Chapter 17 Answers to discussion questions

17.1 The horizontal axis on a response curve diagram represents an ordinal variable. As a result, the curves are not as smooth as those depicted in figure 17.3. Can you think of an alternatively means of representing the horizontal axis that would produce smoother curves?

The horizontal axis in a response curve diagram represents the level of certainty of a cytologist as they form a diagnostic decision. For clarity, four levels of decision certainty have been shown in figure 17.3: definitely normal, probably normal, not sure, probably abnormal and definitely normal. Since this variable has a discontinuous scale and has intrinsic order it should be considered to be an ordinal variable. A more realistic (although visually less pleasing) representation of figure 17.3 would therefore be a frequency histogram rather than a curve with a continuous distribution. The obvious way to produce smooth curves in a response curve diagram would be to replace the ordinal variable with a continuous scale variable, such as percentage certainty.

 

17.2 Discuss the effect that each of the biases described in box 17.1 will have on the position of the criterion line in figure 17.4.

Types of bias	Effect on criterion
Anchoring	Failure to adjust criterion when new information becomes available
Availability effect	Premature adjustment of criterion before all relevant information is considered
Inflexibility bias	Same as anchoring
Expectancy effect	Criterion is inappropriately adjusted to match expectations
Base-rate neglect	Failure to adjust criterion to match the expected base rate (prevalence) of disease
Feedback sanction	Unable to make use of feedback to make appropriate adjustments to criterion
Hindsight bias	Inappropriate upward or downward shift of criterion to match the known outcome
Overconfidence bias	Criterion shifts to an extreme position
Satisfaction of search	Same as availability effect
Bias blind spot	Failure to adjust criterion to compensate for one’s own biases

 

17.3 Discuss the relationship between criterion and the traditional measures of diagnostic accuracy (sensitivity and specificity).

To appreciate the effect of a criterion shift on diagnostic sensitivity and specificity you will need to refresh your memory of the formulae for these two measures of diagnostic accuracy (consult chapter 16). Now take a good look at the areas under the curves in figure 17.3 representing the number of true positives (the area under the abnormal response curve to the right of criterion) and true negatives (the area under the normal response curve to the left of criterion). Shifting criterion to the left increases the number of true positives as a proportion of the total number of abnormals, thereby increasing diagnostic sensitivity. Likewise, there will be a decrease in the number of true negatives as a proportion of the total number of normals, resulting in a decrease in specificity. On the other hand, a right criterion shift decreases the proportion of abnormals that are report as abnormal (thus decreasing sensitivity) and increases the proportion of normals that are reported as abnormal, which increases specificity. If you are feeling brave you might like to consider how such criterion shifts affect the positive predictive value and negative predictive value!

 

17.4 Biases in cytological decision making are generally seen as detrimental to performance. Can you think of any situations in which a certain amount of reporting bias is beneficial in patient care?

There are indeed situations in which a certain amount of reporting bias is clinically beneficial. In diagnostic cytology, which focuses on symptomatic patients – the base rate (prevalence) of neoplastic conditions in patients with the relevant symptoms and clinical history is relatively high and such patients are under close clinical scrutiny. The possibility that treatment may be instigated on receipt of a positive cytology result requires the minimization of false positive errors and maintenance of high levels of test specificity, even though this entails some loss of sensitivity. An upward criterion shift is therefore considered appropriate in such scenarios. Cytology may also be used as a screening test for detecting relatively low prevalence neoplastic conditions (e.g. cervical intraepithelial neoplasia). A general aim of screening is to achieve high levels of test sensitivity, even though this may be at the expense of low specificity, and this requires cytologists to set a relatively low criterion.  It is not unusual for cytologists to divide their working day between diagnostic and screening cytology, so adjustments of criterion are commonplace.

 

17.5 Using the terminology that has been adopted in this chapter, identify some of the key mental skills and performance characteristics of a cytology expert.

• Good visual search and cell discrimination skills
• Good decision making skills
• High discriminability index
• High sensitivity and specificity
• Low susceptibility to unwanted bias
• Appropriate adjustment of criterion for diagnostic and screening scenarios
• Good attentional control
• High capacity visual memory