Hybrid Econometric–Deep Learning Framework for Estimating the Economic Impact of Disease Burden

Akinrotimi Akinyemi Omololu, Atoyebi Jelili Olaniyi, Omotosho Israel Oluwabusayo

Abstract


Despite the widespread use of Disability-Adjusted Life Years (DALYs) to measure health burden, generally well-known econometric models often fail to account for nonlinear relationships and hidden confounders in panel data. This study introduces a hybrid framework that integrates latent embeddings (z) learned via a feedforward neural network (FFNN) with a panel regression structure enhanced by correlated random effects (CRE), and implements coefficient inference through double/debiased machine learning (DML). Using country-level panel data from 1990 to 2021 covering DALYs, human capital, capital stock, and population growth—the model extracts nonlinear representations and produces a more accurate estimate of the elasticity of labor productivity relative to disease burden. The hybrid CRE‑DML model achieves superior predictive accuracy (R² ≈ 0.62; MAE = 0.105; RMSE = 0.141), performing better than both pure econometric and standalone machine learning models. The estimated elasticity (β₁) of DALYs on productivity is –0.038 (SE = 0.010), much tighter than the –0.052 estimate obtained from the traditional regression. Robustness tests—encompassing leave-unit-out cross-validation, regularization variations, and alternative embedding learners (Random Forest and XGBoost), confirm that accuracy fluctuates within ±0.02 and elasticity remains stable within ±0.002 across configurations. This approach unites the predictive strengths of representation learning with the interpretability of econometric modeling, offering a scalable, policy-relevant tool for analyzing health‑economic relationships in data-scarce environments.


Full Text:

PDF

References


Ahrens, A., Chernozhukov, V., Hansen, C., Kozbur, D., Schaffer, M., & Wiemann, T. (2025). An introduction to double/debiased machine learning. arXiv. https://arxiv.org/abs/2504.08324

Bach, P., Chernozhukov, V., Kurz, M. S., Spindler, M., & Klaassen, S. (2021). DoubleML – An object-oriented implementation of double machine learning in Python. arXiv. https://arxiv.org/abs/2104.03220

Bohlke Schneider, C., et al. (2022). Resilient neural forecasting systems for labor planning. arXiv. https://arxiv.org/abs/2203.08492 Ouci

Chen, F., Wang, J., Li, H., & Zheng, L. (2024). Modeling productivity loss under heat stress using machine and deep learning. BMC Public Health, 24, Article 16498. https://doi.org/10.1186/s12889-023-16498-7 PMC

Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., & Robins, J. (2018). Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal, 21(1), C1–C68.

Clarke, P. S., & Polselli, A. (2023). Double Machine Learning for Static Panel Models with Fixed Effects. arXiv. https://arxiv.org/abs/2312.08174

Clarke, P. S., & Polselli, A. (2023). Double machine learning for static panel models with fixed effects. arXiv. https://arxiv.org/abs/2312.08174

Cybenko, G. (1989). Approximation by superpositions of a sigmoidal function. Mathematics of Control, Signals and Systems, 2(4), 303–314.

Daroudi, S., et al. (2021). Cost per DALY averted across income strata: GBD-based estimates for cost-effectiveness thresholds. Cost Effectiveness and Resource Allocation, 19, Article 38. https://doi.org/10.1186/s12962-021-00260-0

Desai, A. (2023). Machine learning for economics research: When, what, and how? arXiv.

Fuhr, J., & Papies, D. (2024). Double machine learning meets panel data: Promises, pitfalls, and potential solutions. arXiv. https://arxiv.org/abs/2409.01266

Global Burden of Disease 2021 Collaborators. (2021). Global incidence, prevalence, DALYs, and HALE for 371 diseases in 204 countries (GBD 2021). The Lancet. https://doi.org/10.1016/S0140-6736(24)00757-8

Grinsztajn, L., Oyallon, É., & Varoquaux, G. (2022). Why do tree-based models still outperform deep learning on tabular data? arXiv. https://arxiv.org/abs/2207.08815

Italian COVID 19 DALY burden group. (2021). Productivity loss estimation due to COVID 19 DALYs in Italy. International Journal of Environmental Research and Public Health, 17, Article 4233. https://doi.org/10.3390/ijerph17124233

Kirigia, J. M., & Kubai, P. K. (2023). Monetary value of disability-adjusted life years and potential productivity losses associated with neglected tropical diseases in the East African Community. Research and Reports in Tropical Medicine, 14, 35–47. https://doi.org/10.2147/RRTM.S382288 infontd.org+6PubMed+6SciProfiles+6

Li, Y., Zhang, X., & Chen, Z. (2022). Comparison of ARIMA, deep neural networks, and LSTM models for forecasting DALYs in Tianjin, China. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2022.848370

Li, Y., Zhang, X., & Chen, Z. (2022). Comparison of ARIMA, DNN, and LSTM models for predicting pneumoconiosis DALYs in Tianjin, China. BMC Public Health.

Liu, A. J., Mukherjee, A., Hu, L., Chen, J., & Nair, V. N. (2022). Performance and interpretability comparisons of supervised machine learning algorithms: An empirical study. arXiv. https://arxiv.org/abs/2204.12868

Masini, A., et al. (2020). Machine learning advances for time-series forecasting: Linear, nonlinear, and hybrid methods. arXiv. https://arxiv.org/abs/2012.12802

McElfresh, D. C., et al. (2023). When do neural nets outperform boosted trees on tabular data? arXiv. https://arxiv.org/abs/2305.02997

Multilayer perceptron. (2025). In Wikipedia. Retrieved from https://en.wikipedia.org/wiki/Multilayer_perceptron

Murray, C. J. L., & Lopez, A. D. (2019). Disability-adjusted life years (DALYs) and their use in global burden of disease studies. Lancet.

Nkosana, T. K., & Ferrer Worient, M. T. (2022). Health burdens and labor productivity in Africa’s middle and low income economies: Implications for the COVID 19 era. Journal of the Knowledge Economy.

Our World in Data. (2025). Burden of Disease [Dataset].

Patel, R., Desai, P., & Singh, A. (2020). Chronic disease burden and workforce participation in India: a micro-level economic study. Journal of Development Economics.

Smith, H. K., & Abbas, H. (2024). Hybrid models combining econometric, machine learning and deep learning approaches in economic forecasting. arXiv. https://arxiv.org/abs/2505.19617

Stempień, D., & Ślepaczuk, R. (2025). Hybrid models combining econometric, machine learning and deep learning approaches in economic forecasting. arXiv preprint arXiv:2505.19617. arxiv.orgarxiv.org

Tariq, M. U., Ismail, S. B., Babar, M., & Ahmad, A. (2024). Deep learning for prediction of population health costs: COVID 19 forecasting accuracy across models. Osong Public Health and Research Perspectives, 15(2), 115–136.

Yang, B., Long, W., & Cai, Z. (2024). Machine learning based panel data models: Debiased machine learning for dynamic panel estimation. Working Paper Series in Theoretical and Applied Economics 202402. https://kuwpaper.ku.edu/2024Papers/202402.pdf.


Refbacks

  • There are currently no refbacks.