87 Output File Formats

1 Format of NONMEM Report File

The format of the NONMEM report file has been slightly modified, with improvements to allow third party software to more easily identify portions of the result file. As described above, the user has now the ability to request a series of classical or new estimation methods within the same problem if he so chooses. Each of the new methods produces slightly different banner text and termination status text in the report file. For example, an iterative two stage analysis may be requested, followed by an MCMC Bayesian method, followed by an FOCEI method. The theta, sigma, and omega results of the iterative two stage method will be passed on as initial values for the MCMC Bayesian method, to facilitate the MCMC Bayesian analysis, which in turn can supply initial values for the FOCEI method. Each of these intermediate analyses will provide output to the NONMEM report file, and will be identified by unique text for that method. To allow a program to consistently find the appropriate positions in the file without having to search for specific words in the text, the report file is augmented with special tag labels that remain constant, regardless of the method used.

The tags always begin with #, followed by four letters to indicate the tag type, followed by a colon (:). The following tags are presently defined:

1.1 #PARA: (NM72)

This tag identifies the parallelization file and number of nodes used, if parallel estimation is performed.

1.2 #TBLN: (NM72)

This tag specifies that following it, on the same line, will be found an integer that refers to the number of this estimation method. This number is also the table number listed in the title to tables in the various output files (raw output file, .cov, .cor, etc). The table number is incremented for each $EST statement, across all problems in the control stream file.

1.3 #METH:

This tag specifies that following it, on the same line, will be found a text that describes the method, for example First Order Conditional Estimation Method with Interaction.

1.4 #TERM:

This tag indicates that beginning on the next lines, text describes the termination status of the analysis. Included in the results are average of the individual etas (ETABAR), its standard error (SE), P-value on the null hypothesis that ETABAR is not statistically different from 0, and eta and epsilon shrinkage. Shrinkage is not reported after an FO analysis. As of nm75, shrinkage information is reported for the BAYES method. See below for more information on shrinkage.

The individual etas used to assess ETABAR/SE/p-value/Shrinkage are modes of the posterior density for ITS/FOCE/Laplace for each individual, or conditional mean etas for IMP/SAEM for each individual, as of the last iteration.

ETABAR, SE, P-Value, and Shrinkage are not always accurately calculated after an SAEM analysis, as these are averaged over the entire set of iterations of the reduced stochastic mode (assuming NITER>0), during which the estimates of THETAs, OMEGAs, and SIGMAs are also averaged. After an SAEM analysis, run a $EST METHOD=IMP EONLY=1 to obtain good post-analysis estimates of shrinkage, standard errors, and objective function, as described earlier.

1.5 #TERE:

This tag indicates the end of the lines describing the termination status of the analysis. Thus, a software program may transfer all lines between #TERM: and #TERE: to a summary file.

1.6 #OBJT:

Indicates that following it, on the same line, is the text describing the objective function, such as Minimal Value Of Objective Function.

1.7 #OBJV:

Indicates that following it, on the same line, is the objective function value. However, a more efficient way of extracting numerical results from the analysis is from the raw output file (see below).

1.8 #OBJS:

Indicates that following it, on the same line, is the objective function standard deviation (MCMC Bayesian analysis only). However, a more efficient way of extracting numerical results from the analysis is from the raw output file (see below).

1.9 #OBJN: (nm73)

Indicates that following it, on the same line, is the nonparametric objective function value.

1.10 #CPUT: (nm73)

Total cpu time in seconds. This is an accurate assessment of CPU usage of the entire problem, whether done in single or parallel mode.

1.11 Shrinkage and ETASTYPE (NM74)

Inter-subject shrinkage for each eta is calculated as:

\[\text{ETASHRINKSD} = 100\% \times \left[1 - \frac{\text{SD}(\eta_j)}{\sqrt{\Omega_{jj}}}\right]\]

where $\eta_j$ is the individual’s empirical bayes estimate (FOCE/LAPLACE/ITS) or sampled conditional mean (IMP/SAEM) for the jth eta. The above definition of eta shrinkage is designated ETASHRINK in versions earlier than nm74. The variance version is also calculated as of nm74:

\[\text{ETASHRINKVR} = 100\% \times \left[1 - \frac{\text{VAR}(\eta_i)}{\Omega_{ii}}\right]\]

The reason for the two types of calculations is that both types appear in the literature.

As of nm73, additional shrinkage information, called EBVSHRINK (nm73) or EBVSHRINKSD (nm74), is the ETA shrinkage based on the average empirical Bayes variance (EBV), the etc(j,j), or phc(j,j) listed in the .phi or .phm table, standard deviation version:

\[\text{EBVSHRINKSD} = 100\% \times \left(1 - \sqrt{1 - \text{etcave}(j,j)/\Omega(j,j)}\right)\]

\[\text{EBVSHRINKSD} = 100\% \times \left(1 - \sqrt{1 - \text{phcave}(j,j)/\Omega(j,j)}\right)\]

Where etcave(j,j) is average etc(j,j) among included subjects, and phcave(j,j) is average phc(j,j) among included subjects, for eta(j) or phi(j). Again, etc/phc is evaluated as first order approximation of the posterior variance around the mode (FOCE/ITS), second order approximation around the mode (LAPLACE), or Monte Carlo assessed posterior variance around the conditional mean (IMP/SAEM).

As of nm74, the variance version is also available, called EBVSHRINKVR:

\[\text{EBVSHRINKVR} = 100\% \times (\text{etcave}(j,j)/\Omega(j,j))\]

\[\text{EBVSHRINKVR} = 100\% \times (\text{phcave}(j,j)/\Omega(j,j))\]

As of nm75, the relative information is available, evaluated as

\[\text{RELATIVEINF} = 100\% / \text{diag}\left[(I - \Omega^{-1} V_{\text{ave}})^{-1}\right]\]

where $V_{\text{ave}}$ is the matrix containining the elements of the phcave(i,j) matrix. Typically, RELATIVEINF is approximately equal to 100-EBVSHRINKVR.

As of nm73, if the eta shrinkage is less than 0, it will reported as a value of 1.0E-10. Less than 0 shrinkage can occur due to limited precision evaluations, and/or sometimes with classical NONMEM methods.

Eta shrinkage is averaged for all individuals if ETASTYPE=0. As of nm73, should you wish to correct for some individuals not contributing at all to one or more etas (this may or may not be desirable, depending on your needs), the shrinkage can be calculated by NONMEM to not include these etas by setting ETASTYPE=1 in the $EST record. This will average shrinkage information only among individuals that provided a non-zero derivative of their data likelihood with respect to that eta, and will not include subjects with a non-influential eta, that is in which the derivative of the data likelihood is zero. Furthermore, you may specify eta i of particular subjects to be excluded, by setting a reserved variable ETASXI(i) to 1 in $PK or $PRED, or specify eta i of certain subjects to be included, by setting ETASXI(i)=2 (ETASXI stands for eta shrinkage exclude/include):

IF(ID==3) ETASXI(1)=1
IF(ID==23) ETASXI(3)=2

The results outputted in the NONMEM report file refer to average eta shrinkage. See the section I.65 $EST: Additional Output Files Produced on root.phi, for additional information one can obtain about eta shrinkage for each subject.

Residual error shrinkage standard deviation version for each residual error is evaluated for simple problems as

\[\text{EPSSHRINKSD} = 100\% \times [1 - \text{SD}(\text{IWRES})]\]

(see (Karlsson and Savic 2007)).

In nm73 and earlier, the above definition was called EPSSHRINK. As of nm74, the variance version is also now available:

\[\text{EPSSHRINKVR} = 100\% \times [1 - \text{VAR}(\text{IWRES})]\]

For more complicated problems, the data and individual predicted values that contribute to assessing the shrinkage for each epsilon is not as straight-forward. For example, if EPS(1) is proportional error to PK data, and EPS(2) is proportional error to PD, and they are not connected by an off-diagonal sigma, then EPS1 shrinkage pertains to PK data residuals, and EPS2 shrinkage pertains to PD data residuals. If they are related by an off-diagonal SIGMA, then their shrinkage is related, and they will have similar or identical shrinkage values.

If two epsilons pertain to the same data, such as proportional EPS and additive EPS for PK data:

Y=F+F*EPS(1)+EPS(2)

Then the same epsilon shrinkage is associated with EPS(1) and EPS(2). However, if F=0 for some data, then such values contribute to EPS(2) shrinkage assessment, but not to EPS(1) shrinkage assessment. In such cases, shrinkage to EPS(1) and EPS(2) may differ slightly, where EPS(1) shrinkage incorporates only residuals to data with predicted values that are non-zero, and EPS(2) shrinkage incorporates residuals to all PK data.

See also ETASAMPLES=0 (default) (nm74) in section I.40 Stochastic Approximation Expectation Maximization (SAEM) Method for other methods of covariate model diagnostics.

The reported shrinkage information from an MCMC Bayesian analysis is based on the average of the shrinkage across all individuals across all stationary iterations (positive numbered iterations in the .ext file).

2 `$EST`: Format of Raw Output File

A raw output file will be produced that provide numerical results in a columnar format. The raw output file name is by default root.ext, where root is the root name of the control stream file. Or, the name is provided by the user using a new FILE= parameter added to the $EST record. A raw output file has the following format:

A header line that begins with the word Table, such as:

TABLE NO. 4: MCMC Bayesian Analysis: Goal Function=AVERAGE VALUE OF LIKELIHOOD FUNCTION

This header line provides the analysis text (same as given on the #METH: line in the main report file), followed by the goal function text (same as given on the #OBJT: line in the report file).

The next line contains the column headers to the table, such as (this is actually all on one line in the file):

ITERATION   THETA1      THETA2      THETA3      THETA4      SIGMA(1,1)  OMEGA(1,1)
OMEGA(2,1)  OMEGA(2,2)  OMEGA(3,1)  OMEGA(3,2)  OMEGA(3,3)  OMEGA(4,1)  OMEGA(4,2)
OMEGA(4,3)  OMEGA(4,4)  OBJ

This is followed by a series of lines containing the intermediate results from each printed iteration (six significant digits), based on the PRINT= option setting:

                  10 1.73786E+00 1.57046E+00 7.02200E-01 2.35533E+00 6.18150E-02 1.82955E-01
-3.18352E-03 1.46727E-01 -4.38860E-02 2.58155E-02 1.45753E-01 -4.58791E-02 6.28773E-03
 5.06262E-02 1.50017E-01 -2301.19773603667

For the above example, each of the values, up to the next to last one, occupies 13 characters, including the delimiter (in this example the delimiter is a space). The last value is the objective function, which occupies 30 characters, to allow for the largest range of objective function values, and the greatest expression of precision.

The iteration number, which is the first value in every line, is typically positive, but also may be negative under the following conditions:

The burn-in iterations of the MCMC Bayesian analysis are given negative values, starting at –NBURN, the number of burn-in iterations requested by the user. These are followed by positive iterations of the stationary phase.
The stochastic iterations of the SAEM analysis are given negative values. These are followed by positive iterations of the accumulation phase.
Iteration -1000000000 (negative one billion) indicates that this line contains the final result (THETAs, OMEGAs, and SIGMAs, and objective function) of the particular analysis. For BAYES analysis, this is the mean of the non-negative iterations (stationary samples) listed before it.
Iteration -1000000001 indicates that this line contains the standard errors of the final population parameters. For BAYES, it is the sample standard deviation of the stationary samples.
Iteration -1000000002 indicates that this line contains the eigenvalues of the correlation matrix of the variances of the final parameters.
Iteration -1000000003 indicates that this line contains the condition number, lowest, highest, Eigen values of the correlation matrix of the variances of the final parameters.
Iteration -1000000004 indicates this line contains the OMEGA and SIGMA elements in standard deviation/correlation format
Iteration -1000000005 indicates this line contains the standard errors to the OMEGA and SIGMA elements in standard deviation/correlation format
Iteration -1000000006 indicates 1 if parameter was fixed in estimation, 0 otherwise.
Iteration -1000000007 lists termination status (first item) followed by termination codes. See I.65 $EST: Additional Output Files Produced under root.xml (NM72) for interpreting the codes.
Iteration -1000000008 lists the partial derivative of the log likelihood (-1/2 OFV) with respect to each estimated parameter. This may be useful for using tests like the Lagrange multiplier test.
Additional special iteration number lines may be added in future versions of NONMEM.

The raw output file is provided automatically, independent of the formatted files that may be requested by the user using the $TABLE command.

For the output files generated during the $EST step, the following parameters may be specified:

2.1 FILE=my_example.ext

Parameters/objective function printed to this raw output file every PRINT iterations. Default is control.ext, where control is name of control stream file.

2.2 DELIM=s or FORMAT=t or FORMAT=,

Delimiter to be used in raw output file FILE. The character defines the delimiter, which may be s for space, t for tab, comma for a comma delimited file with aligned fields (so, padded with spaces), c for comma delimited file with no spaces, or q for comma delimited file with no spaces and double quotes around column names that have commas (such as "OMEGA(2,1)").

2.3 DELIM=s1PE15.8 or FORMAT=s1PG15.8 or FORMAT=tF8.3

In addition to the delimiter, a format (FORTRAN style) may be defined for the presentation of numbers in the raw OUTPUT file.

Default format for additional output files produced by the $EST step is s1PE12.5

The variables DELIM and FORMAT are interchangeable.

If you specify FORMAT=QCSV then this is equivalent to FORMAT=q1PG23.16 If you specify FORMAT=CSV then this is equivalent to FORMAT=c1PG23.16

The lines produced in the ext file may be very long. You may optionally provide a line length, followed by a continuation marker to be tagged at the end of each line (e), and/or a continuation marker to be tagged at the beginning of the continuing line.

FORMAT=s1PE15.8:160&

will print lines of at most 160 characters, followed by a & for each line that needs to be continued (if using an ampersand, and it is at the end of the line in the control stream file, place a ; after it so it is not interpreted as a continuation indicator by the NM-TRAN control stream file reader).

FORMAT=s1PE15.8:160&c

Will print lines of at most 160 characters, with & tagged at the end of the line to be continued, and a c at the beginning of the continued line.

FORMAT=s1PE15.8:160sc

Will print lines of at most 160 characters, with no character at the end of each line to be continued, and a c at the beginning of the continued line. S represents “space”, and a space may not serve as a continuation marker because of its ambiguity, so it serves here as a place holder in the FORMAT definition. These line continuation formats are ignored in $TABLE records, but are used in the $EST record for all additional file formats, and can are used in $EST CHAIN=METHOD and $CHAIN records.

2.4 NOTITLE=[0,1]

If NOTITLE=1 (default=0), then the Table header line will not be written to the raw output file specified by FILE=.

2.5 NOLABEL=[0,1]

If NOLABEL=1 (default=0), then the column label line will not be written to the raw output file specified by FILE=.

2.6 ORDER (NM72)

The order in which the THETAs, OMEGAs, and SIGMAs are listed in the output file is by default as follows: THETAs (T), SIGMAS (S), OMEGAS (O). The SIGMA and OMEGA matrices are listed in lower triangular order, row-wise:

You may change the order in which these are displayed, by specifying the ORDER option. The THETAS are referenced with a T, SIGMAS with S, OMEGAS with O, lower triangular with L, upper triangular with U. The first three letters given in the ORDER option refer to which parameters are listed in order (T, S, O), and the fourth letter is U or L to indicate matrix element order for SIGMAs and OMEGAs. Thus,

ORDER=TSOL

Is the default ordering. This is different from the ordering that is given in the report file for displaying the variance matrix, which is TOSU. In TOSU ordering, THETAs are listed first in the raw output file, followed by OMEGAs, followed by SIGMAs, and the OMEGAs and sigma elements are listed in row-wise upper-triangular order (or column-wise, lower triangular order):

3 `$EST`: Additional Output Files Produced

In addition to the raw output file described in the previous section, the following files are created automatically, with root name based on the root name of the control stream file. The delimiter and numerical format is whatever is specified with $EST … FORMAT=, for the raw output file.

3.1 root.cov

Full variance-covariance error matrix to THETAs, SIGMAs, and OMEGAs

3.2 root.clt (NM74)

Lower triangular form of Variance-covariance error matrix to THETAs, SIGMAs, and OMEGAs, in TOSL order, for easy cut-and-paste transfer to place into control stream files as prior information.

3.3 root.cor

Full correlation matrix to THETAs, SIGMAs, and OMEGAs

3.4 root.coi

Full inverse covariance matrix (Fischer information matrix) to THETAs, SIGMAs, and OMEGAs

3.5 root.phi

Individual phi parameters (phi(i)=mu(i)+eta(i), for ith parameter), and their variances phc(,). For parameters not MU referenced phi(i)=eta(i). When a classical method is performed (FOCE, Laplace), then mode of posterior eta(i) are printed out, along with their Fisher information (first order expected value for FOCE, second order for Laplace) assessed variances etc(,).

For ITS, these parameters are the modes of the posterior density, with first-order approximated expected variances (or second order variances if $EST METHOD=ITS LAPLCE is used).

For IMP, IMPMAP, SAEM methods, they are the Monte Carlo evaluated mean parameters and variances of the parameters under the posterior density.

For MCMC Bayesian, before nm75, they are random single samples of phi(), as of the last position. Their variances are zero. As of nm75, the .phi table contains the average of phi() values collected throughout the stationary distribution (positive iterations) phase, so they also contain variances.

Individual objective function values (obji) are also produced. In the case of MCMC, before nm75, the individual likelihood of the last sample is reported. As of nm75, average the MCMC individual likelihood among during the stationary phase is reported. Note that the variances of the phis (phc(,)) include variability contributed by THETAs for MCMC analysis, unlike the other methods where the variances of the phis are evaluated with THETAs fixed.

As of nm75, variances phc/etc for phis/etas of levels above the subject are now available in the phi table.

As of nm74, if $EST PHITYPE=1 is specified, then conditional mean etas are reported in the phi table, regardless of the analysis method. See PHITYPE=0 (NM74) in section I.40 Stochastic Approximation Expectation Maximization (SAEM) Method.

3.6 root.phm (NM72)

These values have similar meaning as the results in root.phi, but individual phi/eta/obji parameters per sub-population are recorded. This file is only produced in $MIXTURE problems.

The PHITYPE option also acts on this file.

As of nm75, for MCMC Bayesian analysis, the items listed in this table consist of the average values among all the iterations during the stationary distribution phase. Before nm75, the values consisted of values from the last MCMC iteration.

The conditional variances in the root.phi and root.phm files can represent the information content provided by a subject for a given eta or phi. For example, if data supplied by the subject is rich, then the variance tends to be smaller. If little data is supplied by the subject for that eta, then the conditional variance will approach its omega. In fact, a subject’s shrinkage can be evaluated as follows:

In accordance with the SD formula:

\[\text{ETAshrinkage}_i \% = 100\% \times \left(1 - \sqrt{1 - \text{phc}_i(j,j)/\Omega(j,j)}\right)\]

Or by the variance formula:

\[\text{ETAshrinkage}_i \% = 100\% \times \text{phc}_i(j,j)/\Omega(j,j)\]

For subject i, eta or phi j.

3.7 root.iph (NM75)

As of nm75, samples of phi/eta are collected at each BAYES iteration, and summarized to provide mean phi and phc() in the root.phi table, as described above. By default, the individual phi values from each iteration are not stored. However, if you set

$EST … BAYES_PHI_STORE=1

then phi and eta values from each BAYES iteration will be stored in root.iph. For non-mixture problems, only records of SUBP=0 are recorded, as there are no sub-population divisions. For mixture problems, the SUBP=0 records contain the composite phis and etas (the average of these across all non-negative iterations are in the root.phi table), and the SUBP>0 records contain the phis and etas appropriate to each sub-population SUBP (the average of these across all non-negative iterations are in the root.phm table). The root.iph file can become quite large, so it should be used only on the final analysis.

3.8 root.shk (NM72)

This file presents composite ETA shrinkage and EPSILON shrinkage information, the same as given in the report file between the #TERM: and #TERE: tags, but in rows/column format, and with adjustable formatting.

Type 1 = ETAbar
Type 2 = ETAbar SE
Type 3 = P val
Type 4 = %ETA shrinkage SD version
Type 5 = %EPS shrinkage SD version
Type 6 = %ETA shrinkage based on empirical Bayes Variance (SD version)
Type 7 = number of subjects used
Type 8 = %ETA shrinkage variance version
Type 9 = %ETA shrinkage based on empirical Bayes Variance (variance version)
Type 10 = %EPS shrinkage variance version
Type 11 = %Relative information

3.9 root.shm (NM73)

As of NM73, the .shm table (which stands for shrinkage map) will contain information which etas were excluded in the eta shrinkage assessment. The syntax is as follows:

For each subject, sub-population, the value listed in column ets(j) contains the information about whether and how that eta was included in the etabar/shrinkage calculations. It is a binary value of the format x.abcdef, where each of the letters may be 0 or 1. If the eta is excluded from the etabar/eta shrinkage summary that is recorded in the main NONMEM report file or the .shk file, then x=1, otherwise it is 0. The remaining binary digits after the decimal point describes conditions about this eta that were involved in deciding whether to exclude this eta:

a: set to 1 if NONMEM assessed this eta as non-influential (the derivative of the data likelihood with respect to that eta is 0). This exclusion criterion is only acted on (that is, actually excludes this eta, indicated by x=1), if etastype=1.

b: set to 1 if NONMEM excluded this eta for this sub-model (sub-population), for this subject, because this was not the best fitting sub-model for this subject. Thus all etas of that subject for all sub-models that are not the optimally fitting will have this bit set, and only the optimal sub-model will have B cleared (0) for all its etas.

c: set to 1 if NONMEM determined that this eta had no influence for this sub-model. This bit is not set to 1 if bit B is 1. This bit is not set to 1 for non-population-mixture models. Also, this exclusion criterion is set and acted upon when FOCE/Laplace are used, but is not set or acted on for the EM methods. If NONINFETA is set to 1, then FOCE/Laplace behave similarly to EM methods, and will not set this bit even if the eta has no influence.

d: set if the eta is excluded based on selecting the hybrid option in $EST.

e: Set if the user requested an exclusion based on ETASXI(i)=1 setting in $PK or $PRED for eta i.

f: Set if the user requested an inclusion based on ETASXI(i)=2 setting in $PK or $PRED for eta i. Be careful about using this, as it overrides all other exclusion criteria except bit B. The F bit is the only one that indicates inclusion when set, rather than exclusion.

3.10 root.grd (NM72)

This file contains gradient values for classical NONMEM methods.

The format of these files are subject to FORMAT, ORDER, NOLABEL, and NOTITLE options in the $EST command, the same as for the raw output file.

3.11 root.xml (NM72)

An XML markup version of the contents of the NONMEM report file is produced automatically. The rules (schema, document type definition) by which it is constructed are given in output.xsd and output.dtd, in the NONMEM ..\util or ..\run directory.

In NM73, termination_textmsgs catalogs termination text messages by number, which can be mapped to ..\source\textmsgs.f90.

In nm73, termination status catalogs the error status:

For traditional analyses, an error number is listed. If negative, the analysis was user-interrupted.

For EM/Bayes analysis, error numbers map as follows (changed for nm75):

0, 8: optimization was completed
1, 9: optimization not completed (ran out of iterations)
2, 10: optimization was not tested for convergence
3, 11: optimization was not tested for convergence and was user interrupted
4, 12: optimization was not completed and was user interrupted
16, 24: objective function is infinite or all individual objective functions are zero. problem ended
32, 40: All individual objective functions are zero. problem ended
8, 9, 10, 11, 12, 24, 40: reduced stochastic/stationary portion was not completed prior to user interrupt

The expectation only step may also be considered an “optimization” in the context of these codes.

These codes are also listed on line -1000000007 in the .ext file (NM74).

3.12 root.cnv (NM72)

This file contains convergence information for the Monte Carlo/EM methods, if CTYPE>0:

-2000000000 = mean of last CITER values.
-2000000001 = standard deviation of last CITER values (for objective function, STD of second to last CITER values)
-2000000002 = linear regression p-value of last CITER values against iteration number.
-2000000003 = Alpha used to assess statistical significance (p-value < alpha)

Please note the following:

The Sigma values are in their Cholesky format, as this is the form in which convergence of these values are tested.

The Alpha are those based on ones actually used for convergence test of that parameter, or which would have been used on that parameter if CTYPE were of proper type. The alpha may be Bonferroni corrected because of multiple comparisons, depending on number of parameters that were tested or would have been tested. Objective function alphas are not Bonferroni corrected.

For importance sampling and iterative two stage, the average objective function listed in root.cnv could be used as an alternative to the final objective function for likelihood ratio tests.

3.13 root.smt (NM72)

S matrix, if $COV step failed.

3.14 root.rmt (NM72)

R matrix, if $COV step failed.

3.15 root.imp (NM73)

The root.imp file is produced if the user selects importance sampling with option IACCEPT=0.0. In such cases, this file lists the final IACCEPT and DF values that NONMEM selected for each subject.

Three files are produced providing nonparametric information:

3.16 root.npd (NM73)

Each row contains information about a support point: The support point number, the ID from which the support point was obtained as an EBE of that subject (ID is -1 if this support point was randomly generated because NSUPP/NSUPPE was greater than number of subjects). The eta values of the support point are listed, followed by the cumulative probability (CUM) associated with each eta, followed by the joint density probability of that support point, if default or MARGINALS was selected. If ETAS was selected, then instead of cumulative probabilities, the support point eta vector that best fits that subject (ETM) is listed.

3.17 root.npe (NM73)

The expected value etas and expected value eta covariances (ETC) are listed for each problem or sub-problem. Because only one line is written per problem or sub-problem, the column header is displayed (unless NOLABEL=1) only once for the entire NONMEM run. However, each line contains information of table number, problem number, sub-problem number, super problem and iteration number.

3.18 root.npi (NM73)

The individual probabilities are listed in this file. The header line (unless NOLABEL=1) is written only once, at the beginning of the file, per NONMEM run. Each line contains information of table number, problem number, sub-problem number, super problem, iteration number, subject number, and ID. This is followed by the individual probabilities at each support point (of which there are NSUPP/NSUPPE or NIND of them, whichever is greater). The line with Subject number=0 contains the sum of the probabilities of all the subjects, and is similar or exactly equal to the joint probability of each support point listed in root.npd under the column PROBABILITY. That they are not equal is due to the convergence limit of the non-parametric analysis. Row of subject number I, column of support K, contains the individual probability IPROB(I,K). The sum of the individual probabilities over all support points for any given line (subject), is equal to 1/NIND. Thus, the sum of all items across rows and columns (not including subject 0) sums to 1. The format of the file is fixed at (,1PE22.15), and cannot be changed. It is intended for use in further analysis by analytical software, and is designed to report the full double-precision information of each probability.

3.19 root.npl (NM74)

The individual data likelihoods (not including the parameter density) are listed in this file. The header line (unless NOLABEL=1) is written only once, at the beginning of the file, per NONMEM run. Each line contains information of table number, problem number, sub-problem number, super problem, iteration number, subject number, and ID. This is followed by the individual likelihoods at each support point (of which there are NSUPP/NSUPPE or NIND of them, whichever is greater). Unlike the .npi file, there is no line with Subject number=0. The row of subject number I, column of support K, contains the individual likelihood LIK(I,K). The format of the file is fixed at (,1PE22.15), and cannot be changed. It is intended for use in further analysis by analytical software, and is designed to report the full double-precision information of each probability.

The LIK(I,K) of the .npl file and the IPROB(I,K) are related to each other as follows:

IPROB(I,K)=PI(K)*LIK(I,K)/exp(-1/2*OBJ(I))/NIND

where NIND is number of subjects, PI(K) is the probability of a support point (found as item PROBABILITY in the .npd file), and OBJ(I) is the objective function contribution of subject I (found under OBJ column of the .npl or .npi file).

3.20 root.fgh (NM73)

This file is produced if the user selects $EST NUMDER=1 or 3. The file lists the numerically evaluated derivatives of Y with respect to eta, where

G(I,1)=partial Y with respect to eta(i))
G(I,J+1)=Second derivatives of Y with respect to eta(i),eta(j)
H(I,1)=partial Y with respect to eps(i)
H(i,j+1)=partial Y with respect to eps(i),eta(j))

3.21 root.agh (NM73)

This file is produced if the user selects $EST NUMDER=2 or 3. The file lists the analytically evaluated derivatives of Y with respect to eta, from the PK(), ERROR(), and/or PRED() routines in FSUBS, where

G(I,1)=partial Y with respect to eta(i))
G(I,J+1)=Second derivatives of Y with respect to eta(i),eta(j) (not always evaluated by FSUBS)
H(I,1)=partial Y with respect to eps(i)
H(i,j+1)=partial Y with respect to eps(i),eta(j))

3.22 root.cpu (NM73)

The cpu time in seconds is reported in this file. It is an accurate representation of the computer usage, whether single or parallel process. The same problem when run singly or in parallel will report a similar cpu time. This is in contrast with elapsed time, which is improved with parallelization.

3.23 root.vpd (NM74)

Variance-covariance among user-defined parameters and PREDPP parameters are placed in this file, associated with conditional individual variances of the .phi file. The COMACT column value is usually 2, or 3 if $NONPARAMETRIC ETAS was implemented. The subsequent columns contain the values of user-defined table items, followed by variance-covariance of these items (labeled VAR1_VAR2, where VAR1 and VAR2 are the particular table items).

3.24 root.vpt (NM75)

Variance-covariance among user-defined parameters and PREDPP parameters are placed in this file, associated with conditional individual variances of the .phi file, as well as the variance-covariance of the fixed effects parameters (THETAS, SIGMAS, OMEGAS) of the .cov file. The COMACT column value is 1 to indicate the record contains the values during a COMACT=1 pass, that is with ETAs set to 0. Thus, an individual predicted value (IPRED) on this record would in fact by the population predicted value (PRED). The subsequent columns contain the values of user-defined table items, followed by variance-covariance of these items (labled VAR1_VAR2, where VAR1 and VAR2 are the particular table items). The variance IPRED_IPRED for a record with COMACT=1 would then have essentially the statistical error associated with the variance-covariance of the appropriate THETAs. When COMACT=2, table variables are evaluated having the etas set to the empirical bayes estimates (depending on the last estimation method used, and what the FNLETA value is set to), variance-covariance of these items are evaluated based on conditional individual variances of the .phi table, and variance-covariance of the appropriate THETAs, SIGMAs, and OMEGAs, as applicable. If COMACT=3, then non-parametric estimated etas are used.

3.25 root.ets (NM74)

Random samples of individual etas when $EST METHOD=SAEM ETASAMPLES=1 or $EST METHOD=BAYES ETASAMPLE=1. ETASAMPLES=1 causes individual ISAMPLE random eta samples per subject, to be written to root.ets, where root is the root name of the control stream file.

3.26 root.bfm (NM75)

When OFVTYPE=8 during optimal design ($DESIGN OFVTYPE=8) the progress of average individual conditional variances are shown in the root.bfm file (where root is name of control stream file), and the final one on the -1000000000 line, during an optimization. The RAW file (by default named root.ext) which shows only the starting, and final values of the standard errors of population parameters, has extra information.

3.27 msfroot_ETAS

When MSF or MSFO option is used to specify an MSFO file in the $EST record:

$EST … MSFO=msfroot.msf

then in addition to the main MSF file msfroot.msf, file msfroot_ETAS.msf containing individual etas and phis generated during estimation (except FO, which does not generate non-zero etas during estimation), will also be produced, and provide additional information when a $MSFI record is used in a subsequent problem or control stream. In addition, this files stores information useful for resuming an interrupted EM/BAYES estimation. The _ETAS file does not contain POSTHOC evaluated etas, only estimated (during $EST) evaluated etas. Also, ensure that the ordering of subjects in the present problem matches that of the problem when the msf file was generated (for example, do not change the data file structure or IGNORE/ACCEPT filtering between the two problems).

3.28 msfroot_RMAT, msfroot_SMAT (nm73)