Presentation on theme: "PP (Study Design) for 2nd Year"— Presentation transcript:
1 PP (Study Design) for 2nd Year Dave GarberaF1 Arrowe Park Hospital
2 Learning Objectives Population Measures Study Design Statistical DevicesProblems with AnalysisSensitivity, Specificity and Positive Predictive Value
3 Why PP?Provides a good understanding of the basics of Evidence Based MedicineUnderstand study designs, statistics and the strength of evidenceShows you how best to manage your patientEXAMS!Paper 1 (30/150) and Paper 2 (30/100)Critical Analysis of an article4 weeks preparation time
4 Types of Data Population Data Health Event Data CensusDeprivation IndexBirth/Death ratesHealth Event DataHospital Episode StatisticsNational Cancer RegisterGP Research DatabaseThis data allows you to assess NEED
5 Prevalence and Incidence Number of people in a population with a disease at any given point in timeE.g. The prevalence of asthma in Liverpool now is 40 per 1000 peopleTells you how widespread a disease isIncidenceNew cases of a disease in a given time frameE.g. The incidence of asthma in Liverpool from 1 January – 31 December 2012 was 5 per 1000 peopleTells you about RISKWhen would you see a high prevalence but low incidence and vice versa?
6 Risk Absolute Risk Relative Risk The risk of getting a disease in a given populationE.g. the risk of having an MI in Liverpool is 1 in 50Relative RiskThe probability of getting a disease in one group compared to anotherRelative Risk = exposed groupnon-exposed groupE.g the risk of having an MI in smokers is 1/5 and the risk in non-smokers is 1/20, so the relative risk is 4.
7 Risk Reduction Absolute Risk Reduction Numbers Needed to Treat The reduction in absolute risk when an intervention is applied to a population groupE.g. Introduction of a new drug reduces risk of MI from 20% to 10%ARR = 10% = 0.1Numbers Needed to TreatThe number of people that must be treated using a particular intervention to prevent a bad outcome1 / ARR1 / 0.1 = 1010 people must be treated with the new drug to prevent one MI
9 Study Design Case Report Case Series Population Case Series HierarchyCase ReportCase SeriesPopulation Case SeriesCross-sectional StudyCase Control StudyCohort StudyRandomised Controlled Trial
10 Observational Studies Everything except RCTCase ReportSingle case study from one patientCase SeriesSeries of single patient reportsPopulation Case SeriesCase series in a defined geographical area
11 Observational Studies Cross-sectional StudyLooks at one characteristic at a point in timeAllows calculation of prevalenceCase Control StudyThese studies are RETROSPECTIVECompares those with the disease to those withoutCANNOT PROVE CAUSALITY
12 Case Control Studies Smoker? Crohn’s Non-smoker? 1980 Present Smoker? Don’t have Crohn’sNon-smoker?
13 Observational Studies Cross-sectional StudyLooks at one characteristic at a point in timeAllows calculation of prevalenceCase Control StudyThese studies are RETROSPECTIVECompares those with the disease to those withoutCANNOT PROVE CAUSALITYCohort StudyPROSPECTIVE studiesFollow two groups and record outcomes
14 Cohort Studies Smokers Get Crohn’s Don’t get Crohn’s Present 2020 Non-smokersGet Crohn’sDon’t get Crohn’s
15 Randomised Controlled Trial The gold standardThe only trial where YOU interveneDirect comparison of two standardised groupsControl group and interventional groupMost effective when patients researchers don’t know which group is whichBLINDING
16 Randomised Controlled Trial Group receiving current best treatmentMeasure mortality rate = 20%Present2020Group receiving new experimental treatmentMeasure mortality rate = 10%Shows that the new drug reduces mortality by 10% (absolute risk reduction)
18 Measures of Central Tendency MeanUseful if all values are similar 50, 51, 53, 53, 54, 56, 56MedianEliminates extreme values 22, 51, 53, 53, 54, 56, 98ModeAnalyses peaks in data , 22, 22, 51, 53, 98, 98, 98
19 Standard Deviation You don’t need to know how to calculate SD! Allows you to see the spread of dataA small SD shows that data is central around the mean and is, therefore, accurateA large SD shows data dispersion across a range of values and is, therefore, innaccurateYou don’t need to know how to calculate SD!
21 Statistical Devices 95% Confidence Intervals Student’s t-test Gives the range of data you are confident the true result lies inE.g. You expect 50% of the population to vote Labour95% CI says the true value lies between 45% and 55%Student’s t-testStatistical test to determine how significant the results of a study areUses a value known as a p-valueA p-value of less than 0.05 shows statistical significance and demonstrates that the probability the results are due to chance is less than 5%A value of over 0.05 means study results are invalid
22 Measures of RiskOddsThis is a measure of how likely something is to happenIT IS NOT THE SAME AS PROBABILITYIt describes the chance of something happening versus it not happeningE.g. the probability of rolling a six on a dice is 1 out of 6The odds of rolling a dice is 1 out of 5
23 Odds Ratio This is the ratio of two odds It tells you the odds of an event happening in one group compared to the same event in another groupE.g we roll a dice. What is the odds ratio of rolling a six both times?(1/5) divided by (1/5) = 1The odds are the same in both groups
24 Risk RatioThis is similar to odds ratio, except probabilities are usedThe risk ratio (relative risk) tells you the risk of developing a disease related to a given risk factorIt is calculated by dividing the exposed group by the non-exposed groupIf there is no increase in risk, the risk ratio is 1E.g. 30 people in the smoker group develop Crohn’s and 6 people in the non-smoker group develop Crohn’s30/6 = 5 – You are 5 times more likely to get Crohn’s if you smoke
25 cannot, therefore, estimate risk Important!Odds ratio is used for case control studiesRelative Risk is used for cohort studiesYou cannot use relative risk for case control studies because they do not prove causality, and youcannot, therefore, estimate risk
27 Bias Selection Bias Volunteer Bias Information Bias Recall Bias Bias in selecting study participantsVolunteer BiasBias when only certain types of people volunteerInformation BiasBias resulting from errors in in measurements of dataRecall BiasBias created when patients are asked to remember information
28 ConfoundersFactors that may skew results as they correlate with both variables
29 Sample Size This is important (and actually quite obvious!) You cannot draw pertinent conclusions from your data unless the sample size is large enoughThis helps to eliminate results that are due to chanceThere are ways of working out how large your sample size must be for any given studyCovered in 3rd Year Critical Thinking Module
30 Error Type 1 Type 2 Due to factors such as bias or chance Finding a difference between two datasets that isn’t really thereType 2Missing a significant difference between two datasetsDue to factors such as bias or chance
31 Sensitivity and Specificity Measures how good a screening test is at identifying TRUE cases of a diseaseTrue positives / True positives + false negatives (x100)SpecificityMeasures how good a screening test is at identifying healthy individuals with no diseaseTrue negatives / True negatives + false positives (x100)Very subtle difference between the two measurements
33 Positive Predictive Value Tells us how good a diagnostic test is at identifying positive patientsSimilar to sensitivityRepresented as a decimal rather than percentageCalculated by True Positives / All positives70 / 77 = 0.91A result of 1 would indicate a perfect testNegative predictive value also works in exactly the same way
34 Other parts of PP Bradford-Hill criteria Wilson and Jugner criteria Maxwell’s criteriaImpairment, Disability and HandicapEcological FallacyUtility and Opportunity CostRatesKaplan-Meier PlotsPrevention ParadoxCase MixWe can go through these another time if you’d like!