Summary of Measures and Design Hein Stigum Presentation, data and programs at: http://folk.uio.no/heins/ May-16 H.S. 1 Epidemiological measures • Frequency – prevalence – incidence How much disease? • Association – Risk difference – Risk ratio – Odds ratio More disease among exposed? • Potential impact – Attributable fraction May-16 Important cause? H.S. 2 Time E,D E → D E ← D No time prospective retrospective Present time May-16 H.S. 3 Cohorts Closed cohort start Open cohort end start Count persons end Count person-time Closed cohort with time varying covariates start end Count person-time May-16 H.S. 4 Mathematical concepts • Proportion (risk) a p N • Rate a r t • Odds p Disease o 1 p notDisease May-16 H.S. 5 Frequency measures Name Prevalence Incidence Proportion (Cumulative Incidence) Incidence Rate Odds P Type Unit Existing Cases Population risk cases/person New Cases Healthy risk cases/person New Cases Healthy Time rate cases/person-time odds cases/person IP IR odds May-16 Equation p Disease 1 - p NotDisease H.S. 6 ! Disease frequency depicted 100 Existing Cases t a New Healthy Healthy*time 0 50 Cases time H.S. ASSOCIATION MEASURES May-16 H.S. 8 Association measures • More disease among exposed? – Compare frequency among exposed1 and unexposed0 – Difference: – Ratio: RD=IP1-IP0 RR=IP1/IP0 Frequency 0=no effect 1=no effect Association or Effect Difference Ratio Risk Risk Difference, RD Risk Ratio, RR Rate - Rate Ratio, RR, IRR, (HRR) Odds - Odds Ratio, OR May-16 H.S. 9 Adjusted measures Remove the effect of confounders in regression models Frequency Association or Effect Difference Ratio Linear-binomial Risk RD: RR: Rate - IRR: Odds - OR: Log-binomial Cox, Poisson Logistic RD glm y x1 x2 x3, family(binnomial) link(identity) RR glm y x1 x2 x3, family(binomial) link(log) OR glm y x1 x2 x3, family(binomial) link(logit) Cox regression glm y x1 x2 x3, family(poisson) link(log) IRR Linear-binomial Log-binomial Logistic Cox Poisson glm=generalized linear models May-16 H.S. 10 Being bullied, 3 models glm bullied Island Norway Finland Denmark sex age, family(binomial) link(logit) glm bullied Island Norway Finland Denmark sex age, family(binomial) link(log) glm bullied Island Norway Finland Denmark sex age, family(binomial) link(identity) Bullied=17% May-16 H.S. 11 Designs • Aims – Disease occurrence – Exposure-Disease association • Designs – Cross-sectional studies – Cohort studies – Case-control studies • Case-Cohort • Nested Case Control • Traditional Case Control May-16 H.S. 12 THE 2 BY 2 TABLE May-16 H.S. 13 The 2 by 2 table Exposure + - Disease + 100 100 10 100 OR= 1 1 1 1 se(ln( OR )) a b c d .01+.01+.1 +.01 =.13 10.0 The lowest number sets the precision To increase power: Cohort: balance exposure Case-Control: balance disease May-16 H.S. 14 COHORT May-16 H.S. 15 Disease frequency depicted 100 Existing Cases t a New 50 Cases Healthy Healthy*time 0 = Risk time time May-16 H.S. 16 Cohort: Risk, Odds and Rate Full Cohort, 3 year Risk: CHD Exercise Inactive + 100 800 1900 7200 2000 8000 10000 Frequency Risk Odds 0.05 0.05 0.10 0.11 Association RD RR OR -0.05 0.50 0.47 0 1 1 Balance? Full Cohort, Rate: CHD Exercise Inactive May-16 + 100 800 1900 7200 3 years follow up time Risk time Frequency Rate 0.017 0.035 5 850 22 800 28 650 H.S. Association IRR 0.49 1 17 Case-Control studies: 1) Case-Cohort Inside an existing cohort 2) Nested Case-Control At the end of an imaginary cohort 3) Traditional Case-Control 1) CASE-COHORT May-16 H.S. 18 Case-Cohort Existing cases t . . . . New cases a+b controls Healthy N1+N0 Healthy*time py start c+d end time May-16 H.S. 19 Case-Cohort, Risk Full Cohort, Risk: CHD Exercise Inactive N + 100 1 900 800 7 200 2 000 8 000 10 000 Frequency Risk 0.05 0.10 Association RR 0.50 1 Case-Cohort, Risk: CHD Exercise Inactive N + 100 1 900 800 7 200 2 000 8 000 10 000 Controls f=0.1 200 800 Sample controls from healthy at start May-16 H.S. Frequency Pseudo Risk 0.50 1.00 Association RR 0.50 1 In practice: Count: person time Analyze: Cox model 20 2) NESTED CASE-CONTROL May-16 H.S. 21 Nested Case-Control Existing cases . case . case controls Healthy N1+N0 . . t New cases a+b controls Healthy*time py . risk set c+d . risk set start end time May-16 H.S. 22 Nested Case-Control, Rate Full Cohort, Rate: CHD Exercise Inactive Risk time + 100 1900 800 7200 5 850 22 800 28 650 Frequency Rate 0.017 0.035 Association IRR 0.49 1 Nested Case-Control: CHD Exercise Inactive + 100 800 Risk time 1900 7200 5 850 22 800 28 650 Controls f=0.1 585 2 280 Sample controls from risk time May-16 H.S. Frequency Association Pseudo Rate IRR 0.171 0.49 0.351 1 Analyze: Conditional logistic model 23 3) TRADITIONAL CASE CONTROL May-16 H.S. 24 Traditional Case-Control Existing cases Healthy N1+N0 t Healthy*time py start . . . . New cases a+b controls c+d end time May-16 H.S. 25 Traditional Case-Control Full Cohort, Odds: CHD Exercise Inactive N + 100 1 900 800 7 200 2 000 8 000 10 000 Frequency Odds 0.05 0.11 Association OR 0.47 1 Trad. Case-Control, Odds: CHD Exercise Inactive + 100 1 900 800 7 200 Controls f=0.1 190 720 Frequency Pseudo Odds 0.53 1.11 Sample controls from non-disease at end May-16 Association OR 0.47 1 Analyze: Logistic model H.S. 26 Case-Control studies • Cohort studies – Measure the exposure experience of the entire population • Case-Control studies – Measure the exposure experience of a sample of the source population of cases – Key assumption • Sample controls independent of exposure ( same sampling fraction) – Prospective or retrospective May-16 H.S. 27 Summing up • Measures Frequency Association or Effect Difference Ratio Risk Rate ! Odds RD RR - IRR - OR • Designs – Cohort – Case-Control full cohort all cases + sample of healthy • Case-Cohort • Nested Case-Control • Trad. Case-Control May-16 sample at start sample during sample at end H.S. of follow up 28