Accuracy of Visually Estimated Home Blood Pressure: Result from the Estimating Yield of Eyeballed BP Study

Home blood pressure monitoring is the preferred method for long-term monitoring and medication titration among individuals already receiving treatment, according to the recent AHA/ACC Hypertension guidelines¹. Although many digital devices can transmit data directly into electronic medical records, paper logs remain the most practical method used to record home BP readings. Even when home BP data are electronically transmitted, many electronic health records do not automatically calculate averages. Consequently, clinicians often rely on visual estimation (i.e., “eyeballing”) of average BP from these logs to guide medical decisions. However, the accuracy of visual BP estimation and its impact on clinical decision-making remain unknown.

Method:

We surveyed healthcare providers at UT Southwestern Medical Center using institutional email distribution lists from the Internal Medicine and Family Medicine departments; and the School of Health Professions. All physicians, trainees, and advanced practice providers (APP) on these lists during the April–July 2025 sampling period were invited, and 194 completed the survey. Participants were asked to review five simulated BP logs and estimate the average systolic BP (SBP) and diastolic BP (DBP), example Figure C. The final two BP logs represented a hypothetical patient with untreated stage-2 hypertension, for which participants were asked to provide treatment recommendations. In addition, the treatment recommendation for each stage of hypertension is shown in question 4–5 as per the guideline recommendation (e.g. start 2-drug regimen for stage-2 hypertension). The study was granted exempt status from the Institutional Review Board, as the survey was anonymous. Bland-Altman analysis was used to assess agreement between visually estimated and actual mean BP. Fisher’s exact test was used to compare incorrect clinical decisions based on BP underestimation versus accurate or overestimation. Linear mixed regression models with random effects at the participant level were used to determine the relationship between predictors and accuracy of SBP estimation. The intraclass correlation coefficient was calculated to quantify variance attributable to between-participant differences. Univariate logistic regression models were used to examine the association between predictors and initiation or intensification of treatment. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

Figure A. Bland-Altman analysis showing agreement between visually estimated and actual mean SBP

Figure B. Bland-Altman analysis showing agreement between visually estimated and actual mean DBP

Figure C. Example of questions as seen on the survey provided.

Figure D. Tables showing univariate predictors of BP estimation, treatment initiation, and treatment intensification using linear mixed regression model and logistic regression models, respectively.

Figure E. Bar graphs showing percent of incorrect clinical decisions based on BP underestimation versus accurate or overestimation using Fisher’s exact test.

Results:

The participants consisted of 60% females, 61% physicians (38% general internal medicine), 2% APPs, and 19% medical or APP students, with a mean age of 35 years (range 22–69). The mean and standard deviation of actual home SBP/DBP in question 1–5 are 123±11.3/ 67±11.2, 107±17.2/ 55±7.7, 137±15.5/ 80±8.6, 144±11.5/ 76±8.4, 139±9.4/ 70±6.2, respectively. The distribution of differences between eyeballed and actual mean BP values in the Bland–Altman analysis showed only mild tail deviation in the residual and Q–Q plots, with skewness values for SBP and DBP of 0.15 and 0.18, respectively, indicating approximate normality. The eyeball method overestimated SBP by 1.8 mmHg (95% CI: -7.4,11.0) and DBP by 3.1 mmHg (95% CI: -4.5,10.7, Figure A-B). A total of 31.1% of SBP and 27.1% of DBP estimates fell outside the acceptable threshold (±5 mmHg). Percentage of providers who underestimated home BP in question 1–5 are 21.3, 33.6, 29.1, 35, and 38, respectively. Older age and years of practice were associated with overestimation of BP (Figure D). Among those who underestimated home SBP (eyeball < actual SBP), in the hypothetical case with untreated stage-2 hypertension, a higher proportion of participants failed to recommend prompt initiation with 2-drug regimen when compared with the group who accurately estimated or overestimated home BP (eyeball ≥ actual BP, 77% vs 45%, respectively, p=0.0006, Figure E). Importantly, 35% of participants who underestimated home SBP did not intensify therapy in the last question for the same patient with persistently uncontrolled BP, when compared with 6% of the group who reported eyeball ≥ actual BP (p <0.0001, Figure E).

Discussion:

Our study results suggest that the accuracy of visual estimation of mean home BP is limited and falls outside the 5-mmHg threshold used in the validation of home BP devices according to the guidelines ² in >30% of participants. Furthermore, this inaccuracy is associated with failure to initiate antihypertensive medication in the untreated hypertensive patient, or to intensify therapy in the treated but uncontrolled patient, despite clinical indication. Our findings underscore the importance of accurately calculating mean home BP levels to support appropriate hypertension management and minimize therapeutic inertia. Our scenarios did not include low average home BP values; therefore, we were unable to assess potential overtreatment or opportunities for deintensification, which remains an area without clearly defined guideline thresholds. Although the study is limited by a modest sample size in a single academic center, the results highlight the need for development of software tools or practical mobile applications that allow accurate mathematical calculation of mean home BP at the point of care. Future work should also evaluate whether clinicians can accurately interpret additional home BP patterns including diurnal variability and longitudinal trends.