Memorandum by North of Tyne Communicable
Disease Control Unit, Dr Nicol Black
Surveillance information should be
the life-blood that powers clinical practice and public safety.
It is not.
Current surveillance does not provide
an effective early warning system for outbreaks.
Front-line agencies need the tools
to do the job of health protection.
Inaction may be costly and fatal.
Prototype systems are available and
evaluation could commence immediately.
Dr Black has been a full-time Consultant in
Communicable Disease Control since 1989, at Regional and District
level. The North of Tyne Communicable Disease Control Unit is
a busy team, comprising 12 people, serving a population of 775,000
in Northumberland, Newcastle and North Tyneside. It has been innovative
in developing early-warning surveillance systems, risk assessment,
teaching outbreak investigation, prevention and adult immunisation.
(i) Current communicable disease surveillance has got
itself into a conceptual backwater, which is stifling its potential
to influence front-line decision-making (whether in clinical practice
or public health), to make new discoveries about novel threats
that arise or to exploit the vast resources of clinical data in
the NHS by modern electronic technologies.
(ii) In general, surveillance is not "owned"
by front-line staff to prioritise actions but by remote institutions,
patiently answering yesterday's questions. Much surveillance is
reactive, out-dated, monolithic, slow to adapt and frequently
(iii) Current surveillance does not provide
an effective early warning system for outbreaks.
(iv) There is no point in building a better
smoke-detector if the fire service cannot respond adequately.
The Department of Health strategy, set out in Getting Ahead
of the Curve,
sets out proposals for a Health Protection Agency (HPA), to improve
the resources and organisation available, which should improve
the "fire service". Though it envisages increased surveillance,
it contains no clear surveillance strategy beyond "more of
the same", ie it has not been far-sighted enough to identify
the need for early warning systems.
(v) The perspective of active, innovative,
surveillance systems for early warning and local risk analysis
simply is not there. This contrasts with the dramatic changes
in the USA since the revision of the bioterrorist threat since
11 September 2001. At a National Syndromic Surveillance Conference
in New York in September 2002, with 450 delegates, which focused
on early warning systems, there was only one participant from
(vi) The battle for protecting the health
of communities will be won or lost by the front-line agencies.
They need the tools to do the job. This surveillance perspective
needs urgent action to disseminate and develop if costly and fatal
inaction is to be prevented.
(vii) In a world increasingly concerned
with a deliberate bioterrorist threat, early detection of clusters
of non-specific prodromal symptoms may well be the sternest test
of how well designed our surveillance systems are.
(i) Surveillance information should be the life-blood
that powers clinical practice and public safety. It is not. It
should monitor changing risk factors, to allow focused prevention
activities, and assessment of the effectiveness of interventions
in such a timely manner as to influence them. It does not. It
should be sensitive to the early appearance of new hazards as
they occur, whether local or national, and be an effective alerting
system. It is not.
(ii) Why not? Current systems are antiquated,
mainly reflecting bygone needs. They have become centralised in
institutions divorced from day-to-day decision-making in the field,
with a gulf between them and local public health practitioners.
This has led to a steady decrease in interest and participation
by the very people who should be the "eyes and ears"
of an alert intelligence system. Local communicable disease control
systems are under-resourced, under-appreciated, marginalised and
often stagnant. Surveillance should be such an essential informative
component of clinical and public health practice, that not being
an integral part of these activities would be unthinkable. Just
as we need established knowledge (eg from books and papers) we
need contemporaneous information about the nature and level of
risk and about how well our care systems are functioning in practice.
In short, we need to be much more data-driven than ever before.
3. WHAT DO
(i) In general, the twin aims of surveillance of communicable
disease are to detect outbreaks and monitor trends. It is a mixture
of statutory clinical notification and voluntary laboratory reporting.
(ii) Data gets into the system by meeting
a case definition, which can be explicit (as in the US) or implicit,
as in a clinical diagnosis. Something is either in or not in;
there is no gradation of probability (see 4.i). If a diagnostic
label turns out later to be something else, there is no easy way
of refining or changing it (see box 1). There is little "depth"
in the current case-based systems; "best guess" clinically
is indistinguishable from immutable truth.
BOX 1: MEASLES NOTIFICATION
Measles is clinically notifiable. A few years
ago, serology, by collecting gingival exudates, was introduced.
More than 90 per cent of these notifications were found to be
conditions other than measles. However, there is no method of
re-classification and they are not de-notified, to preserve continuity
with notifications from previous years. We thus have a system
that has measles notifications with a caveat that they are not
(iii) Trend analysis. Most current activity
is directed towards trend analysis. To ensure comparability the
data collected is very prescriptive and usually laboratory based.
To ensure continuity it does not change its definitions over time.
Analysis is usually at large population level and seldom gets
into underlying behaviours. It is highly ritualised, slow, frequently
obsolete and often divorced from decision-making. New, or enhanced,
topic-based, non-statutory surveillance schemes have been introduced
in recent years which are an improvement, eg TB, HIV and sexually
(iv) Outbreak detection. There is a clinical
notification system which was designed for rapid alerting of public
health. In practice it does not work very well as most busy clinicians
ignore it and seldom notify except in rare or serious cases. Indeed,
most notifications are post-hoc following receipt of a positive
laboratory report. Laboratory results provide firm data, but their
usefulness is hampered by two factors. Firstly, for many infections
the efficiency of detection is likely to be very low. Of all those
exposed, the numbers of those infected, who then become ill, present
to their doctor, have specimens collected, which turn out to have
positive isolates, which are then notified to public health progressively
diminishes, eg for common enteric illnesses, to probably around
5-6 per cent. Secondly, the delay between exposure and notification
is often eight to nine days, which is much too long to be effective
as an alerting mechanism. Any significant hazard will have had
a lot of people exposed to it by the time laboratory results become
available. Laboratory reporting is, however, particularly useful
for picking up rare organisms or unusual subtypes both locally
(v) Only a relatively small proportion of
outbreaks come to light from routine surveillance. Most come to
light by complaints from the public, observations of unusual occurrences
by health professionals, the media or serendipity. Why does routine
surveillance not play a bigger role? The two functions of detecting
outbreaks and monitoring trends have very different information
needs. Outbreak detection demands timely information, which will
of necessity be incomplete and provisional. Trend analysis, by
contrast, needs precise, accurate, information confirmed by laboratory
tests, which inevitably means it will be delayed (fig 1).
(vi) Current surveillance practice has emphasised
trend analysis at national level, where the need for precision
takes precedence over the need for promptness. As such, most surveillance
activity exists to support this with local public health needs
in a subordinate role. Outbreak alertness has mainly relied on
informal networks of clinicians and environmental health officers
to identify symptomatic people with suspicious circumstances.
4. ANALYSIS OF
(i) Trend analysis needs precise data, shorn of ambiguity,
to make it comparable between different places and over time.
The filtering by tight case-definitions, or by mandating laboratory
confirmation, means much partial, but potentially useful, information
is rejected and does not even enter the system (see box 2).
BOX 2: LOSS OF DATA WHICH HAS A PROBABILITY <50
A condition with a probability of 40 per cent
will be rejected by a system using binary logic. Hence three (or
more) cases, each with a 40 per cent probability, will all be
rejected. However, in real life, three patients in the same hospital,
each with a 40 per cent probability of having smallpox, would
have emergency alarm bells ringing.
(ii) The alternative is to create a people-based
rather than a case-based system, to divorce observations from
inferences, to record all the observations, with confidence estimates
if appropriate, and then apply whatever range of case-definitions
is appropriate to the accumulated data. This is the essence of
a syndromic surveillance database (see section 7). This gives
great scope for manipulating indices of probability or changing
the constituent components of the case definition. As always with
identifiable data, good security and confidentiality safeguards
(iii) This is the epidemiological approach
taken with outbreak investigations. Time, place, symptom and behaviour
data are collected for everyone potentially involved in the outbreak,
which is then subjected to a series of increasingly sharper case-definitions
as the precise nature of the hazard becomes more clearly defined.
This allows control over the degree of ambiguity in various case-definitions.
Essentially, this is the basis of a syndromic database. As it
has no pre-fixed data model it is hugely versatile and can be
used for any type of hazard (not just microbiological) and even
in novel situations where the parameters of the condition have
not yet been established (see section 7).
5. THE PROBLEM
(i) The main problem facing infectious disease in the
UK is lack of data exploitation "at the coal-face" by
health-care workers and public health professionals. Lack of contextual
data blunts diagnostic and treatment precision and prevention
activities lack focus. Technology is not a problem; imagination,
insufficient applied research and a long-term perspective, locally
and nationally, is.
(ii) Surveillance is divorced from local
epidemiology, particularly early warning of outbreaks and recognition
of non-laboratory-confirmed conditions. It does not illuminate
local variation in diagnosis and treatment, which varies widely
from place to place, especially antibiotic usage and resistance,
whose monitoring is still embryonic. Similarly, prevention is
often the Cinderella of communicable disease control, threadbare
and neglected, not least because of poor focus due to inadequately
identified specific risk or local transmission factors.
(iii) Historically, there has always been
an underlying pressure to collect "the minimum data-set"
ie only sufficient data for the immediate purpose intended. This
was rooted in sensible principles of thrift, economy of data collection
and handling and confidentiality. However, as soon as one question
is elucidated, the next is waiting to take its place and parsimony
has often prevented surveillance systems evolving to meet the
continually developing need. The obvious alternative is the ability
to "drill down" from the macro to the micro as and when
required. This means exploiting existing data collected for other
purposes eg clinical morbidity data or, logically, constructing
future data systems with multi-use clearly borne in mind. NHS
Direct was such a missed opportunity when it was initially set
up (see box 3).
BOX 3: CASE STUDYNHS DIRECT
When a caller contacted NHS Direct, basic demographic
data (age, sex, postcode) was collected, then the person proceeded
down sets of algorithms, specific to the nature of the complaint,
to an outcome, which was recorded. It was possible to extract
lists of these outcomes ie how many callers clinically had an
influenza-like illness, but this was not linked to the demographic
data. As the call centres could be receiving calls from anywhere
in the UK depending on demand, no linkage could be made to a specific
geographic population, hence no inferences could be made about
localised incidence, such as the spread of an epidemic.
(i) The policy intervention with the single greatest
impact on prevention and containment would be to create modern,
intelligent, surveillance systems, active and visible in every
locality, addressing both local and national priority issues.
(ii) A national surveillance strategy has
never before been defined. Though Getting Ahead of the Curve33,
identifies the very laudable case for "world class surveillance",
5.3-5.17, it does not articulate any strategic aims or "mission
statement". If health protection is in reality to be successfully
delivered, it will only be by alert, competent field teams making
sound risk assessments, based on relevant, up-to-date locally
applicable data. It is not clear in Ch six, section three that
this is recognised. As such, it does not go far enough. It appears
to be advocating "more of the same", enshrining many
of the limitations of current institutionalised practice, rather
than any clear-sighted vision of locally driven health protection,
alert to early warning signals and responsive to locally determined
variations in risk or transmission.
(iii) Unless such a forward-looking responsive
system is put in place we will fail to deliver public protection
effectively. We need the emphasis shifted towards fire-watching
and fire-prevention and away from elaborately crafted post-hoc
descriptions of the blaze.
(iv) Much of the conceptual work for this
has already been done and could be ready for widespread implementation
in the near future. The bedrock of successful health protection
is successful risk assessment and management. A conceptual model
for risk management in communicable disease control, based on
a quasi-probabilistic model, has been developed by a working party
with a grant from the Department of Healthwith
further work refining a dynamic risk management priority index
risk analysis depends on information acquisition (fig 2).
Fig 2Risk Assessment in the communicable
disease control function
(v) Methods for detecting emerging foci
of infection at local level, which dovetail with formal risk assessment,
have been developed.
In brief, these methodologies include developing systems of early
warning markers (see appendix 1) and improved space-time cluster
and pattern analysis. As they are syndromic-based, they have the
advantage of versatility and are adaptable for bioterrorist activities
or other conditions where the case-definition is unclear, eg in
newly emerging infections.
(i) A syndrome is a complex of signs and/or symptoms
whose coexistence is recognisable without necessarily pinpointing
the precise nature of the underlying disease, eg AIDS was an identifiable
new syndrome before the aetiological identification of HIV infection.
New diseases invariably are recognised firstly as syndromes, eg
BSE, Lyme Disease, Legionnaire's Disease, etc. It follows that
syndromic surveillance is a tool for identifying which signs and
symptoms are found to coexist, in a given time and place, than
would be expected by chance on the assumption of independence.
(ii) This gives unprecedented flexibility
in the type of data that can be collected, at whatever level of
accuracy is available and analyse it for known and unknown patterns.
(iii) A well-designed syndromic system will
reflect public health workflow, by capturing prevailing doubt
and uncertainty by dynamically coding for changing levels of confidence
in individual data items and by readily changing the search criteria.
This allows previously unidentified patterns to be searched for.
Mathematically rigorous belief diagrams make the analysis visible
and explicit. Also, it could incorporate rule-based inference
to run as an automated "expert system" in the background.
(iv) Because it reflects public health workflow,
it also provides a means of reconciling the spectrum of risk (a
continuous variable) to the outcome options (a discrete variable)
in an explicit, mathematically rigorous way, which is intuitively
identical to the existing process of epidemiological reasoning.
(v) Though syndromic surveillance is a very
recent development internationally,
particularly in response to bioterrorist threats, it has been
and in use for meningococcal disease and waterborne outbreaks
in a busy communicable disease control unit in the North-East
of England since 1995.
Another application being developed there is a detailed analysis
of clinical data (in meningococcal infection) for risk recognition
and its feasibility for smallpox surveillance in the US is being
A similar application could potentially track adverse effects
of vaccines or pharmaceuticals, with appropriate pattern recognition
and primary care morbidity data. Its applicability to pattern
recognition in unexpected situations, eg non-terrestrial environmental
syndromes in astronauts is under discussion with NASA.
(vi) A major consequence of the versatility
of a generic, person-based syndromic system is that a single system
could cover all aspects of all infectious diseases rather than
multiple specific applications, with their inevitable multiple
learning and support requirements.
(vii) The limitation of such a system is
that until it is mature as a decision-support system, ie semi-automated,
it is relatively labour-intensive. This confines its deployment
to teams with sufficiently sophisticated epidemiological skills
and capacity to conduct the analyses. Development work is currently
underway on this, but much more work is needed to simplify this
process and make it more efficient.
8. NATIONAL OR
It is not a case of either/or. Both are needed.
National/international surveillance is needed for the "big
picture", ie trend monitoring. Local surveillance is needed
for outbreak detection, targeting prevention and ensuring data
quality. It is the local level which is data-rich; eliminating
ambiguity in the data requires clarification with the information
providers, ie the clinicians and laboratories. Improved interest,
skill and competency in the local field teams will not only improve
local performance but is essential to improve the quality and
depth of nationally provided data.
9. THE FUTURE
(i) The future of infectious disease surveillance (and
many other health protection activities) will be syndromic surveillance
generating a risk priority index for local action.
(ii) The HPA surveillance strategy must
be balanced between local and national needs. This means in practice
being much more "bottom-up" than before.
(iii) As the main problem is that surveillance
is not seen as an important everyday tool and is not practised
by a sufficiently large number of people to illuminate operational
issues, it follows that the first objective is to increase capacity
and competence in the local field teams, both of which are envisaged
in the proposed Health Protection Agency.
(iv) The second objective is motivation
and enablement, which can readily be achieved with modest resources
by setting up "best practice" networking, skills workshops
or training placements in development units.
(v) At the same time, prototype systems
should be fully evaluated in practice, developing sources of risk
markers for early warning systems and the necessary analytical
methodologies and clarifying the scope of their utility.
(vi) These would be enhanced by developing
automated decision-support using rule-based inference. This would
constitute a major advance in supporting best practice risk-management
at local level and improve public safety.
(vii) In the long term, a strategy for linkage
between surveillance systems and other sources of NHS data should
The author gratefully acknowledges the invaluable
contribution of other colleagues, particularly Dr J. Michael Smith,
of Novosystems Software, and Mrs Liz Stokle, Senior Nurse Surveillance
Co-ordinator, North of Tyne Communicable Disease Control Unit.
The author also gratefully acknowledges financial
support for the project from Northumbria Water plc.
APPENDIX 1 EARLY WARNING SURVEILLANCE
(i) This is a system of increasing sensitivity
by identifying potential early markers of infection and following
up suspicious clusters by investigation, eg by active case-finding
(ii) There are many indicators which can
be used as provisional markers of infection, (see box2).
BOX 4: POTENTIAL PROVISIONAL MARKERS OF INFECTION
||Over-the-counter pharmacy sales|
|NHS Direct||Accident/Emergency attendance
|Nursing/residential home surveillance||Water utility customer complaints
|Food histories from Environmental Health Officer visits
(iii) Many of these are with people who have not yet
seen a health-care worker. These markers have the advantage of
being much more numerous than laboratory isolates and occur much
earlier. As such they increase the sensitivity of the system.
An aberration in one marker is inconsequential but simultaneous
aberrations in multiple markers, or space-time clustering, are
suspicious and warrant being analysed for consistency and corroboration
and further investigation, as necessary (fig.2, p4).
(iv) Because of their nature, early warning markers are
inherently ambiguous and the systems to store and analyse this
data need special provision to cope with ambiguity, uncertainty
and incompleteness. This has necessitated developing a syndromic
database (see s 7).
(v) The early-warning systems can be used anywhere. They
require no special systems. However, developing automated pattern
recognition would greatly reduce the tedium of manual inspection
and improve efficiency.
Getting Ahead of the Curve, Department of Health 2002. Back
The development of a risk management model for communicable disease
control in the United Kingdom. Kara-Zaitri C, Gelletlie R, Barnes
H, Black N M I, Walker D, Hatton P, Schweiger M, Wilson D. Proceedings,
Fifth Conference on Probabilistic Safety Assessment and Management,
Vol 4, p2247-53, Universal Academic Press, 2001. Back
The design, development and implementation of a dynamic risk management
priority index for communicable disease control in the UK. Kara-Zaitri
C, Black N M I, Gelletlie R, Schweiger M, Barnes H, Walker D,
Hatton P, Wilson D. Sixth Conference on Probabilistic Safety Assessment
and Management, Puerto Rico, June 2002. Back
Dynamic risk Management in Communicable Disease Control-the further
development of a Risk Management Model. Kara-Zaitri C, Black N,
Barnes H, Schweiger M, Gelletlie R. PHLS Annual conference on
Epidemiology and Control of Communicable Disease and Environmental
Hazards (Health Protection) Nov 2002. Back
Implementing risk assessment of emerging foci of infection methods
for early detection and evaluation. Black N M I, Stokle L, Smith
J M. Sixth Conference on Probabilistic Safety Assessment and Management,
Puerto Rico, June 2002. Back
The emerging science of very early detection of disease outbreaks.
Wagner M, Tsui F-C, Espinato J, Dato V, Sittig D, Caruana R et
al. Journal of Public Health Management and Practice 2001; 7;50-8. Back
Syndromic surveillance; sensitivity versus uncertainty and ambiguity.
A local health department system in the North-East of England.
Black N M I, Stokle L, Smith J M. National Syndromic Surveillance
Conference, New York, Sept 2002. Back
Early detection of emerging foci of infection; finding a guiding
star. Black NMI, Smith J M, Aldrich T E. Faculty of Public Health
Medicine, July 1999, Glasgow. Back
Severity of meningococcal disease; a method for peri-admission
risk assessment. Black N M I, Stokle L, Kara-Zaitri C. Project
proposal 2001. Back
Personal communication, Dr Barbara Watson, Philadelphia Department
of Health 2002. Back
Personal communication, Dr Michael Stamatelatos, Head of Safety
and Reliability, NASA, 2002. Back