NonMem and EVID=2

Hello everybody, In order to obtain predicted values in a PopPK analysis without covariates at points in time with no actual values I have inserted extra rows indicated with EVID=2. The covariance step is completed when including these extra rows and precision of the parameter estimates are adequate. When omitting the extra rows {IGNORE=(EVID.EQ.2)}, I get exactly the same parameter estimates, the covariace step can be completed with some difficulty, but precision of the parameter estimates are now inadequate. I use METHOD=1 and just like Ahmed Abbas Suleiman experienced ( https://cognigencorp.com/nonmem/current/2013-February/4440.html 2 https://cognigencorp.com/nonmem/current/2013-February/4440.html), my outcome did not differ between the two conditions when setting METHOD=0. I saw that William Denney has responded to a similar question in 2016 ( https://cognigencorp.com/nonmem/current/2016-February/6085.html 2 https://cognigencorp.com/nonmem/current/2016-February/6085.html). I cannot see any response to Ahmed’s comment; do you Bill or somebody else have any idea why the outcomes differ with METHOD=1 but not with METHOD=0 in NonMem when including extra rows in the data set? Thank you in advance and kind regards, Johan

Hi Johan, Since I answered the original message, I’ll respond here, too, though with a less specific answer at this point. If you could provide the control stream and data (or example of the data), that would help. My guess is generally similar to my guess in 2016. The integration steps could be too long and the problem too stiff without the EVID=2 rows. Usually, this is automatically detected by the integrator, but if there is a major change in the problem dynamics (e.g. TMDD), the EVID=2 rows could assist the integrator with an intermediate step. This may show up in the covariance step because the maximum likelihood estimate has good convergence properties, but nearby there is a space that has poor properties (such as the TMDD behavior changing significantly). This could also occur if you do not have bounds on parameters that require bounds (such as a linear estimate of clearance which is not restricted to be > 0). For a more specific answer, I think that I’d need to see the code and data. Thanks, Bill *From:* [email protected] <[email protected]> *On Behalf Of *Johan Rosenborg *Sent:* Tuesday, April 23, 2019 11:33 AM *To:* [email protected] *Subject:* [NMusers] NonMem and EVID=2 Hello everybody, In order to obtain predicted values in a PopPK analysis without covariates at points in time with no actual values I have inserted extra rows indicated with EVID=2. The covariance step is completed when including these extra rows and precision of the parameter estimates are adequate. When omitting the extra rows {IGNORE=(EVID.EQ.2)}, I get exactly the same parameter estimates, the covariace step can be completed with some difficulty, but precision of the parameter estimates are now inadequate. I use METHOD=1 and just like Ahmed Abbas Suleiman experienced ( https://cognigencorp.com/nonmem/current/2013-February/4440.html 2 https://cognigencorp.com/nonmem/current/2013-February/4440.html), my outcome did not differ between the two conditions when setting METHOD=0. I saw that William Denney has responded to a similar question in 2016 ( https://cognigencorp.com/nonmem/current/2016-February/6085.html 2 https://cognigencorp.com/nonmem/current/2016-February/6085.html). I cannot see any response to Ahmed’s comment; do you Bill or somebody else have any idea why the outcomes differ with METHOD=1 but not with METHOD=0 in NonMem when including extra rows in the data set? Thank you in advance and kind regards, Johan

Hi Johan, If you could share your code and a relevant snippet of your dataset you might get more specific help. Without such detail: Bill's suggestion (time-varying covariate) could extend to the covariance step as well - those calculations might be more sensitive to the different values generated (values may be much smaller or larger because they depend on individual eta's). If you use differential equations, the additional time steps requested by EVID=2 might be more sensitive in a similar way. BTW, with nm7.3 onwards there is a system to prevent EVID=2 calculations in the est/cov steps, by using MDV=101 or larger. In intro nm7 guide I66: "I.66 Ignoring Non-Impact Records During Estimation (NM73) Typically users may produce data files that are augmented with additional non-dose, non- observation records in order to output predicted values at additional times to create high resolution curves. However, too many of such records tend to slow down the estimation analysis. As of NM73, if an MDV is set to a value greater than or equal to 100, it is converted to that value minus 100 upon input, but will not be used during estimation or covariance assessment, only for table outputting. This option allows you to use the same file for estimation and table outputs, without significantly slowing down the estimation. So if MDV=101, it will be converted to 1 upon use for final evaluations, and the records will be ignored during estimation." Hope this helps & best regards, Jeroen http://pd-value.com [email protected] @PD_value +31 6 23118438 -- More value out of your data!