Influence of Confounding Control on Liver Fibrosis Modeling: A Propensity Score Matching and Lasso-Based Logistic Regression Approach

Abstract

In large-scale observational datasets, associations between clinical outcomes and biomarkers are highly sensitive to confounding control and statistical modeling strategies. This study aimed to evaluate how confounding-control and variable-selection strategies influence the observed association between retinopathy and liver fibrosis in a large observational dataset, thereby providing a methodological demonstration of how statistical modeling decisions may influence epidemiological inference. Data were obtained from the National Health and Nutrition Examination Survey (NHANES) 2005-2008 cycles. Liver fibrosis was defined using the FIB-4 index and classified as no/mild fibrosis (FIB-4 < 1.45) or significant fibrosis/cirrhosis (FIB-4 ≥ 1.45). Retinopathy status and severity were determined using standardized retinal imaging data. To improve baseline comparability between fibrosis groups, propensity score matching (PSM) was implemented based on age and sex, and covariate balance was assessed using standardized mean differences. Variable selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) with 10-fold cross-validation, and the optimal penalty parameter (λ) was selected based on the minimum cross-validated error (λmin), followed by multivariable logistic regression modeling. Variables used in the calculation of the FIB-4 index were excluded from regression analyses to avoid circular inference. A total of 5,364 participants were included. Before matching, substantial imbalance was observed between fibrosis groups, particularly for age and sex. After 1:1 propensity score matching, adequate covariate balance was achieved. LASSO-based variable selection identified hepatitis C virus infection, body mass index, height, race, and retinopathy as candidate predictors of liver fibrosis. In the final multivariable logistic regression model, hepatitis C virus infection showed the strongest association with significant liver fibrosis (OR = 2.70, 95% CI: 1.66-4.39), while body mass index and height demonstrated modest but statistically significant associations. Retinopathy was not independently associated with liver fibrosis after multivariable adjustment. The results demonstrate that the apparent association between retinopathy and liver fibrosis in observational data is highly dependent on statistical modeling choices and confounding-control strategies. Rather than supporting a direct clinical relationship, the findings emphasize how analytical design and variable selection methods can substantially shape conclusions derived from large-scale health datasets. From a methodological perspective, this study illustrates how confounding-control and variable-selection strategies can alter epidemiological inference in observational research. This study highlights the importance of transparent and rigorously justified statistical modeling frameworks in applied data-driven research.

Keywords

• Propensity score matching
• LASSO regression
• Logistic regression
• Confounding
• Covariate balance

References

1. Asero, C., Giandalia, A., Cacciola, I., Morace, C., Lorello, G., Caspanello, A. R., Alibrandi, A., Squadrito, G., & Russo, G. T. (2023). High Prevalence of Severe Hepatic Fibrosis in Type 2 Diabetic Outpatients Screened for Non-Alcoholic Fatty Liver Disease. J Clin Med, 12(8). https://doi.org/10.3390/jcm12082858
2. Austin, P. C. (2011). An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate behavioral research, 46(3), 399-424.
3. Bangchang, K. N. (2024). Application of Bayesian variable selection in logistic regression model. AIMS Mathematics, 9(5), 13336-13345. https://doi.org/10.3934/math.2024650
4. Bo, J., & Yang, M. (2025). Dose-response relationship between physical activity and visceral fat mass: a cross-sectional study based on NHANES 2011-2018. BMC Public Health, 25(1), 3113. https://doi.org/10.1186/s12889-025-24393-6
5. Burgos-Ochoa, L., & Clouth, F. J. (2025). Causal Inference and Survey Data in Paediatric Epidemiology: Generalising Treatment Effects From Observational Data. Paediatr Perinat Epidemiol. https://doi.org/10.1111/ppe.70042
6. Cheang, I., Zhu, X., Zhu, Q., Li, M., Liao, S., Zuo, Z., Yao, W., Zhou, Y., Zhang, H., & Li, X. (2022). Inverse association between blood ethylene oxide levels and obesity in the general population: NHANES 2013-2016. Front Endocrinol (Lausanne), 13, 926971. https://doi.org/10.3389/fendo.2022.926971
7. Chhabra, S., Singh, S. P., Singh, A., Mehta, V., Kaur, A., Bansal, N., & Sood, A. (2022). Diabetes Mellitus Increases the Risk of Significant Hepatic Fibrosis in Patients With Non-alcoholic Fatty Liver Disease. J Clin Exp Hepatol, 12(2), 409-416. https://doi.org/10.1016/j.jceh.2021.07.001
8. Deravi, N., Dehghani Firouzabadi, F., Moosaie, F., Asadigandomani, H., Arab Bafrani, M., Yoosefi, N., Poopak, A., Dehghani Firouzabadi, M., Poudineh, M., Rabizadeh, S., Kamel, I., Nakhjavani, M., & Esteghamati, A. (2023). Non-alcoholic fatty liver disease and incidence of microvascular complications of diabetes in patients with type 2 diabetes: a prospective cohort study. Front Endocrinol (Lausanne), 14, 1147458. https://doi.org/10.3389/fendo.2023.1147458

9. Ellis, G. M., & Souza, P. E. (2021). Using Machine Learning and the National Health and Nutrition Examination Survey to Classify Individuals With Hearing Loss. Front Digit Health, 3, 723533. https://doi.org/10.3389/fdgth.2021.723533
10. Elsabaawy, M. (2024). Liver at crossroads: unraveling the links between obesity, chronic liver diseases, and the mysterious obesity paradox. Clinical and experimental medicine, 24(1), 240. https://doi.org/10.1007/s10238-024-01493-y
11. Erman, H., Boyuk, B., Arslan, S., Akin, S., & Keskin, O. (2024). Noninvasive Liver Fibrosis Indices as Indicators of Microvascular and Macrovascular Complications in Type 2 Diabetes. Metab Syndr Relat Disord, 22(8), 619-625. https://doi.org/10.1089/met.2024.0022
12. Feng, G., Zheng, K. I., Li, Y. Y., Rios, R. S., Zhu, P. W., Pan, X. Y., Li, G., Ma, H. L., Tang, L. J., Byrne, C. D., Targher, G., He, N., Mi, M., Chen, Y. P., & Zheng, M. H. (2021). Machine learning algorithm outperforms fibrosis markers in predicting significant fibrosis in biopsy-confirmed NAFLD. J Hepatobiliary Pancreat Sci, 28(7), 593-603. https://doi.org/10.1002/jhbp.972
13. Guo, Y., Shen, B., Xue, Y., & Li, Y. (2023). Development and validation of a non-invasive model for predicting significant fibrosis based on patients with nonalcoholic fatty liver disease in the United States. Front Endocrinol (Lausanne), 14, 1207365. https://doi.org/10.3389/fendo.2023.1207365
14. Issanov, A., Karim, M. E., Aimagambetova, G., & Dummer, T. J. B. (2022). Does Vaccination Protect against Human Papillomavirus-Related Cancers? Preliminary Findings from the United States National Health and Nutrition Examination Survey (2011-2018). Vaccines (Basel), 10(12). https://doi.org/10.3390/vaccines10122113
15. Jacob, M., Joseph, M., & Idiculla, J. (2023). Non-alcoholic fatty liver disease and diabetic retinopathy: Is there an association? J Family Med Prim Care, 12(9), 2028-2031. https://doi.org/10.4103/jfmpc.jfmpc_2327_22
16. Jeong, J., & Kim, Y.-M. (2024). Robust estimation of a marginal causal effect on the binary outcome using propensity score matching. Journal of the Korean Data And Information Science Society, 35(1), 161-177. https://doi.org/10.7465/jkdi.2024.35.1.161
17. Li, J., Xiang, Y., Han, J., Gao, Y., Wang, R., Dong, Z., Chen, H., Gao, R., Liu, C., Teng, G. J., & Qi, X. (2024). Retinopathy as a predictive indicator for significant hepatic fibrosis according to T2DM status: A cross-sectional study based on the national health and nutrition examination survey data. Ann Hepatol, 29(4), 101478. https://doi.org/10.1016/j.aohep.2024.101478
18. Mantovani, A., Mozrieri, M. L., Aldigeri, R., Palmisano, L., Masulli, M., Bonomo, K., Baroni, M. G., Cossu, E., Cimini, F. A., Cavallo, G., Buzzetti, R., Mignogna, C., Leonetti, F., Bacci, S., Trevisan, R., Pollis, R. M., Cas, A. D., de Kreutzenberg, S. V., & Targher, G. (2024). MASLD, hepatic steatosis and fibrosis are associated with the prevalence of chronic kidney disease and retinopathy in adults with type 1 diabetes mellitus. Diabetes Metab, 50(1), 101497. https://doi.org/10.1016/j.diabet.2023.101497
19. Rogers, P. (2023). Controlling for Confounding in Complex Survey Machine Learning Models to Assess Drug Safety and Risk. In Machine Learning and Deep Learning in Computational Toxicology (pp. 355-374). Springer.
20. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41-55.
21. Salerno, S., Roberts, E. K., Needham, B. L., McCormick, T. H., Li, F., Mukherjee, B., & Shi, X. (2025). What's the Weight? Estimating Controlled Outcome Differences in Complex Surveys for Health Disparities Research. Statistics in medicine, 44(23-24), e70289.
22. Schonlau, M. (2023). Lasso and Friends. In Applied Statistical Learning: With Case Studies in Stata (pp. 73-95). Springer.
23. Shaji, N., Singhai, A., Sarawagi, R., Pakhare, A. P., Mishra, V. N., & Joshi, R. (2022). Assessment of Liver Fibrosis Using Non-invasive Screening Tools in Individuals With Diabetes Mellitus and Metabolic Syndrome. Cureus, 14(2), e22682. https://doi.org/10.7759/cureus.22682
24. Storz, M. A. (2023). Does Self-Perceived Diet Quality Align with Nutrient Intake? A Cross-Sectional Study Using the Food Nutrient Index and Diet Quality Score. Nutrients, 15(12). https://doi.org/10.3390/nu15122720
25. Woodard, J. S., & Abrams, G. A. (2024). Increased Prevalence of Advanced Metabolic Dysfunction-Associated Steatotic Liver Disease Fibrosis in Type 2 Diabetics Despite Low-Risk Fibrosis-4 Index Scores. Journal of Endocrinology and Metabolism, 14(1), 40-47. https://doi.org/10.14740/jem935
26. Zhang, Z., Wang, J., Wang, H., Qiu, Y., Zhu, L., Liu, J., Chen, Y., Li, Y., Liu, Y., Chen, Y., Yin, S., Tong, X., Yan, X., Xiong, Y., Yang, Y., Zhang, Q., Li, J., Zhu, C., Wu, C., & Huang, R. (2023). An easy-to-use AIHF-nomogram to predict advanced liver fibrosis in patients with autoimmune hepatitis. Front Immunol, 14, 1130362. https://doi.org/10.3389/fimmu.2023.1130362