment and caution against deploying region- and culture-naïve models without thorough prior evaluation.Key pointsMachine learning-driven clinical tools are becoming more prevalent in low-resource settings; however, their general performance across regions is not fully established. Given their potential impact, it is crucial models are robust, safe and appropriately deployedModels perform poorly when making predictions for regions that were not included in their training data, as opposed to sites that wereModels trained on different regions can have different optimal parameters and thresholds for performance in practiceThere is a trade-off between internal and external performance, where a model with better external performance usually has worse internal performance but is generally more robust overallSEEDS collaboratorsAgincourt HDSS, South Africa: Ryan Wagner, Rhian Twine, Myles Connor, F. Xavier Gómez-Olivé, Mark Collinson (and INDEPTH Network, Accra, Ghana), Kathleen Kahn (and INDEPTH Network, Accra, Ghana), Stephen Tollman (and INDEPTH Network, Accra, Ghana)Ifakara HDSS, Tanzania: Honratio Masanja (and INDEPTH Network, Accra, Ghana), Alexander MathewIganga/Mayuge HDSS, Uganda: Angelina Kakooza, George Pariyo, Stefan Peterson (and Uppsala University, Dept of Women’s and Children’s Health, IMCH; Karolinska Institutet, Div. of Global Health, IHCAR; Makerere University School of Public Health), Donald NdyomughenyiKilifi HDSS, Kenya: Anthony K Ngugi, Rachael Odhiambo, Eddie Chengo, Martin Chabi, Evasius Bauni, Gathoni Kamuyu, Victor Mung’ala Odera, James O Mageto, Isaac Egesa, Clarah Khalayi, Charles R NewtonKintampo HDSS, Ghana: Ken Ae-Ngibise, Bright Akpalu, Albert Akpalu, Francic Agbokey, Patrick Adjei, Seth Owusu-Agyei, Victor Duko (and INDEPTH Network, Accra, Ghana)London School of Hygiene and Tropical Medicine: Christian Bottomley, Immo KleinschmidtInstitute of Psychiatry, King’s College London: Victor CK DokuUCL Queen Square Institute of Neurology, London: Josemir W SanderSwiss Tropical Institute: Peter Odermatt

show abstract

Section: Data Acquisition Study Design and Pre-processingmentioning

confidence: 99%

Section: Data Pre-processingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Evaluating the generalisability of region-naïve machine learning algorithms for the identification of epilepsy in low-resource settings

Duta,

Kariuki,

Ngugi

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Ensemble methods include Gradient Boosting Machines (GBM) [42], Random Forest (RF) [43], Light Gradient Boosting Machine (LightGBM) [44], Adaptive Boosting (AdaBoost) [45], Extreme Gradient Boosting (XGBoost) [46], Categorical Boosting (CatBoost) [47], and Histogram-based Gradient Boosting (HistGB) [48]. Linear models include Logistic Regression (LR) [49], Multinomial Logistic Regression (MLR) [50], Stochastic Gradient Descent (SGD) [51], and Linear Discriminant Analysis (LDA) [52]. Bayesian methods include Gaussian Naive Bayes (GNB) [53], Multinomial Naive Bayes (MNB) [54], Complement Naive Bayes (CNB) [55], and Quadratic Discriminant Analysis (QDA) [56].…”

Section: Building Machine Learning Prediction Models Using Screened C...mentioning

confidence: 99%

Optimizing a Interpretable Diagnostic Model for Colorectal Cancer Based on Yin Deficiency Pattern Characteristic Genes Using 21 Machine Learning Algorithms and Bayesian Opitimization

Li,

Huang,

Liu

et al. 2023

Preprint

View full text Add to dashboard Cite

In this study, we aimed to identify a core set of Yin Deficiency Pattern (YDP) genes for predicting colorectal cancer (CRC) and to construct reliable machine learning models optimized by Optuna. Comprehensive analysis was performed on nine datasets, totaling 1,680 samples. A CRC diagnostic prediction model was developed by comparing 21 machine learning classification models with Optuna hyperparameter optimization and validated across six independent external cohorts. Additionally, the expression patterns of the core diagnostic genes were experimentally validated. Linear Discriminant Analysis (LDA), along with four other machine learning models (please specify these models), ranked as the top five performing models across six cohorts, demonstrating superior performance with AUC values exceeding 0.99 and all other performance metrics surpassing 0.899. This study marks the first utilization of four specific novel machine learning models (again, please specify these models) in CRC diagnosis. The robust performance of the top models across multiple external validation sets underscores the reliability and generalizability of our diagnostic model. These results hold potential implications for the development of personalized medicine approaches in CRC treatment, offering a new avenue for early detection and prognosis improvement.

show abstract

An epilepsy type algorithm developed in India is accurate in Sudan: A prospective validation study

El-Sadig,

El-Amin,

Mohamed

et al. 2023

Seizure: European Journal of Epilepsy

View full text Add to dashboard Cite

Development and validation of a diagnostic aid for convulsive epilepsy in sub-Saharan Africa: a retrospective case-control study

Cited by 9 publications

References 25 publications

Evaluating the generalisability of region-naïve machine learning algorithms for the identification of epilepsy in low-resource settings

Evaluating the generalisability of region-naïve machine learning algorithms for the identification of epilepsy in low-resource settings

Optimizing a Interpretable Diagnostic Model for Colorectal Cancer Based on Yin Deficiency Pattern Characteristic Genes Using 21 Machine Learning Algorithms and Bayesian Opitimization

An epilepsy type algorithm developed in India is accurate in Sudan: A prospective validation study

Contact Info

Product

Resources

About