Exam 3 SG BIOSTATS PDF

--  Simple & Multiple Linear Regression *assess differences across test a control groups , accounting for factors that do not "average out" could cause bias.- > single or multiple predictors (X) -* ↳indep...

--  Simple & Multiple Linear Regression *assess differences across test a control groups , accounting for factors that do not "average out" could cause bias.- > single or multiple predictors (X) -* ↳independent variable Correlation vs. Regression need to show causation # > only one Y value (outcome) & ↳ ↳ dependent variable description/inferential Prediction/Designed ↳ continuous (#) statistics experiments * linear ↳ R2-adjusted (more acc than R2). ↳ Y 112X +5 Glinexplanatory = ↳ intercept always Y axis dependents preciatis * simple linear regression ↳ estimating sloped intercept of a trendline thrn the middle of When Reading Data! a scatterplot bX-Pluginford multiple R correlation R2 0 1 > 4Y = - = = a+ ↳ close to 0-poor fit (no pattern( ↳ 70 3 = weak crefficient intercept coeff ↓.. ↳ close to l-excellent fit > ALWAYS Use - Adjusted R2 ↳ ) < 0 33 (precise pattern tightly clustered poor fit > - ↳ Null Ho = slope of RL is Zero. = ↳ < 0 33-0 04 > moderate , - ↳no sig linear relationship bt 2variables... 4) 0 ne > strong fit. - > if ANOVA given-look at p-value * Ordinary least squares (OLS) ↳ 05 significant = ↳ simplest form of estimation. for regression < look at coefficients * analysis > use P-value of variables * not of intercept * cofounders * adjust for cofounders > factor ↳ 1 05 preventing appropriate ↳include them in. interpretation of statistical result multiple linear regression as control variables. no causation M independent alone > dependent Multiple Linear Regression (MLR) = more than 1 indvariable * accounts for cofounders & assess for mediation cofounder Enervou Y= a + b ,x, + bzXz... + ↓ or controlled values > mediators intercept - type of cofounder ↳ intermediate variable in a * removes possible source of bias causal chain relating a * use p-value of variables ind. a dep. ↳ ) < 0 05 =. Sig > predictor value. ind > changes on mediator > ↳ changes on dep. * > 0 05 not sig > Control value - ->... =. > if a variable moves from predictor - control in MRS, eX : Hi BP it is a 71 confounding variable > hull : Ho = atleast one of bl bz , be not equal to zero. smoking CVD mediator * results that omit important control variables are considered biased/inefficient  Binary LogisticRegression Dependent Variable probability of occuring = ↑ binary or dichotomous probability of not = 1- Y yes/no 'deathlalive can recode variables to be binary eX : All levels 27 = has diabetes * 9ggo Cl & does NOT include I All levels < 7 = no diabetes ↳ reject null *99%0 Cd includes I ↳ ) BLR fail to reject - - probability of Yi = I Ho BK = 0 : J 05 = fail 2 reject ·. I = has o= does not have.os = reject 4 * alwaysIt od 1 what is I ? > interval = 1. 002-1 016. logit or ii = odds of having ↳ doesn't include I · odds. ↳reject null ! probability of success probability of failure use 95% confidence limits odds ratio IP & - odds of Yi 1 = 1. 002-1 011. if >I coefficient will be + ↓ B> O > OR greater than one - = hired - - - - BLO - OR smaller than one... o = not hired experience  StudyDesign Overview EVIDENCE based medicine pyramid observational studies strength & quality ↑ as moves up "Observed" Critical Appraisal 1. Conort Meta-analyses strongest < prospective > retrospective systematic reviews. Cross-sectional 2 (surveys critically appraised lit. evidence based practice guidelines. case control 3. 4 case series a report experimental studies. randomizedcontrollers as a non-randomized controlled Trials Case Series Case Report observational studies > descriptive > detailed report of diagnosis , · novel or unusual char. frmt , response & follow up of cohort studies individual pt a) prospective b) retrospective - seen over short period of time (small # of Pts) educational , but selection bias · hypothesis generating · rare manifestations observational , case series or studies * group of case reports (3-10) (case control) individual case reports weakest Case Control * MATCHING > begins w/ cases of disease (outcome) amt. smoke observational · looks backward' retrospectively' to ex : 200 cases widisease 50% detect risk factors 200 cases wo disease starts Wh outcome & - 10 %0 · * case : looks at risk factors > look at smoking in pts ↳ind WI disease or outcome ↳ can say. smoking causes !! * control : ↳ ind wo disease or outcome. cross Sectional Conort Studies < analyze data from ind at one point in time. >a group of people who have something surveys epidemiology in common exc call residents dask now much they · · follow forward or 'Prospective' to see what will happen ↳ will exercise weekly & if they have been diagnosed they get? WI heart disease · commonality may= risk factor for disease or other health effect · ex : smokers vs non smokers. 1. objective is to establish 2 groups ↳ exposed vs nonexposed. observational ↳followed either in past or future startswt risk factory > Prospective looks for consequences > follows forward in time Retrospective < also uses historical info (database)  experimental studies RCT * new drug * crossover studies ~ provides strongest evidence causation > same group for intervention & control > involving humans= clinical trials groups treated same in every way except · washout period = no trmt received 1. uncontrolled : for intervention * SWAP ↳ no comparison (control · each pt serves as their own control 1. Single blind : ↳ less validity ↳pts don't know. nonrandomized : 2. double blind : 2 ↳ trials we controls 4) Pt & investigator don't know nonrandomized trials 3 RCT. :. triple blind : 3 > char not , equally distributedbt ↳ trials wi control a randomization 4 Pt , investigator intervention & control , a data analysis personnel unaware questionable conclusions. 4 crossover : * randomization= causation · No causation ↳ trials w/self controls & randomization trials w/ historical controls > use results of another investigators research for comparison · must determine if other factors have changed since controls were treated - L Combine results from studies quantitatively Reviews & Meta Analyses > both qualitative & quantitative assessment * meta-analyses = reviews for overall conclusion * reviews # met analyses