3503 Imaging Science Lecture 5
January 23 th
Recall Case 1: Low p, High U(TP)
• We have a rare disease (low prevalence)
• We assigned a high value to the utility of true
positives which is meant that we were
running the analysis using the point of view of
people who would favour maintaining the old
screening mammography guidelines
• We compared two different imaging systems and got two different expected
utilities. One was excellent detection accuracy and one was merely good
detection accuracy. We had observed that if you go from the merely good to the
excellent system, it would improve the maximum utility that you could get from
screening mammography. The exact improvement was from a maximum utility of
3 (merely good) to a maximum utility of 4 (excellent); a factor of 4/3.
If we take the point of view from the other side of the debate: point of view of people
who are concerned about overdiagnosis. So we must make adjustments of the utilities;
make the utility of true positive smaller and utility of false negatives less negative.
• Overdiagnosis would reduce U(TP)
o U(TP) must be weighted average (based on what we think the fractions of
each section is) of:
Benefit to patients whose lives are saved (last at least 5 years past
the date of diagnosis)
Benefit to patients whose lives are prolonged
Benefit to patients who do not respond to treatment
Harm to over-treated patients
o Point of view that there are a lot more over-treated patients that would pull
down the overall utility of true positives in the expected utility analysis.
Patient’s age may be important; the benefit to society to saving
someone who is 40 is greater than saving someone who is 75, and
there may be oncology factors that depend on age
• Younger patients in saved subgroup will have very high
U(TP) due to individual’s large gain of QALY
• Proportion of patients not responding to treatment may
depend on age
• Evidence for Overdiagnosis o Graph tells the cases of new cancer per 10,000 women per year. Red
curve is new cases of breast cancer discovered that were advanced
cancers (metastatic), and green curve is new cases of premalignant and
early stage cancers.
o The early 1980’s correspond to the point at which universal screening
mammography started to be pushed aggressively.
o You would think that by detecting more cancer at an early stage, this
would reduce the detection of cancers at an advanced stage.
What we see is that after screening mammography became widely
used the number of premalignant and early staged cancers
detected went up, but the rate of metastatic cancers didn’t change.
A lot of the patients on this green curve would have never made it
in to the red curve. The increase in green that was not balanced by
a decrease in red indicates that overdiagnosis is not just
theoretically possible but is more common than actual diagnosis.
o If we assume that overdiagnosis is really common then, then we want a
really dramatic reduction in U(TP) and a really dramatic change in utility of
Case 2: Low p, Low U(TP) • In comparison to case 1, we still have a rare disease but now we have low utility
of true positive indicating potentially that we are concerned about overdiagnosis.
• Excellent imaging system on
the left and merely good on
• This reduction in utility of true
positive has reduced the
maximum utility of excellent
system from 4 to 1.1 and
reduced maximum utility of
merely good system from 3 to
o Makes sense; if utility
of true positive is going
down, then maximum
utility will go down too