Statistical Packages for Biomedical Data Analysis

1. Statistical packages most often used for data management and statistical analysis of biomedical data IRCCS San Raffaele Pisana, Rome, Italy,28 February - 2 March 2018

2. Statistical packages https://www.capterra.com/statistical-analysis-software/ https://en.wikipedia.org/wiki/Comparison_of_statistical_packages 2.500 USD/year

4. R is a programming language and free software environment for statistical computing and graphics that is supported by the R Foundation for Statistical Computing. The R language is widely used among statisticians and data miners for developing statistical software and data analysis. Polls, surveys of data miners, and studies of scholarly literature databases show that R's popularity has increased substantially in recent years. While R is an open-source project supported by the community developing it, some companies strive to provide commercial support and/or extensions for their customers. This section gives some examples of such companies.

5. A powerful combination of biostatistics, curve fitting (nonlinear regression) and scientific graphing. GraphPad Prism is a commercial scientific 2D graphing and statistics software published by GraphPad Software, Inc., a privately held California corporation. Prism is available for both Windows and Macintosh computers. • Features • Provides statistical guidance for novices. • Analysis checklists review if an appropriate analysis was performed. • Nonlinear regression with many options (remove outliers, compare models, compare curves, interpolate standard curves, etc.). • Live links. When data are edited or replaced, Prism automatically updates the results and graphs. • Analysis choices can be reviewed, and changed, at any time. • Automatic error bars. Raw data (replicates) can be entered, and then plotted as mean with SD, SEM or confidence interval. • 30 Days free at https://www.graphpad.com/demos

6. Statistical comparisons Paired or unpaired t tests. Reports P values and confidence intervals. Nonparametric Mann-Whitney test, including confidence interval of difference of medians. Kolmogorov-Smirnov test to compare two groups. Wilcoxon test with confidence interval of median. Perform many t tests at once, using False Discover Rate (or Bonferroni multiple comparisons) to choose which comparisons are discoveries to study further. Ordinary or repeated measures one-way ANOVA followed by the Tukey, Newman-Keuls, Dunnett, Bonferroni or Holm-Sidak multiple comparison tests, the post-test for trend, or Fisher’s Least Significant tests. Many multiple comparisons test are accompanied by confidence intervals and multiplicity adjusted P values. Greenhouse-Geisser correction so repeated measures one-way ANOVA does not have to assume sphericity. When this is chosen, multiple comparison tests also do not assume sphericity. Kruskal-Wallis or Friedman nonparametric one-way ANOVA with Dunn's post test. Fisher's exact test or the chi-square test. Calculate the relative risk and odds ratio with confidence intervals. Two-way ANOVA, even with missing values with some post tests. Two-way ANOVA, with repeated measures in one or both factors. Tukey, Newman-Keuls, Dunnett, Bonferron, Holm-Sidak, or Fishers LSD multiple comparisons testing main and simple effects. Three-way ANOVA (limited to two levels in two of the factors, and any number of levels in the third). Kaplan-Meier survival analysis. Compare curves with the log-rank test (including test for trend). Column statistics Calculate min, max, quartiles, mean, SD, SEM, CI, CV, Mean or geometric mean with confidence intervals. Frequency distributions (bin to histogram), including cumulative histograms. Normality testing by three methods. One sample t test or Wilcoxon test to compare the column mean (or median) with a theoretical value. Skewness and Kurtosis. Identify outliers using Grubbs or ROUT method.

7. Linear regression and correlation Calculate slope and intercept with confidence intervals. Force the regression line through a specified point. Fit to replicate Y values or mean Y. Test for departure from linearity with a runs test. Calculate and graph residuals. Compare slopes and intercepts of 2 or more regression lines. Interpolate new points along the standard curve. Pearson or Spearman (nonparametric) correlation. Analyze a stack of P values, using Bonferroni multiple comparisons or the FDR approach to identify significant" findings or discoveries. Clinical (diagnostic) lab statistics Bland-Altman plots. Receiver operator characteristic (ROC) curves. Deming regression (type ll linear regression). Simulations Simulate XY, Column or Contingency tables. Repeat analyses of simulated data as a Monte-Carlo analysis. Plot functions from equations you select or enter and parameter values you choose. Other calculations Area under the curve, with confidence interval. Transform data. Normalize. Identify outliers.  Normality tests. Transpose tables. Subtract baseline (and combine columns). Compute each value as a fraction of its row, column or grand total.  Nonlinearregression Fit one of our 105 built-in equations, or enter your own. Enter differential or implicit equations. Enter different equations for different data sets. Global nonlinear regression – share parameters between data sets. Robust nonlinear regression. Automatic outlier identification or elimination. Compare models using extra sum-of-squares F test or AICc. Compare parameters between data sets. Apply constraints. Differentially weight points by several methods and assess how well your weighting method worked. Accept automatic initial estimated values or enter your own. Automatically graph curve over specified range of X values. Quantify precision of fits with SE or CI of parameters. Confidence intervals can be symmetrical (as is traditional) or asymmetrical (which is more accurate). Quantify symmetry of imprecision with Hougaard’sskewness. Plot confidence or prediction bands. Test normality of residuals. Runs or replicates test of adequacy of model. Report the covariance matrix or set of dependencies. Easily interpolate points from the best fit curve.