…it’s really interesting

Slides:



Advertisements
Similar presentations
How would you explain the smoking paradox. Smokers fair better after an infarction in hospital than non-smokers. This apparently disagrees with the view.
Advertisements

PP (Study Design) for 2nd Year
1 Case-Control Study Design Two groups are selected, one of people with the disease (cases), and the other of people with the same general characteristics.
CRITICAL APPRAISAL Dr. Cristina Ana Stoian Resident Journal Club
Epidemiology in Medicine Sandra Rodriguez Internal Medicine TTUHSC.
Statistics By Z S Chaudry. Why do I need to know about statistics ? Tested in AKT To understand Journal articles and research papers.
Statistics for Health Care
The Bahrain Branch of the UK Cochrane Centre In Collaboration with Reyada Training & Management Consultancy, Dubai-UAE Cochrane Collaboration and Systematic.
Are the results valid? Was the validity of the included studies appraised?
 Mean: true average  Median: middle number once ranked  Mode: most repetitive  Range : difference between largest and smallest.
Multiple Choice Questions for discussion
OKU 9 Chapter 15: ORTHOPAEDIC RESEARCH Brian E. Walczak.
Academic Viva POWER and ERROR T R Wilson. Impact Factor Measure reflecting the average number of citations to recent articles published in that journal.
Basic statistics 11/09/13.
Understanding real research 4. Randomised controlled trials.
EBCP. Random vs Systemic error Random error: errors in measurement that lead to measured values being inconsistent when repeated measures are taken. Ie:
Epidemiology: Basic concepts and principles ENV
Causal relationships, bias, and research designs Professor Anthony DiGirolamo.
Measuring associations between exposures and outcomes
Medical Statistics as a science
Diagnostic Tests Studies 87/3/2 “How to read a paper” workshop Kamran Yazdani, MD MPH.
Organization of statistical research. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and.
Compliance Original Study Design Randomised Surgical care Medical care.
BIOSTATISTICS Lecture 2. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and creating methods.
Biostatistics Board Review Parul Chaudhri, DO Family Medicine Faculty Development Fellow, UPMC St Margaret March 5, 2016.
2 3 انواع مطالعات توصيفي (Descriptive) تحليلي (Analytic) مداخله اي (Interventional) مشاهده اي ( Observational ) كارآزمايي باليني كارآزمايي اجتماعي كارآزمايي.
Measures of disease frequency Simon Thornley. Measures of Effect and Disease Frequency Aims – To define and describe the uses of common epidemiological.
Epidemiological Study Designs And Measures Of Risks (1)
Chapter 9: Case Control Studies Objectives: -List advantages and disadvantages of case-control studies -Identify how selection and information bias can.
GP ST2 Group, 28/9/11 Tom Gamble
DR.FATIMA ALKHALEDY M.B.Ch.B;F.I.C.M.S/C.M
Health Indicators.
Instructional Objectives:
Statistics 200 Lecture #9 Tuesday, September 20, 2016
Sample size calculation
Study Designs Group Work
An Idiots Guide to Statistics Curriculum 3.6
Question 1 A new ‘Super test’ claims to have a superb capability to diagnose disease X. Its sensitivity is 99% and specificity is 90%. Which of the following.
Present: Disease Past: Exposure
How many study subjects are required ? (Estimation of Sample size) By Dr.Shaik Shaffi Ahamed Associate Professor Dept. of Family & Community Medicine.
Epidemiologic Measures of Association
How to read a paper D. Singh-Ranger.
Epidemiological Studies
Biostatistics Case Studies 2016
Class session 7 Screening, validity, reliability
Relative Values.
Understanding Results
Interventional trials
ASPIRE Workshop 5: Application of Biostatistics
ASPIRE Workshop 5: Application of Biostatistics
Copyright © 2003 Delmar Learning, a Thomson Learning company
NAPLEX preparation: Biostatistics
Lecture 1: Fundamentals of epidemiologic study design and analysis
Inferential statistics,
Public Health Phase 3A Abigail Aitken
11/20/2018 Study Types.
Review – First Exam Chapters 1 through 5
Narrative Reviews Limitations: Subjectivity inherent:
Interpreting Basic Statistics
Appraisal of an RCT using a critical appraisal checklist
Measures of risk and association
Epidemiological Measurements of health
How to assess an abstract
Interpreting Epidemiologic Results.
ASPIRE Workshop 5: Application of Biostatistics
HEC508 Applied Epidemiology
Cohort and longitudinal studies: statistics
Dr Luis E Cuevas – LSTM Julia Critchley
Risk Ratio A risk ratio, or relative risk, compares the risk of some health-related event such as disease or death in two groups. The two groups are typically.
Basic statistics.
Presentation transcript:

…it’s really interesting PP …it’s really interesting

Things to try and cover Vital statistics Confidence Intervals Rates and Ratios OR vs RR Specificity, sensitivity, PPV, NPV Extracting info out of flow charts to make calculations Apply to some relevant questions Other things worth having a look at

Vital statistics P value: probability of event occurring by chance Null hypothesis – no significant difference Chi-sq: Hypothesis test for categorical data (O/E) Type I error ~ False positive ~ alpha value (p value) ↑p value will ↑false positive Type II error ~ False negative ~ beta value Statistical power = 1- Beta Sample size is key Power detecting what you want to detect ↑sample size = ↑Power

Confidence intervals Confidence Interval: a range which is likely to contain the popn mean Usually 95% CI quoted i.e. 95% of time the mean will be in this range Does not denote significance unlike a p value Related to sample size: ↑sample size equates to narrow CI

Options: a) 95% confidence interval b) statistical measure of dispersion of continuous data c) hypothesis test for categorical data d) statistical power- ability of study to find what you want to find e) type I error f) type II error g) statistical significance level This is: the failure to detect a true difference the Chi-squared test 3. the statistical assessment of a statement formulated in the negative, such as “there is no association between osteoarthritis and nutrition” 4. illustrated by looking at the difference between the 5th and the 95th centile of a set of blood pressure measurements in millimetres of mercury 5. a measure of the precision of an estimated proportion of patients on an orthopaedic waiting list who are likely to be over 65 years old f) c) c) b) a)

Rates and raw numbers Learn these- one of the favoured questions Most on the hand out Fertility and death rates Will always be over a specified time period Often specified per 1000 population Rates will be over time / mention proportion Raw numbers if asking for things to make a rate i.e. if ask for the numerator / denominator Almost certainly ask what represents numerator or denominator Remember: Numerator Denominator

Options: For a given geographic area: a) number of women at a specified time b) number of livebirths in specified time period c) number of stillbirths in specified time period d) number of deaths in the first month of life in specified time period e) average number of women aged 15-44 years in specified time period (mid-year estimate) f) number of deaths in the first year of life in specified time period g) total number of births (live and still) in specified time period This is the: 1. denominator for general fertility rate Things you know: It’s something to do with birth i.e. not death! Exclude d) f) It’s general (all births regardless of outcome). Exclude b) c) It’s the denominator so should be lower portion of equation It’s regarding fertility (always women aged 15-44 years). Exclude a) What’s left? E) & G) Does G fit the definition? No as it’s the numerator Does E fit the definition? Yes General fertility rate = no. live and still births in a time period number of women 15-44yrs in spec time

Options: For a given geographic area: a) number of women at a specified time b) number of livebirths in specified time period c) number of stillbirths in specified time period d) number of deaths in the first month of life in specified time period e) average number of women aged 15-44 years in specified time period (mid-year estimate) f) number of deaths in the first year of life in specified time period g) total number of births (live and still) in specified time period This is the: 2. denominator for infant mortality rate Things you know: It’s to do with infants. Exclude all options with women i.e. a) e) Infant deaths: first year of life. Exclude c) d) It’s a denominator so can’t be what’s defined in the question i.e. number of infant deaths. Exclude F) What’s left? B) & G) Does G fit the definition? No Does B fit the definition? Yes(must know the definition here) Get to a 1in2 chance rather than 1in7 by using these approaches

Options: For a given geographic area: a) number of women at a specified time b) number of livebirths in specified time period c) number of stillbirths in specified time period d) number of deaths in the first month of life in specified time period e) average number of women aged 15-44 years in specified time period (mid-year estimate) f) number of deaths in the first year of life in specified time period g) total number of births (live and still) in specified time period This is the: 5. numerator for infant mortality rate Use previous workings and apply to this Previously got to B), F) and G) It’s a numerator so will be what’s defined in the question i.e. number of infant deaths. Select F)

Study design (observation to association to causality) Descriptive (hypothesis generating) Case report (single event / single patient Case series (collection of patients) Cross-sectional (survey / population studies) Correlation (aka ecological) Analytical (Observational looking at hypothesis) Case-control (Disease / no disease with exposures) Cohort (follow up over time) Meta-analysis: Combine results of many studies Experimental (test hypothesis) RCT (Gold standard vs new intervention)

Relate to Bradford-Hill criteria (causality) Temporal relationship – Does the cause precede the effect Plausibility – Is the association consistent with other knowledge Consistency – Have similar results been shown in other studies Strength – What is the strength of the association between the cause and the effect (relative risk) Dose-response relationship – Is increased exposure to the possible cause associated with increased effect Reversibility - Does the removal of a possible cause lead to reduction of disease risk Study design – Is the evidence based on a strong study design Judging the evidence – How many lines of evidence lead to the conclusion

Options: a) undertaking record linkage: bringing two pieces of information together b) population screening: opportunistic / targeting high risk pops c) Surveillance: regular / continuous d) undertaking a case-control study e) matching: e.g. making sure that two groups in case control are as similar apart form disease f) undertaking a service evaluation study g) reporting a case study This would be illustrated by: 1. estimating point prevalence regularly for disease control purposes 2. publishing about an episode of adverse reaction to a new oral contraceptive 3. checking regularly the weight of pregnant women 4. offering a test for Down’s syndrome to all older pregnant women checking childhood immunization uptake using a continuous system   checking the blood pressure of all women of child-bearing age whenever they consult a general practitioner c) g) c) b) c) b)

Approaches for minimising bias Confounder: factor related to both exposure and outcome- not an approach but important to identify so that you can adjust Adjustment: statistical approach to minimise population differences Matching: try and make cases and controls as similar by demographics e.g. age, sex, ethnicity Stratification: sub-sections of populations (homogenous groups) Randomisation: Used in clinical trials Standardisation: used commonly to account for confounding effect of age

Standardisation

Direct Standardisation Standard population structure Direct method of standardisation - calculation of the number of expected deaths for countries A and B applied to a standard population.

Indirect Standardisation Indirect method of standardisation is used to calculate how many deaths would be expected in Country B if it had the same age-specific mortality rates as Country A.

Direct Vs Indirect Standardisation Break down age spec groups in both pops into mortality rates Apply rate to standard population numbers given Standard mortality rate Number of deaths / 1000 if were in standardised population structure DIES Death rate Index population Exposure Sample populations Comparative Mortality Ratio =Pop A Pop B Can compare age-adjusted rates (but not crude rate) Pop A has x% mortality > Pop B Used when age spec mortality rate unavailable (more realistic) Set of age spec rate from standard population to unkn population Get a estimated death / morbidity rate for each age group Add altogether to have total estimated deaths Compare this to what observed Observed number of deaths Expected number of deaths Standardised Mortality Ratio (SMR) SMR = 100 No deviation SMR < 100 Less comparative deaths SMR > 100 More comparative deaths

OR Vs RR Odds is number of times event happens vs not Odds ratio = Odds of exposure in cases Odds of exposure in controls OR used in case control as retrospective selection based on outcome not exposure In a way you can calculate the probability of exposure given outcome / status Risk is probability an event will occur Absolute risk: probability that an event will occur in spec time Attributable risk is the excess risk due to the risk factor Relative risk (RR) = Risk in exposed group Risk in unexposed group RR used in cohort as prospective selection and outcome observed with regards to exposures Can estimate probability of outcome from exposure

Odds ratio: It’s comparing outcome in exposure groups Exposed to risk factor (e.g. Morbid obesity) BMI>40 BMI<40 Total Disease status T2DM A 15 B 6 A+B 21 Control C 4 D 18 C+D 22 A+C 19 B+D 24 Odds ratio: odds of T2DM if morbid obese (exposure) / odds of T2DM if not (non exposure) Odds of T2DM if morbidly obese = A/C Odds of T2DM if not morbidly obese = B/D Odds ratio = (A/C)/(B/D) = AD/BC OR=1: no difference OR>1: T2DM more likely in exposed group OR<1: T2DM less likely in exposed group

Confidence intervals and ORs If 95%CI intersects 1 then it is not significant If 95%CI of two groups overlap they are not significant

Relative risk: Risk of disease given exposure Exposed to risk factor (e.g. Morbid obesity) BMI>40 BMI<40 Total Disease status T2DM A 15 B 6 A+B 21 Control C 4 D 18 C+D 22 A+C 19 B+D 24 RR: Risk of T2DM in BMI>40 (exposure) / Risk of T2DM in BMI<40 (non exposure) Risk of T2DM in BMI>40 (exposure) = A/A+C Risk of T2DM in BMI<40 (non exposure) = B/B+D RR = (A/A+C)/(B/B+D) RR=1: no difference RR>1: Obesity increases T2DM risk RR<1: Obesity reduces T2DM risk

OR != RR Both measure likelihood of event between groups Very similar in low prevalence but as it increases they are not A 2% mortality Vs B 1% mortality RR of A = 2 (what most people would deduce) OR = 2.02 i.e. A (2/98)/(1/99) Most common diseases reflect this A 50% mortality Vs B 25% mortality RR of A= 2 (what most people would deduce) OR = 3 because its always chance of getting versus not in each group i.e. A (50/50)/(25/75)=1/0.333=3 A 90% mortality Vs B 10% mortality RR of A = 9 (what most people would deduce) OR = 81 because its always chance of getting versus not in each group i.e. A (90/10)/(10/90)

Absolute risk reduction and NNT Absolute risk reduction (ARR) = risk(non-exposure) – risk(exposure) Number needed to treat (NNT) average number of patients who need to be treated to prevent one additional bad outcome effectiveness of a health-care intervention NNT = 1 / ARR *Remember dealing with patient numbers so round up NNT=1 everyone treated recovers ↑NNT = ↑poor intervention i.e need to treat lots of people to gain one recovery

ARR & NNT ARR = 14-8 = 5 (0.05) Therefore NNT = 1/0.05 = 20

a) sample size calculation b) calculation of odds c) calculation of a relative risk d) surveillance e) calculation of an odds ratio f) minimizing bias in a cohort study g) matching This is: 1. illustrated by comparing incidence between exposed and non-exposed groups 2. illustrated by a school health service implementing an ongoing system of hearing tests at school entry 3. illustrated by focusing on ‘loss to follow-up’ 4. a way of adjusting for confounders 5. crucial for deciding whether a study has the power to detect a particular difference c) d) f) g) a)

Sensitivity, specificity, PPV, NPV True positive: Correctly identified as +ve False positive: Identified as +ve but actually -ve True negatives: Correctly identified as -ve False negatives: Identified as -ve but actually are +ve Sensitivity: correctly identifying true positives Specificity: correctly identifying true negatives Relevant for population screening etc Don’t want a test that either does not pick up disease or gives people reason to worry for no reason

Sensitivity, specificity, PPV, NPV Disease / Outcome DVT No DVT Total Test +ve True Positive A False Positive B PPV A/A+B -ve False Negative C True Negative D NPV D/C+D Sensitivity A/A+C Specificity B/B+D

Sensitivity aka True positive rate Disease / Outcome DVT No DVT Total test +ve True positive A -ve False Negative C Sensitivity A/A+C aka True positive rate Measures those correctly identified as positive e.g., the percentage of sick people who are correctly identified as having the condition Quantifies avoiding false negatives Sensitivity 100%: all DVTs identified as DVTs A negative result in a test with high sensitivity is useful for ruling out disease (SnOUT) A positive result in a test with high sensitivity is not useful for ruling in disease Overdiagnosis

Specificity aka True negative rate Disease / Outcome DVT No DVT Total Test +ve False Positive B -ve True Negative D Specificity B/B+D Specificity aka True negative rate Measures those correctly identified as negative e.g., the percentage of healthy people who are correctly identified as having the condition Quantifies avoiding false positives Specificity 100% - all non-DVTs are identified A positive result in a test with high specificity is useful for ruling in disease (SpIN) A negative result in a test with high specificity is not useful for ruling out disease Underdiagnosis Does not take into account false negatives

PPV and NPV PPV & NPV dependent on prevalence describe performance of diagnostic test Sensitivity and specificity are not PPV = TP / (TP + FP) will increase proportionally with prevalence If we test in a high prevalence setting, it is more likely positive test truly have disease vs low prevalence population NPV = TN / (FN + TN) will decrease with increased prevalence ↑prevalence ~ ↑PPV ↑prevalence ~ ↓NPV

Disease / Outcome DVT No DVT Test +ve 67 (52+25) -ve True +ve: VTE predicted by tests and VTE found= 67 25 (D-dimer +ve) + 52 (Wells score >4)

Disease / Outcome DVT No DVT Test +ve 67 (52+25) -ve 4 False –ve: VTE not predicted by test but VTE found = 4 4 subjects with VTE but had Wells score ≤4 & D-dimer –ve

Disease / Outcome DVT No DVT Test +ve 67 (52+25) -ve 4 268 (272-4) True –ve: VTE not predicted and not found = 268 272 (Wells score ≤4 & D-dimer –ve ) – 4 (VTE found)

Disease / Outcome DVT No DVT Test +ve 67 (52+25) 249 (125+124) -ve 4 268 (272-4) False +ve: VTE predicted but VTE not found = 249 124 (Wells score >4 but no VTE) = 176 (Wells score >4) – 52 (VTE found) 125 (D-dimer +ve but no VTE) = 150 (D-dimer +ve) – 25 (VTE found) Total = 124 + 125

Sensitivity = TP / All with DVT (TP+FN) = 67/71 (94%) Disease / Outcome DVT No DVT Total Test +ve 67 (TP) A 249 (FP) B PPV A/A+B -ve 4 (FN) C 268 (TN) D NPV D/C+D Sensitivity A/A+C Specificity B/B+D Sensitivity = TP / All with DVT (TP+FN) = 67/71 (94%) Specificity = TN / All without DVT (TN+FP)= 268/517 (52%) Positive predictive value = TP / All that were positive (TP+FP) = 67/316 Negative predictive value = TN / All that were positive (TN+FN) = 268/272

Other things to cover / consider Look up options of answers on vital – very similar questions often used Definitions / explanations on health knowledge website Watch out for double negative questions (especially in null hypothesis based questions) Have a look at immunisation schedule (make sure updated version) Normal distribution, mean, median mode, range Intention-to-treat (include in analysis of clinical trial) 3Es 3As Look up different types of bias Favourites are lead-time / length-time Primary, secondary and tertiary prevention Screening is secondary (as they already have it – picking up disease already there) Screening diabetics for retinopathy is tertiary (they already have DM and minimising impact of complications) Health economics (on hand out)

Thanks hlrsmit4@liv.ac.uk http://www.healthknowledge.org.uk/public-health-textbook