GATE: Graphic Approach To Epidemiology

GATE: Graphic Approach To Epidemiology
Rod Jackson 2013 1 picture, 2 formulas & 3 acronyms

The GATE frame: Graphic Appraisal Tool for Epidemiological studies – a framework for appraising studies Graphic Architectural Tool for Epidemiological studies – a framework for designing studies more.

1 picture, 2 formulas & 3 acronyms
Presentation outline a framework for study design a framework for study analysis a framework for study error a framework for practicing EBP 1 picture, 2 formulas & 3 acronyms

every epidemiological study can be hung on the GATE frame
1. GATE: design of epidemiological studies: the picture & 1st acronym: PECOT The shape of every epidemiological studies, so a useful design tool every epidemiological study can be hung on the GATE frame

longitudinal (cohort or follow-up ) study
GATE Frame picture a cohort of British doctors smoking status measured (observed) smokers non-smokers follow for 10 years yes A cohort study starts with a cohort of people (the triangle) – in the example this is British drs – real study …all Br drs …. , the cohort is then allocated to different groups based on the factors being studied (the circle) …smokers and non smokers. They are then followed over a period time – illustrated by the downward pointing arrow which identifies the study as longitudinal And the Outcomes are counted – eg lung cancer Useful study design for investigating risk factors for disease - both causes of disease and predictors of disease (including prognostic factors) lung cancer events counted no longitudinal (cohort or follow-up ) study observational studies: allocated to groups by measurement

Population/Participants
1st acronym: PECOT P British doctors Population/Participants smoking status measured Exposure Comparison E C smokers non-smokers Outcomes this slide presents the same study as in the previous slide to introduce the first acroynym PECOT which identified the 5 main components of most epi studies. yes O Lung cancer Time no T 10 years

randomised controlled trial
GATE Frame picture & 1st acronym P cohort of British doctors randomly allocated to aspirin or placebo aspirin E C placebo yes Every design fits on the GATE frame RCTs are a special version of a cohort study where participants are randomly allocated to E or C RCT – best design for answering questions about the effectiveness of interventions O heart attacks counted follow for 5 years no T randomised controlled trial RCT: allocated to E & C by randomisation process

Cross-sectional (prevalence) study
GATE Frame picture & 1st acronym P Middle-aged Americans Body mass index measured overweight E C ‘normal’ weight Diabetes status measured in all participants yes The horizontal arrow identifies the study as cross-sectional O no T Cross-sectional (prevalence) study

Cross-sectional study (multiple categories)
GATE Frame picture & 1st acronym P Middle-aged Americans Body mass index measured obese ‘normal’ weight overweight E1 E2 C yes Can have multiple categories of exposure and outcomes Diabetes pre- O no T Cross-sectional study (multiple categories)

Cross-sectional study (with numerical outcome)
GATE Frame picture & 1st acronym P Middle-aged Americans Body Mass Index (BMI) measured High BMI E C Low BMI high Exposure and outcomes can be numerical / continuous O mean (average) Blood glucose T low Cross-sectional study (with numerical outcome)

Diagnostic (prediction) study
GATE Frame picture & 1st acronym P Middle-aged American women Receive Mammogram screening Test Mammogram positive E C Mammogram negative yes Another study design you will come across is the diagnostic study. There are 2 versions of this design that we will discuss later but here is the one that is the easiest to introduce you to Note the arrow goes sideways O Breast cancer T no Diagnostic (prediction) study

2. GATE: analysis of epidemiological studies: the 1st formula: outcomes ÷population
the numbers in every epidemiological study can be hung on the GATE frame

1st formula: Occurrence of outcomes =
number of outcomes ÷ number in population/group British doctors P Participant Population smoking status measured Exposed Group EG CG Comparison Group smokers non-smokers The first formula is shown at the top of the slide and goes like this: To illustrate this formula I need to introduce a little additional jargon. 1st, instead of the E and C in PECOT, the first formula uses EG and CG. These are the populations used in the 1st formula and are the numbers of people with the study exposure and the number of people with the study comparison exposure.. Also the top quadrants of the Outcomes square - the number of people who develop the main study Outcome are labelled a and b Epidemiology is the study of the occurrence of dis-ease in populations and in different groups within populations. The 1st formula defines the goal of every epidemiological study Occurrence is a generic term that covers risk, probability, incidence, prevalence, likelihood etc PPV, NPV, sensitivity, specificity Outcomes a b yes O Time Lung cancer no T 10 years

1st formula: occurrence = outcomes ÷ population
British doctors smoking status measured Exposed Group EG CG Comparison Group smokers non-smokers Outcomes a b The most important GATE jargon/abbreviations you need to learn to understand the analysis of epidemiological studies are shown in the next 2 slides. In this slide I want to introduce you to EGO - the exposed group occurrence or a/EG which is the first formula applied to the exposure group = which in this example would be the number of smokers who developed lung cancer divided by the total number of smokers in the study. This is the occurrence of lung cancer in smokers yes O Time Lung cancer no T 10 years Exposed Group Occurrence (EGO) = a÷EG = number of outcomes (a) ÷ number in exposed population (EG)

1st formula: occurrence = outcomes ÷ population
British doctors smoking status measured Exposed Group EG CG Comparison Group smokers non-smokers Outcomes a b and In this slide I want to introduce you to CGO - the comparison group occurrence or b/EG which is the first formula applied to the comparison group = which in this example would be the number of non-smokers who developed lung cancer divided by the total number of non smokers in the study. This is also the occurrence (or risk or incidence)of lung cancer in non- smokers yes O Time Lung cancer no T 10 years Comparison Group Occurrence (CGO) = b÷CG = number of outcomes (b) ÷ number in comparison population (CG)

goal of all epidemiological studies is to measure (& compare) the occurrence of outcomes in (different) populations (EGO compared with CGO) P British doctors smoking status measured smokers EG CG non-smokers The downward pointing arrow identifies the study as longitudinal 10 years yes a b O EGO: Occurrence (risk) of cancer in smokers CGO: Occurrence of cancer in non-smokers T no Lung cancer

goal of all epidemiological studies is to measure (& compare) the occurrence of outcomes in (different) populations (EGO compared with CGO) P British doctors Randomly allocated to aspirin or placebo EG CG aspirin placebo The downward pointing arrow identifies the study as longitudinal 5 years yes a b O EGO: Occurrence of MI if taking aspirin T CGO: Occurrence of MI if not taking aspirin no Heart attack (MI)

Average blood glucose in EG Average blood glucose in CG
goal of all epidemiological studies is to measure (& compare) the occurrence of outcomes in (different) populations (EGO compared with CGO) P Middle-aged Americans Body Mass Index (BMI) measured High BMI EG CG Low BMI Here is an example of calculating EGO and CGO when the outcome is a numerical variable rather than a categorical variable. EG and CG are categorical variables - that is high BMI and low BMI, but in this example the Outcome - blood glucose - is presented as a numerical variable - the blood glucose measures in every person in EG and in CG. You can see on this slide that when the Outcome is a numerical variable the calculation of EGO and CGO produce an average. The 1st formula is given at the bottom of the slide ..... as an aside, the Exposure/Comparison groups in this example actually started a a numerical variable - that is BMI - but you can convert numerical variables into a categorical variables. I could also have categorised the blood glucose values into 2 or more categories. EGO: Average blood glucose in EG CGO: Average blood glucose in CG high O low EGO = sum of all glucose levels in EG ÷ number in EG

goal of all epidemiological studies is to measure (& compare) the occurrence of outcomes in (different) populations (EGO compared with CGO) P Middle-aged American women receive Mammogram screening Test EG CG mammogram positive mammogram negative yes a b O EGO: Occurrence of cancer if mammogram +ve T CGO: Occurrence of cancer if mammogram -ve no Breast cancer

its all about EGO and CGO
comparing EGO & CGO Risk Ratio or Relative Risk (RR) = EGO ÷ CGO Risk Difference (RD) = EGO – CGO Number Needed to Treat/’expose’ (NNT) = 1 ÷ RD its all about EGO and CGO Measures of occurrence include: risk; rate; likelihood; probability; average; incidence; prevalence

3. GATE: identifying where errors occur in epi studies: the 2nd acronym: RAMBOMAN
Recruitment Allocation Maintenance Blind Objective Measurements We use the acronym RAMboMAN to help readers identify the major errors in epi studies ANalyses GATE frame with RAMBOMAN can be used to identify risk of error in most/all epidemiological studies

RAMBOMAN were Recruited participants relevant to the study objectives?
Study setting RAMBOMAN Eligible population recruitment process P P were Recruited participants relevant to the study objectives? who are the findings applicable to? The R is for recruitment and one of the first question one should ask when appraising a study is: were the participants recruited for the study relevant to the study objectives? You would hope they are but lets say you were wanting to determine if a rather strict and bland diet was effective in helping obese people lose weight. You may start by trying to recruit typical obese people in your community but the people who finally agree to take part may be a very small proportion of the people you invited and may be highly motivated. You carry out the study and determine that the diet is very effective. But you are left asking - just who are the findings applicable to? Perhaps only highly motivated volunteers are likely to stick to such a strict bland diet but most of the obese people you deal with, dont fit this description. So in order to answer the Recruitment question, I've provided a list of things to consider: .... Generalisability of findings will be influenced by how well recruitment is described - not always relevant

RAMBOMAN: how well were participants Allocated to exposure & comparison groups?
was Allocation to EG & CG successful? RCT: Allocated by randomisation (e.g to drugs) Cohort: Allocated by measurement (e.g. smoking) E & C measures accurate? EG & CG similar? EG CG EG CG In RCTs the goal is to allocate participants to two (or more) groups that are very similar prior to the EG & CG interventions are initiated – they would have the same occurrence of outcomes if there were treated the same In non-randomised studies the goal is to allocate participants to two ore more groups so that they are accurately categorised by their exposure/comparison status and to also measure other factors relevant to the study outcome that may differ between the groupss O O T T

completeness of follow-up compliance contamination co-interventions
RAMBOMAN P how well were Participants Maintained in the groups they were allocated to (i.e. to EG & CG) throughout the study? EG CG completeness of follow-up compliance contamination co-interventions The first M in Ramboman stands for maintenance. If the study has done a good job of allocated participants to EG and CG at the beginning of the study, you want to determine if most of the participants remained in their allocated groups. In other words were they maintained in the groups to which they were initially allocated throughout the study? This can certainly be a problem in studies that have a long follow-up period. For example many of the British doctors who were smokers in the 1950s when the study begain, stopped smoking during the study. Also in randomised trials that expect people to take study drugs or maintain a diet every day for a number of years, it can be hard work to keep participants motivated to take all their drugs or maintain their diet. To help answer this question I have provided a list of topics to consider. I'll let you read through this list yourself, as we will come back to the specific topics in subsequent presentations. If you have problems with any of the definitions, there is an excellent glossary on the Cochrane Collaboration site - just google Cochrane Collaboration glossary if you want a definition of contamination, co- intervention etc. finally one the best way to minimise differences in the level of compliance/contamination/co-interventions/follow-up is to keep participants and study staff blind to whether the participants are in EG or CG. O T

RAMBOMAN O Were outcomes measured
P Were outcomes measured blind to whether participant was in EG or CG ? EG CG As you can see, the b in Ramboman gets used twice. In the previous slide I discussed the benefits of blinding participants in relation to Maintenance and here I want comment on the importance of blinding in relation to the second M in RamboMan which stands for Measurement ofoutcomes. Unfortunately many outcome measurements are not clear cut or black and white and one of the ways we can reduce differences in the way outcomes are measured between different groups within the study is to keep the outcome assessors blind to the exp/comp status of the participants. Let me give you an example of where this may be important., if a study was investigating the effect of a new drug (the E) compared to an old drug (the C) on the outcome depression. Depression is not a clear cut diagnosis and outcome assessors normal score participants on a scale. Now if the outcome assessors are aware that particpants are on the new antidepressent, they may subconsciouly score the particpants lower on the depresession scale because they think the new drug is likely to be more effective and vice versa for the older drug. If this happened then there would be an error favouriing the new drug, keeping the outcome assessors blind to the participants exposure status is a useful way to reduce outcome measurement errors O T

Were outcomes measured
RAMBOMAN P Were outcomes measured Objectively? EG CG O T

Were the Analyses done appropriately?
RAMBOMAN P Were the Analyses done appropriately? EG CG Adjustment for confounding The AN at the end of RamboMAN stands for analyses and there are several analysis questions that are covered in the final 3 slides of this presentation. The first relates to how the investigators addressed any important differences between the EG and the CG at the beginning /during the study that could have influenced the number of outcomes over and above the specific exposure and comparison factors being investigated. Differences between EG and CG are far more likely to be present in non randomised studies than in randomised trials. As I mentioned the whole point of randomisation is to start a study with two groups that are as similar as possible, but in non ramdomised studies where participants are allocated by measurement of a specific exposure factor, there are almost certainly going to be other differences between the EG and CG that could influence the outcome. For example if we did a study of smoking and cancer today we would find that the smokers are also more likely to be exposed to a number of other risk taking behaviours than non smokers because everyone now knows that smoking is a risky behaviour and smokers are more likely to be risk takers (in regards to their health) than non smokers. So if the investigators find that smokers are at increased risk of cancer, they wont know how much is due to the smoking and how much is due to the associated risk taking behaviours which are called confounders if they also cause cancer. There are a number of ways investigators can address this very important source of error in non randomised epidemiological studies, in particular using multivariate ananalytical tecniques that mathematically attempt to make the EG and CG similar. O T

Were the Analyses done appropriately?
RAMBOMAN P Were the Analyses done appropriately? Intention to treat? EGA CGA EGC CGC This slide covers two other analytical question that are important to consider when critcally apraising a study. The first question simply asks whether the numbers you would need to calculate EGO and CGO are given in the paper and if so, whether you get a similar answer. In this course we provide you with a excel-based calculator that you can use to do these calculations. The second question asks whether all the participants were analysed in the groups to which they were initially allocated. I'm going to discuss this further under the topic 'intention-to-treat' analysis in a subequent presentation on appraising randomised controlled trials. a b O T

the 2nd formula: random error = 95% confidence interval
In this final slide on the second acronym - Ramboman - I want to introduce the 2nd formula. It is an ANalytical issue but it relates to a different type of error than all the other errors I have discussed so far, and so it has its own formula. I'm pushing the boundaries of the definition of a formula but the second formula is as follows: ' random error equals the 95% confidence interval.' There is a separate presentation in this course about random error and all I want to say at this stage is that every measurement and calculation done in epidemiological studies will be associated with a degree of random error. Increasingly, the standard way of describing the amount of random error in epidemiological measures is using the 95% CI, and every measure of EGO and CGO has a 95% CI, as does every relative risk or risk difference or number needed to treat. I've given a definition of a 95% CI in this slide and will come back to this in a subsequent presentation. So this concludes my overview of the major sources of error in epidemiological studies that need to be considered when appraising a study EGO ± 95% CI CGO ± 95% CI There is about a 95% chance that the true value of EGO & CGO (in the underlying population) lies somewhere in the 95% CI (assuming no non-random error)

Critically appraising a systematic review
the 3rd acronym: FAITH Critically appraising a systematic review Find – were all potentially relevant studies found? Appraise – were studies appraised for validity? Include – were only appropriate studies included in the final analyses? Total-up – were studies pooled appropriately? Heterogeneity – were studies too heterogeneous (i.e. too different) to pool? These points are not found in narrative reviews •Comprehensive list of all studies identified •Clear presentation of characteristics of each included study •Clear analysis of methodological quality of each included study •Comprehensive list of all studies excluded and justification for exclusion •Clear analysis of results using meta-analysis if appropriate and possible •Sensitivity analysis of synthesised data •Structured report stating aims, describing methods and materials and reporting results

4. GATE : a framework for the 4 steps of EBP
Can also be

ask access 3. appraise 4. apply [5. audit your practice]
the steps of EBP: ask access 3. appraise 4. apply [5. audit your practice] The final step of EBP is what QI is all about

EBP Step 1: ASK - turn your question into a focused 5-part PECOT question
1. Participants 2. Exposure 3. Comparison E C this slide presents the same study as in the previous slide to introduce the first acroynym PECOT which identified the 5 main components of most epi studies. 4. Outcomes yes O 5. Time no T

Participants Exposure Comparison Outcome Time frame
EBP Step 2: ACCESS the evidence – use PECOT to help choose search terms Participants Exposure Comparison Outcome Time frame Once you have written down your 5-part question - and please dont spend any time trying to make it grammatically correct - you can then use the relevant parts of the question to identify search terms. You will usually find the most important search terms under participants, exposure and outcome, althpough sometimes the comparison (for example placebo) and the time frame can provide search terms. There is a separate presentation on searching the literature in this course and I encourage you to view it.

EBP Step 3: APPRAISE the evidence – with the picture, acronyms & formulas
Recruitment Allocation Maintenance blind objective Measurements ANalyses E C O Mention RAMMbo T Occurrence = outcomes ÷ population Random error = 95% Confidence Interval

EBP Step 4: APPLY the evidence by AMALGAMATING the relevant information & making an evidence-based decision:’ the X-factor

The inspiration for the X was a student who told me ‘all you need now is an ’X’ to go with your triangle, circle and square!”

epidemiological evidence patient’s clinical circumstances
X-factor: making evidence-based decisions epidemiological evidence economic person values & preferences system features family legal community political patient’s clinical circumstances practitioner So here is the X which provides a great framework for the 4 components of EBP decisionmaking that I discussed in my introduction to EBP presentation. I call this the X-factor because decisionmakers who can put it all together well are good decisionmaker so have the x-factor and I guess that being able to put all these issues together is a way of defining what expertise is all about See excellent article: Clinical expertise in the era of evidence-based medicine and patient choice. EBM 2002;736-8 (March/April) social physical health psychological Practitioner eXpertise: ‘putting it all together’ - the art of practice Clinical expertise in the era of evidence-based medicine and patient choice. EBM 2002;736-8 (March/April)

6-page GATE CATs (word docs)

1-page GATE Calculators (excel files)

1-page GATE-lites (writeable pdf files)
all forms on Cecil EBM site

extra slides

Case-control study GATE Frame picture & 1st acronym O
smokers E C non-smokers smoking status measured cases yes Here I show the stucture of a case-control study which is basically a very efficient version of a cohort study in which the investigators wait until people in a virtual cohort have had the study outcome - these people with the study outcome are known as the cases and the investigators also typically recruit a group of non cases, ideally from the same virtual cohort as the controls.. As with randomised trials and cohort studies, there is a specific presentation on case control studies later in the course. Useful study design for investigating risk factors for disease - both causes of disease and predictors of disease (including prognostic factors) O Lung cancer controls no T Case-control study Observational study: allocated by measurement

Diagnostic test accuracy study
GATE Frame picture & 1st acronym P Middle-aged American women Measured with ‘gold standard’ for breast cancer Breast cancer positive E C Breast cancer negative positive Another study design you will come across in this course is the diagnostic study. There are 2 versions of this design that we will discuss later but here is the one that is the easiest to introduce you to Note the arrow goes sideways O Mammogram T negative Diagnostic test accuracy study

Mammogram if breast cancer Mammogram if no breast cancer
The goal of all epidemiological studies is to measure (& compare) the occurrence of outcomes in (different) populations (EGO compared with CGO) P Middle-aged American women Measured with gold standard for breast cancer EG CG Breast cancer No breast cancer a b EGO: Likelihood of +ve Mammogram if breast cancer positive O CGO: Likelihood of +ve Mammogram if no breast cancer T negative Mammogram

1st formula (with time): occurrence = (outcomes ÷ population) ÷ Time
British doctors smoking status measured Exposed Group EG CG Comparison Group smokers non-smokers I want to complete this presentation on the first formula by mentioning how time fits into the first formula. In this example, which is a cohort study conducted over 10 years, there are 2 ways of calculating EGO and CGO. In the previous slides I have used the first of the 2 formulas shown at the bottom of the slide which involves defining the time period over which the outcomes 'a' were counted - in this example lung cancer cases were counted over a 10 year period. Because this measure involves counting outcomes over a period of time, it is known as a measure of incidence in epidemiology and the first of these two formlas is known as cumulative incidence and you can see from the formula that Time is not actually used in the formula, rather the time period over which outcomes were measured is specified. So in this example the cumulative incidence of lung cancer in smokers over 10 years would be the number of lung cancer events that occurred in smokers over the 10 year study period divided by the number of smokers in the study when it begain. However in the second of the two formulas, Time is actually part of the calculation. I have presented a simplified approximate version of this calculation here because the exact version of the calculation is a little complex, but the key point is that time is in the calculation and this is called an incidence rate. In this example if you divided a/EG by 10 years, you would have calculated lung cancer incidence rate per year. Most of the studies covered in this course that measure incidence use cumulative incidence measures. Outcomes a b yes O Time Lung cancer no T 10 years EGO = (a ÷ EG) during time T (a measure of cumulative incidence) EGO = (a ÷ EG) ÷ T (a measure of incidence rate)

1st formula (with time): occurrence = (outcomes ÷ population) ÷ Time
Middle-aged American women Receive Mammogram screening Test EG CG Mammogram positive Mammogram negative When introducing the 1st equation I didn’t mention time ……if study cross-sectional – time = 1 so EGO = a/EG (horizontal arrow) yes a b O T EGO: Occurrence of cancer if mammogram +ve CGO: Occurrence of cancer if mammogram -ve no Breast cancer EGO = (a ÷ EG) at time T (a measure of prevalence)

GATE: Graphic Approach To Epidemiology

Similar presentations

Presentation on theme: "GATE: Graphic Approach To Epidemiology"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

GATE: Graphic Approach To Epidemiology

Similar presentations

Presentation on theme: "GATE: Graphic Approach To Epidemiology"— Presentation transcript:

Similar presentations

About project

Feedback