a/c ad

Odds ratio = =
b/d bc
a/(a + b)
260
Relative risk =
SECTION II Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology
c/(c + d)and Biostatistics
a c
Attributable risk =
a+b c+d

Disease or outcome
Quantifying risk Definitions and formulas are based on the classic
2 × 2 or contingency table.

or intervention
Exposure
a b

c d

TERM DEFINITION EXAMPLE FORMULA

Odds ratio Typically used in case-control If in a case-control study, 20/30 patients a/c ad
OR = =
studies. Represents the odds of with lung cancer and 5/25 healthy b/d bc
exposure among cases (a/c) vs odds individuals report smoking, the OR is a b
of exposure among controls (b/d). 8; so the patients with lung cancer are 20 5
OR = 1 Ž odds of exposure are 8 times more likely to have a history of c d
10 20
equal in cases and controls. smoking.
OR > 1 Ž odds of exposure are You take a case to the OR.
greater in cases.
OR < 1 Ž odds of exposure are
greater in controls.
Relative risk Typically used in cohort studies. If 5/10 people exposed to radiation are a/(a + b)
RR =
Risk of developing disease in the diagnosed with cancer, and 1/10 people c/(c + d)
exposed group divided by risk in not exposed to radiation are diagnosed a b
the unexposed group. with cancer, the RR is 5; so people 5 5
RR = 1 Ž no association between exposed to radiation have a 5 times c d
1 9
exposure and disease. greater risk of developing cancer.
RR > 1 Ž exposure associated with For rare diseases (low prevalence), OR
 disease occurrence. approximates RR.
RR < 1 Ž exposure associated with
 disease occurrence.
Relative risk The proportion of risk reduction If 2% of patients who receive a flu RRR = 1 − RR
reduction attributable to the intervention as shot develop the flu, while 8% of
compared to a control. unvaccinated patients develop the flu,
then RR = 2/8 = 0.25, and RRR = 0.75.
Attributable The difference in risk between If risk of lung cancer in people who a c
AR = −
risk exposed and unexposed groups. smoke is 21% and risk in people who a + b c + d
don’t smoke is 1%, then the attributable RR − 1
AR% = × 100
risk is 20%. RR
Absolute The difference in risk (not the If 8% of people who receive a placebo c a
ARR = −
risk proportion) attributable to the vaccine develop the flu vs 2% of people c+d a+b
reduction intervention as compared to a who receive a flu vaccine, then ARR =
control. 8%–2% = 6% = 0.06.
Number Number of patients who need to NNT = 1/ARR
needed to be treated for 1 patient to benefit.
treat Lower number = better treatment.
Number Number of patients who need to NNH = 1/AR
needed to be exposed to a risk factor for 1
harm patient to be harmed. Higher
number = safer exposure.
Case fatality Percentage of deaths occurring If 4 patients die among 10 cases of deaths
CFR% = × 100
rate among those with disease. meningitis, case fatality rate is 40%. cases

FAS1_2022_06-PubHealth.indd 260 11/4/21 12:18 PM

 Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics SECTION II 261

Quantifying risk (continued)

TERM DEFINITION EXAMPLE FORMULA
Mortality Number of deaths (in general or If 80 people in a town of 10,000 die over Deaths/1000 people per
rate due to specific cause) within a 2 years, mortality rate is 4 per 1000 per year.
population over a defined period. year.
Attack rate Proportion of exposed people who If 80 people in a town are exposed and People who become ill
become ill. 60 people become ill, attack rate is 75%. Total people exposed

Demographic As a country proceeds to higher levels of development, birth and mortality rates decline to varying
transition degrees, changing the age composition of the population.
Population pyramid
Age

Age

Age

Male Female
Male Female Female % Population % Population %
Male Population

Birth rate   

Mortality rate   
Life expectancy Short Long Long

Population Growing Stable Declining

Likelihood ratio probability of positive result in patient with disorder sensitivity TP rate
LR+ = = =
probability of positive result in patient without disorder 1 – specificity FP rate

probability of negative result in patient with disorder 1 – sensitivity FN rate

LR– = = =
probability of negative result in patient without disorder specificity TN rate

LR+ > 10 indicates a highly specific test, while LR– < 0.1 indicates a highly sensitive test.
Pretest probability × LR = posttest odds. Posttest probability = posttest odds / (posttest odds + 1).

Kaplan-Meier curve Graphic representation of event probability

1.0
(y-axis) vs length of time (x-axis). Useful for 0.9 Drug
Survival probability

displaying “time-to-event” data. Outcomes 0.8

examined may include any event, but 0.7
frequently include mortality. 0.6
Control
0.5
Survival probability = 1 – (event probability).
0.4
0.3
0.2

X Y Z
Time

FAS1_2022_06-PubHealth.indd 261 11/8/21 1:32 PM

 262 SECTION II Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics

Evaluation of Sensitivity and specificity are fixed properties Disease

–
diagnostic tests of a test. PPV and NPV vary depending on
disease prevalence in population being tested. PPV
TP FP = TP/(TP + FP)

Test
NPV
– FN TN = TN/(TN + FN)

Sensitivity Specificity Prevalence

TP + FN
= TP/(TP + FN) = TN/(TN + FP) (TP + FN + FP + TN)

Sensitivity (true- Proportion of all people with disease who test = TP / (TP + FN)
positive rate) positive, or the ability of a test to correctly = 1 – FN rate
identify those with the disease. SN-N-OUT = highly SeNsitive test, when
Value approaching 100% is desirable for ruling Negative, rules OUT disease
out disease and indicates a low false-negative High sensitivity test used for screening
rate.
Specificity (true- Proportion of all people without disease who = TN / (TN + FP)
negative rate) test negative, or the ability of a test to correctly = 1 – FP rate
identify those without the disease. SP-P-IN = highly SPecific test, when Positive,
Value approaching 100% is desirable for ruling rules IN disease
in disease and indicates a low false-positive High specificity test used for confirmation after a
rate. positive screening test
Positive predictive Probability that a person who has a positive test PPV = TP / (TP + FP)
value result actually has the disease. PPV varies directly with pretest probability
(baseline risk, such as prevalence of disease):
high pretest probability Ž high PPV
Negative predictive Probability that a person with a negative test NPV = TN / (TN + FN)
value result actually does not have the disease. NPV varies inversely with prevalence or pretest
probability
Possible cutoff values for vs – test result
Disease Disease A = 100% sensitivity cutoff value
Number of people

absent present B = practical compromise between specificity and sensitivity

C = 100% specificity cutoff value

TN TP Lowering the cutoff value: ↑ Sensitivity ↑ NPV

↑ ↑ ↑
B A (↑ FP FN) Specificity PPV
↑

FN FP
Raising the cutoff value: ↑ Specificity ↑ PPV
A B C B C ( ↑ FN FP)
↑ ↑
Sensitivity NPV
↑
↑

Test results

Receiver operating ROC curve demonstrates how well a diagnostic Ideal test (AUC = 1)
1
characteristic curve test can distinguish between 2 groups (eg, 1)
<
disease vs healthy). Plots the true-positive rate UC
<A
(sensitivity) against the false-positive rate .5
t (0
TP rate (sensitivity)

0.5)
(1 – specificity).
s
l te

C=
ua

The better performing test will have a higher AU

e(
Act

lu
area under the curve (AUC), with the curve va
it ve
ic
closer to the upper left corner. pr
ed
o
In diseases diagnosed based on low lab values N
(eg, anemia), the curve is flipped: lowering the
cutoff further Ž  FP,  FN; raising the cutoff
FP rate (1 – specificity) 1
Ž  FN,  FP.

FAS1_2022_06-PubHealth.indd 262 11/4/21 12:18 PM

 Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics SECTION II 263

Precision vs accuracy
Precision (reliability) The consistency and reproducibility of a test. Random error  precision in a test.
The absence of random variation in a test.  precision Ž  standard deviation.
 precision Ž  statistical power (1 − β).
Accuracy (validity) The closeness of test results to the true values. Systematic error  accuracy in a test.
The absence of systematic error or bias in a test.

Accuracy Accuracy
High Low High Low

High High Low Low

Precision Precision

Incidence vs # of new cases Incidence looks at new cases (incidents).

Incidence = (per unit of time)
prevalence # of people at risk
# of existing cases (at a point in Prevalence looks at all current cases.
Recurrence Prevalence =
Total # of people    time)
Incidence
in a population
Prevalence Prevalence
= Incidence rate × average duration
1 – prevalence of disease
Mortality Cure
Prevalence ≈ incidence for short duration disease Prevalence ∼ pretest probability.
(eg, common cold).  prevalence Ž  PPV and  NPV.
Prevalence > incidence for chronic diseases, due to
large # of existing cases (eg, diabetes).
SITUATION INCIDENCE PREVALENCE

 survival time — 
 mortality — 
Faster recovery time — 
Extensive vaccine administration  
 risk factors  
 diagnostic sensitivity  
New effective treatment started — 
 contact between patients with and — —
without noninfectious disease
 contact between infected and  
noninfected patients with airborne
infectious disease

FAS1_2022_06-PubHealth.indd 263 11/4/21 12:18 PM

 264 SECTION II Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics

Bias and study errors

TYPE DEFINITION EXAMPLES STRATEGIES TO REDUCE BIAS
Recruiting participants
Selection bias Nonrandom sampling Berkson bias—cases and/ Randomization (creates groups
or treatment allocation or controls selected from with similar distributions
of subjects such that hospitals (bedside bias) are of known and unknown
study population is not less healthy and have different variables)
representative of target exposures Ensure the choice of the right
population Attrition bias—participants lost comparison/reference group
Most commonly a sampling to follow up have a different
bias prognosis than those who
complete the study
Performing study
Recall bias Awareness of disorder alters Patients with disease recall Decrease time from exposure
recall by subjects; common in exposure after learning of to follow-up; use medical
retrospective studies similar cases records as sources
Measurement bias Information is gathered in a Using a faulty automatic Use objective, standardized,
systemically distorted manner sphygmomanometer and previously tested methods
Hawthorne effect—participants of data collection that are
change behavior upon planned ahead of time
awareness of being observed Use placebo group
Procedure bias Subjects in different groups are Patients in treatment group Blinding (masking) and
not treated the same spend more time in highly use of placebo reduce
specialized hospital units influence of participants and
Observer-expectancy Researcher’s belief in the An observer expecting researchers on procedures and
bias efficacy of a treatment changes treatment group to show signs interpretation of outcomes
the outcome of that treatment of recovery is more likely to as neither are aware of group
(also called Pygmalion effect) document positive outcomes assignments
Interpreting results
Confounding bias Factor related to both exposure An uncontrolled study shows Multiple/repeated studies
and outcome (but not on an association between Crossover studies (subjects act
causal path) distorts effect drinking coffee and lung as their own controls)
of exposure on outcome (vs cancer; however, people who Matching (patients with
effect modification, in which drink coffee may smoke more, similar characteristics in both
the exposure leads to different which could account for the treatment and control groups)
outcomes in subgroups association Effect modification—
stratified by the factor) association is shown
differently in individual
subgroups due to stratification
by given factor even when
there is association between
exposure and outcome
Lead-time bias Early detection interpreted as Breast cancer diagnosed early Measure “back-end” survival
 survival, but the disease by mammography may (adjust survival according to
course has not changed appear to exaggerate survival the severity of disease at the
time because patients are time of diagnosis)
known to have the cancer for
longer

FAS1_2022_06-PubHealth.indd 264 11/4/21 12:18 PM

 Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics SECTION II 265

Bias and study errors (continued)

TYPE DEFINITION EXAMPLES STRATEGY TO REDUCE BIAS
Interpreting results (continued)
Length-time bias Screening test detects diseases A slowly progressive cancer A randomized controlled trial
with long latency period, is more likely detected by a assigning subjects to the
while those with shorter screening test than a rapidly screening program or to no
latency period become progressive cancer screening
symptomatic earlier

Statistical distribution
Measures of central Mean = (sum of values)/(total number of values). Most affected by outliers (extreme values).
tendency Median = middle value of a list of data sorted If there is an even number of values, the median
from least to greatest. will be the average of the middle two values.
Mode = most common value. Least affected by outliers.
Measures of Standard deviation = how much variability σ = SD; n = sample size.
dispersion exists in a set of values, around the mean of Variance = (SD)2.
these values. SE = σ/√n.
Standard error = an estimate of how much SE  as n .
variability exists in a (theoretical) set of sample
means around the true population mean.
Normal distribution Gaussian, also called bell-shaped.
–1σ +1σ
Mean = median = mode.
For normal distribution, mean is the best –2σ +2σ
–3σ +3σ
measure of central tendency.
For skewed data, median is a better measure of 68%

central tendency than mean. 95%

99.7%
–1σ +1σ
Nonnormal distributions –2σ–1σ +1σ +2σ
Bimodal distribution Suggests two different populations (eg, –3σ
–2σ +2σ
+3σ

metabolic polymorphism such as fast vs –3σ 68% +3σ

slow acetylators; age at onset of Hodgkin 68%

95%
lymphoma; suicide rate by age).
95%
99.7%
Positive skew Typically, mean > median > mode. Mode
Median
99.7%
Asymmetry with longer tail on right; mean falls Mode
MedianMean
closer to tail. Mean
Mode
Median

Negative skew Typically, mean < median < mode. Mean

Median
Mode

Asymmetry with longer tail on left; mean falls Mean

closer to tail.

FAS1_2022_06-PubHealth.indd 265 11/4/21 12:18 PM

 266 SECTION II Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics

Statistical hypothesis testing

Null hypothesis Also called H0. Hypothesis of no difference or relationship (eg, there is no association between the
disease and the risk factor in the population).
Alternative hypothesis Also called H1. Hypothesis of some difference or relationship (eg, there is some association between
the disease and the risk factor in the population).
P value Probability of obtaining test results at least as extreme as those observed during the test, assuming
that H0 is correct. Commonly accepted as 0.05 (< 5% of results occur due to chance).

Outcomes of statistical hypothesis testing

Correct result Stating that there is an effect or difference when Reality
one exists (H0 rejected in favor of H1). H1 H0
Stating that there is no effect or difference when
none exists (H0 not rejected). Power α
Study rejects H0
( 1 – β) Type I error

Study does not reject H0 β

Type II error

Blue shading = correct result.

Testing errors
Type I error (α) Stating that there is an effect or difference when Also called false-positive error.
none exists (H0 incorrectly rejected in favor of 1st time boy cries wolf, the town believes there
H1). is a wolf, but there is not (false positive).
α is the probability of making a type I error You can never “prove” H1, but you can reject the
(usually 0.05 is chosen). If P < α, then H0 as being very unlikely.
assuming H0 is true, the probability of  α level ( statistical significance level).
obtaining the test results would be less than
the probability of making a type I error. H0 is
therefore rejected as false.
Statistical significance ≠ clinical significance.
Type II error (β) Stating that there is not an effect or difference Also called false-negative error.
when one exists (H0 is not rejected when it is 2nd time boy cries wolf, the town believes there is
in fact false). no wolf, but there is one.
β is the probability of making a type II error. β is If you  sample size, you  power. There is power
related to statistical power (1 – β), which is the in numbers.
probability of rejecting H0 when it is false. Generally, when type I error increases, type II
 power and  β by: error decreases.
ƒ  sample size
ƒ  expected effect size
ƒ  precision of measurement

Statistical vs clinical Statistical significance—defined by the likelihood of study results being due to chance. If there is a
significance high statistical significance, then there is a low probability that the results are due to chance.
Clinical significance—measure of effect on treatment outcomes. An intervention with high clinical
significance is likely to have a large impact on patient outcomes/measures.
Some studies have a very high statistical significance, but the proposed intervention may not have
any clinical impact/significance.

FAS1_2022_06-PubHealth.indd 266 11/4/21 12:19 PM

 Public Health Sciences   PUBLIC HEALTH SCIENCES—Epidemiology and Biostatistics SECTION II 267

Confidence interval Range of values within which the true mean H0 is rejected (and results are significant) when:
of the population is expected to fall, with a ƒ 95% CI for mean difference excludes 0
specified probability. ƒ 95% CI OR or RR excludes 1
CI = 1 – α. The 95% CI (corresponding to ƒ CIs between two groups do not overlap
α = 0.05) is often used. As sample size H0 is not rejected (and results are not significant)
increases, CI narrows. when:
CI for sample mean = x ± Z(SE) ƒ 95% CI for mean difference includes 0
For the 95% CI, Z = 1.96. ƒ 95% CI OR or RR includes 1
For the 99% CI, Z = 2.58. ƒ CIs between two groups do overlap

Meta-analysis A method of statistical analysis that pools summary data (eg, means, RRs) from multiple studies
for a more precise estimate of the size of an effect. Also estimates heterogeneity of effect sizes
between studies.
Improves power, strength of evidence, and generalizability (external validity) of study findings.
Limited by quality of individual studies and bias in study selection.

Common statistical tests

t-test Checks differences between means of 2 groups. Tea is meant for 2.
Example: comparing the mean blood pressure
between men and women.
ANOVA Checks differences between means of 3 or more 3 words: ANalysis Of VAriance.
groups. Example: comparing the mean blood pressure
between members of 3 different ethnic groups.
Fisher’s exact test Checks differences between 2 percentages or Example: comparing the percentage of 20 men
proportions of categorical, nominal outcomes. and 20 women with hypertension.
Use instead of chi-square test with small
populations.
Chi-square (χ²) Checks differences between 2 or more Pronounce chi-tegorical.
percentages or proportions of categorical Example: comparing the proportion of
outcomes (not mean values). members of 3 age groups who have essential
hypertension.

Variables to be compared

Numerical (means) Categorical (proportions)

2 groups ≥ 3 groups Small sample size Large sample size

t-test ANOVA Fisher’s exact text Chi-square test

FAS1_2022_06-PubHealth.indd 267 11/4/21 12:19 PM

 268 SECTION II Public Health Sciences   PUBLIC HEALTH SCIENCES—Ethics

Pearson correlation A measure of the linear correlation between two variables. r is always between −1 and +1. The
coefficient closer the absolute value of r is to 1, the stronger the linear correlation between the 2 variables.
Variance is how much the measured values differ from the average value in a data set.
Positive r value Ž positive correlation (as one variable , the other variable ).
Negative r value Ž negative correlation (as one variable , the other variable ).
Coefficient of determination = r 2 (amount of variance in one variable that can be explained by
variance in another variable).
r = –0.8 r = –0.4 r=0 r = +0.4 r = +0.8

Strong negative Weak negative No correlation Weak positive Strong positive

correlation correlation correlation correlation

` PUBLIC HEALTH SCIENCES—ETHICS

Core ethical principles

Autonomy Obligation to respect patients as individuals (truth-telling, confidentiality), to create conditions
necessary for autonomous choice (informed consent), and to honor their preference in accepting
or not accepting medical care.
Beneficence Physicians have a special ethical (fiduciary) duty to act in the patient’s best interest. May conflict
with autonomy (an informed patient has the right to decide) or what is best for society (eg,
mandatory TB treatment). Traditionally, patient interest supersedes.
Nonmaleficence “Do no harm.” Must be balanced against beneficence; if the benefits outweigh the risks, a patient
may make an informed decision to proceed (most surgeries and medications fall into this
category).
Justice To treat persons fairly and equitably. This does not always imply equally (eg, triage).

Decision-making Physician must determine whether the patient is Capacity is determined by a physician for a
capacity psychologically and legally capable of making specific healthcare-related decision (eg, to
a particular healthcare decision. refuse medical care).
Note that decisions made with capacity cannot Competency is determined by a judge and
be revoked simply if the patient later loses usually refers to more global categories of
capacity. decision-making (eg, legally unable to make
Intellectual disabilities and mental illnesses are any healthcare-related decision).
not exclusion criteria for informed decision- Four major components of decision-making:
making unless the patient’s condition presently ƒ Understanding
impairs their ability to make healthcare ƒ Appreciation
decisions. ƒ Reasoning
ƒ Expressing a choice

FAS1_2022_06-PubHealth.indd 268 11/4/21 12:19 PM