| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Point/Counterpoint |
Department of Molecular Endocrinology, University College London Medical School, University College London, London W1 N 8AA, UK.
a Author for correspondence. Fax +44 171 580 2737; e-mail r.ekins{at}ucl.ac.uk
Abstract
The term "sensitivity" (as applied to an analytical method's performance) has again become a subject of controversy. Certain authorities (e.g., IUPAC) define a system's sensitivity as the response curve slope (or response/dose), others (e.g., IFCC) in terms of the detection limit. Many investigators have failed to perceive the contradiction between these concepts, wrongly assuming that maximizing "sensitivity" in the first sense maximizes it in the second (i.e., that they are inversely related). The existence of different meanings for this term (when used in the present context) is a source of confusion that has, among other things, led to erroneous ideas relating to immunoassay design. Such confusion should be terminated by adoption of one or the other of the definitions. However, the definitions are not of equal merit. We advance arguments against retention of the "slope" definition, which conflicts with the word's common meaning and is meaningless as an indicator of the performance of a measuring system.
Recent correspondence in Clinical Chemistry relating to the meaning of the term "sensitivity" has fortuitously coincided with an e-mail discussion on the same topic (ACB Newsgroup; acb-clin-chem-gen@mailbase.ac.uk), in the course of which—in contrast to comments by the Editor of this journal (1)—one of us (R.E.) has argued that the present IUPAC definition ("response curve slope") leads to obvious absurdities, renders the term meaningless as an indicator of assay or instrument performance, and should be abandoned. Moreover, past misunderstanding relating to the concept of "sensitivity"—a consequence of definitions of the kind promulgated by IUPAC—has, in our view, constituted a principal source of mythology in the immunoassay field and significantly retarded its development. But workers in other areas of science have also either disregarded or opposed the "slope" definition, on the grounds that it leads to confusion or is unworkable (e.g. (2)(3)). Indeed, reliance on the detection limit as the more logical indicator of a measuring system's "sensitivity" long predates the emergence of modern immunoassay and related ligand assay techniques in the late 1950s/early 1960s (see, e.g., (4)(4A). Thus the many immunoassayists who now adopt the detection limit definition (in accordance with the past recommendations of the IFCC (5)) are neither unique nor revolutionary in this respect."
The meaning of the term "sensitivity" (when used to describe the performance of an analytical instrument or system) is not a trivial semantic issue of interest only to theoreticians, but is fundamental to analytical science and of major practical importance. A clear understanding of, and agreement on, the concept it represents is crucial in assessing which of two or more instruments or assay kits is the most "sensitive," or the means by which a system's "sensitivity" can be maximized. Because this term is among a number for which formal definitions are currently being considered or reconsidered by NCCLS, this is a particularly opportune time to attempt to clarify its meaning and to resolve disagreements concerning its use that have persisted for 40 years or more. We therefore hope that a full debate on this topic will occur both among clinical chemists in this journal and elsewhere. As an initial contribution, we—though not endowed with the wisdom of Solomon [1]—feel it is appropriate to summarize some of the principal arguments against the slope definition, and to propose a general approach to the assessment of a measuring system's sensitivity.
Much semantic confusion in science stems from defining a term in a way that conflicts with its normal use. Such confusion is well illustrated by the formal identification of "accuracy" with "absence of bias" (5), in contrast with its common (and original) English meaning, "absence of error (as a result of care)"1 . This has led to the recommendation by some authors that the word should be discarded from the scientific lexicon (6)(7)(8). Thus the starting point of any consideration of a scientific definition must be the generally accepted use of the word in the native language, albeit the formal definition may need to be more restricted or rigorous.
As specifically applied to a scientific instrument or measurement, the adjective "sensitive" is defined by the Oxford English Dictionary (OED)2 as: "Indicating slight changes of condition, easily moved or affected by the external forces which it is constructed to detect or record." In a more general sense, Webster's Dictionary defines the word as: "easily hurt, ... easily offended, disturbed, shocked, irritated, etc., by the action of others, ... responding or feeling readily and acutely, ... [and] very keenly susceptible to stimuli." Clearly, in both cases the fundamental concept underlying the word's original and conventional meaning is the ability of an instrument or person to sense or perceive, and to respond to, a small stimulus (e.g., a remark seen as offensive) or a slight change in condition of an external force.
Sensitivity, in turn, is generally defined as the quality or degree of being sensitive. More specifically, the OED defines the term as: "the degree to which a device, test, or procedure responds to small amounts of, or slight changes in, that to which it is designed to respond." In the special case of a radio receiver, it is similarly defined as "the ability of a receiver or other part of a radio system to pick up or respond to weak radio signals." It follows that a more sensitive device, test, or procedure responds to smaller amounts or weaker signals. However, the OED also defines sensitivity as "the ratio of the response of a device to the stimulus." This ancillary definition, which is of recent origin, properly reflects the term's occasional use in the scientific literature. However, by implication, the latter definition is a corollary of (or consistent with) the word's basic meaning; i.e., an increase in the response/stimulus ratio (or response curve slope) is implicitly seen as permitting the device to "sense," or "pick up," smaller amounts. In short, it is evident that the perception by a person of a small stimulus, or in the case of a measuring instrument, the device's ability to detect a small amount or slight change in the measured quantity, is the central concept underlying this term's common use.
Countless examples exist in the scientific literature of its use in this conventional sense. Indeed, notwithstanding the fact that this journal has declared that it "reserves the term ... for the IUPAC definition" (1) (after which the word appears to have largely disappeared from the Journal's pages), a cursory glance through recent issues reveals examples of its continuing use in the original sense. For example, Yi Qian et al. (9), discussing the detection of PSA-ACT complex in prostatic fluid specimens, indicate that "using a highly sensitive ELISA technique, España and associates ... were able to detect PSA-ACT complex in the prostatic fluid and seminal plasma." This sentence can only be interpreted as meaning that the ELISA technique referred to was capable of detecting/measuring low PSA-ACT complex concentrations. Moreover, it is evident that an increase in a measuring system's "sensitivity" is generally regarded as advantageous, on the (implicit) grounds that smaller amounts of, or changes in, the measured quantity can be detected or determined. Immunoassay has, for example, frequently been extolled as an important analytical technique by virtue of its "exquisite sensitivity." This meaning also underlies related terms such as "ultrasensitive," widely accepted as indicative of a method's ability to determine even lesser amounts or concentrations than previously (or commonly) used techniques.
Nevertheless, as noted above, sensitivity has also more recently been used to denote the quotient "response/stimulus" (R/S)—or, in the case of an assay system, the slope (dR/dD) of the dose–response curve. It is, of course, arguable that this second usage is totally unconnected to the first, and that, when used in this second sense, the term represents no more than a "transduction" or conversion factor unrelated to the ability of an assay or instrument to respond to weak stimuli, or to any other (putatively advantageous) aspect of the system's performance (i.e., that the term is entirely neutral in its implications and carries no meritorious connotation). For example, an oscilloscope's "sensitivity" is generally expressed as the deflection of the electron beam caused by a unit electrical potential applied to the deflector electrodes. Though examples of such usage are not uncommon, we are specifically concerned in the present context with the word's use to describe the overall performance of an instrument or method for which the prime purpose is to measure a physical or chemical quantity. In these circumstances the second (slope) definition appears, as indicated above, to have been generally regarded as a logical consequence (or corollary) of the first. Underlying this apparent belief is the implicit but generally unstated assumption that the sole determinant of a system's ability to reveal small amounts of, or changes in, the measured quantity is the slope dR/dD (or the R/S ratio), and that other factors (such as errors in the response measurement) are either constant or irrelevant.
This assumption was explicitly enunciated by Yalow and Berson (10) in their section relating to immunoassay design. In the subsection Conditions for optimal sensitivity, they state, "There has been some controversy concerning the proper definition of sensitivity (Ekins and Newman [ref. 11 in the present paper]). Sensitivity can be defined either as the minimal detectable concentration or the slope of the dose–response standard curve. Let us suppose that the slope of the dose–response curve is a 10% change in response per picogram and that the statistical error in the determination is 10%. Then the minimal detectable quantity would be about 1 picogram. If the slope is 10-fold greater, i.e., 100% change in response per picogram, then with the same 10% error in measurement, the minimal detectable quantity would be 0.1 pg. Thus assuming the experimental error is unchanged, increasing the sharpness of the dose–response curve results in a reduction in minimal detectable quantity. Accordingly we have defined sensitivity in terms of the slope of the dose–response curve" (emphasis added).
It is instructive to analyze this superficially persuasive but, as advanced by Yalow and Berson in the course of a prolonged controversy relating to immunoassay theory (see, e.g., (10)(11)(12)(13)(14)(15)(16)(17)), totally erroneous argument. It rests on two premises: (1) that the statistical error in the determination of the response (at zero dose) is unaffected by changes in assay design, e.g., a change in antibody concentration, and (implicitly) (2) that the response curve is plotted in a coordinate frame such that a constant error in the response variable is represented by an error bar of constant length.
Premise 1 is clearly open to doubt, and is, in practice, generally
invalid. However, even if it can be shown to hold in a particular
system, premise 2 must also be fulfilled. For example, if the
measured response in a conventional RIA is the fraction of labeled
antigen bound (b) and b is plotted against the
analyte concentration ([H]) with linear coordinates (as
posited by Berson and Yalow in their analysis (10)), the
slope of the curve for b vs [H] (at zero dose)
will correlate (inversely) with the detection limit only if the
standard deviation (SD) of bo (i.e.,
b at zero dose) remains constant. If, on the other hand, the
coefficient of variation (CV) of bo remains
constant in the face of changes in assay design (as assumed by Berson
and Yalow), the response curve slope will correlate with the detection
limit only if the logarithm of b or
b/bo is plotted against [H] (see
Fig. 1
).
|
In other words, if the response curve slope is relied on to determine the conditions yielding maximal "sensitivity" (i.e., minimum detection limit), false conclusions will be drawn if the curve is drawn by using dose and response coordinates such that error bars at zero dose are not of constant length. For example, it is readily demonstrable that the maximal slope (of the log b vs [H] curve)—and hence minimization of the detection limit, assuming the CV of measurements of b remains constant—is achieved as the antibody concentration (and hence bo) approaches 0. In contrast, the maximal slope of the b vs [H] curve is achieved when the antibody concentration is 0.5/K, and bo = 0.33 (where K = the equilibrium constant). However, use of this antibody concentration will minimize the detection limit only if the SD of bo remains constant.
In short, the proposition (10) that increasing the slope of the b vs [H] curve implies reduction in the detection limit when the CV in bo is constant is specious. Nevertheless, despite this fundamental flaw in Berson and Yalow's arguments, the myth that an RIA is most sensitive (i.e., that the detection limit is least) when an antibody concentration is used that binds 33% of labeled antigen at zero dose remained a basic tenet of the RIA field for many years, and even now continues to be accepted by many practitioners.
This example highlights the underlying assumptions that have apparently led to the belief that the response curve slope (or the R/S ratio) is an indicator of a measuring system's ability to sense or determine "small amounts of, or slight changes in, that to which it is designed to respond," and is therefore a measure of the system's sensitivity (as originally defined). As indicated below, the assumption that the error in the response variable remains unaffected by changes in assay or instrument design is generally false. But the example also illustrates another important and related point, i.e., that the response curve slope depends on the particular response and dose variables selected (notwithstanding a reliance on the same experimental observations), and that, depending on this choice, it is possible to reach entirely contradictory conclusions regarding which of two or more systems yields the greater slope. Thus, if sensitivity is identified with response curve slope, conflicting conclusions are likely to be reached regarding the conditions yielding maximal assay sensitivity, depending on the particular coordinate frame in which the response curve is drawn.
Such absurdity is especially conspicuous in (albeit not confined to) the immunoassay field. As is well known, conventional RIAs rely on observation of the distribution of radiolabeled antigen between free and antibody-bound moieties, this distribution constituting the assay response. Its determination invariably requires the experimental measurement of two quantities (both of which are subject to random error): the free (F) and bound (B), the free and total (T), or the bound and total labeled antigen activities. (Note: It is generally current practice to introduce a nominally constant amount of labeled antigen into the assay system, and to measure the signal emitted only by the bound or free fraction; nevertheless, the measurement of the total labeled antigen—either by a pipette, or by counting—is subject to error, and must be considered as a component of the assay response.) Whichever of these fractions is experimentally determined, the labeled antigen distribution may be expressed in many different (albeit equally legitimate) ways: e.g., as F/B, B/F, F/T, B/T (i.e., b; see above), T/B, b/bo, or as the logs or logits of these ratios. The dose variable may likewise be variously expressed, e.g., as analyte concentration, or log analyte concentration.
Response curves may, in turn, be plotted in terms of a combination of
any of these dose and response variables. However, a change in assay
design will generally produce different—often opposing—effects on
curve shapes, positions, and slopes. For example, increasing the amount
of antibody used in the assay system causes the slope of the B/F
response curve to increase but that of the F/B curve to decrease (see
Fig. 2
). Meanwhile, the slope of the logit b/bo vs
log concentration curve (which generally approximates unity) is
essentially unaffected by changes in antibody amount. Clearly, it is
nonsensical to suggest that the assay is simultaneously rendered more
sensitive or less sensitive, or that its sensitivity is unchanged, as a
consequence of the change in antibody concentration, depending on the
way the response and dose variables are plotted. But absurdities of
this kind inevitably follow from a reliance on the response curve slope
per se as an indicator of assay sensitivity (or, indeed, of any other
aspect of assay performance). Such reliance nevertheless constitutes a
conceptual trap into which many immunoassay practitioners and
theoreticians have fallen, and underlies much of the mythology that
exists in this field.
|
Referring again to an example drawn from a recent issue of
Clinical Chemistry (18), construction of an
immunoassay is described essentially on the basis of the slope
definition of sensitivity. The authors claim to have selected
monoclonal antibodies for use in a competitive enzyme immunoassay on
the basis of three criteria: sensitivity, specificity, and correlation
studies. With regard to the first, the authors (relying on plots of
b/bo vs log analyte concentration) selected
antibodies yielding a "sensitivity" such that the value of
b/bo fell to 50% in response to an analyte
concentration of <10 µmol/L. (Note: The b/bo
vs log concentration curve is sigmoid and the slope is therefore
variable3
; hence the authors' definition of sensitivity does not in
fact conform to the IUPAC definition.) More importantly, it is well
known that the analyte concentration required to reduce
b/bo to 50% of the zero dose value varies with
antibody concentration (see Fig. 1
, right); thus the authors' method
of antibody selection does not reveal the physicochemical property
(i.e., the antibody affinity constant) of key importance in a
competitive immunoassay and is intrinsically unsound. They might, in
short, have reached quite different (and more valid) conclusions had
they based their choice of antibodies on plots of B/F vs bound analyte
concentration (i.e., Scatchard plots), whose slopes would have
indicated the relative affinities of the antibodies under examination.
Fortunately, however, in this paper (18) (as in many other similar articles) the authors' meaning is relatively clear, even though they use sensitivity in different senses, and their method of antibody selection is highly questionable (illustrating the dangers of designing assays or selecting assay reagents on the basis of response curve slopes). However, the term sensitivity appears in several articles in this and other recent issues of Clinical Chemistry, where its meaning is indeterminate.
Though immunoassay represents an analytical technique that (because of the variety of ways legitimately used to plot dose–response curves) exposes the contradictions stemming from the slope definition, many other analytical methods and instruments can be shown, on examination, to illustrate the same point. For example, the response of a simple balance can be portrayed either as the distance moved by the pointer tip, or the movement of the balance beam, or the angle through which the pointer rotates, when a weight is placed on the pan. Because, among other things, the length of the pointers may differ, the perception of which of two balances is the more sensitive or responsive (or whatever other term is used to represent the R/S ratio) will clearly depend on the choice of response on which the comparison is based.
This example illustrates another (albeit related) problem associated with the slope definition. An increase in the pointer length of a balance or galvanometer increases the response/dose ratio (assuming movement of the pointer tip across a scale is deemed to constitute the instrument's response); however, beyond a certain point, such a stratagem is unlikely to enable the instrument to respond to, and reveal, smaller changes in mass or electric current. Likewise, turning up the volume control on a radio receiver does not necessarily enhance the receiver's ability "to pick up or respond to weak radio signals." In other words, the basic and ancillary definitions of sensitivity given in the OED are often contradictory. For these reasons, Mettler, the well-known balance manufacturer, strongly criticized the slope definition adopted by the American Chemical Society (19) on the grounds of its unworkability and abandoned the use of the term sensitivity in its own descriptive literature. Likewise, Jones, writing (in 1959) in a radio engineering journal (3), recommended that, for much the same reasons, this term should be discarded from the scientific vocabulary.
A further criticism of the slope definition is that it precludes comparison of the sensitivities of two analytical methods or instruments (such as ultraviolet and immunoassay methods to determine steroid hormone concentrations) that rely on different response variables. Clearly, the physical dimensions of the response curve slopes differ in these circumstances, implying that a statement such as "the immunoassay method is more sensitive" would be inadmissible were the slope definition to be retained.
As a final illustration of this definition's impracticality, we note that many modern instruments (including analytical balances and immunoanalyzers, e.g.) provide a direct numerical read-out of the measured quantity. The user may be oblivious of the instrument's internal operation or of the "responding element" on which it depends and may not have the slightest interest in the exact manner in which results are calculated. But the degree to which the instrument responds to and reveals "small amounts of, or slight changes in, that to which it is designed to respond" is an aspect of the instrument's performance that is likely to be of great interest to the user and that can be readily determined by experiment. The slope definition precludes reference to this feature as sensitivity.
All such anomalies and contradictions are immediately dispelled once it is recognized that the statistical error incurred (or the uncertainty) in the determination of the response is as crucial as the response curve slope to an instrument's ability to reveal small amounts or slight changes in that to which it is designed to respond; in other words, this property is determined by the quotient [error in response]/[response curve slope], not by the slope alone. This quotient possesses the physical dimensions of the dose or stimulus, and represents the statistical error in the dose measurement, i.e., the latter's precision (or imprecision). This quantity is independent of the manner in which the response curve is plotted; thus the same conclusion will be reached in regard to which of two immunoassay methods yields the more precise measurement of a specified analyte concentration irrespective of the coordinate frame used to plot the dose–response curve.
Clearly, the precision of measurement of any defined baseline amount
(or concentration) determines the smallest difference (or change) in
amount to which the measurement system yields a perceptible (i.e.,
statistically distinguishable) response. Some authors (e.g., Reeves and
Calhoun (20)) have therefore implicitly identified
precision with sensitivity, suggesting that the latter term should be
defined so as to "include the ability of the measuring system to
distinguish between slightly different (analyte) concentrations over
its entire applicable range, i.e., its resolving power." Though this
suggestion has merit, it creates a number of problems. In particular,
the resolving power of many analytical systems (including RIAs), a
quantity implicitly determined by the standard deviation of the
selected baseline analyte concentration, varies with concentration (see
Fig. 3
). Hence it is impossible to represent the system's resolving
power (or sensitivity, as thus defined) by a single numerical index, or
to compare the sensitivities of two systems without specifying the
analyte concentration to which the comparison relates. Moreover, this
definition conflicts with the basic meaning of sensitivity (i.e., a
measuring device's ability to respond to a small amount or a weak
signal).
|
For this reason, we and our past coworkers have, in our theoretical
studies relating to ligand assay design and optimization (e.g.,
(15)(21)), restricted
use of the term sensitivity to what may be described as "the
resolving power at zero dose." In other words, we have invariably
based our analyses on the proposition that maximal sensitivity is
achieved when the imprecision of the zero dose measurement (i.e.,
[D]o) is least, on the grounds that this quantity
represents the most appropriate numerical indicator of a measuring
system's ability to respond to a small amount or weak signal. This
statistic essentially determines the system's detection limit,
although for the reasons given below, we believe it to be preferable to
differentiate between the two concepts.
The proposition that, in comparing two systems X and Y, the conclusion
is reached that system X is more sensitive than system Y because
([D]o)X is less than
([D]o)Y (or, implicitly, that system X's
detection limit is less than that of system Y) is logical, unambiguous,
scientifically sound, and fully accords with the concepts underlying
the common English (and American) usage of the term.
[D]o is, in short, a measure of a system's ability to
"sense" a small amount. In contrast, the proposition that a system
is more sensitive because the slope of the response curve is greater
leads not only to numerous anomalies and absurdities of the kind
discussed above, but also to the semantic contradiction that the less
sensitive of two instruments may be capable of sensing "smaller
amounts of, or smaller changes in, that to which the instruments are
designed to respond."
The lower limit of detection (LOD) of an assay system is frequently
defined as a multiple of [D]o (commonly 2, 2.5,
or 3x, although no agreement exists on this point, and practice
varies—itself a source of confusion). However, there seems little
advantage in relying on the LOD (as thus defined) as a formal indicator
of sensitivity, because the use of a multiplier contributes no further
information regarding the system's ability to respond to small amounts
of the measured quantity. In other words, knowledge of the magnitude of
[D]o per se is both necessary and sufficient to permit
evaluation of the relative "sensitivities" of different assay
systems or instruments. On the other hand, an analyst for whom it is
important to distinguish between samples that contain and those that do
not contain a specific analyte (e.g., hepatitis B surface antigen) is
likely to wish to assess the actual detection limit of the system under
the actual conditions in which it is used (as calculated, for example,
by Rodbard (22)). This depends on the number of replicates
used, the confidence demanded, etc. In other words, the detection limit
as thus determined depends both on the performance characteristics of
the measuring system itself (i.e., on the value of
[D]o), and on the particular way in which the analyst
chooses to use it and to interpret the results it yields.
Because the detection limit of an analytical system of practical
importance to the analyst depends on such "local" or subjective
factors, it seems inappropriate to rely on the LOD (as defined above)
as an indicator of sensitivity. In short, the possibility of confusion
in this area would probably be reduced if the numerical index of the
sensitivity of any measuring system were defined as "the imprecision
of measurement of a zero amount of that to which the system is designed
to respond" (i.e., as [D]o), and if the term
detection limit, or LOD, were restricted to the quantity calculated in
the manner described by Rodbard (22).
The objection to the inclusion of local factors or analytical
requirements in formal definitions of terms describing assay or
instrument performance likewise applies to the in-vogue term
"functional sensitivity" introduced by Spencer (23)
and recently criticized by Sadler et al. (24). In reality,
this quantity simply constitutes what Spencer evidently (though
doubtless with good reason) regards as the effective lower limit of the
working range of thyrotropin assays used in clinical practice—the
working range being defined as the range of concentrations or amounts
over which measurements are of a precision acceptable to the analyst
(see (8)(17); Fig. 3
). Spencer and her
colleagues are, of course, free to decide when thyrotropin measurements
are of acceptable precision in the particular context in which they use
such assays; however, the needs of other investigators may be more or
less stringent than hers, and their definitions of the working range of
particular thyrotropin assay systems will vary accordingly. Moreover,
an analyst concerned with, e.g., the measurement in transfusion blood
of hepatitis B surface antigen is likely to be governed by different
considerations and criteria in deciding an assay's working range. In
short, the working range of an assay is defined by numerical limits
that are a matter of local choice; these limits should not, therefore,
be incorporated into a formal definition.
A second objection to the use of the term functional sensitivity is
that it introduces yet further semantic confusion into an area that is
already widely misunderstood and controversial. As correctly emphasized
by Sadler et al. (24), the so-called functional
sensitivity of an assay is essentially the analyte concentration given
by 5x [D]o (although [D]o refers in
this context to the interassay variance in measurements of samples
containing zero analyte). It is therefore not surprising that this
quantity is greater than that obtained by multiplying the intraassay
value of [D]o by 2 or 3.
Our own view is that the performance characteristics of an immunoassay
kit or method can be best represented by its precision profile
(8)(25), an approach being adopted by many
immunodiagnostics companies and investigators (e.g.,
(26)). (Note: Assay bias—in so far
as this is a valid concept—is potentially the subject of another
debate and is not addressed here.) Fig. 3
portrays the basic concept as
applied to intraassay measurements, but the concept can clearly be
extended to embrace (e.g.) interassay precision, intersample precision,
and interlaboratory precision (6). It is likewise
applicable in principle to all other analytical techniques. Knowledge
of the precision profiles yielded by two such techniques immediately
reveals which yields the lower value of [D]o and hence
which is the more "sensitive"—even if one may be a bioassay and
another an immunoassay, each yielding response curves with slopes that
cannot be compared.
Though certain secondary assay parameters (e.g., the working range, lower limit of detection) can be derived from the basic performance characteristics represented by the assay precision profile, these involve consideration of local requirements and practice, and it would therefore be inappropriate to define such parameters in numerical terms. Whether or not it is useful to devise additional nomenclature that identifies, e.g., the upper or lower limits (or both) of the working range, as Spencer et al. have implicitly done, depends in part on any additional confusion this might cause. Clearly, such terminology must be carefully chosen to avoid conceptual ambiguity.
In summary, we have here attempted to demonstrate that the response curve slope definition of sensitivity is untenable, that it leads to many absurdities and confusion, and that its retention would effectively exclude the term from the scientific lexicon, leading (as has been claimed to be a consequence of the formal identification of accuracy with absence of bias (6)) to a depletion of linguistic resources. In other words, no English term would be available to describe an instrument or method capable of measuring small quantities of that which it is designed to measure, nor would any adjective exist describing a method capable of determining smaller amounts than another. It is therefore incumbent on the protagonists of the slope definition to justify its utility and, most particularly, to indicate if they regard an improvement in sensitivity (as thus defined) as advantageous—and, if so, on what grounds.
Footnotes
1 1 Accurate: "Of things and persons: exact, precise, correct, as the result of care" (Oxford English Dictionary). 
2 Nonstandard abbreviations: OED, Oxford English Dictionary; B, bound labeled analyte; F, free labeled analyte; T, total labeled analyte; b, fraction of labeled analyte bound; b0, fraction bound at zero dose; [H], analyte concentration; [D]o, standard deviation of zero dose estimate; b0, standard deviation of response at zero dose; LOD, lower limit of detection; R/S, response/stimulus.
3 3 We are indebted to a reviewer of this article for reminding us that immunoassays' response curves as generally portrayed are nonlinear, and that sensitivity (slope definition) therefore varies as a function of analyte concentration.
References
1-antichymotrypsin complex in serum but not in prostatic fluid. Clin Chem 1997;43:352-359.
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  K. Skogstrand, P. Thorsen, B. Norgaard-Pedersen, D. E. Schendel, L. C. Sorensen, and D. M. Hougaard Simultaneous Measurement of 25 Inflammatory Markers and Neurotrophins in Neonatal Dried Blood Spots by Immunoassay with xMAP Technology Clin. Chem., October 1, 2005; 51(10): 1854 - 1866. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. Kokko, N. Johansson, T. Lovgren, and T. Soukka Enzyme Inhibitor Screening Using a Homogeneous Proximity-Based Immunoassay for Estradiol J Biomol Screen, June 1, 2005; 10(4): 348 - 354. [Abstract] [PDF]
|
|
|
|
 |
|
 A. Halperin, A. Buhot, and E.B. Zhulina Sensitivity, Specificity, and the Hybridization Isotherms of DNA Chips Biophys. J., February 1, 2004; 86(2): 718 - 730. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Ekins Ligand assays: from electrophoresis to miniaturized microarrays Clin. Chem., September 1, 1998; 44(9): 2015 - 2030. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. L. Pardue Dr. Pardue responds: Clin. Chem., August 1, 1998; 44(8): 1776 - 1778. [Full Text]
|
|
|
|
|
|
R. Ekins and P. Edwards On the Meaning of ""Sensitivity"": A Rejoinder Clin. Chem., August 1, 1998; 44(8): 1773 - 1776. [Full Text]
|
|
|
|
|
|
H. L. Pardue Counterpoint The inseparable triad: analytical sensitivity, measurement uncertainty, and quantitative resolution Clin. Chem., October 1, 1997; 43(10): 1831 - 1837. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
