You can fit a frailty model for clustered data in PROC PHREG by using a RANDOM statement. Prior to SAS 9.4 TS1M4 (SAS/STAT 14.2), if the total number of parameters specified in a frailty model is greater than 400, the procedure might incorrectly issue a log message such as the following:

When this happens, the results could be less than optimal or biased because a true solution has not yet been found.

The recommended approach with models of this size, and the circumvention, is to specify the TECH=QUANEW option in an NLOPTIONS statement. The quasi-Newton method outperforms the default optimization method when there are more than 400 parameters.

Beginning in SAS 9.4 TS1M4 (SAS/STAT 14.2), the default optimization algorithm for the RANDOM statement is changed to the quasi-Newton algorithm. Because of this improvement, it is no longer necessary to include the TECH=QUANEW option when you have more than 400 parameters in the model.

You can reproduce the SAS/STAT 14.2 and later results by using the TECH=QUANEW option in prior releases.

Operating System and Release Information

Product Family	Product	System	Product Release		SAS Release
			Reported	Fixed*	Reported	Fixed*
SAS System	SAS/STAT	Microsoft Windows Server 2003 Datacenter Edition	9.3		9.3 TS1M0
		Microsoft Windows Server 2003 Enterprise Edition	9.3		9.3 TS1M0
		Microsoft Windows Server 2003 Standard Edition	9.3		9.3 TS1M0
		Microsoft Windows Server 2003 for x64	9.3		9.3 TS1M0
		Microsoft Windows Server 2008	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Microsoft Windows Server 2008 R2	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Microsoft Windows Server 2008 for x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Microsoft Windows XP Professional	9.3		9.3 TS1M0
		Windows 7 Enterprise 32 bit	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Enterprise x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Home Premium 32 bit	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Home Premium x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Professional 32 bit	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Professional x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Ultimate 32 bit	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows 7 Ultimate x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Windows Vista	9.3		9.3 TS1M0
		Windows Vista for x64	9.3		9.3 TS1M0
		64-bit Enabled AIX	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		64-bit Enabled HP-UX	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		64-bit Enabled Solaris	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		HP-UX IPF	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Linux	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Solaris for x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Microsoft® Windows® for x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		Linux for x64	9.3	14.2	9.3 TS1M0	9.4 TS1M4
		z/OS	9.3	14.2	9.3 TS1M0	9.4 TS1M4

* For software releases that are not yet generally available, the Fixed Release is the software release in which the problem is planned to be fixed.

---

Prior to SAS 9.4 TS1M4 (SAS/STAT 14.2), PROC PHREG with a RANDOM statement and more than 400 parameters might incorrectly report that the model converged. To address this, the default optimization algorithm for random effects changed in SAS/STAT 14.2, and different results might be produced.

Type:	Problem Note
Priority:	medium
Topic:	SAS Reference ==> Procedures ==> PHREG Analytics ==> Survival Analysis

Date Modified:	2019-05-02 12:53:32
Date Created:	2019-03-29 14:04:28