Only $35.99/year

Terms in this set (63)

Toxin=substance naturally produced by a living organism
Toxicant=any toxic substance, human made and put into the environment, example: pesticides. Toxicity=Ability to harm human health or environment. One of the criteria that is used to determine whether waste is hazardous waste.
Contaminant=substance that enters a system (the environment, human body, food, etc.) where it is not normally found. Usually used in the negative sense in environmental health and synonymous with pollutant
Background concentration=baseline level of toxicant or pollutant that all are exposed to in a given area
Dose response - relationship between exposure/dose of stressor and adverse health effect
Safe/free from risk vs defining the level of risk where should be concerned vs take action
Adverse health effects may be acute/delayed, clinical or subclinical, and reversible or irreversible
Maximum contaminant levels - EPA/state/local regulations determine maximum levels based mainly on health risks and then on non-health concerns (e.g. odor/taste of water). May incorporate ecological risk.
Measured via parts per million - a concentration of substance in air, water, or soil (1 ppm = 1 mg/kg ~ 1 drop of dye in 18 gallons of water)
Particulate matter (PM) - dust, dirt, soot, smoke, drops of liquid, etc. from wood stoves, forest fires, construction sites, coal fires, factory emissions, cars/trucks, power plants
PM10 (particulate matter 10 micrometers wide) is small enough to irritate mucous membranes, while PM2.5 (2.5 micrometers wide) is small enough to enter deep into lungs

Risk Assessment:
1.Hazard Identification
2.Dose-Response Assessment - can be linear or hyperbolic
3.Exposure Assessment - ingestion, inhalation, absorption
4.Risk Characterization
About 90 people a day die from gun violence in the US
About 33,000 people a year
Another about 85,000 people a year are injured
It affects people of all ages, races/ethnicities, classes. 1 in 3 people know someone who has been shot.
Unintentional injury (which includes by firearms) is the #1 cause of death in ages 1-44, is #3 in 44-64 range, and #4 across all ages.
Homicide - Most homicides involve less than 4 people killed (definition of a mass shooting) but there is a recent rise in numbers of mass shootings (tripled since 2011)
In 2015, there was 330 mass shootings killing 367 people, injuring 1,317
Per year there are:
21,058 gun deaths due to suicide
11,726 gun deaths due to homicide
546 gun deaths due to accidents
269 undetermined
Suicide - Though most suicide attempts are by drug overdose, guns are responsible for about half of all deaths by suicide because gunshots are much more lethal. Groups at risk for suicide attempts include:
American Indians/Alaskan Natives, people bereaved by suicide, justice/child welfare settings, non-suicidal self-injury, previous attempts at suicide, w/ medical conditions/mental/substance use disorders, LGBT, military/veterans, midlife/older men
statewide gun ownership correlated with higher suicide rate
time from deciding to commit suicide to the attempt is less than an hour in 70% of people. An impulse decision.
Children - many deaths/injuries self-inflicted, other children as shooters; more rarely older teens/parents
73% of children under 10 know where guns are kept at home, 36% of them handled by children contrary to parents' reports
Intimate Partner Violence - homicide by intimate partner accounts for about half the deaths of women by homicide (risk of death increases greatly if a gun is present)
financial cost - $229 billion in 2015, $700 per gun, from work loss, medical/mental health care, decreased quality of life, insurance/police/criminal justice, etc.

US has more firearms and more gun deaths than comparable nations
Firearm homicide rate is 25x higher
Firearm suicide rate is 8x higher
Note that physician gag rules have been proposed in other states (FL -- struck down in Feb 2017, MO, MT), but not in MA.
Florida Privacy of Firearm Owners Act of 2011: barred physicians from discussing guns with their patients
Especially important in those most at risk—with young children at home, teens, patient or family member at risk of suicide, history of violence or risk of altered mental state (drugs, dementia)
Remember that most gun owners are knowledgeable and committed to gun safety already
Focus on health—what we are experts at—and ways to reduce risk—a mutual goal
Provide context—include screening for firearms in routine child safety screening/anticipatory guidance for accident prevention
Open-ended: "Do you have any concerns about the accessibility of your gun?" vs "Is your gun safely secured?"
Avoid accusatory statements: "Some of my patients have guns at home, and some gun owners with suicidal thoughts choose to make their guns less accessible. Are you interested in talking about that?" vs "Do you own a gun?"
Brainstorm harm reduction with patients: many ways to make a gun less accessible
Safe storage-locked, trigger locks, away from ammunition, temporarily with someone else, buy-backs, deactivate, loading indicator, new technology specific to owner
Gunshot wounds are required to be reported
Providers need to complete a form for the Weapon-Related Injury Surveillance System or contact the state police's Criminal Information Division and local police
Abuse reporting
Child abuse or neglect
Elder abuse
Abuse of persons with disabilities
Mental health
Suicidal or homicidal patients can be committed
HIPAA allows health care providers to report information to any person, including family members and law enforcement officials, who would be reasonably able to prevent or lessen a threat if they have a good faith belief that there is a serious and imminent threat (to the patient or another person)
Licensed mental health providers have a special duty to warn or take steps to protect a patient's potential victim in certain circumstances
Two groups, one with the outcome/disease of interest and one without (controls).
Controls must be taken from the same population as the cases, but should not be chosen for their exposure
For example, if testing whether exposure to firecrackers while pregnant lead to VSD in babies, choose a group of babies with VSD and one without VSD (regardless of firecracker exposure) from the same hospital.
Usually use this method when the disease/outcome is rare.
Potential for bias in selection of controls
A group of people without the outcome are selected and followed over time. If they do get the outcome they are compared with those that didn't.
Prospective or retrospective
Eg, measuring social integration in group of men and followed through to see who became suicidal and why (prospective)
Eg, Looking at medical records for patients who took vitamin D and those who did not, then investigating whether they later developed obesity (retrospective)
Potential issues: no randomization and need to account for confounding factors
Provide information on risk factors and health outcomes at one point in time
Eg: Are patients with public insurance more likely to have schizophrenia than patients on private insurance? Measure this using national health survey and compare percentages of people on private vs public insurance with schizophrenia
Note this is a correlation not a causation
observational study in which at least one variable is measured at the group level
Eg, take a group of patients with prostate cancer and plot their sugar consumption vs their mortality and find the line of best fit
Note this is an association not a causation, temporal relationships unclear and limitations of existing databases
Communicating Risk Methods
Framing - be aware of how you frame interventions and screenings
Interventions are seen as more beneficial if positive framing is used e.g. saying "there is a 91% of survival at 5 years with treatment" (positive) versus "there is a 9% chance of dying without treatment" (negative)
Screening is viewed as more effective if you use a loss format e.g. "screening for colorectal cancer will improve your survival" (gain) vs "not getting screened for colorectal cancer will reduce your chance of survival" (loss)
Presenting risk reduction - relative risk ratio and absolute risk ratios are better understood than number needed to treat, RRR is seen as "larger" and more persuasive, and ARR is seen as indicating a larger effect than when expressed as an NNT but is not more persuasive. The National Cancer Institute recommends presenting data using absolute risk.
Personalizing risk reduction - using risk tools/online calculators such as the breast cancer risk tool on's website can allow patients to key in their information and get more accurate sense of risk
Natural frequencies e.g. "4 in 100"
Words not numbers - plain words work better. Be aware that for different people, terms like "common" and "rare" are associated with different frequencies (e.g. though pancreatitis is a side effect of a medication in 0.04% of patients, 18% of patients taking the medication thought that they might experience this "rare" side effect (versus 2.1% told there was a "4 in 10,000 chance").
Decision Aids - risk pictographs and other visuals can help patients
Accuracy and precision
Accuracy (aka validity)
How close is the measurement to the true value?
affected by systematic errors
Precision (aka reliability)
How close are the measurements to each other?
Related to square root of the sample size
So to double the precision of an estimate, you must quadruple the sample size (sample size * 4)
affected by random error

Null and alternative hypotheses
In statistics we are employing hypothesis testing
The null hypothesis (H0) is always that there is no difference between or among the groups being compared
The alternative hypothesis (H1 or Ha) is always that there is a difference between or among the groups being compared
Always start with the assumption that the null is true. We want to reject the null hypothesis because we are seeking a difference between the compared groups.

P values
The p value is the probability of getting a difference as big as, or bigger than, what we observed if the null hypothesis were true. It is a measure of the study sample, not the population. The smaller the p-value, the more evidence there is against the null hypothesis.
Indicates whether a result is statistically significant or not; by convention, the threshold is <0.05 (if p≥0.05, not statistically significant)
It DOES NOT mean that the true population value is 95% likely to be the same as what we found in our sample, nor does it mean that there is a 5% likelihood that the null hypothesis is true.
Influenced by:
Sample size (larger sample size is more likely to give a significant p value)
Random chance
Effect size (larger effect size is more likely to give a significant p value)
Results can be statistically significant without being clinically significant

Confidence intervals
If we repeated the experiment 100 times in the same population, 95% of the time the result we would get would be within the 95% confidence interval
A larger sample size yields a tighter confidence interval (note the inverse relationship between confidence interval and n)

A confidence interval that excludes the null implies statistical significance, or a p value <0.05. If the confidence interval crosses the null value (e.g. if RR = 1.5 and the 95% confidence interval is 0.9-2.1), then generally p≥0.05.
Type I and Type II error
Type I error
Incorrectly reject the null hypothesis
Probability of type I error is Alpha
Type II error
Fail to reject the null hypothesis when you should have rejected the null
Probability of type II error is Beta

T tests
Used to compare means
e.g. if Group A = Japanese-made cars, Group B = US-made cars, what is the difference in the mean number of miles per gallon between Group A and Group B?
H0: There is no difference in the mean MPG between the groups; H1: There is a difference in the mean MPG between the groups
Two sample t test = unrelated samples
Paired t test = related samples (ie before/after data)
Larger t = more likely to be significant
Significance of the test is influenced by:
Sample size (larger n = larger t)
Difference between the means (larger difference = larger t)
Standard deviation (smaller SD = larger t)

ANOVA (analysis of variance)
Compares the means among more than 2 groups
e.g. Compare the mean pain rating (among five patients) for six different brands of acetaminophen
H0: none of the sample means differ from any of the others, H1: at least one sample mean differs from at least one of the others
Avoids problem of multiple comparisons
H0 is that none of the means differ from each other, H1 is that at least one mean differs from at least one other mean (doesn't tell you which one(s)!)

Chi square (Χ2)
Compares proportions between or among groups
Can be used for two or more groups
e.g. association of pregnancy complications with exposure to video terminals (two groups are exposed vs unexposed, comparing their pregnancy complication proportions)
H0: there is no association between exposure to video terminals and pregnancy complications, H1: there is an association between exposure to video terminals and pregnancy complications
As with ANOVA, H0 is that none of the proportions differ from each other, H1 is that at least one proportion differs from at least one other proportion (doesn't tell you which one(s)!)
Linear regression
Used to predict a continuous outcome based on one or more predictor variables
Predictors can be continuous or categorical
For simple regression, the regression line is the "best fit line" through the points; the slope is β and the y-intercept is α. If you plug in a known α and β, you can get an estimate of the outcome variable.

Outcome variable = α + β(predictor variable)
Where α is the constant and β is the regression coefficient
P values for the regression coefficient(s) tell you if that predictor is significantly associated with the outcome
Sign of the regression coefficient tells you if the association is positive or negative
Multiple linear regression is used when you have multiple predictor variables
Outcome = α + β1(Predictor 1) + β2(Predictor 2) + β3(Predictor 3)...
Can be used to control for confounding - the p value for the β coefficient tells you if that predictor contributes significantly to the outcome

Determine what factors are associated with birth weight. Looking at p-values: infant's sex and gestational age are associated. Both associations are positive (i.e. male infants weigh more vs female, and higher gestational age infants weight more)

Logistic regression
Used to predict a dichotomous outcome based on one or more predictor variables (the predictor variables can be categorical or continuous and the outcome variable is binary/categorical)

Gives odds of an event happening based on the values of the predictor variable(s)
Log odds of the outcome = α + β(Predictor variable)
log is natural log (ln)
Exponentiate the log odds (raise elog odds) to get the odds
Convert the odds into probability with the formula probability = odds/(1+odds)
α is the constant and β is the coefficient
β tells you the odds ratio associated with that predictor variable, controlling for all other predictors
Multiple logistic regression is used when you have multiple predictor variables
Log odds of outcome = α + β1(Predictor 1) + β2(Predictor 2) + β3(Predictor 3)...
Can be used to control for confounding and to give the odds ratios associated with different predictor variables and the outcome. The p value for the beta coefficient (which is the log odds ratio, so the paper will report an odds ratio) tells you if the relationship is significant.
Case control studies: Choose two groups of people: One with the outcome, one without. Assess their exposure retrospectively and compare the exposure between the two groups. Particularly useful when the outcome is rare because using an observational study for a rare outcome would require following a prohibitively large number of people. Controls should be selected from the same population that gave rise to the cases. Measure of interest = odds ratio
e.g. Exposure to firecrackers while pregnant associated with ventricular septal defects in babies - children with VSD in pediatric cardiology clinic VS children without VSD in general pediatrics at same hospital
Cohort studies: Find a group of people who have not had the outcome of interest and follow them over time. Compare those who get the disease with those who do not. Measure of interest = risk ratio, rate ratio, risk difference, or rate difference
can be prospective or retrospective
e.g. in a group of men, look at high social integration (exposure) vs low social integration and suicide vs not (outcome) over 24 years
Cross sectional studies: Assess exposure and outcome at the same time and only look at one point in time. Inexpensive, useful for hypothesis generation, not useful for determining cause-effect relationships. Measure of interest = prevalence ratio
e.g. are patient with public insurance more likely to have schizophrenia than those with private insurance? (exposure = insurance type, outcome = schizophrenia)
Ecological studies: Measure an outcome, exposure, or both on a group level. May be useful for hypothesis generation when combined with evidence from other studies, cannot determine cause-effect relationships. Measure of interest = correlation
e.g, take a group of patients with prostate cancer and plot their sugar consumption vs their mortality and find the line of best fit
Investigators manipulate exposure in some way
RCTs are "gold standard"
Pros of experimental studies
No need to control for confounders - they are taken care of by randomization
Cons of experimental studies
Must have equipoise to do them ethically
Often very resource intensive and difficult to perform
Key concepts in experimental studies
Quasi-experimental studies: Investigators manipulate exposure but it is not randomly assigned. Outcomes are measured and calculated just like RCTs.
e.g. police officer-led curriculum for middle school students meant to prevent gang involvement. Measured how the intervention affected future victimization (being a victim of a crime). Classrooms that have intervention vs not were not randomly assigned.
Important to compare groups at baseline.
Randomized clinical trials: Patients are assigned randomly to one exposure or another
Randomized controlled trials: A type of randomized clinical trial with a control group (placebo or no treatment); patients are assigned to an exposure or a placebo
Crossover studies: Each individual is his/her own control; they are randomly assigned to treatment A or treatment B, then switch to the other treatment after some time
Blinding: Participants and/or investigators are unaware of what group they've been assigned to. Not always possible. Ideally experiments are double blinded, meaning both the researchers who are assessing the outcome and the participants are unaware of their treatment assignment. Minimally, researchers assessing outcomes should be blinded to prevent observer bias (misclassification of outcomes consciously or unconsciously).
Intent-to-treat: All RCTs should report ITT analyses, in which participants are analyzed according to the treatment to which they were assigned, rather than the treatment they actually got.
Per protocol: Reports of RCTs may describe associations based on what treatment the groups actually got (rather than the treatment to which they were assigned); however the benefit of randomization is lost and such analyses should be considered observational studies
Competing risk: an event that removes a subject from being at risk for the outcome of interest (ie. death)
Cumulative incidence: (aka incidence proportion, risk, attack rate)

Number of new cases in a time period/number of people in population at risk at that time. Remember that the denominator is number of people at risk of the outcome; if someone can have an outcome only once, they are then out of the risk pool when incidence is determined at later time points. The problem with incidence (as opposed to an incidence rate) is that you can't account for competing risks (e.g. if you are measuring cancer incidence, but some people die of heart disease - heart disease is a competing risk).

It is the number of individuals who developed the outcome in the study periodNumber at risk - do not get confused with incidence rate (see below).

Person-time: #people*timetoperiod
(time can be in days, months, years, etc.)

allows people to be removed if they die or move away, but still count the time they were there

Person-time: When a cohort is followed over time, each person contributes one person-year for each year that they are in the study and eligible to "get" the disease or outcome. People stop contributing person years if they: get the outcome of interest, die, move away, or become ineligible for some other reason

Incidence rate (aka incidence density or hazard rate): # incident cases in a specified period of time t0 to t1sum of person-time at risk accumulated to t0 to t1
Number of incident cases in a specified period of time/ sum of person-time at risk in that same period. We need to know WHEN the incidence occurred.
Internal validity: The study measures what it's supposed to - it is valid if there aren't any systematic errors. RCTs tend to have more internal validity, but patients tend to be less like the general population.

External validity: The extent to which the study is applicable to other populations (ie. can the results be extrapolated to my patients?)
-RCTs tend to have more internal validity, but patients tend to be less like the general population

Confounding: The relationship between exposure and outcome is affected by a third variable that is related to both the exposure and the outcome but not in the causal pathway. It may make it appear that there is a relationship when there is not, or vice versa

A confounder is a variable that is associated with both the exposure and outcome for example, birth order and Down syndrome may appear to be related, but it can be explained by the confounder of maternal age.

Confounders can be prevented via randomization, or matching in case-control studies (i.e. finding controls with similar age, neighborhood, race to the cases). Controlling using regression methods and stratified analysis can help to deal with confounders during analysis.
e.g. in studying alcoholism and lung cancer, note that smoking is more prevalent among alcoholics than nonalcoholics, and smoking is a cause of cancer. So exposure=alcoholism, outcome=lung cancer, confounder=smoking. Separate out results by smokers vs nonsmokers.

Effect modification: The exposure has a different impact in different circumstances. Not a bias or something you can control for. Should be described when presenting the data.
e.g. exposure = peri-operative beta-blocker therapy, outcome=mortality; however, if you look at people's cardiac risk index, you will notice that mortality varies across cardiac risk groups.
Proposed framework to infer causation - Gordis 2004 (you do NOT need to memorize this list, however you should know the concepts for each one)
Temporal relationship
Strength of association
Dose-response relationship
Replication of findings
Biologic plausibility
Consideration of alternative explanations
Cessation of exposure
Specificity of association
Consistency with other knowledge

Temporal relationship: Does the effect come after the cause?

Strength of association: How high is the risk ratio/odds ratio? If it's 10, there is a stronger argument than if it's 1.2

Dose-response relationship: Does more exposure lead to increased likelihood of the outcome? (Note that in some cases, the absence or slight presence of the factor can trigger the outcome OR greater exposure leads to lower likelihood of the outcome.)

Replication of findings: Have the same results been found in multiple studies done in multiple ways?

Biologic plausibility: Is there a biological mechanism to explain the association?
Example with lung cancer:

Consideration of alternative explanations: Does the association remain when confounders are adjusted for?

Cessation of exposure: Does removing exposure decrease risk of disease?

Specificity of association: Is one exposure associated with one disease? (Not always applicable though, since many diseases have multiple causes.)

Consistency with other knowledge: Is the finding consistent not just across epidemiologic study types but also across biological studies like in vitro and animal studies?