Lecture 6 – July 16, 2003

Linking populations, prevention, and risk assessment

Objectives

To understand the operating premises of risk assessment

To be familiar with the different types of risk factors and with risk predictors

To understand the relationship between risk factors and levels of evidence

To be able to differentiate between risk models and prediction models

To be familiar with criteria for prediction models

To understand the notion of targeting and how it applies to teeth as well as individuals

Operating premises of risk assessment

For decades our medical colleagues have been developing methods to identify individuals at high risk of diseases such as heart disease, stroke, and cancer

High risk individuals are then targeted for special programs, such as early detection and treatment for various types of cancer, and risk reduction efforts for heart disease and stroke

Operating premises of risk assessment

The majority of risk assessment efforts applied to the two major dental diseases, caries and periodontitis, began in the early 1980’s

Interest in risk assessment for dental conditions arose out of a change in prior paradigms regarding the etiology and progression of these diseases

Operating premises of risk assessment

Under the old paradigm prevalence of the conditions was extremely high and little error in prediction resulted when everyone was categorized as having the disease

Under the new paradigm, the prevalence of the disease is known not to be as high and it is known that some people are more likely to be affected by the condition than others

Operating premises of risk assessment

Now from a perspective of assigning risk, caries and periodontitis can be thought to be more like some of our common medical conditions

Certain people or subgroups of the population are at higher risk than others and efforts at prevention and intervention involve a combination of personal behaviors and professional practices

Operating premises of risk assessment

That: Clinical, behavioral, and etiological factors can be used to determine caries risk

That: Not all patients require the same level of prevention

Thus: The extent of prevention can be appropriate to the level of risk

Operating premises of risk assessment

Distribution

60% of caries occurs in 20% of children

Question

Can children at high risk be identified prior to disease development?

Operating premises of risk assessment

Studies undertaken:

to develop models to identify children at high risk prior to development of the disease

Multivariate (multivariable) perspective

Variables used in multivariable models

There are essentially two types of variables that can be used in the development of multivariable models:

1) risk factors

those that can be modified (e.g., levels of pathogens)

those that are immutable to change (background factors)

2) risk predictors

Risk Factors

Recently the World Workshop on Periodontics adopted a working definition of the term risk factor:

An environmental, behavioral, or biologic factor

confirmed by temporal sequence,

usually in longitudinal studies

if present, directly increases the probability of a disease occurring and, if absent or removed, reduces the probability

Risk factors are part of the causal chain, or expose the host to the causal chain

Example - variables in multivariable models(I)

Clinical conditions - referral status, caries in primary teeth, orthodontic care

Microbiological tests - S mutans, lactobacilli

Sociodemographic factors - snack consumption, dental health practices, fluoride history, antibiotic use

Immutable risk factor

Demographic risk factor/genetic risk factor (background characteristic)

Meets the definition of a risk factor, but currently is immutable to change

Perhaps more likely to expose the host to the causal chain than be part of the causal chain

Not useful for intervention, but often useful as a group characteristic when targeting people to apply another intervention

May be informative in modeling because of potential for interaction effects

Example - Variables in multivariable models (I)

Clinical/dental - age, water fluoridation, gingival recession, n teeth with perio pockets over 3 mm in depth, n teeth with calculus

Measures of general health and physical function

Behavioral and psychosocial - smoking, sugar consumption, anxiety, social integration, depression, stress

Risk predictor (risk marker)

Risk predictor (risk marker)

A characteristic associated with elevated risk of disease (i.e., it predicts well)

Not thought to be part of the causal chain (e.g., tooth loss is a good predictor of future disease)

Usually either a byproduct of the causal chain (usually called a risk marker)

Or some historical measure of the outcome, such as number of missing teeth, or baseline caries

Useful to identify who is at risk, but not useful in identifying likely interventions

Risk predictor (risk marker)

Usually biologic markers indicative of disease process, but are currently thought not to be etiologic

In dentistry tend to be alternative historical measures of the disease being studied, such as the number of missing teeth or past evidence of dental caries or periodontal disease

The terms risk predictor and risk marker often used synonymously

Risk marker tends to be used when describing biological predictors, such as C-reactive protein being a marker for inflammation

Risk predictor often used for any non-etiologic variable that is a good predictor

Example: Variables in multivariable models (I)

Clinical/dental - age, water fluoridation, gingival recession, n teeth with perio pockets over 3 mm in depth, n teeth with calculus

Measures of general health and physical function

Behavioral and psychosocial - smoking, sugar consumption, anxiety, social integration, depression, stress

Risk factors - criteria

Three criteria need to be satisfied in order to identify a characteristic as a risk factor:

1. the factor must be observed to covary with the disease

i.e., the factor must be statistically associated with the development of the disease

or equivalently, the frequency of disease must be observed to differ by category or value of the factor

2. the presence of the risk factor ( or a relevant change in the risk factor) must precede occurrence of the disease

Risk factor – criteria:

3 – the observed association must not be entirely due to any source of error

including chance or sampling error

the involvement of other (extraneous) risk factors

or other problems with the study design or data analysis

E.g., study should reflect design and analytic methods

that make it less likely to produce:

biased findings

or associations unadjusted for potential confounders

Risk factors – level of evidence

If the criteria for identifying a risk factor are to be met, longitudinal studies must be used

However longitudinal studies are expensive to conduct and may take many years to complete

Consequently variables thought to be risk factors frequently are uncovered through associations seen in prevalence (cross-sectional) studies

Risk factors – level of evidence

Indicator for Caries Management -

From the patient history:

past and present fluoride availability

Risk factors – Level of evidence

Level 1 -

Strong evidence from at least one published systematic review of multiple well-designed randomized controlled trials

(multiple,reviewed) ( rct’s) (pop-based)

Risk factors – level of evidence

Indicator for Caries Management -

From the patient history:

dietary component in smooth surface caries

Risk factors – level of evidence

Level 3 -

Evidence from published well-designed trials without randomization

single group pre-, post- comparisons

cohort, time series or matched case controlled studies

(no randomization) (population-based)

Risk factors – level of evidence

Term “risk indicator”

used to differentiate factors that have only been identified by means of prevalence data and can be defined as a probable or putative risk factor

Often detected in cross-sectional studies, that has not been confirmed by longitudinal studies

Other terms used to label factors derived from prevalence studies are putative (or potential) risk factors

Example - variables in multivariable models (I)

Clinical/dental - age, water fluoridation, gingival recession, n teeth with perio pockets over 3 mm in depth, n teeth with calculus

Measures of general health and physical function

Behavioral and psychosocial - smoking, sugar consumption, anxiety, social integration, depression, stress

An example: caries risk assessment (CRA)

Screen populations for model risk variables

Use the model to predict risk

An example: caries risk assessment (CRA)

A total of nine factors considered

Each factor has assigned weighting score

Weights summed to arrive at a CRA score: 0-3, low; 4-8, moderate; 9, high risk

An example: caries risk assessment (CRA)

Evidence of Prior Infectious Disease

4. Number of filled surfaces = 5 or > (2)

5. Last filling for caries was placed less than 1 year ago (1)

An example: caries risk assessment (CRA)

Secondary factors that influence the rate of progression of the carious lesion

8. Fluoride exposure is/was adequate (1/2)

9. Unstimulated saliva flow is below normal (<0.2 ml/min) (2)

An example: caries risk assessment (CRA)

Three risk categories: high, middle, low

Weights summed to arrive at a CRA score:

0-3, low

4-8, moderate

9, high risk

Risk model vs. prediction model

Distinction based on the intended use of the model:

prediction of people at high risk

or prediction of people at high risk and delineation of risk factors

Risk model

If the purpose is prediction and delineation of risk factors

in order to develop the most effective prevention or treatment interventions

then a “risk” model should be developed

Prediction model

If are mainly interested in identifying who is at high risk

The main goal is to maximize sensitivity and specificity of the prediction,

so any good predictor may be included in the model

Prediction models

Some situations favor the use of a prediction model

When the appropriate interventions are known

the objective may be to maximize the ability of the model to identify high-risk and low-risk individuals

i.e., maximize sensitivity and specificity

the proportions of people who have

and do not have the disease, respectively, who are correctly classified by the model

Prediction models

While prediction models may predict more accurately (greater sensitivity and specificity) than risk models

they contain predictor variables that will not influence the incidence of disease if changed

or are characteristics that are immutable to change

An example: caries risk assessment (CRA)

The following series of factors are considered: Presence of active caries

1. Presence of frank lesions in mouth (3)

2. Frank carious lesions = 3 or more (5)

3. Incipient caries surfaces = 3 or more (4)

Prediction models

These same predictors (such as past history of disease or number of teeth) may be powerful predictors that are easy and inexpensive to obtain

In contrast, risk factors, such as microbiologic activity, salivary buffering capacity, and immune status, are more expensive to determine and increase the cost of assessment

An example: caries risk assessment (CRA)

Factors consistent with the current paradigm of the cause and progression of caries

6. Mutans streptococcus count in saliva is high (>= 106 cfu/ml) (2)

7. Sugar/diet history (2)

Prediction models

Thus prediction models often may be the models of choice

However when using prediction models, one should always remember that the models may not make much sense

because the presence of powerful predictors in a model may mask the effects of related risk factors

E.g., the presence of “number of teeth” or “baseline DMF score” in a model may mask the role of microorganisms in diseases known to be infectious in nature

Prediction model criteria

Criteria for predictive model for high-caries risk:

quick

inexpensive

easy to use

limited equipment needed

readily acceptable

Prediction model criteria

Needs to be within acceptable parameters of accuracy, precision:

sensitivity, specificity

positive and negative predictive values

Prediction model criteria -

Assessing Model Utility

Compare actual and predicted scores

Model sensitivity - The % of people that actually have the disease and were correctly predicted to get it

Model specificity - The % of people that did not get the disease and were correctly predicted to be disease free

Prediction model criteria

Choices between sensitivity and specificity must be made:

    Where to draw the line?

    What are the costs of misclassifying             patients?

    Answer is not statistical

Prediction model criteria

When we attempt to improve the sensitivity of procedures, we become less selective

include people who do not have the disease

Can be more restrictive

by raising the specificity of the test …

reduce the # of false-positive determinations

Result: targeting

Directing preventive services to those at risk

1) For the clinician:

Tailor the preventive program to the patient

2) For the community planner:

Target populations

Result: targeting - for the clinician

The CRA score and risk determination help to determine:

an individualized preventive plan incorporated into the overall tx plan

a reevaluation interval determined

Result: Targeting – for a population

Direct preventive services to those at risk

Appropriate levels of care

Greater effectiveness of preventive procedures

Economic efficiency and cost containment

Example: targeting sealant use

Changes in epidemiology of caries during the past decades:

decrease in prevalence

decrease in rate of progression

distribution on different tooth surfaces

Implications

predicting risk

conducting caries preventive programs

Example: targeting sealant use

With limited resources

Need to find ways to predict and target children at higher risk for sealant application

Sealant placement on all sound pit and fissure surfaces

of primary and permanent teeth

on all children is not feasible

Example: targeting sealant use

If caries levels keep declining...

smaller amounts of disease for a sealant program to prevent...

procedure more costly per surface of caries prevented...

unless susceptible surfaces identified

Example: targeting sealant use

Tooth & Surface-Specific Patterns of Caries Attack

Slide 53

Slide 54

Slide 55

Example: targeting sealant use

Figure 3 and 4: occlusal of first molars is the most caries-prone site

Figure 5: increased caries rate of first molars compared to other surfaces

Example: targeting sealant use

Differences in risk among tooth surfaces

allow effective targeting of disease prevention

without incurring cost of population screening

Example: targeting sealant use

Identifying teeth at highest risk of dental caries is more efficient than identifying high-risk individuals

Example - targeting sealant use

Tooth- & Surface-Specific Patterns of Caries Attack

Providers can more easily predict which tooth surfaces are at greater risk

Can better target sealant resources

Example: targeting sealant use

Given: differences in caries risk among teeth/teeth surfaces

considerably larger than among individuals -

By identifying teeth - rather than individuals at highest risk

more effectively target preventive efforts w/o need to screen individuals


	To understand the operating premises of risk assessment
	To be familiar with the different types of risk factors and with risk predictors
	To understand the relationship between risk factors and levels of evidence
	To be able to differentiate between risk models and prediction models
	To be familiar with criteria for prediction models
	To understand the notion of targeting and how it applies to teeth as well as individuals


	For decades our medical colleagues have been developing methods to identify individuals at high risk of diseases such as heart disease, stroke, and cancer
	High risk individuals are then targeted for special programs, such as early detection and treatment for various types of cancer, and risk reduction efforts for heart disease and stroke


	The majority of risk assessment efforts applied to the two major dental diseases, caries and periodontitis, began in the early 1980’s
	Interest in risk assessment for dental conditions arose out of a change in prior paradigms regarding the etiology and progression of these diseases


	Under the old paradigm prevalence of the conditions was extremely high and little error in prediction resulted when everyone was categorized as having the disease
	Under the new paradigm, the prevalence of the disease is known not to be as high and it is known that some people are more likely to be affected by the condition than others


	Now from a perspective of assigning risk, caries and periodontitis can be thought to be more like some of our common medical conditions
		Certain people or subgroups of the population are at higher risk than others and efforts at prevention and intervention involve a combination of personal behaviors and professional practices


	That: Clinical, behavioral, and etiological factors can be used to determine caries risk

	That: Not all patients require the same level of prevention

	Thus: The extent of prevention can be appropriate to the level of risk


	Distribution
		60% of caries occurs in 20% of children
	Question
		Can children at high risk be identified prior to disease development?


	Studies undertaken:
	to develop models to identify children at high risk prior to development of the disease

	Multivariate (multivariable) perspective


	There are essentially two types of variables that can be used in the development of multivariable models:
	1) risk factors
		those that can be modified (e.g., levels of pathogens)
		those that are immutable to change (background factors)
	2) risk predictors


	Recently the World Workshop on Periodontics adopted a working definition of the term risk factor:
	An environmental, behavioral, or biologic factor
		confirmed by temporal sequence,
		usually in longitudinal studies
		if present, directly increases the probability of a disease occurring and, if absent or removed, reduces the probability
	Risk factors are part of the causal chain, or expose the host to the causal chain


	Clinical conditions - referral status, caries in primary teeth, orthodontic care

	Microbiological tests - S mutans, lactobacilli

	Sociodemographic factors - snack consumption, dental health practices, fluoride history, antibiotic use


	Demographic risk factor/genetic risk factor (background characteristic)
		Meets the definition of a risk factor, but currently is immutable to change
		Perhaps more likely to expose the host to the causal chain than be part of the causal chain
		Not useful for intervention, but often useful as a group characteristic when targeting people to apply another intervention
		May be informative in modeling because of potential for interaction effects


	Clinical/dental - age, water fluoridation, gingival recession, n teeth with perio pockets over 3 mm in depth, n teeth with calculus

	Measures of general health and physical function

	Behavioral and psychosocial - smoking, sugar consumption, anxiety, social integration, depression, stress


	Risk predictor (risk marker)
		A characteristic associated with elevated risk of disease (i.e., it predicts well)
		Not thought to be part of the causal chain (e.g., tooth loss is a good predictor of future disease)
		Usually either a byproduct of the causal chain (usually called a risk marker)
		Or some historical measure of the outcome, such as number of missing teeth, or baseline caries
		Useful to identify who is at risk, but not useful in identifying likely interventions


	Usually biologic markers indicative of disease process, but are currently thought not to be etiologic
		In dentistry tend to be alternative historical measures of the disease being studied, such as the number of missing teeth or past evidence of dental caries or periodontal disease
	The terms risk predictor and risk marker often used synonymously
		Risk marker tends to be used when describing biological predictors, such as C-reactive protein being a marker for inflammation
		Risk predictor often used for any non-etiologic variable that is a good predictor


Three criteria need to be satisfied in order to identify a characteristic as a risk factor:
	1. the factor must be observed to covary with the disease
		i.e., the factor must be statistically associated with the development of the disease
		or equivalently, the frequency of disease must be observed to differ by category or value of the factor
	2. the presence of the risk factor ( or a relevant change in the risk factor) must precede occurrence of the disease


3 – the observed association must not be entirely due to any source of error
	including chance or sampling error
	the involvement of other (extraneous) risk factors
	or other problems with the study design or data analysis

	E.g., study should reflect design and analytic methods
		that make it less likely to produce:
		biased findings
		or associations unadjusted for potential confounders


	If the criteria for identifying a risk factor are to be met, longitudinal studies must be used
		However longitudinal studies are expensive to conduct and may take many years to complete
		Consequently variables thought to be risk factors frequently are uncovered through associations seen in prevalence (cross-sectional) studies


	Indicator for Caries Management -
	From the patient history:
	past and present fluoride availability


	Level 1 -
	Strong evidence from at least one published systematic review of multiple well-designed randomized controlled trials
	(multiple,reviewed) ( rct’s) (pop-based)


	Indicator for Caries Management -
	From the patient history:
	dietary component in smooth surface caries


	Level 3 -
	Evidence from published well-designed trials without randomization
	single group pre-, post- comparisons
	cohort, time series or matched case controlled studies
	(no randomization) (population-based)


	Term “risk indicator”
	used to differentiate factors that have only been identified by means of prevalence data and can be defined as a probable or putative risk factor
		Often detected in cross-sectional studies, that has not been confirmed by longitudinal studies
		Other terms used to label factors derived from prevalence studies are putative (or potential) risk factors



	Screen populations for model risk variables

	Use the model to predict risk


	A total of nine factors considered
	Each factor has assigned weighting score
	Weights summed to arrive at a CRA score: 0-3, low; 4-8, moderate; 9, high risk


	Evidence of Prior Infectious Disease

	4. Number of filled surfaces = 5 or > (2)

	5. Last filling for caries was placed less than 1 year ago (1)


	Secondary factors that influence the rate of progression of the carious lesion
	8. Fluoride exposure is/was adequate (1/2)

	9. Unstimulated saliva flow is below normal (<0.2 ml/min) (2)


	Three risk categories: high, middle, low
	Weights summed to arrive at a CRA score:
	0-3, low
	4-8, moderate
	9, high risk


	Distinction based on the intended use of the model:
	prediction of people at high risk
	or prediction of people at high risk and delineation of risk factors


	If the purpose is prediction and delineation of risk factors
	in order to develop the most effective prevention or treatment interventions
	then a “risk” model should be developed