J Cbron Dis Vol. 40, No. 6, pp. 635440, 1987 0021-9681/87 $3.00 +O.

OO
Printed in Great Britain. All rights reserved Copyright 0 1987 Pergamon Journals Ltd

CLINIMETRIC PERSPECTIVES

ALVAN R. FEINSTEIN

Yale University School of Medicine, 333 Cedar Street, P.O. Box 3333, New Haven,
CT 06510-8025, U.S.A.

Clinimetrics Rating scales Clinical indexes

ALTHOUGH the word clinimetrics needs called ailment -oriented phenomena. These are
definition, we need not worry about any major the things that doctors observe as direct clinical
disagreements, because I made up the word. events in the practice of medicine and in the
Like Humpty Dumpty, I can presumably have evaluation of therapy. The events include exis-
it mean whatever I would like. tence and relative magnitude of symptoms,
I use the term clinimetrics in reference to severity of illness, diagnostic existence of dis-
arbitrary ratings, scales, indexes, instruments or ease, and other entities usually associated with
other expressions that have been created as a particular disease or diseases. Several years
“measurements” for clinical phenomena that ago, when my colleagues and I began reviewing
cannot be measured in the customary dimen- the medical literature to see what clinimetric
sions of laboratory data [l]. People who work indexes we could find, we were surprised by the
with laboratory measurements have a regular great many that were out there. For example,
set of methods and instructions that give precise about 220 different indexes have been pro-
directions for how to do each measurement. For posed solely for assessing the function of joints
clinimetric rating scales, however, there are no and patients’ responses in arthritis or other
analogous methods and standards. locomotor diseases.
A different set of indexes can be called gen-
TYPES OF CLINIMETRIC INDEXES
eral, rather than ailment-oriented. These general
indexes refer to disabilities and dysfunctions
An example of a clinimetric rating scale is the that are not specific for a particular ailment or
Apgar Score [2] for the condition of a newborn disease. For example, the New York Heart
baby, which Virginia Apgar created simply and Association Functional classification [lo], al-
effectively, perhaps because she had no consul- though intended for patients with heart disease,
tants to help her. Other indexes are the Killip has a structure and a set of categories that are
Class [3] for myocardial derangement in patients applicable to many other ailments as well. The
with acute myocardial infarction, the Visick Katz Index [l l] and various other functional
index of gastro-intestinal function [4], the TNM indexes for activities of daily living [ 121, are
stages for cancer [5], the Jones diagnostic crite- general indexes. And, of course, general indexes
ria for rheumatic fever [6], the World Health are illustrated by the quality of life indexes
Organization diagnostic criteria for myocardial discussed at the Portugal Conference.
infarction [7], diagnostic criteria that have Regardless of any disagreements on meth-
been prepared for diverse other diseases, the odological issues, I think most scientists would
Lansbury index of activity [8] in rheumatoid immediately agree that clinimetric indexes have
arthritis, and the Glascow Coma Scale used a very important humanistic value. The im-
in head injury and various other activities in provements shown with clinimetric indexes are
neurology [9]. what most people seek in clinical care. Patients
All of those indexes have been directly con- want to know how to reduce their pain or
structed by clinicians to describe what might be distress; how to improve their function; how to
635
636 ALVAN R. FEINSTEIN

prevent something bad from happening. Hu- settings, with different therapeutic activities, it is
manistically, the clinimetric information is what not surprising that many different kinds of
old-fashioned doctors used to think about be- indexes might be needed for adequate descrip-
cause there were no laboratory data to distract tions.
their thoughts. As for therapeutic interventions, at least five
Clinimetric indexes, however, also have im- different kinds of interventions can take place.
portant roles in science as well as humanism. One of them is the classical clinical activity: a
The clinim;;tric details will serve to identify the remedial intervention intended to cure some-
“material” that is under study. When we evalu- thing or at least produce a major improvement.
ate therapeutic interventions, we need to know This is what we do when we give an analgesic
those details in discriminating important dis- for the treatment of headache, or antibiotics for
tinctions [ 131. For example, as Hans Troidl said the treatment of bacterially manifested infec-
at the conference (in effect), “Suppose two tion. We can directly see what we have accom-
operations have identical rates of mortality plished by seeing how well the target (headache
and morbidity. To decide which operation is or bacteria) improves. In rehabilitation therapy,
superior, you will compare various kinds of we may not expect to achieve a cure, but we
clinical distinctions.” want improvement in some kind of dysfunction.
In the rest of this discussion, I do not plan to This is what happens with occupational ther-
solve any of the problems that were mentioned apy, physiotherapy, or such work as teaching
at the Portugal Conference. Instead, I want to people to use a prosthesis. With supportive
develop a catalog that shows why clinimetric therapy, we are not trying to cure, achieve
issues are often difficult and occasionally massive improvement, or rehabilitation. We are
confusing. simply trying to keep someone comfortable in
a terminal course.
With prophylactic therapy, we are trying to
SOURCES OF COMPLEXITY
prevent or retard an adverse outcome. There are
We are discussing an interplay of three two different kinds of prophylaxis. In primary
different entities: groups, interventions, and prevention, we vaccinate against polio, or use
data. In our general intellectual models, a per- special dietary tactics to try to prevent coronary
son described with baseline data receives an heart disease. In secondary prevention, which is
intervention; and the results are identified with what clinicians commonly practice without call-
outcome data. The complexity arises because ing it “prevention”, we can do GI surgery to
the groups under study contain different people, prevent GI bleeding, cardiac surgery to prevent
in different settings, receiving interventions death from coronary heart disease, or use med-
given for different purposes. ication for various reductions in blood pressure,
The type of setting might be natural, for blood sugar, or blood lipids to try to prevent
someone at home; clinical, in a doctor’s office or the vascular complications of hypertension,
hospital; or institutional, in a nursing home or diabetes mellitus, or hyperlipidemia. In fact,
hospice. The type of practice that is going on whenever the results of any clinical treatment
might be ordinary clinical care, or a specific trial are expressed as a survival rate, we are talking
with a specific protocol in a single location, or about secondary prophylactic therapy.
a multi-center clinical trial in many locations. Finally, certain treatments might be called
All of these activities and locations may require salutary. The starting state is not a disease and
very different kinds of indexes for different types the concluding state is not a disease. Instead the
of patients. The patient whom we observe in a goal is to maintain or improve health. This is
particular study might be healthy or have an what happens when people jog, not necessarily
acute illness, a chronic illness, or a chronic trying to prevent heart disease, but simply be-
illness with acute exacerbation. In other circum- cause they feel good after they’ve done it, and
stances, we may be concerned not about a believe that in some way it improves their
specific clinical illness, but about a general state health.
of incapacitation. In yet other circumstances, With so diverse an array of therapeutic
where the patient is moribund, our main goal interventions, you would readily expect that
may be to provide terminal care that makes the different kinds of indexes and rating scales
last part of life as comfortable as possible. would be needed to assess what is being accom-
For these different people, in different clinical plished.
 Clinimetric Perspectives 637

TYPES OF DATA baseline state rating, such as a TNM stage in

cancer, is used to prognosticate an outcome.
Of the different types of information collected For example, we might say that in patients who
in the care of patients, one kind of data can be are in TNM stage 1 for cancer of the omphalos,
called demographic : age, gender, religion, socio- the expected 5year survival rate is 80%. The
economic status, etc. This is the type of informa- predictions that we make in medicine today are
tion that can be acquired by someone without often imprecise because they depend mainly on
any particular medical training. A second kind paraclinical data alone, ignoring the kind of
of information, which might be called para- clincial information that would tell us about
clinical, is the data that have become so se- severity, comorbidity, and auxometry. The
ductive and revered today in clinical care. This prognostic impact of these clinical factors has
is the “hard” information that comes from been well demonstrated in patients with cancer,
substances derived from patients and examined but the distinctions hold true for just about any
away from the bedside by methods other than other major ailment you look at.
that of conventional clinical examination. Para- If you want to think about a biologic vector
clinical information is what we get from radi- that describes the progress of a disease in a
ography and modern imaging, from laboratory particular patient, we often describe that vector
tests, from the microscopic examination of tis- mainly according to the morphology of the
sues and cells, from electrography of heart, disease or its laboratory aspects. In our quest
brain, and muscles, and from the entire phantas- for “hard” data, we ignore crucial aspects of the
magoria of technological methods now being biologic vector that are denoted by the patient’s
used in modern medicine. clinical illness, by the auxometry of its pro-
The availability of this kind of “hard infor- gress, and by the comorbidity of the associated
mation” has helped make doctors generally diseases [14].
delinquent in paying attention to the “soft”
clinical data of symptoms, physical signs, and
FORMATION OF SCALES
clinical judgments about the severity of illness,
the comorbidity of associated diseases beyond As for the indexes themselves, there was a lot
the index disease, and the rate of progress of the of discussion at the Portugal Conference with-
index disease. The combination of paraclinical out strong agreement about language and ideas.
and clinical information is contained in ailment- In certain indexes, which might be called “un-
oriented expressions, but can also appear in calibrated” or “global,” the scale is expressed in
general indexes concerned with physical func- categories that are not accompanied by stan-
tion, emotional function, and social function. dard instructions, agreements, or operational
criteria for how to use the categories. In con-
USES AND TIMING OF INDEXES
trast, a calibrated or demarcated scale has such
instructions. The instructions would say, for
Beyond those distinctions in the types of example, that severe is used for pain that keeps
available data, other distinctions arise from the you awake, but that does not make you move
uses and timing of the indexes. Certain indexes about restlessly; moderate is used for pain that
deal with a single state, describing a baseline does not keep you awake, but that does keep
condition before treatment or an outcome con- you from thinking or eating; etc.
dition afterwards. Other indexes reflect a change In many circumstances, we create composites
in state by denoting the difference between of uncalibrated scales to form what looks like a
outcome and baseline, or by expressing the calibrated scale. For example, people might give
change directly in a transition index that does global (uncalibrated) ratings from 1-5 for a
not contain specific data for two single states. large scale series of items. When the results are
The transition index is obtained in response to added to produce a numerical rating, such as 72,
questions such as, “Are you feeling better?“, the result appears to be a calibrated scale.
“What improvement has there been?“, or In forming scales for complex phenomena, we
“What change have you noticed?” The response may use a simple, direct approach such as the
does not involve two separate ratings for two classical visual-analog global “uniscale” to ask
separate single states and is directed instead at a patient to express quality of life by marking a
an assessment of the transition itself. line between 0 as terrible and 10 as terrific. In
An index can be used for predictions when a a tandem profile, we assemble a series of sub-
638 ALVAN R. FEINSTEIN

scales and put them together in tandem without construct validity. The complexity of the ideas
fusing them into a single rating. A classical and calculations are discussed elsewhere [l].
example of a tandem profile is a TNM index In sensibility, however, the clinician wants to
rating such as T2NOM 1, where the 2 is a rating know things that are not always assessed quan-
for Tumor, the 0 is a rating for Nodes, and the titatively. Is the inquiry aimed at the right
1 is a rating for metastases. The TNM rating target, i.e. does it have face validity? This type
system has a separate set of indexes in which the of validity, as well as content validity, must
T, N and M ratings are aggregated into groups usually be assessed by some form of qualitative
that are called stages. The stages are graded as appraisal because they cannot be expressed
I, II, or III, so that people can better commu- quantitatively. Another important consider-
nicate and analyze their data. ation in clinical sensibility is the ease of usage
In the aggregates of other indexes, the sub- for an index. For example, after looking at a
scales are united either in a summation or in profile index that is magnificently validated and
some form of hierarchial structure. An example reliable, but that has 138 items, clinicians may
of a straight summation is the Apgar score, conclude that the index is too difficult to use.
which contains five different ratings for respira- Many clinicians might therefore prefer that
tory rate, heart rate, color, reflexes, and tone. quality of life be assessed with an index that is
Each of the ratings can range from 0 to 2, and elegantly simple, like the Spitzer Scale [15]. It
the five ratings are added together to form the has five components, each of which is graded as
Apgar Score, which ranges from 0 to 10. The 0, 1, and 2, and which gives a result somewhat
TNM staging system is an example of a hier- analogous to the Apgar Score.
archial structure. In the ascending hierarchy, When sensibility and standardization come
someone is placed in stage III if they have into conflict, clinicians may create an index
evidence of metastasis, no matter what else is that is sensible and then not bother to standard-
going on. Stage II is usually for someone who ize it. Psychosocial investigators, on the other
lacks distant metastasis, but who has in- hand, may create something that is marvelously
volvement of regional nodes. Someone who standardized but that clinicians say is too
lacks evidence for Stage III or Stage II is placed difficult to use, aimed at the wrong targets, or
in Stage I. The Katz scale for activities of daily ineffectual in describing the right targets.
living also contains arrangements for hierarchial These conflicts are not unique in clinimetric
structuring of the rated categories. indexes. An analogous set of major problems
occurs in randomized trials today, because of
the inevitable conflict between two sets of poli-
CONFLICTS IN STANDARDIZATION cies for the design and analysis of the trials [16].
AND SENSIBILITY The inevitable conflicts between the fastidious
and the pragmatic policies often lead to major
Having discussed the formations and struc- controversies in the results. Consequently, ran-
tures that can be used in clinimetric indexes, I domized trials will be magnificant for what they
shall now turn to a particularly interesting can do, but they will often be impossible or
problem: the conflicts that can arise between unsuccessful because of ethics, logistics, costs,
two reasonable sets of strategies and goals. The or controversies. At another fundamental level,
conflicts arise between the scientific goal of conflicts arise about whether randomized trials
standardization and the clinical goal of what are done to answer the question of “Does the
might be called sensibility. For standardization, agent work?, i.e. Does it have therapeutic
we want an index to have reliability and valid- efficacy?“, or the question of “What kind of
ity. Reliability often has two parts: an “exter- patient is helped by this agent, and what is the
nal” part, which refers to observer variability in effect in that patient?” A trial that may answer
using the index; and an “internal” part; which the first question may not always answer the
refers to homogeneity of the different variables second.
(or “items”) contained in the index. The internal
homogeneity of the interrelationship of the vari-
SOURCES OF DISAGREEMENT IN
ables is often measured with a statistical calcu- CLINIMETRIC INDEXES
lation called Cronbach’s Alpha. Validity is often
determined statistically with calculated values I want to comment on another reason why we
for ideas that are called criterion validity and did not reach unanimity at the Portugal Confer-
 Clinimetric Perspectives 639

ence. The different observers looked at different task more slowly and with much less effort. By
parts of the same elephant. We had four neglecting the role of effort and support when
different focal points: ailment-oriented data, we note changes in the magnitude of a per-
physical-function data, social-function data, formed task, we will continue to mismeasure the
and emotional-function data. The diverse tasks.
unions and intersections of all four types of data Another problem is the frequent failure to
can be used to describe the total quality of life. integrate physical, psychic, and social ratings
For many psychosocial scientists, however, for their mutual impact on one another. For
quality of life is described by the specific inter- example, a profile index might have a rating for
section of social, emotional, and physical func- psychic depression and another rating that indi-
tion data. For many clinicians, the clinical qual- cates the patient is not working, without speci-
ity of life is described by an intersection of the fying why the patient is not working. The failure
ailment-oriented and physical function data. to work might be due to a physical impairment,
The two sets of observers may not recognize the psychic depression, the absence of an avail-
that they think different thoughts and look at able job, the person’s incompetence at the job,
different data when they describe the different or the fear that a pension may be lost if the
parts of the same elephant. patient returns to work. It is valuable to have
I believe that one of our main problems at the multi-variable indexes that give separate ratings
Portugal Conference was the general lack of for work and for psyche as separate items, but
agreement on the basic construct of what is we need better ways to show how the psyche
meant by quality of life. We did not achieve affects work or vice versa.
agreement on that construct in this meeting; and Yet another problem is that statistical assess-
I doubt that we could have done so without ments of reliability and validity may be revered
sacrificing all the rest of the agenda in favor of without suitable recognition that the mea-
what might have been a non-productive, incon- surements may be highly restricted to the partic-
clusive debate. Quality of life is something like ular setting and group of people that produced
intelligence. Everyone knows it exists and thinks the statistical data. The results may not be
they can identify it in various ways, but we may “exportable” beyond the particular locale of the
not be able to evoke universal agreement on field trial. At the other end of the intellectual
what it is. We are probably better off letting spectrum, clinicians may revere the sensibility of
people propose indexes, which we can then use certain indexes without recognizing that the
or not use, rather than try to get a multi- indexes are unreliable and that different clin-
individual consensus on what ought to be there. icians may arrive at different judgments for the
The situation is much easier when we try to same things. This last problem occurs, un-
create ailment-oriented indexes for clinical fortunately, for hard data as well as for soft
work. Because clinicians have good general data. The radiologists and pathologists who
agreement on the construct called “congestive provide the hard data are often like policemen
heart failure,” we might fight about how good who do not have a civilian review board. When
a particular index is, but we don’t have to fight checked for observer variability, the radiologists
about the construct itself. Because we do not and pathologists may be much more erratic than
have unanimous agreement about the construct the clinicians.
for quality of life, however, the idea has become On several occasions during the Portugal
a kind of umbrella under which are placed many Conference, the preferences of patients were
different indexes dealing with whatever the user mentioned. Why are clinicians (and psycho-
wants to focus on. social scientists) so refractory about asking the
patient a simple, direct question, such as “What
OTHER SOURCES OF PROBLEMS do you want?” The question is particularly
pertinent when a single disability among several
A separate problem in clinimetric indexes is is to be chosen as the focus of rehabilitation. We
that many indexes of physical function denote can ask, “Do you want us to work most on your
the magnitude of a performed task, without hands or on your feet?” But investigators are
suitable attention to the patient’s effort or social somehow reluctant to ask patients to state their
support in doing the task. It so happens that I own preference about what is important. The
can “cure” angina pectoris, when it occurs on choice may not require any complex strategies
exertion, by telling the patient to do the same for utilities, probabilities, or mathematical mod-

c D.40,&M
640 ALVAN R FEINSTEIN

els; and a simple question such as, “What would two disciplines. To improve the situation, statis-
you like us to emphasize?“, may be highly tical and psychosocial consultants can make
effective and valuable. greater efforts to understand the clinical prob-
lems before applying methodologic solutions,
POLICY ISSUES IN RESEARCH but the main challenge today is for the clinical
investigators themselves. They must recognize
Finally, I should like to point out that some
that their craft offers basic intellectual chal-
of the difficulties I have cited arise from policies
lenges at fundamental levels of science, scholar-
regarding medical science and medical research.
ship, and creativity. Although collaborative
For more than a century, medical research has
consultants can offer invaluable aid, the basic
followed a paradigmatic policy that basic clin-
science that clinicians most need to learn and
ical science consists of explaning what nature
apply must be developed by themselves.
has done, but not intervening to change or
prevent nature’s actions. Since intervention is REFERENCES
the main activity in clinicial practice, clinicians
1. Feinstein AR: Clhdmetrics. New Haven, CT: Yale
are left with nothing basic to think about in University Press, (In Press, 1987)
their own daily work. 2. Apgar V: A proposal for a new method of evaluation
Consequently, clinical medicine has become of the newborn infant. Anesth Analg 32: 266267, 1953
3. Killip T, Kimball JT: Treatment of myocardial in-
regarded as an application of two different sets farction in a coronary care unit: A two year experience
of basic sciences [ 131. The clinician is supposed with 250 patients. Am J Cardiol 20: 457464, 1967
to apply molecular biology, physiology, path- 4. Visick AH: A study of the failures after gastrectomy.
Ann R Coil Surg Engl 3: 266284, 1948
ology, and other “basic” sciences for explaining 5. American Joint Committee for Cancer: Staging and
clinical phenomena; and is then supposed to End Results Reporting: Manual for Staging of Cancer.
apply statistics and other kinds of epidemiologic Chicago, IL: American Cancer Society, 1977
6. American Heart Association Committee on Rheu-
or social-science concepts for assessing inter- matic Fever and Bacterial Endocarditis: Jones criteria
ventions. With clinical medicine achieved by (revised) for guidance in the diagnosis of rheumatic
applying two sets of non-clinical basic sciences, fever. Circulation 69: 203A-208A, 1984
7. Joint International Society and Federation of
clinicians are not challenged to think at a funda- Cardiology/World Health Organization Task Force on
mental intellectual level. In the absence of basic Standardization of Clinical Nomenclature: Nomen-
challenges, it is not surprising that clinicians clature and criteria for diagnosis of ischemic heart
disease. Circulation 59: 607-609, 1979
seem to produce very little profound thinking; 8. Lansbury J: Quantitation of the activity of rheumotoid
they are supposed to get all their profound ideas arthritis: A method for summation of the systemic
indices of rheumotoid activity. Am J Med Sci 232:
from other places.
30&310, 1956
Suppose the makers of scientific policy for 9. Teasdale G, Jennett B: Assessment of coma and
medical research were to recieve the proposal impaired consciousness: A practical scale. Lancet 2:
81-84, 1974
that clinicians themselves have basic ideas to
IO. Criteria Committee of the New York Heart Associ-
contribute, and that an additional but different ation, Inc: Diseases of the Heart and Blood Vessels:
form of basic science must be developed for Nomenclature and Criteria for Diagnosis, 6th edn.
Boston, Mass.: Little, Brown, 1964
clinical practice. The policy makers would
11. Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffee
probably think that anyone who offered this MW: Studies of illness in the aged: The Index of ADL:
wild idea was either ignorant about basic science A standardized measure of biological and psychosocial
function. JAMA 185: 914-919, 1963
or bereft of basic sensibility. The policy makers
12. Feinstein R, Josephy BR, Wells CK: Scientific and
therefore do not give suitable support for clinical problems <n -indexes of functional disability.
research or education that emphasizes basic AM Intern Med 105: 413420. 1986
13. Feinstein AR: An additional basic science for clinical
clinical scholarship.
medicine: I. The constraining fundamental paradigms;
In the absence of such support, investigators II. The limitations of randomized trials; III. The
from the psychologic and social sciences should challenges of comparison and measurement; IV. The
development of clinimetrics. Ann Intern Med 99:
not be surprised to find that clinicians are often
393-397, 544-550, 705-712, 843-848, 1983
excellent at solving problems, but poor at artic- 14. Feinstein AR: On classifying cancers while treating
ulating either the problems or the solutions. patients. Arch Intern Med 145: 1789-1791, 1985
When those clinical problem-solvers, who are 15. Spitzer WO, Dodson AJ, Hall J, Chesterman E, Levi
J, Shepherd R, Battista RN, Catchlove BR: Measuring
not good at specifying methods, collaborate the quality of life of cancer patients: A concise QL-
with psycho-social consultants who can articu- index for use by physicians. J Chron Dis 34: 585-597,
1981
late methods excellently, but who may not
16. Feinstein AR: Clinical Epidemiology: The Architecture
understand the problems, the result may be a of Clinical Research. Philadelphia, PA: WB Saunders,
cross-sterilization rather than fertilization of the I985