Topic 3 Getting the Data! Dr Luke Kane April 2014 Topic 3: Getting the Data 1
Nov 29, 2014
1
Topic 3Getting the Data!
Dr Luke KaneApril 2014
Topic 3: Getting the Data
Topic 3: Getting the Data 2
Outline
• Study design • Data Collection• Populations • Sampling• Types of Study• Confounders• Matching• Placebo
Topic 3: Getting the Data 3
Objectives
• Explain what we mean by study design• Explain what we mean by data collection• Understand a sampling frame, populations, errors,
simple random sample, stratified and systematic sampling, cluster and control sampling
• Understand the types of studies: Case reports, cross sectional, case control, cohort, clinical trials, randomised control trials
• Understand what is meant by confounders and matching• Understand randomisation and placebo
Topic 3: Getting the Data 4
Study Design
• Study design– What is the question?– What is the hypothesis?– What are the variables?
• What is the outcome variable (the main one)?
– How many subjects do we need to include?– Who are the subjects? How do we select them?– How many groups do we need?– Are we going to intervene or observe?– Do we need a comparison group?– When will take measurements? Before, during, after?– How long will the study take?
Topic 3: Getting the Data 5
Data Collection
• How are we going to collect the data from the subjects?
• How do we make sure the sample is as representative as possible?
Topic 3: Getting the Data 6
Sampling: Huh?
• What is sampling?– Selecting a subset of individuals from a population
to estimate characteristics of total population• If you want to study rat behaviour– You can’t watch every rat in the world– “Sampling” is how you choose which rats to look
at– You need to make the rats you look at
representative
Topic 3: Getting the Data 7
Sampling: The Sampling Frame
• The information you use to identify your sample
• Examples:– List of people in a census– Telephone directory– Management list of workers in a plantation– Maps
Topic 3: Getting the Data 8
Sampling: Populations
• Best explained with examples: – Target population: All children with malaria in Cambodia
in 2013– Study population: All children with malaria in the main
hospital in Phnom Penh, Battambang, Siem Reap and Sihanoukville in 2013
– Sample population: 200 children from the paediatric ward of each of the four hospitals in 2013
Topic 3: Getting the Data 9
Sampling: Errors
• Can a sample ever be a perfect replica of the target population?
• NO!– It is an feature of any sample – Unless you could measure every single person in a
population (usually impossible)• Example: – Total population has a TB prevalence of 1.3%– Your sample has a prevalence of 0.8%– The sampling error is 0.5%
Topic 3: Getting the Data 10
Sampling: The Simple Random Sample
• Importance of data being representative– Most representative sample is usually a simple
random sample• Only way it will differ from target population is by
chance
– What do we mean by RANDOM• Each individual has an equal chance of being included
Topic 3: Getting the Data 11
Sampling: Further Types of Random Sampling
• Can also have stratified and systematic random sampling– Stratified: break down sampling frame into strata• E.g. male/female, smoker/non-smoker etc.
– Systematic: Use a system to pick individuals out of a sampling frame• E.g. every 10th on the list• May be patterns on the list – Randomness!
Topic 3: Getting the Data 12
Sampling: Other Types of Sampling
• Cluster sampling– Test households for dengue in Phnom Penh– Difficult to get a list of every house in PP– So you can look at a map, divide the map up and
take samples from different “clusters” of houses• What if you look at houses which are all along a canal?
• Contact or consecutive sampling– Look at patients visiting a clinic– What if the clinic is in a very rich part of town?
Topic 3: Getting the Data 13
Types of Study
• Case reports• Cross sectional studies• Case-control studies (“Retrospective studies”)• Cohort studies (“Prospective studies”)• Randomised controlled trials (RCTs)• Ecological studies
Topic 3: Getting the Data 14
How to Categorise Types of Studies
• Observational Vs. Experimental– Observing is when you measure, ask questions etc– Experimentation is when you make an
intervention – A CHANGE – and see what happens
Observational Experimental
Case Series or Case Report Clinical trials
Cross Section study Randomised controlled trial
Cohort Study
Case Control study
Topic 3: Getting the Data 15
Observational: Case Series/Report
• Case report – experience of on patient• Case series – experience of a group of patients
with a similar diagnosis– Very good for identifying new disease– Accumulation of case reports could point to an
epidemic• Easy, quick• But very limited, no comparison group
Topic 3: Getting the Data 16
Case Report: ExamplesAm J Cardiol. 1968 Dec;22(6):782-90.Transplantation of the heart in an infant and an adult.Kantrowitz A, Haller JD, Joos H, Cerruti MM, Carstensen HE.PMID: 4880223 [PubMed - indexed for MEDLINE]
Topic 3: Getting the Data 17
Observational: Cross Sectional Studies
• Probably the most common type of study– Sample (cross section) of population interviewed,
tested or studied to answer a question• Examples:– What is prevalence of TB in Cambodia?– Is prevalence of TB affected by age or sex?
• Quick and easy, good for measuring scale of problem
Topic 3: Getting the Data 18
Observational: Cohort Studies – “Prospective”
• Descriptive cohort study: follow a group (cohort) of people with a risk factor and see if they develop a disease
• Analytic cohort study: • Prospective – i.e.
they look forward• Incidence of
disease
Topic 3: Getting the Data 19
Example of Cohort Studies
• Is the risk of lung cancer higher among people who smoke compared with non smokers?– Sir Richard Doll’s “British Doctors’ Cohort Study”• 35,000 British doctors – Smoking and Lung Cancer
Topic 3: Getting the Data 20
Observational: Case Control Studies – “Retrospective”
• Compare cases (people with a disease) and controls (people without the disease) to see if they share a past exposure– Look backwards to find a cause
• Cases and controls must be as similar as possible
• This is to account for “confounding” – will talk about this soon
Topic 3: Getting the Data 21
Case Control Studies: Examples
• Are people with lung cancer more likely to be smokers than people without lung cancer?– Define cases:
• people with lung cancer
– Define controls:• People without lung cancer
– Define exposure:• Smoking
• Does working in a plantation increase the risk of malaria?
Topic 3: Getting the Data 22
Example: Malaria & Plantations 1
• Case report:– A patient in Mondulkiri province has P. falciparum
malaria and he works and lives in a rubber plantation• Case series: – There are 15 patients in Mondulkiri with P.
falciparum malaria and they all work in a rubber plantation
• Cross sectional study:– Test the blood of a samples of workers in 20
plantations in Cambodia to see if they have malaria
Topic 3: Getting the Data 23
Example: Malaria & Plantations 2
• Case control study: – Ask 500 people with malaria and 500 people without
malaria where they work• Descriptive cohort study: – Take 100 new plantation workers who have never been to
a plantation and monitor them to see if they develop malaria
• Analytic cohort study:– Take 100 rural workers, assign 50 to work in a rice paddy,
and 50 to work in a plantation. Monitor them to see if they develop malaria
Topic 3: Getting the Data 24
Confounding
• Before looking at experimental designs…• Cases and controls must be similar– Example: Does smoking cause lung cancer?– Cases: smokers, controls: non-smokers – difficult to tell if smoking causes lung cancer if
controls are all double the age of the cases– Because cancer increases with age– So age is a CONFOUNDER in this example
Topic 3: Getting the Data 25
Confounding
• A confounder is a variable that is associated with the risk factor and the outcome
• Commonly age and sex• Important to adjust for or control for confounders• Rates of drowning increase with ice-cream
consumption– Confounder is the SUMMER– i.e. no real relationship between drowning and ice
cream
Topic 3: Getting the Data 26
Matching
• Matching is a way of making cases and controls more similar– How you do the matching divides case-control studies into
two types:• Matched and unmatched designs
• Matched designs– Each person matched with another person
• Unmatched designs – Use frequency matching to broadly group cases and controls– E.g. same proportion of M/F, same mix of ages
Topic 3: Getting the Data 27
Experimental: Clinical Trials
• Compare treatments between a treatment group and a control group– Example is a new drug to treat asthma– Give half the population the new drug– Half an old drug– See what the difference is
Topic 3: Getting the Data 28
Randomisation
• How do you allocate people to each group?• You can do this randomly– Like tossing a coin– Or a random number generator
• So any differences between the groups will only be by chance
• Gets rid of selection bias– Researchers choose who to put in each group
Topic 3: Getting the Data 29
Experimental: Randomised Control Trials (RCTs)
• Randomised clinical trial is called a randomised control trial
• BLINDING: – better if patient’s don’t know what group they are
in• Reduced placebo effect
– Better still if investigator doesn’t know what group patient is in • Reduces treatment bias ( you think drug is working)• Reduces assessment bias ( you think they are better)
Topic 3: Getting the Data 30
Placebo
• Psychological response which can lead to a physical (i.e. biochemical) response
• Can effect outcomes in studies
Topic 3: Getting the Data 31
Summary
• Study design • Data Collection• Populations • Sampling• Types of Study• Confounders• Matching• Placebo
Topic 3: Getting the Data 32
References• Bowers, D. (2008) Medical Statistics from Scratch: An Introduction
for Health Professionals. USA: Wiley-Interscience.• Grant, A. (2014) “Epidemiology for tropical doctors”. Lecture (S6)
from the Diploma of Tropical Medicine & Hygiene, London School of Hygiene & Tropical Medicine.
• Greenhalgh, T. (1997) “How to read a paper” British Medical Journal. Web, accessed April-May 2014 at <http://www.bmj.com/about-bmj/resources-readers/publications/how-read-paper>
• Hoskin, T (2012) Parametric and non-parametric: Demystifying the Terms. Retrieved from <http://www.mayo.edu/mayo-edu-docs/center-for-translational-science-activities-documents/berd-5-6.pdf>