PROCEEDINGS
Clinical
M.F. Johnson
PharmaNet, Inc., 504 Carnegie Center, Princeton, NJ
08540; mjohnson@pharmanet.com
J Dent Res 83(Spec Iss C):C116-C118, 2004
ABSTRACT
In seeking new and more effective therapies to delay or
prevent caries development, investigators must design
clinical trials focused on high-risk populations with a
predictable incidence of caries over a limited period of
time. In children and adolescents, the strongest
predictors of caries incidence appear to be baseline
levels of caries activity (present caries, e.g., dmfs,
DMFT, caries lesions in first molars). Other predictors
of caries risk typically include oral hygiene level, counts
of cariogenic micro-organisms in plaque and saliva,
fluoride history, sucrose intake, and parent's socio-
economic level. This paper will briefly review existing
literature to address the most useful and relevant
prognostic factors for predicting future caries onset. The
relative merits of identifying high-risk subjects based on
these factors, either singly or in combination, will be
explored in terms of statistical efficiency. Particular
attention will focus on the advantages of covariate
adjustment in the context of survival-based methods for
the analysis of caries data. Further, with the advent of
more sophisticated diagnostic procedures (e.g.,
quantitative light fluorescence) to screen and monitor
study subjects for caries activity, there is the potential
for earlier states of lesion initiation and progression (or
regression) to be detected, with, therefore, improved
experimental sensitivity to treatment effects. The
validity of risk assessment and outcome measurement
on the basis of these new diagnostic tools vs. more
conventional methods will be examined.
KEY WORDS: risk factors, caries clinical trials.
Presented at the International Consensus Workshop on
Caries Clinical Trials, Glasgow, Scotland, January 7-10,
2002
C116
The Role of Risk Factors in the
Identification of Appropriate
Subjects for Caries Clinical Trials:
Design Considerations
INTRODUCTION
T he paper summarizes the oral presentation made to the International Consensus
Workshop on Caries Clinical Trials.
The term 'risk factor' will be used here to refer to any aspect or baseline
characteristic of the study population that affects the likelihood of observing the
clinical event of interest, in this case, dental caries. Baseline characteristics in
general, and risk factors in particular, play a major role in identifying appropriate
subjects for caries clinical trials. From a statistical standpoint, proper adjustment for
risk factors (baseline variables that have prognostic value) will improve the
efficiency of significance tests and also improve the validity (reduce bias) of
estimated treatment effects. But the topic has a much broader meaning when we
consider the modern environment, the low prevalence of dental caries in the
industrial world, and the difficulty of setting up studies in what some may consider a
'rare disease' setting.
This paper will begin with a brief review of caries studies published over the last
few decades to identify, from an historical perspective, population characteristics that
were considered important potential risk factors, i.e., variables that changed the
probability of a caries event. Types of analysis will also be discussed that take
advantage of risk factors as blocking factors to improve the efficiency of statistical
tests. Attention will focus on the relative merits of identifying 'high-risk' subjects for
inclusion in clinical trials and the impact this will have on study designs in the future.
In this context, some of the newer diagnostic methods will be considered for use in
screening and monitoring study subjects for caries activity. These methods have the
potential to detect earlier states of lesion initiation and progression, and this paper
will examine the validity of assessing risk and measuring outcome on the basis of
these methods vs. more conventional methods. These concepts will lead to some
general recommendations about study design vis--vis the selection of high-risk
subjects.
CHALLENGES IN THE DESIGN OF CARIES TRIALS
The nature of clinical trials in any therapeutic area must evolve to keep up with
changes in diagnostic tools, advances in treatment options, and variations in disease
prevalence or etiology. Studies of dental caries are no exception. We have seen this
in many other therapeutic areas, particularly studies in cardiovascular and infectious
diseases, where a vast array of marketed products detracts from our ability to recruit
a representative cross-section of patients, or where we are facing a different etiology
of disease, highly resistant to existing therapies.
Studies in dental caries face a similar challenge. The design of clinical trials to
find new products for the treatment and prevention of dental caries must adapt to the
changing environment. Traditional methods used to diagnose caries lesions during
the high-prevalence caries era will no longer be the sole basis for caries detection.
With the increased presence of fluoride and improved oral hygiene (especially in
developed countries), caries lesions now progress much more slowly and cavitation
occurs much later compared with the early high-prevalence phase of anti-caries
product development. Likewise, we face trouble with the placebo-controlled design,
now that the standard of care includes the use of a medicated dentifrice. In view of
these changes, we may need to re-define what we mean by 'high-risk' subjects and the
manner in which we identify them for clinical studies.
J Dent Res 83(Spec Iss C) 2004 Risk Factors in Caries Clinical Trials
THE ROLE OF RISK FACTORS
Risk factors play an important role in the design and interpretation of
clinical trials. Again, from a statistical perspective, it is important to
account for the influence of baseline characteristics and disease
severity in the analysis of clinical data, either through least-squares
covariate adjustment or through stratification. Depending on the
correlation between baseline factors and the response (typically
measured as either caries increments, incidence densities to account for
time at risk, or time to caries progression, however defined),
appropriate adjustment for covariate effects will typically improve the
efficiency of statistical tests and produce more valid (i.e., unbiased)
estimates of treatment efficacy.
In addition, the baseline data allow for tests of treatment by covariate
interaction and thus address the consistency of treatment effects across
relevant population subgroups. In the presence of interaction, we can
examine treatment effects within relevant baseline strata to determine if
efficacy differs qualitatively or quantitatively for specific high- or low-
risk subgroups. Risk factors also have value for assessing the impact of
loss to follow-up on the analysis of caries increments and will often
suggest the need for more sophisticated longitudinal data analysis to
account properly for variation in the at-risk period (Beck et al., 1997). For
instance, it has been shown that, in the elderly (more so than in children),
subjects who fail to return for final dental examinations are materially
different from more compliant subjects who return and have complete
follow-up information, in terms of both overall health condition and oral
health status. Demographic information, baseline caries status, and
medical history are all critical data for examination of the impact of
subject attrition on estimates of treatment outcome. Finally, these sorts of
baseline measurements can be explored as prognostic factors. If valuable
as predictors of caries onset, the measurements could well serve as
selection criteria in future caries trials.
BENEFITS OF STRATIFICATION
As mentioned, blocking or stratification on important risk factors will
typically improve the efficiency of statistical tests. Relative efficiency
may be estimated by the ratio of variance estimators, i.e.,
RE = 100 x sest2/s2
where the numerator (sest2) is the pooled variance of the response
variable across treatments estimated without stratification or blocking
factors included in the model, and the denominator (s2) is the pooled
variance estimated with blocking factors included in the model. In a
randomized blocks design, the higher the correlation between
responses within blocks, the greater the gain in efficiency. As Fleiss
(1986) has noted, RE should be at least 125-130% to make the effort of
blocking or stratification worthwhile.
Several population characteristics and study design features have
influenced clinical outcome in studies of fluoride dentifrices over the
past few decades (Stookey et al., 1993). Table 1 provides a list of these
factors. Many of these are known risk factors that have served as
effective stratification variables in previous trials, most notably age.
Caries studies typically focus on children, ranging in age anywhere from
6 to 16 yrs, but age-related factors (such as product usage habits, ability
to follow instructions, caries present in erupted teeth, oral hygiene, etc.)
are the underlying risks affecting trial results. Likewise, gender
differences and socio-economic level have a bearing on caries risk (boys
typically have higher rates than girls, as do children in public vs. private
school). Other variables affecting trial outcome are subject exclusion
criteria (e.g., confounding medical problems or orthodontic appliances),
drop-out rates (often a source of bias when attrition is high and differs
between groups), and, most importantly, the caries prevalence/incidence
in the population. This last point is critical, because experimental
sensitivity depends on the level of measurable disease. Latin America is
a good site for caries study today, since the caries prevalence there has
reached a level similar to that in the US in the mid-1960s. On the other
C117
hand, US caries prevalence has dropped over three-fold since that time,
just as it has in other Westernized countries. Other design factors that
affect study outcome include: length of study, fluoride exposure,
brushing compliance, pre-stratification criteria (age, gender, and baseline
DMFS scores are essential for balanced designs), examination methods,
and statistical power/sample size considerations.
NEW DEVELOPMENTS IN RISK ASSESSMENT
Recent dental research in the area of risk assessment is focusing on the
evaluation of new, technologically advanced methods for the diagnosis
of caries, perhaps in the hope that their improved sensitivity and
specificity (compared with those of conventional methods) will
increase the chance of detecting small treatment effects or discriminate
between competing treatment modalities, even in low-risk populations.
These methods may also help to identify caries development sooner,
long before caries is clinically evident. Ultimately, we want to develop
treatments that will delay or suppress caries development, allowing
newer diagnostic methods to provide an earlier signal or marker of
impending caries development. The use of survival or time-to-event
methodologies is clearly relevant in this area, including the work of
Hannigan et al. (2000), exploiting the use of the log-logistic model for
clustered survival data, and that of Hujoel and co-workers (1994),
applying the Poisson regression model to caries incidence. Cox
proportional hazards regression analysis also makes sense to explore
caries risk factors, as recently used to determine the relationship
between salivary mutans streptococci (MS) counts and caries incidence
in Japanese preschoolers (Ansai et al., 2000). Each of these models
allows for the inclusion of subject- and surface-specific explanatory
variables to identify risk factors that affect caries development. Risk
factors identified through the use of survival methods can guide the
selection of appropriate high-risk subpopulations for future studyin
this case, the subset of tooth surfaces and subjects likely to benefit
from preventive therapies.
For example, a recent three-year study in older adults (age 60+)
used the Poisson regression model to show that the risk of coronal
caries was increased in subjects with high baseline root DMFS, high
counts of mutans streptococci and lactobacilli, male gender, and Asian
ethnicity (Powell et al., 1998). Relative risks ranged from 1.2 to 2 for
these factors, indicating only moderate effects on incidence. But
knowledge of these relationships will strengthen the analysis of
treatment effects or even play a role in subject selection for caries
trials. Similar factors were associated with an increased risk of root-
surface caries (baseline coronal DMFS, high bacterial counts, and
Asian ethnicity). Again, the study confirms the value of baseline
DMFS and salivary variables, along with ethnicity, as a useful design
aspect for study of dental caries in the elderly. As in other therapeutic
areas, we will have to change the notion that caries prevention needs to
be based on large population-based cohort trials in favor of more
focused trials in targeted sub-populations of high-risk subjects.
As for alternative diagnostic methods, they will be useful only if
they are truly valid measures of caries risk and offer a reliable means
to select high-risk subjects and/or identify caries earlier than traditional
methods. Diagnostic techniques that have been explored for their
ability to improve the quality of caries prediction, in addition to visual
inspection, include radiography, fiber optic transillumination (FOTI),
Table 1. Factors Affecting Outcome of Caries Trials
Population Factors Design Factors/Variables
Age Length of study
Gender Fluoride exposure
Socio-economic level Brushing instructions
Exclusion criteria Stratification
Drop-out rates Examination techniques
Caries prevalence/incidence Statistical methods
C118 Johnson
Table 2. Validity of Caries Diagnostic Tests: Correlation between Test
Outcome and Lesion Depth in Studies Published Since 1992
(Verdonschot et al., 1999)
Diagnostic Method Surface R*
Visual inspection Occlusal 0.57, 0.76, 0.90
Radiography Occlusal 0.54, 0.77
Approximal 0.70
ERM Occlusal 0.62, 0.82
Quantitative FOTI Approximal 0.87, 0.92
QLF Buccal/lingual 0.78, 0.86
* Spearman rank correlation coefficient.
electrical resistance measurements (ERM), and quantitative laser/light-
induced fluorescence (QLF). The question remains, Will the use of
these techniques to monitor caries progression alter the way we select
patients for study or somehow change the risk profile of patients
identified for caries studies? The literature is not clear on this.
Table 2 summarizes the range of published correlations between
each diagnostic test outcome and actual lesion depth, as validated by
histology or measures of mineral loss in studies published since 1992
(Verdonschot et al., 1999). For approximal surfaces, FOTI had the
highest correlation with lesion depth (r values of 0.87 and 0.92). For
occlusal surfaces, visual inspection had a variable range of correlations
with histology, depending on the scoring system used, whereas ERM
performed quite well, with the correlation of 0.82. Diagnosis of
smooth-surface caries was best for QLF (r = 0.78-0.86).
In general, the new quantitative methods (FOTI, ERM, QLF)
show high correlation with lesion depth and therefore would be quite
suitable for monitoring small changes in lesions over time. Of course,
their use in clinical trials would not only affect the outcome measures
(e.g., time to progression instead of DMFT increments) but could also
affect the risk assessment models. That is, the strongest predictors of
caries incidence, such as present caries activity (typically measured as
baseline dmfs, DMFT, caries lesions in first molars), may have little
bearing on the risk of caries progression defined by the more high-tech
procedures. Weak factors for caries prediction (e.g., cariogenic micro-
organisms in plaque and saliva, saliva flow, and plaque tests) may
perform better as predictors of more sensitive outcome measures. For
example, the signs of caries progression determined through the use of
the quantitative laser-induced fluorescence (QLF) procedure may have
baseline 'risk' indicators somewhat different from those predicting
increments in clinically assessed DMFS as the outcome for analysis.
So, as in other therapeutic areas, the challenge will be to validate these
methods as potential 'surrogate markers' of caries development. This
will involve further exploration of their sensitivity and specificity, as
well as investigation into risk factors that may influence their accuracy
and predictive value for caries onset.
J Dent Res 83(Spec Iss C) 2004
RECOMMENDATIONS
Studies evaluating products that delay or prevent caries development
will be most sensitive to treatment effects if they are conducted in
populations that are 'enriched' with high-risk subjects. Using
conventional methods to quantify caries increments, such studies
should be conducted in areas or countries with high caries incidence
(e.g., Latin America, Central/Eastern European countries, Japan). This
would typically include regions with limited fluoride exposure and
poor oral hygiene, or in special populations of subjects selected for
their pre-disposition to caries development. Consideration should be
given to the use of new diagnostic methods in the screening,
stratification, and monitoring process, provided that these methods are
sufficiently validated, i.e., that they have high sensitivity and
specificity for caries detection. Future studies are likely to require
baseline DMFT or DMFS as covariates or stratification variables. But,
in addition, the newer, more sophisticated diagnostic test results may
reveal additional criteria for selecting and categorizing subjects at high
risk for caries progression. The definition of such 'risk thresholds'
requires validation against known outcomes, and this process of risk
assessment will take further research. Finally, to target the high-risk
groups for future study and prevention programs, we should continue
to explore the use of statistical models that allow us to forecast risk-
groups with high caries activity. This will include use of more
powerful multivariate regression models to adjust for prognostic
factors while evaluating changes in caries status over time.
REFERENCES
Ansai T, Tahara A, Ikeda M, Katoh Y, Miyazaki H, Takehara T (2000).
Caries risk in Japanese pre-school children. Pediatric Dent 22:377-380.
Beck JD, Lawrence HP, Koch GG (1997). Analytic approaches to
longitudinal caries data in adults. Community Dent Oral Epidemiol
25:42-51.
Fleiss JL (1986). The design and analysis of clinical experiments. New
York: J. Wiley & Sons.
Hannigan A, O'Mullane DM, Barry D, Schafer F, Roberts AJ (2000). A
caries susceptibility classification of tooth surfaces by survival time.
Caries Res 34:103-108.
Hujoel PP, Isokangas PJ, Tiekso J, Davis S, Lamont RJ, DeRouen TA, et
al. (1994). A re-analysis of caries rates in a preventive trial using
Poisson regression models. J Dent Res 73:573-579.
Powell LV, Leroux BG, Persson RE, Kiyak HA (1998). Caries risk in the
elderly. Community Dent Oral Epidemiol 26:170-176.
Stookey GK, DePaola PF, Featherstone JDB, Fejerskov O, Moller IJ,
Rotberg S, et al. (1993). A critical review of the relative anticaries
efficacy of sodium fluoride and sodium monofluorophosphate
dentifrices. Caries Res 27:337-360.
Verdonschot EH, Angmar-Mnsson B, ten Bosch JJ, Deery CH, Huysmans
MCD, Pitts NB, et al. (1999). Developments in caries diagnosis and
their relationship to treatment decisions and quality of care. Caries Res
33:32-40.