Artificial Intelligence in determining optimal questions in assessing social socio-economic status of individuals for routine immunisation services in Tanzania

Abstract

Background: Accurate determination of socio-economic status (SES) is crucial for equitable access to immunization services. Existing SES assessment tools, like the DHS wealth index, are comprehensive but impractical for routine clinical settings due to their length.Objective: To identify the minimum number of questions that can validly determine SES using artificial intelligence (AI), and to assess their validity compared to the standard DHS wealth index.Methods: This study applied Principal Component Analysis (PCA), Convolutional Neural Networks (CNN), and Artificial Neural Networks (ANN) using the DHS wealth index as the gold standard. Data were collected from routine RCH clinics in Tanzania. CNN was used to extract weights for each question, and ANN was trained to validate different subsets of questions.Results: Eight questions were identified as optimal for assessing household SES, achieving a sensitivity of 76.9% and specificity of 94.2%. The correlation between the derived SES score and the DHS standard wealth index was R² = 0.76. The study demonstrates that CNN can be an effective method for selecting valid SES indicators in healthcare.Conclusion: AI models, particularly CNNs, can identify a small set of questions with strong predictive validity for SES, enabling practical and accurate equity assessments in immunization programs. Integrating such tools into the Tanzania Immunization Registry (TImR) may enhance targeting and improve vaccine coverage. This study was approved by the ethical review board of MUHAS.