METHODS FOR ANALYZING INDIVIDUAL BLOOD PRESSURE MEASUREMENTS: TRENDS, SEASONALITY, AND FORECASTING

Abstract

The article investigates an approach to analyzing individual daily blood pressure measurements with the aim of identifying long-term trends, periodic fluctuations, and forecasting possible future changes. The study explores the use of linear regression, STL decomposition, and the ARIMA, SARIMA, and Holt-Winters models for analyzing blood pressure as a time series. The application of these methods made it possible to assess changes in a patient’s physiological parameters and to determine regularities in the dynamics of blood pressure. The analysis was conducted on data obtained from long-term home monitoring of a 67-year-old patient diagnosed with atherosclerosis. Linear regression was found to be effective in identifying baseline long-term trends that reflect gradual changes in the average blood pressure level. The STL decomposition method allows for the visualization of the time series structure and the separation of trend, seasonal, and residual components, providing a clearer understanding of the underlying patterns. The results of the study showed that the Holt-Winters model provides the highest forecasting accuracy for both systolic and diastolic blood pressure in cases where seasonal fluctuations are absent or weakly expressed. Conversely, the SARIMA model demonstrated better performance when the data contained pronounced periodic patterns, while ARIMA was found to be most appropriate as a baseline approach for assessing general trends without emphasizing seasonal influences. To quantitatively evaluate the accuracy of the models, standard statistical metrics – Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) – were employed. The comparative analysis of these metrics confirmed the importance of accounting for seasonality when building predictive models of physiological processes. The study highlights the potential of time series analysis methods for use in personalized medical analytics. It demonstrates that these methods can effectively reveal the dynamics and variability of an individual’s cardiovascular parameters, enabling a more nuanced understanding of personal health trajectories. However, the research also indicates that the reliability and accuracy of such models significantly depend on the quantity and duration of collected data. For reliable short-term forecasting, a dataset containing at least one hundred observations is recommended, while more extensive time series are required for identifying stable long-term trends and seasonal cycles. Furthermore, integrating additional physiological and environmental parameters – such as heart rate variability, physical activity, stress levels, and ambient temperature – into forecasting models could significantly enhance their precision and adaptability. These factors may influence blood pressure fluctuations and thus should be incorporated into future predictive frameworks. The study concludes that comprehensive, long-term monitoring combined with advanced time series modeling can serve as a foundation for developing precise individualized risk profiles that reflect the true dynamics of a patient’s physiological condition and support data-driven medical decision-making.