The FDA asked us to use a method other than Wald and strongly suggested that we use Miettinen-Nurminen 1985 (MN) for computing confidence limits (CL) at summary level. This method is becoming increasingly popular and is used by many pharmaceutical companies, including Merck, Abbvie, and Moderna (cf. simple Google search). SAS has provided a partial implementation of MN scoring through COMMONRISKDIFF(CL=score) where the within-stratum calculation follows MN 1985 but inverse-variance stratum weights are used at the summary level. This method comes from the 2013 Agresti book. Has this method ever been published in a peer-reviewed article? Has it ever been used in a clinical trial submission? Is this method popular, i.e., number of citations? The lack of visibility in the use of this "SCORE" method prevent the use of this algorithm in a submission to health authorities.   The current alternative is to use R. Indeed, two packages have been developed to calculate CL based on MN 1985 method: metalite.ae (Merck) and scoreci. SAS users have already written about this 4 years ago.  I am a biostatistical manager at Bristol Myers Squibb. This issue has risen to the VP level as we plan to adopt MN 1985 in all our submissions in all the therapeutic areas. Confidence limits play an important role because they influence key analyses such as futility analysis for the primary outcome. Will you provide this algorithm and if so, when? ... View more

Currently when using the autoname option it adds a prefix to the end of the variable name. It would be helpful to have an option to add it to the beginning of the name instead. Depending on the type of analysis this can be important to later use variable shortcut lists appropriately.    There are currently many ways to accomplish this but it would be a nice to have feature/option to do this automatically.    ... View more

SAS should write a PROC to do meta-analysis.   1. There's a shortage of good software to do this, so SAS would add to its market. 2. There are lots of articles on it, but they often contradict each other, so the expertise of the people at SAS Statistics would be valuable. 3. LOTS of people do meta-analysis. 4. There is a book on doing this in SAS but it is very dated. 5. There are some articles on doing it with PROC MIXED but they are scattered, not fully coherent, use old versions of SAS (I found good one that uses v 6). And no source that I found has a complete method. E.g. one that includes combining effects, forest plots, diagnostics and so on.   ... View more

When it comes to fitting multivariable regression models (linear, logistic, time-to-event, etc) the objective may be to find the "best" parsimonious model from a set of categorical and continuous independent variables using a stepwise variable selection method. Some of these variables may be predictive of the outcome, others may not. Similarly of the continuous covariates, some may fit adequately with a simple linear relation, others may require a polynomial transformation to best predict the outcome.   The Multivariable Fractional Polynomial (MFP) approach to model fitting is essentially a backward elimination procedure in which all effects are fit, and considered for deletion. When a continuous covariate is considered, the best-fitting fractional polynomial (FP) transformation is identified (usually with a maximum of two polynomial terms, and with powers from the set (-2, -1, -0.5, 0, 0.5, 1, 2, 3), with 0 representing a logarithm ter), and tested against a model without any term, against a model with a simple linear effect, and finally a model with a simpler FP form. This process to identify the most appropriate form (if any) can be done using a closed testing procedure outlined in the references (eg Section 3 of [4]).   SAS Macros are available to implement this method (described in reference [4]), but it would be far more preferable if regression modelling procedures (REG, LOGISTIC, PHREG, GLMSELECT) could provide this selection method natively. Those SAS procedures that support programming statements can accommodate a univariable FP selection approach where a FP has only 1 polynomial term (FP1), for example:     proc phreg data=analysis; title 'Explore functional form of x, where x>=0'; fp1_n2 = 1 / (x+1)**2; fp1_n1 = 1 / (x+1)**1; fp1_n0_5 = 1 / sqrt(x+1); fp1_0 = log(x+1); fp1_0_5 = sqrt(x); fp1_1 = x; fp1_2 = x**2; fp1_3 = x**3; model Time*event(0) =fp1_n2 fp1_n1 fp1_n0_5 fp1_0 fp1_0_5 fp1_1 fp1_2 fp1_3 / details selection=FORWARD stop=1; run; quit; ... but fitting a "univariable" FP model where the polynomial has two polynomial terms requires fitting all models with 2 of the above terms. What's more, the set of two-term FP models (FP2) includes models with repeated powers, specified in the form beta1 * x^p1 + beta2 * (x^p1) * log(x), so adjusting the above step to force exactly two effects into the final model excludes this set of FP2 models with repeated powers. The full set of FP2 transformations might be specified one by one in the programming statements as done above...   proc phreg data=analysis; title 'Explore functional form of x, where x>=0'; fp2_n2_n2 = 1 / (x+1)**2 + log(x) * 1 / (x+1)**2; fp2_n1_n2 = 1 / (x+1)**1 + 1 / (x+1)**2; ...   however the procedure would fit the model beta1 * (x^p1 + x^p2), rather than the required model beta1 * x^p1 + beta2 * (x^p2) (in the case where p1 =/= p2).    References: [1] Royston P, Altman DG. Regression using fractional polynomials of continuous covariates - parsimonious parametric modelling. Applied Statistics 2006; 43(3😞 429–467. [2] Multivariable model-building : a pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables. Patrick Royston, Willi Sauerbrei. Wiley, 2008. ISBN 9780470028421  [3] Morris, T. P., White, I. R., Carpenter, J. R., Stanworth, S. J., and Royston, P. (2015) Combining fractional polynomial model building with multiple imputation. Statist. Med., 34: 3298–3317. http://dx.doi.org/10.1002/sim.6553 [4] W. Sauerbrei, C. Meier-Hirmer, A. Benner, P. Royston, Multivariable regression model building by using fractional polynomials: Description of SAS, STATA and R programs, Computational Statistics & Data Analysis, Volume 50, Issue 12, 2006, Pages 3464-3485, http://dx.doi.org/10.1016/j.csda.2005.07.015 ... View more

In HPSPLIT, if the dependent variable is unbalanced and dichotomous, there is a fix (albeit a somewhat awkward one) that lets you give better balance to sensitivity and specificity.  This should be improved by making it a part of HPSPLIT itself.  However, when there are more than 2 categories, there is nothing to do. For instance, if we divide birthweight into 3 categories (as is often done, even if that is a bad idea):     data new; set sashelp.bweight; count + 1; if weight < 1500 then do; bwcat3 = "1: Very low"; bwcat2 = "Low"; end; else if weight < 2500 then do; bwcat3 = "2: low"; bwcat2 = "Low"; end; else do; bwcat3 = "3: Normal"; bwcat2 = "Normal"; end; run; and then run HPSPLIT     proc hpsplit data = new seed = 123; class black boy married momedlevel momsmoke bwcat3; model bwcat3 = black boy married momedlevel momsmoke momage momwtgain visit cigsperday; output out=hpsplout; run; we get absolutely awful categorization, with pretty much all babies being labeled "normal". ... View more

PROC LIFETEST will report quartiles of survival times, but it’d be nice to have option to request other percentiles and associated CIs in a option similar in syntax to TIMELIST=.   See:  https://www.linkedin.com/pulse/kaplan-meier-survival-time-percentiles-chris-smith/ ... View more

Customer suggestion and reference:   Provide the ability for PROC MI to perform so-called "jump to reference" multiple imputation.   reference: LaVange, L.M. and Permutt, T. (2016) "A Regulatory Perspective on Missing Data in the Aftermath of the NRC Report", Statistics in Medicine, 25, 2835-2864 ... View more

Procedures like PROC GLIMMIX have a DDFM=method option on the MODEL statement to:   "specify the method for computing the denominator degrees of freedom for the tests of fixed effects that result from the MODEL, CONTRAST, ESTIMATE, LSMEANS and LSMESTIMATE statements."      Specifics on these methods can be found in the Details => Degrees of Freedom Methods - - http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_glimmix_details36.htm of the GLIMMIX procedure documentation.     PHREG does not have this feature and, indeed, does not handle a multi-level clustered situation beyond two levels, e.g. - county nested in state.    A received request is paraphrased as follows "I am doing some tests for fixed effects of level 2 (cluster level) using data of two levels with varying sample size in different clusters.   If you look for a reference and see how other software has done, refer to STATA 14 and the  MESTREG command - https://www.timberlake.co.uk/software/stata-new-multilevel-and-small-sample "     ... View more

Please bring back the Narratives from the PSS (Power and Sample Size) application. I found this very useful for rapidly generating a sensible, accurate report.   According to SAS Technical Support, the PSS application was replaced by a set of tasks in SAS Studio 3.6, but does not include the Narratives (described near the bottom of this page:https://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_clientpss_sect006.htm )   More discussion here: https://communities.sas.com/t5/Community-Suggestion-Box/Bring-back-the-Narratives-from-the-PSS-Power-and-Sample-Size/idc-p/831196  ... View more

After building a bayesian model it would be great to have simple way to run counterfactual analysis. Perhaps a procedure where we could specify the model and a variable that we want to simulate different values for. The output could include the target variable values for the different input variable values and a sensitivity (tornado) diagram. ... View more

I was surprised to see that there is no option to perform a non-inferiority power analysis when using TWOSAMPLESURVIVAL in PROC POWER.  It is a design layout that is becoming increasingly common in oncology outcomes research, and I think it would be a useful tool for several folks in the medical community.  As it stands, I am going to need to go to the expense of purchasing separate software that has integrated this method (like PASS from NCSS). ... View more

Hello,   The results displayed with PROC UNIVARIATE are described as descriptive statistics. However, the standard error of the mean is included in this display. While the standard deviation is certainly a descriptive statistic, the standard error is not. It is an inferential statistic that is only appropriate when the data comprise a random sample selected from a population of interest.   It would be helpful as a statistics instructor to have the SE not reported in this way with the PROC UNIVARIATE output, since it is not a descriptive statistic.   Thank you for considering.   Matt Hayat, PhD Professor of Biostatistics Georgia State University ... View more

This applies to all versions of SAS (at least back to version 6 and up to 9.4!) and, as far as I know, all operating systems. I am using 9.4 on Windows 10.   Often, with paired data, one variable is a later version of the other.  So you might have, e.g. test1 and test2, where there is training in between.   If you run: title "Paired t tests"; proc ttest data = DSNAME; paired test1*test2; run; then you get (among other output), the mean and sd and t test and so on of test1 - test2.  This is backwards of the way I think (and I think I am usual). What you want is test2 - test1, so that a positive value means it went up.   But you also get a graph called "paired profiles" which has test1 on the left and test2 on the right - which makes sense.   There should be an option to change this so that the output makes more sense.   (Yes, I know it's easy to just change the sign for the results, but .... we shouldn't have to do that!)   ... View more

The %margins macro from SAS is very useful, but it requires SAS/IML. Can the macro be revised to not require SAS/IML?   http://support.sas.com/kb/22/604.html   (Posted by suggestion of SAS: https://twitter.com/SAS_Cares/status/1122944288999059458)   Thank you! ... View more

To whom it may concert,   I am running an Example 87.11 Analysis of Clustered Data, and data listed as below: data Blind;    input ID Time Status dty trt @@;    Type= put(dty, type.);    Treat= put(trt, Rx.);    datalines;    5 46.23 0 1 1    5 46.23 0 1 0   14 42.50 0 0 1   14 31.30 1 0 0   16 42.27 0 0 1   16 42.27 0 0 0   25 20.60 0 0 1   25 20.60 0 0 0   29 38.77 0 0 1   29  0.30 1 0 0   46 65.23 0 0 1   46 54.27 1 0 0      ... more lines ...     1705  8.00 0 0 1 1705  8.00 0 0 0 1717 51.60 0 1 1 1717 42.33 1 1 0 1727 49.97 0 1 1 1727  2.90 1 1 0 1746 45.90 0 0 1 1746  1.43 1 0 0 1749 41.93 0 1 1 1749 41.93 0 1 0 ;   But it is very helpful for the users to run the codes on their own to compare the  results with SAS help, so can you include the whole data in all the examples instead of partial data only?   I am using SAS v9.4   Thanks   Wei ... View more

Hi, I would love to see c-statistics added to proc phreg for Fine and Gray competing risk outcomes.   Thanks, Anders ... View more

I would like to nominate something to be on the SAS ballot.  I want SAS to print out the effect size for any t-test or ANOVA and anything else they want.  I know that it is possible to calculate or program this ourselves, but a lot of people who publish in the applied fields are not up to doing this.  If SAS were to print it, then more people who publish in the applied fields would be more likely include this when they report statistics (nothing is guaranteed, but I think this would help a lot).  Both the American Statistical Association and the American Psychological Association have recommended this practice instead of only reporting P values which get misinterpreted as effect sizes which they most definitely are not.  If SAS were to do this, then it would move the reporting of statistical results forward.  ... View more

It would be wonderful if you could have SAS tell you the half width of a confidence interval for a proportion if given a coverage probability, true outcome proportion, and sample size in PROC POWER ONESAMPLEFREQ. This would be useful for choosing a sample size for a single-arm clinical trial.   Example code:   proc power; onesamplefreq ci=wilson /* or exact, wald, etc. */ halfwidth = . /* want to solve for this! */ proportion = .5 .7 .85 ntotal = 25 50 75 probwidth = .95; run;     ... View more

SUGGESTION:  Add the START= option for model selection in PROC GLMSELECT. This option is already available in PROC REG, PROC HPGENSELECT, and PROC PHREG.   The only way to automate backward stepwise model selection in SAS in to use SELECTION=BACKWARD and set START equal to the number of independent variables in the MODEL statement. PROC HPGENSELECT produces different variance estimates than PROC GLM and PROC REG, and PROC REG does not allow CLASS variables, so adding this feature to PROC GLM will fill a hole for a basic statistical procedure in SAS. ... View more