End-Digit Preference and the Quality of Blood Pressure Monitoring in Diabetic Adults

We conducted a non-concurrent prospective cohort study of 404 adults with type 2 diabetes seen at 16 academically affiliated clinics during the years 1999–2001. The cohort was comprised of federal employees and their dependents enrolled in a managed care program. Patients were eligible for inclusion if their charts were coded with ICD-9 codes 250.xx, 357.2, 362.0, or 366.41, or if they had been prescribed insulin or oral hypoglycemic agents and if they had been seen for two or more primary care encounters or one emergency room or hospital stay during 1 January 1999 to 31 December 2001. Chart abstraction was performed by two trained registered nurses and reviewed by two physicians. Abstracted data included BP readings by the medical assistant and/or by the physician, age, sex, race, height, weight, and presence of other comorbidities.

STATA software (9) was used to perform all statistical analyses. Distributions of end digits of systolic and diastolic BPs were plotted and a single sample test of proportion was used to test for significant differences. Student's t test and χ² analyses were used to compare demographic variables in BPs ending in zero and BPs ending in nonzero. Finally, multivariate analysis was performed using generalized estimating equations to fit logistic regression models to account for longitudinal correlation between multiple measurements in the same patients and to simultaneously adjust for confounding factors.

The majority of study subjects were male (58%), white (59%), older (63 ± 8 years), seen in urban clinic locations (83%), overweight (mean BMI 32 ± 5.8 kg/m²), and hypertensive (70%) (Table 1). There was 5,677 BPs taken with nonautomated sphygmomanometers (4,330 by nonphysicians; 1,347 by physicians). BP taken by nonphysicians was 139 ± 19/78 ± 11 (mean ± SD) (n = 4,330) and 142 ± 19/80 ± 10 taken by physicians (n = 1,347). Compared with the expected frequency of end-digit zeros of 20% (since zero is one of five even gradations on scales of manual sphygmomanometers), there was strong evidence of EDP among nonphysicians (50% of systolic and 50% of diastolic readings ended in zero; Fig. 1) and even stronger evidence among physicians (69% of systolic and 64% of diastolic readings ended in zero; Fig. 2).

Characteristics of the study population by BP end digit are presented in Table 2. In multivariate logistic regression, nonphysicians showed greater EDP for systolic blood pressure with older patients (odds ratio [OR] 1.07 per 5 years) and women (OR 1.36 vs. men) and for diastolic blood pressure with African-Americans (OR 1.25 vs. whites; all adjusted ORs, P < 0.05). In contrast, physicians were less influenced by age, sex, or race, but did show greater EDP in patients with lower BMI (OR 0.97 per 5 kg/m², P = 0.02; Table 3).

These analyses were repeated after pooling EDP data across systolic and diastolic readings among physicians and among nonphysicians. In these pooled analyses (Table 4), nonphysicians showed greater EDP for female patients (OR 1.26 vs. men; P = 0.006), and physicians showed greater EDP for African-American patients (OR 1.35 vs. whites; P = 0.03) and for patients with lower BMI (OR 0.97 per 5 kg/m²; P = 0.02).

Our data indicate that low-quality BP measurement remains common in the primary care of adults with type 2 diabetes and also suggest that it may be more common in women and African-Americans. Widely accepted practice guidelines recommend a target BP of <130/80 mmHg in adults with diabetes (10). Such guidelines heighten the need for accurate BP assessment in this population. However, the high observed frequency of EDP indicates that medical staff are rounding BPs to the nearest zero. Insofar as they round down (e.g., from 134 mmHg systolic to 130 mmHg), this practice may lead to underdiagnosis and undertreatment. Improving BP measurement may therefore represent a potential strategy for reducing health care disparities.

Practitioners might justify their behavior on the basis of the underlying variability of BP itself and the imprecision inherent in sphygomanometry: why record BPs that imply greater precision than is actually achievable? BP is more variable than other diabetes-related risk factors such as A1C and fasting lipids. Nonetheless, guidelines incorporate variability into their conceptual framework by suggesting that treatment be based on several BP measurements at successive visits (10). Within this framework, measures to reduce variability related to technique and equipment should increase underlying precision, inspire greater confidence, and enhance decision-making.

To our knowledge, our study is the first to quantify EDP in diabetic adults in a primary care setting. Limitations to this study include the inability to compare the level of EDP in diabetic patients compared with nondiabetic patients in the same practice (could the level of EDP actually be worse in nondiabetic patients?) and the inability to assess the impact of BP inaccuracy over time in terms of provider prescribing patterns or cardiovascular outcomes. The results of this study, however, raise some interesting questions regarding measurement bias in BP based on a patient's sex, race, and weight. Our findings of increased EDP in older patients and patients with lower BMI support previous findings (8). These results could indeed be explained by concern of missing a diagnosis of hypertension in young or obese patients, but additionally, lower EDP in heavier patients could also be due to the use of better fitting cuffs and less confidence in taking BPs in patients with larger arm circumferences, leading to more careful BP measurements. Additionally, the finding that there is more EDP in BPs taken in women and African-American patients is particularly worrisome, since women comprise 51% of all heart disease deaths (11) and African-Americans have the highest rate of hypertension among all racial groups (12), with substantially higher rates of cardiovascular mortality (13). Design of further studies on the topic of BP measurement accuracy should take into account these findings in an effort to decrease cardiovascular morbidity and mortality overall, but also to better discern causes of cardiovascular health disparity between sexes and among races in an effort to eliminate them altogether.

EDP in BP measurements taken in a diabetic population is common. The main implication is that better quality standards and/or alternative methods for BP measurement deserve attention as part of an overall strategy to tighten BP control in diabetic adults. While some sources of error, such as pervasive hearing loss, are not immediately amenable to systematic remediation, faulty measurement technique (14,15) and faulty equipment (16,17) can be addressed by retraining or better equipment. Automated sphygmomanometers have emerged as an alterative to reduce imprecision related to technique (18–20), although they may introduce new sources of inaccuracy (21,22). At the very least, manual sphygmomanometers, which are still in use in many physician offices, should be well maintained and calibrated to obtain the most accurate readings. Additionally, effort should be made to reduce observer and technical error (23,24) by improving training and standardization of BPs measured by physicians and clinic staff.

This study was funded by National Institutes of Health Grants K24-DK62222 and R01-DK48117 (F.L.B. was the principle investigator for both grants).

E.S.H.K. initiated/conducted/completed this work while a medical resident at the Johns Hopkins Hospital and was supported by the Osler Fund for Scholarship, Department of Medicine, Johns Hopkins University.