Assessing the Value of Diagnostic Tests

We comprehensively analyze testing methods to determine their benefits and risks. Only by including all of these values can we evaluate the true diagnostic value of a test, to ensure that you have access to accurate information about its overall usefulness and proper role in diagnosis and treatment.

Some tests ultimately have no positive effects for the general population. Others, such as the PSA test, can be useful if their results are understood within a Bayesian framework, but are harmful if interpreted in the standard manner. Some tests, usually ones involving radiation or surgery, are themselves dangerous.  

There is a difference between a test’s sensitivity (how frequently it correctly identifies actual positives) and its specificity (how frequently it correctly identifies actual negatives), making it difficult to judge efficacy in a meaningful way1. In practice, tests make a tradeoff between sensitivity and specificity, and neither of these features alone fully capture the value of a test. An informed patient needs to understand this distinction.

The probability that a patient with a positive test result has a disease is different from the probability that a patient with a disease will have a positive test result. A test may accurately detect 95% of cases of a condition, but a patient with a positive result may not have a 95% chance of the condition - depending on its prevalence their actual chance could be 1% or lower. Extensive research has revealed that even the majority of medical professionals fail to appreciate the distinction2. In one experiment, only 15% of doctors were able to calculate these probabilities correctly. Accurate interpretation of test results requires taking into account both how likely it is overall that someone will receive a positive test regardless of their disease status, and how likely it is that someone has a disease regardless of their test result. The relationship between these probabilities is expressed as 

Bayes’ theorem:

P(disease|positive result) = ( P(positive result|disease) * P(disease) ) / P(positive result)

Only by including all of these values can we evaluate the true diagnostic value of a test, to ensure that you and your patients have access to accurate information.

1. Pearson TA, Manolio TA (March 2008). "How to interpret a genome-wide association study". JAMA 299 (11): 1335–44.
2. Ge D et al. (September 2009). "Genetic variation in IL28B predicts hepatitis C treatment-induced viral clearance". Nature 461 (7262): 399–401.