Screening for novel risk factors related to high myopia using machine learning

Abstract

Background: High myopia-related complications have become a major cause of irreversible vision loss. Evaluating the association between potential factors and high myopia can provide insights into pathophysiologic mechanisms and further intervention targets for myopia progression. Method: Participants aged 12–25 years from National Health and Nutrition Examination Survey 2001–2006 were selected for the analysis. Myopia was defined as spherical equivalent (sum of spherical error and half of the cylindrical error) of any eyes ≤-0.5 diopters. High myopia was defined as the spherical equivalent of any eye ≤ − 5.00 diopters. Essential variables were selected by Random Forest algorithm and verified by multivariable logistic regression. Results: A total of 7,033 participants and 74 potential factors, including demographic (4 factors), physical examination (6 factors), nutritional and serological (45 factors), immunological (9 variables), and past medical history factors (10 factors), were included into the analysis. Random Forest algorithm found that several anthropometric, nutritional, and serological factors were associated with high myopia. Combined with multivariable logistic regression, high levels of serum vitamin A was significantly associated with an increased prevalence of high myopia (adjusted odd ratio = 1.46 for 1 µmol/L increment, 95% confidence interval [CI] 1.01–2.10). Furthermore, we found that neither C-reactive protein nor asthma increased the risk and severity of myopia. Conclusion: High levels of serum vitamin A was seemingly associated with an increased prevalence of high myopia. This borderline significant association should be interpreted with caution because the potential increased type I error after the multiple testing. It still needs further investigation regarding the mechanism underlying this association. Neither C-reactive protein nor asthma increased the risk and severity of myopia.