The J-Curve Phenomenon and theTreatment of HypertensionIs There a Point Beyond WhichPressure Reduction Is Dangerous?Lisa Farnett, PharmD; Cynthia D. Mulrow, MD, MSc; William D. Linn, PharmD; Catherine R. Lucey, MD; Michael R. Tuley, PhD

We critically appraised the medical literature to evaluate whether there is a pointbeyond which blood pressure reduction in hypertensive subjects is no longerbeneficial and possibly even deleterious. Thirteen studies that stratified cardio-vascular outcomes by level of achieved blood pressure in treated hypertensivesubjects who had been followed up for at least 1 year were critiqued by fourindependent reviewers. Data addressing population, protocol, and methodologi-cal characteristics were evaluated. Studies did not show a consistent J-shapedrelationship between treated blood pressure and stroke, but they did demon-strate a consistent J-shaped relationship for cardiac events and diastolic bloodpressure. The beneficial therapeutic threshold point was 85 mm Hg. We con-clude that low treated diastolic blood pressure levels, ie, below 85 mm Hg, areassociated with increased risk of cardiac events.

(JAMA. 1991;265:489-495)

THERE is ample scientific evidence toconclude that uncontrolled hyperten¬sion increases cardiovascular morbidityand mortality, and clinical trial datahave shown that lowering elevatedblood pressure levels decreases morbid¬ity and mortality. The question that re¬mains is: "To what level should bloodpressure be lowered to optimize treat¬ment?" Recently, some authors have fo¬cused on the dangers of lowering blood

pressure below certain levels and havesuggested that excessive reductions inblood pressure may explain why majorclinical trials have not shown greatereffects in reducing coronary artery dis¬ease.1"8 These authors have proposed a

"J-shaped curve" relationship betweenblood pressure and cardiac morbidityand mortality, whereby lowering bloodpressure below a critical point is no

longer beneficial and possibly even dele¬terious. Other authors have maintainedthe traditional premise of "the lower theblood pressure the better."4

Support for the traditional therapeu¬tic goal of "the lower the better" is basedlargely on the results of large observa¬tional studies and actuarial data. Al¬though these studies often includednormotensive subjects and stratifiedcardiovascular outcomes based on base¬line rather than treated levels of blood

pressure, they illustrate the controver¬sy addressed in this review. For exam¬

ple, data from the massive Build andBlood Pressure Study,5 the mortalitysurveillance study of the Multiple RiskFactor Intervention Trial screenees,6subsets of the Pooling Project,7 the Fra-mingham study,8 and the CoronaryDrug Project study,9 as well as datafrom several lesser known studies,10"15have shown positive linear relationshipswithout threshold points between base¬line blood pressure levels and cardio¬vascular events. In general, these stud¬ies have analyzed data using best fitsmooth curves between the points thatrelate blood pressure and cardiovascu¬lar events. These methods do not allowdetection of a J-curve as to reveal a

J-shaped curve individual points mustbe connected. Regardless of these limi¬tations, the magnitude and consistencyof these data have prompted many au¬thorities to accept low blood pressuretargets for hypertensive patients.

In fact, in a recent meta-analysis,MacMahon et al4 combined nine obser¬vational studies and concluded that"there is no evidence of any thresholdbelow which lower levels of blood pres¬sure were not associated with lower lev¬els of stroke and coronary heart dis¬ease." Their methods of analysis andresults have been criticized.16"18 Theanalysis was based on baseline ratherthan treated blood pressure levels, andfive observational studies that did notsupport a continuous linear relation¬ship were excluded. 19"2S Moreover, at

From the College of Pharmacy, The University ofTexas at Austin (Drs Farnett and Linn); the Division ofGeneral Internal Medicine, the Department of Medi-cine, The University of Texas Health Science Center atSan Antonio (Drs Mulrow and Lucey); and the GeriatricResearch, Education, and Clinical Center at the AudieL. Murphy Memorial Veterans Hospital at San Antonio(Tex) (Drs Mulrow and Tuley).

Reprint requests to Ambulatory Care (11C), Audie L.Murphy Memorial Veterans Hospital, 7400 Merton Mint-er Blvd, San Antonio, TX 78248 (Dr Mulrow).

by guest on July 18, 2010 www.jama.comDownloaded from

least five population-based observation¬al studies have shown increases indeaths at lower levels of diastolic bloodpressure. Additional support for a

J-shaped relationship comes from thereexamination by Anderson24 of datafrom the Framingham study. Althoughthese data initially seemed to support alinear relationship,8 Anderson pointedout that the original data were logisti-cally smoothed, and, therefore, the ex¬istence of a J-shaped relationship couldbe masked. He then reanalyzed the dataand clearly showed an increase in car¬diovascular events with diastolic bloodpressure levels below 85 to 98 mm Hg.

In summary, large observationalstudies have been used to support thepolicy that blood pressure should belowered as far as possible. The appro¬priateness of using these studies to settreatment goals for hypertensive pa¬tients is questionable, given their fre¬quent inclusion of normotensive sub¬jects, inappropriate analysis strategies,and classification of subjects based onbaseline blood pressure levels only. Inaddition, reanalysis of data from theFramingham study, as well as originaldata from several European studies,has suggested that even in observation¬al studies, the relationship betweenblood pressure and cardiovascular mor¬

bidity and mortality may not be linear.Hence, the specific question of this re¬view is: "Is there a point beyond whichblood pressure reduction in hyperten¬sive subjects is no longer beneficial andpossibly even deleterious?"METHODSData Acquisition

A search of the English medical lit¬erature was performed using theMEDLINE on-line computer databasefor the years 1966 through 1989. Termsfor the MEDLINE search included hy¬pertension, blood pressure, and mor-

bidity and mortality. Other sources ofdata included references identified frompertinent articles and written communi¬cations with international experts in thefield of hypertension.

A total of 478 articles were identifiedfrom the MEDLINE database search.Of the 478 articles, 134 were deemedpotentially relevant by two indepen¬dent reviewers. "Potentially relevant"was interpreted broadly to include anystudy that correlated any blood pres¬sure level with a cardiovascular or totalmortality outcome. The methods sec¬tions of these 134 articles and a 10%random sample of the remaining 344were reviewed by two reviewers to as¬sess whether studies met four specificselection criteria. First, study popula¬tions had to involve hypertensive sub-

jects who were followed up for at least 1year while undergoing antihyperten-sive treatment. Second, accepted out¬comes included stroke and myocardialinfarction incidence and total mortalityand death due to stroke and cardiacevents. Third, to detect a J-shaped rela¬tionship, outcomes had to be stratifiedby at least three treated blood pressurelevels and had to include at least onelevel with achieved pressures of lessthan 90 mm Hg. Fourth, prospective or

retrospective cohort and randomizedtrial designs were allowed.

Thirteen studies were identified thatmet selection criteria.25"37 (None of thesewere identified from the random sam¬

ple.) One other potentially relevantstudy38 was cited by experts and articlereferences, but published data from thisstudy, which met selection criteria,were unavailable. The remaining 121articles included 24 that provided dupli¬cate and supplementary data for the 13primary studies, one small study thathad less than 1 year of follow-up, and 96studies that failed multiple selection cri¬teria. None of the excluded studies hadoutcomes presented by treated bloodpressure levels (ie, all failed the thirdselection criteria).Data Extraction

Articles were abstracted by four in¬dependent reviewers (L.F., C.D.M.,W.D.L., and C.R.L.) using a standard¬ized form. The form addressed subjectpopulation characteristics, study proto¬col characteristics, whether a J-shapedrelationship was found and for what out¬come measures, and eight items thatwere used to generate a quality score.Items rated for this score were(1) whether the study population wasreferred or community based,(2) whether outcomes were assessed in a"blinded" manner, (3) whether out¬comes were assessed using an equal fol¬low-up time (or adjusted for unequalfollow-up times), (4) whether the per¬centage of the dropout rate was greateror less than 25%, (5) whether cointer-ventions were addressed, (6) whetherresults were adjusted for demographicfactors (age and sex) and comorbid dis¬eases or risk factors (preexisting is¬chémie disease, diabetes, hyperlipid-emia, and smoking status), (7) whetherprecardiovascular event vs postcardio-vascular event blood pressures were an¬

alyzed, and (8) whether statistical sig¬nificance was evaluated. Scores couldrange from 0, indicating poor quality, to15, indicating excellent quality. Thisquality score was weighted specificallyto address methodological issues rele¬vant to detect a J-shaped relationship.The scores do not necessarily reflect the

ability of studies to address their origi¬nal hypotheses. The overall weighted kvalue among the four assessors for theserankings was. 62.

Data SynthesisThe data were reviewed qualitatively

by specifically integrating study resultswith subject characteristics, study pro¬tocols, and study methods or quality.Quantitative estimates of the effects oftreated diastolic blood pressures onischémie heart disease events and mor¬

tality and total mortality were derivedusing least-squares regression modelsthat weighted studies by person-yearsof observation.39 (Longer studieswith more patients received greaterweights.) Models with higher-orderpolynomials were fit. Mallows C(P) sta¬tistic was used to determine the bestfitting model.39 The statistical signifi¬cance of the model and its parameterswere tested using F and t tests, respec¬tively. Cerebrovascular events werenot pooled because only three studies26"28gave these data separate from cardiacdata, and one of these27 was presentedby changes in blood pressure ratherthan by treated level stratifications.

RESULTSSubject and Protocol Characteristics

More than 48 000 subjects have beenincluded in the 13 studies (Table 1). Ap¬proximately 54% have been men. Mostsubjects have been middle-aged, al¬though some elderly subjects have beenstudied. No studies specifically ex¬cluded smokers or subjects with elevat¬ed cholesterol levels. Subjects often hadknown ischémie disease at entry, al¬though four studies were designed toinvestigate patients who were relative¬ly free of cardiovascular disease.30,33"31Most entry blood pressure levels havebeen in the mild to moderate range. Oneearly, very small study (N = 58) dealtexclusively with severely hypertensivesubjects; it is not presented in the tablesor discussed further.38

Approximately half of the studieswere designed as cohort studies and halfas randomized trials. Average follow-uplengths varied from 4 to 12 years. Bloodpressure measurements have been per¬formed in the sitting position usingphase V Korotkoff's sounds in all butone study.30 Antihypertensive agentsmost commonly used were diuretics,ß-blockers, or both, though some stud¬ies have included calcium channelblockers and angiotensin-convertingenzyme inhibitors. Most studies haverelated achieved diastolic blood pres¬sure levels to cardiovascular outcomes,though some have looked at achieved

by guest on July 18, 2010 www.jama.comDownloaded from

Table 1 .—Subject and Protocol Characteristics of Studies That Have Stratified Cardiovascular Outcomes by Levels of Achieved Blood Pressure*

IncludesMean Mean Entry Subjects With Mean Quality

No. of Age, % Diastolic Cardiovascular Follow-up BP BP Primary ScoreStudy Subjects_y_Men BP, mm Hg_Disease_Time, y_Phase Position_Drugs_(0-15)

Crulckshank et aP 939 55 61 109 Yes 6.1 V Sitting AB, TD, 5_PSD_DHCCP»1_2145_51_50_107_Yes_4_V_Supine_Unclear_4HDFP"_10053_51_54_90-104_Yes_4_V_Sitting_AB,TD_8_Waller et al28_3350_50_50_110_Yes_6j>_V_Unclear_Unclear_5HEP"_884_68_31_98_Yes_4A_V_Sitting_AB,TD_4Stewart"_169_44_71_124_No_&25_IV_Unclear_AB,TD_7NYEC31 1765 51 72 102 Yes 4.2 V Sitting AB, TD, 11_CCB,ACE_EWPHE"_840_71_30_90-119_Yes_-L7_V_Sitting_TD, PSD_7PPT33_686_52 100_106_Not_12_V_Sitting_AB,TD_14IPPPSH34_6357_52_50_108_No_3-5_V_Sitting_AB, PSD_5ANBP35_3931_50_55_101_Not_4_V_Sitting_TD_4_MRC36 17 354 52 52 90-109 Yes 5 5 V Sitting AB, TD 6

*BP indicates blood pressure; DHCCP, Department of Health and Social Security Hypertension Care Computing Project; HDFP, Hypertension Detection and Follow-upProgram; HEP, Hypertension in Elderly Patients; NYEC, the New York Employee Cohort Study; EWPHE, European Working Party on High Blood Pressure in the Elderly; PPT,Primary Prevention Trial; IPPPSH, International Prospective Primary Prevention Study in Hypertension; ANBP, Australian National Blood Pressure Study; MRC, MedicalResearch Council; AB, adrenerglc blocker; TD, thiazide diuretic; CCB, calcium channel blocker; ACE, angiotensin-converting enzyme inhibitor; and PSD, potassium sparingdiuretic.

tExcluded subjects with myocardial infarction or stroke within the past 2 years.tExcluded subjects with stroke and angina, as well as subjects with myocardial Infarction within the past 3 months.

Table 2.—Summary of Findings for and Against a J-Curve Relationship for Various Cardiovascular End Points*

Myocardial Infarction DiastolicStroke Stroke •-"-> Total Blood Pressure

Studyt Incidence Mortality Incidence Mortality Mortality J-Point, mm HgCruickshank etal»_.___PRO«_._^_PRO_CON_85-90DHCCP«_ _CON_.___PRO_CON_86-91HDFP"_.^_^_.„_„._PRO_26#Waller etal"_ _COT*_._^_PRO_CON_91-98HEP"_.___1_LJ_PROt_PRO_LJJ_80-89Stewart30_L__ _PROt_PROt_._^_100-109NYEC"_CON_CON_PRO_PRCj_._^_Aa17EWPHE3"_L_J_PRO§_ _PRO§_PRO_.. .**

PPT33_PROt§_PROt§_PROt_PROt_„._86-89IPPPSH34_ _.___PROtH_PROtll_._^__92ANBP35_PROt§ll_PROt§ll_PROt§ll_PROt§ll_^_85-89MRC36

... ...

CON... ...

Not shown

'PRO indicates findings that support the J-curve hypothesis; and CON, findings that oppose the J-curve hypothesis.tAbbreviations are expanded in a footnote in Table 1.¿Incidence and mortality data combined.§Cardiovascular and cerebrovascular data combined.||A flattened curve rather than an actual J-shaped curve was seen.«¡Shown for systolic blood pressure only.#Change in blood pressure from baseline to follow-up."Inverse relationship without actual threshold identified.

systolic blood pressure levels also.Most studies have presented cardiovas¬cular outcomes stratified by averageachieved treated blood pressure levels.Three studies have presented results bythe change in the level ofblood pressureachieved from initial readings to follow-up readings.27'30,31Methodological Considerations

To determine whether study resultscould be used to identify a treatmenttarget level of blood pressure, severalissues were considered. One of the mostimportant was whether measurementsof treated blood pressure were made

prior to ischémie events such as myocar¬dial infarctions. This is because lowblood pressure levels often result fromthese events and, thus, serve as a mark¬er of an underlying prognostic event. Inthese instances, low blood pressure lev¬els could be the result of ischémie eventsrather than the cause of future ischémieevents. Only three studies limited theiranalyses to blood pressure measure¬ments that were made prior to cardio¬vascular events.30,31,33

Adjustments for cardiovascular risksand treatment effects associated withcointerventions also were important.Levels of achieved blood pressure may

have varying implications depending onwhether a subject is a male smoker withan elevated cholesterol level comparedwith a female who does not smoke andwho has a normal cholesterol level.Likewise, a treated blood pressure of80 mm Hg with associated hypokalemiaand uncontrolled diabetes could have adifferent implication if the hypokalemiaand/or hyperglycemia were absent. Toassess the true impact of achieved bloodpressure control, these factors must beaddressed. Although several studiesadjusted their results for known base¬line cardiovascular risk factors, noneconsidered changes in risk factors or

by guest on July 18, 2010 www.jama.comDownloaded from

competing effects of treatment cointer¬ventions made during the course of thestudy.

Finally, observation times for sub¬jects either must be equal or must notvary with blood pressure control. If fol¬low-up time is shorter or longer in sub¬jects with good or poor control, the out¬come of interest could be related to thelength of follow-up rather than theachieved blood pressure. Three studiesformally addressed inequality in obser¬vation times.27,31'33

Study FindingsAn overview of study findings is giv¬

en in Table 2. Although some studieshave been interpreted previously asnot showing a threshold point and/orJ-shaped relationship, we did not relyon previous interpretations. Rather,judgments were made by consensus ofthe four raters after their evaluation ofthe data stratifications available in thearticles. Raters considered a thresholdpoint to be present if the relationshipbetween blood pressure and cardiovas¬cular outcomes was not linear and if thelowest stratification of achieved level ofblood pressure was associated with nodecrease in cardiovascular events. Noconsistent J-shaped relationship be¬tween treated blood pressure level andstroke incidence or mortality was seenin the studies. However, studies consis¬tently demonstrated a J-shaped rela¬tionship between treated diastolic bloodpressure and cardiac event incidenceand mortality (Fig 1). Only one study36failed to show such a relationship, whiletwo34,35 showed flattened curves at dia¬stolic blood pressures of approximately85 to 92 mm Hg rather than an actualJ-shaped or upswinging curve. Systolicblood pressure relationships often werenot addressed or gave conflictingresults.

Three studies that received the high¬est quality ratings and, thus, avoidedmany of the methodological pitfalls al¬ready outlined will be reviewed in de¬tail. The one study that failed to docu¬ment a J-shaped relationship will bedetailed. Finally, a quantitative analy¬sis of the cardiac event data will bepresented.

The first study is the Göteborg (Swe¬den) Primary Prevention Trial,33 whichinvolved 686 middle-aged men whowere followed up for 12 years. Subjectswere recruited from a random commu¬nity sample. They had mild to moderatehypertension and were treated initiallywith a thiazide, ß-blocker, or both. Re¬sults were adjusted forbaseline risk fac¬tors (age, smoking status, cholesterollevel, and baseline blood pressure), andonly precardiac event blood pressure

30 -|

25

20

15

gs io

•2(2—

o

II5H

o

Cruickshank et al!s

<85 85-90 >90Treated Diastolic Blood Pressure, mm Hg

a .

70 n

60

50oi c

b"t:CL|S 30Xoo »*-

E»O) Q.

40

20

Stewart10

<90 90-99 100-109110-119 >119Treated Diastolic Blood Pressure, mm Hg

wraoi

a>>en >,Pc_ë2*2isx°

30

25

20 H15

10-

5

Department of Health and SocialSecurity Hypertension Care

Computing Project2*

—I-1-I-1-1—<86 86-91 92-96 97-101 »101

Treated Diastolic Blood Pressure, mm Hg

ni _.

01_

_

01Ora__.

"o»s

E o>0) Cl

Primary Prevention Tr¡al3:

-|-I-1-1-1-<86 86-89 90-93 94-97 >97

Treated Diastolic Blood Pressure, mm Hg

S 30 noi>-ül '

«c 25-s|QP 20-*ix§(J ._

E oi

-c ****

15-

10-

5-

Waller et alM

—i-1-1-i-r*<86 86-90 91-98 99-106 >106

Treated Diastolic Blood Pressure, mm Hg

01 c« 01Q're.ojoXo

_

*~

ÛJ Q.

—

01>

LU

International ProspectivePrimary Prevention Study

in Hypertension'*

80 90 100—I—

110Treated Diastolic Blood Pressure, mm Hg

a 30 no 01

Sé 25-lS|to.

Xo

20-

15-

01 CL

LU

h 10-

5H

Hypertension inElderly Patients"

c80 80-90 •»90Treated Diastolic Blood Pressure, mm Hg

Summary Curve Model

i-1-1-1-1-1-1-1-1-165 70 75 80 85 90 95 100105110115Treated Diastolic Blood Pressure, mm Hg

Fig 1.—Studies that stratified cardiac events by treated diastolic blood pressure levels. Ischémie heartdisease events is a combination of morbidity and mortality. The summary curve was calculated using a

systematic decrease of 10 mm Hg In diastolic blood pressure levels in the study by Stewart because of hisuse of phase IV diastolic blood pressure measurements; 95% confidence intervals are shown by thesurrounding dashed curves.

by guest on July 18, 2010 www.jama.comDownloaded from

measurements were included. Differ¬ences in mean follow-up times for peoplewith varying levels of blood pressurecontrol were not statistically signifi¬cant. Changes in multiple risk factorsand associated cointerventions that oc¬curred during the 12-year follow-upwere not addressed. A J-shaped rela¬tionship that was not statistically signif¬icant was found when relating treateddiastolic and systolic blood pressureto cerebrovascular and cardiovascular

Large Moderate Small(>17mmHg)(7-17mmHg)(<7mm Hg1Change in Diastolic Blood Pressure

Fig 2.—J-curve relationship between incidence ofmyocardial infarction and change in diastolic bloodpressure (adapted from Alderman et al31).

events defined by combining incidenceand mortality data (Fig 1). The relation¬ship was evident in subjects both withand without preexisting coronary heartdisease. The nadir of the curve occurredat a treated diastolic blood pressure lev¬el of 86 to 89 mm Hg.

The second study is the New YorkEmployee Cohort study,31 which in¬volved 1765 middle-aged men and wom¬en with mild to moderate hypertension.In this study, subjects received a

stepped-care approach to treatment us¬

ing a thiazide, a ß-blocker, a calciumchannel blocker, or an angiotensin-con-verting enzyme inhibitor for initial ther¬apy. They were followed up for a meanof 4.2 years. Results were stratified ac¬

cording to change in blood pressureachieved. Varying baseline risks wereaddressed (age, sex, preexisting cardio¬vascular disease, race, smoking status,body mass index, cholesterol level,baseline blood pressure, and electrocar-diographic results). Only precardiacevent blood pressure measurementswere related to outcomes, and analysestried to account for variable lengths offollow-up. A statistically significantJ-shaped relationship was found be¬tween myocardial infarction incidenceand mortality and changes in diastolicblood pressure (Fig 2). A similar trendthat was not statistically significant

was observed for systolic blood pres¬sure changes. The J-shaped relation¬ship was observed in people both withand without preexisting cardiovasculardisease, but it was more marked in peo¬ple with known cardiovascular disease.It seemed to be independent of the typeofdrug therapy used. The nadir or turn¬ing point of the curve occurred in sub¬jects who had a moderate drop in bloodpressure level with treatment. Thismoderate drop ranged from 7 to17 mm Hg.

The third high-quality study is theHypertension Detection and Follow-upProgram,27 which is a randomized trialthat involved more than 10 000 middle-aged men and women with mild to mod¬erate hypertension. Subjects receivedeither stepped care or referred care fortheir hypertension. After following uppatients for 5 years, a statisticallysignificant J-shaped relationship wasshown for total mortality related to dia¬stolic blood pressure changes (Fig 3) butnot for systolic blood pressure changes.The nadir of the J-shaped curve oc¬curred at a 10- to 19-mm Hg drop indiastolic blood pressure. For peoplewith mild hypertension, minimum risksfor mortality occurred with a 10-mm Hgdrop. Data to show a J-shaped relation¬ship for cardiac event incidence and/ormortality were not given. Results were

Fig 3.—An Increase In mortality is shown In patients with low blood pressurelevels (baseline-annuals) enrolled in the Hypertension and Detection Follow-upProgram (adapted from Cooper et al ).

0.09

0.08 H

S 0.07 -

(0er

1 0.06-

S 0.05 -\

0.04 H

5 0.03 -\0.02-

0.01 -1-1-1-1-1-20.00-10.00 0.00 10.00 20.00 30.00 40.00 50.00

Change ¡n Diastolic Blood Pressure, mm Hg

Fig 4.—A J-curve relationship between mean treated blood pressure and deathfrom myocardial infarction is shown only for patients with preexisting heartdisease (adapted from Cruickshank et al ).

20-,

15-

«H

5*H

o-1

Nonischemic

<85—I-85-90

11>90

Mean Treated Diastolic Blood Pressure, mm Hg

by guest on July 18, 2010 www.jama.comDownloaded from

adjusted as necessary for unequal fol¬low-up times and for varying baselinerisks and comorbid diseases (age,race, sex, end-organ damage, diabetes,smoking status, ideal body weight,baseline blood pressure, and antihyper-tensive medication status). Some post-event blood pressure measurementswere used.

The only study that failed to confirm anonlinear relationship between treateddiastolic blood pressure and cardiacevents was the large Medical ResearchCouncil Trial, which involved 17 354male and female hypertensive sub¬jects.36 Propranolol and thiazide diure¬tics were the principle therapeuticagents, and subjects were followed upfor 5.5 years. Postevent blood pressurelevels and unequal follow-up times wereused. Although actual data were notprovided in the article, it was reportedthat there was no significant quadraticeffect for diastolic blood pressure in ac¬

tively treated subjects. The levels ofstratification of achieved diastolic bloodpressure that were used were unclear,but it seemed that all subjects withpressures less than 90 mm Hg were con¬sidered together. Graphs that plottedstroke rates against treated systolicblood pressures suggested a J-shapedrelationship in subsets of subjects whowere smokers.

Most studies, including those re¬viewed in detail above, stratified re¬sults according to demographic factors(ie, age and sex) and according to riskfactors (ie, smoking status). In general,it was impossible to identify whetherthe J-shaped relationship was specific tocertain subgroups such as men, smok¬ers, or middle-aged people. Some stud¬ies performed subset analyses limited tosubjects with preexisting heart disease.In these analyses, a J-shaped relation¬ship was found for treated blood pres¬sure and cardiac events.25'26,28'31'33 Thisfinding was most marked in the study byCruikshank et al,25 which found a

J-shaped curve only in patients withpreexisting ischémie disease (Fig 4).However, some of the studies haveshown a J-shaped curve, albeit oftenattenuated, even in subjects withoutpreexisting heart disease.26'28'30,31,33

Of particular note, four studies29,32'36'36that included placebo or control sub¬jects stratified results by treated anduntreated groups. All these studiesshowed either a J-shaped relationshipor a flattened curve in the untreatedgroups. The European Working Partyon High Blood Pressure in the Elderlystudy29 showed the J-shaped relation¬ship for total mortality and combinedcardiac and cerebrovascular mortalityin the placebo group. The Hypertension

in Elderly Patients study32 and the Med¬ical Research Council Trial36 had similarresults for cardiac mortality and myo¬cardial infarction in a control and place¬bo group, respectively. In the MedicalResearch Council Trial, the J-curve re¬

lationship was reported only in womenin the placebo group; it was not seen intreated subjects.

The graphs of the seven stud¬ies25,26'28-30,3334 that evaluated cardiacevents by levels of treated diastolicblood pressure are seen in Fig 1. (TheMedical Research Council Trial is notgraphed because actual diastolic bloodpressure data were not published.)Four studies26262829 evaluated ischémieheart disease mortality independentfrom morbidity, whereas three30,33,34 pre¬sented only combined event data. Thebest-fitting summary curves for boththe cardiac event (Fig 1) and cardiacmortality data were J shaped. The sum¬

mary cardiac event curve is givenin Fig 1 where person-time eventrates equaled -0.001085 diastolicblood pressure3+ 0.3286 diastolic bloodpressure2-32.26 diastolic blood pres¬sure +1039.5. All coefficients were sig¬nificant (P<.05), and the model ex¬

plained 64% (r2) of the variability in thedata (P<.05). The derived nadir of thecurve was 84 mm Hg. The predictedcardiac event incidence rates per 1000person-years were 10.6 and 5.2 at 75 and85 mm Hg, respectively. This repre¬sented a two times higher rate of eventsfor treated blood pressure levels of75 mm Hg compared with 85 mm Hg. Acalculated summary curve for the stud¬ies that presented total mortality databy treated blood pressure level was flat,with a plateau of 85 to 90 mm Hg. Ofnote, this summary curve excluded datafrom one study27 that supported aJ-curve because such data were pre¬sented only by change in treated bloodpressure.COMMENT

In summary, 13 studies that involvedmore than 48 000 subjects with treatedhypertension have stratified their out¬comes by achieved blood pressure lev¬els. These studies have shown thatthere is not a consistent J-shaped rela¬tionship between treated blood pres¬sure level and stroke, but that a consis¬tent J-shaped relationship is apparentfor cardiac events and treated diastolicblood pressure level. It is possible thatthis relationship is most marked in sub¬jects with preexisting cardiac disease.In addition, the J-shaped relationshiphas been observed consistently in con¬trol subjects in these studies. This sug¬gests that the J-curve phenomena,whether occurring biologically or thera-

peutically, may explain increased cardi¬ac mortality. More important, it sug¬gests that the J-curve is "probablyindependent of treatment and that itreflects a relation between cardiacevents and absolute diastolic bloodpressure with treatment merely shift¬ing ischémie hypertensive subjectsfrom a fairly safe part to a less safe partof the curve."40

The exact mechanism by which low¬ering blood pressure levels beyond acritical point could increase cardiovas¬cular complications is unknown. How¬ever, several pathophysiological mech¬anisms have been proposed.2 First, as

many as 50% of hypertensive patientshave left ventricular hypertrophy. Ahypertrophied ventricle has increasedmyocardial oxygen consumption com¬

pared with a normal ventricle. If a de¬crease in coronary flow occurs as a re¬sult of lowered diastolic blood pressure,a hypertrophied ventricle may becomeischémie before a normal ventricle andmay be prone to ischémie events. Sec¬ond, in the coronary circulation, oxygenextraction at baseline may be at nearmaximum capacity. In low flow situa¬tions the heart may not compensate byincreasing oxygen extraction. Third,autoregulation, the primary defensemechanism in patients with coronaryartery disease, is adversely affected bylow diastolic blood pressures. Normal¬ly, when a large coronary artery be¬comes stenotic, the smaller distal arter¬ies dilate or autoregulate. This distaldilation decreases resistance and im¬proves flow. If coronary perfusion pres¬sure is lowered by overaggressive hy¬pertensive therapy, the vessels may notfurther autoregulate and ischemia maydevelop. Fourth, excessive lowering ofthe diastolic blood pressure leading tolow coronary flow may cause increasedblood viscosity and increased plateletadhesiveness that could lead to throm¬bus formation, a keystone event in thedevelopment of myocardial infarction.Last, areas of well-perfused myocardi¬um adjacent to areas of ischémie myo¬cardium precipitated by low diastolicblood pressures may cause a buildup ofmetabolic gradients, giving rise to ven¬tricular dysrrhythmias.

Whether the data examined in thisreview can be used to infer an optimaltreatment goal in hypertensive subjectsis debatable. The studies were not origi¬nally designed to specifically establish atreatment goal. No clinical trial hasbeen performed that has randomizedhypertensive subjects to study armswith different treatment targets, al¬though one such trial is in progress.41Thus, the data considered in this revieware subject to various limitations often

by guest on July 18, 2010 www.jama.comDownloaded from

inherent in any secondary data analysis.Although several of the studies at¬tempted to control for potential limita¬tions, most did not control for metabolicside effects of therapy such as hypokale¬mia, hyperlipidemia, and hyperglyce-mia, and none controlled for changes inrisk factors or for associated cointer-ventions that could have affected car¬diovascular outcomes. These factorscould account for the observed J-shapedrelationships and, thus, could serve asconfounders in using the study findingsto determine treatment goals.

In summary, it is clear from random¬ized trials that treating hypertensionreduces cardiovascular events. Wheth¬er there is a threshold point beyondReferences

which subjects should not be treated isnot established. However, cliniciansmust make decisions even in the face ofimperfect data. Lacking a randomizedtrial that evaluates different treatmentthresholds, we can and should take ad¬vantage of existing information to guideus in our therapeutic decision making.As existing data do not suggest a consis¬tent J-shaped relationship betweenstroke and treated blood pressure, it isreasonable to consider reducing bloodpressure to conventional levels (70 to90 mm Hg) to prevent strokes and theirattendant mortality. Unfortunately,such reductions may be deleterious inprecipitating cardiac events, particular¬ly in patients with known cardiac dis-

ease. One may then be faced with theuncomfortable choice of whether toattempt to prevent cerebrovascularevents at the expense of cardiac eventsor vice versa. As increased cardiac mor¬

bidity and mortality related to bloodpressure treatment, if it occurs, may bemost marked in subjects with preexist¬ing cardiac disease, a reasonable cur¬rent compromise is to be cautious inlowering blood pressure levels below85 mm Hg in patients with known isch¬émie heart disease. The prudence of thistactic is accentuated by the conspicuouslack of evidence of benefit for therapeu-tically lowering blood pressure levelsbeyond this threshold.

1. Cruickshank JM, Pennert K, Sorman AE, et al.Low mortality from all causes, including myocardi-al infarction, in well-controlled hypertensivestreated with a beta-blocker plus other anti-hyper-tensives. JHypertens. 1987;5:489-498.2. Berglund G. Goals of hypertensive therapy: isthere a point beyond which pressure reduction isdangerous? Am J Hypertens. 1989;2:586-593.3. Kaplan NM, Alderman MH, Flamenbaum DA, etal. Guidelines for the treatment of hypertension.Am J Hypertens. 1989;2:75-77.4. MacMahon S, Peto R, Cutler J, et al. Bloodpressure, stroke, and coronary heart disease, I:prolonged differences in blood pressure: prospec-tive observational studies corrected for the regres-sion dilution bias. Lancet. 1990;335:765-774.5. Society of Actuaries and Association of Life In-surance Medical Directors of America. Build andBlood Pressure Study. Chicago, Ill: Recording andStatistical Corp; 1979.6. Kannel WB, Neaton JD, Wentworth D. Overalland coronary heart disease mortality rates in rela-tion to major risk factors in 325,348 men screenedfor the MRFIT.Am HeartJ. 1986;112:825-836.7. Paul O. Risks of mild hypertension: a ten-yearreport. Br Heart J. 1971;33(suppl):116-121.8. Kannel WB, Gordon T, Schwartz MJ. Systolic vsdiastolic blood pressure and risk of coronary heartdisease. Am J Cardiol. 1971;27:335-346.9. Coronary Drug Project Research Group. Bloodpressure in survivors of myocardial infarction.JAm Coll Cardiol. 1984;4:1135-1147.10. Rabkin SW, Mathewson FAL, Tate RB. Pre-dicting risk of ischemic heart disease and cerebro-vascular disease from systolic and diastolic bloodpressures. Ann Intern Med. 1978;88:342-345.11. LeHeuzy J, Guize L. Cardiac prognosis in hy-pertensive patients: incidence of sudden death andventricular arrhythmias. Am J Med. 1988;84(suppl1B):65-68.12. Svensson GI, Lundstrom A. Hypertension andmortality. J Chronic Dis. 1982;35:165-172.13. Taguchi J, Freis ED. Partial vs complete con-trol of blood pressure in the prevention of hyperten-sive complications. Circ Res. 1975;37(suppl 1):1257\x=req-\1260.14. Trafford AP, Horn CR, O'Neal H, et al. Five\x=req-\year follow-up of effects of treatment of mild tomoderate hypertension. BMJ. 1981;282:1111-1113.

15. Nobrega FT, Morrow GW, Smoldt RK, OffordKP. Quality assessment in hypertension: analysisof process and outcome methods. N Engl J Med.1977;296:145-148.16. Cruickshank JM, Fox K, Collins P. Meta-analy-sis of hypertension treatment trials. Lancet.1990;235:1092.17. Alderman MH. Meta-analysis of hypertensiontreatment trials. Lancet. 1990;235:1092-1093.18. Kaplan NM. Meta-analysis of hypertensiontreatment trials. Lancet. 1990;235:1093.19. Aromaa A. Blood pressure level, hypertensionand five-year mortality in Finland. Acta MedScand. 1981;646:43-50.20. Miall WE. Systolic or diastolic hypertension:which matters most? Clin Exp Hypertens A. 1982;A4:1121-1131.21. Holme J, Waaler HJ. Five-year mortality in thecity of Bergen, Norway, according to age, sex, andblood pressure. Acta Med Scand.1976;200:229-239.22. Mattila K, Haavisto M, Rajala S, HeikinheimoR. Blood pressure and five-year survival in thevery old. BMJ. 1988;296:887-889.23. Ashcroft MT, Desai P. Blood pressure and mor-

tality in a rural Jamaican community. Lancet.1978;2:1167-1170.24. Anderson TW. Reexamination of some of theFramingham Blood Pressure Data. Lancet. 1978;2:1139-1141.25. Cruickshank JM, Thorpe JM, Zacharias FJ.Benefits and potential harm of lowering high bloodpressure. Lancet. 1987;1:581-584.26. Fletcher AE, Beevers DG, Bulpitt CJ, et al.The relationship between a low treated blood pres-sure and IHD mortality: a report from the DHSSHypertension Care Computing Project. J HumHypertens. 1988;2:11-15.27. Cooper SP, Hardy RJ, Labarthe DR, et al. Therelation between degree of blood pressure reduc-tion and mortality among hypertensives in the Hy-pertension Detection and Follow-up Program. AmJ Epidemiol. 1988;127:387-402.28. Waller PC, Isles CG, Lever AF, Murray GD,McInnes GT. Does therapeutic reduction of diastol-ic blood pressure cause death from coronary heartdisease? J Hum Hypertens. 1988;2:7-10.29. Coope J, Warrender TS. Randomized trial oftreatment of hypertension in elderly patients inprimary care. BMJ. 1986;293:1145-1148.

30. Stewart IMcDG. Relation of reduction in pres-sure to first myocardial infarction in patients re-

ceiving treatment for severe hypertension. Lancet.1979:1:861-865.31. Alderman MH, Ooi WL, Madharan S. Treat-ment induced blood pressure reduction and the riskof myocardial infarction. JAMA. 1989;262:920-924.32. Staessen J, Bulpitt C, Clement D, et al. Rela-tion between mortality and treated blood pressurein elderly patients with hypertension: report of theEuropean Working Party on High Blood Pressurein the Elderly. BMJ. 1989;298:1552-1556.33. Samuelsson 0, Wilhelmsen L, Anderson OK,Pennert K, Berglund G. Cardiovascular morbidityin relation to change in blood pressure and serumcholesterol levels in treated hypertension: resultsfrom the Primary Prevention Trial in G\l=o"\teborg,Sweden. JAMA. 1987;258:1768-1776.34. The IPPPSH Collaborative Group. Cardiovas-cular risk and risk factors in a randomized trial oftreatment based on the beta-blocker oxprenolol:the International Prospective Primary PreventionStudy in Hypertension (IPPPSH). J Hypertens.1985;3:379-392.35. Report by the Management Committee of TheAustralian Therapeutic Trial in Mild Hypertension:untreated mild hypertension. Lancet. 1982;1:185-191.36. Medical Research Council Working Party.Stroke and coronary heart disease in mild hyper-tension: risk factors and the value of treatment.BMJ. 1988;296:1565-1570.37. Pererra GA. Antihypertensive drug vs symp-tomatic treatment in primary hypertension: effecton survival. JAMA. 1960;173:11-13.38. Wilhelmsen L, Berglund, G, Elmfeldt D. Beta\x=req-\blockers versus diuretics in hypertensive men:main results from the HAPPHY trial. JHypertens.1987;5:561-572.39. Neter J, Wasserman W, Kutner MH. AppliedLinear Statistical Models. 2nd ed. Homewood, Ill:Richard D. Irwin Inc; 1985.40. Coope J. Hypertension: the cause of theJ-curve. J Hum Hypertens. 1990;4:1-4.41. The BBB Study Group. The BBB Study: aprospective randomized study ofintensified antihy-pertensive treatment. J Hypertens. 1988;6:693-697.

by guest on July 18, 2010 www.jama.comDownloaded from