Accuracy ofrisk assessments in practice
A perfectrisk marker would have a sensitivity of 100% and a specificity of 100%, implying noerrors in risk assessment. Consequently, the false-positive and falsenegative rates wouldbe 0%, and positive and negative predictive values would be 100%.Having perfect accuracy means that the predicted high-risk group would consist ofonly true high-risk individuals and that only true low-risk individuals would be includedin the predicted low-risk group. Unfortunately, no such marker is availablefor the assessment of caries risk. A certain proportion of errors have to be accepted.However, there are no generally accepted rules of what the acceptable level of errormight be.
It has beensuggested that, in a risk model, the sum of sensitivity and specificity be at least 160%before a caries risk marker can be considered a legitimate candidate for targetingindividualized prevention (Kingman, 1990). This is in agreement with an alternativesuggestion that a sensitivity and specificity of 80% would be acceptable forpractical use in the community. Although neither of these suggestions takesinto account thefact that errors related to poor sensitivity do have consequences that are differentfrom those related to poor specificity, both proposals can be used as a startingpoint for evaluating the performance of proposed markers for high caries risk.
What woulda combined sensitivity and specificity of 160% mean in practice? If both the sensitivityand specificity were 80%, every fifth individual with a true high risk wouldremain undetected in a risk assessment and thus fail to receive the intensified protectionagainst caries that he or she needs. Correspondingly, every fifth individual with a truelow risk would erroneously be included in the high-risk group and receive preventivemeasures to no or little purpose. Thus, even the proposed minimum acceptablelevel of accuracy would result in an uninvitingly high rate of
misclassifications.
If theproportion of caries-risk individuals in a population is close to half or more,this clearlyimplies that the occurrence of caries is not low enough to justify the effortand expense ofidentifying key-risk individuals. In such a situation, the preventive efforts should betargeted to the whole population.
Theproportion of the target population that can be given individual protectionagainst furthercaries development naturally varies from one setting to another. In most cases, risk groupsof a size exceeding 30% seem to be unworkable. In a thorough review by Hausen etal (1994), an effort was made to compare the predictive power of risk markers ina situation where the aim was to select the 30% of the target population with the highestrisk of developing new lesions. For none of the markers aimed at assessingthe risk for coronal caries did the predictive power reach the proposed combinedsensitivity and specificity of 160% (Kingman, 1990). This level was surpassedin one study only, where a combination of several predictors had been used forassessing the risk of root caries (Scheinin et al, 1992).
Thedifficulty of predicting caries is not unexpected. The multifactorialetiological
andmodifying factors of dental caries make it likely that even the mostsophisticated
risk modelswill be of limited value in predicting future caries development very
accurately.Furthermore, even a perfect marker is capable of predicting a person ‘s
futurecaries experience only if the conditions on which the prediction is basedremain
stable. Inmost industrialized countries, where virtually all the prediction studies have
beenconducted, the populations are exposed to a variety of professional prevention
andtreatment regimens as well as self-care, which, if applied selectively, most
probablyreduce the observed power of such studies. The living conditions and oral
healthbehaviors may change over time, thus modifying a person’s caries risk in either
direction.In addition, the rational and ethical consequence of risk prediction in
clinicalpractice is to introduce needs-related measures for caries prevention and
cariescontrol. The optimal outcome therefore should be no new carious lesions. For
thesereasons, it is not likely that, even in the future, caries risk can be assessed
accuratelyby using one single risk marker.
Past cariesexperience (caries prevalence¾the number of decayed, missing, or filled
teeth andsurfaces¾and caries incidence¾the number of new carious teeth and
surfaces ina year) has so far been the most powerful single predictor for future caries
incidence,at least in children and young adults. That is because carious lesions
representthe sum result of all the etiologic and modifying risk factors to which the
individualhas been exposed.
Forexample, in a recent 3-year longitudinal study, Bjarnason and Kohler (1997)
achieved89% sensitivity value in a group of Swedish adolescents by comparing the
prevalenceof non-cavitated enamel caries and DFS at the baseline as predictors.
Together,the sensitivity and specificity values reached 160% or more. High salivary
MS andlactobacilli scores resulted in 71% sensitivity and 75% specificity,
respectively(cutoff level for high caries risk was 5 or more new carious surfaces in 3
years).However, only baseline values of incipient enamel caries were significantly
correlatedto the caries incidence.
The use ofpast caries experience as an indicator of future incidence has justly been
criticizedby the argument that the aim should be to determine the high-risk
individualsbefore there are any signs of past caries experience. In other words, efforts
shouldfocus on primary prevention instead of secondary prevention. In particular this
isimportant in infants and children with erupting permanent teeth. Wendt et al(1994)
found thatthe caries incidence in infants and toddlers aged 1 to 3 years was strongly
correlatedto the plaque scores and oral hygiene regimens even at 1 year of age.
