better health by designLearn More

A test has most value when the prevalence is high

4647883696_c564082a24_zA perennial problem in primary care is whether to schedule laboratory tests. This is an issue that is often very perplexing early in clinical careers because there is often a lot of pressure or perceived pressure not to miss the rarest of rare conditions. The secret, if there is one, is to know the prevalence of the condition before testing for it. Supposing a condition has a prevalence of 20% (1 in 5) in a population, this would be considered very high. What if the recommended test detects 95% of cases (sensitivity) and identifies 80% of people who did not have the condition (specificity). If the test were applied to 200 people:

  • 70 would be identified as positive and of these 38 would be ‘true positives’. Therefore 32 people would be told they might be sick when they were not (positive predictive value = 54.29%).
  • On the other hand 130 would be told they were not sick when two of them were ‘false negatives’.

If the same test were applied to the same population but the prevalence was 2% (1 in 50), in other words pathology is unlikely, then:

  • 43 people would be identified as abnormal and of these 39 people would NOT have the condition (positive predictive value = 9.3%).
  • On the other hand 157 would be reported well, which was everyone with a negative test.

The impact of a test is not only on those who are found to have pathology but also on those who are told they might have pathology and need more, often invasive tests, for a very unlikely condition. Prostate cancer testing has been studied in this respect.

In primary care tests for life limiting illness in particular can be calibrated with a high sensitivity (more true positives), so that pathology is not missed and a modest specificity (more false positives) which means that more people might be subjected to further investigations because their symptoms need explanation and they may need more investigations. However a false positive result has a significant impact on the patient’s life. The numbers above illustrate the impact of prevalence on the proportion of people without pathology who would be subject to further tests. The higher the prevalence, the more worthwhile the test and the better the positive predictive value of a test. You can play with these numbers using this on line calculator.

A question to ask whenever requesting a test is how common is this condition in people like the person to be tested?  If it isn’t very common what harm could be caused by multiple tests to ‘prove’ this person doesn’t have this condition? In the business of healthcare no patient, client or customer should be subjected to tests without the practitioner having a firm grasp on how the test will help to manage the case. By corollary there is no short cut to taking a detailed history and examining the patient in order to make a diagnosis. Tests can never compensate for poor practice, nor should they be used to try to impress a patient that ‘everything is being done’, often what is being done is iatrogenic harm.

Picture by National Library of Medicine.


  1. Jayne ingham says:

    I agree with rational use of tests. I always ask the patient particularly when they want an Xray/ Scan/ MRI is if this test will make any difference to the way we treat the condition.

  2. Most patients or ill people don’t have the opportunity to get tested or undergo medical reviews. Specially in men, prostate cancer is more like 22 out of 100,000 per year in terms of death. So, with this alone, raising an awareness campaign seems like unnecessary with males being busy at work. But I agree.. Test is valued with the prevalence is high.


  1. […] because of an injudicious use of tests in search of the needle in the haystack. But that’s another […]

  2. […] a previous post I explained why  tests can be harmful with respect to the limited positive predictive value of […]

Speak Your Mind