Neural network analysis of the relationships of risk factors with a fatal event depending on prospective follow-up duration

Aim . To compare the significance of risk factors (RF) in neural network modeling of a fatal outcome for prospective follow-up periods of 10, 20, 30 and 40 years. Material and methods . From the Russian Lipid Research Clinics Study of 1975-1982, 13263 men and 5691 women were included in the current follow-up until 2017. The end point was all-cause death. Sex, age, blood pressure, heart rate, body mass index, blood lipid levels, smoking and education status, hypertension and hypotension were analyzed. Artificial neural network simulators were used to build multivariate models. Results . According to the sensitivity analysis, the significance of all input variables included in the models increases with follow-up duration extension. The minimum significance of studied risk factors is observed with a 10-year follow-up in women. Conclusion . Neural network prediction of a fatal event using the studied risk factors reaches maximum information content by 30 years of prospective follow-up.