data in Figure 13-4,

, which is 0.81. This means that among the women who were

not pregnant, the home test was negative only 81 percent of the time, and 11 percent of women who

were truly negative tested as positive. (You can see why it is important to do studies like this before

promoting the use of a particular screening test!)

But imagine you work in a lab that processes the results of screening tests, and you do not usually have

access to the gold standard results. You may ask the question, “How likely is a particular screening

test result to be correct, regardless of whether it is positive or negative?” When asking this about

positive test results, you are asking about positive predictive value (PPV), and when asking about

negative test results, you are asking about negative predictive value (NPV). These are covered in the

following sections.

Sensitivity and specificity are important characteristics of the test itself. Observe that the

answers depend on the prevalence of the condition in the background population. If the study

population were older women, then the prevalence of being pregnant would be lower, and that

would impact the sensitivity and specificity. The prevalence will also impact the PPV and NPV,

which we discuss in the next section. For these reasons, it is important to use natural sampling in

such a study design.

Positive predictive value and negative predictive value

The positive predictive value (PPV, also called predictive value positive) is the fraction of all

positive test results that are true positives. In the case of the pregnancy test scenario, the PPV would be

the fraction of the time a positive screening test result means that the woman is truly pregnant. PPV is

the likelihood that a positive test result is correct. You calculate PPV as

. For the data in Figure

13-4, the PPV is

, which is 0.73. So, if the pregnancy test result is positive, there’s a 73 percent

chance that the woman is truly pregnant.

The negative predictive value (NPV, also called predictive value negative) is the fraction of all

negative test results that are true negatives. In the case of the pregnancy test scenario, the NPV is the

fraction of the time a negative screening test results means the woman is truly not pregnant. NPV is the

likelihood a negative test result is correct. You calculate NPV as

. For the data in Figure 13-4,

the NPV is

, which is 0.93. So, if the pregnancy test result is negative, there’s a 93 percent

chance that the woman is truly not pregnant.

Investigating treatments

In conditions where there are no known treatments, one of the simplest ways to investigate a new

treatment (such as a drug or surgical procedure) is to compare it to a placebo or sham condition using

a clinical trial study design. Because many forms of dementia have no known treatment, it would be

ethical to compare new treatments for dementia to placebo or a sham treatment in a clinical trial. In

those cases, patients with the condition under study would be randomized (randomly assigned) to an

active group (taking the real treatment) and a control group (that would receive the sham treatment), as

randomization is a required feature of clinical trials. Because some of the participants in the control

group may appear to improve, it is important that participants are blinded as to their group assignment,

so that you can tell if outcomes are actually improved in the treatment compared to the control group.