Screening for alcohol problems: what makes a test effective?

Alcohol Research & Health, Wntr, 2004 by Scott H. Stewart, Gerard J. Connors

Like positive predictive value, negative predictive value depends on the validity of the screening test and the prevalence of the disorder. However, in this case the relationship is inverse: the higher the prevalence of the disorder in the population, the lower the negative predictive value. This means that a negative screening result is less helpful in ruling out the disease if the prevalence of the disease in the population is high. Continuing with the AUDIT example from Volk and colleagues, the postscreen probability for at-risk drinking among patients screening negatively with a cutoff of 4 was 2 percent, given 10 percent prevalence in the population. At a prevalence of 20 percent, the postscreen probability for a negative result was 4 percent. Analogous to the positive predictive values, negative predictive values illustrate that a negative screening result does not necessarily rule out a disorder. The extent to which a negative screening result indicates that a person has a decreased risk of actually having the disorder under investigation depends on the prevalence of the disorder in the population and the test's validity.

Limitations of Predictive Values

Positive and negative predictive values are useful when assessing postscreen probabilities for disorders with a known prevalence in the screened population. In these situations, predictive values provide average postscreen probabilities for all members of the screened population with a particular test result. For example, based on a known prevalence for current alcohol dependence in the screened population of 6 percent, a positive screen for dependence may then increase the probability that a person is alcohol dependent from 6 percent to 25 percent. Both the prescreen probability of 6 percent and the postscreen probability of 25 percent, however, represent an average risk for members of the population. Predictive values do not consider a person's additional risk factors, such as a family history of alcohol dependence, that may modify both the prescreen and postscreen risk for dependence in that person. A method for incorporating individual risk factors in clinical settings is based on likelihood ratios, which are discussed in the next section.

LIKELIHOOD RATIOS

In clinical settings, the physician often has additional information on a patient relevant to that patient's risk for drinking problems. The use of likelihood ratios allows the clinician to incorporate a specific patient's prescreen risk for a drinking problem into estimating postscreen probabilities. A likelihood ratio is the ratio of two probabilities--the probability of a given test result among people with the disease divided by the probability of that test result among people without the disease. For example, a likelihood ratio for at-risk drinking would be the probability that an at-risk drinker has a certain test result on the AUDIT divided by the probability that a nonrisk drinker has that result on the AUDIT. Depending on whether one assesses patients with positive test results or negative test results, the resulting likelihood ratios are known as positive likelihood ratio and negative likelihood ratio. The following sections discuss the clinical use of likelihood ratios because they are frequently presented as characteristics of screening tests. In actual practice, however, the results of screening tests applied to individual patients who are not already clinically suspected of having a drinking problem are interpreted dichotomously (i.e., positive or negative). A positive result will lead to additional diagnostic evaluation, and a negative result will preclude further evaluation.