Stroke 1994 D'Agostino 40 3

8/12/2019 Stroke 1994 D'Agostino 40 3

1/5

R B D'Agostino, P A Wolf, A J Belanger and W B KannelStroke risk profile: adjustment for antihypertensive medication. The Framingham Study.

Print ISSN: 0039-2499. Online ISSN: 1524-4628Copyright 1994 American Heart Association, Inc. All rights reserved.

is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231Strokedoi: 10.1161/01.STR.25.1.40

1994;25:40-43Stroke.

http://stroke.ahajournals.org/content/25/1/40World Wide Web at:

The online version of this article, along with updated information and services, is located on the

http://stroke.ahajournals.org//subscriptions/

is online at:StrokeInformation about subscribi ng toSubscriptions: http://www.lww.com/reprints

Information about r eprints can be found online at:Reprints:

document.Permissions and Rights Question and Answeravailable in thePermissions in the middle column of the Web page under Serv ices. Further information about this process isOnce the online version of the published article for which permission is being requested is located, click Request

can be obtained via RightsLink, a service of the Copyright Clearance Center, not the Editorial Office.Stroke Requests for permissions to reproduce figures, tables, or portions of articles originally published inPermissions:

by guest on November 14, 2013http://stroke.ahajournals.org/ Downloaded from by guest on November 14, 2013http://stroke.ahajournals.org/ Downloaded from
http://stroke.ahajournals.org/content/25/1/40http://stroke.ahajournals.org//subscriptions/http://stroke.ahajournals.org//subscriptions/http://stroke.ahajournals.org//subscriptions/http://www.lww.com/reprintshttp://www.lww.com/reprintshttp://www.lww.com/reprintshttp://www.ahajournals.org/site/rights/http://www.ahajournals.org/site/rights/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org//subscriptions/http://stroke.ahajournals.org//subscriptions/http://www.lww.com/reprintshttp://www.lww.com/reprintshttp://www.ahajournals.org/site/rights/http://www.ahajournals.org/site/rights/http://stroke.ahajournals.org/content/25/1/40


2/5

40

Stroke Risk Profile: Adjustmentfor Antihypertensive MedicationThe Framingham Study

Ralph B. D Agostino, PhD ; Philip A. Wolf, MD ;

Albert J. Belanger MA; William B. Kannel, MD

Background and Purpose We sought to modify existingsex-specific health risk appraisal functions (profile functions)for the prediction of first stroke that bett er assess th e effects ofth e use of antihypertensive medication.

Methods Hea lth risk appraisal functions were previouslydeveloped from the Framingham Study cohort. These func-tions were Cox proportional hazards regression models relat-ing age, systolic blood p ressu re, dia betes mellitus, cigarettesmoking, prior cardiovascular disease, atrial fibrillation, leftventricular hypertrophy by electrocardiogram, and the use ofantihypertensive medication to the occurrence of stroke.Closer examination of the data indicated that antihypertensivetherapy effect is present only for systolic blood pressuresbetween 110 and 200 mm Hg. Adjustments to the regressions

to better fit the observed data were developed and tested forstatistical significance and goodness-of-fit of the modelresiduals.

Results Modified functions more consistent with the datawere developed, and, from these, tables to evaluate 10-yearrisk of first stroke were computed.

Conclusions The stroke profile can be used for evaluationof the risk of stroke and suggestion of risk factor modificationto reduce risk. The effect of antihypertensive therapy in theevaluation of stroke risk can now be better evaluated. (Stroke.1994;25:40-43.)

Key Words antihypertensive agentsdisorders risk factors

cerebrovascular

Wolf et al 1 presented sex-specific health risk

appra isal functions (or profile functions)relating risk factors including systolic blood

pressure and the use of antihypertensive medication tothe probability of developing a stroke. Those functionswere Cox proportional hazards regression models. Formen there was an elevation of risk for those on antihy-pertensive medication. For women there was an eleva-tion of risk and also an interaction of systolic bloodpressure level and being on antihypertensive medica-tion. In both sexes these effects were presented asapplying to all levels of systolic blood pressures. Closerexamination of the data indicated that an added effect ispresent only for systolic blood pressures between 110and 200 mm Hg. Modifications to the h ealth risk ap-praisal functions were developed to take this into ac-count. These modifications are more consistent with thedata and supply better statistical fits to the data. In thefollowing the modified functions are presented alongwith tables that can be used to estimate 10-year prob-abilities of first stroke.

Received July 28, 1993; final revision received September 29,1993; accepted September 29, 1993.

From the Department of Mathematics (R .B.D., A.J.B.), Collegeof Liberal Arts, and the Department of Neurology (P.A.W.),School of Medicine, Boston University; and the Section of Pre-ventive Medicine and Epidemiology (P.A.W., W.B.K.), EvansMemorial Department of Clinical Research and Department ofMedicine, University Hospital, Boston, Mass.

Reprint requests to Ralph B. D'Agostino, PhD, Boston Univer-sity, Mathematics Department, 111 Cummington St, Boston, MA02215.

Subjects nd M ethodsThe data used were the same as those used to develop the first

functions.1 Health risk appraisal functions were developed usingsubjects aged 55 to 84 years, free of stroke at the time of twoexamination cycles (examinations 9 and 14), and followed for 10years from each of these examinations. Stroke risk factorsincluded age, systolic blood press ure, use of antihypertensivetherapy, diabetes mellitus, cigarette smoking, prior cardiovascu-lar disease (coronary heart disease, cardiac failure, or intermit-tent claudication), atrial fibrillation, and left ventricular hyper-trophy by electrocardiogram. All these factors are the same aspreviously u sed and are defined elsewhere .12

The Cox proportional hazards regression model was used asthe model for the stroke profile.3 In the analysis of Wolf et al, 1

use of antihypertension medication was entered into theanalysis as a dummy variable, and the interaction of it withsystolic blood pressure was investigated as the product of thedummy variable for the antihypertensive medication and sys-

tolic blood pressure. More consistent with the data is thereplacement of the latter two variables w ith a variable of theform

NEWHRXSBP=HRX {SBP- 110)*(200-S5/>)

for systolic blood pressures between 110 and 200 mm Hg. HereHRX is a dummy variable defined as 1 if the individual is onantihypertensive medication and 0 if not, and SBP is theindividual's systolic blood pressure. Th e variable NE-WHRXSBP equals 0 for systolic blood pressures below 110 orgreater than 200 mm Hg.

The effect of this variable on the stroke profile is illustrated inthe F igure. For women the regression coefficient for SBP in thenew stroke profile is 0.0161, and for NEWHRXSBP it is 0.00026.For systolic blood pressures below 110 mm Hg or above 200

mm Hg, the stroke profile function gives the same weight toindividuals whether they are on or off antihypertensive medica-tion, and it is 0.0\6\*SBP. For those with systolic blood pres-

by guest on November 14, 2013http://stroke.ahajournals.org/ Downloaded from
http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/http://stroke.ahajournals.org/


3/5

3 T

ontri ution

* not on iretf cation

- on media t ion

- m ediati on has no effect

80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240

Systolic Blood Pressure (SBP)

D Agostino et al S troke Risk Profile 41

Graph shows contribution of systolic blood pressure(SBP) on female stroke profile function. The contri-bution is 0.0161 SBP for wom en w ith SBP below

110 mm Hg or above 200 mm Hg regardless of theirantihypertensive medication use and for those withSBP between 110 and 200 mm Hg if they are not onantihypertensive m edication. The contribution is0.0161 *SBP+0.000 26*(SBP-11 0)*(200-SBP) forthose with SBP between 110 and 200 mm Hg andalso on antihypertensive medica tion.

sures between 110 mm Hg and 200 mm Hg and not on medica-tion, the contribution of the blood pressure to the s troke profileis 0.0161*55/*, while for those with systolic blood pressuresbetween 110 mm Hg and 200 mm Hg and on medication thecontribution is 0.0161*SBP+0.00026*(SBP-110)*(200-SBP).

The Figure depicts the contribution to the stroke profile forsystolic blood pressures between 80 mm Hg and 240 mm Hg.

The values 110 mm Hg and 200 mm Hg were selected foruse in the NEWHRXSBP variable because most systolic bloodpressures of those on antihypertensive medication fell in thatinterval (96 ) and because they repres ented discontinuities inthe frequency distributions. The selection of other values didnot improve the fit to the data.

The Statistical Analysis System's Cox proportional hazardsmodel procedure, PROC PHREG,4 was used for analysis. Modelsusing the antihypertensive therapy variable and the interaction ofit with systolic blood pressure levels were contrasted with modelsusing the NEWHRXSBP variable, for both statistical significanceand goodness-of-fit as judged by analyses of the m odels'residuals.4

ResultsDuring 10 y ears of follow-up from examinations 9 and

14, 472 stroke events occurred in 2372 men and 3362women. Descriptive statistics of these are given in theprevious article.1 The new variable NEWHRXSBP wasstatistically significant for both sexes in Cox regressionmodels that included all the risk factors listed aboveexcept for antihypertensive medication (/'=.0006 forwomen and />=.O5 for men).

For both the female and male models the addition ofthe dummy variables for antihypertensive medication andthe interaction of this with systolic blood pressure did notadd significantly to the regressions that already included

NEWHRXSBP. This latter variable was sufficient to re -place them. Further, analysis of the residuals indicatedthat the use of NEWHRXSBP provided the best mod el fitto the data. F inally, selection of v lues other than 110 and200 mm Hg for NEWHRXSBP did not improve statisticalsignificance and model fit. Table 1 presents the Coxregression coefficients and their standard errors, relativerisks, and 95 confidence intervals for the risk factors ofthe new hea lth risk appraisal functions. Tables 2 and 3 givethe probabilities of stroke within 10 years for men andwomen aged 55 to 84 years and free of previous stroke,respectively, com puted from these functions. Table 4 givesthe average 10-year probabilities of first stroke by agegroups for each sex. Table 4 can be used for comparison

with probabilities computed from Tables 2 and 3. Thesetables replace tab les of the previous article1 as follows: th enew Table 1 here replaces the previous Table 3, Table 2

replaces the previous Table 5, Table 3 replaces the previ-ous Table 6, and th e present Table 4 replaces the previousTable 4.

As an illustration, say a 70-year-old woman presentsherself to a physician with the following profile: systolicblood pressure of 135 mm Hg, on antihypertensivetherapy, nondiabetic, cigarette smoker, previous history

TABLE 1 . Regress ion Coefficients, Relative Risk s, andConfidence Intervals for Risk Factors in CoxProportional Haza rds Regres sions for Stroke Profile inSubjec ts Aged 55-84 Years and Free of S t roke a tExaminations 9 and 14

Variable

Me n

Age, y*

SBP, mm Hg*

NEWHRXSBPCVD

LVH

Cigs

AF

Diabetes

Women

Age, y*

SBP, mm Hg*

NEWHRXSBP

CVD

LVH

Cigs

AF

Diabetes

0.0488

0.0152

0.00019

0.5460

0.7864

0.5224

0.5998

0.3429

0.0699

0.0161

0.00026

0.4404

0.8055

0.5419

1.1173

0.5604

SE

0.0103

0.0031

0.00010

0.0151

0.2846

0.1429

0.3011

0.1894

0.0089

0.0024

0.00007

0.1462

0.2429

0.1453

0.2302

0.1706

RR

1.63

1.16

N/A1.73

2.20

1.69

1.82

1.41

2.01

1.17

N/A

1.55

2.24

1.72

3.06

1.75

95 Cl

1.33-1.99

1.10-1.24

1.68-1.78

1.26-3.84

1.27-2.23

1.01-3.29

0.97-2.04

1.69-2.40

1.12-1.23

1.17-2.07

1.39-3.60

1.29-2.29

1.95-4.80

1.25-2.45

RR indicates relative risk; Cl, confidenc e interval; and N/A, notapplicable. Variables were defined as follows: SBP, systolicblood pressure; NEWHRXSBP =0 if SBP 20 0,else, =HR X*(SB P-11 0)*(200 -SBP ) where HRX=1 if on medi-catio n, 0 if not; CVD (cardiovascular disea se), history of myoca r-dial infarction, angina pecto ris, corona ry insufficiency, intermit-tent claudication, or congestive he art failure (yes =1 , no =0 );LVH, left ventricular hypertrophy o n electrocardiogram (ye s= 1,no=0); Cigs, smokes cigarettes (yes=1, no=0); AF, history ofatrial fibrillation (yes=1, no=0); Diabetes, history of diabetes(yes=1, no=0) .

*For age and SBP, the RR is computed for an increase of 10units. All other variables compare presence vs absence.



4/5

42 Stroke Vol 2 5, No 1 January 1994

TABLE 2. Probability of Stroke Within 10 Years for Men Aged 5 5-8 5 Years and Free of Previous Stroke in theFramingham Heart Study

Age, yUntreated SBP,mm Hg

Treated SBP,mm Hg

Diabetes

Cigs

CVD

AF

LVH

Points

1

2

3

4

5

6

7

8

9

10

0 + 1

54-56 57-59

97-105 106-115

97-105 106-112

No

No

No

No

No

10-YearProbability,

3

3

4

4

5

5

6

7

8

10

+2

60-62

116-125

113-117

Yes

Points

11

12

13

14

15

16

17

18

19

20

+3

63-65

126-135

118-123

Yes

10-YearProbability,

11

13

15

17

20

22

26

29

33

37

Points

+ 4 + 5

66-68 69-72

136-145 146-155

124-129 130-135

Yes

Ye s

Yes

Points

21

22

23

24

25

26

27

28

29

30

+6 +7 +8 +9 + 1 0

73-75 76-78 79-81 82-84 85

156-165 166-175 176-185 186-195 196-205

136-142 143-150 151-161 162-176 177-205

10-YearProbability,

42

47

52

57

63

68

74

79

84

88Variables were defined as follows: SBP, systolic blood pressure; Diabetes, history of diabetes; Cigs, smokes cigarettes; CVD

(cardiovascular disea se), history of myoca rdiai infarction, angina p ectoris, coronary insufficiency, intermittent ciaud ication, or con gestiveheart failure; AF, history of atrial fibrillation; LVH, left ventricular hypertrophy on electrocardiogram.

of cardiovascular disease, no atrial fibrillation present,and no left ventricular hypertrophy. Using Table 3, thiswoman receives 5 points for being aged 70, 6 points forhaving a systolic blood press ure of 135 mm H g while onantihypertensive therapy, no points for diabetes, 3points for smoking, 2 points for history of cardiovasculardisease, no points for the presence of atrial fibrillation,and no points for left ventricular hypertrophy. Thisperson therefore receives a total of 16 poin ts. A score of

16 points yields a 10-year stroke probab ility of 19 .Relatin g this 19 to the average probability for awomen this age provides perspective; the average 10-year probability of stroke for a 70-year-old women is10.9 (Table 4). Thus, this hypothetical woman has astroke risk that is 1.7 times the average. Her physiciancan urge her to stop smoking, which would result in atotal of 13 points and a probability of 11 for a strokerisk of 1.0 relative to the average risk.

Computation of the risk for the aforementioned70-year-old woman by the previous function1 produceda probability of 23 , which is a difference of 4 fromthe new estimate. Part of this difference is due torounding in generating Table 3 but also reflects a slight

overestimate by the previous function for women onantihypertensive medication. The exact estimates pro-duced directly by the Cox regressions are 20.2 for the

new function and 23.2 for the old function, accountingfor 3 of the 4 difference. Ro und ing prod uced bygenerating the tables accounts for 1 of the difference.

DiscussionIn a previous article1 we discussed in detail the

importance of control and intervention of the riskfactors of stroke and described in detail how to use thestroke profiles or health risk appraisal functions. The

focus of this article has been the adjustm ent for antihy-pertensive medication. It involved determining the re-gion of blood pressures attained by those on antihyper-tensive medication and then approximating the riskprofiles for these subjects. While the adjustments madefor antihypertensive medication in this article do notdiffer appreciably from those generated by the previousfunction, they are more consistent with the data andreflect a substantive advancement in the generation ofhealth risk appraisal functions. These functions havetraditionally ignored the effects of treatments, assumingthat, for example, blood pressures attained on medica-tion are equivalent to uncontrolled blood pressures.This is not the case with our data. Further health risk

appraisal functions will have to explore and investigatetreatments not only for blood pressure but for otherimportant risk factors such as cholesterol and glucose.



5/5

D Agostino et al Stroke Risk Profile 43

TABLE 3. Probability of Stroke Within 10 Years for Wom en Aged 5 5-8 4 Years and Free of Previous Stroke in theFramingham Heart Study

Age, yUntreated SBP,m m H g

Treated SBP,m m H g

Diabetes

Cigs

CVD

AF

LVH

Points

1

2

3

4

5

6

7

8

9

10

0 + 1

54-56 57-59

95-106

95-106

No

No

No

No

No

10-YearProbability,

1

1

2

2

2

3

4

4

5

6

+2

60-62

107-118

107-113

Ye s

Points

11

12

13

14

15

16

17

18

19

20

+3

63-64

119-130

114-119

Yes

Yes

10-YearProbability,

8

9

11

13

16

19

23

27

32

37

Points

+ 4 + 5

65-67 68-70

131-143 144-155

120-125 126-131

Ye s

Points

21

22

23

24

25

26

27

+6 +7 +8 +9 +10

71-73 74-76 77-78 79-81 82-84

156-167 168-180 181-192 193-204 205-216

132-139 140-148 149-160 161-204 205-216

Yes

10-YearProbability,

43

50

57

64

71

78

84

Variables were defined as follows: SBP, systolic blood pressure; Diabetes, history of diabetes; Cigs, Smokes cigarettes; CVD(cardiovascular disease), history of myocardial infarction, angina pectoris, coronary insufficiency, intermittent ciaudication, or congestiveheart failure; AF, history of atrial fibrillation; LVH, left ventricular hypertrophy on electrocardiogram.

Finally, in our previous article we discussed in detailhow to use the health risk appraisal functions forestimates different than 10 years. The instructions giventhere are appropriate here except that the antihyper-tension variable and the interaction of this with systolic

TABLE 4. Average 10-Year Probability of StrokeAccording to Age in Men and Women

Age Group, y Men Women

4.7

7.2

10 9

15.5

23.9

6.5

Values are percentages.

55-59

60-64

65-69

70-74

75-79

80-84

Age-adjusted

5.9

7.8

11.0

13.7

18.0

22.3

9.6

blood pressure level are replaced by the new variableNEWHRXSBP described above.

AcknowledgmentsThis study was supported in part by National Institute of

Neurological Disorders and Stroke grant 2-RO1-NS-17950-09(Dr Wolf), National Heart, Lung, and Blood Institute grantRO1-H140423-04 (Dr D'Agostino), and National Heart, Lung,and Blood Institute contract NIH-NO1-HC-38038 (Dr Wolf).

References1. Wolf PA, D'Agostino R B, Belanger AJ, Kannel WB. Probability of

stroke: a risk profile from the Framingham Study. Stroke. 1991;22:312-318.

2. Cupples LA, D'Agostino RB. Some risk factors related to theannual incidence of cardiovascular disease and death using pooledrepeated biennial measurements: Framingham Heart Study,30-year follow-up. In: Kannel WB, Wolf PA, Garrison RJ, eds. Th eFramingh am Study: An Epidemiologica l Investigation of Cardio-vascular Disease, Section 34. Bethesda, Md: National Heart, Lung,and Blood Institute publication NIH 87-2703; 1987.

3. Kalbfleisch JD, Prentice RL. T he Statistical Analysis of Failure TimeData. New York, NY: John Wiley & Sons, Inc; 1980.

4. SAS Institute Inc. SA S Supplemental Library User s Guide, Version 5Edition. Cary, NC: SAS Institute Inc; 1986.


Documents

Stroke 1994 D'Agostino 40 3