Sources of Measurement Variation in Blood Pressure in Large-scale Epidemiological Surveys with Follow-up

Abstract

The Copenhagen City Heart Study (CCHS) is a longitudinal epidemiological study of 19698 subjects followed up since 1976. Variation in blood pressure (BP) measurement in the first three CCHS surveys is evaluated by assessing two components, systematic variation and random variation [daytime and seasonally variation, observer bias, non-response bias, variation with explanatory variables, such as diabetes, hypertension, body mass index (BMI), height, plasma cholesterol and smoking] for the purpose of identifying relevant errors in population surveys. BP was measured in the seated position after a 5 min rest, with the cuff around the non-dominating arm, in accordance with recommended guidelines. The participation rate fell from 74% in survey 1 to 63% in survey 3. Significant non-response bias with respect to BP values was not found. No daytime variability was noted either in systolic (SBP) or diastolic (DBP) BPs. A trend towards a lower BP was seen during the summertime. Random variation, expressed as the standard deviation of the measured values, increased with increasing BP values (SBP: 11.9-13.4 to 21.2- 25.1 mmHg; DBP: 10.6-11.2 to 11.9-13.4 mmHg). SBP was positively correlated to BMI and plasma cholesterol. SBP was 5-10 mmHg higher in diabetics ( p = 0.000-0.04) than in age- and sex-matched nondiabetics. DBP did not differ between the two groups. Smokers from the age of 50 years had a 2-4 mmHg lower SBP ( p = 0.000-0.01) and 1-3 mmHg lower DBP ( p = 0.000-0.005) than had non-smokers. In addition, significantly fewer smokers took antihypertensive medication than did non-smokers ( p = 0.000). In conclusion, judging from the degree of association with BP and/or differences between the three surveys, the most important factors to consider were seasonal variation, BMI, the use of antihypertensive drug therapy, plasma cholesterol, smoking status and diabetes. An inter-survey comparison of BP in population cohorts requires controlling for these factors.