You are on page 1of 7

The patient medical record as a database

B. E. Jones and M. A. Ould


King's College Hospital Computer Centre, 97 Denmark Hill, London SE59RS

The patient medical record, viewed as a life-time's accumulation of health care information,
presents interesting problems in database design. The requirement may be a database where
structure is partially controlled by users at terminals.
(Received March 1974)

Introduction doctors and nurses, must have a clear idea of what is in the data-
An early real-time filing system for medical records at King's base, how it is organised, and how to use it—just as they have
College Hospital soon demonstrated a need for planning with respect to the elements of the paper record system. The

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


patient-based information on a grand scale. The filing system database may prove to be very complicated in fact, but as far
was adequate for the early real-time applications but soon all as possible, this should not cloud the users' conception of it.
kinds of patient-based batch systems began to appear, and even
real-time applications acquired their own archive files. This The day-to-day information system
fragmentation of related data sets suggests a database solution. This is by far the most important area of design because it is
Computerised patient records may be a 'natural' for database where the bulk of the database is gathered and distributed.
technology since there is so much replication of data across the Here the quality of the database is established and so is its value.
information systems in the health care services. The content of The day-to-day information system is already well understood
a referral letter, say, from General Practitioner to Clinic, or in the sense that computers have already made some impact,
vice versa, should largely be recorded somewhere already. particularly in hospital laboratories. Here we are concerned
There are numerous examples of this kind both within the units with a larger pattern and less with the detailed systems in wards,
of health care—hospitals, health centres, etc.—and across them. clinics, surgeries, etc.

Primary objectives The long-term information system


The database approach is so fundamental that the starting The intention is to build up a computer-based medical record-
point of design must be 'better patient care'. Perhaps the prim- for-life for each patient. Superficially this is an obvious strategy
ary purpose of acquiring a computer-based medical record is which becomes more feasible with the advent of computers.
that one may thereby achieve acceptable measures of aspects of It should be possible to derive valuable supplements to the
patient care—even if the underlying concept remains elusive. traditional patient history, saving doctors' time or improving
If such measurements prove possible, then the potential for patient care or both. The long term record should also add
improving these aspects is not in doubt. another dimension to the possibilities of clinical analysis of the
A step towards such an information system was embodied in database. In practice, there are various problems and the kind
the aims of the King's College Hospital Medical Computing of benefits expected must take many years to be realised.
project. The chief characteristics defined for this system were to Our objective in this area must be to develop an additional
be: patient data structure, within the database, called the front
1. The storage of medical records containing information of end, which is to provide the essential data required of each
both disciplined and unrestricted nature. life-long medical record but not the detail. Of course, we have
yet to establish more than a vague notion of what the essential
2. The handling of data flow around the hospital. data is but we do expect to generate some of it via the day-to-
3. The gathering of statistics for a management information day information system. Thus, for the time being, the structure
service. and content of the long-term information base will be mainly
To provide a generalised solution to these objectives we derived from the day-to-day system—although the latter is
recently sought to establish a database which would provide: logically a subset of the former.
1. A day-to-day, patient-based medical information system Preservation of data for long periods—perhaps 70 years—
gradually to replace much of the paper-based medical introduces two further problems of maintaining relevance and
records, and to speed theflowof information throughout the compatibility, which are beyond the scope of this paper.
health care services.
The statistical information system
2. A longer-term information storage system supporting the The challenge of clinical and administrative analysis requires
patient medical record for life and interfacing with all forms careful consideration. The problem is that an information
of medical care. system that supports rather than dictates to the medical record
3. An information base suitable for various types of clinical and cannot ever be either wholly accurate or complete. Computer
administrative analysis. techniques may compare favourably with paper systems in this
Additionally, the database design must take seriously the respect, but that is not saying much.
matters of privacy and legality, which are associated somewhat If the day-to-day information system integrates well with
informally with the paper-based medical record. Inattention to users, there will be two-way information flow between the data-
these aspects at this stage, in the face of public sensitivity to base and users and it is reasonable to suppose that the quality
database technology, could be disastrous. of information recorded will improve dramatically. If the
There is, perhaps, one further major objective which is much computer system collects data with no real feed-back to users,
less tangible. Users of the information system, particularly then the database cannot be regarded as good for analysis.

Volume 17 Number 4 295


Whatever the quality of data the statistical techniques used on for the same patient. If such trouble can be detected by the
the database should be more akin to census studies than system, this is all to the good, but even if a user discovers the
direct counts of admissions, etc. Then, up to a point, the sheer flaw (or causes it) it is essential that it be possible to put things
volume of the database can be expected to eliminate problems straight. The natural solution is to provide simple batch facili-
of data quality. ties for adjusting or editing the database under strict supervision.
Analysis of the database does raise specific problems of the
compatibility of data and whether data should be coded as is Structure as seen by the user
sometimes argued, but we ignore these issues here. General
Users, in this context, could be any medical or paramedical
Privacy persons, inside or outside a hospital, with a valid interest in the
The Younger Committee in England and the Department of patient medical record. It is convenient though, to consider the
Health, Education and Welfare in the United States have both hospital patient medical record as representing most of the
recently reported the findings of their respective investigations problems for this section. For users the hospital medical
into the subject of database privacy and the citizen. record folder is an accepted fact of hospital life. It may be
It is a telling comment on the ambiguity of the subject that the disorganised, incomplete, illegible or unobtainable but its
Younger Committee feels there is no need for urgency in respect shortcomings, and the reasons for them, are largely accepted.
of detailed controls and have presented an advisory document Moreover, what structure is intended is simple and users know
full of 'shoulds' whilst the Americans come to the opposite how to handle it; they know what to expect and how to find
conclusion and arm their proposals with imperatives. In the what they want.

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


final analysis, data will only be as private as its users are When we introduce a computer-based medical record, we are
prepared to make it. suppressing the familiar in favour of the unknown. If users have
Many of the Younger Committee principles are matters for little or no concept of what the new system is about, we can
the Medical Profession with respect to patient records, rather confidently expect trouble in various forms. In particular, it is
than requirements for database design. In particular, database possible that the early medical recording experiments at King's
content, how it is used, how long it is kept, etc. are not easily College Hospital became largely a 'data collection system'
defined as Younger would recommend. On the other hand, the because the doctors had a poor conception of the structure of
database design can ensure, at some cost, proper access controls the record they were creating and how it could be accessed.
and other mechanisms in support of Younger. This is clearly not just a problem for the database; it affects all
systems more or less. But with the database we introduce, of
Legal requirements necessity, more structure to the medical record than ever before.
This area is not well defined, either. For example, hospital There is a strong requirement, then, that the database should
inpatient and outpatient information must be held for six years be portrayed to users in a form they can understand and which
after the last attendance at the hospital by the patient. A medical can reflect their thought processes.
record-for-life appears to meet this kind of requirement almost The discussion of structure is therefore divided into two
by definition. parts—that presented to the user, and the internal software
However, the legal problem is essentially one of providing structure which will represent the medical record for the user.
evidence in cases of litigation. Then the database must provide
information with the legal status of 'notes taken at the time'. What and where is the total patient record?
This implies the reconstruction of a patient's record to the state At the simplest and highest level, there should be a simple
it was in at a particular moment, and identification of the source division of patient data between, on the one hand, the database,
of all the content. and, on the other, the continuing but diminished data storage
In practical terms, then, the database must record who makes systems currently used such as the patient folder, headboards
additions and changes to the patient data and when. Changes, and the Kardex system in the hospital environment. This
comments, etc. must be effected without loss of the amended division will, of course, change as the database becomes more
data so that the earlier state can be reconstructed. comprehensive and is really controlled by applications
development, but the patient database will exclude, as policy
Resilience if not necessity, certain types of information (e.g. graphic,
It is logically 'impossible' for a patient to be re-admitted to photographic, etc.).
hospital unless he has been already discharged in some sense. Within the database, there are some more difficult conceptual
The designer of a computer-based admission system might problems. There seem to be three distinct areas:
therefore wish to force users to record a discharge before per- 1. The front end and patient medical record-for-life concept.
mitting the re-admission procedure. While this may be accept- 2. Summary creation to provide communication between
able to the user, it is a matter for detailed systems study for the parallel episodes of treatment.
particular application and not an area where the database
should dictate the possibilities (as it might well do otherwise). 3. Structure of data within individual episodes of treatment.
This should demonstrate some of the dangers in database At this point we begin to wonder whether the database should
design. It is essential to relate the database to medical recording concern itself with such a detailed level of user requirement,
and other hospital practices, but, in so doing, one must not especially bearing in mind the requirements for resilience. Our
build in too many constraints on individual applications. An answer is that the medical record must embody structure to
elusive balance is required to achieve a level of what might be permit practical methods of access. It is possible for the same
called resilience without 'going overboard' for endless flexi- technical structure to support distinct views of the database
bility. structure.
Another consideration, which is also a matter of resilience, Without valid structure, the database will be disastrous—like
is manual correction of the database. So far we have been a library with books distributed randomly on the shelves.
concerned with the ability of the system to sort itself out In the ensuing discussions, it should be apparent that any
despite minor 'abuses' or disruptions to the normal information structure will impinge on current medical practice. By attempt-
flow. The database may nonetheless accept some significantly ing to minimise this effect we may destroy the potential of the
erroneous data. For example, two separate records may develop database, whereas the more radical approaches, for example,

296 The Computer Journal


that of Weed's problem-oriented record (Weed, 1971), could be ment of the patient can be recorded in terms of other segments
symbiotic with the database technique. of the record (e.g. investigations, diagnosis, therapy, clinical
comment, etc.). These management segments may not provide,
Summaries and the front end of themselves, adequate structure.
There is a presumption here that users, particularly doctors, Earlier medical recording experiments at King's have shown
require resumes of the current episode, of other current treat- that there is considerable useful structure in the history and
ment and of past episodes. The GP and hospital summaries are physical examination segments. In other words, there can be a
steps in this direction, although a computer system can, hope- visible pattern in these segments and not just a series of state-
fully, go much further. ments. This was not the case for management statements and
In simple terms, the hospital summary provides a resume of a there seems to be a need to bring out the medical logic in the
hospital episode and similar levels of summary are conceivable management segments. Two possible approaches have been
for other episodes (e.g. outpatients, health centres, etc.). We considered.
can then extrapolate this principle to provide the front end as If one retains the primary segmentation of the statements
a summary of such summaries. This is a tidy concept, provided relating to one episode, it could be made possible for users
that there are workable ways of maintaining each level of explicitly to cross-link statements in the management segments
summary, because the user gets a simple picture of the past and possibly those in the history and physical examination. In
from the front end and otherwise has only the detail of the simple terms, when a user ordered an investigation, for example,
current episode to deal with. he could link it to some other statement in the record. The
Unfortunately such a simple view collapses once one attempts resulting sequences of statements then should reflect some

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


to cope with two or more parallel episodes (e.g. attendance at 'line' of clinical thought. Unfortunately it is not only difficult
distinct outpatient clinics). The paper record may not be to provide an adequate pattern of links technically but we find
effective in this situation either; but then users are not expecting it difficult to comprehend the real advantage to the users of
resumes. The essence of the problem is to find a balance thinking in these terms. We feel that this kind of additional
between keeping information entirely separate and combining structure is conceptually difficult for users to grasp.
it for the distinct but parallel episodes. This appears to require An alternative may be some derivative of the problem oriented
the propagation of all salient points of each episode among all record (Weed, 1971) for the management segments. After
those episode records running in parallel. Now we have a recording conventional history and physical examination
situation where salient information must be recognised when it segments, management data is recorded by 'problems' (i.e.
is generated, so that it can be propagated for other users. signs, symptoms, diagnoses. . .) which the doctor has elected to
In this more complex database structure, the user 'sees' a manage. So, apart from history and physical examination
front end summary, plus the statements that make up the segments, users see a problem list for each patient. Within each
episode with which he is associated, plus the salient statements problem are recorded comment, investigations, treatment,
from other episodes which may be current for the same patient. management plans, etc. Since Weed undoubtedly designed this
It should be possible to integrate the key information from kind of structure bearing in mind a computer-based information
other episodes into the topical episode. This is not denying the system, it is not surprising that the whole approach seems more
doctor, for example, access to the entire medical record for life; viable in computing terms than the kind of linking structure
it is promoting a simple way of accessing a large and complex suggested above. In particular, it is easier to see how to combine
information structure. parallel episodes through a common problem list, and how to
If in practice it were not possible to establish the salient manage summaries and the front end by problem lists.
information in such a way that it can be 'propagated' across Whatever view of an episode is presented to a user, the data-
parallel episodes then the database medical record structure base can only support a few options unless it becomes exces-
would revert to one of two degenerate forms: sively complex.
1. Information from a number of parallel episodes must be A reasonable compromise seems to be to limit the user's view
maintained as one whole, so that users concerned with of the database to segmentation with either simple statement
distinct episodes 'see' all of each other's information. linking or a problem oriented management structure super-
This makes it more difficult to create summaries and thence imposed. In either case, the exposition of structure at this level
the front end. becomes a matter for medical expertise.
2. Users concerned with distinct episodes 'see' only their own
information for their particular episodes. The system cannot Structure within the database
help them to be aware of other courses of investigation and Here we take the structure seen by the user and discuss it in
treatment, so users must themselves explore the possibility general terms, leaving a formal definition to the Appendix.
of relevant data outside their 'own' episodes. Of course, this The four structural concepts with which the user must become
latter situation is inevitable to some extent until most aspects familiar are:
of medical care are recorded in the database. 1. the episode and the episode group
2. segmentation of episodes
Discussion of episode structure 3. summaries
The record of a particular episode of medical care is made up of 4. the front end.
a number of what may be termed statements of various types.
In this context, virtually every type of entry in the patient We consider these in turn, and then discuss other aspects.
medical record is implied, including orders, reports, comment,
etc. The episode and episode group
There are broad categories of statements such as history, An episode may be defined in different ways. Possibly the most
physical examination, diagnoses, etc. but the distinctions tend natural definition is that which treats
to be arbitrary although probably not contentious. This view a stay in hospital
of the episode we call segmented and the individual categories a sequence of clinic visits
are referred to as segments. In practical terms, the history a sequence of outpatient visits
and physical examination segments represent the state of the and a set of health survey results
patient at the start of the episode. Subsequently the manage- as separate episodes.
Volume 17 Number 4 297
However, it seems reasonable to assume that the following solve it, the database summary mechanism must allow workers
two statements hold for any definition of an episode: in concurrent episodes within an episode group to 'pass around'
(a) an episode is an interaction between the patient and the significant findings, without lumbering each other with all
NHS, possibly for a considerable period and possibly for a statements from each episode. 'Starring' data in your episode
number of reasons, and with reference to the hospital summary category has the special
property that that data is starred relative not only to your
(Jb) there is a definite possibility that several episodes may be episode (in which it is generated) but also to all other current
concurrent. episodes—automatically.
Statement (b) leads us on to define an episode group (EG) as a Hence any user working within episode A and requesting all
set of episodes during the total time span of which there is significant data about the patient sees not only what has been
always at least one current (active) episode. (This definition is starred in that episode but also any other data in episodes B, C,
introduced because, in practical situations, it is desirable that etc. which has been starred there. The system must also provide
workers in different but possibly concurrent episodes have him with a mechanism for editing the starrings, i.e. for remov-
available vital data from each other's episode—this is pursued ing the star from any data he sees as starred (be it from his
later.) episode or another). This latter action will have no effect on the
Furthermore, as we are obliged by expense of on-line backing starrings seen from other episodes.
store to remove some of the database to less easily accessible Now these two summaries are standard, but we also want to
devices (tapes and so on), we introduce a split in the database offer users the ability to create such summary categories as
between current (on-line) data and archived (off-line) data. would be useful to them. These summaries would probably be

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


To make this a clean division we adopt the convention that confined to one episode within a group, and the categorisation
neither the front end nor any EG can straddle it, so that any of data involved would have no effect on other episodes.
EG is either wholly current—a CEG—or archived—an AEG. A doctor will want to be able to mark certain data as signi-
Clearly there can be at most one CEG but any number of ficant with respect to a given summary category—either at
AEG's, and since the latter are effectively read-only we can input or retrospectively. He may also wish to remove the
consider them to be eternally in our archives, even though we summary marks from data. This marking and unmarking of
may want to get a copy of one on-line for perusal. (Fig. 1) data must be recorded by the system and machinery is required
Each episode is further divided into segments. for abstractions and presentation of data in any summary
category; in fact, this may well be the most used form of
Segmentation of episodes interrogation.
The statements in a given episode may be divided into broad There are two important features then of summaries. Firstly
categories: the division of data in the patient record according to episode
History of origin or segment within episode must not hamper the
Physical examination important database function of summarisation—a process that
Diagnoses cuts across episodes and segments. Secondly, since summaries
Investigations are one of the main access routes to patient data the
Patient education, and so on. 'threading' of summarised data must not only be flexible
These categories are called segments, and the breakdown of an but also fast to access.
episode into the divisions suggested clearly says something
about the definition of 'episode' being used. The front end
The way that the system handles segments in the structure of This is a group of statements containing, inter alia,
the patient record must ensure that access is fast to all data (a) a highly condensed summary of past episode groups plus
within a segment (of a given episode) as soon as we have vital data from them about operations, confirmed diagnoses
'fixed' on that segment. and treatment (where this might affect later episodes).
Any further subclassification of data within a given segment
(b) personal medical data of general importance such as blood
is achieved by the use of simple data types—each data item has
such a data type. They would: group, tissue type, abnormal reactions to drugs, etc.
(c) ongoing therapy (where this may affect later episodes).
(a) be defined by applications software
It is maintained quite separately from all episodes and could
(b) be local to a given segment as regards meaning well consist of a simply structured collection of statements,
(c) be stored by database software with each data item grouped under such headings as were described earlier. The
(d) permit a finer selection mechanism for application software
when extracting data from records.
As an illustration, within the segment 'Investigations' we might
ARCHIVES CURRENT
define the data types 61, 62 and 63 for Chemical Pathology,
Haematology and Radiology results, respectively.
(a). In this case the pat ent has no interaction with the
hospital/Health Serv ce currently.
Summaries | AEG's 1 Front End
A summary within an episode is a selection of data statements.
For instance, the Hospital Summary aims to give a precis of a (b). Here the patient is j n contact with the hospital , and
patient's stay in hospital and contains only those facts which not for the first tilae»
are considered significant and useful should the patient re-
| AEG's CEG Front End
appear in the hospital. The GP Summary orientates the data
more towards what the patient's GP will want to know to help (c). This patient is in ec ntact with the hospital for the
him in the management of the patient after discharge from first time.
hospital.
| CEG | Front End
The hospital summary has an important place in the user's
view of the patient, giving a concise description of an episode
in terms of 'significant findings'. We clearly have a data dis-
semination problem when we have concurrent episodes. To Fig. 1 The different possible states of a patient record

298 The Computer Journal


split into the various data areas. (This shows only one problem
PATIENT
structure although there may be more than one.)
This introduces a division between two different types of
segment:
History
Physical Examination Data = Initial Data Baso * (a) Initial Patient Data (IPD) segments, consisting of history,
Admission Laboratory Results physical examination and such like,
(b) Management Data (MD) segments, consisting of diagnoses,
PROBLEM LIST treatment, investigations, etc.
Problem 1 (P ) • • Management Data for P^ Using this division we can envisage a structure where the
Problem 2 (Pg) • • Management Data for P2 problem list is a further level of division between episode and
segment, so that an episode looks like Fig. 3.
Such a system has the disadvantage that frequently certain
Problem n (P )
n
Management Data for P data in an MDS is relevant to more than one problem and we
would become involved in some sort of crosslinking or dupli-
Initial action plans cation of records to reflect the sharing of data between prob-
Progress notes lems.
Subjective Data The situation is made simpler if we refer to the structure of
Objective Data * Fig. 4 and regard a problem as a selection of data from the MD

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


Interpretation segments—in other words, as a special sort of summary of
Treatment and Therapy
Immediate plans the episode. Several points follow immediately:
etc. 1. The name of the summary corresponding to a problem is the
Note: terms marked with an asterisk are defined in Weed (1971).
statement of the problem itself,
Fig. 2 A representation of the problem-oriented record
2. Problems can now overlap quite happily,
3. Problems do not require special mechanisms within the
I SOD database to deal with them as they use the same machinery
I as other summaries. Creation, merging and suppression
IPD Segments Problem List come under functions that we have already decided to adopt,
4. The user is not obliged to make any use of problems,
I 1 I
Problem 1 Problem 2 Problem 3 .. 5. The concept of a problem, which may fall into disuse at any
time, is embedded purely in application software and its use
MDS for P, MDS for P, MDS for P, ..
or non-use in no way affects database structure.

H i II! Ill
Fig. 3 An unsatisfactory structure for an episode
Classes and groups
We now return to the subject of summaries and, in retrospect,
consider the two different types of summary which are sup-
ported by the database:
exact content and structure is very much a medical matter, (a) system-defined summaries, hereafter to be called 'classes'
but we can see certain data types being automatically copied, on (hospital summary, GP summary),
generation, from the originating episode into the front end—
admission and discharge data, operations performed, confirmed (b) user-defined summaries, hereafter to be called 'groups
diagnoses and so on. (private summary categories, problems, etc.).
We should also mention here two other appendages to the (It is perhaps worth noting in passing that there is a third type
patient record—patient details and patient diary—which have of summary which does not concern the database, namely that
a structural status similar to that of the front end. Patient which is formed from statements selected at the time of output
details consists of a collection of statements, possibly coded for of the summary and which does not involve any explicit
privacy in some way or separate from the main file and indices, marking, in the database, of the statements involved. An
giving the standard personal data for the patient. 'episode summary' consisting, say, of the current active problem
The patient diary is, generally speaking, a simple time- list plus related ongoing treatment and investigation plans
sequential collection of statements describing the time, date might be an example—the relevant data statements are selected
and location of the patient's appointments at clinics, say, or from the patient record not on the basis of any explicit episode
any other events known to be happening at a prescribed time. summary marker but because those statements are known to
It is relevant only to the CEG but cuts across all CEs and is represent ongoing material.)
used by them all. Generally speaking, every episode for every patient will have
all classes automatically set up, whereas groups are defined
locally within a given patient episode and vary from episode to
Handling the problem-oriented record (POR)
episode. In consequence, classes do not vary in number (except
We now consider how the concepts of the 'problem' and
'problem management' fit into our structure. We presuppose:
EPISODE
(a) that the individual problem is episode-specific,
(b) that a doctor may choose to use or not to use the problem-
IPD Segments MD Segments
oriented technique. So facilities must be available for the
doctor to work in 'problem-context',
(c) that the problem structure may be reviewed and re-
assembled repeatedly.
We might summarise Weed's description of the POR (Weed,
1971) with Fig. 2, in which, as an example, problem Pn has been Fig. 4 Weed-based segmentation of an episode

Volume 17 Number 4 299


perhaps once in several years) and every patient episode has 3. Long-term compatibility
that number of classes; on the other hand the number of Techniques and standards used will promote long term com-
groups denned within any given patient episode may vary from patibility with future information standards. As far as possible,
day to day. records should still be valuable after, say, 50 years.
At this point we face a logical conflict with the propagation
property of the hospital summary. If a data item, X say, is 4. Contemporaneity
starred in episode A, we are forced by technical considerations The database should be as up-to-date as possible. This implies
(and by the problem of complete reconstruction of a patient that any working subsets of the database have to be correlated
record on any given date) to copy Zinto all concurrent episodes, with the main database regularly so that all the data is self-
say B and C. But suppose episode B is being handled in a consistent at midnight every day, for example.
problem oriented way. Then if X is not 'put under' some prob-
lem within B, workers in that episode will not be aware of it 5. Manipulation
except by interrogating the entire record. Here the database A manipulative capability is required for re-organising the
must rely on applications software to provide a group status for database at various levels. It could happen, for instance, that
X perhaps by instruction from users. the same patient might have records stored by accident under
two different unit numbers; the database software will be able
Implementation to merge the two sets of records. It should also be possible with
suitable controls to correct errors in the individual statements
Introduction
of the database where normal access to the database cannot

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


Earlier sections describe the database largely in medical terms
achieve the desired result.
and the resulting structure is expressed formally in the Appen-
dix. Technical judgments are implicit in this design, being
mainly concerned with achieving the maximum capability for 6. Monitoring
an acceptable level of complexity. An over ambitious design The database is a general purpose structure rather than a
could become a Pandora's Box of technical trouble and so there mechanism for storing a predictable quantity of data in a
must be a comfortable credibility for what is proposed. particular fashion. It is therefore desirable to provide for some
analysis of how the database is used as a guide to enhancement
Having achieved a desired structure with the minimum of and also a cross check on other analyses of system usage. The
technical constraint, there are some further important technical main requirements are for a breakdown of the database into its
considerations to be fitted into the picture. constituent parts and for a study of how the database access
machinery is used by on-line users.
Technical objectives
One of the obvious benefits of database design, as opposed to Technical constraints
evolution of a series of adhoc file structures, is the opportunity In practice, we must plan for a real information system on a
to provide features which might be clumsy or even prohibitively simple configuration using current technology.
costly if developed for isolated applications.
1. Security 1. Archives
A general strategy must be adopted such that data does not It is not possible at present to hold all the data in on-line direct
get lost as a result of peripheral failure or breakdown of the accessfilesfor reasons of bulk, expense, etc. Although direct
processor for example. Recovery must be total and accurate access storage will undoubtedly become cheaper and more
within reasonable limits. We adopt the objective that data in compact in future it is almost certainly true that a considerable
the database should have at least two levels of backup. That is, price differential will still exist between such techniques and
if a version of the database fails at any stage in its daily usage mass serial storage such as magnetic tape. In view of this, the
then there should always be sufficient backup information to archival of low activity patient records is and will remain a
rebuild the database completely. If this backup information more economical approach to storage than attempting to hold
should itself fail at any stage of database recovery there should all data on-line.
be recourse to a second level of backup information. Only if The database software must therefore include facilities for
this second level fails should data loss be a possibility. moving patient medical records from the current database to
Since the database is common to all applications, errors in it the archive database and vice versa as required.
caused by technical shortcomings can have widespread effects
but the centralisation has the advantage that the database can 2. Accessibility
be validated as a whole and utilities can be justified for this Access to the database takes finite time and resources, and this
purpose. must be compatible with the needs of the applications. On-line
users of the database are the main problem since they compete
2. Transparency for the resources. A particular access to the database must take
The system will be able to store many distinct types of data place quickly enough to satisfy the on-line user who instigated
for many kinds of patient and this data may have its own it; it must also not tie up too much of the computer resources
inter-relationships and structure which the database will else it will interfere with other users of the computer. Of course,
maintain. However any structure imposed on data by the the more common the type of access the more critical each of
system—in pursuit of access speed, storage economy (in time these factors becomes.
and space) as well as the mechanisms of the basic structure— This will affect the structure of the on-line database. Common
must not be apparent to applications accessing the database. modes of on-line access to the database must operate efficiently
Equally the database structure must be independent of the and the structure can make all the difference. Only in an
content to the extent that the structure of the database can be idealised hardware situation is this just a mapping problem.
altered with minimum upheaval to the applications which
access it. 3. Recovery of the on-line database
In practical terms this means that applications should have Security of the archive database can be provided by grandfather-
easy access to the data content of statements within the rigid father-son techniques but the back-up strategy for the on-line
framework of segments, summaries, episodes and patients. database is inextricably related to other design considerations.

300 The Computer Journal


According to the requirements of security, at least two copies ever, what challenges a standard database management package
of each update to the database must be filed, in addition to the would face. The real-time element presents the usual problems
update itself. This provides the two levels of back-up but of recovery systems, access management and performance.
places an additional burden on the time and resources involved Possibly the 'audit trail' requirements of updating and the
in database access. This influences considerations of internal open size of the structures could present difficulties. The cre-
structure. ation and manipulation of problem structures in real time may
From another point of view, the speed of the recovery process be related to creating and manipulating files under a multi-
must be taken into account. Thus if the database is maintained access operating system.
entirely in duplicate, the operation of first level back-up is
trivial but the duplication itself may tie up excessive resources Acknowledgements
for normal access. We would like to thank Professor John Anderson, Department
of Medicine, King's College Hospital Medical School as our
mentor in the medical aspects of this paper and also our
4. Control of access to on-line data
colleagues at King's College Hospital Computer Centre who
A system of access management is required to avoid certain have contributed to the problems and solutions discussed.
types of overlapped access to the database by independent
users. The 'lockout' system adopted is also related to both Appendix—Formal structure
access and recovery considerations. Ideally the database The following is a formal definition of the structure of the
structure should avoid logical dependencies across the actual proposed database.
data statements, although this is quite impossible to achieve in Note that round brackets have their usual set-theoretic

Downloaded from https://academic.oup.com/comjnl/article/17/4/295/443627 by guest on 13 October 2020


practice. However, the less the logical relationships stretch meaning, and that square brackets are used to enclose 'optional'
across the database the better. In the worst extreme, the entire expressions.
database has to be 'locked' out from other users while each on-
line access is performed. This would presumably have a serious <database> = «patient-record»
effect on a highly conversational system. Lock-outs at patient (patient-record) = <patient-identity>[<front-end>[<pra>,
level might also be impracticable on such a system. Ideally, we <prc>]]
would like to work as low as segment level. <pra> = «archived-episode-group»
(prc) = <patient-diary>[<current-episode-group>]
(current-episode-group), (archived-episode-group) =
Software «episode»
The database discussed requires a great deal of software to <episode> = (initial-data-segments)(management-data-
support it in the following areas: segments)
(a) housekeeping (archive interfacing, file reorganisation, etc.) <initial-data-segments> = (segment, : / e Iid)
(Jb) archive maintenance (current database interfacing, <management-data-segments> = (segment;: ielmd)
statistical analysis) (segment-set) = (segment,-: iel)
(c) batch access (input and output) <patient-identity> = ((statement))
(d) on-line database recovery <patient-diary> = ((statement))
(e) data manipulation. (statement) = type, datatype,, origin date data . . .
Further discussion of these is outside the scope of this paper. . . . ([classification,.] : c) ([grouping^] : g)
where the statement is in segment; of episode^.
Conclusion <front-end> = ((statement)) possibly segmented,
Very little has been said about implementation. A phased class,. = ((statement: classification,., type; e statement) : i e /)
approach was essential because of the scale of the project. group ti , = ((statement : grouping^, type, e statement) : i e I)
The detailed design of the first phase, including bridging —a subset therefore of episode*,
existing systems, had been established before the medical segment^ = (statement : typet e statement)
recording project was closed down. Note that the definitions of 'group' and 'class' imply that they
There was no question of using manufacturers' software in are both segmented structures.
the circumstances of the project. It is interesting to note, how- Also, Iid and Imd are index sets, their union being the set /.

Reference
WEED, L. L. (1971). Medical records, Medical education, and patient care: the problem-oriented record as a basic tool, Cleveland, Ohio:
Case Western Reserve University Press.

Volume 17 Number 4 301

You might also like