1DIASTOLIC BLOOD PRESSURE, CARDIOVASCULAR DISEASE, and MORTALITYSidney Port, Linda Demer, Robert Jennrich, Noel Boyle, Alan GarfinkelDepartments of Mathematics (S Port PhD), Statistics (S Port, R Jennrich PhD),Medicine-Cardiology (L Demer MD PhD, N Boyle MD PhD, A Garfinkel PhD), Physiology(L Demer) and Physiological Science (A Garfinkel), University of California, Los Angeles,CAPrepublication managementNoel BoyleUCLA School of MedicineDivision of CardiologyRoom 47-123 CHS10833 LeConte AvenueLos Angeles, CA 90095-1679Phone: 310-794-2165Fax: 310-206-9133Email: [email protected] to:professor Sidney PortDepartment of MathematicsUniversity of California, Los AngelesLos Angeles, CA. 90025(e-mail:[email protected])2SummaryBackground There are two views of the relation between diastolic blood pressure andrisks of cardiovascular disease and death. The most widely accepted is that risk issteadily rising with diastolic blood pressure. The other is that the relation follows a “J-curve”, in which risks are also increased at low as well as high pressures. This view iscontroversial because it is not consistently found. We reanalyzed data from theFramingham study to determine the nature of the relation of cardiovascular risk todiastolic blood pressure, to see why the J-curve is elusive, to seek justification for taking90 mm Hg as the cut-point for diastolic hypertension, and to determine if the systolic ordiastolic pressure is a better predictor of risk.Methods Reanalysis of the Framingham data on diastolic blood pressure usinglogistic splines.Results & Interpretations (1) The Framingham data rejects the linear logistic model;the risk - diastolic blood pressure relation is not continuous and strictly increasing. (2) Thebasic relation of risk to diastolic blood pressure is the same as we previously found forsystolic blood pressure, namely risk is constant to a threshold at the 70th percentilepressure (about 90 mm Hg) and steadily increases thereafter. (3) Rather than beingarbitrary, 90 mm Hg is a natural threshold for hypertension.(4) The J-curve phenomenonis a barely detectable effect, hovering on the boundary of statistical significance, that maycause a rise in risk for pressures less than 70 mm Hg. The weakness of this effect mayexplain why its detection is so elusive (5). Systolic blood pressure and diastolic bloodpressure are equivalent predictors of risk. (6) For the soft endpoint of cardiovasculardisease incidence there is a sharp jump at 90 mm Hg, with risk being constant to the leftof 90 mm Hg, again constant between 90 mm Hg and 104 mm Hg, and a increasingthereafter. The difference between this outcome and the ‘hard’ outcomes suggests thatthe presence or absence of hypertension may influence the diagnosis of cardiovasculardisease.3INTRODUCTIONBoth JNC VI 1 and WHO/ISH 2 support the view that the relation of both diastolic andsystolic blood pressure to endpoints such as death due to cardiovascular disease iscontinuous, strictly increasing, and with no lower bound (Fig. 1). This, they say, is basedon the preponderance of eveidence from epidemiological studies and randomized trials.Previous analysis from the Framingham study 3,4 was instrumental in propagating thisview of the risk - blood pressure relation, primarily by the use of linear logistic smoothingof the data. We previously reported that the relation between cardiovascular risks andsystolic blood pressure was not linear but has a threshold level 5. Risk was not increasedat any pressures except for those in the upper 30% of pressure for their age and sex. Wealso found that the use of 140 mm Hg as the universal division between normal andelevated systolic blood pressure was unjustified; it needed to be replaced by age and sexdependent cut-points.Unlike the case with systolic pressure, there has long been some opposition to thestrictly increasing model of cardiovascular risk with diastolic blood pressure. Starting withAndersen’s observation 6 that the same Framingham data considered here apparentlyshowed that the risk of cardiovascular disease decreases (rather than increases) withincreasing diastolic pressure to 89 mm hg, this controversy has centered on the “J-curveeffect”. That effect is an apparent increase in risk at low as well as high diastolic bloodpressure 6-21. This view is controversial, primarily because it is not consistently found.We reassessed the diastolic pressure data from the Framingham 18 year follow upreport 4 for the following outcomes: cardiovascular disease incidence, death due tocardiovascular disease, and overall mortality. Our primary goal was to determine thenature of the relation of these risks to diastolic blood pressure. Secondary goals were toexplain why the J-curve effect was so elusive, to see what justification there was for 90mm Hg being a universal cut-point for diastolic hypertension, and to find whether thediastolic blood pressure or systolic blood pressure was a better predictor of risk.We chose to reanalyze the Framingham 18-year follow up data because: (10 it wasaccurately gathered, (2) it was unconfounded by antihypertensive drug intervention, (3) itcontained women, (4) it contained older people.4MethodsThe Framingham data 4 are presented separately for each sex divided into three agegroups 45-54, 55-64, and 65-74 years. There are 10 blood pressure categories (Table 1)Unlike the systolic pressures, for persons aged 45-74, the distribution of diastolicpressures does not change much with age and is about the same in both sexes Forpractical purposes we can take these distributions to be the same.We first determined if the relation of overall and cardiovascular disease death todiastolic blood pressure is homogeneous across the six groups, i.e. if the additive modelholds. In that model the risk for a person is simply the sum of two effects, one based onthe group the person is in and the other from the person’s blood pressure. Our analyseswere carried out using both the specific rates and the direct group adjusted rate. Modelsfor the specific rates were viewed as sub-models of the additive model.We modeled the relation of both overall and cardiovascular death to diastolic bloodpressure by curves known as logistic splines. A logistic spline is a curve that results fromcontinuously joining two or more logistic curves. The points where two curves join arecalled knots and the curves are called segments. A logistic spline is parameterized by thelocation of the knots,