In an ideal world, we would be able to offer tests that picked up cancer before it reached the more serious stages, allowing early intervention and a much greater prospect of cure. Such a process is called screening and is now available for a number of cancers: breast, uterine, cervix, and bowel. In addition, the PSA blood test is a potential screening test for prostate cancer, but its use remains controversial. It is helpful to describe the characteristics of an ideal screening test and then examine how these tests shape up in practice. This is illustrated in the following example:
Table 1 – features of an ideal screening test (Source: WHO)
- The target disease should be a common form of cancer, with a high associated death rate
- Effective treatment, capable of reducing the risk of death if applied early enough, should be available
- Test procedures should be acceptable, safe, and relatively inexpensive In addition, we need to consider:
- True positive rates: Sick people correctly diagnosed as sick
- False positive rates: Healthy people wrongly identified as sick
- True negative rates: Healthy people correctly identified as healthy
- False negative rates: Sick people wrongly identified as healthy.
Thus the sensitivity (Patients with liver disease and abnormal scans/all patients with a positive scan) = 231/231+31 = 0.88 and the specificity (Patients with normal scans and no disease/all patients with normal scans) = 54/ (27+54) = 0.67
A further measure is the positive predictive value (the proportion of patients with abnormal scans who have liver disease) = 231/ (231+27) = 0.89 and the negative predictive value of a negative scan (the proportion of patients with normal scans who have no liver disease) = 54/ (32+86) = 0.45
For a test in the clinic, this is pretty good – a positive scan in someone suspected of having a liver disease is a pretty good indicator that the person has the disease. How does this fare as a screening test, then?
To illustrate the difference between using a test for diagnosis in someone already known to be ill and screening for disease in people with no symptoms, we can look at the figures for breast cancer. Let us suppose that the rate of missed cases (the specificity) in those who we test is 10% and that the level of early, undetected disease is 1 in 500 people. If we now test 100,000 subjects, an ideal test would yield 200 positive tests in the cancer sufferers and 999,800 negative tests in those without the disease. However, our test, though good, is not perfect and will only detect 180 of the 200 cases, leaving 20 people wrongly reassured. Conversely, the test is also not completely specific. Let us say that 95% of those without the disease will test negative but 5% will wrongly test positive. When we apply this to our screening population, we see that this means that 5% of the 99,800 without the disease will falsely test positive. This works out as 4,999 false positive tests in people without the disease. This means that only a minority (180/4,999 = 4%) of those with a positive test actually have the disease, but 4,999 – 180 = 4,819 people have had a nasty scare. Furthermore, 20 have been falsely reassured and will go on to present with cancer anyway, possibly detected late as they may ignore the symptoms, believing they to be cancer-free. However, the overwhelming majority of those with a negative test (99,800 – more than 99%) really were free of the disease, so a negative test is pretty reassuring.
These worked examples are important, as they illustrate the limitations of screening tests which at first sight sound pretty good. In point of fact, the figures above are the best figures available – sensitivity and specificity fall in younger women (probably because their breast tissue is denser, making it harder to see abnormal lumps), leading to more incorrectly categorized cases. Furthermore, while the cost of the test itself is small, the cost of chasing up the false positives is much larger and needs to be factored into the costs of the screening programme.
There is a further problem when working out the benefit of screening. In our example above, we will identify cases of cancer earlier than would have happened without screening, potentially improving treatment prospects. However, with breast cancer, the cure rates are good, with three-quarters of diagnosed women being long-term survivors. This leaves the quarter that is destined to do badly, who are the main potential beneficiaries of screening. This is a relatively small number in relation to the numbers of tests carried out, and the downside is over-investigation of healthy women without breast cancer.