Risk of Arrhythmia and Sudden Death pdf

● the importance of the different risk factors and envi-ronment in the appearance of SCD.1–3 Population subgroups and sudden cardiac death According to Myerburg,13 if we consider all cas

Risk of Arrhythmia

and Sudden Death

Edited by Marek Malik

BMJ Books

All rights reserved No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in anyform or by any means, electronic, mechanical, photocopying, recording and/or otherwise, without the prior writtenpermission of the publishers.

First published in 2001

by BMJ Books, BMA House, Tavistock Square,

London WC1H 9JR

www.bmjbooks.com

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

ISBN 0-7279-1581-9

Typeset by Phoenix Photosetting, Chatham, Kent

Printed and bound by Creative Print and Design Ltd

Part I: Problem and methodology

Roberto Elosua, Josep Guindo, Xavier Viñolas, Antonio Martinez-Rubio, Toni Bayés-Genis and Antoni Bayés de Luna

Steen Z Abildstrom, Christian Torp-Pedersen and Lars Køber

Part II: Techniques of risk assessment

Branco Mautner

Steen Z Abildstrom, Christian Torp-Pedersen and Lars Køber

11 Electrocardiographic assessment of myocardial ischaemia

Shlomo Stern

Michael Cusack and Simon Redwood

Peter Taggart and Peter Sutton

Piotr Kulakowski

Antonio Michelucci, Luigi Padeletti, Andrea Colella, Maria Cristina Porciani, Paolo Pieragnoli, Alessandro Costoli and Gian Franco Gensini

Katerina Hnatkova and Marek Malik

Morrison Hodges and James J Bailey

Mark L Brown

Federico Lombardi

Xavier Copie, Olivier Piot, Thomas Lavergne, Louis Guize and Jean-Yves Le Heuzey

Robert E Kleiger and Phyllis K Stein

Maria Teresa La Rovere, Andrea Mortara and Gian Domenico Pinna

Georg Schmidt

Pierre Maison Blanche and Philippe Coumel

Daniel M Bloomfield and Richard J Cohen

26 Social and psychosocial influences on sudden cardiac death, ventricular

Harry Hemingway

30 Ventricular arrhythmias in apparently healthy athletes 316

Francesco Furlanello, Fredrick Fernando, Amedeo Galassi and Annalisa Bertoldi

Johan EP Waktare

Joseph T Dell’Orfano and Gerald V Naccarelli

Isabelle C Van Gelder and Harry JGM Crijns

Peter J Schwartz

Patrick Lam and Paul Schweitzer

Josef Kautzner

Part IV: Antiarrhythmic trials

Arthur J Moss

Michiel J Janse

James J Bailey

Center for Information Technology, National Institutes of

Health, Bethesda, Maryland, USA

Velislav Batchvarov

Department of Cardiological Sciences, St George’s

Hospital Medical School, London, UK

Antoni Bayés de Luna

Departamento de Cardiología, Hospital de la Santa Creu i

Sant Pau, Barcelona, Spain

Assistant Professor of Medicine, Division of Cardiology,

College of Physicians and Surgeons, Columbia University,

USA

Mark L Brown

Staff Scientist, Tachyarrhythmia Research Department,

Atrial Fibrillation Research Group, Medtronic,

Minneapolis, Minnesota, USA

Rory Childers

Professor of Medicine, Section of Cardiology, University

of Chicago Medical Centre, Chicago, USA

Timothy R Church

Division of Environmental and Occupational Health,

School of Public Health, University of Minnesota,

Minneapolis, Minnesota, USA

Technology, Massachusetts, USA

Roberto Elosua

Unidad de Lípidos y Epidemiología Cardiovascular,Instituto Municipal de Investigación Médica, Barcelona,Spain

Fredrick Fernando

Sports Science Institute, Italian National OlympicCommittee, Rome, Italy

Francesco Furlanello

S Raffaele Scientific Institute, Milan-Rome, Italy

Amedeo Galassi

Arrhythmia and Electrophysiological Center, S Donato

Institute, Milan, Italy

Gian Franco Gensini

Istituto di Clinica Medica e Cardiologia, Università di

Firenze, Florence, Italy

Louis Guize

Department of Cardiology, Broussais Hospital, Paris, France

Josep Guindo

Departmento di Cardiologia, Hospital de la Santa Creui i

Sant Pau, Barcelona, Spain

Harry Hemingway

Senior Lecturer in Epidemiology, International Centre for

Health and Society, Department of Epidemiology and

Public Health, University College London Medical School,

London; Director of Research and Development,

Department of Research and Development, Kensington &

Chelsea and Westminster Health Authority, London, UK

Morrison Hodges

Minneapolis Heart Institute Foundation, Minnesota, USA

Stefan H Hohnloser

Department of Medicine, Division of Cardiology, J.W

Goethe University, Frankfurt, Germany

Katerina Hnatkova

Department of Cardiological Sciences, St George’s

Hospital Medical School, London, UK

Michiel J Janse

Cardiovascular Research, Academic Medical Center,

Amsterdam, The Netherlands

Josef Kautzner

Department of Cardiology, Institute for Clinical and

Experimental Medicine, Prague, Czech Republic

Robert E Kleiger

Professor of Medicine, Washington University School of

Medicine; Medical Director of Washington University

School of Medicine HRV Laboratory, St Louis, USA

Beth Israel Medical Center, New York, USA

Maria Teresa La Rovere

Fondazione “Salvatore Maugeri”, IRCCS, Divisione diCardiologia, Centro Medico Montescano, Pavia, Italy

Pierre Maison Blanche

Hôpital Lariboisière, Paris, France

Marek Malik

Professor of Cardiac Electrophysiology, Department ofCardiological Sciences, St George’s Hospital MedicalSchool, London, UK

Non-William J McKenna

Professor of Cardiac Medicine, Department ofCardiological Sciences, St George’s Hospital MedicalSchool, London, UK

Antonio Michelucci

Istituto di Clinica Medica e Cardiologia, Università diFirenze, Florence, Italy

Chief, Division of Cardiology, MS Hershey Medical Center

of the Pennsylvania State University, Hershey,

Pennsylvania, USA

Jim Nolan

Consultant Cardiologist, Cardiothoracic Centre, North

Staffordshire Hospital, Stoke on Trent, UK

Luigi Padeletti

Istituto di Clinica Medica e Cardiologia, Università di

Firenze, Florence, Italy

Paolo Pieragnoli

Istituto di Clinica Medica e Cardiologia, Università di

Firenze, Florence, Italy

Gian Domenico Pinna

Servizio di Bioingegneria, Fondazione “Salvatore

Maugeri”, IRCCS, Centro Medico Montescano, Pavia,

Italy

Olivier Piot

Department of Cardiology, Broussais Hospital, Paris,

France

Maria Cristina Porciani

Istituto di Clinica Medica e Cardiologia, Università di

Firenze, Florence, Italy

Simon Redwood

Senior Lecturer / Honorary Consultant Cardiologist,

Department of Cardiology, Rayne Institute, St Thomas’

Hospital, London, UK

Georg Schmidt

Deutsches Herzzentrum und Medizinische Klinik der

Technischen Universität München, Munich, Germany

Isabelle C Van Gelder

Department of Cardiology, Thoraxcenter, UniversityHospital Groningen, Groningen, The Netherlands

Yee Guan Yap

Specialist Registrar in Cardiology, Department ofCardiological Sciences, St George’s Hospital MedicalSchool, London, UK

Gang Yi

Department of Cardiological Sciences, St George’sHospital Medical School, London, UK

quality of life and prolonged patient survival but also a

sig-nificant burden on healthcare providers irrespective of

whether they are privately organised or government

con-trolled Despite the differences in the economies and

healthcare arrangements of the Western world countries,

the discussions about the ever rising cost of medical care

are everywhere similarly heated Clearly, there are no easy

solutions since removing a potentially life-saving treatment

from any human being is not ethical At the same time, no

country of the world is so wealthy that it could afford the

best available treatment and care for all its citizens

Cardiology is no exception to this trend For instance,

studies that have confirmed the efficacy and

appropriate-ness of the prophylactic use of implantable defibrillators

are examples of findings that have both clinical and

finan-cial implications Still, compared to comprehensive safety

nets and resuscitation programs designed to save patients

only after they have suffered from malignant ventricular

arrhythmias, prophylactic use of implantable defibrillators

is most likely a cost-saving option

In general, prophylactic medicine is one of the most

effective ways of reducing the overall costs of medical care

while, at the same time, maintaining adequate quality of

life and improving the survival of the overall population It

is therefore not surprising that prophylactic methods are

presently receiving substantial attention from both clinical

community and healthcare providers The advances in

pro-phylactic medicine are clearly dependent on improved risk

markers and advanced risk stratification approaches Again,

cardiology is no exception Cardiac risk stratification is

presently being investigated not only to identify patients

who might benefit from more cost-effective modes of

pro-phylactic treatment but also, even more importantly, to

save patients who would otherwise succumb to cardiac

death and sudden cardiac death in particular

For all these reasons, I was very pleased when I was

offered the opportunity of editing this book on risk

strati-fication of arrhythmias and sudden cardiac death Indeed,

some of the principles and methodologies of arrhythmia

comprehensive coverage of the field, I have divided thecontents of the book into four sections The first part con-tains chapters dealing with the general principles of riskstratification, explaining the individual facets of methodol-ogy and technology and summarising the goals of arrhyth-mic risk stratification studies The second part is devoted

to a detailed description of individual investigation niques aimed at identification of patients at high risk ofarrhythmia or sudden death The third part describes thepresent experience with applying the risk stratificationtechnologies and tests to patients of different clinicallydefined groups Finally, the two chapters of the final partsummarise the presently conducted clinical trials utilisingthe risk stratification techniques

tech-As with every multi-authored book, I faced the usual toral dilemma of finding the proper balance between havingthe book compact with cross-references within individualchapters and having the chapters suited for separate read-ing Eventually, I felt that with a book of this size aimed atproviding a source of standard references, each chaptershould contain a standard coverage of its subject Hence, I

edi-am happy to recommend the reader to select chapters responding to his/her particular needs and interest.Needless to say, reading the book in its entirety offers muchmore comprehensive learning of the whole field

cor-Without the kind positive response of the authors of theindividual chapters, this book would have never been writ-ten I truly appreciate the efforts of the individual contribu-tors and am very grateful for their kind help My deepthanks also go to my secretary, Mrs Melanie Monteiro, whocarefully organised the editorial office of this book and whohelped me in many other ways I am grateful to the pub-lisher for their useful suggestions, significant technicalhelp, and kind flexibility My apologies go to my wife andchildren since far too frequently I have devoted the timethat I should have spent with them to the editing of thistext

Marek Malik

AT attributable risk

BRS baroreflex sensitivity

CHF congestive heart failure

DAD delayed after-depolarisations

EAD early after-depolarisations

ERP effective refractory period

FRP functional refractory period

LVEF left ventricular ejection fraction

LVH left ventricular hypertrophy

MAPs monophasic action potentials

MI myocardial infarction

intervals that differ ≥50 ms from thepreceding interval

PPV positive predictive value

PSVT paroxysmal supraventricular tachycardia

of normal to normal RR intervals

ROC receiver-operating characteristic

SACT sinoatrial conduction time

SDNN standard deviation of normal to normal RR

in this chapter on aspects related to epidemiological and

clinical issues to stratify risk of SD We consider SD as a

syndrome that may be associated with different diseases

and situations The term “sudden death” has been used in

different ways by epidemiologists, clinicians, forensic

pathologists, etc Clinically, the term is used for deaths

due to natural causes that occur within 1 hour of the

onset of symptoms in a person with or without

pre-exist-ing heart disease, but in whom the time and mode of

death are unexpected In approximately one-third of cases

the death is instantaneous, without symptoms or

coincid-ing only durcoincid-ing a few seconds with the presence of

symp-toms If the patient is found dead, death is considered to

be “sudden” if the subject was seen alive and well in the

preceding 24 hours.1

In this chapter we shall discuss the need for

identifica-tion of high-risk populaidentifica-tions for malignant arrhythmias

and sudden death It is especially important to emphasise

the paradox that the highest-risk subgroups (cardiac arrest

survivors or patients with an ejection fraction <30%)

account for a small proportion of SD events in a

popula-tion We emphasise how different is the clinical approach

for risk stratification in the general population and in

patients with already clear evidence of heart diseases, for

example in post-myocardial infarction patients

Clinical epidemiological aspects

Sudden death usually appears in the presence of clinical

or silent heart disease.2,3Ischaemic heart disease (IHD) is

by far the most common associated situation found in

patients presenting with sudden cardiac death (SCD) (Box

1.1) Nevertheless, other heart diseases may present with

SCD, especially in the presence of heart failure and/or left

ventricular hypertrophy.4,5 In some cases, there are only

isolated electrophysiological abnormalities (WPW, long QT

syndrome, Brugada syndrome, etc.)6–8 or congenital

defects of coronary arteries.9In a few cases (≅3%), there

malities Such cases, now named “idiopathic SCD”, will

be fewer in the future.10

SCD is considered to be of ischaemic origin if itappears during an acute ischaemic event or in the pres-ence of severe IHD From a pathological point of view,SCD is considered ischaemic when stenosis of at least75% of the area of the lumen of one coronary artery isfound Intraluminal or intra-intimal thrombi may or maynot be present.11

Severe coronary atherosclerosis mayexist whether or not there are clinical symptoms (acutemyocardial infarction or angina) before SCD.12

We will comment on the following epidemiologicalaspects related to SCD:

● the size of denominator pools in different populationsubgroups that determine the ability to identifypotential victims within population subgroups of vari-ous sizes;13

● the time dependence of risk, which expresses risk asnon-linear function of time after a conditioning clini-cal event;13,14

Box 1.1 Principal causes of sudden cardiac death

dis-ease

● the importance of the different risk factors and

envi-ronment in the appearance of SCD.1–3

Population subgroups and sudden cardiac death

According to Myerburg,13 if we consider all cases of SCD

occurring in the USA, the overall incidence is in the range

of 1–2/1000 population per year (0.1–0.2%) This large

population base contains both those victims where SCD

occurs as a first cardiac event, in which the possibility to

predict sudden death is very limited (i.e the general

pop-ulation), and those victims whose deaths may be predicted

with greater accuracy because they come from higher risk

subgroups (i.e survivors of sudden cardiac death) (Fig

1.1) Because of the size of the denominator, any

inter-vention designed for the general population must be

applied to more than 99% of population who will not have

an event during the course of a year These numbers limit

the nature of a broad-based intervention and encourage

the identification of specific higher-risk clinical subgroups

Nevertheless, with increasing specificity of subgroups, the

absolute number of potential victims who can be

identi-fied decreases Figure 1.1 expresses the incidence

(per-cent per year) of SCD among various subgroups

In clinical epidemiology it is very important to consider

two different levels of risk: attributable risk versus relative

risk.15,16 Relative risk (RR) is the ratio of the incidence of

an event, SCD in this case, in persons with and without a

risk factor; whereas, attributable risk (AR) refers to the

proportion of the incidence of an event that can beexplained by the presence of the risk factor.1

By moving from the total adult population to a group with higher risk there may be a more than 10-foldincrease in the incidence of events annually, an important

sub-RR increase As shown in Fig 1.1, there is a progressiveincrease in the incidence of SCD as one moves from thegeneral population to subgroups having had a prior coro-nary event and to those with low ejection fractions andheart failure, and to survivors of out-of-hospital cardiacarrest, and to those experiencing ventricular tachycardia

or fibrillation (VT/VF) during the convalescent phase aftermyocardial infarction However, the correspondingabsolute number of deaths becomes progressively smaller

as the subgroups become more focused, an important ARdecrease Thus, in these selected subgroups although the

RR is higher the AR is lower

Clinical application of new technologies and proceduresappears to have had a favourable impact on the highestrisk subgroups Progress measured in terms of prevention

of large numbers of sudden cardiac deaths in the generalpopulation, however, will be limited until it is possible toidentify more easily higher risk individuals in the generalpopulation At the moment it is impossible to know whichpatients in the general population present vulnerableplaques Thus, the only way to prevent SCD in this popu-lation is to fight against classical cardiovascular risk factors(smoking, hypertension, lipid disorders)

Time dependence of risk

Furukawa and Myeburg13,14have also emphasised that risk

of SCD does not appear to be linear as a function of timeafter a change in cardiovascular status Survival curvesafter major cardiovascular events (cardiac arrest survivors,post-myocardial infarction patients, patients with recentonset of heart failure, etc.), which identify populations athigh risk for both sudden and total cardiac death, generallydemonstrate that the most dangerous period occurs duringthe first 6–18 months after the index event By 24months, the slope of a survival curve is not very differentfrom one describing a similar population that has remainedfree of an interposed major cardiovascular event Thus,there is a time dependence of risk after a major event,which stresses the importance of the most intensive inter-vention in the early period The higher risk subgroup, ashappens in survivors of cardiac arrest, has not only a differ-ent overall mortality rate but also a different pattern ofattrition as a function of time In Fig 1.2 the actuarialanalysis of recurrences among a population of 101 cardiacarrest survivors with IHD is shown The risk is high in thefirst 6 months (11.2%) and then falls to 3.3% in the next

Figure 1.1 Bar graphs showing the relation between

inci-dence and total numbers of sudden cardiac deaths in overall

and population subgroups The numbers are calculated for

the US population (Adapted form reference 13.)

(Abbreviations: EF = ejection fraction; MI = myocardial

infarction; VF = ventricular fibrillation; VT = ventricular

three 6-month blocks thereafter Low ejection fraction is

the most powerful predictor of death during the first 6

months, subsequently, persistent inducibility during

pro-grammed stimulation is the most powerful predictor of

death.13,14In a study of a post-myocardial infarction

popula-tion, Moss et al.8pointed out that 50% of the deaths that

occur during the 48 months after myocardial infarction

occur within the first 6 months Therefore, the use of time

as a dimension for measuring risk is extremely important

Risk factors and the environment

The incidence of SCD varies considerably from country to

country with relation to the prevalence of IHD Since IHD

is characteristic of industrialised societies and its ageing

population, the prevalence of SCD is greater in the

west-ern countries than in the rest of the world According to

the World Health Organization,17the incidence of SCD in

industrialised areas varies between 19 and 159 per

100 000 inhabitants per year among men between the

ages of 35 and 64

In the USA and other industrialised countries, the

inci-dence of SCD is decreasing parallel to the overall decrease

in IHD mortality.18 This decrease has been attributed to a

decrease in the incidence of IHD and to a lower fatality

related to an improved medical therapy and a more

effec-am than during the rest of the day A circadian variation

of SCD, similar to that of the occurrence of non-fatalmyocardial infarction and episodes of myocardialischaemia, was observed

Although the incidence of SCD parallels the incidence

of IHD, which increases with age, the proportion of thosewho die suddenly is higher in younger patients This dif-ferential effect of age may be due to older patients havingmore advanced coronary heart disease, and therefore theincidence of death due to heart failure is greater

SCD is more common in males than in females by aratio of 3 to 1 This is due to the lower prevalence of IHD

in premenopausal women Females attain a comparableincidence of SCD 20 years older than men However,more females than males die suddenly without clinical evi-dence of IHD Since the frequency of SCD is much higher

in postmenopausal women compared to premenopausalwomen of the same age, it is likely that a hormonal factorinfluences the result

Multivariate logistic analysis in the FraminghamStudy,21

including all coronary risk factors, indicates that,

in males, age, systolic blood pressure, cigarette smoking,and relative bodyweight are all independently related tothe incidence of sudden death In females, aside fromage, only hypercholesterolaemia and vital capacity areassociated independently with an increased risk of sud-den death Using these parameters, there is a wide varia-tion in the risk of sudden death Forty-two percent ofsudden deaths in males and 53% in females occur in one-tenth of the population in the top decile of multivariaterisk

It is important to conceptualise sudden cardiac death

as a manifestation of IHD occurring as a consequence of avariety of risk factors and living habits Small changes inblood pressure, cholesterol levels, and cigarette smokingpattern may not be of great importance if they are iso-lated, but they can interact to increase the risk of SCD.Genetic factors are also important in less commoncauses of SCD, such as in the congenital long QT syn-drome,22

hypertrophic cardiomyopathy,23

arrhythmogenicright ventricle dysplasia, or Brugada syndrome,24

etc.Parental sudden death has been also identified as a riskfactor for SCD.25

0

Months ( F/U )

among survivors of cardiac arrest Actuarial analysis or

recur-rences among a population of 101 cardiac arrest survivors

with coronary heart disease Risk was higher in the first 6

months and then fell to 3.3% for the next three 6-month

peri-ods After 24 months, the rate fell to 0.8% for each 6-month

period thereafter The most powerful predictor of death was

a low ejection fraction (EF) during the 6 first months and

persistent inducibility during programmed stimulation

after-wards (Adapted from reference 14.)

Pathophysiology of sudden cardiac death

Ischaemic heart disease

Most cases of SCD occur in patients with IHD In these

patients the chain of events leading to SCD may occur in

two ways:

● an acute ischaemic syndrome (AIS), especially acute

myocardial infarction (AMI), or

● a primary arrhythmic event (PAE) (Fig 1.3)

It is practically impossible to predict the appearance of

acute ischaemic syndrome (acute MI) when this is the

first manifestation of the disease without previous

herald-ing symptoms This would only be possible if we could

detect, easily and non-invasively, not only the presence of

coronary atherosclerosis, but especially the existence of

vulnerable plaques Such plaques are at high risk of

rup-ture or thrombosis.26–28From a pathological point of view,

plaques are considered vulnerable when they have large

lipid cores occupying more than 40% of the overall plaque

volume, thin and inflamed fibrous caps separating the

cores from the arterial lumen, a high density of

macrophages, and a low density of smooth-muscle cells in

the caps.26

On the other hand, the danger of a primary arrhythmic

event often occurs in post-MI patients In these cases, the

danger of an acute ischaemic syndrome (severe and

persis-tent ischaemia, usually appearing as acute myocardial

infarction) is not the only cause of the vulnerability of the

myocardium to SCD, although some degree of residual or

transient ischaemia may be present The most important

markers that express this arrhythmic risk type of

vulnera-bility are:

● clinical/ECG

● morphological (anatomic substrate), and

● related to autonomic nervous system imbalance (Fig.1.4)

In all these circumstances these markers represent anincrease, often very important, in electrical instability,which bears no relation to the presence of an acuteischaemic syndrome This instability is a substrate todevelop a malignant ventricular arrhythmia in the pres-ence of different triggers

The comparative role of acute ischaemic syndrome versus primary arrhythmic event in the presentation of SCD: a 50 % responsibility

As we stated at the beginning of this article (Fig 1.1), thepopulation pool of SD represents an overall incidence of1–2/1000 per year This large population base include vic-tims of SCD as a first manifestation of disease; in themajority of patients with first MI, in whom SCD is related

to acute ischaemic syndrome (severe and persistentischaemia), the danger of SCD is very difficult to predict

In this group there are also included a small number ofcases of hypertrophic cardiomyopathy, primary ventricularfibrillation, etc This group represents 40–60% of SCDcases On the other hand, the remaining cases of SCD may

be predicted with greater accuracy because they areincluded in higher-risk subgroups The majority of thesepatients present IHD, but often the cause of the presenta-tion of the final event is not an acute ischaemic syndrome,although some transient ischaemia may be present There

is considerable evidence29,30 that, in patients withimplanted defibrillator, thus in patients at a higher risk topresent SCD, the cause of death is, in the majority ofcases, a primary arrhythmic event and not an ischaemic-

Ischaemic

heart

disease

Vulnerable plaque

Anatomic substrate

Vulnerable myocardium

Ventricular fibrillation

Ventricular fibrillation

Final arrhythmia

Acute ischaemic syndrome Sustained ventricular tachycardia

Final step

Sudden cardiac death

Figure 1.3 The chain of events that leads to final

arrhyth-mia and sudden cardiac death Some triggers and/or

modu-lators acting in a vulnerable myocardium lead to the final

step and final arrhythmia The triggers and modulators may

be similar in the two ways but the vulnerable myocardium is

different: vulnerable plaque in the case of acute ischaemic

syndrome leading to a ventricular fibrillation and anatomic

substrate leading to a sustained ventricular tachycardia and

ventricular fibrillation.

Anatomic Substrate

Electrical instability

RISK

Trigger Malignant Ventricular Arrhythmia Clinical and ECG

markers

Autonomic nervous system

Left ventricular dysfunction

Residual ischaemia

Figure 1.4 The triangle of risk in post-myocardial tion patients with satellite triangle in the angle of electrical instability with different markers of bad outcome for a pri- mary arrhythmic event.

infarc-tricular arrhythmia recorded in device-incorporated

elec-trograms during follow-up This adds further support to

the consideration that the primary cause of VF was a

sus-tained VT not triggered by an acute ischaemic syndrome

On the whole we may consider that acute ischaemic

syndrome with thrombosis accounts for approximately

40–50% of cases of SCD There are several arguments to

support this Clinically, the incidence of symptoms

sugges-tive of ischaemia in cases of SCD is between 30 and

50%.1,2 Furthermore, in our survey of patients who died

while wearing a Holter device,31 evidence of ischaemia

detected by Holter monitoring was around 30% From an

angiographic point of view, Spaulding demonstrated32 that

the presence of acute coronary artery occlusion in a

coro-nariography performed immediately in survivors of

out-of-hospital cardiac arrest was 48% Lastly, from a pathological

point of view the frequency of active lessions varies greatly

(from 13% to 95%)11,12,33,34 depending on the type of

patients studied (ambulatory versus CCU patients,

methodological differences, etc.) Burke et al.33 recently

demonstrated that, in patients with IHD and presenting

with a SCD, acute thrombosis was present in 52% of cases

(WPW) and congenital long QT syndrome

Clinical goals

The clinical goals are to prevent the appearance of the ease if possible, and when the disease exists to avoid pro-gression from silent to clinical, and when it has alreadybeen diagnosed to decrease the incidence of new eventsand complications We will comment on some aspectsfocusing on IHD but also consider this problem in othersituations

dis-In case of ischaemic heart disease

To detect the presence of ischaemic heart disease and especially of vulnerable plaque before the clinical appearance of the disease

At present, as previously described, the presence of nerable plaque by non-invasive methods cannot be investi-

vul-Triggers &

modulators

Vulnerable myocardium

Final step

Final arrhythmia

Sudden cardiac death Hypertrophic

· Supraventricular arrhythmias

· Ionic/Metabolic imbalance

· Drugs

· Supraventricular arrhythmias

· Pulmonary embolism

· Atrial fibrillation with rapid ventricular response

· psychologial stress

Physical-Hypertrophy (disarray)

Electrical instability

Ventricular fibrillation

Ventricular fibrillation

Ventricular fibrillation

Ventricular fibrillation

Electrical instability

Electrical instability

Torsade de pointes

Dilated (fibrosis)

High risk bypass tract

Repolarization abnormalities

Figure 1.5 Sequence of events leading to sudden cardiac death in other associated diseases:

hypertrophic cardiomyopathy, heart failure, Wolff–Parkinson–White, and congenital long QT syndrome.

gated There is some evidence that magnetic resonance

may be useful in the future.35

What is possible is to pect the presence of coronary atherosclerosis by special

sus-computed scanner and when some non-invasive ECG tests

are very pathological (e.g exercise testing) The sensitivity

and specificity are also very high if this pathological

non-invasive test is positive in patients with many risk factors

or evidence of atherosclerotic disease in other locations In

this high-risk subgroup of asymptomatic patients, it may be

necessary to perform other tests, even in some cases

coro-nariography to rule out IHD, and if possible, to identify

vulnerable plaques with more sophisticated techniques.36

Preventive campaigns are very important because, if we

treat hypertension, decrease level of lipid disorders and

smoking, we will decrease the incidence of IHD If IHD is

present, we can slow atherosclerosis progression or

sta-bilise vulnerable plaques and consequently decrease the

incidence of SCD Nevertheless, there are still 20–30% of

cases in which IHD is present in the absence of classical

risk factors Thus it is also very important to identify

genetic aspects or other non-conventional risk factors that

increase the risk of IHD

To avoid recurrences of ischaemic syndrome

and the presentation of primary arrhythmic

event

When IHD is already established, our goal is to fight

against the factors that are related to a negative outcome:

residual ischaemia, left ventricular dysfunction, and

elec-trical instability (Fig 1.4) Therefore to decrease SCD we

have to proceed with medical treatment or

revascularisa-tion procedures to decrease or, if possible, abolish

resid-ual ischaemia, and to give the best treatment

(beta-blockers, ACE inhibitors, etc.) to decrease left

ven-tricular dysfunction If patients show markers of a

high-risk group for SCD from a primary arrhythmic event (low

ejection fraction, evidence of substrate, ventricular

cardia in Holter, or previous sustained ventricular

tachy-cardia, especially if it occurs in the first month after

myocardial infarction), we may advise an ICD as a primary

preventive measure

In other associated diseases

In other cases, as in those of SCD associated with other

different diseases (hypertrophic cardiomyopathy, electrical

disorders, etc.) we have to proceed to stratify the risk of

SCD In case of high risk, the solution as you will see in

different chapters of this book may be different, but

usu-ally we will have enough information to solve, sometimes

definitively, the problem as happens in WPW syndrome,

congenital abnormalities, etc In other cases, as in

hyper-trophic cardiomyopathy, dilated cardyomyopathy, tal long QT syndrome, or idiopathic ventricular fibrillation,the only solution at the present time in high-risk groupsmay be an ICD

congeni-Conclusion

In the majority of cases the approach to reduce the ber of SD victims has been based on secondary preventionamong patients who have survived potentially fatalarrhythmias, or those who are identified as being at extra-ordinarily high risk because of a recent major cardiovascu-lar event or specific clinical risk factors Nevertheless, thisapproach does not emphasise that a large number of fatali-ties occur as a primary event In patients with high risk ofSCD, as in those who have already presented with out-of-hospital cardiac arrest, the most optimistic figures avail-able for 1-year survival are in the range of 25–30% Inthese cases an ICD implant is mandatory However, pri-mary prevention of cardiac arrest should have a muchgreater impact on overall outcome, but at the moment it

num-is practically impossible to identify individual patients athigh risk from amongst large population pools who usually

do not present with markers of any real danger of SCD.Therefore we have to increase our ability to detect candi-dates of SCD in the global population if we want to fightefficaciously for SCD prevention in the future

References

1 De Vreede-Swagemakers JJ, Gorgels AP, Dubois-Arbouw WI

et al Out-of-hospital cardiac arrest in the 1990s: a

popula-tion-based study in the Maastrich area on incidence,

char-acteristcs and survival J Am Coll Cardiol 1997;30:1500–5.

2 Hinkle LE, Thaler HT Clinical classification of cardiac

deaths Circulation 1982;65:457–4.

3 Goldstein S, Bayés de Luna A, Guindo Soldevila J Sudden

cardiac death Mount Kisko, NY: Futura Publishing Co.,

1994.

4 Maron BJ, Fananapazir L Sudden death in hypertrophic

cardiomyopathy Circulation 1992;85:I57–63.

5 Stevenson WG, Stevenson LW, Middlekauff HR, Saxon LA Sudden death prevention in patients with advanced ven-

tricular dysfunction Circulation 1993;88:2953–61.

6 Torner P, Brugada P, Smeets J et al Ventricular fibrillation

in Wolff–Parkinson–White syndrome Eur Heart J

1991;12:144–50.

7 Brugada J, Brugada R, Brugada P Right bundle branch block and ST-segment elevation in leads V1 through V3: a marker for sudden death in patients without demonstrable

structural heart disease Circulation 1998;97:457–60.

8 Moss AJ, Schwartz PJ, Crampton RS, Locati E, Carleen E The long QT syndrome: a prospective international study.

Circulation 1985;71:17–21.

9 Cheitlin MD, De Castro CM, McAllister HA Sudden death

left ventricular scarring and heart weight Am J Cardiol

1984;54:65–73.

13 Myerburg RJ, Kessler KM, Castellanos A Sudden cardiac

death: structure, function, and time-dependence of risk.

Circulation 1992;85(Suppl I):I2–I10.

14 Furukawa T, Rozanski JJ, Nogami A, Moroe K, Gosselin AJ,

Lister JW Time-dependence risk of and predictors for

car-diac arrest recurrence in survivors of out-of-hospital carcar-diac

arrest with chronic coronary artery disease Circulation

1989;80:599–608.

15 Nieto FJ, Peruga A Riesgo atribuible: sus formas, usos e

interpretación Gaceta San 1990;18:112–17.

16 Greenland S, Rothman KJ Measures of effect and

mea-sures of association In: Rothman KJ and Greenland S eds.

Modern epidemiology Philadelphia: Lippincott-Raven

Publishers, 1998, pp 47–66

17 World Health Organization Technical Report 726 Sudden

cardiac death Basel: WHO, 1985, pp 3–26.

18 Feinleib M The magnitude and nature of decrease in

coro-nary heart disease mortality Am J Cardiol 1984;54:2C–6C

19 Muller JE, Ludmer PL, Willich SN et al Circadian variation

in the frequency of sudden cardiac death Circulation

1987;75:131–8.

20 Willich SN, Levy D, Rocco MB, Tofler GH, Stone PH,

Muller JE Circadian variation in the incidence of sudden

cardiac death in the Framingham Heart Study population.

Am J Cardiol 1987;60:801–6.

21 Kannel WB, Schatzkin S Sudden death: lessons from

sub-sets in population studies J Am Coll Cardiol

1985;5:141B–9B.

22 Zareba W, Moss AJ,Schwartz PJ et al Influence of

geno-type on the clinical course of the long-QT syndrome.

International Long-QT Syndrome Registry Research Group.

New Engl J Med 1998;339:960–5.

23 McKenna WJ, Coccolo F, Elliot PM Genes and disease

acteristics to degree of stenosis in human coronary

arter-ies Circulation 1996;84:928–31.

28 Gronholdt ML, Dalager-Pedersen S, Falk E Coronary

atherosclerosis: determinants of plaque rupture Eur

Heart J 1998;19(Suppl C):C24–C29.

29 Henry PD, Pacifico A Sudden cardiac death: still more

questions than answers G Ital Cardiol 1997;27:1319–24.

30 Mont L, Valentino M, Sambola A, Matas M, Aguinaga L, Brugada J Arrhythmia recurrence in patients with a healed myocardial infarction who received an implantable defibril-

lator: analysis according to the clinical presentation J Am

Coll Cardiol 1999;34:351–7.

31 Bayés de Luna A, Coumel P, Leclercq JF Ambulatory den cardiac death: mechanism of fatal arrhythmia on the

sud-basis of data from 157 cases Am Heart J 1989;117:151–9.

32 Spaulding CM, Joly LM, Rosenberg A et al Immediate

coronary angiography in survivors of out-of-hospital cardiac

arrest New Engl J Med 1997;336:1629–33.

33 Burke AP, Farb A, Malcom GT, Liang YH, Smialek J, Virmani R Coronary risk factors and plaque morphology in

men with coronary disease who died suddenly New Engl J

Med 1997;336:1276–82.

34 Farb A, Tang AL, Burke AP, Sessums L, Liang Y, Virmani R Sudden coronary death Frequency of active coronary lesions, inactive coronary lesions, and myocardial infarc-

tion Circulation 1995;92:1701–9.

35 Toussaint JF, LaMuraglia GM, Southern JF, Fuster V, Kantor HL Magnetic resonsance images lipid, fibrous, cal- cified, hemorrhagic and thrombotic components of human

atherosclerosis in vivo Circulation 1996;94:932–8.

36 Feld S, Ganim M, Carell ES et al Comparison of

angioscopy, intravascular ultrasound imaging and tive coronary angiography in predicting clinical outcome

quantita-after coronary intervention in high-risk patients J Am Coll

Cardiol 1996;28:97–105.

Defining arrhythmic risk is the art of classifying clinical

events, most often death, as being primarily caused by or not

caused by a cardiac arrhythmia In rare occasions this may

be straightforward, but in most cases the amount of data

available is modest and the classifications are presumptuous

Some classifications wisely take some distance to the goal of

assuming arrhythmia, by using the concept of sudden death,

but even this is presumptuous in many cases The

classifi-cations all rely on the fact that sudden unexpected death can

be classified as arrhythmic in many of those cases where

documentation of the cardiac rhythm is available at the time

of death or syncope This is particularly true in the general

population where there is ample documentation that

unex-pected (circulatory) collapse often is caused by ventricular

fibrillation or ventricular tachycardia However,

classifica-tion is not focused in the general populaclassifica-tion, but in groups

of heart disease patients where the risk of death is high and

where specific interventions can be attempted to prevent

certain causes of death Death is in general related to a

car-diac arrhythmia, eventually asystole, and the classifications

must in some way try to identify those cases where a cardiac

arrhythmia was likely to have been the primary event This

may in general be highly problematic in patients who already

have a cardiac disease and where any event can be regarded

as secondary to the disease and not completely unexpected

Primary arrhythmic cause becomes a more uncertain term

relating to the circumstances immediately prior to death

The uncertainty has naturally generated numerous

defini-tions and to some degree each group of investigators have

in their dissatisfaction tried to improve classifications in the

hope of being more accurate

In the following we will summarise how arrhythmic risk

has been defined We will make no claim of presenting a

completely exhaustive list of all definitions used, but will

instead try to outline the circumstances where the

defini-tions have been implied, the amount of data the

investiga-tors required, and the extent to which the classifications

have been validated

Sudden death

Sudden death (SD) is the more common and more modest

of classifications, the name including the uncertainty of

the classification Often, but not always, SD has a timefactor included, and death is classified as sudden whenthe time between death and new symptoms is below acertain figure The time base for a definition of SD in rela-tion to new symptoms has varied considerably from 24hours to immediate, with a clear trend over the yearstoward using 1 hour or less as the limit Before consider-ing the time course of the terminal event, deaths with adocumented non-cardiac or non-cardiovascular reason, orwhere the underlying reason is obscure, should be sepa-rated from the cardiac or cardiovascular deaths In popula-tions with manifest heart disease, this may have only littleimpact on the final result since most of the mortality isdue to cardiac or cardiovascular reasons, but in the gen-eral population the results may be influenced

In the mid-1960s Kuller using a time limit of 24 hoursfound that the leading finding in sudden unexpecteddeath among 20–39-year-olds was alcoholism and fattyliver.1 In a 1970 report from the World HealthOrganization Scientific Group, SD was referred to, but itwas considered premature to define it.2SD was conceived

as a very general concept describing unexpected vascular death The report describes other World HealthOrganization studies, where a time-based definition of SD

cardio-is a death occurring within 6 hours from the onset of newsymptoms Nine years later in a report from WHO con-cerning the nomenclature for primary cardiac arrest, a def-inition of SD was “purposely omitted”; it stated that “Thedefinition used should be operational.”3 Several studiesevaluating the prognostic value of different non-invasivemeasures of arrhythmogenicity as a predictor of SD fol-lowing myocardial infarction have set the time limit to 1hour, giving the first indications that SD in some cases isarrhythmic In order to understand the mechanism of SD,Goldstein in 1982 suggested that only witnessed deathsshould be considered and a time limit of 1 hour used.4Ro-

berts gave a restricted suggestion in an editorial in The

American Journal of Cardiology.5Here SD was defined asdeath, which is non-violent, instantaneous, and occurringwithin a few minutes of an abrupt change in previous clin-ical state The concept was specifically suggested to omitinclusion of patients dying during sleep or found dead.This concept is currently most often described as instanta-

ening heart failure However, patients suffering an

out-of-hospital collapse may be brought to the out-of-hospital before

they are declared dead and patients dying out-of-hospital

may do so because they wish a peaceful end to a long

ill-ness Place of death should be viewed in a broader

con-text Consider a case where a patient is entered in a

post-MI study during hospitalisation for the

index-infarc-tion and dies still in-hospital Such a death could be

sud-den but is likely to be secondary to the MI and can hardly

be classified as both sudden and unexpected, unless the

patient was about to be discharged When deaths are

clas-sified in such studies, mortality should be divided into

death during index-hospitalisation and death following

dis-charge

Patients dying in coma following a cardiac arrest are

usually classified as SD even if some authors have

required death to be declared within a certain time limit

(for example 24 hours).7

The reason that SD is an appropriate entity is the

antici-pation that this in many cases represents arrhythmic

death At least the proportion of SDs being arrhythmic is

larger than the proportion of non-SDs being arrhythmic,

but this relation is highly dependable on the population

studied Hinkle and Thaler8

found indeed that SD may not

be arrhythmic, and arrhythmic death may not be sudden

Using data from a monitoring study in high-risk patients in

New York they found that all 37 deaths occurring within

5 minutes could be classified as arrhythmic Sixteen of 20

deaths occurring between 5 minutes and 1 hour were also

classified as arrhythmic leading to 93% SDs (<1 h) being

arrhythmic However 29 of 84 deaths (35%) occurring

later than 1 hour were also classified as arrhythmic

sug-gesting an underestimation of the actual number of

arrhythmic deaths when the classification based solely on

the time from debut of symptoms till death was used In

the AIRE study all deaths among randomised patients

were classified by both a time-based and a more

descrip-tive algorithm.7

Cleland et al found that, even if 50% of

the cardiovascular deaths were sudden, only

approxi-mately 50% of the SDs in this particular group of patients

were arrhythmic.9

However, proof of arrhythmic deathwas common in none of the definitions used in AIRE, and

the different classifications used serve to illustrate that

The Seattle studies and similar studies in the general ulation indicate that instantaneous observed death is verylikely to be preventable arrhythmic death; this is not nec-essarily the case in the populations of patients with heartfailure and severe left ventricular dysfunction, where therisk of dying is sufficiently grave to warrant prophylacticintervention

pop-Arrhythmic death

Arrhythmic death will often be the focus of interest when

SD is studied An attempt to improve classification wasprovided by Hinkle and Thaler in 1982.8

This work isimportant since it appears to be the first attempt to pro-vide a surrogate endpoint for arrhythmic death, with dataused that are more likely to be available than ECGs at thetime of death They proposed a clinical definition ofarrhythmic death as “abrupt loss of consciousness and dis-appearance of pulse without prior collapse of the circula-tion” They examined four general methods of classifyingcardiac deaths One classification was made based on “thecondition of the circulation immediately before death”(Box 2.1), and was considered to be the most useful.Other classifications examined were “classification in rela-tion to clinical evidence of acute ischaemic heart disease

at the time of death”, “classification by the duration of theterminal acute illness”, and classification by “location ofthe death” When a classification by state of the circula-tion at the time of death was found most useful, this is anassumption, since the truth is not known This definitionhas had a major impact on later definitions and on theframework that classification committees have worked in

If the Hinkle and Thaler surrogate classification is correct,then SD within 1 hour used as synonym with arrhythmicdeath underestimates the occurrence of arrhythmic death Death classification in the Cardiac ArrhythmiaSuppression Trial (CAST) as developed in the pilot study11

was specifically designed to detect arrhythmic deaths sibly preventable by antiarrhythmic drugs A surrogate def-inition of arrhythmic death/cardiac arrest was similar tothe Hinkle/Thaler classification defined as “the abruptspontaneous cessation of respiration and blood circulation

pos-(pulse) and loss of consciousness in the absence of other

progressive severe medical conditions likely to cause

death” Documented arrhythmic death/cardiac arrest was

considered present if ventricular tachycardia or fibrillation

was recorded within 10 minutes of clinical death Other

cases were classified as “presumed arrhythmic” In

evalua-tion of “other progressive medical condievalua-tions likely to

cause death” in the definition of arrhythmic death, the

event committee only classified death as arrhythmic if the

patient would probably have survived at least 4 months

had the arrhythmia not occurred The resemblance to the

Hinkle/Thaler criteria has in many cases led to the term

“modified Hinkle/Thaler criteria” when the criteria used in

CAST are being referred to

In the CAST pilot study12 the relation between SD

(within 60 minutes of new symptoms) and arrhythmic

death (documented or presumed) was studied Of 29

cases of SD, nine were classified as non-arrhythmic death,

and of 16 patients with non-SD three were classified as

arrhythmic This result differs from the classification by

Hinkle and Thaler as described above, mainly because a

significant proportion of SD was not classified as

arrhyth-mic It may be important that the populations studied

were different, but a more likely cause of the discrepancy

is that the CAST criteria of arrhythmic death require a

clinician to assume that the patient would have been able

to live for 4 months, had the arrhythmia not occurred

Nevertheless, there is overall agreement between the

CAST criteria and the Hinkle/Thaler criteria, that SD

within 1 hour of new symptoms is a reasonable marker for

arrhythmic death as illustrated in Table 2.1 where the

numbers of death giving the sensitivity and specificity are

listed

In a report from 1996 Narang et al.7 demonstrated

beyond doubt the heterogeneity among classificationsused for deaths in chronic heart failure patients They rec-ommended that SD should not be time based, but rathershould be death not attributable to terminal pump failure,stroke, or non-cardiac cause of death The so-called ACMEcriteria take several factors into consideration:

This new set of criteria has been developed for theAcute Infarction Ramipril Efficacy (AIRE) study and, asmentioned, a rather poor association between sudden andarrhythmic death was observed in this particular patientpopulation

Box 2.1 Hinkle and Thaler classification of death

disabling Witnessed directly Not witnessed directly

disabling Witnessed directly Not witnessed directly

Witnessed directly Not witnessed directly

Table 2.1 Sensitivity and specificity between sudden death (SD) and arrhythmic death (AD) in Hinkle and Thaler, and CAST Pilot Study, where AD is “the truth”

Hinkle & Thaler

uniform tabulation of certain data Clearly the

tion should now be modified to include also a

classifica-tion as described by Hinkle and Thaler

Arrhythmic events

In order to boost the number of endpoints investigators have

tried to combine arrhythmic death and non-lethal

ventricu-lar arrhythmic events This will be particuventricu-larly obvious in

implantable cardioverter defibrillator (ICD) settings The

problem is to define clearly when an arrhythmic event is

potentially lethal If just the number of shocks without

elec-trogram (EGM) is used, documentation is inadequate In the

Coronary Artery Bypass Graft (CABG)-Patch trial, there was

huge discrepancy between number of shocks in the ICD

group and mortality in the control-group.14By the end of 4

years of follow-up, 70% of the ICD patients had received a

shock, but the absolute arrhythmic death rate in the control

group was 6.2%.15 This relation has probably improved as

ICD technology has developed better algorithms detecting

truly ventricular arrhythmias Even when EGM

documenta-tion is available, some kind of clinical manifestadocumenta-tion of the

arrhythmia should be warranted Syncope in a patient with

an arrhythmic disposition may by default be interpreted as

an arrhythmia but in the general population this is invalid

Use of classifications in studies

Over the years the different classifications have been

applied to a number of patient populations and subsets of

the general population Table 2.2 lists some of the major

randomised interventional studies that have tried to

subdi-vide mortality into sudden or arrhythmic death versus

non-sudden or non-arrhythmic death The aim of the table

is to show the heterogeneity, since the actual numbers

are highly biased by the differences in patient subgroup,

patient selection and follow-up, and randomisation

proce-dures In some cases we now know that applying the

intervention actually increased the risk of dying, especially

sudden or arrhythmic death, making the picture even

more complex

ventable arrhythmic death It also provides someindication that the classification (or the classifiers) mayunderestimate preventable arrhythmic death since cardiacnon-arrhythmic death was also reduced

ICD trials are very interesting since an ICD can onlytreat arrhythmias efficiently and are only rarely pro-arrhythmic or induce non-arrhythmic death In theAntiarrhythmics versus Implantable Defibrillators (AVID)study, the Canadian Implantable Defibrillator Study(CIDS), and the Cardiac Arrest Study Hamburg (CASH)study, the impact of ICD versus amiodarone was a 50%reduction in arrhythmic death and no influence on non-arrhythmic death.17 These three trials all used the modi-fied Hinkle/Thaler criteria being somewhat like the CASTcriteria This gives a notion that classification of deaths isindeed possible and meaningful at least in a patient popu-lation having suffered and survived a cardiac arrest

Hartikainen et al.18demonstrated that heart rate variabilitywas highly efficient in predicting arrhythmic death inpatients with preserved ejection fraction, and that lowejection fraction was efficient in predicting non-arrhyth-mic death in patients with preserved heart rate variability

In such an evaluation there is a risk of a circular ment For death to be classified as arrhythmic by theCAST criteria it is required that the patient could beassumed to survive for a further 4 months if arrhythmicdeath had not occurred If the ejection fraction is verylow, this could be the reason for assuming that thepatient would not survive 4 months, and thus deathwould be classified as non-arrhythmic Thus, the low ejec-

argu-tion fracargu-tion could per se be the reason that the patient

was classified as having non-arrhythmic death

Perhaps a time-based definition for SD should also be

studied in such studies, although it would be no

guaran-tee, since the criterion of unexpectedness could include

similar risk of a circular reasoning Knowledge of a very

low ejection fraction could raise the probability that death

was not considered unexpected

Fig.2.1 shows the decision process of a CAST-like

clas-sification This slightly simplified flowchart demonstrates

that lack of data will increase the likelihood of death

being classified by default as presumed arrhythmic In

descriptions of evaluations, it is never forgotten that the

members of the event committees are experts, whereas

the fact that autopsies are rare and few data are available

in many cases is passed lightly

An exact description of the handling of cases, where

the amount of data is so poor or contradictory that

classifi-cation is truly random, is important since differences inthese procedures may lead to different results The lack ofinformation may make it impossible to decide whether the

Abbreviations: ACE-I = ACE-inhibitor; ca-blocker = calcium channel antagonist; CHF = congestive heart failure; d = days; EF = ejection fraction;

h = hour; HF = clinical heart failure; m = months; nsvt = non-sustained ventricular tachycardia; Post-MI = following myocardial infarction; PVC = premature ventricular complex; SAECG = signal averaged ECG; spiron = spironolactone.

Death Cardiac death Sudden death Arrhythmic death Presumed arrhythmic death

Documented non-cardiac death Documented non-sudden death Documented non-arrhythmic death Documented arrhythmic death

Figure 2.1 The decision process of a CAST-like cation.

classifi-If the concept of SD is going to be used in the

foresee-able future, it must in each case be defined whether the

concept is used as a (relatively) strict time-based definition

or whether more descriptive modes of death are included

in the concept Particular attention must be paid to the

description of the handling of unobserved death Some

classifications have requested that the time from when

the patient was last seen should be used Most

classifica-tions state that suddenness should be classified by a

“detailed review of the circumstances of death” Most

likely, circumstances where the patient appears to have

died instantaneously or has peacefully died during sleep

are classified as sudden

Validation of classifications

The basic problems inherent in any classification of

arrhythmic death are related to validation of classifications

and the quality of available data Currently we have

large-scale validation from the Seattle studies of cardiac arrest

in the community that sudden, observed collapse in

soci-ety is often preventable arrhythmic death With MADIT,

AVID, CIDS, and CASH we have small-scale evidence that

death classified as arrhythmic by the Hinkle/Thaler or

CAST criteria is also to a large extent preventable

There are major problems with the quality of available

data In many studies autopsy reports are only available in

a few percent of cases and, even in studies where specific

efforts have been made, autopsy reports are only available

in a minority of the cases Because of the many difficulties

of practical classification, committees are used to perform

the classification Usually, cases are classified by at least

two members and various mechanisms of reaching

con-sensus are used in cases of discrepancies In a recent

investigation of patients in different ICD studies, Pratt et

al.19

studied the inaccuracies of evaluation With the use

of CAST criteria, committee consensus was reached in

classifying a 2% rate of arrhythmic death Nearly twice as

many, 3.6% was classified as SD by at least one committee

member In addition, the opinion of the committee was

not in agreement with that of the investigator in 40% of

the cases Uncertainty of interpretation of the data is not

nal EGM led to erroneous classification of a pulmonaryembolism as arrhythmic death This shows that document-ing an arrhythmia in the final phase of dying does not nec-essarily make the death arrhythmic These recordingsshould probably be viewed in connection with the generalcircumstances of the terminal event

Grubman20

and Pires21

have reported some data frompatients dying with SD despite having an ICD The elec-trograms showed that in most cases SD in these patientswas not related to arrhythmia and that none of the SDswas due to ICD malfunction These data cannot be used

to comment on the relation between sudden and mic death, since these patients are highly selected andsince the events, where the ICD did actually treat apotentially lethal arrhythmia, were not in the analysis.However, it is comforting to know that ICD malfunction israre even in patients dying SD

arrhyth-Future perspectives

Mortality studies of medical therapies and devices to treat

or prevent arrhythmias must include a classification ofmode of death Current classifications have a major draw-back in that the quality of data is not apparent from thetabulation Even the finest committee can be forced todefault classification in the majority of cases, if data aresparse resulting in an overreporting of arrhythmic death.For this reason, we would recommend that future tabula-tion of arrhythmic deaths include a tabulation to indicatethe certainty of available data as shown in Box 2.2 Alsothe exact handling of cases where data are insufficient(death during sleep, unwitnessed death etc.) should beapparent from the report

No study has yet provided sufficient data in their report

to allow the reader to classify deaths by such strict ria Use of committees and/or tabulation of the committeeconclusions cannot replace a presentation of the availabledata It is to be hoped that investigators and editors willrequire that a breakdown as suggested or similar isincluded in future reports of death classifications

crite-Apart death being classified as arrhythmic or arrhythmic, it is of major interest to subclassify, in order

non-to examine groups where the outcome of therapy might

be different The Hinkle/Thaler classification suggests one

important classification based on the general severity of

disease It could also be of major interest to examine

whether ischaemia is present at the time of arrhythmia,

since arrhythmias provoked by ischaemia could very well

have a different outcome than those unrelated to acute

ischaemia Hinkle and Thaler provide one suggestion for

such a classification simply by pooling clinical information

as to whether the patient in immediate relation to the

arrhythmia had ischaemia or acute MI The value of such

a subclassification is currently unknown, but should be

attempted

Since all current evaluations of mode of death rely on

unproved assumptions, no classification has yet the

poten-tial to be used as a primary endpoint in a trial In

accor-dance with this, recommendations for evaluation of ICDs

should use total mortality as an endpoint.22

Given all the uncertainties of classification, proof of

arrhythmic death can only be achieved by proving that

pre-vention or treatment of arrhythmia saves life Such proof

has been established in the general population by the

Seattle studies, and with MADIT, AVID, CIDS, and CASH

we have emerging evidence of proof in at least part of the

high-risk population However, with the CABG-Patch

study,14 we also have an indication that the situation is

more complex in high-risk patients At this time a number

of studies are being initiated, randomising wider groups of

patients to receive a cardioverter defibrillator or standard

treatment The success or failure of these studies will

pro-vide critical insight We can further hope that technical

problems with devices as well as their prices will fall

fur-ther in order to allow true wide-scale intervention studies.These studies could very well resolve the uncertainties ofclassification of high-risk patients, yet perhaps obviate theneed for future classification In the meantime we mustlive with the uncertainties and hope that reports in thecoming years will present their results in such a way thatthe uncertainties of each study are clearly presented

References

1 Kuller L, Lilienfeld A, Fisher R Sudden and unexpected

deaths in young adults An epidemiological study JAMA

1966;198:248–52.

2 World Health Organization Scientific Group The

pathologi-cal diagnosis of acute ischemic heart disease Wld Hlth Org

Techn Rep Serv 1970;5–27.

3 Rapaport, E Nomenclature and criteria for diagnosis of

ischemic heart disease Circulation 1979;59:607–8.

4 Goldstein S The necessity of a uniform definition of den coronary death: witnessed death within 1 hour of the

sud-onset of acute symptoms Am Heart J 1982;103:156–9.

5 Roberts WC Sudden cardiac death: definitions and causes.

Am J Cardiol 1986;57:1410–13.

6 Goldstein S, Friedman L, Hutchinson R et al Timing,

mechanism and clinical setting of witnessed deaths in

post-myocardial infarction patients J Am Coll Cardiol

1984;3:1111–17.

7 Narang R, Cleland JG, Erhardt L et al Mode of death in

chronic heart failure A request and proposition for more

accurate classification Eur Heart J 1996;17:1390–403.

8 Hinkle LE J, Thaler HT Clinical classification of cardiac

deaths Circulation 1982;65:457–64.

9 Cleland JG, Erhardt L, Murray G, Hall AS, Ball SG Effect

of ramipril on morbidity and mode of death among vivors of acute myocardial infarction with clinical evidence

sur-of heart failure A report from the AIRE Study

Investigators Eur Heart J 1997;18:41–51.

10 Greene HL Sudden arrhythmic cardiac death – nisms, resuscitation and classification: the Seattle perspec-

mecha-tive Am J Cardiol 1990;65:4B–12B.

11 Greene HL, Richardson DW, Barker AH et al Classification

of deaths after myocardial infarction as arrhythmic or

non-arrhythmic (the Cardiac Arrhythmia Pilot Study) Am J

Cardiol 1989;63:1–6.

12 CAPS Effects of encainide, flecainide, imipramine and moricizine on ventricular arrhythmias during the year after acute myocardial infarction: the CAPS The Cardiac

Arrhythmia Pilot Study (CAPS) Investigators Am J Cardiol

1988;61:501–9.

13 Epstein AE, Carlson MD, Fogoros RN, Higgins SL,

Venditti-FJ J Classification of death in antiarrhythmia trials J Am

Coll Cardiol 1996;27:433–42.

14 Bigger JTJ Prophylactic use of implanted cardiac tors in patients at high risk for ventricular arrhythmias after coronary-artery bypass graft surgery Coronary Artery

defibrilla-Bypass Graft (CABG) Patch Trial Investigators New Engl J

Med 1997;337:1569–75.

Box 2.2 Classification of arrhythmic death by

cer-tainty of data

with documented VF or VT within 10 minutes of

death

absence of non-cardiac death should be

recorded

compatible with instantaneous death where

his-tory does not indicate severe disease

incompati-ble with continued life for some time

absence of non-cardiac death should be

recorded

of pulse without prior collapse of the circulation

but no further data

18 Hartikainen JE, Malik M, Staunton A, Poloniecki J, Camm

AJ Distinction between arrhythmic and non-arrhythmic

death after acute myocardial infarction based on heart rate

variability, signal-averaged electrocardiogram, ventricular

arrhythmias and left ventricular ejection fraction J Am Coll

Cardiol 1996;28:296–304.

19 Pratt CM, Greenway PS, Schoenfeld MH, Hibben ML,

Reiffel JA Exploration of the precision of classifying

sud-den cardiac death Implications for the interpretation of

clinical trials Circulation 1996;93:519–24.

20 Grubman EM, Pavri BB, Shipman T, Britton N, Kocovic DZ.

Cardiac death and stored electrograms in patients with

third-generation implantable cardioverter–defibrillators J

Am Coll Cardiol 1998;32:1056–62.

21 Pires LA, Lehmann MH, Steinman RT, Baga JJ, Schuger

CD SD in implantable cardioverter–defibrillator recipients:

clinical context, arrhythmic events and device responses J

Am Coll Cardiol 1999;33:24–32.

22 Guidelines Guidelines for Clinical Intracardiac

Electrophysiological and Catheter Ablation Procedures A

report of the American College of Cardiology/American

Heart Association Task Force on practice guidelines.

(Committee on Clinical Intracardiac Electrophysiologic and

Catheter Ablation Procedures) Developed in collaboration

with the North American Society of Pacing and

Electrophysiology Circulation 1995;92:673–91.

23 Swedberg K, Kjekshus J Effects of enalapril on mortality in

severe congestive heart failure: results of the Cooperative

North Scandinavian Enalapril Survival Study

(CONSEN-SUS) Am J Cardiol 1988;62:60A–66A.

24 Swedberg K, Held P, Kjekshus J, Rasmussen K, Ryden L,

Wedel H Effects of the early administration of enalapril on

mortality in patients with acute myocardial infarction.

Results of the Cooperative New Scandinavian Enalapril

Survival Study II (CONSENSUS II) New Engl J Med

1992;327:678–84.

25 Goldman S, Johnson G, Cohn JN, Cintron G, Smith R,

Francis G Mechanism of death in heart failure The

Vasodilator-Heart Failure Trials The V-HeFT VA

Cooperative Studies Group Circulation 1993;87:VI24–

VI31.

26 Cohn JN, Ziesche S, Smith R, et al Effect of the calcium

antagonist felodipine as supplementary vasodilator therapy

in patients with chronic heart failure treated with

enalapril: V-HeFT III Vasodilator-Heart Failure Trial

(V-HeFT) Study Group Circulation 1997;96:856–63.

27 SOLVD-t Effect of enalapril on survival in patients with

30 O’Connor CM, Carson PE, Miller AB et al Effect of

amlodip-ine on mode of death among patients with advanced heart failure in the PRAISE trial Prospective Randomized Amlodi-

pine Survival Evaluation Am J Cardiol 1998;82:881–7.

31 Packer M, Bristow MR, Cohn JN et al The effect of

carvedilol on morbidity and mortality in patients with chronic heart failure U.S Carvedilol Heart Failure Study

Group New Engl J Med 1996;334:1349–55.

32 BHAT A randomized trial of propranolol in patients with

acute myocardial infarction I Mortality results JAMA

35 Singh SN, Fletcher RD, Fisher SG et al Amiodarone in

patients with congestive heart failure and asymptomatic ventricular arrhythmia Survival Trial of Antiarrhythmic

Therapy in Congestive Heart Failure New Engl J Med

38 Olsson G, Wikstrand J, Warnold I et al Metoprolol-induced

reduction in postinfarction mortality: pooled results from

five double-blind randomized trials Eur Heart J

41 Kober L, Torp PC, Carlsen JE et al A clinical trial of the

angiotensin-converting-enzyme inhibitor trandolapril in patients with left ventricular dysfunction after myocardial infarction Trandolapril Cardiac Evaluation (TRACE) Study

Group New Engl J Med 1995;333:1670–6.

42 Echt DS, Liebson PR, Mitchell LB et al Mortality and

mor-bidity in patients receiving encainide, flecainide, or

placebo The Cardiac Arrhythmia Suppression Trial N Engl

J Med 1991;324:781–8.

43 Waldo AL, Camm AJ, deRuyter H et al Effect of d-sotalol

on mortality in patients with left ventricular dysfunction

after recent and remote myocardial infarction The

SWORD Investigators Survival With Oral d-Sotalol Lancet

1996;348:7–12.

44 Burkart F, Pfisterer M, Kiowski W, Follath F, Burckhardt D.

Effect of antiarrhythmic therapy on mortality in survivors

of myocardial infarction with asymptomatic complex

ven-tricular arrhythmias: Basel Antiarrhythmic Study of Infarct

Survival (BASIS) J Am Coll Cardiol 1990;16:1711–18.

45 Ceremuzynski L, Kleczar E, Krzeminska-Pakula M et al.

Effect of amiodarone on mortality after myocardial

infarc-tion: a double-blind, placebo-controlled, pilot study J Am

Coll Cardiol 1992;20:1056–62.

46 CIBIS A randomized trial of beta-blockade in heart failure.

The Cardiac Insufficiency Bisoprolol Study (CIBIS) CIBIS

Investigators and Committees Circulation 1994;90:1765–

73.

47 CIBIS The Cardiac Insufficiency Bisoprolol Study II

(CIBIS-II): a randomised trial Lancet 1999;353:9–13.

48 Pitt B, Zannad F, Remme WJ et al The effect of

spironolac-tone on morbidity and mortality in patients with severe heart failure Randomized Aldactone Evaluation Study

Investigators New Engl J Med 1999;341:709–17.

49 Julian DG, Camm AJ, Frangin G et al Randomised trial of

effect of amiodarone on mortality in patients with tricular dysfunction after recent myocardial infarction: EMIAT European Myocardial Infarct Amiodarone Trial

left-ven-Investigators Lancet 1997;349:667–74.

50 Cairns JA, Connolly SJ, Roberts R, Gent M Randomised trial of outcome after myocardial infarction in patients with frequent or repetitive ventricular premature depolari- sations: CAMIAT Canadian Amiodarone Myocardial

Infarction Arrhythmia Trial Investigators Lancet 1997;349:

675–82.

Patients recovering from or afflicted with severe illness

face many potential complications that cloud their future

and make the choice of treatments difficult This is

partic-ularly true of sudden cardiac death syndrome,1 chronic

congestive heart failure,2 and myocardial infarction.3 A

common concern among these conditions is predicting

whether the patient will experience sudden

tachyarrhyth-mic death.4 Although many clinical and

electrocardio-graphic factors have been identified that have some

degree of association with the chances that an individual

will experience a life-threatening tachyarrhythmia, much

work still needs to be done, and the appropriate design

and execution of clinical studies is a necessary foundation

on which this work must be built The following material

attempts to outline the biostatistical (that is, inferential)

requirements for identifying and validating risk

stratifica-tion and its use in selecting and validating the appropriate

therapy for different risk strata In addition, although the

appropriate use of statistical methods that address the

needs of risk-stratification studies can be no more than

outlined in a single chapter such as this, such methods

are described, reasons for their selection given, and

help-ful references are cited for the interested reader to learn

more details regarding their implementation, including the

underlying assumptions, computational considerations,

and pitfalls

Definitions

Risk stratification

Usually, risk stratification is the process of assigning

indi-viduals to homogeneous groups based on assessed risk in

order to differentiate treatment efficacy or to predict

com-mon outcomes.5Those groups with high risk can be given

more aggressive prevention or monitoring and those with

lower risk can be spared treatment Ideally, risk

assess-variables and the events can be expected For the purpose

of this discussion, we will expand the definition of riskstratification to refer to the assignment of an individualprobability or risk of sudden cardiac death

Other activities resemble risk stratification, in that theyattempt to categorise individuals based upon a set of fac-tors associated with a condition of interest These activi-ties include mass screening, diagnosis, identification ofrisk factors, and prognostic staging (for example, in can-cer) Mass screening6 identifies asymptomatic individualsundergoing diagnosis of an underlying disease Diagnosis7identifies precisely (or rules out) an underlying disease,but does not necessarily dictate a particular therapy (forexample, one may not exist, or the therapeutic decisionmay require additional indications) Like screening, identi-fication of risk factors8 assigns to (usually) asymptomaticindividuals a probability of developing disease, but, asopposed to screening, diagnosis or risk stratification,before any diagnosable disease is present for the purpose

of preventing the development of disease Prognostic ing,9 which focuses on the extent of disease progression(as in staging cancer) rather than specific mechanisms offurther morbidity, is perhaps the closest to risk stratifica-tion; it is also used to predict outcomes and tailor thera-pies Although in the following we will not focus on thesimilarities of these activities to risk stratification, theirassociated methods can guide the development of meth-ods for risk stratification

stag-Terminology

In the following, measurement will refer to any ment of a characteristic of a patient, whether it isnumeric (for example, heart rate, age in years, number ofoccluded vessels, or QTC) or categorical (for example, sex,race, New York Heart Association Class, or presence ofmitral valve regurgitation) Any function of one or moremeasurements is also a measurement Any measurementthat may be associated with the probability of a patient

assess-experiencing SCD is a predictor; and specifying a

cut-point for the predictor, so that patients are divided into

two groups, one at high risk and one at low risk, creates a

test

Statistical considerations and methods

Prediction

Some claim that the main challenge in modern clinical

medicine is to make predictions regarding the patient.10

Risk stratification, in which the challenge is to employ

sta-tistical methods that can emphasise the most pertinent

data and filter irrelevant information, fits this description

exactly However, both misuse of and failure to use

appro-priate methods can lead to either serious error or

under-use of information Whatever under-useful information is

collected should be used optimally to predict sudden

car-diac death A number of pitfalls, well known in the

biosta-tistical literature, can be identified in advance and,

therefore, avoided by the careful investigator These

pit-falls are often overlooked as fine points, even by able and

experienced researchers Traditionally, statistical methods

have focused on parameter estimation and hypothesis

test-ing, as opposed to prediction, but in the last decade some

attention has been paid to predictive probabilities.11

Some of the problems peculiar to prediction are

overfit-ting, model misspecification, and variable selection

Although these are highly interrelated, a brief description

of each separately is in order Overfitting refers to the

tendency of a predictive model to appear to be more

accu-rate in the set of data on which it is developed (the

learn-ing sample) than can be expected in a new set of data

For example, if a population of CHD patients are

mea-sured at baseline and followed until a number of them

experience SCD, and if a predictive model using multiple

independent predictor variables is developed, the model

will correctly predict a certain fraction of SCD and a

cer-tain fraction of those not experiencing SCD However,

both theory and empirical evidence dictates that these

fractions will, on the average, be higher than the rates of

accurate prediction in a new set of patients, even though

they are drawn from the same general population.12

Thisstems from the tendency of model fitting methods to

overfit the model to the patients in the “learning sample”,

so that not only will the model adapt to the characteristics

of the learning sample that are representative of the larger

population, but it will also adapt to those characteristics

peculiar to that particular sample Thus, the investigator

must be chary of being overly optimistic about a model

extracted from a particular data set and use methods that

minimise this optimistic tendency

The second problem of model misspecification simply

refers to the potential that a parametric model (for ple, a multiple logistic model) does not truly represent theunderlying relationship between the predictor variablesand the probability of SCD If the true relationship is log-normal, say, instead of logistic, then the prediction will beoff, especially at the tails of the distribution Anotherpotential for misspecification comes when it is decidedhow the variables interact when they enter the model Dothe effects add, multiply, or exponentiate? Using modelsthat are flexible enough to adapt to the form that bestrepresents the underlying relation between the predictorsand the probability of SCD can combat this error On theother hand, unless care is taken, such flexible models canexacerbate the problem of overfitting by allowing themodel to reflect even more closely the idiosyncrasies ofthe learning sample

exam-Finally, the problem of variable selection, i.e decidingwhich potential predictor variables to include in the pre-dictive model, involves elements of the other two prob-lems, with the additional complication that in commonstatistical software there are many available methods ofaccomplishing selection, each with its own appropriatecontext, but rarely is it accompanied by good advice onwhen to use it Many variables potentially predict SCD incardiac patients, but only a few are likely to be helpful incombination If too many are used, the problem of overfit-ting becomes acute; if enough variables are used, appar-ently perfect prediction in the learning sample will result,with absolutely no usefulness in a new sample On theother hand, if too few are used and especially if the appro-priate set is not used, valuable predictive power will beleft on the table with no one benefiting

Use of multiple predictors

When several univariate predictors are available, for ple, heart-rate variability and late potentials, frequentlytheir individual predictive values are compared and the

exam-“loser” discarded in favour of the winner On the otherhand, when multiple predictors are considered in combi-nation, it is usually after the optimal cut-point has beendetermined for each predictor and the resulting tests com-bined However, both head-to-head comparison and com-bining tests can grossly underuse the predictive power ofthe combined predictors, as can be illustrated with a pre-viously published simplified example.5

Suppose, in a population with a low-risk group and ahigh-risk group, that Predictor 1 and Predictor 2 are underconsideration for discriminating members of one groupfrom members of the other Fig 3.1 shows the hypotheti-cal distributions of Predictor 1 for the two risk groups andFig 3.2 shows those for Predictor 2 Predictor 1 by itselfpredicts SCD only moderately well, but Predictor 2 ispractically useless by itself Casual observation would lead

to the conclusion that Predictor 2 should be discarded,

and only Predictor 1 used to evaluate a patient’s risk

based on the vertical line separating the two groups

However, suppose the predictors are correlated, and the

correlation coefficient (Pearson’s r) = 0.80 Fig 3.3

rep-resents the joint distributions of Predictor 1 and Predictor

2 for each risk group by two ellipses, which enclose about

95% of each group The reader can imagine them as

enclosing the base of two peaks that represent where the

majority of individuals in the two groups are located by

their values of Predictor 1 and Predictor 2 Determining

optimal cut-points for each predictor to create two tests

and combining them would define the four quadrants

(defined by the beaded horizontal and vertical lines)

shown in the Fig 3.3 It is easy to see that prediction

would be less than optimal no matter how those

quad-rants are combined On the other hand, an almost perfect

separation of the low-risk group from the high-risk group

is indicated by the diagonal dashed line This line is based

on a discriminant function.12 Although this simplified

example is based on idealised conditions, the general

prin-ciple that poor to moderate predictors can be combined toproduce an excellent prediction holds true in less idealcircumstances

This example can be used to illustrate four ciated, but valuable points:

underappre-● Measurements with little apparent discriminatingability by themselves may be powerful predictorswhen combined with other predictors

● High correlation with other predictors is not a goodreason for discarding a predictor

● Combining predictors can produce improved resultsover using tests based on individual predictors

● Dichotomising predictors prior to combining themcan weaken predictive power

and Cox regression.14

Figure 3.1 Distribution of hypothetical Predictor 1 for

high-risk and low-risk groups, showing moderate

discrimina-tion.

Figure 3.2 Distribution of hypothetical Predictor 2 for

high-risk and low-risk groups, showing little discrimination.

Figure 3.3 Joint distribution of Predictors 1 and 2 for high-risk and low-risk groups, showing high discrimination for linear combination.

Selection bias

Patients entering studies are highly selected by the time

period they enroll in a study; by consenting to a study; and

by study entry criteria intended to make patients easier to

follow or to ensure the patient will provide a worthwhile

amount of data In several studies, selection effects have

been demonstrated to affect even mortality For example,

in the Multiple Risk Factor Intervention Trial, which

ran-domised subjects to intervention or usual care to see if

modifying their risk factors could reduce death from

coro-nary heart disease, the number of cardiac deaths and

non-fatal events in the control group were much lower than

predicted by the risk equation formulated from the

Framingham Study and accounting for the explicit entry

cri-teria.15

Each successive eligibility screen reduced mortality

relative to the predicted rate by 27, 40, and 50%,

respec-tively In the Minnesota Colon Cancer Control Study, the

control group was 46% lower than the general population

from which they were recruited.16

Similar selection effectsshow up in other studies17,18

and one group of investigatorsfound selection to be common in prevention trials.19

Regression to the mean

If a predictor measurement fluctuates over time, even as

the basic risk for the individual stays constant, a relatively

high measurement of a predictor may more likely indicate

a true value closer to the average, whereas a low

measure-ment may be more likely to represent a higher value This

is a phenomenon known as “regression to the mean” For

example, in hypertension, an individual blood pressure

measurement may vary as much as 15 mmHg from time to

time during the day,20

but each individual measurementdoes not influence long-term risk; only the average about

which the variation occurs is believed to be relevant In a

previous publication,5

it has been shown that regression

to the mean could result in an apparent drop of about

10 mmHg and, with a moderate sample size, this drop

would appear to be a highly “statistically significant” result

even if no treatment is applied

By definition, risk stratification selects individuals with

extreme values of inherently variable biological

measure-ments, so regression to the mean invariably occurs For

example, if patients with a measured heart-rate variability

(HRV) in the lowest quartile on a 24-h Holter recording

are entered into a study, a comparison of the follow-up

HRV to the baseline values will be biased; even the

con-trol group will show an artefactual increase in HRV

Further, if the relationship between HRV and outcome

events is estimated from these biased baseline data, the

actual number of events will be greater than the number

of events predicted, if unbiased values are used for

predic-tion Care must be taken to avoid misinterpreting the

comparison of risk-stratification scores when they areremeasured after selection

Low structure methods

Although the development of optimal predictions usingmultiple predictors presents substantial technical chal-lenges from a statistical point of view, there are a variety

of approaches available Discriminant analysis, a statisticaltool closely related to logistic regression that optimallycombines multiple variables to predict the status of indi-viduals under certain assumptions, has been available for

30 years.21

Because it is often difficult to verify thoseassumptions, one should use more recent “low-structure”methods, such as classification and regression trees,22

eralised linear models,23

gen-generalised additive models,24

neural networks,25

and general smoothing techniques26

;these require fewer assumptions and provide flexibility toaccount for more complex relations between predictorsand the probability of an event

Prior knowledge

The prediction equations that are developed must beupdated as new information becomes available, either torefine predictions using existing predictors, or to incorpo-rate new predictors Bayesian methods not only provide aframework for previous information to be updated by newinformation, but also provide a coherent foundation forpredictive inferences.27

Such methods lend themselves toevidence-based medical practice.28

The acceptability ofBayesian approaches in science is evidenced by the factthat their use in the literature has nearly doubled over thelast 10 years, marking a turn in the practice of statistics.29

Combining Bayesian methods with the low-structuremethods described provides the solution to the problem

of predicting sudden cardiac death from multiple tors without precise physiologic theory to guide the speci-fication of parametric statistical models Cross-validationtechniques can be used to assure that the selection of theprediction function is unbiased.30,31

predic-Confidence intervalscan be created for prediction probabilities that adjust formissing data and still guarantee nominal coveragerates.32,33

The Appendix to this chapter gives a cal example of an estimated predictor function and itsresulting positivity (the fraction of results that are posi-tive) and positive predictivity

hypotheti-Development of risk stratification

Setting out to risk-stratify a defined population (forexample, all CHD patients), the investigator wishes to useclinical, laboratory, and electrocardiographic information

● to develop a candidate scheme, and

● to validate the scheme

Each of these phases requires separate specific steps

The development phase includes:

1 Identifying a target population

2 Selecting candidate variables

3 Collecting data on a representative sample of the

pop-ulation

4 Developing a risk-stratification scheme with

associ-ated risk estimates

5 Incorporating prior information

6 Updating the risk estimates in the stratification

scheme

Target population

The target population must be one that is readily identified

by history or some other symptom or sign, and in which

we want to subcategorise the members by their risk of

SCD The population can be as narrowly defined as, for

example, those with long-QT syndrome, or as broad as the

population of the world What is important is that it be

well defined, so that the results can be analysed and

gener-alised appropriately For example, in a narrowly defined

population like the long-QT example, which has a high

underlying rate of SCD, more costly and invasive tests can

be used to risk stratify, and, since the population is clearly

defined, the applicability of the stratification to others with

long-QT is fairly straightforward For the population of the

world, the rate of SCD is fairly low relative to the long-QT

group, so less invasive and costly tests could be used, and,

again because of the clear definition of the population, the

generalisation would be straightforward as well Once a

suitable target population is identified, then a

representa-tive sample of that population must be selected to provide

information about predictors and SCD rates

Candidate variables

Note that the presence of long-QT syndrome, which

defines the first population, might be used, in part, to

triglyceride levels;

● electrocardiographic parameters, such as heart-ratevariability, signal-averaged ECG, or QT interval What is important is that the chosen variables havesome plausible association with SCD and that they beobtainable from members of the target population Although it is assumed that all of the candidate vari-ables have some plausible connection with the chances ofSCD, it will be useful in distinguishing between testsused in clinical decision-making and the underlying predic-tors that are used to form the test In the following, thenotion of a “predictor” will be distinguished from that of a

“test”, by defining a test to be any procedure that results

in a binary decision such as “positive/negative”, “rulein/rule out”, “at risk/not at risk”, and the like Predictors,

on the other hand, can be binary, polytomous (having tiple categories, either with some ordering or unordered),continuous, or any combination of these characteristics, aslong as they contain information about the chances ofSCD To illustrate the difference, note that heart rate vari-ability (HRV) expressed as a standard deviation is a contin-uous predictor of SCD but, by defining a critical value forHRV below which the patient is considered at high risk,creates a test A second example is the number of VTepisodes on 24-h Holter monitoring As a predictor ittakes on ordered categories 0, 1, 2, etc but, if it isdichotomised by noting the presence or absence of VT, itcreates a test

mul-Data collection

Data must be collected twice on the sample of the tion to develop risk strata First, baseline information,consisting of the value of each candidate predictor consid-ered for risk stratification, must be collected on each indi-vidual at the beginning of the follow-up period Collectingthe results of tests, as will be illustrated, is inadequate,because some information could be lost in the collapsing

popula-of the candidate predictors Although risk stratificationcan be performed when some data are missing for someindividuals, minimising the number of missing values willmaximise the usefulness of the risk stratification