There is a growing awareness of the broader HIV epidemic’s enormous mental health burden, particularly in sub-Saharan Africa. Studies have shown that 5 – 20% of people living with HIV (PLHIV) have mental health problems, with the highest rate occurring in adolescence. A World Health Organization 2022 report found that depression and anxiety were among the main causes of illness and disability among adolescents, and suicide was the fourth leading cause of death among people aged 15–29 years. Rates of self harm and suicidal ideation are even higher among adolescents living with HIV. Mental health considerations are crucial in understanding both pathways to risk, and interventions that can help improve adolescent health outcomes. Significant improvements in access to antiretroviral therapy (ART) among women of child-bearing age has substantially reduced vertical HIV transmission to children. However, rates of perinatal depression among mothers living with HIV in South Africa have been shown to be between 20 – 50%. Evidence for adverse effects on child development from both HIV and maternal depression, including delayed development, behavioural, and emotional problems, raises a major public health concern.

In sub-Saharan Africa, the situation is made more complex by the relative lack of resources for mental health support, and the need for locally validated, culturally relevant screening tools to identify mental health problems. Since 2000, the Africa Health Research Institute has operated a HDSS platform in a population of 90,000 (10,000 households) in rural KwaZulu Natal, South Africa. A number of different research studies on maternal and adolescent mental health have enrolled participants from the surveillance area. This project aims to bring together these different data sources, to create a common data resource that will be accessible to other mental health researchers in Africa. 

The overall objective of this project is to create an integrated data resource for research on adolescent mental health that will be accessible to the wider scientific community. This thematic data resource will leverage existing datasets to tackle complex research questions, apply a range of analytical approaches, and identify new areas of scientific discovery.  

Our specific aims are: 

• To harmonise the data from the Africa Health Research Institute’s (AHRI) existing studies of mental health
• To merge the harmonised datasets with data from AHRI’s health and demographic surveillance system (HDSS) and clinical platform to provide information on long-term outcomes among study participants
• To link the harmonised datasets with routine service delivery data sources to which AHRI has access
• To make the data available on UZIMA-DS and AHRI’s data sharing platform, promote access to longitudinal population data on adolescent mental health

The current diagnostic approaches for psychosis often rely on subjective assessments and single data types, which can result in misclassification and suboptimal treatment outcomes. The integration of multimodal data and machine learning (ML) techniques holds immense potential for achieving highly accurate and comprehensive identification of psychosis cases. Despite the promising potential improving diagnostic accuracy using machine learning, current literature has pinpointed substantial limitations including data integration challenges, small sample sizes, lack of diversity in the data, and lack of clinical utility and relevance of models. This application seeks to address these gaps by; i) combining data types, ii) including multimodal data from clinical assessments, patient history, and genetic data from a varied dataset, iii) utilizing unsupervised and supervised machine learning techniques to refine our understanding and predictive capabilities of psychosis subtypes, and thereby fostering the practical clinical implications of machine learning models in psychosis diagnosis and treatment. 

Aims:

• A combination of clinical, demographics, and genetics data within the NeuroGAP-Psychosis study, which is a multisite case-control study comprised of 42,943 participants recruited from within Ethiopia, Kenya, South Africa, and Uganda will be fed into choice supervised ML algorithms as feature inputs (input data) to identify and categorize psychosis.  
• The exploration of the associations between key clinical and genetic features and the outcome of psychosis using standard statistical and unsupervised ML methods. This aim will also provide an opportunity to investigate temporal patterns and relationships within the data, specifically related to changes in clinical features and by considering trajectories in hospitalizations. 
• Assess the performance and generalizability of ML models for predicting psychosis and demonstrates a commitment to ensuring that the developed models can be applied beyond the immediate research context. This will be achieved through utilizing the varied dataset that will enable us to train and validate our models on a representative sample as well as testing the performance of our models on independent datasets from different clinical and research settings to assess their robustness and ability to generalize.

Every year, it is projected that approximately 2% to 10% of the pregnant mothers suffer from gestational diabetes mellitus (GDM) globally. In Africa, the average prevalence of GDM is 13.6%, though slightly higher in the sub-Saharan regions. Women with GDM are more likely to suffer from complications such as prolonged labor, development of type 2 diabetes, operative delivery and pre-eclampsia while the infants are at risk of macrosomia as a result of the accelerated fetal growth due to neonatal hyperglycemia, preterm delivery and maternal hyperglycemia.

While there exist several studies that attempted to use ML algorithms for early prediction of GDM, it is worth noting that most of the studies were done in China with very few studies using data collected in other low and middle-income countries and no studies available on data collected in Kenya. The primary objective of this research is to create precise and resilient ML algorithms that can promptly predict Gestational Diabetes Mellitus (GDM) in Kenya. By achieving this, we aim to enhance maternal and fetal health outcomes through timely interventions. Additionally, the research findings will contribute to knowledge of GDM risk factors, potentially leading to further research and advancements in prenatal care.

Aims:

• To summarize GDM risk factors at first and second trimester of pregnancy via descriptive statistics for patients enrolled at Aga Khan University, Kenya for antenatal visits and delivery.
• To develop a robust ML algorithm for early prediction of GDM in the Kenyan setting.
• To validate the performance of the developed machine learning algorithm for early prediction of gestational diabetes mellitus.