Concentration-Dependent CL in Pumas

1

Hello - I have a prior model with concentration-dependent clearance to capture an autoinduction effect. However, now that I’m fitting a new, very large & sparse dataset this model is too computationally expensive to feasibly run. Below is how I’m currently describing concentration-dependent CL in the vars block.

  @vars begin 
    conc = Central / Vc
    CL = CL0 * (1+((Emax*conc)/(EC50+conc))) 
  end

This model fits my original, “reduced” dataset well and is well qualified. I’m wondering if using a piecewise function may be a better solution. I’ve tried calling the below function in the @vars block but haven’t had much luck. Is this supported in Pumas or is there another way I could approach this problem of computational burden? As a side note, I’ve tried using both FOCE and SAEM. FOCE run times are astronomical and SAEM converges to infinity.

function piecewise_CL(C, CL_base)
    if C < 521.12
        return CL_base
    elseif C < 1042.23
        return CL_base + 0.25 * 10.8207
    elseif C < 2084.46
        return CL_base + 0.5 * 10.8207
    elseif C < 4168.92
        return CL_base + 0.75 * 10.8207
    else
        return CL_base + 10.8207
    end
end

Any suggestions are greatly appreciated - thank you!

2

@adunn34 before you actually fit the new model, can you try the following

  1. Simulate using the reduce model parameter estimates into the full dataset and see if the reduced model captures the distribution of your larger dataset?
  2. Given that your large dataset has a lot of sparse information, would fixing some of the parameters from the reduced model help, especially those related to time varying clearance?