Democratising Early Diabetes Detection: A Human-Integrated AI Approach to Overcome Representation Bias in the Global South

Introduction:

Indonesia faces critical metabolic health challenges, with an estimated 73% of diabetes cases undiagnosed. Traditional diagnosis systems that rely on invasive and costly blood tests are structurally difficult to scale across the archipelago. Many digital screening tools like photoplethysmography (PPG) exist but suffer from ethnophenotypic bias across diverse skin tones, creating a massive diagnostic gap and barrier for early detection. We present a human-integrated artificial intelligence (HIAI) framework as a solution to these diabetes diagnostic challenges. Unlike reactive clinical testing, HIAI transforms smartphones into proactive and equitable tools, establishing a racially inclusive diagnostic infrastructure.

Methods:

We conducted a robust in silico validation to evaluate the HIAI system using 158,355 profiles matched to Indonesia's national health demographics. Our framework encompasses multimodal inputs that synergise human intelligence (digital anamnesis) with objective AI sensing using smartphone-derived PPG biomarkers. Our signal extraction algorithm is dynamically calibrated with the Monk Skin Tone scale to actively minimise ethnophenotypic bias. Diagnostic performance was evaluated using the Area Under the Curve (ROC-AUC), overall accuracy, Negative Predictive Value (NPV), and phenotypic-stratified performance deviation.

Results & Conclusion:

The HIAI framework achieved an ROC-AUC of 0.867 and 89.1% accuracy. This model also delivered a 95.7% NPV to be suitable for early pre-diagnosis and screening. Importantly, phenotypic-stratified analysis exhibited less than 1% performance variation across all ethno-racial subgroups, indicating its algorithmic fairness. This study demonstrates that the combination of human and unbiased physiological AI can be an inclusive pre-diagnosis tool. While current results are based on in silico validation, they provide a robust basis for future work. The next stage of development will be to validate these potential diagnostic capabilities through real-world clinical trials. It paves the way to an equitable diagnostic infrastructure for the Global South, consistent with global inclusive health goals.