P8130: Biostatistical Methods I

Fall 2021
Instructor: Vahe Khachadourian, MD, MPH, PhD
Outline
• Introduction
• Syllabus
• Structure of the course
• Principle of Biostatistics
• Type of studies
• Type of variables

2
Fall 2021 - Hybrid Course
• This hybrid course is structured into modules
• Each module contains:
• Recorded lectures – you need to listen to these prior to live sessions!
• Required readings – textbook chapters
• Learning checks – easy practice problems

• Interactive Zoom sessions will be offered live twice a week:

• Quick recap of the lectures
• R code review and implementation
• Open for YOUR questions
3
Fall 2021 - Hybrid Course
• Course info can be found in the syllabus (see Canvas)

• Main expectations:
• Class engagement (especially via Zoom)
• Complete all assignments on time (late submissions will not be accepted/graded)
• Group work is encouraged for homework, but NOT for EXAMs
• Cheating will not be tolerated!!

4
Software: R and RStudio
• R is a FREE, open-source software used for statistical computing and graphics
• Installing R: [Link] (Windows, Mac, Linux)
• RStudio is a user-friendly development environment for R
• Installing RStudio: [Link]
• R is more interactive, can be more time consuming, but you are in control!!
• Some online resources:
• [Link]
• [Link]
• R.D., Peng Exploratory Data Analysis With R: [Link]

5
Introduction to Biostatistics
In this lecture you will learn about the role, importance and
applications of Biostatistics.
Key Questions/Objectives:
1. What is Biostatistics?
2. Identify different types of data and collection techniques
3. What are the strengths and limitations of different study designs

6
What is Biostatistics?
• The branch of applied statistics targeted towards biological and life
(health) sciences.

• Applied Statistics: application of statistical methods to solve real-life

inspired problems involving generated data, and also develop new
methodologies motivated by real problems.

• Applications and contributions include, but not limited to:

• Healthcare, Medicine, Genetics, Epidemiology, Translational Research…

7
Descriptive and Inferential Statistics
• A collection of observations, systematically arranged is called DATA
• DATA is used for:
1. Descriptive Statistics: methods of organizing and summarizing data

2. Statistical Inference: methods of making generalizations about a

population based on a sample

Estimation (confidence intervals) vs Hypothesis Testing

8
The Scientific Method
Science is a process of eliminating hypotheses that are inconsistent with
observed data
H0, H1, H2, …, Hn
Example: Competing hypothesis about a coin
H0: coin has 2 heads (0 tail)
Competing hypothesis about a coin
H0: coin has 2 heads (0 tail)
H1: coin has 1 head and 1 tail
H2: coin has 2 tails (0 heads)
Toss the coin once and obtain head (data: 1 head observed)

Conclusion: H2 is ruled out (impossible event) plus data favor H0 over H1 (head under
H0 is twice as likely than head under H1)
9
The Scientific Method
Example: Newborns’ problems noticed by OB/GYNs.
2 competing hypothesis were created:
H0: Children are born healthy Weight ≥ 2500 g
HA: Children are born underweight Weight < 2500 g
Define the outcome of interest: Birthweight in grams
Define the population: All babies born in the past 5 years in the area of interest
Research approach and design: Find the birthweight of every newborn from the
past 5 years and then calculate the average?
à Very time consuming, expensive and long
Better: select an appropriate subset of children born in the past 5 years and
investigate only those

10
Principle of (bio)statistics
Were babies born in the past 5 years underweight?
H0: ≥ 2500g vs. HA: <2500g

Sample
All babies born X
in past 5 years selected

Data collection,
Generalization analysis

The true average of x Average

selected new borns is Median
so low that HA wins Conclusion SD
11
Principle of (bio)statistics

Population
parameter Sample
µ

Sample
Conclusion statistics
X¯
Inference: Confidence intervals to estimate the population mean μ, or
Hypothesis test.
12
Some Definitions
• Population: the complete collection of units (individuals or objects) of
interest in a study
Parameter (the truth): any descriptive measure based on a population

• Sample: a smaller subset of the population of interest

Statistic (approximation of the truth, prone to error): any descriptive
measure based on a sample

• Variable: a characteristic of each element of a population or sample

13
Sampling Populations
• The ability to obtain reliable measures of population depends critically on
how we sample populations

• Properties of good samples:

• Low sampling error, focus on precision
• Accurate (or unbiased) estimates

• Some types of sampling:

• Random sampling
• Convenience sampling
• Volunteer sampling
Whitlock et al, 2015

14
Some problems in sampling
Sampling
Population Sample

Sampling in practice:
Target Source Sampled
population population Sample
population

Inference
Population Sample
parameter statistic
15
Sampling
Statistical inference allows generalization from the sample only to the
population from which the sample came from (assuming no sampling and
response bias!).
If we want to make inference about the target population, we have to make
sure that the population from which the sample came from is similar to the
target population.
Example:
We want to learn about all university students in NYC
We randomly sample 100 students from Columbia
We can only generalize to all Columbia students
In order to generalize to all NYC students, we have to first show that Columbia
students are ‘similar’ to students from other universities in NYC.
16
 Examples: Population-Sample
A researcher wanted to determine the prevalence of Tuberculosis in New York.
She randomly selected 10000 subjects from the New York phone book and
questioned them about their TB status. She found out that only 3 had TB.
Describe the following
Target population:
Source population:
Sampled population:
Sample:
Parameter:
Statistic:

17
Examples: Population-Sample
A researcher wanted to determine the prevalence of Tuberculosis in New York.
She randomly selected 10000 subjects from the New York phone book and
questioned them about their TB status. She found out that only 3 had TB.
Describe the following
Target population: People living in NY
Source population: All individuals in the NY phone book
Sampled population: 10,000 randomly selected individuals from the list of the
individuals in the NY phone book
Sample: Individuals who agreed to be included in the study (ideally, 10,000)
Parameter: Proportion of people in NY who have TB
Statistic: 3 out of 9,700 = 0.031%
18
Types of Data
• Qualitative data: measurements expressed not in terms of numbers

• Qualitative variables can be subdivided into:

Nominal variables: no inherent order or ranking (e.g., gender)
Ordinal variables: ordered series (e.g., preference, disease severity)
Binary variables: only two options (e.g., pass/fail, yes/no)

• Usually qualitative data are recoded as numerical values

19
Types of Data
Quantitative data: measurements expressed in terms of numbers:
e.g., weight, blood pressure, survival time, etc.

Quantitative variables can also be subdivided into:

Discrete variables: usually there are gaps between the values
e.g., # of pregnancies
Continuous variables: have a set of all possible values within an
interval
e.g., body mass index (BMI)

20
Sources of Data
• Published Source: government, business, sports statistics are collected
and presented in press, online, etc.

• Experimental Study: researchers deliberately influence events and

investigate the effects of an intervention

• Survey: researchers select sample of individuals and record their

responses to questions

• Observational Study: researchers collect information on the attributes

or measurements of interest, without influencing the events

21
Study Design
• Experimental

• Observational

22
Experimental Studies
• Randomized Clinical Trial (RCT)
• Interventions are allocated at random and subjects are followed prospectively
until an outcome is observed
• Randomization
• Ensures that the two (or more) groups are comparable with respect to all
nuisance variables (e.g., confounders)
• Example: testing a new diet for weight loss - want the distribution of body
weight to be similar at baseline between the groups
• Blinding:
• Guards against placebo and experimenter effects, respectively
• Example: researcher might be biased and measure responses differently – why??

23
Observational Studies
Observational studies are to be contrasted with experiments.
• No intervention
• Data collected on an already existing system (practical, less expensive, feasible)
Types of observational studies:
• Case study: descriptive characteristics of a single subject
• Case-control study
• Cross-sectional study
• Cohort studies

24
Case-Control Study
Select subjects with disease, find matching PROs:
controls and test the association between • Useful for rare diseases and long-latency
exposure and disease/outcome of interest.
• Can explore several risk factors
simultaneously

CONs:
• Prone to bias
• Difficult to select controls
• Statistical methods can get complicated

25
Cross-Sectional Studies
Collect data from a group of subjects at PROs:
one specific point in time. • Based on a sample of a general population
• Relatively short time

CONs:
• Difficult to separate the cause-effect
because of the time singularity
• Biased to determine cases with longer
disease
• Snapshots in time can be misleading

26
Cohort Studies
Prospective cohort: select subjects based PROs:
on exposure and follow through time to
observe the development of the disease. • Useful for rare exposures
• Cause-effect easier to establish because
of temporality
• Direct measurement of incidence of
disease

CONs:
• Lost to follow-up (death, drop-outs)
• Can be expensive and long

27
What is the role of a (bio)statistician?
Statisticians tell or fail to tell the truth, the whole truth, and nothing but the truth…

• Study design
• Power/sample size calculation
• Create the data analysis plan
• Data management
• Study implementation
• Data analysis
• Manuscript writing

28
For Next Time
1. Download and get familiar with R and RStudio
2. Funny, but true dialogues:
• Statistics vs Biology: [Link]
• Statistics vs Marketing
[Link]
• Others …

29