One of the most intensely debated topics in the dental literature today is the contribution of oral conditions, such as periodontal disease, to the risk of developing nonoral diseases. Statistical associations between oral and non-oral diseases have been demonstrated primarily from observational epidemiologic studies using measures such as odds ratios (ORs) and relative risks (RRs).^1–3 ORs are the odds of an outcome’s occurring in one group divided by the odds of an outcome’s occurring in another group; RRs are the probability of an outcome’s occurring in one group divided by the probability of the outcome’s occurring in another group.

Some of the authors of these studies conclude that statistically significant associations between oral diseases and nonoral diseases imply that oral disease is a clinically useful predictor of nonoral conditions in individual patients. But can predictive or prognostic information applicable to individual patients be derived from these types of studies? This is a major concern that has significant consequences for overall health and well-being, as well as the future role of oral health care providers as screeners for nonoral diseases.

The key question is this: when will a statistical association between the marker (oral disease) and the outcome (nonoral disease) imply that the marker is useful in predicting nonoral disease outcomes in the individual patient? Phrased another way, when can the presence of oral disease predict—or discriminate between—patients who will develop and those who will not develop nonoral disease?

Two important statistical indexes are used to determine the accuracy and utility (or express the strength) of a risk factor as a predictor for a disease outcome: sensitivity and specificity. Sensitivity is the probability of getting a positive test result from a patient who has the disease. For instance, when the sensitivity is 95 percent, 95 percent of patients with the disease will have a true-positive result and 5 percent of patients with the disease will have a false-negative result. Specificity is the probability of getting a negative test result from a patient who does not have the disease. For instance, when the specificity is 90 percent, 90 percent of patients without the disease will have a true-negative result and 10 percent of patients without the disease will have a false-positive result.

Let us apply these indexes to the risks of developing pancreatic cancer or coronary heart disease (CHD) as examples. If we screen for pancreatic cancer, a highly deadly cancer, the test preferably should have high sensitivity—that is, few false-negative results. CHD is associated with expensive invasive and noninvasive work-ups and long-term therapy. To predict such an outcome—CHD—it would be advantageous to have a test with high specificity—that is, few false-positive results. Before using a screening or prognostic test, we need to determine the appropriate sensitivity and specificity for this particular test, or our optimal threshold for accepting the test as useful for our particular purpose. As a rule, if we increase the sensitivity of a test to identify more cases (true positive), the ability to identify true-negative cases (specificity) is diminished.

The association between oral and nonoral diseases typically has been calculated using ORs and RRs. Are ORs and RRs good measures of the prognostic value of a test? If a screening test for a marker of interest has a sensitivity of 90 percent and an associated OR of 3, the specificity would be only 25 percent. In other words, 75 percent of the disease-free patients would be labeled incorrectly as having the disease (false positive) (OR = [sensitivity/1-sensitivity] x [specificity/1– specificity]). If a test has a specificity of 60 percent and an associated OR of 3, the sensitivity would be 66 percent, which would mean that 34 percent of the disease-free cases would not be identified. Thus, because the same OR occurs at various levels of sensitivity and specificity, the OR by itself is not sufficient to characterize the predictive value of a diagnostic test for an individual patient.

Even a diagnostic test with a large OR may not be a good predictive test. A test with a sensitivity of 90 percent and an associated OR of 36 still would fail to predict 20 percent of all true-negative cases. With the same associated OR of 36, a test with a sensitivity of 60 percent would have a specificity of 96 percent, thus missing 4 percent of all true-negative cases.

RR can quantify a risk factor in one group of people as compared with another group. However, RR still cannot be used to predict a person’s risk of developing disease. It only can imply that the risk of developing disease is a specific amount (for instance, three times) more likely in one group than in another group.

In summary, ORs and RR values reported in available epidemiologic studies on the association between oral and nonoral diseases and conditions are not sufficient to distinguish people who will develop nonoral maladies from those who will not.

To find the optimal threshold for a test for diagnostic or screening purposes, it would be helpful to plot sensitivity and specificity over a range of values of interest.⁴ (For more in-depth information, see van Schalkwyk.⁵) This type of analysis has been used to evaluate measurement of bone density of the jaws as a diagnostic test for osteoporosis, but it has yet to be used to assess the predictive value of other oral conditions in ascertaining the presence of nonoral diseases or conditions.

Constructing risk models and identifying markers to predict the likelihood of developing disease is a worthwhile undertaking and should be pursued. It is an endeavor that should engage and challenge us in the oral health care field, as more and more data become available on the association between oral and nonoral diseases. We should pursue this, not only out of academic interest, but also to get answers to questions that eventually can lead to improved diagnostic performance and therapeutic guidance.

A good predictive marker or risk factor should have incremental prognostic value, indicating severity or quantity denoting different levels of risk. It also should generate additional power to existing risk stratifications. Here lies a challenge if we truly want to identify patients at risk of developing CHD. A patient’s socioeconomic status (SES) is a significant risk factor for developing coronary artery disease,⁶ yet we do not routinely refer patients with low SES for a cardiovascular work-up.

At present, there is some evidence that periodontal disease may be more prevalent in specific groups of people with cardiovascular disease or in women with adverse pregnancy outcomes. However, the presence of periodontal disease in a specific person has not shown to add any additional discriminatory power to determine if this particular person’s risk of developing cardiovascular disease or having adverse pregnancy outcomes is increased.

Editors of professional journals have a responsibility to ensure that the conclusion of a published study is supported by the presented data. Professional organizations and associations have a responsibility to provide their constituencies with legitimate interpretations of scientific findings based on valid interpretation of available data. And all health care professionals have a responsibility to incorporate into their practice important new findings that will benefit their patients. If the presence of oral diseases truly predicts the development of nonoral diseases, we have an obligation to ascertain this association and act on it.

Trying to predict a patient’s propensity for developing a nonoral disease is a risky business. As custodians of the oral cavity, oral health care professionals need to be able to interpret the biomedical literature on this subject correctly. We owe that to our patients.