nlmixr

Multiple endpoints

Joint PK/PD models, or PK/PD models where you fix certain components are common in pharmacometrics. A classic example, (provided by Tomoo Funaki and Nick Holford) is Warfarin.

In this example, we have a transit-compartment (from depot to gut to central volume) PK model and an effect compartment for the PCA measurement.

Below is an illustrated example of a model that can be applied to the data:

pk.turnover.emax <- function() {
  ini({
    tktr <- log(1)
    tka <- log(1)
    tcl <- log(0.1)
    tv <- log(10)
    ##
    eta.ktr ~ 1
    eta.ka ~ 1
    eta.cl ~ 2
    eta.v ~ 1
    prop.err <- 0.1
    pkadd.err <- 0.1
    ##
    temax <- logit(0.8)
    #temax <- 7.5
    tec50 <- log(0.5)
    tkout <- log(0.05)
    te0 <- log(100)
    ##
    eta.emax ~ .5
    eta.ec50  ~ .5
    eta.kout ~ .5
    eta.e0 ~ .5
    ##
    pdadd.err <- 10
  })
  model({
    ktr <- exp(tktr + eta.ktr)
    ka <- exp(tka + eta.ka)
    cl <- exp(tcl + eta.cl)
    v <- exp(tv + eta.v)
    ##
    #poplogit = log(temax/(1-temax))
    emax=expit(temax+eta.emax)
    #logit=temax+eta.emax
    ec50 =  exp(tec50 + eta.ec50)
    kout = exp(tkout + eta.kout)
    e0 = exp(te0 + eta.e0)
    ##
    DCP = center/v
    PD=1-emax*DCP/(ec50+DCP)
    ##
    effect(0) = e0
    kin = e0*kout
    ##
    d/dt(depot) = -ktr * depot
    d/dt(gut) =  ktr * depot -ka * gut
    d/dt(center) =  ka * gut - cl / v * center
    d/dt(effect) = kin*PD -kout*effect
    ##
    cp = center / v
    cp ~ prop(prop.err) + add(pkadd.err)
    effect ~ add(pdadd.err)
  })
}

Notice there are two endpoints in the model cp and effect. Both are modeled in nlmixr using the ~ “modeled by” specification.

To see more about how nlmixr will handle the multiple compartment model, it is quite informative to parse the model and print the information about that model. In this case an initial parsing would give:

ui <- nlmixr(pk.turnover.emax)
ui
#> ▂▂ RxODE-based ODE model ▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂ 
#> ── Initialization: ───────────────────────────────────────────────────────────── 
#> Fixed Effects ($theta): 
#>       tktr        tka        tcl         tv      temax      tec50      tkout 
#>  0.0000000  0.0000000 -2.3025851  2.3025851  1.3862944 -0.6931472 -2.9957323 
#>        te0 
#>  4.6051702 
#> 
#> Omega ($omega): 
#>          eta.ktr eta.ka eta.cl eta.v eta.emax eta.ec50 eta.kout eta.e0
#> eta.ktr        1      0      0     0      0.0      0.0      0.0    0.0
#> eta.ka         0      1      0     0      0.0      0.0      0.0    0.0
#> eta.cl         0      0      2     0      0.0      0.0      0.0    0.0
#> eta.v          0      0      0     1      0.0      0.0      0.0    0.0
#> eta.emax       0      0      0     0      0.5      0.0      0.0    0.0
#> eta.ec50       0      0      0     0      0.0      0.5      0.0    0.0
#> eta.kout       0      0      0     0      0.0      0.0      0.5    0.0
#> eta.e0         0      0      0     0      0.0      0.0      0.0    0.5
#> ── Multiple Endpoint Model ($multipleEndpoint): ──────────────────────────────── 
#> ┌─────────────────────────┬─────────────────────────┬─────────────────────────┐
#> │ variable                │ cmt                     │ dvid*                   │
#> ├─────────────────────────┼─────────────────────────┼─────────────────────────┤
#> │ variable                │ cmt                     │ dvid*                   │
#> ├─────────────────────────┼─────────────────────────┼─────────────────────────┤
#> │ cp ~ …                  │ cmt='cp' or cmt=5       │ dvid='cp' or dvid=1     │
#> ├─────────────────────────┼─────────────────────────┼─────────────────────────┤
#> │ effect ~ …              │ cmt='effect' or cmt=4   │ dvid='effect' or dvid=2 │
#> └─────────────────────────┴─────────────────────────┴─────────────────────────┘
#>   * If dvids are outside this range, all dvids are re-numered                  
#>   sequentially, ie 1,7, 10 becomes 1,2,3 etc                                   
#> 
#> ── μ-referencing ($muRefTable): ──────────────────────────────────────────────── 
#>                              ┌─────────┬──────────┐
#>                              │ theta   │ eta      │
#>                              ├─────────┼──────────┤
#>                              │ tktr    │ eta.ktr  │
#>                              ├─────────┼──────────┤
#>                              │ tka     │ eta.ka   │
#>                              ├─────────┼──────────┤
#>                              │ tcl     │ eta.cl   │
#>                              ├─────────┼──────────┤
#>                              │ tv      │ eta.v    │
#>                              ├─────────┼──────────┤
#>                              │ temax   │ eta.emax │
#>                              ├─────────┼──────────┤
#>                              │ tec50   │ eta.ec50 │
#>                              ├─────────┼──────────┤
#>                              │ tkout   │ eta.kout │
#>                              ├─────────┼──────────┤
#>                              │ te0     │ eta.e0   │
#>                              └─────────┴──────────┘
#> 
#> ── Model: ────────────────────────────────────────────────────────────────────── 
#>     ktr <- exp(tktr + eta.ktr)
#>     ka <- exp(tka + eta.ka)
#>     cl <- exp(tcl + eta.cl)
#>     v <- exp(tv + eta.v)
#>     ##
#>     #poplogit = log(temax/(1-temax))
#>     emax=expit(temax+eta.emax)
#>     #logit=temax+eta.emax
#>     ec50 =  exp(tec50 + eta.ec50)
#>     kout = exp(tkout + eta.kout)
#>     e0 = exp(te0 + eta.e0)
#>     ##
#>     DCP = center/v
#>     PD=1-emax*DCP/(ec50+DCP)
#>     ##
#>     effect(0) = e0
#>     kin = e0*kout
#>     ##
#>     d/dt(depot) = -ktr * depot
#>     d/dt(gut) =  ktr * depot -ka * gut
#>     d/dt(center) =  ka * gut - cl / v * center
#>     d/dt(effect) = kin*PD -kout*effect
#>     ##
#>     cp = center / v
#>     cp ~ prop(prop.err) + add(pkadd.err)
#>     effect ~ add(pdadd.err) 
#> ▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂

In the middle of the printout, it shows how the data must be formatted (using the cmt and dvid data items) to allow nlmixr to model the multiple endpoint appropriately.

Of course if you are interested you can directly access the information in ui$multipleEndpoint.

ui$multipleEndpoint
variable cmt dvid*
variable cmt dvid*
cp ~ … cmt='cp' or cmt=5 dvid='cp' or dvid=1
effect ~ … cmt='effect' or cmt=4 dvid='effect' or dvid=2
* If dvids are outside this range, all dvids are re-numered sequentially, ie 1,7, 10 becomes 1,2,3 etc

Notice that the cmt and dvid items can use the named variables directly as either the cmt or dvid specification. This flexible notation makes it so you do not have to rename your compartments to run nlmixr model functions.

The other thing to note is that the cp is specified by an ODE compartment above the number of compartments defined in the RxODE part of the nlmixr model. This is because cp is not a defined compartment, but a related variable cp.

The last thing to notice that the cmt items are numbered cmt=5 for cp or cmt=4 for effect even though they were specified in the model first by cp and cmt. This ordering is because effect is a compartment in the RxODE system. Of course cp is related to the compartment central, and it may make more sense to pair cp with the central compartment.

If this is something you want to have you can specify the compartment to relate the effect to by the | operator. In this case you would change

cp ~ prop(prop.err) + add(pkadd.err)

to

cp ~ prop(prop.err) + add(pkadd.err) | central

With this change, the model could be updated to:

pk.turnover.emax2 <- function() {
  ini({
    tktr <- log(1)
    tka <- log(1)
    tcl <- log(0.1)
    tv <- log(10)
    ##
    eta.ktr ~ 1
    eta.ka ~ 1
    eta.cl ~ 2
    eta.v ~ 1
    prop.err <- 0.1
    pkadd.err <- 0.1
    ##
    temax <- logit(0.8)
    tec50 <- log(0.5)
    tkout <- log(0.05)
    te0 <- log(100)
    ##
    eta.emax ~ .5
    eta.ec50  ~ .5
    eta.kout ~ .5
    eta.e0 ~ .5
    ##
    pdadd.err <- 10
  })
  model({
    ktr <- exp(tktr + eta.ktr)
    ka <- exp(tka + eta.ka)
    cl <- exp(tcl + eta.cl)
    v <- exp(tv + eta.v)
    ##
    emax=expit(temax+eta.emax)
    ec50 =  exp(tec50 + eta.ec50)
    kout = exp(tkout + eta.kout)
    e0 = exp(te0 + eta.e0)
    ##
    DCP = center/v
    PD=1-emax*DCP/(ec50+DCP)
    ##
    effect(0) = e0
    kin = e0*kout
    ##
    d/dt(depot) = -ktr * depot
    d/dt(gut) =  ktr * depot -ka * gut
    d/dt(center) =  ka * gut - cl / v * center
    d/dt(effect) = kin*PD -kout*effect
    ##
    cp = center / v
    cp ~ prop(prop.err) + add(pkadd.err) | center
    effect ~ add(pdadd.err)
  })
}
ui2 <- nlmixr(pk.turnover.emax2)
ui2$multipleEndpoint
variable cmt dvid*
variable cmt dvid*
cp ~ … cmt='center' or cmt=3 dvid='center' or dvid=1
effect ~ … cmt='effect' or cmt=4 dvid='effect' or dvid=2
* If dvids are outside this range, all dvids are re-numered sequentially, ie 1,7, 10 becomes 1,2,3 etc

Notice in this case the cmt variables are numbered sequentially and the cp variable matches the center compartment.

DVID vs CMT, which one is used

When dvid and cmt are combined in the same dataset, the cmt data item is always used on the event information and the dvid is used on the observations. nlmixr expects the cmt data item to match the dvid item for observations OR to be either zero or one for the dvid to replace the cmt information.

If you do not wish to use dvid items to define multiple endpoints in nlmixr, you can set the following option:

options(RxODE.combine.dvid=FALSE)
ui2$multipleEndpoint
variable cmt
variable cmt
cp ~ … cmt='center' or cmt=3
effect ~ … cmt='effect' or cmt=4

Then only cmt items are used for the multiple endpoint models. Of course you can turn it on or off for different models if you wish:

options(RxODE.combine.dvid=TRUE)
ui2$multipleEndpoint
variable cmt dvid*
variable cmt dvid*
cp ~ … cmt='center' or cmt=3 dvid='center' or dvid=1
effect ~ … cmt='effect' or cmt=4 dvid='effect' or dvid=2
* If dvids are outside this range, all dvids are re-numered sequentially, ie 1,7, 10 becomes 1,2,3 etc

Running a multiple endpoint model

With this information, we can use the built-in warfarin dataset in nlmixr:

summary(warfarin)
#>        id             time             amt                dv          dvid    
#>  Min.   : 1.00   Min.   :  0.00   Min.   :  0.000   Min.   :  0.00   cp :283  
#>  1st Qu.: 8.00   1st Qu.: 24.00   1st Qu.:  0.000   1st Qu.:  4.50   pca:232  
#>  Median :15.00   Median : 48.00   Median :  0.000   Median : 11.40            
#>  Mean   :16.08   Mean   : 52.08   Mean   :  6.524   Mean   : 20.02            
#>  3rd Qu.:24.00   3rd Qu.: 96.00   3rd Qu.:  0.000   3rd Qu.: 26.00            
#>  Max.   :33.00   Max.   :144.00   Max.   :153.000   Max.   :100.00            
#>       evid               wt              age            sex     
#>  Min.   :0.00000   Min.   : 40.00   Min.   :21.00   female:101  
#>  1st Qu.:0.00000   1st Qu.: 60.00   1st Qu.:23.00   male  :414  
#>  Median :0.00000   Median : 70.00   Median :28.00               
#>  Mean   :0.06214   Mean   : 69.27   Mean   :31.85               
#>  3rd Qu.:0.00000   3rd Qu.: 78.00   3rd Qu.:36.00               
#>  Max.   :1.00000   Max.   :102.00   Max.   :63.00

Since dvid specifies pca as the effect endpoint, you can update the model to be more explicit making one last change:

cp ~ prop(prop.err) + add(pkadd.err)
effect ~ add(pdadd.err) 

to

cp ~ prop(prop.err) + add(pkadd.err)
effect ~ add(pdadd.err)  | pca
pk.turnover.emax3 <- function() {
  ini({
    tktr <- log(1)
    tka <- log(1)
    tcl <- log(0.1)
    tv <- log(10)
    ##
    eta.ktr ~ 1
    eta.ka ~ 1
    eta.cl ~ 2
    eta.v ~ 1
    prop.err <- 0.1
    pkadd.err <- 0.1
    ##
    temax <- logit(0.8)
    tec50 <- log(0.5)
    tkout <- log(0.05)
    te0 <- log(100)
    ##
    eta.emax ~ .5
    eta.ec50  ~ .5
    eta.kout ~ .5
    eta.e0 ~ .5
    ##
    pdadd.err <- 10
  })
  model({
    ktr <- exp(tktr + eta.ktr)
    ka <- exp(tka + eta.ka)
    cl <- exp(tcl + eta.cl)
    v <- exp(tv + eta.v)
    emax = expit(temax+eta.emax)
    ec50 =  exp(tec50 + eta.ec50)
    kout = exp(tkout + eta.kout)
    e0 = exp(te0 + eta.e0)
    ##
    DCP = center/v
    PD=1-emax*DCP/(ec50+DCP)
    ##
    effect(0) = e0
    kin = e0*kout
    ##
    d/dt(depot) = -ktr * depot
    d/dt(gut) =  ktr * depot -ka * gut
    d/dt(center) =  ka * gut - cl / v * center
    d/dt(effect) = kin*PD -kout*effect
    ##
    cp = center / v
    cp ~ prop(prop.err) + add(pkadd.err)
    effect ~ add(pdadd.err) | pca
  })
}

Run the models with SAEM

fit.TOS <- nlmixr(pk.turnover.emax3, warfarin, "saem", control=list(print=0),
                  table=list(cwres=TRUE, npde=TRUE))
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:01 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [1] "CMT"

print(fit.TOS)
#> ── nlmixr SAEM(ODE); OBJF by FOCEi approximation fit ───────────────────────────
#> 
#>           OBJF      AIC      BIC Log-likelihood Condition Number
#> FOCEi 1388.433 2314.127 2393.548      -1138.064         697.1191
#> 
#> ── Time (sec $time): ───────────────────────────────────────────────────────────
#> 
#>            saem   setup table  cwres covariance    other
#> elapsed 169.139 15.5767 5.107 20.834      0.085 1.052299
#> 
#> ── Population Parameters ($parFixed or $parFixedDf): ───────────────────────────
#> 
#>              Est.     SE  %RSE Back-transformed(95%CI) BSV(CV% or SD)
#> tktr        0.228  0.274   120      1.26 (0.734, 2.15)           94.4
#> tka       -0.0435   0.18   414     0.957 (0.673, 1.36)           43.2
#> tcl         -1.96 0.0514  2.62     0.14 (0.127, 0.155)           26.8
#> tv              2 0.0467  2.33       7.42 (6.78, 8.14)           21.0
#> prop.err    0.121                                0.121               
#> pkadd.err   0.813                                0.813               
#> temax        2.96  0.265  8.94      0.951 (0.92, 0.97)         0.0877
#> tec50      -0.186  0.117  62.8       0.83 (0.66, 1.04)           48.4
#> tkout       -2.91 0.0327  1.12 0.0545 (0.0512, 0.0581)           7.39
#> te0          4.57 0.0111 0.243       96.6 (94.5, 98.7)           4.86
#> pdadd.err    3.69                                 3.69               
#>           Shrink(SD)%
#> tktr           52.1% 
#> tka            64.6% 
#> tcl            7.57% 
#> tv             17.9% 
#> prop.err             
#> pkadd.err            
#> temax          89.6% 
#> tec50          8.93% 
#> tkout          42.2% 
#> te0            18.3% 
#> pdadd.err            
#>  
#>   Covariance Type ($covMethod): linFim
#>   Fixed parameter correlations in $cor
#>   No correlations in between subject variability (BSV) matrix
#>   Full BSV covariance ($omega) or correlation ($omegaR; diagonals=SDs) 
#>   Distribution stats (mean/skewness/kurtosis/p-value) available in $shrink 
#> 
#> ── Fit Data (object is a modified tibble): ─────────────────────────────────────
#> # A tibble: 483 x 42
#>   ID     TIME CMT      DV EPRED  ERES   NPDE    NPD  PRED   RES   WRES IPRED
#>   <fct> <dbl> <fct> <dbl> <dbl> <dbl>  <dbl>  <dbl> <dbl> <dbl>  <dbl> <dbl>
#> 1 1       0.5 cp      0    1.80 -1.80 -0.553 -0.553  1.41 -1.41 -1.04  0.512
#> 2 1       1   cp      1.9  4.46 -2.56  2.94   2.94   3.99 -2.09 -0.789 1.65 
#> 3 1       2   cp      3.3  8.40 -5.10 -1.79  -1.79   8.47 -5.17 -1.40  4.38 
#> # … with 480 more rows, and 30 more variables: IRES <dbl>, IWRES <dbl>,
#> #   CPRED <dbl>, CRES <dbl>, CWRES <dbl>, eta.ktr <dbl>, eta.ka <dbl>,
#> #   eta.cl <dbl>, eta.v <dbl>, eta.emax <dbl>, eta.ec50 <dbl>, eta.kout <dbl>,
#> #   eta.e0 <dbl>, depot <dbl>, gut <dbl>, center <dbl>, effect <dbl>,
#> #   ktr <dbl>, ka <dbl>, cl <dbl>, v <dbl>, emax <dbl>, ec50 <dbl>, kout <dbl>,
#> #   e0 <dbl>, DCP <dbl>, PD <dbl>, kin <dbl>, tad <dbl>, dosenum <dbl>

SAEM Diagnostic plots

plot(fit.TOS);



v1s <- nlmixr::vpc(fit.TOS, show=list(obs_dv=T), scales="free_y") +
  ylab("Warfarin Cp [mg/L] or PCA") +
  xlab("Time [h]")

v2s <- nlmixr::vpc(fit.TOS, show=list(obs_dv=T), pred_corr = TRUE) +
  ylab("Prediction Corrected Warfarin Cp [mg/L] or PCA") +
  xlab("Time [h]")

library(patchwork)
v1s / v2s

FOCEi fits

## FOCEi fit/vpcs
fit.TOF <- nlmixr(pk.turnover.emax3, warfarin, "focei", control=list(print=0),
                  table=list(cwres=TRUE, npde=TRUE));
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:01 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [====|====|====|====|====|====|====|====|====|====] 0:00:00 
#> 
#> [1] "CMT"
#> calculating covariance matrix
#> [====|====|====|====|====|====|====|====|====|====] 0:02:14 
#> done

FOCEi Diagnostic Plots

print(fit.TOF)
#> ── nlmixr FOCEi (outer: nlminb) fit ────────────────────────────────────────────
#> 
#>           OBJF      AIC      BIC Log-likelihood Condition Number
#> FOCEi 1339.898 2265.592 2345.013      -1113.796         1311.187
#> 
#> ── Time (sec $time): ───────────────────────────────────────────────────────────
#> 
#>            setup optimize covariance table    other
#> elapsed 21.94584 134.5003   134.5003 4.072 525.1666
#> 
#> ── Population Parameters ($parFixed or $parFixedDf): ───────────────────────────
#> 
#>              Est.     SE  %RSE Back-transformed(95%CI) BSV(CV% or SD)
#> tktr        0.373  0.208  55.9      1.45 (0.965, 2.18)           131.
#> tka       -0.0656 0.0672   102     0.937 (0.821, 1.07)           83.7
#> tcl         -2.03 0.0545  2.69    0.132 (0.119, 0.147)           28.8
#> tv           2.07  0.036  1.74       7.91 (7.37, 8.49)           24.9
#> prop.err    0.112                                0.112               
#> pkadd.err   0.374                                0.374               
#> temax        4.45  0.286  6.44     0.988 (0.98, 0.993)           1.01
#> tec50       0.143 0.0718  50.3          1.15 (1, 1.33)           47.9
#> tkout       -2.94   0.02  0.68   0.053 (0.051, 0.0551)           9.62
#> te0          4.57 0.0113 0.247       96.6 (94.5, 98.7)           7.08
#> pdadd.err    3.77                                 3.77               
#>           Shrink(SD)%
#> tktr           61.1% 
#> tka            63.2% 
#> tcl            1.33% 
#> tv             12.4% 
#> prop.err             
#> pkadd.err            
#> temax          88.1% 
#> tec50          7.67% 
#> tkout          33.0% 
#> te0            27.6% 
#> pdadd.err            
#>  
#>   Covariance Type ($covMethod): r,s
#>   Some strong fixed parameter correlations exist ($cor) :
#>        cor:tka,tktr    cor:tcl,tktr     cor:tv,tktr  cor:temax,tktr  cor:tec50,tktr 
#>          0.888          -0.160           0.453           0.521           0.185  
#>  cor:tkout,tktr    cor:te0,tktr     cor:tcl,tka      cor:tv,tka   cor:temax,tka 
#>          0.350           0.275          -0.168           0.140           0.286  
#>   cor:tec50,tka   cor:tkout,tka     cor:te0,tka      cor:tv,tcl   cor:temax,tcl 
#>          0.117           0.209           0.221          0.0696          -0.142  
#>   cor:tec50,tcl   cor:tkout,tcl     cor:te0,tcl    cor:temax,tv    cor:tec50,tv 
#>         -0.326        -0.00851          -0.135           0.631          0.0327  
#>    cor:tkout,tv      cor:te0,tv cor:tec50,temax cor:tkout,temax   cor:te0,temax 
#>         0.0159         -0.0706           0.286           0.301          -0.134  
#> cor:tkout,tec50   cor:te0,tec50   cor:te0,tkout 
#>          0.583        -0.00789          0.0574  
#>  
#> 
#>   No correlations in between subject variability (BSV) matrix
#>   Full BSV covariance ($omega) or correlation ($omegaR; diagonals=SDs) 
#>   Distribution stats (mean/skewness/kurtosis/p-value) available in $shrink 
#>   Minimization message ($message):  
#>     false convergence (8) 
#>   In an ODE system, false convergence may mean "useless" evaluations were performed.
#>   See https://tinyurl.com/yyrrwkce
#>   It could also mean the convergence is poor, check results before accepting fit
#>   You may also try a good derivative free optimization:
#>     nlmixr(...,control=list(outerOpt="bobyqa"))
#> 
#> ── Fit Data (object is a modified tibble): ─────────────────────────────────────
#> # A tibble: 483 x 42
#>   ID     TIME CMT      DV EPRED  ERES   NPDE    NPD  PRED   RES   WRES IPRED
#>   <fct> <dbl> <fct> <dbl> <dbl> <dbl>  <dbl>  <dbl> <dbl> <dbl>  <dbl> <dbl>
#> 1 1       0.5 cp      0    2.06 -2.06 -1.79  -1.79   1.46 -1.46 -0.944 0.440
#> 2 1       1   cp      1.9  4.57 -2.67  2.47   2.47   4.03 -2.13 -0.635 1.45 
#> 3 1       2   cp      3.3  7.92 -4.62 -0.367 -0.367  8.30 -5.00 -1.13  3.95 
#> # … with 480 more rows, and 30 more variables: IRES <dbl>, IWRES <dbl>,
#> #   CPRED <dbl>, CRES <dbl>, CWRES <dbl>, eta.ktr <dbl>, eta.ka <dbl>,
#> #   eta.cl <dbl>, eta.v <dbl>, eta.emax <dbl>, eta.ec50 <dbl>, eta.kout <dbl>,
#> #   eta.e0 <dbl>, depot <dbl>, gut <dbl>, center <dbl>, effect <dbl>,
#> #   ktr <dbl>, ka <dbl>, cl <dbl>, v <dbl>, emax <dbl>, ec50 <dbl>, kout <dbl>,
#> #   e0 <dbl>, DCP <dbl>, PD <dbl>, kin <dbl>, tad <dbl>, dosenum <dbl>
plot(fit.TOF)


v1f <- nlmixr::vpc(fit.TOF, show=list(obs_dv=T), scales="free_y") +
  ylab("Warfarin Cp [mg/L] or PCA") +
  xlab("Time [h]")

v2f <- nlmixr::vpc(fit.TOF, show=list(obs_dv=T), pred_corr = TRUE) +
  ylab("Prediction Corrected Warfarin Cp [mg/L] or PCA") +
  xlab("Time [h]")

v1f / v2f