OBJECTIVE

To compare the informativity of different types of variables characterizing impairments of the blood pressure regulation level, with regard to the neural network models for prediction of the fatal outcome probability during long-term prospective observation from 10 to 40 years.

MATERIALS AND METHODS

The research included 13 263 men and 5 691 women with a duration of prospective observation up to 42 years from the Russian part of the Lipid clinics study in 1975—1982. We analyzed the quantitative values of systolic, diastolic and mean dynamic blood pressure (BP), as well as nominative variables characterizing the presence or absence of arterial hypertension and hypotension. The end point was death from all causes. Models of prediction of the fatal outcome probability with a prospective observation of 10, 20, 30, 40 years were built using the programs — simulators of artificial neural networks built.

RESULTS

The significance of the input variables included in the models was compared using general sensitivity analysis. The importance of nominative variables characterizing the presence of arterial hypertension and hypotension in all periods of observation was higher than the significance of quantitative variables — systolic, diastolic and mean dynamic BP.

CONCLUSION

In the context of neural network models of the prediction of the fatal event probability, relative significance of quantitative variables characterizing measured blood pressure level is lower compared to the significance of nominative variables characterizing the presence or absence of arterial hypertension and hypotension.