Exam Study Designs and Final Exam Notes

Study Designs Review

• This video provides a comprehensive review of study designs and information about the final exam.
• It is recommended to take notes while watching.

Plan of Attack

• Cover all study designs learned since the midterm.
• Go over information about the note card for the final exam.
• Reminder about the online course evaluation.
• Week 10 in-class session will be an open lab for review problems and questions.

Review of Study Designs

• Consider a scenario:
  • A study began in 1965, recruiting 3,000 adults in Baltimore and asking about their alcohol consumption.
  • The group was assessed to see who was diagnosed with cancer between 1981 and 1995.
  • What study design was used?

Definitions from Gordis

• Prospective cohort study
• Retrospective cohort study
• Cross-sectional study
• Randomized trials
• Case-control studies

Cohort Studies

• Recruit a sample of individuals with or without an exposure (etiologic factor).
• Follow up over time to compare the incidence of a health outcome between exposed and unexposed groups.
• Ensure that nobody has the outcome at the start of the assessment to watch new cases develop.

Prospective Cohort Study

• Recruit the sample and assess for exposure at the time of assessment or shortly thereafter.
• Move forward in time with participants to see if an outcome occurs.

Retrospective Cohort Study

• Start by recruiting individuals based on them being part of a pre-identified group from the past.
• The initial group will have all been held together by belonging to some form of group.

Cross-Sectional Study

• Complete an assessment of a population at a single point in time.
• Assess both exposure status and outcome status simultaneously.
• Often uses surveys, interviews, or biological measures.
• When the sample is recruited, the researcher doesn't know about participant exposure or outcome.

Randomized Trial

• A form of experimental design.
• Researchers have a defined population for whom a particular intervention could be useful.
• Randomize participants to either be in a treatment group or a control group.
• Assess participants for the outcome, then administer an intervention or a placebo, and then reassess at a later time point post intervention to see if there has been change outcome measure.
• The analysis looks at that change.

Case-Control Studies

• Identify and recruit a group of individuals with an outcome (cases).
• Recruit a group of individuals without the outcome (controls).
• Look back in time to assess if a particular exposure occurred.
• Compare the odds of exposure in cases to the odds of exposure in controls.

Applying Study Designs to Scenarios

Scenario 1

• A study began in 1965 with 3,000 adults in Baltimore, asking about alcohol consumption.
• Cancer occurrence was tracked between 1981 and 1995.
• What was the exposure, and what was the outcome?
  • Alcohol use leads to cancer.
• Who did the researchers recruit at the start of the study (1965)?
  • A general population of 3,000 adults in Baltimore.
  • Exposure info (alcohol consumption) was collected at the start of the study.
• Study Design:
  • Prospective cohort study.

Scenario 2

• Physical examination records of an entire incoming freshman class of 1935 at the University of Minnesota were examined in 1987 to see if they recorded height and weight at the time of admission to the university was related to the development of coronary heart disease by 1986.
• Study Design:
  • Retrospective cohort study.

Scenario 3 - Team Approaches to Studying Children's Reading and SAT Scores

• Association between children aged two-seven being read to regularly and having those same children score 550 or higher on the English SAT in high school.
• Exposure: Being read to regularly.
• Outcome: High SAT English score in high school.

Team 1

• Randomly selected and recruited 500 parents of graduating seniors for a survey.
• Survey questions: Frequency of reading to students and academic progress (including SAT scores).
• Study Design:
  • Cross-sectional research design. (Data collected at one point in time.)

Team 2

• Contacted researchers from a prior prospective study of children born in 2007 about early childhood education.
• The prior study included data on how much parents read to their children between the ages of one and eight.
• Team two researchers were able to get a participant list of all the parents who participated in this prior study, and then they called the parents and the students from the prior study and asked them to participate in this new study being conducted by Team two. It was amazing!
• The entire group said yes and agreed to participate.
• Team two researchers then looked at the old data from the prior study to get their reading information, in terms of how much the children read were read to, and then they looked up until today to see who has scored well on the SAT test.
• Study Design:
  • Retrospective cohort design.
  • Recruited based on membership in a past research study.
  • Exposure data from the prior study; moving forward in time to collect information about SAT score outcomes.

Team 3

• Recruited 200 students who scored well on the English SAT and 400 students who scored below the high-level mark.
• Interviewed parents about how much time they spent reading to their kids when their kids were young.
• Study Design:
  • Case-control study.
  • Recruitment based on the outcome (SAT score).

Interpreting Data from Randomized Clinical Trials

• A table presents outcome measures at different times, with the exposure variable as rows (treatment group vs. placebo group).
• The outcome generally has a baseline measurement and an end-of-treatment measurement.
• The analysis asks the question: Was the change from baseline to end of treatment different in the experimental group compared to the control group?
• Example: Fear of Epidemiology score as the outcome.
  • Experimental group: Decrease of 9 points.
  • Placebo group: Decrease of 5 points.
• Need to determine if these two numbers are statistically different by looking at the p-value.
  • For the p value to be statistically significant at a level where there would only be a 5% chance or less of making a type I error when we were saying the groups differ, we would need for that p value to be less than or equal to 0.05.
• Example: If p-value is greater than 0.05, you cannot say that the epiphobia drug group had a greater change from the placebo group because the p value was larger than 0.05.

Error, P-Value, Power, and Sample Size in Randomized Trials

*If the epiphobia group had a decrease of 9 where the other group had a decrease of only 5, then we might say, Wow, it looks like the two groups really should be different, but in this situation, we did not have enough power to detect a real difference.

• If the p-value is 0.06, meaning that there was a 6\% chance of making an error if we called the groups differ, the researchers would not conclude that the groups are different per their data.
• Power: The ability to detect a difference that actually exists.
  • Equation: Power = 1 - \beta
• Relationship:
  • Larger sample size decreases the chances of making either type of error.
  • Decreasing the chances of making a type one error will also have a smaller p value.
  • As sample size increases, the p-value generally decreases.
  • A smaller p-value indicates the groups differ, and power increases with study size.

Final Assessment Overview

• The final assessment is on the last day of class, in the same room, time, and place.
• Like the midterm, it will be made up of multiple choice questions, and I will add in a few extra credit opportunities as well.
• The exam will cover class from the midpoint, or mid midterm forward
• Bring a 4x6 note card with notes, both front and back, and a calculator.
  • The note card can include a study design grid.

Week-by-Week Review

Randomized Trials

• Key feature: Individuals are randomized into their study group.

Sub-Designs

• Randomized Crossover Trial: Individuals are randomly assigned into the experimental group or the control group at the beginning of the study, and then halfway through the study, their exposure condition is swapped so that those who were receiving the placebo now get the actual experiment and those who were receiving their experimental condition get swapped into the control condition.
• Masking/Blinding:
  • Single-blind: Participants don't know if they are receiving the treatment or placebo, but the researcher does.
  • Double-blind: Neither the participant nor the assessor knows the study condition.
• Analysis: Change in outcome from pre- to post-intervention.
  • Did the experimental group change more than the control group?

Error, P-Value, Power, and Sample Size Revisited

• Type I Error (Alpha): Researchers conclude that groups differ when they do not.
  • P-value: Provides information about the probability of making a Type I error.
  • Example: A p-value of 0.05 means there is a 5% chance of making an alpha error.
• Type II Error (Beta): Researchers conclude that groups do not differ when they actually do.
• Power: ability to detect a difference when one truly exists out there in the world. Equation: 1 - \beta

Cohort Studies

• Key feature: Assess exposure at one point in time and look forward to a different point in time to assess if an outcome has occurred.
• Temporality: Ensuring that exposure occurs before the outcome.
• Calculations: Since we can find the incidence of the outcome among different groups, like the exposed group, the unexposed group, and the total population, then there are several different types of analyses that we can compute. We can calculate different measures of risk. don't do that because the relative risk is considered a stronger statistical measure than an odds ratio risk.
  • Absolute Risk: the incidence of the outcome in your exposed group.
  • Relative Risk.
  • Attributable Risk.
  • Population Attributable Risk.
• Cohort Study Design:
  *Prospective Design: The researchers are defining and identifying a group of people, currently and are sorting them based on exposure and then following them forward in time so they're moving forward in time with the research participants from exposure to outcome.
  *Retrospective Design: Researchers are identifying that group of people from the past or some cohort from the past, and that could either be a group of individuals that, for example, all graduated at the same time or were admitted into college at the same time or babies all born at a particular hospital in a particular year, etc. Any of those could be considered retrospective cohorts.

Case-Control Studies

• Used to study rare outcomes.
• Recruit people based on their outcome.
• Start with the outcome (cases) and recruit people without the outcome (controls).
• Look back in time to see who did and did not experience some exposure.
• Control Selection: Controls should be representative of people with the outcome and similar to the general population.
  • Methods: Controls from the same hospital or neighborhood; friends or family.
• Matching: Recruiting controls who are similar to cases regarding age, sex, or socioeconomic status.
  • By matching, we what what this does is it makes it so that the cases and the controls look the exact same in relation to the variables that we matched on. So those characteristics can no longer explain any differences that we see between the cases and the controls.
• Analysis: Odds Ratio.

Sub-designs Options

*These generally occur when researchers have developed a big cohort study with some other focus on other variables of interest, but they also happen to be collecting data that researchers later deem useful for a case control study, and those two sub designs differ based on the controls that are selected and when the controls are selected from the initial pool of participants from the cohort study to serve as controls in the case the nested case control design. All right, and a final study design type that I want to review here is the cross sectional design.

Cross-Sectional Designs

• Collect data from a group of people without recruiting based on exposure or outcome.
• Gather data at one time point.
• At that time people time point, people either will or will not have experienced the exposure, and people either will or will not have experienced the outcome.
• Data can be gathered anonymously.
• More cost-effective.
• Analysis: Odds Ratio.

Evidence-Based Public Health

• Use of data as evidence to inform prevention, interventions, regulations, public policies, and laws.
• Not all data is created equal.
• Individuals should be their own judge regarding evidence.

Final Note Card

• Use the provided grid as a portion of your notes.
• Notes must fit on a 4x6 note card.
• Include other information or examples of equations on the back.
• Review questions at the end of study design chapters in the textbook provide excellent examples.

Two By Two Tables

• Show all two-by-two tables in calculations for partial credit.
• Used for:
  • Prevalence.
  • Incidence.
  • Incidence in the exposed group (Absolute Risk).
  • Relative Risk.
  • Attributable Risk.
  • Population Attributable Risk.
  • Odds Ratio.
  • Sensitivity, specificity, and predictive values are not included in the final exam, since this material was covered in the first half of the course.

Additional Tips for the Test

*Please show me all of your two by two tables.

• Know your equations and how they relate to the two by two tables.
• Draw two by two tables with complete examples of the calculation on the back of your note card.
• Know what it means to have, in terms of interpretation for your relative risk and your odds ratios, if they are equal to one, less than one, or greater than one.
• Do look back at your interpretation of relative risk and odds ratios.
• Review questions at the end of study design chapters in the textbook provide excellent examples.

Final Exam Scenarios

Scenario 1: Measles and Vaccination

• In February, the US declared that measles had been eliminated. However, in 2014, there were six forty four cases of measles in The US. Let's say that there were a number of anti vaccine proponents who were claiming that the vaccine is not effective.
• A researcher wants to analyze if individuals with measles were less likely to have been vaccinated than those without the measles.
  *What design could you use for this particular scenario?
  *Restriction: There's only one design that we could use to be that that would be appropriate for this scenario.

Scenario 2: Scored Wristbands

• Fifth-grade students were given the opportunity to wear scored wristbands (activity trackers).
• Researchers want to know if kids who wear the band are more likely to average 60 minutes of recommended daily exercise than kids who do not wear the band.