ACMT Board Review Course Population Health and Assessments Jeffrey Brent, M.D., Ph.D. Toxicology Associates University of Colorado School of Medicine.

ACMT Board Review Course

Population Health and Assessments

Jeffrey Brent, M.D., Ph.D.Toxicology AssociatesUniversity of Colorado

School of Medicineand

Colorado School of Public Health

Topics for this Lecture1. Exposure monitoring and sampling2. PPE3. Study designs and measures of

association4. Statistical concepts5. Bias and confounding6. The Hill “criteria”7. Sensitivity, specificity, predictive

values

Exposure monitoring and sampling

Exposure monitoringEnvironmental sampling:

Wipe samplingWater samplingAir samplingBreathing zone measurements are best for

inhalational exposuresBiological monitoring e.g.:

Blood PbUrine mercury

Personal protective equipment

RespiratoryChemically protective clothing

Respiratory protection

Classification by sizeQuarter faceHalf faceFull face

Classification by functionAir-purifying

Uses chemical specific cartridges

Supplied airSCBA

Protection Factor The factor by which exposure is

reduced by use of a respiratorAmbient/protection factor = exposureFor example:

Ambient of 100 PPMProtection factor of 10Exposure = 100/10 = 10 PPM

Protection factors range form 5 – 10,000

The goal is to get exposure to below safe limits

Chemically protective clothing

Simple protectionEx. Aprons, boots, gloves

Nonencpasulating suits1 or 2 pieces, for example:

1 piece hooded coverallsHooded jacket + chem protective pants

Encapsulating suitHighest level of protection

Chemically protective clothing is usually designated by the EPA rating system

Level A = Max protection Encapsulating suit SCBA

Level B Supplied air respirator (or SCBA) Non-encapsulating garment

Level C Air purifying respirator Non-encapsulating garment

Level D = standard work clothes

STUDY DESIGN AND MEASURES OF ASSOCIATION

STATISTICAL CONCEPTS

NEXT

Types of human data

AnecdotalCase-reports and series

Controlled observationalControlled epidemiological studies

Controlled interventionalTrials

Controlled observational studies Cohort Cross-sectional Mortality Case-control Ecologic

For all epi studies: Groups should be matched for relevant

variables (e.g.)AgeSexAnything else that can affect results

Cohort studiesCompares exposed group to an

unexposed groupCan be retrospective or prospectiveCan assess incidence rates

Incidence = Rate of new cases (e.g. Cases/100,00/yr)Prevalence = Number of cases in the

population (e.g. cases/100,000)

Results expressed as Relative Risk (aka risk ratio or rate ratio)

Cross-sectional studies

Compares exposed group to an unexposed group at one snapshot in time

Provides prevalence dataExample: Prevalence of drug abuse in

medial toxicologists taking the board exam v those that are not

Results expressed as Relative Risk (aka risk ratio or rate ratio)

Mortality StudiesTypically a variation of a cohort studyAssesses diagnoses at time of deathResults expressed as mortality rates

corrected for relevant factors (“Standardized mortality rates”)

Usually expressed as a percentage(Mortality rate of exposed/rate in

unexposed) X 100 = SMR)Thus an SMR of 100 = no difference

btw exposed and unexposed

Case-control studies Compares individuals with a specific condition

with individuals that do not have that condition and compares exposures (or other risk factors)

For example: Comparing medical toxicologists with alcohol abuse (the cases) with those w/o this dx (controls) to see if there is a higher likelihood of alcoholic abuse if preparing to take the boards.

Thus assesses risk factors (e.g. exposures) related to specific conditions

Recall bias major problem Results expressed as Odds Ratios

Ecologic StudiesAssesses population numbers, not

individualsExample: Rate of admission for asthma

exacerbations in a city with high airborne PM10 compared to a city with low PM10

Results expressed as Relative Risk (aka risk ratio or rate ratio)

Ex: Snow’s study of cholera rates in London districts

Assessment of results of epi studiesBy convention a result is

statistically significant if the likelihood that it is chance result is < 5% or approx 2SDs from the mean

Interpretation of EPI Data

You can never assess the degree of association based only on the magnitude of the RR, OR or SMR

These values have an inherent uncertainty that is determined by the nature of the data

In modern epi this uncertainty is expressed as Confidence Intervals

The 95% Convention In science the

uncertainty in a result is expressed as that range of data in which there is a 95% likelihood that the real value exists

CIs express this range Ex: RR 1.6 (CI 0.7 –

2.4)

The importance of confidence intervals

If RR 1.6 (0.7 – 2.4)Than there is a 95% likelihood that the real

value lies between 0.7 – 2.4 If the real RR is:

>1= association 1= Non-association <1 = negative association (protective effect)

The 95% rule defines “statistical significance”

Thus, in order to be a statistically significant result the CI must not include 1

What about “p values”?

p Values are an older way of describing statistical significance

P < 0.05 means a result is statistically significant

OR 1.6 (0.7 – 2.4) = OR 1.6 p > 0.05

OR 1.6 (1.1 – 2.1) = OR 1.6 p < 0.05

Now the Bad NewsA statistical relationship never

a priori means a causal relationship

It is not the falling of the leaves that causes winter to

come

There are many more statistical associations in toxicology than there are causal relationships

How to get from association to causation

Requires specific rigorous methodology

Stems from Doll and Hills’ observation of an association between smoking and lung cancer

Hill’s ViewpointsTo be applied if a

statistical association is shown to exist

Does not account for quality of studies showing such an association

Hill’s “Viewpoints” Strength of association Consistency Specificity Biological gradient Temporal precedence Coherence Plausibility Experimental support Analogy

Also must consider the quality of the study

Bias and confoundingBias = systematic error

Ex: You are doing a study on childhood bl Pb concentrations and behavior. However, your lab technique inflates blood lead values by 20% = a bias.

Confounding = uncontrolled for factor affecting results.Ex: You are doing a retrospective cohort

study on chronic exposures to phosgene in laboratory workers and the incidence of lung cancer but you do not control for smoking.

Smoking is a confounder

A LITTLE TIP -KNOW HOW TO CALCULATE SENSITIVITY, SPECIFICITY, AND PREDICTIVE VALUES

Sensitivity The likelihood of a test being positive if the

condition is present Ex: Being under 16 yrs old has 100% sensitivity for

the detection of childhood Pb poisoning. Good for screening (few false negatives (FN))

Sensitivity = True positives (TP)/(TP + FN) Sensitivity is often expressed as a %

In example above if screen 100 individuals and 10 had Pb poisoning: Sens = 10/(10+0) = 10/10 = 1 (or 100%)

Another exampleTo determine the sensitivity of a terminal R in

lead AVR for the detecting of Na+ channel antagonist toxicity in all OD patients.

Screen 1,000 EKGs of OD patients, 100 had OD’d on Na+ channel blockers and 80 had a terminal R wave (TPs). 50 had a terminal R wave but did not OD on these agents.

TP = 80 FN = 20 Sens = TP/(TP + FN) = 80/(80 + 20) = 80/100 = 0.8 (80%)

SpecificityThe likelihood of the unaffected

individuals correctly having a negative test

Test: using criteria of being under 16 for dx of childhood Pb poisoning.N = 10010 with Pb poisoning - the other 90 are

false positives (FP)Specificity = True neg (TN)/(TN + FP)= 0/0+90

= 0

The second experimentScreen 1,000 EKGs of OD patients,

100 had OD’d on Na+ channel blockers and 80 had a terminal R wave. 50 others had a terminal R wave but did not OD on these agents (FPs).

TN= 850 FP = 50Sp = TN/(TN+FP) = 850/(850+50) = 850/900 =0.94 (94%)

Comparison btw Sensitivity and specificity

Both= True/(True + False)Sens = TP/(TP+FN)Specificity is the mirror image Spec = TN/(TN+FP) For both the “trues” in the numerator

and denominator terms are the same.The other denominator term is the

complete opposite

Positive predicative value

PPV = likelihood that the test will correctly Dx the condition

Test: using criteria of being under 16 for dx of childhood Pb poisoning.N = 10010 with Pb poisoning (TP) - the other 90

are false positives (FP)

PPV = TP/(TP+FP) = 10/(10 + 90) = 0.1 So 10% PPV

PPV – the second experiment

Screen 1,000 EKGs of OD patients, 100 had OD’d on Na+ channel blockers and 80 had a terminal R wave (TP). 50 others had a terminal R wave but did not OD on these agents (FPs).

TP = 80FP = 50PPV = TP/(TP + FP) = 80/(80+50)=

80/130= 0.6

Negative predicative value

The likelihood that the disease is not present if the test is negative

Test: using criteria of being under 16 for dx of childhood Pb poisoning.N = 1000 are TN0 are FN

NPV = TN/(TN+FN) = 0/(0+0) = 1 (100%)

NPV – a more rational study

Screen 1,000 EKGs of OD patients, 100 had OD’d on Na+ channel blockers and 80 had a terminal R wave. 50 others had a terminal R wave but did not OD on these agents (FPs).

NPV = TN/(TN + FN)TN = 850FN = 20NPV = 850/(850+20) = 850/870 = 0.97

Predicative values - summary

PPV uses only positive termsPPV = TP/(TP+FP)NPV uses only negative terms and

is exactly opposite of the PPVNPV = TN/(TN+FN)

If, when you are studying, this you have any questions call me (24/7) @ 303-765-3800 or e-mail me at [email protected]