A collection of proven, probable, possible and control cases of Fungal Disease to develop AI Fungal Disease algorithms – Global Action for Fungal Infections (GAFFI)

Lead applicant

Professor Juan Luis Tudela – Global Action for Fungal Infections (GAFFI)

Co-applicants

Ana Alastruey-Izquierdo – National Centre for Microbiology, Instituto de Salud Carlos III

Guillermo Garcia – The Litoral University, Argentina

Claudia E. Frola – Hospital Juan A. Fernández, Buenos Aires

Geographical focus – Global, but especially LMICs

Research Theme – Innovative platforms

Lay summary

Diagnosing and treating severe fungal diseases is complex. Most of these infections occur in patients with underlying conditions, and since many fungi are opportunistic, an isolated diagnostic test does not always confirm infection. Therefore, a sophisticated classification system has been developed to categorize cases as proven, probable, and possible.For AI to fulfill its promise of improving global health, at least three key challenges need to be addressed. The first is the reliability and availability of data. AI systems must be trained using large volumes of data, and the quality of the output reflects the quality of the input. Although the incidence and prevalence of Fungal Disease is increasing, we lack well-curated, high-fidelity clinical datasets. Passive surveillance will not yield sufficient data as it has for other diseases. Consequently, it is uncertain whether the current information that AI algorithms are using for healthcare is suitable for Fungal Disease. For other diseases, the amount of data has expanded substantially in recent years, but it is still mostly from high-income settings and has largely not been tested or validated in low-income settings. In Fungal Disease, the data that AI health systems are using are often of unknown sources, and some publications have shown its unreliability for histoplasmosis. Using such data can create biases in the AI system’s training and, consequently, its responses.The Fungal Disease community needs to engage in active case finding for proven, probable, and possible cases, as well as for controls, to train and validate healthcare algorithms. This will ensure that Fungal Disease does not remain neglected in the realm of AI. Although this will be beneficial globally, it will dramatically improve the situation in LMICs, where the shortage of clinicians trained in Fungal Disease is evident. Additionally, worldwide data is necessary to attain robust training and validation of healthcare AI algorithms.While these clinical datasets have various applications, one significant use will be to understand the relationship between in vitro susceptibility/resistance classification of isolates and the clinical outcomes of Fungal Disease. Although evidence suggests that treating an infection with an antifungal to which the strain is resistant results in worse outcomes compared to a susceptible isolate, many other variables can influence these outcomes. These variables include patient comorbidities, immune status, the PK/PD of the antifungal used, and the presence of different risk factors. By analyzing comprehensive clinical data sets, researchers can control for these variables and better isolate the impact of antifungal resistance on patient outcomes. Demonstrating a cause-effect relationship, rather than merely an association, would be invaluable for patients with severe Fungal Disease, leading to more effective treatment protocols and better patient management. This would ultimately contribute to personalized medicine approaches, optimizing treatment based on individual patient and pathogen characteristics.