Egret Features

Egret - A Tool for Epidemiological Analysis

• Comes with collection of over 30 ready-to-analyze data sets.
  -Intuitive, Windows-standard graphical environment
  -Use the CaseEditor, EGRET's powerful data editor, to create, modify and view data.
  -Use descriptive menus and dialog boxes to define a regression model for analysis
  -The EGRET Results Workbook window presents analysis results and post-fit diagnostics.
  -The Output Log Window accumulates in the background all output and generated results from the current session.
  -EGRET's default control settings can be easily viewed or changed.
  -Save current workspace with current settings for quick-restart.
• The nearly 300-page manual includes tutorials and descriptions of all windows and dialog boxes, the analysis models and post-fit regression diagnostics, descriptive statistics and the case-editor, and other features.
• Appendices list model fitting procedures and example data sets, and are followed by references and an index.
• Technical support via e-mail, telephone, and fax.

Case Editor

• Spreadsheet-like facility for importing, creating and editing data files.
• The Import option allows EGRET to read Dos Egret, LOGXACT, STATXACT, ASCII, Excel, Excel CSV, SPSS, SYSSTAT, and SAS Transport files.
• Up to 250 variables, with names up to 15 characters long.
• Up to 120,000 records.
• Create or modify variables and observations as needed:
  Create new variables easily from scratch or based on other variables, using any combination of statistical, arithmetic, logical, and column-wise operations
  RECODE function creates new categorical variables from existing categorical and continuous variables.
  Create random variates with normal or chi-squared distributions.
• SELECTIF, REJECTIF functions create subsets of the full data set.
• WEIGHT option specifies regression weights with a variable.
• Descriptive statistics, histograms and scatterplots available for inspection of the raw data or the subset used to fit a model.

• Logistic regression Conditional logistic regression
  Variable case/control ratio across matched sets
  Variably-sized matched sets
• Poisson regression Logistic Regression with Random Effects
  Betabinomial regression
  Logistic-normal regression
  Logistic-binomial regression
  Logistic-binomial regression for distinguishable data
  Test for excess variation.
• Cox proportional hazards regression
  Stratified risk sets
  Staggered entry times
  Three types of time-dependent covariate
  Test for proportional hazards
• Parametric Regression for Failure Time Data
  Exponential Regression
  Weibull Regression
  Model both scale and shape parameters
  Test for Weibull distribution
• Exact 2 x k contingency table analysis:
  Unstratified Data:
  Exact and asymptotic hypothesis tests and confidence intervals for odds ratios.
  Exact and asymptotic test of association against both trend and general alternatives.
  Stratified Data:
  Exact and asymptotic tests for homogeneity of odds ratios,
  Exact and asymptotic Mantel-Haenszel inference for a common odds ratio,
  Asymptotic tests of association against trend and general alternatives.
• Kaplan-Meier analysis:
  Staggered entry times
  Kernel-smoothed hazard functions
  Create matched-sets data file with randomly selected controls from risksets.

Regression Features:

• Automatic creation of dummy-variable main-effect and interaction terms from specified factor variables.
• Additive risk, relative risk, or additive-relative risk, depending upon specified analysis model.
• Deviances, p-values, standard errors, likelihood ratio tests, odds ratios, conditional odds ratios confidence intervals, and conditional confidence intervals, and user-selectable confidence coefficients.
• Term-wise Wald tests, score tests, model extensions, offset term, and time-dependent offset term.
• Forward and backward stepwise regression.
• Time-dependent covariates:
  Multiply regression terms by log(time) or time;
  Time-dependent factors (dummy variables),
  Time-dependent interpolated and step functions;
  Staggered entry times.
  Define up to 20 complex time-dependent covariates at one time.
• Fitting algorithm options: Newton-Raphson, modified Newton-Raphson, quasi-Newton, or Nelder-Mead simplex.

Post-fit Diagnostics:

• Fitted values and delta-betas plotted and in spreadsheet format
• Plots Available:
  Scatterplots
  Standardized/unstandardized delta-beta plots.
  Flag and identify observations with extreme delta-betas.
  Fitted value plots
  Kaplan-Meier and post-Cox regression plots:
  Survival and failure curves,
  Cumulative and log-cumulative hazard curves
  Delta-beta and/or fitted value plots available for logistic, conditional logistic, random-effects logistic regression, Cox Proportional Hazards, and Poisson regression.
• Save plots in both BMP and JPEG formats.

• Easy installation; initialization file can be customized for local preferences.
• Easy to use, screen-oriented.
• On-line, context-sensitive help panels always available. Context-sensitive menu line always at screen bottom.
• Most input parsed character by character so that all input is syntactically correct. All other input via single-key mnemonic commands, choice boxes and menus.
• Optional automatic operation via comprehen-sive multiple keystroke macro facilityö includes easy editing, single stepping, and extension of existing macros.
• A background listing file saves a record of the session. It can be annotated with comments at any time and reviewed from within the program.
• Save and retrieve individual study specifica-tions, including current Monte Carlo results.
• Diskpeek utility: choose files from a menu of those on your hard drive.

Models

• Logistic regression for prospective studies.
• Logistic regression for unmatched case-control studies.
• Poisson regression for subject-time data.
• Conditional logistic regression for 1:N matched sets, including specification of matching variables.
• Cox proportional hazards regression, including specification of hazard, censoring, and accrual period functions. Also log-rank test.

Estimation

• Obtain estimates in a two-way table, for 5 canonical alpha test levels and 5 power levels.
• Obtain estimates across a range of up to 100 steps, so that the values can easily be plotted or saved in a file:
  Power versus sample size, given alpha.
  Relative risk versus sample size, given alpha and the power.
  Relative risk versus power, given the sample size and alpha.
  Case rate versus sample size, given alpha and power.
• Test for overall exposure effect or trend in exposure levels.
• Empirical power versus sample size for specific values of alpha.
  

Monte Carlo

• Use Monte Carlo facility to verify estimates or obtain empirical power directly.
• Carry out Monte Carlo experiments for single sample size or for a series of sample sizes based on alpha level or power. Extend the number of trials and optionally specify seeds for the random number generator.
• Plot Monte Carlo results with confidence intervals along with fitted power vs. sample size curve.
• All Monte Carlo results are saved within a session so that they can be fitted with a generic power vs. sample size curve from which additional estimates can be obtained.
• Create random datasets for immediate analysis by the EGRET Statistical Analysis Package. Plot power versus estimated sample size.

Maximum Study Size

• Retain information on up to 20 variables.
• Up to 10 variables per regression: 1 exposure variable and 9 confounder and matching variables.
• Maximum of 100 parameters in a regression.
• All variables are categorical. Up to 20 categorical levels for each variable.
• Maximum of 4,000 covariate patterns for regression variables; 50 covariate patterns for matching variables.

Display-Only Graphics

• Screen display plots available on standard adapters.
• Limited graphics hardcopy output through DOS GRAPHICS command and Print-Screen functions, and limited output to Postscript files.
• Plot power versus sample size, given alpha.
• Plot relative risk versus sample size, given alpha and the power.
• Plot relative risk versus power, given the sample size and alpha.
• Plot empirical power (with confidence intervals) versus sample size for specific values of alpha; also plot the fitted curve to these values, which is used for fine tuning.