555365

research-article2014
JDRXXX10.1177/0022034514555365Journal of Dental ResearchCAMBRA,Treatment Effect Mechanisms, and Caries Increment

Research Reports: Clinical

Journal of Dental Research
2015, Vol. 94(1) 44­–51
Understanding Treatment Effect © International & American Associations
for Dental Research 2014

Mechanisms of the CAMBRA Randomized Reprints and permissions:

sagepub.com/journalsPermissions.nav

Trial in Reducing Caries Increment

DOI: 10.1177/0022034514555365
jdr.sagepub.com

J. Cheng1, B.W. Chaffee1, N.F. Cheng1, S.A. Gansky1,

and J.D.B. Featherstone1

Abstract
The Caries Management By Risk Assessment (CAMBRA) randomized controlled trial showed that an intervention featuring
combined antibacterial and fluoride therapy significantly reduced bacterial load and suggested reduced caries increment
in adults with 1 to 7 baseline cavitated teeth. While trial results speak to the overall effectiveness of an intervention,
insight can be gained from understanding the mechanism by which an intervention acts on putative intermediate variables
(mediators) to affect outcomes. This study conducted mediation analyses on 109 participants who completed the trial to
understand whether the intervention reduced caries increment through its action on potential mediators (oral bacterial
load, fluoride levels, and overall caries risk based on the composite of bacterial challenge and salivary fluoride) between the
intervention and dental outcomes. The primary outcome was the increment from baseline in decayed, missing, and filled
permanent surfaces (ΔDMFS) 24 mo after completing restorations for baseline cavitated lesions. Analyses adjusted for
baseline overall risk, bacterial challenge, and fluoride values under a potential outcome framework using generalized linear
models. Overall, the CAMBRA intervention was suggestive in reducing the 24-mo DMFS increment (reduction in ΔDMFS:
–0.96; 95% confidence interval [CI]: –2.01 to 0.08; P = 0.07); the intervention significantly reduced the 12-mo overall risk
(reduction in overall risk: –19%; 95% CI, –7 to –41%;], P = 0.005). Individual mediators, salivary log10 mutans streptococci,
log10 lactobacilli, and fluoride level, did not represent statistically significant pathways alone through which the intervention
effect was transmitted. However, 36% of the intervention effect on 24-mo DMFS increment was through a mediation effect
on 12-mo overall risk (P = 0.03). These findings suggest a greater intervention effect carried through the combined action
on multiple aspects of the caries process rather than through any single factor. In addition, a substantial portion of the
total effect of the CAMBRA intervention may have operated through unanticipated or unmeasured pathways not included
among the potential mediators studied.

Keywords: caries management, caries risk, lactobacillus, mediation analysis, Streptococcus mutans, salivary fluoride

Introduction theoretical basis that motivates a particular intervention and

creates opportunities to test competing theories about
In many dental studies, an intervention is designed to mechanisms of action (MacKinnon and Luecen 2011).
change postrandomization (intermediate) variables (e.g., Enhanced understanding of intervention mechanisms and
knowledge, behavior, or biological factors) under the the particular mediating variables involved would allow
hypothesis that changing those intermediate variables will greater opportunities to tailor evidence-based practices to
lead to improvement in final dental outcomes (MacKinnon specifically targeted populations of interest.
and Luecen 2011). In such studies, researchers are inter- The standard analysis, comparing the dental outcomes
ested in not only whether the intervention works to improve between treatment and control groups, provides the overall
final outcomes but also how it works (i.e., mechanisms). treatment effect estimate but does not provide information
That is, interest lies in whether, and how much, the inter-
vention affects the outcome through either (1) measured 1
University of California San Francisco, San Francisco, CA, USA
intermediate variables, as designed, or (2) direct interven-
Corresponding Author:
tion effects on the outcome involving unmeasured or J. Cheng, University of California San Francisco, 3333 California Street,
unknown intermediate variables. Analysis of mediation Suite 495, San Francisco, CA 94143-1361, USA.
pathways obligates researchers to be explicit about the Email: jing.cheng@ucsf.edu

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

CAMBRA, Treatment Effect Mechanisms, and Caries Increment 45

Table 1.  Participant Inclusion and Exclusion Criteria in the CAMBRA Trial.

Inclusion criteria Ability to speak and understand English; 18 y of age or older; planning to stay in the immediate area for
the next 3 y; a minimum of 16 teeth; willing to have needed dental radiographs every 2 y; at least 1 and
up to 7 cavitated carious teeth; no moderate or severe periodontal disease needing periodontal surgery
or chemotherapeutic agents; voluntarily provided written informed consent
Exclusion criteria Significant past or current medical history of conditions that may affect oral health or oral flora (i.e.,
diabetes, HIV, heart conditions that require antibiotic prophylaxis); use of medications that may affect
the oral flora or salivary flow (e.g., antibiotic use in the past 3 mo, drugs associated with dry mouth/
xerostomia); complex dental need; frequent periodontal maintenance; another household member
participating in the study; drug or alcohol addiction, or other conditions that may decrease the
likelihood of adhering to the study protocol

on the mechanism or pathway associated with the treatment Dental Clinics from 1999 to 2001 (see Table 1 for inclusion
effect. Even if a treatment does not have a significant effect, and exclusion criteria); 115 were randomized to control and
a positive mediation effect through intermediate variables 116 to the intervention group. Included for analysis in this
could possibly cancel a direct negative effect through other article were those 109 participants (47% of enrolled partici-
pathways, leading to no overall effect. pants; 52 in the control group and 57 in the intervention
The Caries Management by Risk Assessment group) who completed the CAMBRA trial, with final clini-
(CAMBRA) randomized controlled clinical trial (R01 DE cal examinations completed from 2001 to 2004. Table 2
012455; Featherstone et al. 2012) was conducted at the shows the baseline characteristics for all participants and
School of Dentistry, University of California, San Francisco, participants who completed the trial.
with institutional review board approval. Based on the
hypothesis that caries arises when pathological factors (e.g.,
oral bacteria, such as mutans streptococci [MS] and lacto- Treatment Groups
bacilli [LB]) outweigh protective factors (e.g., salivary flu- Both the intervention and control groups received initial
oride, saliva flow; ten Cate and Featherstone 1991; caries removal and restorations following the baseline
Featherstone 2003), the study aimed to assess whether com- examination; salivary assays for assessment of MS, LB, and
bined antibacterial and fluoride therapy based on risk fluoride at baseline, when initial restorations were com-
assessment has beneficial effects on preventing new caries plete, and every 6 mo thereafter; and radiographs and dental
over 24-mo follow-up in adults with 1 to 7 baseline cavi- examinations at the beginning and end of the study. In addi-
tated teeth. The intervention sought to reduce the pathologi- tion, after initial caries removal and restoration, participants
cal factors and to increase protective factors in high-risk in the control group received conventional treatment per
participants with a high bacteria level and low salivary fluo- usual clinical practices (e.g., oral hygiene instruction, peri-
ride. The primary analyses showed that the intervention odic dental cleaning and oral examination scheduled every
group had a statistically significantly lower caries risk at 6 mo, radiographs scheduled every 24 mo, and restorative
follow-up and suggested a lower average caries increment treatment as needed), while participants in the intervention
(ΔDMFS) compared with control over 2 y (Featherstone et group received a targeted, combined antibacterial (0.12%
al. 2012). Of interest is whether this overall intervention chlorhexidine gluconate mouth rinse) and fluoride therapy
effect was due mainly to lowered overall caries risk, bacte- (1,100 ppm sodium fluoride toothpaste, 0.05% sodium flu-
ria reduction, or fluoride increase, as anticipated. If oride mouth rinse, and topical 1.1% NaF gel application)
observed, mediation through these mechanisms would pro- based on their individual risk assessments. Demographics,
vide further evidence for focusing future caries prevention clinical variables, and outcomes were measured at baseline
efforts on these components. Therefore, in this article, we and at follow-up intervals (Tables 2 and 3).
will examine if, and how much, the intervention reduced
caries increment through actually lowering caries risk,
reducing bacterial levels, or increasing salivary fluoride, as Mediators of Interest
opposed to through other unknown factors/mechanisms. Based on the theoretical mechanism of the CAMBRA inter-
vention to increase protective factors and decrease bacterial
Subjects and Methods challenge (Figure), we are interested in overall risk, bacterial
challenge (salivary MB and LB levels), and salivary fluoride
Subjects value as possible mediators. The overall risk was predefined
The CAMBRA trial enrolled 231 adults with 1 to 7 baseline as low or high based on a composite measure of bacteria
cavitated carious teeth at the School of Dentistry Student challenge and salivary fluoride levels (Table 4) based on

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

46 Journal of Dental Research 94(1)

Table 2.  Baseline Characteristics of All Participants and Participants Who Completed the Trial (Featherstone et al. 2012).

All Participants Participants Who Completed the Triala

Control Intervention Control Intervention

  (n = 116) (n = 115) (n = 52) (n = 57)
Age, y, mean ± standard deviation 38.4 ± 13.5 36.9 ± 13.2 40.9 ± 14.8 39.2 ± 14.7
Sex: female, n (%) 69 (59) 69 (60) 32 (62) 38 (67)
Race/ethnicity, n (%)
  Asian, non-Hispanic 8 (7) 15 (13) 8 (15) 14 (25)
  Black, non-Hispanic 6 (5) 8 (7) 6 (12) 8 (14)
  White, non-Hispanic 18 (16) 28 (24) 18 (35) 27 (47)
 Hispanic 13 (11) 7 (6) 13 (25) 6 (11)
  Other/multiracial, non-Hispanic 4 (3) 2 (2) 3 (6) 1 (2)
 Refused/missing 67 (58) 55 (48) 4 (8) 1 (2)
Education, n (%)
  High school 26 (22) 30 (26) 10 (19) 10 (17)
 College 65 (56) 58 (50) 26 (50) 30 (53)
 Graduate/professional 25 (22) 27 (23) 16 (31) 17 (30)
Works in San Francisco, n (%) 80 (69) 85 (74) 42 (81) 42 (74)
Last dental visit, n (%)
  <6 mo prior 27 (24) 32 (28) 16 (32) 16 (29)
  6 to 11 mo prior combine <12 mo 10 (9) 23 (20) 4 (8) 90 (18)
  12 to 23 mo prior 26 (23) 20 (18) 10 (20) 9 (16)
  24 to 35 mo prior combine 24+ mo 22 (19) 14 (12) 11 (22) 7 (13)
  36+ mo prior 29 (25) 25 (22) 9 (18) 13 (23)
Ever had professional cleaning, n (%) 98 (90) 100 (91) 46 (94) 48 (89)
Brushed 2+ times yesterday, n (%) 89 (77) 85 (74) 40 (77) 42 (74)
Used fluoride toothpaste, n (%) 87 (76) 79 (69) 39 (75) 40 (70)
Fair/poor oral health, n (%) 61 (53) 62 (54) 22 (42) 22 (39)
Drank alcohol in past week, n (%) 61 (53) 57 (50) 25 (50) 27 (47)
Smoked cigarette ≤30 d, n (%) 34 (29) 31 (27) 6 (11) 8 (14)
a
Compared with those who dropped out, participants with complete data were more likely at baseline to work in San Francisco, be more educated,
have better self-rated oral health status, and report less alcohol and tobacco use (all P < 0.05; Featherstone et al. 2012). However, baseline
characteristics between the intervention and control groups for participants with complete data were not significantly different (P = 0.305), indicating
that the 2 comparison groups were balanced in baseline covariates and that ignorability of treatment is a reasonable assumption.

previous findings (Krasse 1988; Leverett, Featherstone, control, no matter whether the effect operates through the
et al. 1993; Leverett, Proskin, et al. 1993). mediator of interest. The overall effect of the intervention is
the sum of 2 components: (1) the effect of the intervention
Outcome of Interest through the mediator of interest, called the indirect or media-
tion effect, and (2) the effect of the intervention around the
The primary outcome of interest in this study was the incre- mediator of interest, called the direct effect of the interven-
ment from baseline in the number of decayed, missing, and tion (Figure). The direct effect may operate through unknown
filled permanent surfaces (ΔDMFS) 24 mo after initial res- or unmeasured variables but not through the mediator of
torations were completed for baseline lesions. DMFS incre- interest. Standard analysis estimates the average overall
ment is a nonnegative integer count or a zero-inflated count effect of an intervention compared with control without effect
if many participants have no new DMFS, with higher val- decomposition into the component indirect and direct effects.
ues indicating worse dental outcomes.
Indirect or mediation effect. The indirect or mediation
effect of an intervention is the influence of the intervention
Effects of Interest through the mediator of interest. This effect can be concep-
Overall Effect. The overall effect of an intervention is the total tualized as the change in the outcome due exclusively to the
effect of the intervention on the outcome compared with effect of the intervention on the mediator. In other words, it

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

CAMBRA, Treatment Effect Mechanisms, and Caries Increment 47

Table 3.  Bacterial Challenge, Salivary Fluoride Concentration, High Overall Caries Risk, and DMFS Increment by Treatment Group.

Mediator

High Overall Caries Log10 MS, Log10 LB, Fluoride ppm,

  Risk, % mean ± SD mean ± SD mean ± SD
Baseline
 Intervention 93 4.27 ± 1.42 3.60 ± 1.95 0.03 ± 0.02
 Control 90 4.51 ± 1.29 3.71 ± 1.98 0.03 ± 0.02
  P 0.74 0.37 0.76 0.73
12 mo
 Intervention 67 3.26 ± 1.95 2.92 ± 2.10 0.08 ± 0.14
 Control 88 4.58 ± 1.49 3.28 ± 2.00 0.04 ± 0.03
  P 0.01 <0.01 0.25 0.02
Mediation effect
  Mean (95% CI) –0.38 (–0.82 to –0.03) –0.31 (–0.90 to 0.11) 0.00 (–0.17 to 0.18) 0.02 (–0.25 to 0.30)
  P 0.03 0.17 0.99 0.88
  DMFS Increment = 0, n (%) DMFS Increment, mean ± SD
Outcome at 24 mo  
 Intervention 7 (12.3) 3.5 ± 3.5
 Control 4 (7.7) 4.6 ± 4.1
  P 0.32 0.02

CI, confidence interval; DMFS, decayed missing filled tooth surface index; LB, lactobacilli; MS, mutans streptococci, SD, standard deviation.

Table 4.  Low and High Overall Caries Risk Predefined as a Function of log10 MS, log10 LB, and Salivary Fluoride (Featherstone et al.
2012).

log10 MS (CFU/mL)

log10 LB (CFU/mL) <4 4 to 5.9 ≥6

Low salivary fluoride: [F] < 0.03 ppm 
<1 Low risk High risk High risk
1 to 2.9 High risk High risk High risk
≥3 High risk High risk High risk
Medium salivary fluoride: 0.03 ppm ≤ [F] < 0.079 ppm 
<1 Low risk Low risk High risk
1 to 2.9 Low risk High risk High risk
≥3 High risk High risk High risk
High salivary fluoride: [F] ≥ 0.08 ppm 
<1 Low risk Low risk High risk
1 to 2.9 Low risk Low risk High risk
≥3 Low risk High risk High risk

LB, lactobacilli; MS, mutans streptococci.

is the expected difference in the outcome if the intervention Direct effect. The direct effect of an intervention is the
changed the mediator variable but with no other alteration influence of the intervention around (not through) the
in treatment status or in alternative variables through which mediator of interest. This effect can be conceptualized as
the intervention might act. This effect provides information the change in the outcome due to any and all actions of the
as to how much, if any, of the overall intervention effect intervention that do not involve altering the mediator of
on the outcome can be attributed to the intervention influ- interest. This effect is the difference in outcome expected
encing the mediator, as opposed to the intervention effect between intervention and control if the mediator had been
through other pathways. fixed so that no treatment effect could be attributed to it.

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

48 Journal of Dental Research 94(1)

Total Effect = (a) Mediation Effect + (b) Direct Effect

Total effect of CAMBRA: -0.96, 95% CI (-2.01, 0.08), p=0.07

Mediator at 12 months
Overall caries risk
Salivary log10 MS
(a) (a)
Salivary log10 LB
Salivary fluoride

(a) Mediation Effect through overall caries risk

-0.38, 95% CI (-0.82, -0.03), p=0.03
Intervention

Anti-bacterial treatment: Outcome at 24 months

Chlorhexidine gluconate rinse
DMFS increment
Fluoride treatment: (b) Direct Effect around overall caries risk
Fluoride gel application, dentifrice, rinse -0.58, 95% CI (-1.57, 0.42), p=0.25

Figure.  Theoretical mechanism, total effect, direct effect, and mediation effect of the CAMBRA intervention. CAMBRA, Caries
Management By Risk Assessment; CI, confidence interval; DMFS, decayed missing filled tooth surface index; LB, lactobacilli; MS,
mutans streptococci.

A significant direct effect of an intervention indicates that more detailed discussion on structural equation models.)
the intervention operates to affect the outcome through 1 or Various methods based on the potential outcome framework
more alternative pathways that do not include effects on the (Rubin 2004; Albert 2008; Imai et al. 2010; Vansteelandt
mediator of interest. 2010) have been recently proposed for mediation analysis.
Compared with conventional mediation analysis (such as
LSEM), popularized since Baron and Kenny (1986), these
Power Estimation newer methods first clearly define causal mediation effects
The CAMBRA trial was designed to have a power of 90% independent of particular statistical models and then specify
with a final evaluable sample size of 122 to detect an overall assumptions to identify the effects. These methods can also
difference in caries incidence proportions of 0.60 and 0.30 be extended to nonlinear models for general outcomes
(Featherstone et al. 2012). Given 109 participants who (other than normally distributed outcomes) when linear
completed the study, there would be 66% power to detect an models do not fit the data well. Dental outcomes such as
overall effect of 1.7 DMFS increment, with 40% power for caries incidence and DMFS increment are often binary and
a direct effect of 1.1 DMFS increment and 57% power for a count or zero-inflated count variables with a relatively large
mediation effect of 0.6 DMFS increment. portion of participants with values of zero to indicate no
new DMFS; such data are not normally distributed, so lin-
ear models usually do not fit well. Albert and Nelson (2011),
Statistical Methods Albert (2012), and Wang et al. (2013) have applied the
Previously, linear structural equation models (LSEM) have potential outcome-based methods to binary and count den-
been used to examine the direct and indirect effects of an tal outcomes. In this article, we used a method for media-
intervention or of measured social status on oral health tion analysis in which Cheng et al. (unpublished data, 2014)
(Donaldson et al. 2008; Tu et al. 2008). Instead of clearly extended the approach of Imai et al. (2010) specifically to
defining causal mediation effects independent of statistical assess count and zero-inflated count dental outcomes.
models, LSEMs interpret coefficient estimates only within The method (Imai et al. 2010; Cheng et al. unpublished
a particular statistical model as causal mediation effects. data, 2014) assumes sequential ignorability of intervention
Although LSEMs are usually presented along with a set of and mediator given baseline covariates; that is, the interven-
strong assumptions on all the variables in the system, the tion and mediator are unrelated to confounders after adjust-
assumptions necessary to identify direct and indirect effects ing for baseline covariates. Since the intervention was
to which causal interpretations can be ascribed are rarely randomized in the CAMBRA trial, assuming ignorability of
articulated explicitly. The LSEMs are also difficult to the intervention is straightforward in this study. Although
extend to nonlinear models. (See VanderWeele 2012 for a the postbaseline mediators (overall risk, bacteria load, and

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

CAMBRA, Treatment Effect Mechanisms, and Caries Increment 49

salivary fluoride at 12 mo) were not randomized, controlling participants and participants who completed the trial.
for relevant baseline covariates allows the assumption of no Compared with those who dropped out, participants with
unmeasured confounding on mediators to be reasonable. complete data were more likely at baseline to work in San
The analyses were based on the quasi-Bayesian Monte Francisco (77.1% vs 66.4%), be more educated (college and
Carlo approximation (King et al. 2000) and involve 5 steps above: 81.7% vs 70.5%), have better self-rated oral health
(Imai et al. 2010; Cheng et al. unpublished data, 2014). Briefly, status (good and above: 59.6% vs 35.3%), and report less
alcohol (never: 19.6% vs 9.1%) and tobacco use (never:
1. Fit the mediator and outcome models with observed 87.2% vs 58.2%; all P < 0.05; Featherstone et al. 2012).
data. However, baseline characteristics between the intervention
and control groups for participants with complete data were
For the continuous mediators (bacterial challenge and not significantly different (P = 0.305), indicating that the 2
salivary fluoride concentration), a linear model was fitted comparison groups were balanced in baseline covariates and
for the mediator. For the binary mediator overall caries risk that ignorability of treatment is a reasonable assumption.
(high or low), a logistic regression model was fitted for the Table 3 shows the intervention effect on potential media-
mediator. All of the mediator models included the treatment tors at 12 mo. The mean log10 MS at 12 mo was reduced
assignment as the independent variable and baseline value about 1.0 CFU/mL from baseline in the intervention group
of the mediator as a covariate. For the outcome DMFS but remained about 4.5 CFU/mL in the control group. The
increment, generalized linear models for Poisson, negative mean log10 LB at 12 mo was reduced about 0.7 CFU/mL
binomial (NB), zero-inflated Poisson (ZIP), and zero- from baseline in the intervention group and about 0.4 CFU/
inflated negative binomial (ZINB) were fitted, each with mL in the control group. The mean salivary fluoride concen-
the treatment assignment and mediator as independent vari- tration was increased about 0.05 ppm in the intervention
ables and baseline value of the mediator as a covariate. The group and about 0.01 ppm in the control group. The inter-
Vuong test (Vuong 1989) was used to compare the 4 out- vention group had a significantly lower MS level (P =
come models (Poisson, NB, ZIP, and ZINB) to select the 0.0002) and higher salivary fluoride level (P = 0.0233) than
best model for the final analysis. the control group at 12 mo, but there was no statistically
significant difference in LB level between the 2 groups. The
2. Simulate model parameters (coefficients) from their percentage of participants classified in the high-caries-risk
sampling distributions. group decreased from most participants at baseline (93%) to
3. Given the coefficients randomly drawn from their two-thirds (67%) at 12-mo follow-up in the intervention
sampling distributions in (2), (a) simulate potential group but remained high (12-mo follow-up: 88%) in the
values of the mediator based on the mediator model, control group (Table 3). The intervention group had a sig-
(b) simulate potential outcomes based on the out- nificantly lower percentage of participants classified as high
come model and simulated potential values of the overall risk at 12 mo than the control group (P = 0.0046).
mediator from (a), and (c) compute the direct, medi- At the end of the trial, at 24 mo, 12% of participants in the
ation, and overall effects of the intervention. intervention group did not have any increment in DMFS,
4. Perform Monte Carlo replications by repeating steps compared with about 8% in the control group (Table 3). The
(2) and (3) 1,000 times each. intervention group had mean DMFS increment about 1 sur-
5. Compute the point estimates and 95% confidence face fewer (3.5) than the control group (4.6), but the mean
intervals for direct, mediation, and total effects of change in DMFS at 24 mo was not significantly different
the intervention and corresponding P values based between the 2 groups in the primary analysis based on the
on the results from the 1,000 repetitions. extended Mantel-Haenszel test (P = 0.101; Featherstone et al.
2012). The DMFS increment in the 2 groups was not nor-
Analysis was completed using R software, version 3.0.3 mally distributed (Kolmogorov-Smirnov test, P < 0.01), indi-
(www.r-project.org). cating a linear model assuming normality may not be a good
fit for the data and that count data methods may fit better.
Mediation analyses were conducted following the 5
Results
steps discussed in the Statistical Methods section with the
Among the 115 participants randomized to control and 116 mediators of interest. Generalized linear models were fitted
to intervention, 52 (45%) participants in the control group separately for the DMFS increment using Poisson, NB, ZIP,
and 57 (49%) in the intervention group completed the and ZINB distributions. The NB model fit best by the Vuong
CAMBRA trial and final dental examination through the test for all the mediators (e.g., goodness of fit for overall
24-mo follow-up period and were included in this analysis. caries risk: NB > Poisson, P = 0.004; NB > ZIP, P = 0.006;
Table 2 shows the major baseline characteristics of all NB > ZINB, P = 0.310). Table 3 shows the mediation effects

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

50 Journal of Dental Research 94(1)

through overall caries risk, log10 MS, log10 LB, and salivary the effects through multiple mediators (log10 MS, log10 LB,
fluoride concentration at 12 mo, in which only the media- and salivary fluoride), suggesting that the CAMBRA inter-
tion effect through 12-mo overall caries risk was statisti- vention transmitted more of its anticaries effect through com-
cally significant (P = 0.03). bined action on multiple mediators than through any single
The Figure shows the pathways for direct, mediation, variable. In the future, a larger study would be needed to
and total effects of the intervention through 12-mo overall understand other potential pathways not included in this
caries risk. The total effect of the CAMBRA intervention study that could account for the remaining total effect of the
was –0.96 tooth surfaces, with a 95% confidence interval CAMBRA intervention on 24-mo DMFS increment. While
(CI; –2.01 to 0.08), indicating that the intervention was sug- we can only speculate as to the nature of these pathways (e.g.,
gestive at reducing DMFS increment by almost 1 surface on ecological shifts in biofilm-residing plaque bacteria or behav-
average compared with control (P = 0.07) in adults who had ioral changes triggered by following the intervention proto-
1 to 7 baseline cavitated teeth and received the initial caries col), an important implication of these findings is that much
removal and restorations. The direct effect of the CAMBRA of the total effect of the CAMBRA intervention was poten-
intervention that excluded any effect on 12-mo overall car- tially transmitted through mediators that were unmeasured or
ies risk was –0.58 tooth surfaces (95% CI, –1.57 to 0.42) unanticipated, speaking to the complexity and multifactorial
but was not statistically significant (P = 0.25). The media- nature of the caries process. Future mediation analyses
tion (indirect) effect of the CAMBRA intervention through applied to other caries prevention trials could help elucidate
reduction in 12-mo overall caries risk was statistically sig- such mechanisms, provided that investigators have collected
nificant with –0.38 tooth surfaces (P = 0.03; 95% CI, –0.82 detailed data on potential mediators.
to –0.03), meaning that approximately 36% (median of Similar to the conventional mediation analysis methods,
 mediation effect  the method used in this study assumes sequential ignorabil-
 % ity of the intervention and mediator; that is, the intervention
 total effect  over 1,000 Monte Carlo replica- and mediator are not associated with confounders. Although
tions) of the intervention effect on reducing 24-mo DMFS the ignorability of the intervention is often satisfied in a trial
increment was through its mediation effect on the 12-mo by random assignment, we must plausibly assume that
overall caries risk (Figure). No other statistically significant ignorability of the mediator is achieved after controlling for
mediation effect was found in this study (Table 3). baseline covariates. However, it is possible that unmeasured
confounding could have some influence on our estimates.
In addition, future directions include extending the method-
Discussion ology to accommodate multiple mediators and applying
Mediation models have attracted considerable interest in those new methods to these data.
health research in general and more recently in oral health In conclusion, our study suggests that the CAMBRA
research in particular (Garcia 2011), to better understand the intervention transmitted more of its anticaries effect through
mechanisms of how an intervention works to improve out- the combined action on multiple mediators than through
comes (Donaldson et al. 2008; Tu et al. 2008). However, the any single variable. This finding is consistent with studies
conventional mediation analysis largely based on linear mod- on multifactorial health problems, which suggest that a pro-
els may not fit many dental outcomes, which are frequently gram that addresses a variety of risk and protective factors
count or zero-inflated count data. This article illustrates a in multiple domains would have a greater impact on improv-
useful new extension of a method based on the potential out- ing outcomes than actions targeting any single factor (Office
come framework (Albert 2008; Imai et al. 2010) for dental of the Surgeon General et al. 2001; NSW Department of
data (Cheng et al. unpublished data, 2014) to further under- Community Services 2007; Saminsky 2010). Likewise,
stand the mechanism of the CAMBRA intervention. future anticaries interventions and clinical approaches are
The CAMBRA intervention was based on the hypothesis likely to be most efficacious when designed to operate
that a combined antibacterial and fluoride treatment would through diverse mediating pathways, consistent with the
reduce oral bacterial challenge and increase salivary fluoride complexity of the caries process.
level (i.e., reduce overall risk), which would then lead to
reduced DMFS increment. Perhaps because of a limited sam- Author Contributions
ple size, the mediation analyses conducted in this study failed J. Cheng, contributed to conception, design, data acquisition, anal-
to show a statistically significant mediation effect through ysis, and interpretation, drafted and critically revised the manu-
bacterial challenge or salivary fluoride level individually. In script; B.W. Chaffee, contributed to data acquisition and
contrast, significant mediation was found through the overall interpretation, critically revised the manuscript; N.F. Cheng, con-
risk at 12 mo, accounting for 36% of the total effect. The tributed to data analysis and interpretation, critically revised the
overall risk itself is an a priori composite measure combining manuscript; S.A. Gansky, J.D.B. Featherstone, contributed to

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014

CAMBRA, Treatment Effect Mechanisms, and Caries Increment 51

conception, design, data acquisition, and interpretation, critically Krasse B. 1988. Biological factors as indicators of future caries.
revised the manuscript. All authors gave final approval and agree Int Dent J. 38(4):219–225.
to be accountable for all aspects of the work. Leverett DH, Featherstone JD, Proskin HM, Adair SM, Eisenberg
AD, Mundorff-Shrestha SA, Shields CP, Shaffer CL, Billings
Acknowledgments RJ. 1993. Caries risk assessment by a cross-sectional discrim-
ination model. J Dent Res. 72(2):529–537.
This study was made possible by grants U54 DE 019285 and R01
Leverett DH, Proskin HM, Featherstone JD, Adair SM, Eisenberg
DE 012455 from the National Institute of Dental and Craniofacial
AD, Mundorff-Shrestha SA, Shields CP, Shaffer CL, Billings
Research (NIDCR), a component of the National Institutes of
RJ. 1993. Caries risk assessment in a longitudinal discrimina-
Health, which is part of the U.S. Department of Health and Human
tion study. J Dent Res. 72(2):538–543.
Services. The content is solely the responsibility of the authors and
MacKinnon DP, Luecen LJ. 2011. Statistical analysis for identi-
does not necessarily represent the official views of the National
fying mediating variables in public health dentistry interven-
Institutes of Health. We thank the study staff and, most impor-
tions. J Public Health Dent. 71(suppl 1):S37–S46.
tantly, the study participants who made this work possible. The
NSW Department of Community Services. 2007. Risk, protection
authors declare no potential conflicts of interest with respect to the
and resilience in children and families (research to practice
authorship and/or publication of this article.
notes); [accessed 2004 Sep 12]. www.community.nsw.gov.au/
docswr/_assets/main/documents/researchnotes_resilience.pdf.
References Office of the Surgeon General (US), National Center for Injury
Albert JM. 2008. Mediation analysis via potential outcomes mod- Prevention and Control (US), National Institute of Mental
els. Stat Med. 27(8):1282–1304. Health (US), Center for Mental Health Services (US). 2001.
Albert JM. 2012. Distribution-free mediation analysis for non- Youth violence: a report of the surgeon general. Chapter 5:
linear models with confounding. Epidemiology. 23(6):879– prevention and intervention. Rockville (MD): Office of the
888. Surgeon General (US); [accessed 2014 Sep 12]. http://www
Albert JM, Nelson S. 2011. Generalized causal mediation analy- .ncbi.nlm.nih.gov/books/NBK44295/.
sis. Biometrics. 67(3):1028–1038. Rubin D. 2004. Direct and indirect causal effects via potential out-
Baron RM, Kenny DA. 1986. The moderator-mediator variable comes. Scand J Stat. 31:161–170.
distinction in social psychological research: conceptual, Saminsky A. 2010. Preventing juvenile delinquency: early inter-
strategic, and statistical considerations. J Pers Soc Psychol. vention and comprehensiveness as critical factors. Student
51(6):1173–1182. Pulse. 2(2):1–3.
Donaldson AN, Everitt B, Newton T, Steele J, Sherriff M, Bower ten Cate JM, Featherstone JD. 1991. Mechanistic aspects of the
E. 2008. The effects of social class and dental attendance on interaction between fluoride and dental enamel. Crit Rev Oral
oral health. J Dent Res. 87(1):60–64. Biol Med. 2(3):283–296.
Featherstone JD. 2003. The caries balance: contribution factors Tu YK, Jackson M, Kellett M, Clerehugh V. 2008. Direct and
and early detection. J Calif Dent Assoc. 31(2):129–133. indirect effects of interdental hygience in a clinical trial. J
Featherstone JD, White JM, Hoover CI, Rapozo-Hilo M, Weintraub Dent Res. 87(11):1037–1042.
JA, Wilson RS, Zhan L, Gansky SA. 2012. A randomized VanderWeele TJ. 2012. Invited commentary: structural equa-
clinical trial of anticaries therapies targeted according to risk tion models and epidemiologic analysis. Am J Epidemiol.
assessment (caries management by risk assessment). Caries 176(7):608–612.
Res. 46(2):118–129. Vansteelandt S. 2010. Estimation of controlled direct effects on
Garcia I. 2011. Message from the NIDCR acting director. J Public a dichotomous outcome using logistic structural direct effect
Health Dent. 71(suppl 1):S1. models. Biometrika. 97(4):921–934.
Imai K, Keele L, Tingley D. 2010. A general approach to causal Vuong QH. 1989. Likelihood ratio tests for model selection and
mediation analysis. Psychol Methods. 15(4):309–334. non-nested hypotheses. Econometrica. 57(2):307–333.
King G, Tomz M, Wittenberg J. 2000. Making the most of statisti- Wang W, Nelson S, Albert JM. 2013. Estimation of causal media-
cal analyses: improving interpretation and presentation. Am J tion effects for a dichotomous outcome in multiple-mediator
Pol Sci. 44:341–355. models using the mediation formula. Stat Med. 32:4211–4228.

Downloaded from jdr.sagepub.com at Universidad de Sevilla. Biblioteca on March 23, 2015 For personal use only. No other uses without permission.

© International & American Associations for Dental Research 2014