Methods
The Mid Devon Medical Practice operates three separate surgeries in a rural area. This study was carried out in the main surgery (list size 1900). Eligible participants were those registered with the author (CEC) and receiving treatment for hypertension. Hypertension was defined by the then current guidelines of the British Hypertension Society (≥160/100 mm Hg or ≥140/90 mm Hg in the presence of target organ damage, diabetes, or coronary heart disease risk score ≥15%). We excluded participants on the basis of anatomical criteria (loss, previous injury, surgery above wrist level, or paralysis of one arm) or for practical reasons—that is, the inability or unwillingness to regularly attend the surgery for review.
Measurements
At recruitment one investigator (CEC) measured blood pressure using a standard mercury sphygmomanometer (Accoson; AC Cossor, Harlow, Essex), which was calibrated every six months. Standard or large cuffs were used as appropriate. Pairs of blood pressure readings were collected sequentially after the participant had been seated for five minutes; measurement was taken in the arm first presented without prompting, and the cuff was then swapped to the other arm and another measurement taken. The arm was supported during each measurement. We obtained single pairs of measurements at the first and subsequent two visits and recruitment ran from 9 November 1999 to 17 June 2002. Return visits were planned every six months if blood pressure was controlled, or at shorter intervals if treatment for high blood pressure was being adjusted. We averaged the three pairs of readings to obtain a mean systolic and diastolic blood pressure for each arm to derive the mean interarm difference. After the first visit we recorded the participant’s medical history and characteristics (age, sex, smoking status, body mass index, glucose level, total cholesterol level, creatinine level, pre-treatment blood pressure, years since diagnosis of hypertension, evidence of left ventricular hypertrophy on electrocardiogram, and Framingham risk score calculated from pre-treatment values extracted from patient records). If such data were missing we undertook the necessary investigations. Drugs were adjusted to achieve optimal blood pressure control according to guidelines, but we did not include use of drugs in the dataset. We prospectively collected data on events until 26 April 2011. Events were defined as death (cardiovascular or all cause using death certification data augmented where available by post mortem findings) or non-fatal cerebrovascular events, and cardiovascular events (myocardial infarction or a new diagnosis of angina), confirmed after referral to secondary care.
Data Analysis
We entered anonymised data on an Excel spreadsheet and used SPSS Predictive Analytics Software Statistics v18.0.0 and Stata v11.1 for analysis. Left ventricular hypertrophy was diagnosed by electrocardiography at recruitment using the Perugia scoring system, and we calculated scores for the risk of coronary heart disease at 10 years from the Framingham equation. We compared the participant’s characteristics at entry to the cohorts using t or χ tests between groups according to predefined cut-off points of interarm differences that have been used in the literature and in our previous report—namely, 10 mm Hg or more and 15 mm Hg or more differences in systolic blood pressure and 10 mm Hg or more differences in diastolic blood pressure. At each of these cut-off points we used Kaplan-Meier survival plots to compare the time to death (all cause and cardiovascular), combined non-fatal cardiovascular and cerebrovascular events, and death or non-fatal events. We used a Cox’s proportional hazards regression model to calculate the unadjusted hazard ratios for these outcomes, fitting interarm difference as either a continuous variable or using the defined cut-off points. A multivariable Cox regression model was used to derive adjusted hazard ratios, which included the Framingham risk score (model 1) and, in addition, included mean blood pressure (systolic for analyses of systolic interarm difference, and diastolic for analyses of diastolic interarm difference, calculated as the mean of three pairs (n=6) of blood pressure measurements at recruitment), presence of diabetes, and pre-existing cardiovascular or peripheral vascular disease on entry to the cohort (model 2). Regardless of their statistical effect we included these variables in the model for their relevance on clinical grounds. To assess the specific contribution of interarm differences in blood pressure, we used the likelihood ratio test to assess the reduction in goodness of fit arising as the result of omitting the interarm difference term from each of the adjusted models. A predefined secondary analysis was undertaken in the subgroup of participants without previous cardiovascular disease at entry to the cohort. We assessed proportionality of hazards over time by plotting -ln(-ln(survival)) versus ln(analysis time), and tested this using Schoenfeld residuals. (Also see web extra on bmj.com.) We found no major violations of the proportional hazards assumption. The competing risk of death was accounted for by censoring at the date of death. In mortality outcome models, we considered any previous non-fatal events to be uninformative.