
Revision For Quiz 1 - Epi
Revision For Quiz 1 - Epi
Week 1
Measuring Disease Frequency (Morbidity and Mortality)
The Goal of Epi:
Improve overall health of the population by reducing morbidity and mortality. To be able to achieve this goal, we need to learn how to quantify morbidity and mortality.
Measures of Morbidity:
Prevalence: What proportion of the population actually has the diseases at a specific time point.
Common types of prevalence:
Point Prevalence: What proportion of the population has the diseases at a specific time point.
Period Prevalence: What proportion of the population has the diseases at any point during a given time period.
Lifetime Prevalence: What proportion of the population had the diseases at some point in life.
Incidence: What proportion of the population newly acquired the diseases during a given time period.
Common types of incidence:
Incidence Proportion (IP) or Cumulative Incidence or Attack Rate: proportion of people who develop the disease or become injured or die or are at risk of getting disease during a given time period (cross-sectional).
Incidence Rate (IR): Proportion of people who develop the disease or cure or becomes injured or dies or is at risk of getting the disease per unit of time (longitudinal).
Calculating IP
N= 10
Total recovered = 7
IP = 7/10 x 100
70% have totally recovered
Calculating IR
Calculate person*time (week sussed in example)
(3 people*1 week) + (1*3) + (3*4) + (3*5) = 33
7 people recovered, 7/33 = 0.21 x 100
21 people per 100 person-weeks
Relationship Between Prevalence and Incidence
Increase incidence-increase prevalence, vice versa
Increased death - decrease cure, vice versa
Case 1 acute: Complete cure:
Increase cure-decrease prevalence - decrease incidence- decrease death.
Case 2 chronic:
Cure constant - constant/increase prevalence - constant/increase incidence - decrease death.
Measures of Mortality
Crude Mortality Rate (crude death rate): The mortality rate from all causes of death for a population.
Total number of people that have died during a given time period/ total number of population in the same time period. (Reported in 1000 or 100,000)
Cause Specific Mortality Rate: The mortality rate from a specific cause for a population during a given time period.
Total number of people that have died during a given time period from a specified cause/ Total number of population in the same time period. (reported in 100 or 100,000).
Proportional Mortality Rate: The proportion of deaths in a specified population during a given time period attributable to different causes.
Total number of people that have died due to a specific cause during a given time period/ Total number of people that have died from all causes during the same time period.
Death to Case Ratio: Total number of people that have died due to a specific cause during a given time period / total number of new cases reported for that specific cause in the same time period.
Infant Mortality Rate: Total number of infants under 1 year of age that have died during a given time period/ total number of live births reported during the same period.
Maternal Mortality Rate: Total Number of deaths of women during pregnancy, during childbirth, or within 42 days of termination of pregnancy during a given time period/ Total number of live births reported during the same time period (reported in 100,000).
Standardised Rates (Direct and Indirect Method)
Helps to compare morbidity and mortality between two or more populations.
Comparison of crude mortality or morbidity rates is often misleading because the populations underlying characteristics may differ, such as age or gender.
Standardization rates is a method for overcoming the effect of few confounding variables commonly age and gender.
Age as a variable generally used to compare or describe mortality.
Two Types of Standardization Methods:
Direct Method: Comparing sampled population to a reference population. Useful with large sampled population e.g. cancer rates across countries.
Indirect Method: Used hwen reference population data not available. Used when sampled population is very small and comparison to reference population can produce incorrect findings eg cancer rate between ACT and NSW.
Practical Implications of Standardization Population Based Morbidity and Mortality Indicators
The usefulness of standardised rates as the terminology suggests are the standardised rate therefore can help to compare data across ecological data sets.
Helps to interpret research data accurately.
Global Health Indicators
Life Expectancy: The average period that a person may expect to live.
Healthy Years OR Disability-free Life Expectancy: The number of years that a person is expected to continue to live in a healthy condition.
Years of Lost Life (YLL): The number of years of life lost due to premature death, defined as dying before the ideal life span.
Disability-adjusted Life Years (DALY): A measure of healthy life lost, either through premature death or living with disability due to illness or injury.
Week 2
Overview of the most common types of epidemiological studies.
Population Based: Descriptive Studies
Provides distribution (percentages, frequencies) of diseases. Does not provide causes of disease.
Eg: Australian Health Survey
Strengths:
Quick and easy
Useful for identifying variations in the distribution of disease or in the distribution of nutritional factors
Limitations:
Limited ability to identify associations- cause-effect relationship.
Population Based: Ecological Studies
An observational study defined by the level at which data are analysed, namely at the population or group level, rather than an individual level.
Eg: Cross-cultural studies on national data
Strengths:
Quick and easy if using secondary data
Useful information about various factors associated with a condition
Useful for forming hypotheses about diet/disease relationship
Limitations:
May not be applicable at an individual level
No cause-effect relationship
Individual Based Studies
Case Report/Case Series: A particular case in a patients, population/sample group discusses. For example, Iodine deficiency and prevalence of goitre in locals living in the Himalayan region.
Cross-Sectional: association between exposure variable and outcomes identified. For example, maternal pressure feeding and picky eating in children are bidirectional in nature.
Case-Control: Comparing an interest group to a reference group (no randomisation). For example, smoker than no-smoker are 8x at the risk of lung cancer.
Strengths for green highlight:
Relatively quick and inexpensive
Can investigate a wide variety of potential risk factors simultaneously
Can be applied to common and rare diseases
Limitations for green highlight:
Retrospective design
Problems collecting accurate dietary intake data
Objective data on past exposures may not be available
Subject to several types of bias
Recall bias
Selection bias
Information bias
No cause-effect associations
Cohort/prospective/longitudinal/retrospective study: sample followed by x amount of time. For example, a 10 years longitudinal study showed that low F&V intake is related to cancer.
Strengths
Information is more reliable and applicable to normal life- participants followed in daily conditions.
Allow determination of the timing of events
Limitations
Time-consuming and expensive
Changes in behaviour may mislead results
High potential for selection bias and confounding factors
Drop out rates, generalisation of the findings
No cause-effect associations.
Non-randomized/quasi experimental/field/community trails
Complex Intervention Designs
Suppose a researcher wants to study the effect of a reading program on reading achievement.
She might implement the reading program with a group of students at the beginning of the school year and measure their achievement at the end of the year.
This simple design is known as a one-shot case study design.
Unfortunately, the students' end of year reading scores could be influenced by other instruction in school, the students maturation, or the treatment.
We also do not know whether the students' reading skills actually changed from the start to end of the school year.
We would improve on this design by giving a pretest at the start of the study.
This is known as a one group pretest-posttest.
Unfortunately, the students' end of year reading scores still could be influenced by other instruction in school, the students maturation, or the treatment.
Our researcher may wish to have a comparison group.
This is a static-group pretest-posttest design.
If our researcher believes that the pretest has an impact on the results of the study, she might not include.
This is a static-group comparison.
Because our researcher did not pretest, she might wish to randomly assign subjects to treatment and control groups.
Random assignment of subjects to groups should spread the variety of extraneous characteristics that subjects possess equally across both groups.
This is a randomized posttest-only, control group design.
Our researcher could also include a pretest with her random assignment.
This is a randomized pretest-posttest control group design.
With the randomized Solomon four-group design, all groups are randomly assigned and given the posttest.
Two of the groups are given pretests.
One of the pretest groups is assigned to treatment and one of the non-pretest groups is assigned to treatment.
Experimental/Intervention Studies (RCT) - Complex Intervention Designs
Randomized control trials (RCTs)
Researchers recruit subjects than randomly assign them to receive or not receive a treatment under investigation
Observe to see whether intervention influences occurrence of disease
Ideally should be:
Randomised
Placebo controlled (preferably double-blind)
Cross-over design
Community Intervention RCT
Intervention trial carried out at the community level (C-RCT)
Eg: water fluoride trails:
Entire communities had fluoride added to their water supplies
Other communities received untreated water
Dental health in both communities was assessed.
Experimental/Intervention Studies Strengths & Weaknesses
Strengths
Tightly controlled
Can provide direct evidence of cause and effect (RCT only)
Limitations
Limited application (ethical issues, nature of food intake)
Subject compliance
Need long-term intervention to detect an effect - cost prohibitive
Can assess only 1 or 2 factors at a time
Strength of Evidence
Weak:
In vitro
Animal
Case-study
Kind of Weak
Descriptive
Ecological
Kind of Strong
Case-control
Cohort
Strongest
Intervention trials
Levels of Evidence: NHMRC
I - Evidence obtained from a systematic review/meta analysis of all relevant RCT’s
II - Evidence obtained from at least one properly designed RCT
III-1 - Evidence obtained from well-designed quasi-randomized controlled trials
III-2 - Evidence obtained from comparative studies with concurrent controls and allocation not randomised, cohort studies, case control studies
III-3 - Evidence obtained from comparative studies with historical control, two or more single-arm studies, or interrupted time series without a parallel control group (not randomized)
IV - Evidence obtained from case series, either post-test or pre-test and post-test (no control group), cross-sectional.
V - Clinical/expert consensus.
Overview of the most common types of research articles.
Original Research: These articles report on a specific/single study or experiment.
Review Articles: Summarize the findings of other studies or experiments; attempts to identify trends or draw broader conclusions. ( A valuable source to find more references in your related area; the content is generally used in writing your literature review and discussion).
Meta-Analysis: This type of research article also summarises the findings of others studies or experiments, however statistical analysis is conducted to derive conclusions. Therefore, a review is a narrative, whereas a meta-analysis has a narration along with numerical conclusions.
Case Study: useful for clinical based research. They describe a single case or a similar trend observed. For example, diabetic nephropathy observed in Alzheimer’s patients in a particular hospital.
Letter to the editor or short communications:
They may describe an original research very briefly
They may respond to a previously published research article, if they have a difference of opinion
They could reply to the authors who had a difference of opinion
Government documents/reports: usually sourced directly from google
Other thesis: Google, library
Week 3- Measures of Association
Understand association between IV & DV: By Quantifying Associations
Understanding associations: Focus on exposure (IV) and health outcomes (DV).
How to understand associations between IV and DV: By quantifying associations.
Various methods to quantify associations for cohort studies: Rate ratios; risk ratios; attributable risk; population attributable risk and fractions.
Various methods to quantify associations for case-control studies, cross-sectional: Odds ratio
Quantifying Associations for Cohort Studies: Rate Ratios, Risk Ratios, Attributable Risk, Attributable Fractions
Rate Ratio or Incidence Density Ratio: the incidence rates, person-time rates, or mortality rates in an exposed group divided by the incidence rates, person-time rates, or mortality rates in an unexposed (or less exposed) comparison group.
Rate ratio = IRe/IRu
IR= incidence rate
e= exposed group
u= unexposed group
Rate ratio = 1 is equal rates in the two groups, >1 indicates increased risk in exposed groups, <1 indicates decreased risk in exposed groups.
Risk ratio or Relative Risk: A risk ratio (RR), also called relative risk, compares the risk of a health event (disease, injury, risk factor, or deaths) among one group with the risk among another group.
Risk ratio = IRe/IRu (from both groups)
Risk ratio = 1 is equal rates in the two groups, >1 indicates increased risk in exposed groups, <1 indicates decreased risk in exposed groups.
Difference in the Definition:
Rate Ratio: ratio of the rate of an event in one group (exposure or intervention) to that in another group (control).
Risk Ratio: ratio of the risk of an event in one group (exposure or intervention) to that in another group (control).
It depends on the definition of rate and risk.
Rate: summary measure which conveys the idea of risk over time.
Risk: probability of occurrence of a given event.
Risk Difference or Rate Difference or Attributable Risk (AR):
AR is the difference between the incidence of disease in the exposed population versus incidence in the unexposed population.
Attributable risk is the extent to which the incidence of disease would be reduced has the individuals not been exposed to the factor.
AR measures the strength of association and indicates causality.
AR is a better indicator than relative risk for preventative interventions when the disease risk or health issue is removed.
AR = IRE-IRu
Attributable Fractions (AFs)
AFs = (IRe-IRu)/IRe
Odds Ratio (Case-Control, Cross-Sectional Studies)
Odds Ratio
The ratio of the odds of an event occurring in the exposed group versus the unexposed group
Calculated in case-control studies as incidence of outcome is not known.
OR>1 indicates increased occurrence of event
OR<1 indicates decreased occurrence of event (protective exposure)
OR= (a*d)/(b*c)
Calculating AF and PAF with OR
In case control studies the disease condition is already existing (disease vs control group) therefore cannot calculate true incidence rate (new cases).
We can therefore not calculate the true attributable risk (AR is the difference between the incidence of disease and the exposed population vs incidence in the unexposed population).
However, we can calculate the proxy attributable fraction (AF), via odds ratio
Key Epidemiological Concepts: Validity and Reliability
Validity: The truth in the measurement/accuracy. Validity is the degree to which an instrument/procedure measures what it is actually supposed to measure.
Types of validity related to the study procedure:
Criterion Validity: Whether the instrument/procedure is measuring what it should measure. Eg, an IQ test should measure IQ.
Face Validity: The degree to which the purpose of the instrument/procedure is clearly understood to the participants. Eg, interview with participants.
Content Validity: Validity of a construct. Type:
Convergent Validity: examines whether data obtained from two different procedures/instruments measure the same construct.
Divergent Validity: examines whether the construct of interest is different from other constructs present in a research.
Internal Validity: The confidence that the independent variable has only caused the outcome/dependent variable. Therefore, control for confounders.
External Validity: The validity of generalization.
Reliability: The consistency in the measurement, precision
Test-retest Reliability: Measures the variation in the measurements taken by the same person or using the same instrument on the same item and under the same conditions.
Inter-rater Reliability: Two or more trained researchers record observations of the subjects independently.
Overall Reliability or Internal Consistency: It examines whether the items proposed to form a construct/scale measures what it is intended to measure.
Refer to notebook for the table on Cronbach’s alpha on internal consistency.
Week 4
Linking Exposures (IV) and Diseases (DV)
Smoking causes cancer. Smoking is the exposure/IV and cancer is the disease/DV.
Before claiming causation, check if the association between IV and DV is not because of:
Chance (Random Error)
Bias (Systematic Error)
Confounders
What is Chance (Random Error)
Chance/Random Sampling Error
You may have a significant association between your IV and DV not because it is a true reflection of what is happening in the wider population but because of chance and random sampling error.
You cannot study the entire population. You take a sample of the population that is most likely to be representative of the entire population and investigate your research question.
Even if you study several different samples from the same population, each time you answer your research question will be significantly different. Sample 1 = strong association, Sample 2 = moderate association, sample 3 = weak association, sample 4 = no association.
The reason for the difference each time is because people in each of your sample are different - their characteristics, behaviours, lifestyle, therefore outcomes are different. This is called Random Sampling Error.
Random sampling error can lead to Type I and Type II errors.
Type I error - H0 is true, but results show significance.
TypeII error - Reject H0, but it is not significant.
How to minimize Random Sampling error and Types I & II errors.
Undertake hypothesis testing.
Null Hypothesis (H0): There is no association between IV and DV. Null hypothesis accepted.
Alternative Hypothesis (HA): There is an association between IV and DV. Null hypothesis rejected.
Minimizing Type I Errors
p-values
Significance of 0.05 reflects that there is a 95% confirmation that your results are accurate and not occurred by chance.
This means there is a 95% less chance of committing a Type I error.
The lower the p-value, the less likely of a Type I error.
Effect Size
An effect size is a standardized measure. Results from different studies can be compared based on the effect size.
Pearson’s correlation coefficients, odds ratio, confidence intervals: The most common effect size generally reported in the field of Health Science.
In both Pearson’s and Odds Ratio, the greater the effect size, the stronger are the results, so the less chance of Type I error.
The narrower the CI, the stronger the results, the less chance of type I error.
Pearson’s correlation effect size is denoted by ‘r’.
r <0.3: small effect size
r= 0.3-0.4: medium effect
r>0.5: large effect
Difference Between Significance and Effect Size
Significance reflects that there is 95% confirmation that your result has not occurred by chance.
Effect size reflects the strength of the model.
Effect Size (95% CI)
Dispersion means to quantify how your data is spread around your central tendency.
Dispersion can be quantified in various ways:
Range
IQR
Variance
SD
Z scores
95% CI
Mean = reference line
Dispersion = The number of people in the sample
Error/deviance = distance between the mean and the actual data
Less dispersion = better are the results
More dispersion = poorer are the results
Q-Q plots can be used to assess dispersion too
Less dispersion is better:
Findings are more generalizable to the larger population with similar characteristics
Less chase of Type I error, fewer confounding factors and/or the degree to which CF are influencing the main variables is less.
Statistical Significance vs Clinical Significance
Statistical significance doesn’t always mean clinical significance.
Clinical significance = significance needs to achieve real life outcomes such as clinical impact.
Sources of Random Error
Biological Variation: Fluctuation in biological processes in the same individual over time
Sampling Error: Error caused by random influences on who is selected for the study
Measurement Error: The error resulting from random fluctuations in measurement.
How to Minimize Type II Error
Sufficient power minimizes type II error.
Calculation of sample size.
Generally studies are powered to 80% eg 80% of the time we can correctly identify that the IV and the DV are related.
Generally studies have a precision of 5%.
Studies generally have a 95% CI.
Sample size calculations for intervention studies also consider drop out rates based on previous literature.
What is Bias (Systematic Error)
Bias exists in all research, across research designs and is difficult to eliminate all biases. Aim here is to minimize bias.
Bias can occur at each stage of the research process
Bias impacts on the validity and reliability of study findings and misinterpretation of data can have important consequences for practice.
Planning Stage
Design Bias: Poor study design and unclear aims and methods. To address it, clearly define risk and outcome, preferably with objective or validated methods. Standardized and blind data collection.
Selection/participant bias: Process of recruiting participants and study inclusion criteria. To address, Select patients using rigorous criteria to avoid confounding results. Participants should originate from the same general population.
Confounding bias: Withdrawal rate; age; sex. To address, control confounders.
Channelling bias: Type of allocation bias. When participants with specific illness/health are allocated to a specific group for favourable results. To address, clearly define risk and outcome, preferably with objective or validated methods. Standardize and blind data collection.
Data Collection Stage
Measurement Bias: Occurs if a tool or instrument has not been assessed for its validity or reliability. To address, use reliable and valid tools.
Interviewer bias: To address, standardize interviewers' interaction with patients. Blind interviewer to exposure status.
Hawthorne effect or social desirability bias: Participants respond differently because they are being studied. To address, data collection by middle-person, participants ensure data confidentiality, non-judgemental, benefiting larger causes.
Chronological bias: Time as a potential confounder, participants recruited during Christmas vs those recruited in January. To address, participants recruited during the same timeframe, control during analysis.
Participant Recall bias: To address, use objective data sources whenever possible. When using subjective data sources, corroborate with medical records.
Transfer bias: Participants lost to follow-up. To address, plan for follow up loss by convenient office hours, personalized patient contact via phone or email, home visit.
Exposure misclassifications: Eg FFQ only capturing intake of F&V cannot be generalised to an overall healthy diet. To avoid, clearly define exposure prior to study. Avoid using proxies of exposure. Use validated measures.
Outcome misclassifications: Eg BMI used as an indicator of adiposity. To address, same as exposure misclassifications.
Performance bias: Knowledge of interventions allocation, in either the researcher or the participant. To avoid, double-blind designs, C-RCT.
Evaluation Stage
Analysis bias: Eg reporting only significant findings or findings supporting hypotheses. To address, report results ethically.
Citation/Publication bias: studies are more likely to be published if reporting statistically significant findings. Industry-funded results publishing findings only if it supports their product. To address, during the planning stage discuss MOU/collaborative agreements/intellectual property agreements.
Revision For Quiz 1 - Epi
Week 1
Measuring Disease Frequency (Morbidity and Mortality)
The Goal of Epi:
Improve overall health of the population by reducing morbidity and mortality. To be able to achieve this goal, we need to learn how to quantify morbidity and mortality.
Measures of Morbidity:
Prevalence: What proportion of the population actually has the diseases at a specific time point.
Common types of prevalence:
Point Prevalence: What proportion of the population has the diseases at a specific time point.
Period Prevalence: What proportion of the population has the diseases at any point during a given time period.
Lifetime Prevalence: What proportion of the population had the diseases at some point in life.
Incidence: What proportion of the population newly acquired the diseases during a given time period.
Common types of incidence:
Incidence Proportion (IP) or Cumulative Incidence or Attack Rate: proportion of people who develop the disease or become injured or die or are at risk of getting disease during a given time period (cross-sectional).
Incidence Rate (IR): Proportion of people who develop the disease or cure or becomes injured or dies or is at risk of getting the disease per unit of time (longitudinal).
Calculating IP
N= 10
Total recovered = 7
IP = 7/10 x 100
70% have totally recovered
Calculating IR
Calculate person*time (week sussed in example)
(3 people*1 week) + (1*3) + (3*4) + (3*5) = 33
7 people recovered, 7/33 = 0.21 x 100
21 people per 100 person-weeks
Relationship Between Prevalence and Incidence
Increase incidence-increase prevalence, vice versa
Increased death - decrease cure, vice versa
Case 1 acute: Complete cure:
Increase cure-decrease prevalence - decrease incidence- decrease death.
Case 2 chronic:
Cure constant - constant/increase prevalence - constant/increase incidence - decrease death.
Measures of Mortality
Crude Mortality Rate (crude death rate): The mortality rate from all causes of death for a population.
Total number of people that have died during a given time period/ total number of population in the same time period. (Reported in 1000 or 100,000)
Cause Specific Mortality Rate: The mortality rate from a specific cause for a population during a given time period.
Total number of people that have died during a given time period from a specified cause/ Total number of population in the same time period. (reported in 100 or 100,000).
Proportional Mortality Rate: The proportion of deaths in a specified population during a given time period attributable to different causes.
Total number of people that have died due to a specific cause during a given time period/ Total number of people that have died from all causes during the same time period.
Death to Case Ratio: Total number of people that have died due to a specific cause during a given time period / total number of new cases reported for that specific cause in the same time period.
Infant Mortality Rate: Total number of infants under 1 year of age that have died during a given time period/ total number of live births reported during the same period.
Maternal Mortality Rate: Total Number of deaths of women during pregnancy, during childbirth, or within 42 days of termination of pregnancy during a given time period/ Total number of live births reported during the same time period (reported in 100,000).
Standardised Rates (Direct and Indirect Method)
Helps to compare morbidity and mortality between two or more populations.
Comparison of crude mortality or morbidity rates is often misleading because the populations underlying characteristics may differ, such as age or gender.
Standardization rates is a method for overcoming the effect of few confounding variables commonly age and gender.
Age as a variable generally used to compare or describe mortality.
Two Types of Standardization Methods:
Direct Method: Comparing sampled population to a reference population. Useful with large sampled population e.g. cancer rates across countries.
Indirect Method: Used hwen reference population data not available. Used when sampled population is very small and comparison to reference population can produce incorrect findings eg cancer rate between ACT and NSW.
Practical Implications of Standardization Population Based Morbidity and Mortality Indicators
The usefulness of standardised rates as the terminology suggests are the standardised rate therefore can help to compare data across ecological data sets.
Helps to interpret research data accurately.
Global Health Indicators
Life Expectancy: The average period that a person may expect to live.
Healthy Years OR Disability-free Life Expectancy: The number of years that a person is expected to continue to live in a healthy condition.
Years of Lost Life (YLL): The number of years of life lost due to premature death, defined as dying before the ideal life span.
Disability-adjusted Life Years (DALY): A measure of healthy life lost, either through premature death or living with disability due to illness or injury.
Week 2
Overview of the most common types of epidemiological studies.
Population Based: Descriptive Studies
Provides distribution (percentages, frequencies) of diseases. Does not provide causes of disease.
Eg: Australian Health Survey
Strengths:
Quick and easy
Useful for identifying variations in the distribution of disease or in the distribution of nutritional factors
Limitations:
Limited ability to identify associations- cause-effect relationship.
Population Based: Ecological Studies
An observational study defined by the level at which data are analysed, namely at the population or group level, rather than an individual level.
Eg: Cross-cultural studies on national data
Strengths:
Quick and easy if using secondary data
Useful information about various factors associated with a condition
Useful for forming hypotheses about diet/disease relationship
Limitations:
May not be applicable at an individual level
No cause-effect relationship
Individual Based Studies
Case Report/Case Series: A particular case in a patients, population/sample group discusses. For example, Iodine deficiency and prevalence of goitre in locals living in the Himalayan region.
Cross-Sectional: association between exposure variable and outcomes identified. For example, maternal pressure feeding and picky eating in children are bidirectional in nature.
Case-Control: Comparing an interest group to a reference group (no randomisation). For example, smoker than no-smoker are 8x at the risk of lung cancer.
Strengths for green highlight:
Relatively quick and inexpensive
Can investigate a wide variety of potential risk factors simultaneously
Can be applied to common and rare diseases
Limitations for green highlight:
Retrospective design
Problems collecting accurate dietary intake data
Objective data on past exposures may not be available
Subject to several types of bias
Recall bias
Selection bias
Information bias
No cause-effect associations
Cohort/prospective/longitudinal/retrospective study: sample followed by x amount of time. For example, a 10 years longitudinal study showed that low F&V intake is related to cancer.
Strengths
Information is more reliable and applicable to normal life- participants followed in daily conditions.
Allow determination of the timing of events
Limitations
Time-consuming and expensive
Changes in behaviour may mislead results
High potential for selection bias and confounding factors
Drop out rates, generalisation of the findings
No cause-effect associations.
Non-randomized/quasi experimental/field/community trails
Complex Intervention Designs
Suppose a researcher wants to study the effect of a reading program on reading achievement.
She might implement the reading program with a group of students at the beginning of the school year and measure their achievement at the end of the year.
This simple design is known as a one-shot case study design.
Unfortunately, the students' end of year reading scores could be influenced by other instruction in school, the students maturation, or the treatment.
We also do not know whether the students' reading skills actually changed from the start to end of the school year.
We would improve on this design by giving a pretest at the start of the study.
This is known as a one group pretest-posttest.
Unfortunately, the students' end of year reading scores still could be influenced by other instruction in school, the students maturation, or the treatment.
Our researcher may wish to have a comparison group.
This is a static-group pretest-posttest design.
If our researcher believes that the pretest has an impact on the results of the study, she might not include.
This is a static-group comparison.
Because our researcher did not pretest, she might wish to randomly assign subjects to treatment and control groups.
Random assignment of subjects to groups should spread the variety of extraneous characteristics that subjects possess equally across both groups.
This is a randomized posttest-only, control group design.
Our researcher could also include a pretest with her random assignment.
This is a randomized pretest-posttest control group design.
With the randomized Solomon four-group design, all groups are randomly assigned and given the posttest.
Two of the groups are given pretests.
One of the pretest groups is assigned to treatment and one of the non-pretest groups is assigned to treatment.
Experimental/Intervention Studies (RCT) - Complex Intervention Designs
Randomized control trials (RCTs)
Researchers recruit subjects than randomly assign them to receive or not receive a treatment under investigation
Observe to see whether intervention influences occurrence of disease
Ideally should be:
Randomised
Placebo controlled (preferably double-blind)
Cross-over design
Community Intervention RCT
Intervention trial carried out at the community level (C-RCT)
Eg: water fluoride trails:
Entire communities had fluoride added to their water supplies
Other communities received untreated water
Dental health in both communities was assessed.
Experimental/Intervention Studies Strengths & Weaknesses
Strengths
Tightly controlled
Can provide direct evidence of cause and effect (RCT only)
Limitations
Limited application (ethical issues, nature of food intake)
Subject compliance
Need long-term intervention to detect an effect - cost prohibitive
Can assess only 1 or 2 factors at a time
Strength of Evidence
Weak:
In vitro
Animal
Case-study
Kind of Weak
Descriptive
Ecological
Kind of Strong
Case-control
Cohort
Strongest
Intervention trials
Levels of Evidence: NHMRC
I - Evidence obtained from a systematic review/meta analysis of all relevant RCT’s
II - Evidence obtained from at least one properly designed RCT
III-1 - Evidence obtained from well-designed quasi-randomized controlled trials
III-2 - Evidence obtained from comparative studies with concurrent controls and allocation not randomised, cohort studies, case control studies
III-3 - Evidence obtained from comparative studies with historical control, two or more single-arm studies, or interrupted time series without a parallel control group (not randomized)
IV - Evidence obtained from case series, either post-test or pre-test and post-test (no control group), cross-sectional.
V - Clinical/expert consensus.
Overview of the most common types of research articles.
Original Research: These articles report on a specific/single study or experiment.
Review Articles: Summarize the findings of other studies or experiments; attempts to identify trends or draw broader conclusions. ( A valuable source to find more references in your related area; the content is generally used in writing your literature review and discussion).
Meta-Analysis: This type of research article also summarises the findings of others studies or experiments, however statistical analysis is conducted to derive conclusions. Therefore, a review is a narrative, whereas a meta-analysis has a narration along with numerical conclusions.
Case Study: useful for clinical based research. They describe a single case or a similar trend observed. For example, diabetic nephropathy observed in Alzheimer’s patients in a particular hospital.
Letter to the editor or short communications:
They may describe an original research very briefly
They may respond to a previously published research article, if they have a difference of opinion
They could reply to the authors who had a difference of opinion
Government documents/reports: usually sourced directly from google
Other thesis: Google, library
Week 3- Measures of Association
Understand association between IV & DV: By Quantifying Associations
Understanding associations: Focus on exposure (IV) and health outcomes (DV).
How to understand associations between IV and DV: By quantifying associations.
Various methods to quantify associations for cohort studies: Rate ratios; risk ratios; attributable risk; population attributable risk and fractions.
Various methods to quantify associations for case-control studies, cross-sectional: Odds ratio
Quantifying Associations for Cohort Studies: Rate Ratios, Risk Ratios, Attributable Risk, Attributable Fractions
Rate Ratio or Incidence Density Ratio: the incidence rates, person-time rates, or mortality rates in an exposed group divided by the incidence rates, person-time rates, or mortality rates in an unexposed (or less exposed) comparison group.
Rate ratio = IRe/IRu
IR= incidence rate
e= exposed group
u= unexposed group
Rate ratio = 1 is equal rates in the two groups, >1 indicates increased risk in exposed groups, <1 indicates decreased risk in exposed groups.
Risk ratio or Relative Risk: A risk ratio (RR), also called relative risk, compares the risk of a health event (disease, injury, risk factor, or deaths) among one group with the risk among another group.
Risk ratio = IRe/IRu (from both groups)
Risk ratio = 1 is equal rates in the two groups, >1 indicates increased risk in exposed groups, <1 indicates decreased risk in exposed groups.
Difference in the Definition:
Rate Ratio: ratio of the rate of an event in one group (exposure or intervention) to that in another group (control).
Risk Ratio: ratio of the risk of an event in one group (exposure or intervention) to that in another group (control).
It depends on the definition of rate and risk.
Rate: summary measure which conveys the idea of risk over time.
Risk: probability of occurrence of a given event.
Risk Difference or Rate Difference or Attributable Risk (AR):
AR is the difference between the incidence of disease in the exposed population versus incidence in the unexposed population.
Attributable risk is the extent to which the incidence of disease would be reduced has the individuals not been exposed to the factor.
AR measures the strength of association and indicates causality.
AR is a better indicator than relative risk for preventative interventions when the disease risk or health issue is removed.
AR = IRE-IRu
Attributable Fractions (AFs)
AFs = (IRe-IRu)/IRe
Odds Ratio (Case-Control, Cross-Sectional Studies)
Odds Ratio
The ratio of the odds of an event occurring in the exposed group versus the unexposed group
Calculated in case-control studies as incidence of outcome is not known.
OR>1 indicates increased occurrence of event
OR<1 indicates decreased occurrence of event (protective exposure)
OR= (a*d)/(b*c)
Calculating AF and PAF with OR
In case control studies the disease condition is already existing (disease vs control group) therefore cannot calculate true incidence rate (new cases).
We can therefore not calculate the true attributable risk (AR is the difference between the incidence of disease and the exposed population vs incidence in the unexposed population).
However, we can calculate the proxy attributable fraction (AF), via odds ratio
Key Epidemiological Concepts: Validity and Reliability
Validity: The truth in the measurement/accuracy. Validity is the degree to which an instrument/procedure measures what it is actually supposed to measure.
Types of validity related to the study procedure:
Criterion Validity: Whether the instrument/procedure is measuring what it should measure. Eg, an IQ test should measure IQ.
Face Validity: The degree to which the purpose of the instrument/procedure is clearly understood to the participants. Eg, interview with participants.
Content Validity: Validity of a construct. Type:
Convergent Validity: examines whether data obtained from two different procedures/instruments measure the same construct.
Divergent Validity: examines whether the construct of interest is different from other constructs present in a research.
Internal Validity: The confidence that the independent variable has only caused the outcome/dependent variable. Therefore, control for confounders.
External Validity: The validity of generalization.
Reliability: The consistency in the measurement, precision
Test-retest Reliability: Measures the variation in the measurements taken by the same person or using the same instrument on the same item and under the same conditions.
Inter-rater Reliability: Two or more trained researchers record observations of the subjects independently.
Overall Reliability or Internal Consistency: It examines whether the items proposed to form a construct/scale measures what it is intended to measure.
Refer to notebook for the table on Cronbach’s alpha on internal consistency.
Week 4
Linking Exposures (IV) and Diseases (DV)
Smoking causes cancer. Smoking is the exposure/IV and cancer is the disease/DV.
Before claiming causation, check if the association between IV and DV is not because of:
Chance (Random Error)
Bias (Systematic Error)
Confounders
What is Chance (Random Error)
Chance/Random Sampling Error
You may have a significant association between your IV and DV not because it is a true reflection of what is happening in the wider population but because of chance and random sampling error.
You cannot study the entire population. You take a sample of the population that is most likely to be representative of the entire population and investigate your research question.
Even if you study several different samples from the same population, each time you answer your research question will be significantly different. Sample 1 = strong association, Sample 2 = moderate association, sample 3 = weak association, sample 4 = no association.
The reason for the difference each time is because people in each of your sample are different - their characteristics, behaviours, lifestyle, therefore outcomes are different. This is called Random Sampling Error.
Random sampling error can lead to Type I and Type II errors.
Type I error - H0 is true, but results show significance.
TypeII error - Reject H0, but it is not significant.
How to minimize Random Sampling error and Types I & II errors.
Undertake hypothesis testing.
Null Hypothesis (H0): There is no association between IV and DV. Null hypothesis accepted.
Alternative Hypothesis (HA): There is an association between IV and DV. Null hypothesis rejected.
Minimizing Type I Errors
p-values
Significance of 0.05 reflects that there is a 95% confirmation that your results are accurate and not occurred by chance.
This means there is a 95% less chance of committing a Type I error.
The lower the p-value, the less likely of a Type I error.
Effect Size
An effect size is a standardized measure. Results from different studies can be compared based on the effect size.
Pearson’s correlation coefficients, odds ratio, confidence intervals: The most common effect size generally reported in the field of Health Science.
In both Pearson’s and Odds Ratio, the greater the effect size, the stronger are the results, so the less chance of Type I error.
The narrower the CI, the stronger the results, the less chance of type I error.
Pearson’s correlation effect size is denoted by ‘r’.
r <0.3: small effect size
r= 0.3-0.4: medium effect
r>0.5: large effect
Difference Between Significance and Effect Size
Significance reflects that there is 95% confirmation that your result has not occurred by chance.
Effect size reflects the strength of the model.
Effect Size (95% CI)
Dispersion means to quantify how your data is spread around your central tendency.
Dispersion can be quantified in various ways:
Range
IQR
Variance
SD
Z scores
95% CI
Mean = reference line
Dispersion = The number of people in the sample
Error/deviance = distance between the mean and the actual data
Less dispersion = better are the results
More dispersion = poorer are the results
Q-Q plots can be used to assess dispersion too
Less dispersion is better:
Findings are more generalizable to the larger population with similar characteristics
Less chase of Type I error, fewer confounding factors and/or the degree to which CF are influencing the main variables is less.
Statistical Significance vs Clinical Significance
Statistical significance doesn’t always mean clinical significance.
Clinical significance = significance needs to achieve real life outcomes such as clinical impact.
Sources of Random Error
Biological Variation: Fluctuation in biological processes in the same individual over time
Sampling Error: Error caused by random influences on who is selected for the study
Measurement Error: The error resulting from random fluctuations in measurement.
How to Minimize Type II Error
Sufficient power minimizes type II error.
Calculation of sample size.
Generally studies are powered to 80% eg 80% of the time we can correctly identify that the IV and the DV are related.
Generally studies have a precision of 5%.
Studies generally have a 95% CI.
Sample size calculations for intervention studies also consider drop out rates based on previous literature.
What is Bias (Systematic Error)
Bias exists in all research, across research designs and is difficult to eliminate all biases. Aim here is to minimize bias.
Bias can occur at each stage of the research process
Bias impacts on the validity and reliability of study findings and misinterpretation of data can have important consequences for practice.
Planning Stage
Design Bias: Poor study design and unclear aims and methods. To address it, clearly define risk and outcome, preferably with objective or validated methods. Standardized and blind data collection.
Selection/participant bias: Process of recruiting participants and study inclusion criteria. To address, Select patients using rigorous criteria to avoid confounding results. Participants should originate from the same general population.
Confounding bias: Withdrawal rate; age; sex. To address, control confounders.
Channelling bias: Type of allocation bias. When participants with specific illness/health are allocated to a specific group for favourable results. To address, clearly define risk and outcome, preferably with objective or validated methods. Standardize and blind data collection.
Data Collection Stage
Measurement Bias: Occurs if a tool or instrument has not been assessed for its validity or reliability. To address, use reliable and valid tools.
Interviewer bias: To address, standardize interviewers' interaction with patients. Blind interviewer to exposure status.
Hawthorne effect or social desirability bias: Participants respond differently because they are being studied. To address, data collection by middle-person, participants ensure data confidentiality, non-judgemental, benefiting larger causes.
Chronological bias: Time as a potential confounder, participants recruited during Christmas vs those recruited in January. To address, participants recruited during the same timeframe, control during analysis.
Participant Recall bias: To address, use objective data sources whenever possible. When using subjective data sources, corroborate with medical records.
Transfer bias: Participants lost to follow-up. To address, plan for follow up loss by convenient office hours, personalized patient contact via phone or email, home visit.
Exposure misclassifications: Eg FFQ only capturing intake of F&V cannot be generalised to an overall healthy diet. To avoid, clearly define exposure prior to study. Avoid using proxies of exposure. Use validated measures.
Outcome misclassifications: Eg BMI used as an indicator of adiposity. To address, same as exposure misclassifications.
Performance bias: Knowledge of interventions allocation, in either the researcher or the participant. To avoid, double-blind designs, C-RCT.
Evaluation Stage
Analysis bias: Eg reporting only significant findings or findings supporting hypotheses. To address, report results ethically.
Citation/Publication bias: studies are more likely to be published if reporting statistically significant findings. Industry-funded results publishing findings only if it supports their product. To address, during the planning stage discuss MOU/collaborative agreements/intellectual property agreements.