Communicating The Uncertainty of Harms and Benefits of Medical Interventions
Communicating The Uncertainty of Harms and Benefits of Medical Interventions
Communicating The Uncertainty of Harms and Benefits of Medical Interventions
EISENBERG CLINICAL DECISIONS AND COMMUNICATIONS SCIENCE CENTER WHITE PAPER SERIES
Background. There is growing interest in shared medical uncertainty and its sources, as well as in its measurement.
decision making among patients, physicians, and policy The few studies that have assessed alternate means of
makers. This requires patients to interpret increas- communicating uncertainty dealt mostly with presenting
ing amounts of medical information, much of which is uncertainty about probabilities. Both patients’ and physi-
uncertain. Little is known about the optimal approaches cians’ interpretation of and responses to uncertainty may
to or outcomes of communicating uncertainty about the depend on their personal characteristics and values and
risks and benefits of treatments. Methods. The authors may be affected by the manner in which uncertainty is
reviewed the literature on various issues related to uncer- communicated. Conclusions. Research has not yet identi-
tainty in decision making: conceptualizing uncertainty, fied best practices for communicating uncertainty to
identifying its potential sources, assessing uncertainty, patients about harms and benefits of treatment. More con-
potential methods of communicating uncertainty, poten- ceptual, qualitative, and quantitative studies are needed
tial outcomes of communicating uncertainty, and current to explore fundamental questions about how people pro-
practices and recommendations by expert groups on com- cess, interpret, and respond to various types of uncer-
municating uncertainty. Results. There are multiple tainty inherent in clinical decisions. Key words:
sources of uncertainty in most medical decisions. There uncertainty; risk communication; decision making. (Med
are conceptual differences in how researchers define Decis Making 2007;27:681–695)
6% for intermediate decisions, and 16.6% for com- could reasonably be attributed to the measurand. The
plex decisions. parameter may be a standard deviation or the width
The rapid growth of medical knowledge has of a confidence interval.’’ Sources of this type of
spurred interest in rating the quality of medical uncertainty include data acquisition, sampling, quan-
evidence. Sophisticated rating systems have been tification, and interpolation. These groups distinguish
developed to stratify evidence according to its scien- between uncertainty and error, in which uncertainty
tific credibility and level of uncertainty. Consider- represents the range of all determinations, whereas
able progress has been made in developing such error refers to the difference between an individual
rating systems by initiatives such as the Cochrane result and the true value of the measurand. Error con-
Collaboration, Evidence-Based Practice Centers, US sists of a random component and a systematic compo-
Preventive Services Task Force, and Consumer Rep- nent but is an idealized concept, and its magnitude
orts on Health. In contrast, researchers have only cannot be known exactly.
begun to investigate ways to present scientific uncer- The variety of ways in which uncertainty has
tainty to health care consumers. been defined and conceptualized reflects its many
With increasing attention paid to helping patients potential sources. Because our primary interest is in
use research evidence to inform personal health informed decision making in the health care
decisions, this review was undertaken to address domain, we have chosen to focus specifically on
the following questions: How do we conceptualize uncertainties that relate to patients’ understanding
uncertainty? How do we assess uncertainty? What of the outcomes of medical interventions. For our
are the best practices for communicating uncertainty purposes, therefore, we differentiate between 5 main
about harms and benefits of treatment? How do types or sources of uncertainty: 1) risk, or uncer-
patients and physicians respond to uncertainty? tainty about future outcomes; 2) ambiguity, or
Where are the gaps in the literature? uncertainty about the strength or validity of evi-
dence about risks; 3) uncertainty about the personal
significance of particular risks (e.g., their severity,
CONCEPTUALIZING UNCERTAINTY AND ITS timing); 4) uncertainty arising from the complexity
SOURCES of risk information (e.g., the multiplicity of risks and
benefits or the instability of risks and benefits over
There are various definitions and interpretations of time); and 5) uncertainty resulting from ignorance.
uncertainty. The Merriam-Webster dictionary defines These different sources of uncertainty are not often
uncertainty as ‘‘the state of being indefinite, indeter- distinguished in the risk communication literature
minate, unreliable, unknown beyond doubt, not but should be clarified when discussing uncertainty
clearly identified or defined, and/or not constant.’’6 in medical decision making.
Another proposed definition of uncertainty is ‘‘a Fundamental uncertainty about the future occur-
cognitive state created when an event cannot be ade- rence or nonoccurrence of a given outcome lies at
quately defined or categorized due to lack of infor- the core of the notion of risk. Risk estimates describe
mation.’’7, 8 In the health domain, Mishel9 has this uncertainty in probabilistic terms and are
broadly defined uncertainty in illness as ‘‘the inability derived from empirical observations of an outcome’s
to determine the meaning of illness related events,’’10 occurrence within a given population. Yet risk esti-
which results from the ambiguity, complexity, and mates embody additional uncertainties as well.
unpredictability of illness10 or deficiency of informa- First, because risk estimates are not truly predictive
tion about one’s illness and its consequences. but rather postdictive (i.e., they explain past pat-
In measurement and science, uncertainty often terns of occurrence in a reference population), their
has a slightly different and technical meaning related use for the purposes of informing decisions regard-
to imprecision in measurement. The National Insti- ing an individual’s outcome requires a leap of faith
tute of Standards and Technology11 defines the uncer- that the future, and all of its deterministic elements,
tainty of a measured result as the standard deviation will be the same as the past.
of the collection of data samples approximating the An equally important problem is that risk esti-
measurand, the quantity being measured. Similarly, mates necessarily have limited applicability at the
the International Vocabulary of Basic and General individual level. Many argue that risk indeed loses
Terms in Metrology12 defines uncertainty as ‘‘a para- precise meaning with respect to individual persons
meter associated with the result of a measurement, and events because individuals either will or will
that characterizes the dispersion of the values that not be affected.13, 14 Although risk estimates might
accurately predict the aggregate number of outcomes appropriate interpretation of a CI requires an under-
in a population, they cannot specify their exact dis- standing of the difference between accuracy, or how
tribution15 nor what is most critical for decision close an estimate is to the actual or true value, and
making at the individual level: the future outcome of precision, or the reproducibility of the estimate. CIs
any given person. No matter how much any given reflect the precision of an estimate but not its accu-
person resembles other individuals statistically iden- racy. As such, CIs capture some—but not all—of the
tified as falling into one risk category or another, ambiguity pertaining to a risk estimate.
any one person’s outcome is unknowable and may The 3rd and 4th types of uncertainty in our
diverge from the categorical norm. The irreducible schema relate to the personal significance of risks
nature of this uncertainty poses a fundamental limit and to uncertainty arising from the complexity of
to the value of risk estimates in individual decision risk information, respectively. In the health care
making. In addition, there are some types of events domain, the clinical significance of risks for various
that are beyond the realm of prediction.16 outcomes is often unclear. For example, clinicians
The 2nd type of uncertainty in our schema relates and patients may be uncertain about the severity of
not to the occurrence of future outcomes but to the these outcomes, their tolerability, scope, timing, or
quality of the risk information at hand. Decision temporal impact. At the same time, uncertainties
theorists have used the terms ambiguity and vague- may arise from the sheer complexity of risk informa-
ness to describe this type of uncertainty, which tion. For example, decision makers often need to
relates to the ‘‘reliability, credibility, or adequacy’’ process and interpret multiple risks simultaneously
of risk information.17;18 Ambiguity or vagueness is and to make sense of risks that change over time
high when risk information is unreliable, conflict- and as a consequence of different actions.20 The
ing, incomplete, unknown, unknowable, or when optimal way of processing risk information in these
expert knowledge is contested. This type of uncer- circumstances is not clear even for expert decision
tainty, which relates to the strength of scientific evi- makers or complex computerized decision support
dence, has several sources, including missing or systems, and it depends on various assumptions
inconsistent empirical data or conflicting expert that are difficult to make explicit. Additional uncer-
opinions and recommendations. tainties about the correct diagnosis and about the
The strength of scientific evidence is affected natural history of a disease may further magnify
by many factors. These include, but are not limited any uncertainty about the outcomes of treatment
to, study design (randomized controlled trial v. because the likelihood of benefit (but not risk) is pre-
observational), blinding, duration of treatment and dicated on the patient’s having the condition in
follow-up, appropriateness of the outcome measures question.
used, controlling of confounders in design and ana- Uncertainty may also arise from ignorance of rele-
lysis, sample size, and sample population. Inade- vant information. Evaluating risks, making clinical
quacies in any of these factors can affect the validity diagnoses, and making treatment decisions within
(i.e., increase uncertainty) of the study’s findings, the confines of a brief clinical encounter require that
making it more prone to bias or chance events. many facts be left unknown, unspoken, or crudely
Extrapolating risks derived from studies examining summarized. This can lead to uncertainty about the
one duration of treatment to another, from one for- current state of a patient. Many doctors fail to
mulation to another, from a composite end point to review21 or document family history,22 and patients
its components, or from one population to another have been shown to inaccurately recall important
also introduce ambiguity, as do differences in how risk factors that are used to calculate risks.22 This
the same studies are interpreted. For example, a type of uncertainty is common in many clinical cir-
study examining congruence among different meta- cumstances yet is difficult to assess.
analyses19 found differences among reported find-
ings to be relatively small in comparison to substan-
tial disagreement in the authors’ interpretations PROBLEMS IN ASSESSING UNCERTAINTY
with regard to clinical applications of the findings.
Uncertainty regarding a risk estimate is often exp- Aside from the conceptual problems involved in
ressed statistically through the use of a confidence defining uncertainty and its sources, assessing
interval (CI) around a point estimate of risk. CIs uncertainty presents an additional challenge. To
express the dispersion around a point estimate promote informed decision making in health care,
arising from sampling issues and sample size. The one would like to be able to specify or quantify the
uncertainty pertaining to information about the risks overestimate the impact and underestimate the
and benefits of a given intervention. Yet this task is uncertainty of treatment on more important end
made problematic not only by the variety of types of points.
uncertainty (some of which may not lend them- A further difficulty in assessing uncertainty is
selves to quantification at all) but also by methodo- introduced by the different types of summary statis-
logical difficulties. tics used to express risk. For example, there is gen-
For example, even if one focuses only on uncer- eral concurrence that the risks of treatment should
tainty pertaining to the statistical precision of a risk be communicated as absolute risks (ARs), not RRs.25
estimate (a type of ambiguity), the assessment and Communicating the benefits and risks of treatment
calculation of uncertainty through the use of confi- in relative rather than absolute terms can influence
dence intervals (or P values) can be challenging. a patient’s perception of a therapy’s effectiveness,26
One problem is that it may be possible to apply making the benefits of a treatment appear more
more than one CI when expressing ambiguity, and favorable27 or conversely, emphasizing its risks.28
the choice of CI is itself ambiguous. A data set can These framing effects concerning the point estimates
be examined in its entirety (larger N, smaller CI) or of risks have been well studied and have been shown
in subgroups (smaller N, larger CI), and risk esti- to affect treatment decisions. However, relatively lit-
mates from these alternative analyses can differ sub- tle is known about how to estimate or communicate
stantially in magnitude and dispersion. Although the associated CI of absolute risks. Converting an RR
the overall risk may appear to be more precise than into an AR is relatively simple, but computing the CI
the risks associated with subgroups because of tigh- associated with the calculated AR is not.29 When
ter CIs, a subgroup analysis may more accurately ARs are not reported, or when the patient differs
reflect the risk pertinent to members of that sub- from the cohort on which the reported ARs are
group (but with less precision). For example, the based, ARs and their associated CIs must be calcu-
National Surgical Adjuvant Breast and Bowel Project lated. To translate RRs into ARs, the individual’s
P-1 trial23 found that tamoxifen increased the risk of baseline (pretreatment) risk for a condition is com-
endometrial cancer 2.5-fold (relative risk [RR] = 2.53, bined with the RR of the treatment on that condition
CI = 1.35 − 4.97) among all women. However, age using simple multiplication. For example, a person
was an effect modifier: The RR was 1.21 (0.41–3.6) with a 10% baseline risk who takes a treatment that
among women younger than 51 years but 4.01 (1.7– has an RR of 2.5 for that condition will experience
10.9) among women 51 years or older. In such cases, an AR of 25% (0.10 × 2.5) with treatment.
it is not clear which CI best reflects the point esti- Estimating the CI for the AR, however, is more
mate and associated uncertainty pertaining to an challenging. Simply finding the CI associated with an
individual woman: the more accurate estimate that individual’s baseline risk estimates can be difficult
factors in her age, or the more precise estimate that because these often appear in monographs30 or older
does not. journals that may be difficult to access. Once the pub-
Another problem in assessing uncertainty is lication is found, identifying the appropriate CI for a
posed by the use of composite end points, which risk estimate typically involves tedious multistep
combine more than one clinical end point into a sin- table look-ups, graphical interpretations, and cross-
gle measure of impact. Composite end points typi- comparisons.29–31 Finally, these published tables and
cally combine more frequent but less clinically graphs require many simplifications from the exact
important events (e.g., laboratory abnormalities) calculations, thus introducing further uncertainty.
with less frequent but more clinically important For example, a 40-year-old woman whose 30-year
events (e.g., death, stroke) to increase statistical breast cancer risk is calculated exactly as being 7.3%
power (by increasing the event rate).24 However, (CI = 7–8) is estimated from the tables and graphs as
these events may differ substantially in terms of having a risk of 7.5% (6–10).32 The developers of key
their effects, the uncertainty surrounding their risk prediction models have emphasized the uncer-
effects, and their clinical significance. These differ- tainty associated with their models and the impor-
ences among the events that are integrated within tance of conveying this uncertainty to patients:
composite end points make the specification of ‘‘All of the estimates discussed here are associated
uncertainty very difficult and may lead to biased with error; this fact should be transmitted to the
estimates. Composite end points that are largely patient should any estimates be given in a clinical
based on less important clinical end points may setting.’’29ðp204Þ Nonetheless, when such risk models
are promulgated,33 their associated CIs are typically literature on approaches to communicating risks (i.e.,
not reported. probabilities or likelihoods), little attention has been
To estimate the CI of a calculated AR, one needs given to communicating uncertainty.
to combine the CI surrounding the patient’s baseline In this section, we review possible approaches for
risk with the CI of the RR. Although these calcula- communicating uncertainty and the limited evi-
tions are intended to measure the uncertainty of the dence for the effectiveness of these approaches. For
AR, they also introduce a substantial amount of the most part, the discussion will relate to the com-
uncertainty. This uncertainty stems from differences munication of statistical uncertainty, or ambiguity,
between the study population used to develop the given the existing literature. However, much of what
risk model and the individual to whom it is being is written on this topic concerns communication of
applied, as well as the extent to which the model risk, not uncertainty.
may be misspecified. Misspecification may result Uncertainty and ambiguity may be communicated
from omission of important but as yet unidentified in a variety of ways, including verbally, numerically,
risk factors associated with the disease of interest or or visually. Verbal methods using subjective descrip-
failure to include specific study design elements tive words such as highly uncertain have the advan-
such as matching schemes.29 Combining the CI of tage that people think they understand what is being
the 2 distributions assumes that the 2 are indepen- said; however, interpretation of such terms has been
dent, an assumption that is unlikely to be true. The shown to be highly variable.3436 The use of num-
CI of risk prediction models are tightest in the mid- bers to depict uncertainty and ambiguity potentially
dle of their prediction range (i.e., for those at average allows for more precision and avoids variable inter-
risk) and widen the further the deviation from the pretation. However, there are many different ways in
average. On the other hand, because most clinical which information about risks may be communi-
trials are conducted on high-risk populations, their cated quantitatively. They can be described using
tightest CIs are typically among those at higher risk. RRs, ARs, frequencies, or number needed to treat;
These and other methodological uncertainties the measure used can affect how the information is
confound the quantification of uncertainty regarding understood and answered. Ambiguity might be com-
risk estimates and pose challenges for the task of municated in terms of any or all of these summary
communicating uncertainty to individual decision statistics; however, these possibilities have not been
makers. evaluated.
Furthermore, many people, including experts,
POTENTIAL APPROACHES TO have difficulty understanding and combining statisti-
COMMUNICATING UNCERTAINTY cal information effectively.37 Numeracy—the ability
to comprehend quantitative information—is manifest
Problems in both the conceptualization and in the way in which people process statistical infor-
assessment of uncertainty have important implica- mation. One study38 found that individuals with high
tions for the task of communicating uncertainty. numeracy were less biased by framing effects and
There are multiple types of uncertainty that clini- more influenced by affective meaning than were those
cians may want and need to communicate to with low numeracy. Another study39 found that indi-
patients, and the assessment of these uncertainties is viduals with low numeracy tend to better compre-
not straightforward. Furthermore, the communica- hend information about the comparative quality of 2
tion of uncertainty to patients may serve different choices when it is simplified (e.g., by highlighting
purposes and goals, for example, to convey doubt or relevant or meaningful information). It is also possible
to increase the level of confidence in a finding, to that numeracy is reflected in people’s ability to under-
inform patients about their estimated disease risk stand and process information about statistical uncer-
and the limitations of these estimates, or to help tainty or ambiguity, but this possibility has not been
patients understand the general complexity or unpre- explored.
dictability of illness and its management. Moreover, Visual depictions of data may facilitate rapid
these goals are not always consistent with one understanding of numerical expressions of uncer-
another and may require focusing on particular types, tainty, particularly if large amounts of numerical
assessments, or approaches to communicating uncer- information must be presented. However, little
tainty. Yet we know very little about the optimal is known about how ambiguity might be depic-
approaches and outcomes of communicating different ted visually. Limited work suggests that the visual
types of uncertainty. Although there is a growing depiction of statistical uncertainty may exploit the
ability of the eye to quickly detect discontinuities in Table 1 Categories of Effectiveness Used by the
an image and to interpret these discontinuities as BMJ Clinical Evidence Series5
areas with distinct data characteristics. Techniques
that have been used to depict uncertainty have Intervention Icon Description
Animation effects have been used to simplify the smallcompared with the benefits.
Table 2 How the US Preventive Services Task Force Grades Its Recommendations48
Good A B C D
Fair B B C D
The US Preventive Services Task Force (USPSTF) grades its recommendations based on the strength of evidence and magnitude of net benefit (benefits
minus harms).
A. The USPSTF strongly recommends that clinicians provide [the service] to eligible patients. The USPSTF found good evidence that [the service]
improves important health outcomes and concludes that benefits substantially outweigh harms.
B. The USPSTF recommends that clinicians provide [the service] to eligible patients. The USPSTF found at least fair evidence that [the service]
improves important health outcomes and concludes that benefits outweigh harms.
C. The USPSTF makes no recommendation for or against routine provision of [the service]. The USPSTF found at least fair evidence that [the service]
can improve health outcomes but concludes that the balance of benefits and harms is too close to justify a general recommendation.
D. The USPSTF recommends against routinely providing [the service] to asymptomatic patients. The USPSTF found at least fair evidence that [the ser-
vice] is ineffective or that harms outweigh benefits.
I. The USPSTF concludes that the evidence is insufficient to recommend for or against routinely providing [the service]. Evidence that [the service] is
effective is lacking, of poor quality, or conflicting, and the balance of benefits and harms cannot be determined.
The USPSTF grades the quality of the overall evidence for a service on a 3-point scale (good, fair, poor).
Good: Evidence includes consistent results from well-designed, well-conducted studies in representative populations that directly assess effects on
health outcomes.
Fair: Evidence is sufficient to determine effects on health outcomes, but the strength of the evidence is limited by the number, quality, or consistency of
the individual studies; generalizability to routine practice; or indirect nature of the evidence on health outcomes.
Poor: Evidence is insufficient to assess the effects on health outcomes because of limited number or power of studies, important flaws in their design or
conduct, gaps in the chain of evidence, or lack of information on important health outcomes.
substantial, moderate, small, zero), assigning letter that imply a CI but that are not explicitly defined
grades (A, B, C, and D) to various combinations of (‘‘between 17% and 25% are pain-free’’). Uncer-
these 2 rating scales (Table 2). Both systems attempt tainty about individual variation is communicated
to surmise the likelihood that the benefits outweigh by stating,
the risks and blend uncertainty with the magnitude
of the net benefit. It’s very important for you to know that people respond
The popular lay publication Consumer Reports differently to the various [treatments]. You may get lit-
Best Buy Drugs49 explains the benefits and risks of tle or no relief from one and respond much better to
treatments to consumers by using subjective terms another . . . . But it’s possible that you would not be
(‘‘about 30%,’’ ‘‘up to 7%’’) and by reporting ranges helped by one of the three [treatments]. In that case,
you should try one of the others. Doctors are very used behavior change in response to information.5658
to trying another [treatment] if the first one they pre- Health information might also be tailored according
scribe for you does not seem to be working. to patients’ individual disease risks or risk factors.
For example, the provision of individualized dis-
Data are lacking regarding outcomes of these ease risk estimates has been found to increase
approaches to communicating uncertainty for either patients’ uptake of screening tests, particularly in
physicians or patients involved in treatment deci- patients at high risk for disease.59 Tailoring the com-
sion making. According to the recent Cochrane munication of health information according to
review of decision aids, for instance, few of the 131 patients’ individual disease risks, however, is pro-
decision aids reviewed included a description of the blematic given the substantial uncertainties implicit
level of uncertainty about the evidence.50 in individual risk estimates.14 Furthermore, the tai-
loring of any type of health information according to
POTENTIAL OUTCOMES OF COMMUNICATING individual patient characteristics still raises a larger
UNCERTAINTY question about the optimal amount and type of
information to be conveyed to patients from an ethi-
cal, as well as a psychological, standpoint. Until
To identify appropriate ways to communicate
these issues are resolved, a critical issue is to better
uncertainty to patients, it is important to understand
understand the potential outcomes of communicat-
the potential outcomes of this endeavor. Although
ing uncertainty to patients in general.
empirical evidence in the health domain is lacking,
Empirical and theoretical work on judgment and
there is a substantial literature on the psycho-
decision making provides insights on the potentially
logical and behavioral effects of uncertainty in other
negative effects of communicating uncertainty about
decision-making domains.
risk information and the mechanisms for these
effects. A large body of past research has shown that
COGNITIVE AND BEHAVIORAL OUTCOMES both laypersons and experts trained in statistics tend
OF COMMUNICATING UNCERTAINTY to simplify and use mental shortcuts, or heuristics,
when interpreting risk information.60 Patients often
One concern about communicating uncertainty is simplify probabilistic information into 2 broad cate-
that doing so may have deleterious as well as benefi- gories (e.g., ‘‘I will get the disease’’ or ‘‘I will not get
cial effects. To the extent that communicating the disease’’61 ). People’s reliance on heuristics and
uncertainty entails the transfer of additional, com- other simplifying strategies points to inherent limita-
plex information, it has the potential to overwhelm tions in human cognitive processing.62 These limita-
and confuse patients and to impair their ability to tions have been described extensively and raise
make truly informed decisions. Ample research in the possibility that people may not always respond
decision psychology has shown that a surfeit of in adaptive ways to the provision of complex prob-
information does not necessarily facilitate informed abilistic information. For example, some have
decision making51 and may in fact hinder it.39 Full argued that decision makers have a fundamental pre-
disclosure of all the benefits and risks associated ference and need to rely on mental representations
with a medical treatment may exceed patients’ capa- of the bottom-line meaning or gist of a choice option,
city to process and use this information effect- rather than the detailed or verbatim representation
ively.52 The provision of added information about that includes the accompanying details.63, 64 For
medical uncertainty may have similar outcomes these reasons, decision makers may find details
with respect to patient understanding and decision about the uncertainty pertaining to risk information
making. of limited relevance or overly difficult to process or
A potentially effective way of mitigating these understand.
negative outcomes of complex health information Also relevant to the discussion of potential out-
may be to tailor the information to individual comes of communicating uncertainty in health care
patients, altering the specific type or amount of is the literature on decision psychology. A number of
information presented according to various charac- studies have shown that people generally prefer to
teristics (e.g., gender, culture, education, psycholo- avoid ambiguity; when confronted by ambiguous
gical factors, or behaviors of interest5355 ) that relate information about risks, people tend to appraise
to patients’ capacity to use or respond to such infor- these risks pessimistically and avoid making deci-
mation. Presumably, such tailoring enhances sions.17, 65, 66 This phenomenon has been termed
ambiguity aversion and has been demonstrated in a uncertainty can motivate patients to seek discus-
number of decision-making settings. Ambiguity aver- sions with their health care providers, which may
sion has been shown to persist even when odds favor comfort patients who prefer to take an active deci-
the ambiguous option,6769 although the level of risk sion-making role in their medical care. However,
appears to affect people’s reactions to ambiguity. At uncertainty may also lead to information avoidance
very low probabilities of the unambiguous option, and confusion if patients lack the proper resources
decision makers become indifferent to ambiguity and to interpret available information and manage
may even become ambiguity seeking.68, 70, 71 Ambigu- uncertainty. In one study, for example, 13% of peo-
ity aversion may also depend on whether gains ple who were tested for HIV never received their
or losses are at stake.65, 66, 72 With potential gains results,85 even though in a separate study, those
(e.g., winning money), people are ambiguity averse, who initially avoided learning their HIV status
whereas with potential losses (e.g., losing money), showed an improvement in mood upon receiving
people are ambiguity seeking. However, this finding their test results (regardless of their HIV status).86
has not been consistently obtained,69 and ambiguity Information avoidance may be used as a coping
aversion itself varies considerably among individuals strategy by people who have difficulty tolerating
and across different decision-making circumstan- potential but uncertain negative health conse-
ces.73 Ambiguity aversion may also vary according to quences. A similar example is those who receive or
how information on ambiguity is framed, for exam- fill prescriptions but do not take medications
ple, in terms of gains versus losses74 or verbally ver- because of their fear of side effects.
sus quantitatively.73
Ambiguity aversion may have various psychologi-
cal and behavioral manifestations. For example, Emotional Outcomes of Communicating
ambiguity regarding estimates of risks of an adverse Uncertainty
outcome may lead people to perceive themselves at
higher risk75 and to have greater distrust in the infor- Most of the literature about emotional responses
mation at hand.76 However, relatively little research to uncertainty pertains to uncertainty in a broad
has directly explored how ambiguity aversion may sense that includes, but is not limited to, the specific
be manifest in the domain of health-related deci- types of uncertainty outlined in this article. Lazarus
sions and outcomes. Using hypothetical scenarios, and Folkman’s87 stress and coping theory posits that
Ritov and Baron77 and Meszaros and colleagues78 the cognitive appraisal of a stressor such as uncer-
showed that increasing people’s awareness of ambi- tainty occurs in 2 steps. First, a person interprets the
guity about a vaccine’s safety made them reluctant to meaning of a stressor and its relationship to his or
receive it. Other studies have shown that percep- her experiences and values. Next, a person assesses
tions of ambiguity regarding health risks and disease his or her resources and capacity for coping with a
prevention recommendations are associated with given stressor.
ambiguity-averse perceptions and emotions and This theory maintains that patients who nega-
lower uptake of screening and preventive interven- tively appraise uncertainty might correspondingly
tions.7981 Intervention studies have demonstrated experience subsequent fear, anxiety, panic, and a
that informing people about uncertainties surround- desire to reduce uncertainty. These negative emo-
ing cancer screening measures decreases their inter- tional responses may lead to heightened vigilance
est in screening, also implying that people are about illness, which may further exacerbate worry
ambiguity averse.82;83 For instance, there is conflict- about illness.88 However, uncertainty might not
ing evidence about the effectiveness of breast self- always have negative emotional effects; when con-
examinations (BSEs) for early detection of breast fronted by uncertainty about an illness or treatment,
cancer. Following the disclosure of the uncertainty some people are able to maintain hope and opti-
about evidence, some patients may ignore the posi- mism. Optimism can encourage patients to accept
tive evidence about BSEs and decide not to perform treatments and maintain an active response to their
BSEs.84 illness if they perceive great potential benefits.84
Communicating uncertainty may also prompt dif- However, false hopes about treatments can lead some
ferent information-seeking behaviors. Some patients patients to ignore real risks of other lifestyle beha-
may respond to uncertainty by actively seeking viors. For instance, diabetic patients who believe that
information. The attempt to resolve uncertainty may insulin alone can control blood sugar levels might
help them to cope with it; knowledge about not additionally alter their dietary habits.
In addition to influencing emotions during Patients’ values and preferences for medical care
decision making, communicating uncertainty may may also affect how they perceive and respond to
influence patients’ emotional responses following a uncertainty about health-related risks. For example,
decision. Past research has shown that patients can those who are averse to medications might perceive
experience 3 types of regret following treatment uncertainty about medication risks more negatively
decisions: outcome regret, which is regret about a than those who are comfortable with prescription
negative health outcome following a decision; medications.35 As a result, they might decide not to
option regret, which is regret about the choice one take the medication because of risk uncertainty.
made; or process regret, which is regret about the Individual personality traits may affect pati-
quality of the decision-making process (e.g., it was ents’ responses to uncertainty. For example, social
too hasty).89 It is possible that communicating psychologists have identified 2 personality types
uncertainty can lead to increased satisfaction in the that may influence information-seeking behavior
quality of the decision-making process, thereby under uncertainty: uncertainty-oriented individuals,
reducing option or process regret following a deci- who tend to process uncertainty and seek out rele-
sion. However, it is also possible that communicat- vant information to allow for resolution of it, and
ing uncertainty can cause patients to blame certainty-oriented individuals, who tend to gravitate
themselves in the case of bad outcomes, leading to toward familiar situations that are less ambiguous
more regret.89 and who tend to rely on respected others to make
Although these findings relate to emotional res- decisions.93 Similarly, cognitive processing styles
ponses to uncertainty in a very general sense, they such as monitoring (vigilantly seeking and attending
may also be applicable to understanding the effects to information) or blunting (distracting oneself from
of particular types of uncertainty, such as ambigu- information, blunting its impact) can affect the man-
ity, as they additionally apply to the particular ner in which patients perceive risks and uncertain-
domain of health care decisions. These outcomes ties surrounding their health condition.94
need to be examined further. Finally, uncertainty may affect physicians as well
as patients, inducing anxiety and excessive concern
about bad outcomes. Physicians and medical trai-
Individual Differences in Patient Responses to nees tend to fear that patients will perceive them as
Uncertainty inadequate or ineffective,95;96 which may influence
their willingness to disclose uncertainty to patients97
Patients’ cognitive, emotional, and behavioral as well as their use of health care resources.98
responses to uncertainty are themselves uncertain Patients’ responses to uncertainty may also be
and may depend on several individual patient char- influenced by physicians’ reactions to it. Physicians
acteristics. For example, the patient’s desired role in are often concerned that acknowledging uncertainty
his or her medical care may be important. Some to patients may undermine patient trust and satis-
hypothesize that patients who consider themselves faction. However, some research has shown that
knowledgeable and competent prefer to make their patient satisfaction is affected by the manner in
own health decisions under uncertainty, and those which the physician handles uncertainty, not
who consider physicians the absolute experts prefer whether or not he or she presents uncertainty,99 con-
to defer decisions to physicians.90, 91 Others have sistent with previous theories.100 When physicians
suggested that ‘‘patients most want to introduce their are comfortable with uncertainty and collaborate with
own extra-medical values when medical factors their patients in their medical care, patient trust and
alone do not seem to be decisive.’’61ðp246Þ Deber and satisfaction are actually high.61
colleagues92 found that the type of decision-making
task can affect patients’ desired role in the process;
patients may desire involvement in tasks such as HELPING PATIENTS COPE WITH UNCERTAINTY
determining the acceptability of risks but may prefer
their physicians to identify treatment options, risks Because of patients’ complex cognitive, emotional,
and benefits of treatments, and probabilities of risks and behavioral responses to uncertainty, many argue
and benefits occurring. When faced with a mismatch that the focus of risk communication should be on
in decision-making preference and physician beha- helping patients tolerate and cope with uncertainty
vior, some patients might distrust their physician or rather than simply helping them understand it.84, 88
become less satisfied with their care. Uncertainty can be stressful and anxiety provoking
for patients throughout the illness experience.9, 61 harms to aid the decision making process. Both sources
Some have suggested practical ways for physi- of uncertainty should be acknowledged in comprehen-
cians to help their patients cope with uncertainty: sive discussions of risks in patient decision aids.’’103ðp14Þ
assure patients that they will answer all questions
about their health, refer patients to other sources In addition, there have been numerous articles in
such as reputable Web sites, remain open and the medical literature advocating that physicians
sympathetic to patients, inform patients of physi- discuss uncertainty. One states,
cians’ own biases and values, and inform patients
of alternative treatments.101, 102 Others propose While often difficult, a discussion of uncertainties is
that physicians should clarify the type of uncer- crucial for a patient’s comprehensive understanding of
tainty that is the most distressing to patients (e.g., the options. Thoughtful discussion can promote trust
uncertainty about probabilities, uncertainty about and encourage adherence. Examples: ‘‘The chance that
sources of information, uncertainty about evidence) this will help is excellent,’’ ‘‘Most patients with this
condition respond well to this medication, but not
and be available to explain the complexities of
all.’’2ðp2314Þ
each.84 Although these tasks make sense and are
ethically justifiable, it remains unclear whether they
On the other hand, some104;105 argue that the dis-
are truly beneficial and feasible for physicians to
persion around a point estimate is irrelevant when a
perform.
choice needs to be made between 2 uncertain options
and that the decision in such cases should be based
RECOMMENDATIONS BY EXPERT GROUPS on which option has the highest expected net bene-
fit.105 The costs of failing to adopt a new treat-
Some organizations have made explicit recommen- ment simply because the difference in net benefit is
dations concerning the need for and possible app- not statistically significant can be substantial. This
roaches to communicating uncertainty. The Interna- approach does not imply that treatment decisions
tional Patient Decision Aids Standards collaboration should be based on poor-quality evidence but rather
recommends communicating uncertainty and provid- that the amount of information that should be
ing concrete examples to help explain it103 : acquired is an empirical rather than ideological ques-
tion, differing across different clinical decisions.
It’s very important to acknowledge uncertainty in prob-
ability estimates. Often the uncertainty is large, espe- CONCLUSIONS: CRITICAL KNOWLEDGE GAPS
cially if evidence is scarce or events are rare. It’s AND FUTURE RESEARCH NEEDS
probably wise to do simple things such as rounding off
numbers (to avoid false illusions of precision), using Although the ideal of informed or shared decision
phrases like ‘‘our best guess is . . . ,’’ give ranges, or pro- making implies a need for communicating uncer-
vide 95% confidence intervals. tainty to patients, this task is problematic for many
Even with the best evidence from large studies, the reasons. From a conceptual standpoint, it is unclear
issue of stochastic uncertainty remains. Essentially, we which of the many types and sources of uncertainty
never quite know who are the patients who are going to clinicians should communicate to their patients.
be affected, and who the treatment is going to be most Exactly what is meant by the term uncertainty and
useful for. One way to deal with this uncertainty might the ethical justification for communicating different
be to say: ‘‘If 100 patients like you are given no treat-
types of uncertainty are themselves uncertain. More
ment for five years, 92 will live and eight will die.
work is needed to define the circumstances in
Whether you are one of the 92 or one of the eight, I do
not know. Then, if 100 patients like you take a certain
which uncertainty ought to be communicated.
drug every day for five years, 95 will live and five will Should this be a function of the magnitude of uncer-
die. Again, I do not know whether you are one of the 95 tainty surrounding these options and/or the magni-
or one of the five . . . .’’ tude of the consequences of these choices? Should it
Despite these limitations from uncertainty, practi- depend on the number of choices available, the
tioners generally feel that we can still try to make deci- novelty of the treatments being considered, or
sions about what the best treatment plan is for an whether there is a clear dominant option? What
individual person, based on what happens to these aspects of uncertainty should be communicated,
groups of patients in the studies. Hence the value, it is given the task? Are there patient characteristics that
thought, of presenting the information about benefits and may influence when we should or should not
include uncertainty in risk communication or how More research is needed on potential tradeoffs in
the subject should be approached? communicating uncertainty. Tailoring information
Many of these questions are conceptual and ethical to the individual’s risks has the advantages of mak-
in nature, rather than empirical, and raise the larger ing information more relevant and reducing the
question of whether there is any level of acceptable volume of information, but the more tailored the
uncertainty. Deciding between various treatment information, the greater the uncertainty associated
options is inherently situation specific, and there is with risk prediction. Thus, there may be an inherent
no universally acceptable absolute level of acceptable tradeoff between the degree of personalization of
risk; one’s acceptance of risk is contingent on many risk information and its uncertainty. There may also
factors.106;107 Acceptable risk refers to the risk asso- be a tradeoff between the completeness of informa-
ciated with the most acceptable option in a particular tion given and the efficacy of this information in
decision.107 Is there an analogous concept of accepta- terms of how it is processed and responded to. In
ble uncertainty? If so, how might different ways of addition, there may be a tradeoff between the preci-
judging thresholds of acceptable uncertainty be sion of a risk estimate and its accuracy. Further
defined, and from whose perspective? exploration of these potential tradeoffs could help
From an empirical standpoint, we know little inform discussion of when and how to communi-
about how to measure and quantify uncertainty and cate uncertainty.
the various factors contributing to it. What types of The increasing focus on personalized medicine108
uncertainty do commonly used measures of uncer- mandates a more sophisticated understanding of the
tainty (such as confidence intervals) actually cap- limitations and errors in applying and communicat-
ture? Can measures of specific contributors to ing population-based, epidemiologic findings to the
uncertainty be combined to estimate the total or individual. Because of the poor positive predictive
composite uncertainty in a finding? Exploration of value of most of the risk factors (including genetic
the use of different characterizations of risk and markers) for common noninfectious diseases, most
uncertainty (i.e., uncertainty about the time to event, people who will eventually get a disease will not be
a continuous variable, rather than the likelihood of designated as high risk by population-based risk
the event, a dichotomous variable at the individual prediction tools.14, 109 Furthermore, how patients
level) may help move the field forward given the respond to personalized risk estimates is poorly
irreducible uncertainty surrounding individual pre- understood. Fischoff writes, ‘‘Asking people about
dictions when using the likelihood of an event as the the risks to others like themselves is not the same as
outcome measure. Research is needed to develop asking them about their personal risk. Nor need
these measures of component and composite uncer- reports about others’ risk levels be taken perso-
tainty and to validate them; they then need to be dis- nally.’’109 Exactly how to use risk estimates and risk
seminated to decision makers and researchers. prediction tools to improve and inform individual
A further empirical problem is that the optimal treatment decisions, while acknowledging and com-
methods and outcomes of communicating different municating their limited power to predict indivi-
types of uncertainty are not known. There is sugges- dual futures, is a critical challenge that will become
tive, but not definitive, evidence about how various even more important as new disease biomarkers are
types of uncertainty are differentially interpreted by discovered.
patients, clinicians, and researchers and how these More work is also needed to differentiate the con-
interpretations affect clinical decision making as struct of uncertainty from risk in various conceptual
well as other patient-centered outcomes such as per- models and theories of health behavior. Divergent
ceptions and well-being. The manner in which conceptual models exist for understanding how
uncertainty is communicated can affect how it is people perceive and respond to uncertainty; further
perceived and responded to, but little is known about exploration of new models and attempts to reconcile
the mechanisms of these framing effects. It is unclear differences across models could help advance these
whether uncertainty is best presented verbally, fields of inquiry.
numerically, graphically, or using multiple formats.
Studies exploring how perceptions of and responses RECOMMENDATIONS
to uncertainty are potentially affected by its framing
(i.e, certainty v. uncertainty, gain v. loss), choice of At this time, there are no clear best practices for
specific terms, measurement units and scale, and gra- presenting uncertainty. The best method of present-
phics would help to clarify this critical issue. ing uncertainty depends on the task required of the
patient and the type of uncertainty presented. There 15. Diamond GA. What price perfection? Calibration and discri-
is little evidence about the effect of alternate types mination of clinical prediction models. J Clin Epidemiol. 1992;
45(1):85–9.
or presentations of uncertainty on patient under-
standing, satisfaction, and informed decision mak- 16. Bernstein PL, ed. Against the Gods: The Remarkable Story of
Risk. New York: John Wiley & Sons; 1996.
ing in clinical medicine.
17. Ellsberg D. Risk, ambiguity, and the savage axioms. Q J Econ.
More conceptual, qualitative, and quantitative
1961;75:643–69.
studies are needed to explore fundamental ques-
18. Kuhn KM, Budescu DV, Hershey JR, Kramer KM, Rantilla
tions about how people process, interpret, and AK. Attribute tradeoffs in low probability/high consequence
respond to various types of uncertainty inherent in risks: the joint effects of dimension preference and vagueness.
clinical decisions. Much more research is also Risk Decision and Policy. 1999;4(1):31–46.
needed to broaden the scope of inquiry beyond the 19. Chalmers TC, Berrier J, Sacks HS, Levin H, Reitman D, Naga-
problems of communicating statistical uncertainty lingam R. Meta-analyses of clinical trials as scientific discipline.
around a point estimate to the many other types of II. Replicate variability and comparison of studies that agree and
uncertainty relevant to the numerous decisions that disagree. Stat Med. 1987;6(7):733–44.
patients and clinicians face. 20. Rockhill B, Colditz GA, Rosner B. Bias in breast cancer ana-
lyses due to error in age at menopause. Am J Epidemiol. 2000;
151(4):404–8.
REFERENCES
21. Wilkins-Haug L, Erickson K, Hill L, Power M, Holzman GB,
Schulkin J. Obstetrician-gynecologists opinions and attitudes on
1. Braddock C III, Fihn SD, Levinson W, Jonsen AR, Pearlman
the role of genetics in women’s health. J Womens Health Gend
RA. How doctors and patients discuss routine clinical decisions:
Based Med. 2000;9(8):873–9.
informed decision making in the outpatient setting. J Gen Intern
Med. 1997;12(6):339–45. 22. Murff HJ, Byrne D, Haas JS, Puopolo AL, Brennan TA. Race
and family history assessment for breast cancer. J Gen Intern Med.
2. Braddock CH, Edwards KA, Hasenberg NM, Laidley TL,
2004;20(1):75–80.
Levinson W. Informed decision making in outpatient practice:
time to get back to basics. JAMA. 1999;282(24):2313–20. 23. Fisher B, Constantino JP, Wickerham DL, et al. Tamoxifen for
prevention of breast cancer: report of the National Surgical Adju-
3. Woolf S, Chan EC, Harris R, et al. Promoting informed
vant Breast and Bowel Project P-1 Study. J Natl Cancer Inst. 1998;
choice: transforming health care to dispense knowledge for deci-
90(18):1371–88.
sion making. Ann Intern Med. 2005;143(4):293–300.
24. Ferreira-González I, Busse JW, Heels-Ansdell D, et al. Pro-
4. Stokey E, Zechouser R. A Primer for Policy Analysis. New
blems with use of composite end points in cardiovascular trials:
York: WW Norton & Company; 1978.
systematic review of randomised controlled trials. BMJ. 2007;334
5. BMJ Clinical Evidence. 14 March 2007. Available from: URL: (7597):786.
http://www.clinicalevidence.com/ceweb/about/guide.jsp
25. Edwards A, Elwyn G, Mulley A. Explaining risks: turning
6. Merriam-Webster. Merriam-Webster’s online dictionary. numerical data into meaningful pictures. BMJ. 2002;324:827–30.
Available from: URL: http://www.m-w.com/
26. Hux JE, Naylor CD. Communicating the benefits of chronic
7. Hilton BA. Perceptions of uncertainty: its relevance to life- preventive therapy: does the format of efficacy data determine
threatening and chronic illness. Crit Care Nurs. 1992;12(2):70–3. patients’ acceptance of treatment? Med Decis Making. 1995;15:
8. McCormick KM. A concept analysis of uncertainty in illness. 152–7.
J Nurs Scholarsh. 2002;34(2):127. 27. Forrow L, Taylor WC, Arnold RM. Absolutely relative: how
9. Mishel MH. Reconceptualization of the uncertainty in illness research results are summarized can affect treatment decisions.
theory. Image J Nurs Sch. 1990;22(4):256–62. Am J Med. 1992;92:121–4.
10. Mishel MH. The measurement of uncertainty in illness. Nurs 28. Bucher HC, Weinbacher M, Gyr K. Influence of method of
Res. 1981;30(5):258–63. reporting study results on decision of physicians to prescribe drugs
11. Taylor BN, Kuyatt CE. Guidelines for evaluating and expres- to lower cholesterol concentration. BMJ. 1994;309(6957):761–4.
sing the uncertainty of NIST measurement results. NIST Techni- 29. Claus EB. Risk models used to counsel women for breast and
cal Note 1297. Gaithersburg (MD): National Institute of Standards ovarian cancer: a guide for clinicians. Fam Cancer. 2001;1:
and Technology; 1994. 197–206.
12. Ellison S, Rosslein M, Williams A. Quantifying Uncertainty 30. Cupples LA, Ramaswamy R, Christiansen JC, Belanger AJ,
in Analytical Measurement. 2nd ed. St. Gallen (Switzerland): D’Agostino RB, eds. The Framingham Study: An Epidemiological
EURACHEM/CITAC; 2000. Investigation of Cardiovascular Disease, Section 38: The Relation-
13. Pearl J. Probabilistic Reasoning in Intelligent Systems: Net- ship Between Sporadically Measured Variables and the Risk of
works of Plausible Inference. San Francisco: Morgan Kaufmann; Cardiovascular Disease and Death: The Framingham Heart Study,
1988. 30-34 Year Follow-up. Bethesda (MD): National Heart, Lung, and
14. Rockhill B. Theorizing about causes at the individual level Blood Institute; 1988.
while estimating effects at the population level: implications for 31. Gail MH, Pfeiffer RM. On criteria for evaluating models of
prevention. Epidemiology & Society. 2005;16(1):124–9. absolute risk. Biostatistics. 2005;6(2):227–39.
32. Benichou J, Gail MH, Mulvihill JJ. Graphs to estimate an indi- 54. Kreuter MW, Oswald DL, Bull FC, Clark EM. Are tailored
vidualized risk of breast cancer. J Clin Oncol. 1996;14:103–10. health education materials always more effective than non-
33. National Cancer Institute. Breast Cancer Risk Assessment tailored materials? Health Educ Res. 2000;15(3):305–15.
Tool (BCRAT). Washington, DC: National Cancer Institute; 2007. 55. Kreuter MW, et al. Achieving cultural appropriateness in
34. Mazur DJ, Hickam DH. Patients’ interpretations of probabil- health promotion programs: targeted and tailored approaches.
ity terms. J Gen Intern Med. 1991;6(3):237–40. Health Educ Behav. 2003;30(2):133–46.
35. Bogardus ST, Holmboe E, Jekel JF. Perils, pitfalls, and possi- 56. Cassell MM, Jackson C, Cheuvront B. Health communication
bilities in talking about medical risk. JAMA. 1999;281:1037–41. on the Internet: an effective channel for health behavior change? J
Health Commun. 1998;3(1):71–79.
36. Mazur DJ, Merz JF. Patients’ interpretations of verbal expres-
sions of probability: implications for securing informed consent 57. Revere D, Dunbar PJ. Review of computer-generated outpati-
to medical interventions. Behav Sci Law. 1994;12(4):417–26. ent health behavior interventions: clinical encounters ‘‘in absen-
37. Hoffrage U, Lindsey S, Hertwig R, Gigerenzer G. Communi- tia.’’ J Am Med Inform Assoc. 2001;8:62–79.
cating statistical information. Science. 2000;290(5500):2261–2. 58. Prochaska JO, Velicer WF, Rossi JS, et al. Stages of change
38. Peters E, Vastfjall D, Slovic P, Mertz CK, Mazzocco K, Dickert and decisional balance for 12 problem behaviors. Health Psychol.
S. Numeracy and decision making. Psychol Sci. 2006;17(5):407–13. 1994;13(1):39–46.
39. Peters E, Dieckmann N, Dixon A, Hibbard JH, Mertz CK. Less 59. Edwards A, Unigwe S, Elwyn G, Hood K. Effects of commu-
nicating individual risks in screening programmes: Cochrane sys-
is more in presenting quality information to consumers. Med Care
tematic review. BMJ. 2003;327(7417):703–9.
Res Review. 2007;64(2):169–90.
40. Wittenbrink C, Pang A, Lodha S. Verity Visualizations: Visual 60. Tversky A, Kahneman D. Judgment under uncertainty: heur-
Mappings. Santa Cruz: University of California, Santa Cruz; 1995. istics and biases. Science. 1974;185:1124–31.
41. Djurcilov S, Kim K, Lermusiaux P, Pang A. Visualizing scalar 61. Parascandola M, Hawkins J, Danis M. Patient autonomy and
the challenge of clinical uncertainty. Kennedy Inst Ethics J. 2002;
volumetric data with uncertainty. Computers and Graphics. 2002;
26:239–48. 12(3):245–64.
42. Pang A, Wittenbrink C, Lodha S. Approaches to uncertainty 62. Simon HA. Bounded rationality and organizational learning.
visualization. Visual Computer. 1997;13(8):370–90. Organ Sci. 1991;2(1):125–35.
43. Potter K. Methods for presenting statistical information: the 63. Reyna VF. How people make decisions that involve risk. Curr
Dir Psychol Sci. 2004;13(2):60–66.
box plot. In: Hagan H, Kerren A, Dannemann P, eds. Visualization
of Large and Unstructured Data Sets, GI-Edition Lecture. Notes in 64. Reyna VF. Fuzzy-trace theory, judgment, and decision-
Informatics (LNI). 2006. making: a dual-processes approach. In: Izawa C, Ohta N, eds.
44. Altman DG. Confidence intervals for the number needed to Human Learning and Memory: Advances in Theory and Applica-
tions. Mahwah (NJ): Erlbaum; 2005. p 239–56.
treat. BMJ. 1998;317:1309–12.
65. Einhorn HJ, Hogarth RM. Decision making under ambiguity.
45. Ibrekk H, Morgan MJ. Graphical communication of uncertain
quantities to nontechnical people. Risk Anal. 1987;7(4):519–29. Journal of Business. 1986;59(4):S225–50.
46. Schapira MM, Nattinger AB, McHorney CA. Frequency or 66. Camerer C, Weber M. Recent developments in modeling prefer-
probability? A qualitative study of risk communication formats ences: uncertainty and ambiguity. J Risk Uncertain. 1992;5:325–70.
used in health care. Med Decis Making. 2001;21(6):459–67. 67. Curley SP, Yates JF. The center and range of the probability
interval as factors affecting ambiguity preferences. Organ Behav
47. British Publishing Group. BMJ Clinical Evidence Concise.
Issue 16. London: BMJ Publishing Group; 2003. Hum Decis Process. 1985;36:273–87.
48. Association for Healthcare Research and Quality. U.S. Pre- 68. Curley SP, Yates JF. An empirical evaluation of descriptive
ventive Services Task Force (USPSTF). 7 February 2007. Avail- models of ambiguity reactions in choice situations. J Math Psy-
able from: URL: http://www.ahrq.gov/clinic/uspstfix.htm chol. 1989;33:397–427.
49. Consumer Reports Best Buy Drugs. Treating migraine head- 69. Keren G, Gerritsen LEM. On the robustness and possible
aches: the triptans. Available from: URL: http://www.bestbuy- accounts of ambiguity aversion. Acta Psychologica. 1999;103:
drugs.org/drugreport_DR_triptans.shtml 149–72.
50. O’Connor AM, Stacey D, Entwistle V, et al. Decision aids for 70. Lauriola M, Levin IP. Relating individual differences in atti-
people facing health treatment or screening decisions. Cochrane tude toward ambiguity to risky choices. J Behav Decis Making.
Database Syst Rev 2003(1):CD001431. 2001;14(2):107–22.
51. Hibard JH, Peters E. Supporting informed consumer health 71. Einhorn HJ, Hogarth RM. Ambiguity and uncertainty in prob-
care decisions: data presentation approaches that facilitate the use abilistic inference. Psychol Rev. 1985;92(4):433–61.
of information in choice. Annu Rev Public Health. 2003;24:413–33. 72. Kahn BE, Sarin RK. Modeling ambiguity in decisions under
52. Mazur DJ, Merz JF. How older patients’ treatment preferences uncertainty. J Consum Res. 1988;15(2):265–72.
are influenced by disclosures about therapeutic uncertainty: sur- 73. Budescu DV, Weinberg S, Wallsten TS. Decisions based on
gery versus expectant management for localized prostate cancer. J numerically and verbally expressed uncertainties. J Exp Psychol
Am Geriatr Soc. 1996;44:934–7. Hum Percept Perform. 1988;14(2):281–94.
53. Miller SM, Exner TM, Williams SP, Ehrhardt AA. A gender- 74. Kuhn KM. Communicating uncertainty: framing effects on
specific intervention for at-risk women in the USA. Aids Care. response to vague probabilities. Organ Behav Hum Decis Process.
2000;12(5):603–12. 1997;71(1):55–83.
75. Viscusi WK. Alarmist decisions with divergent risk informa- of personal relevance on social judgments. J Pers Soc Psychol.
tion. Econ J. 1997;107:1657–70. 1988;55(3):357–71.
76. Johnson BB, Slovic P. Presenting uncertainty in health risk 94. Miller SM, Rodoletz M, Mangan CE, Schroeder CM, Sedlacek
assessment: initial studies of its effects on risk perception and TV. Applications of the monitoring process model to coping with
trust. Risk Anal. 1995;15(4):485–94. severe long-term medical threats. Health Psychol. 1996;15(3):
77. Ritov I, Baron J. Reluctance to vaccinate: omission bias and 216–25.
ambiguity. J Behav Decis Making. 1990;3:263–77. 95. Gerrity MS, Earp JL, DeVellis RF, Light DW. Uncertainty and
78. Meszaros JR, Asch DA, Baron J, Hershey JC, Kunreuther H, professional work: perceptions of physicians in clinical practice.
Schwartz-Buzaglo J. Cognitive processes and the decisions of Am J Sociol. 1992;97(4):1022–51.
some parents to forego pertussis vaccination for their children. J 96. Katz J. Why doctors don’t disclose uncertainty. Hastings Cent
Clin Epidemiol. 1996;49(6):697–703. Rep. 1984;14(1):35–44.
79. Han PK, Kobrin SC, Klein WM, Davis WW, Stefanek M, Taplin 97. Gordon GH, Joos SK, Byrne J. Physician expressions of uncer-
SH. Perceived ambiguity about screening mammography recom- tainty during patient encounters. Patient Educ Couns. 2000;40:
mendations: association with future mammography uptake and per- 50–65.
ceptions. Cancer Epidemiol Biomarkers Prev. 2007;16(3):458–66. 98. Allison JJ, Kiefe CI, Cook EF, Gerrity MS, Orav EJ, Center R.
80. Han PK, Moser RP, Klein WM. Perceived ambiguity about The association of physician attitudes about uncertainty and risk
cancer prevention recommendations: relationship to perceptions taking with resource use in a Medicare HMO. Medical Decis Mak-
of cancer preventability, risk, and worry. J Health Commun. 2006; ing. 1998;18(3):320–9.
11(suppl 1):51–69. 99. Johnson CG, Levenkron JC, Suchman AL, Manchester R.
81. Han PKJ, Moser RP, Klein WMP. Perceived ambiguity about Does physician uncertainty affect patient satisfaction? J Gen
cancer prevention recommendations: associations with cancer- Intern Med. 1988;3(2):144–9.
related perceptions and behaviours in a US population survey. 100. Bursztajn H, Feinbloom RI, Hamm RM, BrodskyA, eds. Medi-
Health Expect. Epub ahead of print. DOI:10.1111/j.1369-7625. cal Choices, Medical Chances. New York: Delacorte Press; 1981.
2007.00456.x 101. Henry MS. Uncertainty, responsibility, and the evolution of
82. Frosch DL, Kaplan RM, Felitti V. Evaluation of two methods the physician/patient relationship. J Med Ethics. 2006;32:321–3.
to facilitate shared decision making for men considering the
102. Hewson MG, Kindy PJ, Van Kirk J, Gennis VA, Day RP. Stra-
prostate-specific antigen test. J Gen Intern Med. 2001;16:391–8. tegies for managing uncertainty and complexity. J Gen Intern
83. Volk RJ, Cass AR, Spann SJ. A randomized controlled trial of Med. 1996;11(8):481–5.
shared decision making for prostate-specific antigen screening. 103. O’Connor AM, Llewellyn-Thomas H, Stacey D. IPDAS
Arch Intern Med. 1999;156:1333–6.
Collaboration Background Document. 30 April 2007. Available
84. Babrow AS, Kline KN. From ‘‘reducing’’ to ‘‘coping with’’ from: URL: http://www.informedhealthchoice.com/pdf/IPDAS_
uncertainty: reconceptualizing the central challenge in breast background_2005.pdf
self-exams. Soc Sci Med. 2000;51(12):1805–16. 104. Claxton K. The irrelevance of inference: a decision-making
85. Tao G, Branson BM, Kassler WJ, Cohen RA. Rates of receiv- approach to the stochastic evaluation of health care technologies.
ing HIV test results: data from the U.S. National Health Interview J Health Econ. 1999;18:341–64.
Survey for 1994 and 1995. J Acquir Immune Defic Syndr. 1999;22 105. Claxton K, Neumann PJ, Araki S, Weinstein MC. Bayesian
(4):395–400. value of information analysis. Int J Technol Assess Health Care.
86. Conley TD, Taylor SE, Kemeny ME, Cole SW, Visscher B. 2001;17(1):38–55.
Psychological sequelae of avoiding HIV serostatus information.
106. Vlek C. Risk assessment, risk perception and decision mak-
Basic Appl Soc Psychol. 1999;21(2):81–90. ing about courses of action involving genetic risk: an overview of
87. Lazarus RS, Folkman S. Stress, Appraisal, and Coping. New concepts and methods. Birth Defects Orig Artic Ser. 1987;23(2):
York: Springer; 1984. 171–207.
88. Brashers DE. Communication and uncertainty management. J 107. Fischhoff B, Lichtenstein S, Slovic P, Derby SL, Keeney RL.
Commun. 2001;51(3):477–97. Acceptable Risk. Cambridge (MA): Cambridge University Press;
89. Connolly T, Reb J. Regret in cancer-related decisions. Health 1981.
Psychol. 2005;24:S29–34. 108. Collins F, McKusick V. Implications of the Human Genome
90. Heath C, Tversky A. Preference and belief: ambiguity and Project for medical science. JAMA. 2001;285:540–4.
competence in choice under uncertainty. J Risk Uncertain. 1991;4 109. Fischhoff B, Bostrom A, Quadrel MJ. Risk perception and
(1):5–28. communication. Annu Rev Public Health. 1993;14:183–203.
91. Levinson W, Kao A, Kuby A, Thisted RA. Not all patients
want to participate in decision making: a national study of public
preferences. J Gen Intern Med. 2005;20(6):531–5.
92. Deber RB, Kraetschmer N, Irvine J. What role do patients
wish to play in treatment decision making? Arch Intern Med.
1996;156(13):1414–20.
93. Sorrentino RM, Bobocel DR, Gitta MZ, Olson JM. Uncertainty
orientation and persuasion: individual differences in the effects