Logistic regression model for predicting failure of dual antihypertensive therapy: a prospective comparative non-randomized clinical trial

Abstract

Background. Initial dual antihypertensive therapy is currently considered as the first management step for the majority of patients with arterial hypertension. However, it often fails to achieve the target blood pressure levels. An approved algorithm for predicting the failure of dual antihypertensive therapy is still to be developed. Objectives. To establish predictors of dual antihypertensive therapy failure in patients with high and very high cardiovascular risk and to create a model for predicting negative outcome of dual antihypertensive therapy. Methods. The paper presents a prospective comparative non-randomized clinical trial. The recruiting of participants and recording of results were carried out in March–December 2019 with 3 months of the follow-up period. The trial involved examination of 88 patients with poor blood pressure control, stage II and III arterial hypertension, high and very high cardiovascular risk of stages 1–3. Clinical and laboratory examination was carried out in compliance with the current regulatory documents. Additional examination included tests for uric acid, high-sensitivity C-reactive protein, as well as respiratory polygraphy and computerized capillaroscopy. All patients were prescribed dual antihypertensive therapy. The primary search for predictors was performed using the binary logistic regression. The predictive model was developed by stepwise variable selection. The diagnostic significance of the binary classifier was assessed by means of ROC-curve analysis; the calculation was performed using MedCalc 20.218 software (MedCalc Software Ltd., Belgium). Results. Administration of two hypotensive drugs appears to be effective in 33% of patients. The final model for predicting negative outcomes of dual antihypertensive therapy included such independent predictors as interventricular septal thickness, daily mean systolic blood pressure, and area density of the capillary network. The odds ratio accounted for 9.1 (95% confidence interval 3.12; 26.82). The area under the ROC curve based on the multiple binary logistic regression model comprised 0.805±0.05 with 95% confidence interval: 0.707-0.882 (p<0.0001). The sensitivity and specificity of the method amounted to 83.1 and 69.0%, respectively. The prediction accuracy comprised 77.3%. Conclusion. The development of patient-oriented algorithms for selection of hypotensive treatment is considered to be essential due to poor blood pressure control during dual antihypertensive therapy. The developed prognostic model may be applied when managing hypertension.