M3 method with SAEM

Estimation
1

Hello,

I am wondering if we can use M3 method with SAEM?
I don’t see such option in the error model for EM models here, not sure if there are any examples of manually calculating likelihood

2

Hi, yes M3 in Censored().

Here’s an example that you can run in Pumas:

using Pumas
using PharmaDatasets

mdl_censored_em = @emmodel begin
    @param begin
        Kraw ~ 1 | LogitNormal
    end
    @random begin
        Ka ~ 1 + SEX | LogNormal
        CLf ~ 1 | LogNormal
    end
    @covariates WT
    @covariance 1 1
    @pre begin
        CL = CLf * WT
        K = Kraw * (0.5 - 0.008) + 0.008
        Vc = CL / K
    end
    @dynamics Depots1Central1
    @post begin
        SC = Vc / WT
        conc = Central / SC
    end
    @error begin
        dv ~ Censored(Normal(conc), 1.0, Inf)
    end
end

df = dataset("pumas/event_data/THEOPP")
df.dv = max.(1.0, df.dv)

param = (
    #Ka MEAN ABSORPTION RATE CONSTANT for SEX = 0(1/HR), and diff for 0
    Ka=(2.77, log(1.5) - log(2.77)),
    Kraw=(0.0781 - 0.008) / (0.5 - 0.008), #K MEAN ELIMINATION RATE CONSTANT (1/HR)
    CLf=0.0363, #SLP  SLOPE OF CLEARANCE VS WEIGHT RELATIONSHIP (LITERS/HR/KG)
)

theopp_censored = read_pumas(df, covariates=[:WT, :SEX])

ft_censored_saem = @time fit(
    mdl_censored_em,
    theopp_censored,
    param,
    SAEM(),
)
2 Likes
3

just run the loglikelihood function, but with LaplaceI instead of SAEM:

loglikelihood(
    mdl_censored_em,
    theopp_censored,
    param,
    LaplaceI(),
)
1 Like