Lecture 1: Introduction

1
Lecture 1 Introduction

• Summer Program
• Brian Healy

2
Welcome!!!

• Important things to know about Boston
• Everyone LOVES the Red Sox
• Very bad drivers
• Easy to get lost
• Good variety of restaurants, but pizza and bagels
  are bad
• Important things to know about Harvard
• Attractive part of campus is in Cambridge
• Food in cafeteria quite good
• Great speakers

3
Who am I?

• Brian Healy
• bhealy_at_hsph.harvard.edu
• Office 412-B (Dave and I share an office)
• Brighton, MA near Cleveland Circle, but am moving
  to Salem, MA
• 5th year graduate student, planning to graduate
  in Oct/Nov
• Just married!!
• No formal office hour, but I am always available
  to answer questions

4
What is this class?

• Designed to help students learn
• Basics of methods including
• Descriptive statistics
• Confidence intervals
• Hypothesis testing
• Regression
• Basics of computing
• If you feel like you know a topic, you are not
  required to attend
• Short HW every 3 classes

5
Syllabus

• Basic outline, but may change depending on how
  quickly we go through topics
• The purpose of the class is to get everyone on
  equal footing at the start of the semester. Any
  and all questions are encouraged. It is never a
  problem to go over a topic a second or third time.

6
Computing background

• Coming into this program I had very limited
  computing background and I have taught myself the
  basics of programming.
• Important to practice
• Google and the internet are great resources
• Other languages

7
General information

• Good book to practice with Rosner, Fundamentals
  of Biostatistics
• For help with R, there are great web resources.
  Books about S-plus are more plentiful and
  information very applicable to R
• If you have any feedback on the course, please
  tell me so that I can make the course as useful
  as possible for all. The only caveat is that I
  will not be able to go too fast because this is
  mostly to ensure everyone has a certain
  background, not as enrichment for people with
  lots of stats

8
Question

• What appeals to you about biostatistics?
• Is there any particular area that you find most
  interesting?

9
What is the difference between statistics and
biostatistics?

• My opinion Statistics is more driven by
  interesting statistical questions, while
  biostatistics is more driven by interesting real
  data analysis problems.
• There is a lot of overlap, but the spectrum looks
  like

Epidemiology
Biostatistics
Statistics
Mathematics
10
Topics for biostatisticians

• Applications driven methods
• Survival analysis
• Multiple testing methods (genetics)
• Causal inference
• Risk / Decision Science
• Computational biology / Bioinformatics

11
What are biostatisticians trying to do?

• We want to answer questions about the population
  by using a sample
• In more technical language, we want to make
  inferences about a population characteristic from
  a sample

12
Clinical study

• Study design
• Hypothesis formulation- Does the treatment have
  an effect?
• Sample selection- Who are we going to study?
• Collection of data
• Mid-study markers- Should we stop the study
  early?
• Analysis of data
• Results- Was there any effect?
• Conclusions- What does this all mean?
• Generalizability

13
What does the statistician do?

• Study design
• Hypothesis formulation- Define outcome and
  exposure
• Sample selection- Sample size, type of sampling
• Collection of data
• Mid-study markers- Toxicity or effect
• Analysis of data
• Results- Estimation and confidence intervals
• Conclusion- Significance of effect or association
• Causality vs. association

14
Hypothesis formulation

• The most important step of any problem, but the
  one that gets the least attention in statistics
  courses.
• What do we want to know?
• The more specific the better because statistics
  is designed to answer very specific questions.
• Ex. Is Treatment A better than Treatment B?
• Does Treatment A lead to an larger decrease in
  the size of a tumor than Treatment B?
• We will discuss hypothesis formulation in every
  class

15
Sample selection

• Exposure and control group (Lecture 4)
• What group of people are we most interested in?
• Can we generalize our conclusions to other cancer
  patients? all men? all people?
• Sample size (Lecture 9)
• How many patients are needed to prove what we
  want to show?

16
Mid-study markers

• May use many of the same techniques we will
  discuss in this course to determine how the
  treatment and placebo group are performing
• Most commonly used in long clinical trials and
  these are written into the protocol
• Ex. For HIV trials, it is unethical to keep
  patients on a placebo treatment once it can be
  shown that the treatment is working, even if the
  study has not finished.

17
Results

• The main results from a study are provided in
  graphs and tables (Lecture 2)
• Every public health paper has a first table which
  describes the demographics of the study
  population
• Summary statistics and confidence intervals for
  the effect measures (Lecture 6-7)
• What is the best analysis for this data?

18
Significance of results

• Descriptive studies / Comparative studies
• In the latter, we want to know if there is a
  difference between groups, especially based on an
  exposure of interest. To do this, we use
  hypothesis testing (Lecture 6). The specific type
  of test depends on the question of interest.
• Have the assumptions of the analysis been met?

19
Generalizability

• Assume that we have found a difference between
  our exposure and control group, what does this
  mean for the general population? Specifically, to
  which group of people can we apply our results?
• This is often based on how the sample was
  originally collected.
• Ex. If exposed group were school children living
  near power lines and the control group were
  school children living elsewhere, can we
  generalize the findings to all children? adults?

20
Example

• We are interested in determining if the men in
  Boston are shorter than average for the rest of
  the United States (it sure seems that way
  according to all of my female friends and wife)
• How can we determine this?

21
Study

• Hypothesis
• Sample
• Mid study marker
• Results
• Significance of results

22
Aside

• Throughout this course, we will go through many
  examples from consulting I have done during my
  time here. I think working with physicians and
  other researchers is extreme valuable because you
  can apply the concepts you learn. Often the
  statistical techniques appear very easy to use
  because the question of interest and the data
  have been given to you in a form that are easy to
  deal with, but this is not how you get the data
  in practice.
• If there is any medical topic or research area
  that you find interesting, I will try to include
  this in some of my examples.

23
Association vs Causation

• Association
• Definition two factors are related, ex. height
  and weight
• There is no implied cause and effect relationship
• Causation
• Definition one factor causes a second factor,
  either directly or indirectly, ex. laying in the
  sun causes sun burn
• The causal factor is required for the outcome to
  occur

24

• It is much easier to demonstrate an association,
  but for public health, it is much more
  interesting to demonstrate causation
• For example, the fact that eating red meat is
  associated with heart disease may show an
  important group to direct interventions towards,
  but we do not know if there is no causal
  relationship. Therefore, it is not clear if
  changing this behavior would change heart disease
  status.

25

• The majority of methods we describe in this
  course are used to demonstrate an association
  between two factors because this is much easier
  to show.
• The important thing to ask yourself when you read
  a study or complete an analysis
• What do these results mean?
• Are the two factors associated or causally
  related?
• What action can be taken based on this?
• There has been significant research into causal
  inference (some of the best in the world happens
  here).

26
Counterfactual

• One common way to look at a causal effect is to
  think about counterfactuals
• Def What would the outcome be if, contrary to
  fact, the exposure was different?
• Ex. We know that a patient who ate red meat got
  heart disease. What would have happened if the
  patient lived exactly the same life, except
  he/she did not eat red meat?
• The causal effect is the difference in the
  counterfactual outcomes, CE Yred meat Yno red
  meat

27

• The causal effect is the thing we are trying to
  find because this shows which exposures affect
  (directly or indirectly) the outcome
• Of course, both counterfactual outcomes cannot be
  observed for each patient, so we have to find
  clever ways to estimate this effect. The type of
  study we have will determine how to estimate this
  causal effect
• When we cannot estimate the causal effect, we can
  describe the association between exposure and
  outcome

28
Difference between this course and the
probability course

• This course
• Focus on analysis of data without too much theory
• Applications
• Better jokes
• Probability course
• Focus on theoretical background for most of the
  data analysis techniques we describe
• Theory

29
Inference vs Methods

• This distinction between data analysis and theory
  holds for the academic year as well
• I think the relationship between these courses
  can sometimes missed in all of the details during
  the semester, but Dave and I will try to make
  connections where possible this summer to get you
  in the mind set that these are separate, but very
  related areas

30
Computing

• Computing resources, SAS, R, etc., have made
  completing analyses to determine the significance
  of results and confidence intervals much simpler.
• Although software allows analyses to be complete
  faster, they do not imply that the correct
  analyses have been done. This is why framing the
  problem is the most important step of the
  analysis.
• R- focus of the courses computing (Lecture 2)
• SAS- secondary package (Lecture 10)