Professional Documents
Culture Documents
Risk Preferences: Consequences For Test and Treatment Thresholds and Optimal Cutoffs
Risk Preferences: Consequences For Test and Treatment Thresholds and Optimal Cutoffs
Risk attitudes include risk aversion as well as higher-order test cutoff value. Risk vulnerability, which combines risk
risk preferences such as prudence and temperance. This aversion, prudence, and temperance, is relevant if there
article analyzes the effects of such preferences on medical is a comorbidity risk: thresholds and optimal cutoff values
test and treatment decisions, represented either by test decrease even more. Since common utility functions imply
and treatment thresholds or—when the test result is not risk vulnerability, our findings suggest that diagnostics in
given—by optimal cutoff values for diagnostic tests. For low prevalence settings (e.g., screening) may be consid-
a risk-averse decision maker, effective treatment is ered more beneficial when risk preferences are taken
a risk-reducing strategy since it prevents the low health into account. Key words: decision analysis; value of
outcome of forgoing treatment in the sick state. Compared information; provider decision making; decision rules;
with risk neutrality, risk aversion thus lowers both the test patient decision making. (Med Decis Making
and the treatment threshold and decreases the optimal 2014;34:33–41)
logarithmic and root functions) imply mixed risk Like Pauker and Kassirer,1,2 we take a normative
aversion.6,7 Only when preferences are risk vulnera- stance and derive test and treatment thresholds that
ble will a DM be more ‘‘risk averse’’ in an overall indicate the DM’s best choice depending on the a pri-
sense. We analyze how these risk preferences affect ori probability of illness. However, the model may
the thresholds in the a priori probability of illness also be used for a positive analysis and can possibly
that define the prevalence range within which testing explain why we observe testing in low probability
is indicated. Furthermore, we demonstrate how ranges, which we would not expect if the DM acted
higher-order risk preferences influence the decider’s in a risk-neutral manner.
choice of the positivity criterion, which is the optimal The remainder of the article is organized as fol-
cutoff value of a diagnostic test when the test charac- lows. We start the second section with a basic model
teristics are not exogenously given. that captures risk neutrality and risk aversion. We
The concept of risk aversion in decisions under then add a comorbidity risk, which allows us to
uncertainty is well established in economics. Gener- address prudence and temperance under the concept
ally speaking, risk aversion can be explained by the of risk vulnerability. The different risk preferences
decreasing marginal utility of income. A risk-neutral can be compared with the ratio of the utility gain
individual is indifferent between a lottery and its (cer- from treatment in the sick state and the utility loss
tain) expected value (i.e., the individual will accept from treatment in the healthy state, representing the
a fair bet). A risk-averse individual, however, requires tradeoff a DM faces when deciding for or against treat-
compensation (the risk premium) to accept the bet. In ment under uncertainty. The third section investigates
the medical setting, risk aversion refers to the decreas- the treatment decision without diagnostic testing and
ing marginal utility of health (i.e., a marginal improve- analyzes how it changes when moving from risk neu-
ment of health is valued more at low health levels than trality to risk vulnerability. The fourth section adds
at high levels). Risk-averse patients would thus be pre- diagnostic testing and presents the effects of risk pref-
pared to forego some of their health if they can thereby erences on the test and treatment thresholds. The fifth
reduce uncertainty over their health state. section discusses the analysis for endogenous test
However, risk preferences are only partially cap- results. The sixth section presents a clinical example,
tured by risk aversion. Higher-order risk attitudes and seventh section discusses the findings and limita-
such as prudence3 and temperance4 also appear to tions of the model and concludes.
be relevant in decision making under uncertainty.
Sometimes prudence and temperance are called
third- and fourth-order risk aversion, indicating their RISK AVERSION, PRUDENCE, TEMPERANCE,
complementarity to (second-order) risk aversion. Eeck- AND RISK VULNERABILITY
houdt and Schlesinger8 characterize prudence and
temperance preferences over particular classes of lot- Our focus here is on the diagnostic risk inherent in
tery pairs. Prudence is defined as a preference for disag- the decision on medical treatment. In our analysis,
gregating a sure loss and a zero-mean background risk. we do not differentiate between physicians and
This means that a DM is more willing to accept an extra patients as DMs but rather assume that physicians
risk when his or her level of health is higher. Temper- decide purely in the interest of their patients. The
ance implies a preference for disaggregating two (inde- patient’s health state i is unknown; he or she can be
pendent) zero-mean background risks. So, intuitively, sick or healthy, i ¼ s; h, where p describes the proba-
a DM facing one unavoidable background risk will bility of the sick state. The DM must decide whether
seek to reduce exposure to another background risk. to treat the patient (j ¼ þ) or not (j ¼ ). The DM can
In the medical context, the background risk can be also use diagnostics and make the treatment decision
a comorbidity risk, a disease or disorder that is pres- dependent on the test result (i.e., will treat if the test
ent in addition to the primary health states. We is positive and will not treat after a negative test
assume that this risk is exogenous (i.e., neither its result). Therefore, we use j, j ¼ þ; , for the test result
existence nor its extent can be influenced by the treat- as well as the treatment decision. We assume that the
ment decision). We develop a model with and with- DM knows the possible consequences of his or her
out considering the comorbidity risk. We will show decision for the patient’s health. In general, the con-
that a risk-averse DM will test and treat at lower a pri- sequences can be described by a vector of health
ori probabilities of illness than the risk-neutral DM. attributes with different levels. For simplicity, we
Taking comorbidity risk into account, the risk-vul- use a single interval-scaled variable H for the
nerable DM will test and treat even earlier. patient’s health state. It follows then that Hh
ORIGINAL ARTICLE 35
absolute prudence) exceeds absolute risk aversion. All For p\ p,~ the expected utility of no treatment exceeds
constant relative risk aversion (CRRA) utility functions the expected utility from treatment, so that not treat-
and all the common decreasing absolute risk aversion ing is indicated. For p > p, ~ by comparison, the opti-
(DARA)
pffiffiffiffiffi utility functions—for example, lnð H Þ and mal strategy is to treat.
H , which exhibit both DARA and CRRA—satisfy The therapeutic threshold will depend on the
the necessary conditions for risk vulnerability.5 DM’s risk preferences. We classified the different
Higher-order risk preferences become relevant if risk preferences with respect to the ratio of utility
a comorbidity risk m ~ exists ‘‘in the background’’ of gains and losses from treatment in the sick and
the health status. In this case, health status H itself healthy states. The effects of risk neutrality, risk aver-
turns into a random variable, H þ m. ~ The comorbidity sion, and risk vulnerability on the therapeutic thresh-
risk is assumed to be exogenous—its outcome is inde- old can be gleaned from (5) as p~ is decreasing in
pendent of the DM’s treatment decision. In the supple- ðG=LÞ. In other words, if either the utility gain from
mental appendix, we show that risk vulnerability treatment in the sick state increases or the loss from
implies EGV =ELV > GA =LA , where E stands for the treatment in the healthy state decreases, the thera-
expected value and V indicates a risk-vulnerable DM. peutic threshold decreases and treatment is indicated
A risk-vulnerable DM facing a background risk puts at lower a priori probabilities. This also makes sense
even more weight on improving his or her level of intuitively: if either the potential benefit from treat-
health in the sick state relative to health loss from treat- ment increases or the potential harm decreases, the
ment in the healthy state than a merely risk-averse DM. treatment option becomes more favorable.
This leads to the following sequence of inequalities: Now, the utility gain from treatment increases for
a risk-averse DM compared with a risk-neutral DM,
EGV =ELV > GA =LA > GN =LN : ð3Þ GA > GN , while the utility loss from treatment
decreases, LA \LN . This leads to GA =LA > GN =LN
When considering the utility gain from treatment in and thus to a lower therapeutic threshold for A than
the sick state and the utility loss from treatment in for N, p~A \ p~N . A uses the treatment strategy as an
the healthy state, the risk-neutral DM is indifferent insurance device (it reduces the spread between the
to marginal changes in the health levels in these 2 possible health states) and thus opts for treatment at
states. The risk-averse DM values a marginal increase lower prevalence rates than N.
in health more in the sick state than in the healthy In the situation with comorbidity risk, the third
state. When facing a background risk, the utility and fourth derivatives of the utility function appear
gain-loss ratio of treatment increases even more for to be relevant as well. Since EGV =ELV > GA =LA , it
a risk-vulnerable DM. follows that the therapeutic threshold is reduced
even further: p~V \ p~A . Although the background risk
is exogenous, the risk-vulnerable DM braces himself
TREATMENT DECISION WITHOUT A DIAGNOS- or herself in the face of it and treats earlier than the
TIC TEST: THE THERAPEUTIC THRESHOLD (merely) risk-averse DM.
Summing up, we conclude that (3) implies
To estimate the value of information provided by
a diagnostic test, one must first consider the DM’s deci- p~V \ p~A \ p~N : ð6Þ
sion regarding whether to treat in a situation in which
no test is available. In this base case without comorbid- The therapeutic threshold will be lower for a risk-
ity risk, expected utility as a function of j, with j ¼ þ for averse DM than for a risk-neutral one. Facing comor-
treatment and j ¼ for no treatment, becomes bidity risk leads to a further reduction of the thresh-
old for a risk-vulnerable DM.
EU j ð pÞ ¼ pU Hsj þ ð1 pÞU Hhj : ð4Þ
TREATMENT DECISION WITH A DIAGNOSTIC
There is an a priori probability of illness p~ at which TEST: TEST AND TREATMENT THRESHOLDS
the DM is indifferent between treatment and no treat-
ment. This treatment threshold, first derived by The availability of a diagnostic test increases the
Pauker and Kassirer,1 satisfies DM’s options. Instead of 2 alternative actions (treat-
L 1 ing and not treating), he or she now has the additional
p~ ¼ ¼ : ð5Þ
L þ G 1 þ G=L option of using diagnostics. The performance of a test
ORIGINAL ARTICLE 37
situation where the DM has to determine a cutoff value If the test result shows a value below the optimal cut-
that distinguishes positive from negative test results. A off value, the physician should not treat. By contrast,
good example is the prostate-specific antigen (PSA) a higher value should lead to treatment. A lower opti-
test for the detection of prostate cancer. The analysis of mal cutoff then implies earlier treatment. The intui-
a blood sample results in a PSA value that the physician tion for the rank order shown in (12) is the same as
declares as a positive or negative test result depending in the last two sections: risk aversion increases the
on his or her chosen cutoff value. Correspondingly, the utility gain in the sick state and decreases the utility
test characteristics sensitivity and specificity are no lon- loss in the healthy state, leading to earlier treatment.
ger given but determined by the chosen cutoff. Given a background risk, risk vulnerability strength-
The test technology can then be plotted on the ens this effect, causing a further decrease in the opti-
receiver operating characteristics (ROC) curve, which mal cutoff value.
represents all the feasible maximal sensitivity-
specificity pairs. In the simplest case where there is
only 1 marker, such as the PSA value, each possible CLINICAL EXAMPLE
cutoff value corresponds to 1 point on the ROC curve.
In general, a higher cutoff value for a test result increases For illustrative purposes, we take a clinical exam-
specificity at the expense of sensitivity. The DM’s task is ple from Sox and others.15 A physician examines
to choose the optimal cutoff value at which the patient’s a 55-year-old man with a headache and progressive
expected utility is maximized at a given a priori proba- unilateral weakness. The physician suspects a brain
bility of illness. The optimal cutoff value corresponds to tumor, which can be treated with brain surgery. For
a point along the ROC curve that satisfies the following simplicity, we consider life expectancy LE to be the
condition (see McNeil and others,12 Metz,13 and Felder only outcome parameter and ignore quality of life
and others14): (which was considered in the original example).
This does not lessen the significance of the analysis,
dSe L=G as we can assume decreasing marginal utility of
¼ : ð11Þ LE.16 If he is healthy, the patient can expect to live
dð1 SpÞ p=ð1 pÞ
another 21 years. Unnecessary brain surgery will
According to this equation, the slope of the ROC reduce his remaining LE to 20 years. If he has a brain
curve for the optimal cutoff value is equal to the prod- tumor and remains untreated, he can expect to live for
uct of the inverse of the a priori odds and L/G. With an only 2 more years, while brain surgery in this case
increase in the prevalence rate p or in the utility gain will increase his remaining LE to 11 years.
G, the slope of the ROC curve decreases. This implies First, we calculate the therapeutic threshold for
an upward move along the ROC curve, which in turn DMs with different risk preferences. The utility func-
increases sensitivity at the expense of specificity. tion of a risk-neutral DM is linear. To compare utility
Provided that the marker is positively correlated values for risk-neutral and risk-averse (risk vulnera-
with the presence of the illness, this implies a lower ble) DMs, we divide LE by 21. The utility gain
optimal cutoff value. This result is intuitive, since G and utility loss L from treatment can then be
the importance of detecting the sick increases in the expressed as GN ¼ ð11 2Þ=21 ¼ 0:429 and LN ¼
probability of illness or with a higher utility gain of ð21 20Þ=21 ¼ 0:048. Therefore, the therapeutic
treatment in the sick state. The opposite applies for threshold for a risk-neutral DM will be p~N ¼ LN =
an increase in the utility loss L, where a downward ðGN þ LN Þ ¼10:00%. Thus, in this situation, without
move along the ROC curve is optimal, leading to further diagnostics, brain surgery is the optimal strat-
lower sensitivity and increased specificity. egy for a risk-neutral DM, provided that the a priori
As the slope of the ROC curve is decreasing in probability of the patient having a brain tumor is at
ðG=LÞ, we can easily qualify the effect of risk preferen- least 10%.
ces on the optimal point along the ROC. Denoting the Sox and others15 report the following utilities for
cutoff value with x—which refers to the diagnostic life expectancies, derived by the standard gamble
marker, rather than to a point on the ROC curve—we method: U ð21Þ ¼ 1; U ð2Þ ¼ 0:25; U ð0Þ ¼ 0. We
then derive the following rank order for the optimal approximated these utility levels by the utility func-
(*) cutoffs as a function of risk preferences: tion U ðLEÞ ¼ LE0:6 6:212, which represents a mixed
risk-averse DM who is also risk vulnerable. The util-
xV \xA \xN : ð12Þ ities of the remaining life expectancies become
U ð21Þ ¼ 1:000, U ð20Þ ¼ 0:971, U ð11Þ ¼ 0:679, and DISCUSSION AND CONCLUSION
U ð2Þ ¼ 0:244 (note that U ð2Þ differs slightly from
0.25). The utility gain G and the utility loss L from In this article, we analyze the impact of risk prefer-
treatment change to GA ¼ 0:679 0:244 ¼ 0:435 and ences on decisions over medical testing and treat-
LA ¼ 1:000 0:971 ¼ 0:029, leading to a therapeutic ment. In particular, we investigate the effect of
threshold of 6.23% (p~A ¼ LA =ðGA þ LA Þ ¼ 0:029= a comorbidity risk in the presence of higher-order
ð0:435 þ 0:029Þ). The risk-averse DM would therefore risk attitudes. A risk-averse DM values the health
prescribe brain surgery at a much lower a priori prob- gain from treatment in the sick state more and the
ability. He or she values the utility gain to utility loss health loss from treatment in the healthy state less
ratio G=L more highly than the risk-neutral DM by than a risk-neutral DM. When facing an exogenous
a factor 1.67 (GA =LA ¼ 15:1; GN =LN ¼ 9:0). comorbidity risk, the utility gain-loss ratio for
Next, we assume that the patient’s health status a risk-vulnerable DM will increase even further.
depends not only on the presence or absence of the Risk attitudes affect the test and treatment thresh-
primary illness but also on the presence or absence olds. Both thresholds decrease with an increasing
of a comorbidity. We add a symmetric zero-mean utility gain-loss ratio of treatment, leading to both
comorbidity risk E½m ~ ¼ 0, where m ~ can take on the testing and treatment at lower a priori probabilities
values 1 2 and –2 with 50% probability each. Hence, of illness. A risk-averse DM will thus test and treat
the patient’s life expectancy in the healthy states at lower probabilities than a risk-neutral DM. Facing
(sick states) becomes 23 or 19 (4 or 0) with equal prob- a comorbidity risk, a risk-vulnerable DM will test and
ability if he is not treated and 22 or 18 (13 or 9) if he is treat even earlier.
treated. Corresponding results apply for endogenous test
For a risk-neutral DM, the additional comorbidity results. Since a risk-averse DM realizes a higher util-
risk changes nothing, as the expected utility gain G ity gain-loss ratio of treatment, he or she will lower
and expected loss L of treatment remain at 0.429 the test’s cutoff value to increase sensitivity at the
(GN ¼ 0:5 ð13 þ 9 4 0Þ=21) and at 0.048 expense of specificity. In the presence of a comorbid-
ðLN ¼ 0:5 ð23 þ 19 22 18Þ=21Þ. For the risk- ity risk, risk vulnerability strengthens this effect,
vulnerable DM,
EGV increases to 0.491 causing an even further decrease in the optimal
’ð0:5=6:212Þ 130:6 þ 90:6 40:6 00:6 and EGV cutoff.
increases to 0.02893 ½’ð0:5=6:212Þ 230:6 þ 190:6 The threshold analysis is particularly relevant in
220:6 180:6 Þ, which results in a higher utility gain the low-prevalence setting of screening programs. If
to utility loss ratio EGV =ELV ¼ 17:0 and thus in a fur- DMs are risk averse (and risk vulnerable), they will
ther decrease in the therapeutic threshold to realize a larger utility gain from treatment in the
p~V ¼ 5:57%. sick state and a smaller utility loss in the healthy
Finally, we take into account that the physician state. They will therefore test at lower a priori proba-
can use a computed tomography (CT) scan with a sen- bilities than if they were risk neutral. Risk aversion
sitivity of 95% and a specificity of 97% for further and higher-order risk preferences might thus explain
diagnostics. The CT produces false-positive and some of the observed excess testing (e.g., for breast
false-negative results, which must be considered cancer19 and prostate cancer20). Of course, prospect
before using the test. The likelihood ratios for the theory21,22 and defensive medicine23 are other possi-
CT scan amount to LRþ ¼ 31:67 and LR ¼ 0:05. ble explanations.
Using for the test and treatment thresholds and the The commonly used utility functions with CRRA
utility gain to utility loss ratios from treatment (e.g., logarithmic and root functions) imply mixed
reported above, we obtain the following values: risk aversion—that is, have derivatives with progres-
p~Dx
N ¼ 0:35%, p ~Dx
A ¼ 0:21%, p ~Dx
V ¼ 0:19% and p ~Rx
N ¼ sively alternating signs6,7 and feature risk vulnerabil-
Rx Rx
68:31%, p~A ¼ 56:30%, p~V ¼ 53:34%. We note that ity. Constant absolute risk aversion (CARA) utility
risk aversion in particular has a strong effect on the functions (i.e., exponential utility functions) also
therapeutic threshold as compared with risk neutral- exhibit mixed risk aversion but do not show risk vul-
ity. Moreover, recent empirical studies show that nerability, implying that a DM with a CARA utility
higher-order risk preferences can have a sizable function does not react to the zero-mean comorbidity
impact on decisions in general (see Ebert and Wie- risk. However, although exponential utility functions
sen17) and on medical treatment decisions in particu- are often more convenient analytically than
lar (see Krieger and Mayrhofer18). CRRA functions, they are usually considered less
ORIGINAL ARTICLE 39
plausible.24 Interestingly, assuming that the comor- to predict individual behavior, especially with regard
bidity risk applies only to the sick state, the only to possible probability weighting. Alternatives to
assumption necessary for EG=EL > G=L to hold is expected utility theory, such as rank-dependent
that the DM be prudent—in this case, even DMs choice models, have been shown to be more in line
with CARA utility functions will appear to be more with actual behavior.30,31 Nonetheless, recent stud-
risk averse.25 ies, such as the one by List,32 challenge the validity
We model health as a continuous variable, of rank-dependent theory. They claim that experi-
although it clearly has multidimensional aspects. enced individuals behave largely in accordance
The theory of multivariate risk aversion, which takes with expected utility theory. Since physicians rou-
into account that utility is a function of many varia- tinely decide on medical tests and treatments, they
bles (e.g., Duncan26), could be applied to the health should be able to make unbiased estimations of a pri-
setting. However, since there are thousands of possi- ori probabilities of illnesses and thus make coherent
ble combinations of health variables, it is feasible to treatment decisions. Moreover, this study’s nature
think of overall H as being continuous. is normative, and ‘‘expected utility is the best theory
Recent experimental evidence confirms risk-averse to determine which decisions to take.’’33(p697)
and prudent behavior among individuals facing uncer- The second limitation is the restriction to diagnos-
tainty.17,27,28 Empirical results for temperance are tic risk. In general, DMs face not only diagnostic risk
ambiguous so far. While Deck and Schlesinger27 find but also therapeutic risk, which refers to uncertainty
(weakly) intemperate behavior, studies by Ebert and over the treatment outcome. Eeckhoudt34 shows that
Wiesen17 and Noussair and others29 point in the other risk aversion increases the critical probability of suc-
direction, in line with our theoretical assumptions. cessful treatment in this case. The intuition here is
Ebert and Wiesen17 also measure the intensity of that treatment is a risk-increasing action, so a risk-
higher-order risk preferences and find that the com- averse DM treats later than a risk-neutral one. There-
pensation subjects are willing to pay to avoid third- fore, the effects of higher-order risk preferences on the
order risk is significantly higher than for second- thresholds may cancel each other out when diagnostic
order risk, emphasizing the importance of measuring and treatment risk are analyzed simultaneously.
higher-order risk preferences in general. Bleichrodt and others35 analyze the effect of
Krieger and Mayrhofer18 study whether individu- comorbidities on the treatment decision in the con-
als in an experimental laboratory display risk-averse text of therapeutic risk. They mirror the QALY model
and prudent behavior in medically framed treatment and assume that treatment only affects quality of life,
decisions, where DMs’ payoffs are monetary. Their that comorbidity only has an impact on life duration,
theoretical approach is similar to the one presented and that patients can influence treatment intensity.
in this article. Their experiment provides evidence They find that comorbidities do not affect the treat-
for risk-averse and prudent behavior. Such risk pref- ment decision, since the prudence premium3 can be
erences have a considerable impact on the therapeu- interpreted as the DM’s risk premium for a longer
tic threshold, leading to its reduction by 41% relative life. However, in contrast to Eeckhoudt’s34 and our
to the risk-neutral position. Risk aversion accounts models, the background risk in Bleichrodt and others
for three-fourths of this effect and prudent behavior is imposed on a second dimension (i.e., life duration)
for one-fourth. Furthermore, they find that the medi- rather than on the patient’s health level.
cal framing of the decision as compared with a neutral The third limitation concerns our rather abstract
framing does not affect second-order risk preferences clinical example, which should be regarded as
but is associated with more frequent and stronger merely illustrative. Further empirical and experi-
prudent behavior. Since—assuming that the comor- mental research is needed to produce reliable infor-
bidity risk applies only to the sick state—the impact mation on higher-order risk preferences regarding
of higher-order risk preferences on test and treatment medical decision making.
thresholds as well as on optimal cutoffs is analogous Pure health outcomes such as mortality or survival
to the therapeutic threshold, the results of Krieger and rates are often used instead of utility measures. This
Mayrhofer illustrate the potential importance of mea- may lead to unreasonably high test and treatment
suring risk preferences, particularly risk aversion and thresholds and cutoff values. The use of QALYs
prudence, for medical treatment recommendations. instead would reduce this bias, as QALYs reflect
This article has at least 3 limitations. First, our patient preferences over health statuses when they
model is rooted in the expected utility framework, are calculated by methods grounded in the expected
which has been criticized for its insufficient capacity utility theory and using the standard gamble method,
which reflects higher-order risk preferences if pres- 14. Felder S, Werblow A, Robra BP. A-priori risk and optimal test
ent. However, often only the primary illness is taken accuracy in prenatal diagnostics. Med Decis Making. 2003;23(5):
406–13.
into account when eliciting QALYs but not the possi-
bility of comorbid conditions. 15. Sox HS, Blatt MA, Higgins MC, Marton KI. Medical Decision
Making. Philadelphia, PA: American College of Physicians; 2007.
We conclude that medical decisions should be Reprint, Boston: Butterworth-Heinemann; 1988.
based on the accurate measurement of preferences
16. Pratt JW, Zeckhauser RJ. Willingness to pay and the distribu-
in an extended standard gamble approach. Our tion of risk and wealth. J Polit Econ. 1996;104(4):747–63.
results are relevant for clinical guidelines. In particu- 17. Ebert S, Wiesen D. Joint measurement of risk aversion, pru-
lar, our findings suggest that diagnostic tests in low- dence and temperance. J Risk Uncertain. Forthcoming 2013.
prevalence settings (e.g., screening) may appear 18. Krieger M, Mayrhofer T. Patient Preferences and Treatment
more beneficial if (higher-order) risk preferences are Thresholds under Diagnostic Risk: An Economic Laboratory
taken into account. Experiment. Ruhr Economic Papers No. 321. Essen, Germany:
RWI; 2012.
19. Breen N, Gentleman JF, Schiller JS. Update on mammography
ACKNOWLEDGMENTS
trends: comparisons of rates in 2000, 2005, and 2008. Cancer. 2011;
117(10):2209–18.
We are grateful to the associate editor and 2 anonymous ref- 20. Farwell WR, Linder JA, Jha AK. Trends in prostate-specific
erees for most helpful comments regarding not only spe- antigen testing from 1995 through 2004. Arch Intern Med. 2007;
cific points but also the organization and presentation of 167(22):2497–502.
the material. We also thank Miriam Krieger and the partic- 21. Kahneman D, Tversky A. Prospect theory: an analysis of deci-
ipants of the 13th Biennial European Conference of the sion under risk. Econometrica. 1979;47(2):263–91.
Society for Medical Decision Making in Hall in Tirol, Aus- 22. Tversky A, Kahneman D. Advances in prospect theory: cumu-
tria (2010), for their remarks. lative representation of uncertainty. J Risk Uncertainty. 1992;5(4):
297–323.
REFERENCES 23. DeKay ML, Asch DA. Is the defensive use of diagnostic tests
good for patients, or bad? Med Decis Making. 1998;18(1):19–28.
1. Pauker SG, Kassirer JP. Therapeutic decision making: a cost 24. Friend I, Blume ME. The demand for risky assets. Am Econ
benefit analysis. N Engl J Med. 1975;293(5):229–34. Rev. 1975;65(5):900–922.
2. Pauker SG, Kassirer JP. The threshold approach to clinical deci- 25. Felder S, Mayrhofer T. Medical Decision Making—A Health
sion making. N Engl J Med. 1980;302(20):1109–17. Economic Primer. Berlin, Germany: Springer; 2011.
3. Kimball MS. Precautionary saving in the small and in the large. 26. Duncan GT. A matrix measure of local multivariate risk aver-
Econometrica. 1990;58(1):53–73. sion. Econometrica. 1977;45(4):895–903.
4. Kimball MS. Precautionary motives for holding assets. In: New- 27. Deck C, Schlesinger H. Exploring higher order risk effects. Rev
man P, Milgate M, Falwell J, eds. New Palgrave Dictionary of Econ Stud. 2010;77(4):1403–20.
Money and Finance. London: Macmillan; 1992. p. 158–61. 28. Ebert S, Wiesen D. Testing for prudence and skewness seeking.
5. Gollier C, Pratt JW. Risk Vulnerability and the tempering effect Manage Sci. 2010;57(7):1334–49.
of background risk. Econometrica. 1996;64(5):1109–23. 29. Noussair CN, Trautmann ST, van de Kuilen G. Higher order
6. Brocket PL, Golden LL. A class of utility functions containing risk preferences, demographics, and financial decisions. 2011.
all the common utility functions. Manage Sci. 1987;33(8):955–64. Available from: http://ideas.repec.org/p/dgr/kubcen/2011055.
html
7. Caballé J, Pomansky A. Mixed risk aversion. J Econ Theory.
1996;71:485–513. 30. Abdellaoui M. Parameter-free elicitation of utility and proba-
bility weighting functions. Manage Sci. 2000;46(11):1497–512.
8. Eeckhoudt L, Schlesinger H. Putting risk in its proper place. Am
Econ Rev. 2006;96:280–9. 31. Bleichrodt H, Pinto JL. A parameter-free elicitation of the prob-
ability weighting function in medical decision analysis. Manage
9. Arrow KJ. Essays in the Theory of Risk-bearing. London: North-
Sci. 2000;46(11):1485–96.
Holland; 1971.
32. List J. Neoclassical theory versus prospect theory: evidence
10. Pratt J. Risk Aversion in the small and in the large. Econometr-
from the market place. Econometrica. 2004;72(2):615–25.
ica. 1964;32:122–36.
33. Wakker PP. Lessons Learned by (from?) an economist working
11. Eeckhoudt L, Gollier C, Schlesinger H. Changes in background
in medical decision making. Med Decis Making. 2008;28(5):690–8.
risk and risk taking behavior. Econometrica. 1996;64(3):683–9.
34. Eeckhoudt L. Risk and Medical Decision Making. Boston:
12. McNeil BJ, Keeler M, Adelstein SM. Primer on certain ele-
Kluwer; 2002.
ments of medical decision making. N Engl J Med. 1975;293(5):
211–5. 35. Bleichrodt H, Crainich D, Eeckhoudt L. The effect of comorbid-
ities on treatment decisions. J Health Econ. 2003;22:805–20.
13. Metz CE. Basic principles of ROC analysis. Semin Nucl Med.
1978;8(4):283–98.
ORIGINAL ARTICLE 41