Study Designs: Hierarchy of Evidence

Introduction to Study Designs

• A study design is a specific method for collecting, analyzing, and interpreting data in research.

Learning Objectives

• The goal is to explain the hierarchy of evidence in study designs.
• The goal is to understand the differences between various study designs.

Hierarchy of Evidence in Study Designs

• Systematic reviews are at the top of the hierarchy because they give a complete and organized overview of a research question.
• Critically-appraised topics are evidence syntheses.
• Critically-appraised individual articles refer to article synopses.
• Randomized controlled trials (RCTs) are a type of study design.
• Cohort studies and Case-Controlled studies are types of study designs.
• Case series and reports are types of study designs.
• Background information and expert opinion are at the bottom of the hierarchy.

Case Report/Case Series

• Observational studies collect detailed information about an individual (case report) or a small group of patients (case series) who have similar diagnoses or treatments.
• These studies can be like detective stories, and they reveal what happened, not why.
• They do not compare patients, so cannot confirm the effect of an intervention due to unseen factors
• This means, there is risk of bias

Cross Sectional Studies

• Exposure and outcome are measured at the same time.
• It cannot be determined if the exposure caused the outcome, or the outcome caused the exposure
• As an example, obesity and diabetes can be assessed simultaneously.

Measuring Health and Disease

• When measuring health and disease, it is important we understand who gets the disease.
• What are risk factors.
• Where do people get the disease.
• When do people get the disease.
• Goal to intervene.

Measuring Disease Frequency

• Three things must be defined when measuring disease frequency
• WHAT: Defining cases means defining who is counted as a case and how that is determined
• WHO: The size of the population at risk determines the numbers being studied
• WHEN: The time point or period during which data are collected

Defining Cases

• Case definitions should consider the level of severity whether all cases, hospitalized cases, mild cases, and deaths are included.
• Whether all episodes or just the first time an illness is diagnosed are included.
• Outcomes are not always discrete for example Hypertension such as SBP ≥140 mmHg or DBP ≥90 mmHg.
• Seropositivity.
• Intensity of infection.
• A cut-off point may need to be defined to separate cases from non-cases.

Defining the "Population at Risk"

• The population from which the cases originate must be defined.
• It can be defined by several factors like geographically with a region, country, or refugee camp.
• Also can be any age, sex, or race / ethnicity specified.
• The 'population at risk' should only include people who would be counted as a case if they were to have the condition.
• Someone should only be counted as a case if they are from the population and become a case during the risk period.

Defining the Time Period

• Disease frequency measures relate to a particular point in time and relate to what happens over a particular time-period.
• It is important to specify the time-period and / or include it in the calculation for measures of disease frequency.

Two Measurements of Disease Frequency

• Prevalence (prevalent cases) frequency (or number) in a well-defined population who are specified with that disease or other condition within a defined period.
• Which includes all previous cases such as diabetes, hypertension, etc.
• Incidence (incident cases) frequency (or number) new events of the outcome from each study population based on what period is examined.

Summary for Measuring Health and Disease

• Measure the what the who and the when
• Incidence captures new cases within a specific time
• Prevalence captures all existing cases within a specified period

Case Control Studies

• Participants with outcome in a matched group, that do not have the same outcome.
• For instance, participants with, VS, participants without diabetes.
• There is some uncertainty on measurement, as measurement timing is often not defined

Cohort studies

• Exposures are measured in the present, with participants being monitored over time.
• Obesity is measured before diabetes presents.

Randomized Clinical Trial

• During the process, participants are randomly allocated to either an exposure or no exposure, these differences are then monitored.
• Gold standard to predict causality

Study Design Summary

Data Types and Numerical Summaries

• A data type can be either categorical or numerical.
• Numerical data has continuous or discrete aspects
• Categorical data has nominal/ordinal aspects

Data Type: Categorical

• Nominal Categories with no inherent order.
• Customer's gender is male or female.
• Ice cream flavours of chocolate, vanilla, or strawberry.
• Ordinal: Ranked categories such as customer satisfaction, and intervals between aren't consistent Very dissatisfied or Dissatisfied or Neutral or Satisfied, Very satisfied.

Bar Charts

• For categorical data, this involves the ice cream flavours and there counts.

Numerical Summary: Categorical data

• This refers to the frequency which is counts or observations
• As well as the percentages, based on the proportion relative to the total number of observations.

Numerical Data Types

• Continuous numerical data can take any value within a range, and represent measurements.
• Customer's age or amount spent on ice cream.
• Discrete is countable with distinct values with distinct values
• Number of ice cream scoops can be counted.

Graphical Display: Numerical Data

• Displays for nominal scale are histograms and boxplots.

Graphical Display: Boxplot Components

• Box plots show several components, and the minimum values.
• The first quantile.
• Whiskers.
• Median.
• 3rd quantile
• Interquartile range.
• Outliers.
• Maximums.

Graphical Display: Data Distribution

• Data distributions can be either symmetric or skewed.
• A symmetric data set is balanced on each side.
• Imbalanced sets lead to data skew.

Skewed Data Example

• Right-skewed data can occur when age is clustered towards the younger ones, where there are fewer older customers.
• Left-skewed data can occur when age is clustered towards the more matured, where there are fewer younger customers.
• The interpretation of distributions, often are examples of customer experience.
• Normal distribution is a balanced mix of customers across age groups.
• Right skewed data show a larger proportion of younger customers.
• Left skewed data, show larger proportion of older customers.

Numerical Summary

• A "numerical summary" indicates something about central tendencies.
• What's the customer's average age? Is the main question.
• Or the most common ice cream flavour.
• Or the number of ice cream scoops bought?
• The mode is the most frequently occurring value.
• The median is the middle value.
• And the average is what is the mean

Data Distribution

• In numerical data there are different means, modes and medians depending on the distributions
• In a normal distributions the median, is about equal to the mode
• Otherwise it is either negative or positive

Numeric Variability

• Besides central tendencies, there is data on numerical variability
• Order same number of scoops, or different
• Range is defined to show variability
• This includes an inter quartile range
• or Standard deviation also show variability
• There is small area showing the range, and an inter quartile range that is in the middle

Standard Deviation

• Average distance from the means
• Consider scenario with scooper "A" and "B", that scoop ice cream, but are inconsistent
• "Scooper A" scoops, have a lower standard deviation (the range is smaller) compared to "B"

Numerical Summary

• In numerical data there is a variability portion and also a central tendency, which can be looked at through various means

Sampling Methods

• Different data collection and description, include describing the benefits, plus their limitations

Sampling Method

• Study designs, include different ways to define that and also sampling methods. Where there are
• Errors, that can be random or systematic.
• Population versus sample

Selection of Participants

• Simple random sample
• Cluster
• Probability methods
• Systematic
• Stratified
• Non probability methods
• Sampling methods
• Convenience
• Purposive
• Quota
• Snowball

Key Definitions

• Sampling Unit: A person, household, or school, etc.
• Target (or Reference ) Population: The population of interest.
• Sampling Design (or Scheme): The procedure for selecting the sampling units from the study population.
• Sampling frame: List of sampling units from which the sample is selected.

Non-Probability Samples

• Convenience are individuals that are chosen without any specific pattern, such as a clinic goer
• With Purposive samples, investigators use balanced criteria for select individuals
• With "snowball" the initial individual refers to the researcher to another
• Quatos select predetermined number of individuals
• They are easy and cheap, but not representative

Simple Random Sampling

• Equal probability: Everyone has the same chance of being selected
• Probability of selecting each unit is known
• Removes possibility of bias
• Drawn by listing population, and randomly selected

Simple Random: Benefits and Limitations

Sampling

Equal probability: Everyone has the same chance of being selected

• Benefits:*
• Representative samples are a good representation on your populations Limitations are sampling frame needs a list of people in your population
• It can easily and logically become difficult with a larger population.

Systematic Sampling

• Sampling units need to be in a sequence
• Determine k
• Population / sample size
• Choose Random and begin.
• Select unit at regular intervals
• EXAMPLE: A sample of 600 is wanted, out of 10000
• The intervals are, rounded to 17
• A random person is chosen
• 15 was chosen, and the 15th person is to be sampled
• The subsequent ppl, are to be sampled 32, 49

Systematic Sampling: Benefits and Limitations

• Works well when units are already in Sequence.
• Logistically sometimes easier than simply random sampling.
• Susceptible to errors based on season. and time variations.
• Are ppl being studied in an are more likely to visit, during lunch or later.

Stratified Sampling

• Separate the population based on set characteristics
• Such as sex, age, ethnicity, geographic region.
• The number of people selected per strata can be proportional or nonproportional
• If no proportional, then an adjustment statement must be input into the analysis.

Stratified Sampling Benefits and Limitations

• Benefits*
• Practical for large graphic areas
• You're interested in specific subgroups
• Can be more varied on sampling based population.
• Limitations*
• Be cautious of the strata statement
• The strata must be inputted in the formula analysis

Cluster Sampling

• Most populations exist in clusters.
• Often there is no information on all possible Units, but there is some knowledge on the unit(s)
• Cities, buildings exist.

Cluster Benefits and Limitations

• Useful when populations are widely dispersed*
• Naturally occurring units are more homogenous than those of the total populations*
• Smaller effective size sample will be smaller*

Inclusion and Exclusion Criteria

• Inclusion is used to identify the main group within a population to study
• Exclusion helps eliminate interference that creates issues

Summary Sampling