1. Analysis Section Research Design

2. Protocol Overview Background4-5 pages Question/Objective/Hypothesis4 lines Design4-20 lines Study Population0.5-1 page Measurement3.5-4 pgs. Outcomes Exposures/predictors Confounders Analysis0.5-1 page Other (ethics/procedure/expected contribution) 0.5-1 page

3. Expected Components of the Analysis Section: Primary Objective Primary statistical approach to address study objective/ hypothesis (main results table) ◦ Statistical assumption assessment-changes in main approach Model-Fitting Issues ◦ Inclusion of predictor and confounding variables ◦ Linearity assessment ◦ Effect modification assessment (interactions) ◦ Quality of Model Fit ◦ Management of cluster/correlation in the data ◦ Sensitivity analysis ◦ Multiple testing/Missing value management Relevance / Quality of Prediction Assessment ◦ Proportion of Variance Explained ◦ Area under the Curve ◦ Population Attributable Risk/Benefit

4. Expected Components of the Analysis Section: Secondary Issues Assessment of “Within Study” Methodological Issues ◦ Unblinding ◦ Randomization integrity ◦ Inter-rater, intra-rater agreement ◦ Equivalence of data collection at different test sites, time periods ◦ Reliability of data-collection methods, questionnaires ◦ Construct/content validity of measures Analysis of Secondary Objectives ◦ Main methodological approach ◦ Predictors, outcomes and confounders

5. Examples of Expected Components Expected Components Example Restate Primary Study Objective To test the association between “x” and “y” To estimate the incidence of” x” To estimate the reproducibility of “q” Main approach to the analysis..multivariate logistic regression..intra-class correlation Model-fitting issues -inclusion of predictors and potential confounders All predictors and potential confounders included.. Full model fit with backward elimination based on p-value > 0.1 -linearity..a quadratic term will be added to assess linearity -effect modification to assess the hypothesized modification in the association between n”x” and “y” by “R”, an interaction term between “x” and ” R” will be included in the model and the Likelihood Ratio Test will be used to determine if it improves model fit

6. Define Main Approach to the Analysis and 2 Issues in Model- Fitting & Within Study Analysis Objective To determine whether outcomes after in-hospital cardiac arrest differ during nights and weekends compared with days/evenings and weekdays. Design and Setting We examined survival from cardiac arrest in hourly time segments, defining day/evening as 7:00 AM to 10:59 PM, night as 11:00 PM to 6:59 AM, and weekend as 11:00 PM on Friday to 6:59 AM on Monday, in 86 748 adult, consecutive in-hospital cardiac arrest events in the National Registry of Cardiopulmonary Resuscitation obtained from 507 medical/surgical participating hospitals from January 1, 2000, through February 1, 2007. Main Outcome Measures The primary outcome of survival to discharge and secondary outcomes of survival of the event, 24-hour survival, and favorable neurological outcome were compared using odds ratios and multivariable logistic regression analysis. Point estimates of survival outcomes are reported as percentages with95%confidence intervals (95% CIs). JAMA. 2008;299(7):785-792

7. Working Discussion Example ComponentJAMA. 2008;299(7):785-792 Main Approach to the AnalysisPredictor Outcome Confounders Analysis Method Issue #1 Model-Fitting Issue #2 Model-Fitting Within-Study Methodological Analysis Issue #1 Issue #2

8. JAMA. 2008;299(7):785-792 To examine the association between hour of day and outcomes, we used event hour as our exposure variable. The model included prospectively designated, clinically important potential confounders or their class (sex, race, illness category, combination of preexisting condition and cause variables, interventions in place at time of event, weekend, hospital size, event location, monitored status, witnessed status, first documented rhythm, initial or subsequent VT/VF, CPR duration, delay in defibrillation, delay in CPR, delay in vasopressor use, use of epinephrine, and time from hospital admission to event). Main Analytic Approach We examined 7 variables for evidence of effect modification by including their interactions with time of day in logistic regression models (first documented rhythm, event location, whether the event was monitored, whether the event was witnessed, delay in defibrillation, race [specifically black vs white], and illness category). Prospectively designated clinically important variables (age, Hispanic ethnicity, month of year, other cardiac arrest medication use, and induced hypothermia) were then entered into a stepwise multivariable logistic regression for the primary end point of survival to hospital discharge. The criterion for the stepwise selection of variables was P.25. Model Fitting Plan Effect Modification Assessment