search

Open-Systems-Pharmacology / Forum

Discussion forum for the Open Systems Pharmacology Project
65 stars 19 forks source link

Pediatric extrapolation: Overprediction of Vss (issue #354) #89

Closed vbalbas closed 6 years ago

vbalbas commented 6 years ago

Dear all,

I would like to share some issues I have with my project and maybe anyone could have some suggestions about how can I improve my PBPK model.

I am trying to extrapolate to pediatric population an already existing model for adults in PKSim (Kuepfer L, Niederalt C, Wendl T, Schlender JF, Willmann S, Lippert J, et al. Applied Concepts in PBPK Modeling: How to Build a PBPK/PD Model. CPT Pharmacometrics Syst Pharmacol. 2016;5(10):516–31).

The drug (Ciprofloxacin), present its CL splitted in three different processes: Enzymatic: CYP1A2, Bile, and Renal (GFR + TS). I would like to know if there is any way in which I can know the CL prediction per CL pathway.

The adult model predicts very well plasma concentration for IV and PO administration. Also CL and Vss for IV administration are well predicted as well, comparing them with the PK parameters for adult population obtained from the different literature trials. But for PO administration, Vss/F is overpredicted.

ADULT IV

PO

Pred vs observed for the adult population (dashed line refers to the twofold time prediction interval): image

When I extrapolate the model to pediatric population (in-house data with customized fu and GFR per ID), the model overpredicts Vss in IV and much more in PO (Vss/F) parameters, so we should figure out why (In that case, the reference parameters come from a popPK model that is being developed).

CHILDREN IV

PO

Pred vs observed for the adult population (dashed line refers to the twofold time prediction interval): image

It seems that there is a problem overpredicting Vss, specially after PO administration. Any suggestions?

I also would like to know if there is any trick to obtain as an output of the simulation the CL prediction per CL pathway.

Thank you in advance Violeta

Aedginto commented 6 years ago

Hi Violeta, A few comments first. The bioavailability of cipro from the popPK model is about 50% in adults (which is lower than the expected of 60-70%) and 94% in children. At least that is what I am getting from your popPK estimates. This is a high F in kids....from a previously published popPK model in a rather large age range of kids, F in kids (61%) is similar to adults (Rajagopalan. Population pharmacokinetics of ciprofloxacin in pediatric patients. J Clin Pharmacol). Using CL as the reference in children, the bioavailability as predicted from PK-Sim is 65%. If the model is built on adults (F = 61% in the adult model as reported from your numbers) and the drug is about 35% hepatically cleared (low extraction ratio of about 0.2) in adults (biliary is fully mature and 1A2 is not), 65% in children is a reasonable prediction. Considering that cipro is 2/3rds renally cleared in adults (more in kids), it would be suspect to predict 94% in kids.

The Vss/F is wonky. I recreated a cipro model (in general) and I can't recreate such as huge Vss difference between IV and oral. On a side note, the V is set in PK-Sim mechanistically meaning that there is one V that is derived for each compound, regardless of the administration. The V that you are looking at in the PK-Analysis window is derived as one would derive it from a concentration vs. time curve. Thus the V from the time curve is dependent on the how long the simulation goes for. Always make sure that you you are finishing the simulation at the same time as the observed data! For cipro, this makes a big difference because oral uptake occurs along the length of the intestines and, as such, 12 hours vs. 72 hours changes the dose arriving systemically (changing all PK parameters).

I have a few questions before I can figure out what's going on:-)

  1. How old are the kids in the trial? By the numbers I'm guessing in the 2-6 yr range?
  2. Did you create one virtual child for every real child and predict the profile and compare to observed levels? I think so, just checking.
  3. How did you incorporate the GFR for each kid? If you could give me an example, that would be helpful. As a check, you could just leave the GFR as estimated for the age of the child and see if the under-prediction in concentrations remains.
  4. How did you incorporate tubular secretion ontogeny?

Take care! Andrea

vbalbas commented 6 years ago

Hi Andrea,

First of all, thank you very much for your reply. I am answering you back under your comments. If something is not clear, just let me know.

A few comments first. The bioavailability of cipro from the popPK model is about 50% in adults (which is lower than the expected of 60-70%) and 94% in children.

The issue here is that I did not find a popPK model for cipro for healthy subjects (only for ICU patients: Optimizing ciprofloxacin dosing in intensive care unit patients through the use of population pharmacokinetic-pharmacodynamic analysis and monte carlo simulations. J Antimicrob Chemother. 2011). So the PK parameters for adult population (IV and PO) are obtained as an average from the different literature parameters from trials which underwent non parametric analysis. Anyway, it seems that the F is not right.

At least that is what I am getting from your popPK estimates. This is a high F in kids....from a previously published popPK model in a rather large age range of kids, F in kids (61%) is similar to adults (Rajagopalan. Population pharmacokinetics of ciprofloxacin in pediatric patients. J Clin Pharmacol).

l completely agree with you, but I did not explain myself correctly. The popPK parameters I had provided as reference (CL, CL/F, Vss and Vss/F), come from a popPK model which is currently being developed in our lab with the same pediatric data I am using to simulate in PKSim: CHILDREN IV

PO

The typical F from the popPK is 0.733. To obtain the reference value for CL, CL/F, Vss and Vss/F, I had used the popPK model and I have calculated the parameters including the model covariates (Sex, Weight, age, Kidney function, prediction of Fat-Free Mass..) from our pediatric population. The covariates values were different in the case of PO and IV groups of children. As a consequence, we cannot compare the parameters of CL and CL/F because they are based on different population, but we can compare them with the CL or CL/F from PKSim as the simulated population was the same.

Using CL as the reference in children, the bioavailability as predicted from PK-Sim is 65%. If the model is built on adults (F = 61% in the adult model as reported from your numbers) and the drug is about 35% hepatically cleared (low extraction ratio of about 0.2) in adults (biliary is fully mature and 1A2 is not), 65% in children is a reasonable prediction. Considering that cipro is 2/3rds renally cleared in adults (more in kids), it would be suspect to predict 94% in kids.

The Vss/F is wonky. I recreated a cipro model (in general) and I can't recreate such as huge Vss difference between IV and oral.

How did you calculate the Vss/F? because I understood that it is not possible to obtain the parameters after a population simulation directly from PKsim (https://github.com/Open-Systems-Pharmacology/PK-Sim/issues/354 ).

On a side note, the V is set in PK-Sim mechanistically meaning that there is one V that is derived for each compound, regardless of the administration. The V that you are looking at in the PK-Analysis window is derived as one would derive it from a concentration vs. time curve. Thus the V from the time curve is dependent on the how long the simulation goes for. Always make sure that you you are finishing the simulation at the same time as the observed data!

To calculate the parameters I have simulated only one administration (either for IV or PO) for 24 h. Later I have obtained the .csv (export PK analysis to csv) from the simulation and I have calculated with NPA analysis (PKSolver) the parameters, extrapolating the AUC to inf.

For cipro, this makes a big difference because oral uptake occurs along the length of the intestines and, as such, 12 hours vs. 72 hours changes the dose arriving systemically (changing all PK parameters).

I have a few questions before I can figure out what's going on:-)

How old are the kids in the trial? By the numbers I'm guessing in the 2-6 yr range?

The children's age range is: 0.5-13.4 for IV administration and 0.3-15.3 for PO administration.

Did you create one virtual child for every real child and predict the profile and compare to observed levels? I think so, just checking.

Yes, it is exactly what I did. I had created a trial of 200 ID per real children (with the same age, weight, height, fu and GFR) and after running the simulation, I´ve compared the simulation to the observed levels (the parameters were calculated only using the media from the 200 id simulation).

How did you incorporate the GFR for each kid? If you could give me an example, that would be helpful. As a check, you could just leave the GFR as estimated for the age of the child and see if the under-prediction in concentrations remains.

For each ID: I have created a individual “individual_GFR” (with my ID height, weight and age) specifying its GFR (Anatomy & Physiology→ Physiology→ GFR specific). Afterwards, I have created a population with the same anthropometric characteristics and I choose as reference ID the “individual_GFR”. image (The GFR has been calculated using the Schwartz formula non normalized by BSA, as we have the serum creatinine: image

How did you incorporate tubular secretion ontogeny?

I had not incorporated it. How could it be done?

Thank you very much for your complete reply and for taking your time in understanding my issue. Best,
Violeta

Aedginto commented 6 years ago

Hi again!

l completely agree with you, but I did not explain myself correctly. The popPK parameters I had provided as reference (CL, CL/F, Vss and Vss/F), come from a popPK model which is currently being developed in our lab with the same pediatric data I am using to simulate in PKSim: CHILDREN IV Vss popPK model = 34.60 L REFERENCE CL popPK model =9.72 L/h REFERENCE PO Vss/F popPK model = 35.47 L REFERENCE CL/F popPK model =10.32 L/h REFERENCE The typical F from the popPK is 0.733. To obtain the reference value for CL, CL/F, Vss and Vss/F, I had used the popPK model and I have calculated the parameters including the model covariates (Sex, Weight, age, Kidney function, prediction of Fat-Free Mass..) from our pediatric population. The covariates values were different in the case of PO and IV groups of children. As a consequence, we cannot compare the parameters of CL and CL/F because they are based on different population, but we can compare them with the CL or CL/F from PKSim as the simulated population was the same.

For the popPK modeling, I would have combined the IV and oral data as opposed to making a model for each administration; although I am unsure of the model purpose.

How did you calculate the Vss/F? because I understood that it is not possible to obtain the parameters after a population simulation directly from PKsim (Open-Systems-Pharmacology/PK-Sim#354 ).

I made an individual and looked at V after IV and V/F after oral. The population mean is going to be in the same ballpark as the mean individual. You could also export the population analysis to .csv (right click on simulation and select "Export PK-Analysis to .csv") and look at the values for V/F in there for each individual if you want. Vss is V/F. F is one for IV. There is no need to calculate V yourself. If you want to get a better V estimate, simulate longer than 24 hours for both IV and oral (at least 72 hours).

Yes, it is exactly what I did. I had created a trial of 200 ID per real children (with the same age, weight, height, fu and GFR) and after running the simulation, I´ve compared the simulation to the observed levels (the parameters were calculated only using the media from the 200 id simulation).

OK, interesting. What is the goal of the model creation here? That will drive what evaluation metrics you use. If it is the question: Does the PBPK model accurately simulate a pediatric population? Create a pediatric population that looks like your population and check to see if 90% of the observed data points are within the 90th prediction interval. You can also change your question a bit to ask: Does the PBPK model accurately simulate a pediatric population between 0-2 yrs, 2-6 yrs, 6-12 years and 12-16 yrs? Here you just do the same pop generation as above but in the time profile analysis section, parse your ages. Making 200 kids of exactly the same age/weight...etc as the observed child is not answering the above questions.

Are these kids renally impaired? If not, I would not alter the GFR.

Also, tubular secretion. Cipro undergoes tubular secretion which would need to be added to the adult model (make sure you get the right fraction excreted unchanged in urine in the adult model before moving onto kids. This will necessitate a GFR fraction of 1 plus tubular secretion.). To add ontogeny to tubular secretion, you need to create a transport process. This is an influx transporter on the apical side of the kidney (do this in your individual and create your population from this individual). In the adult, the value of the clearance due to the kidney needs to add up to fraction excreted unchanged in the urine. If one is just describing adult data, you can use a 'Renal Plasma Clearance' but if you are going to extrapolate, this is insufficient and you need to mechanistically model renal clearance. Tubular secretion ontogeny can be added (it is not automatically in PK-Sim as it is transporter dependent....and most people don't know which transporter is involved for their compound!). I use the Hayton ontogeny function (Hayton 2002. Maturation and Growth of Renal Function: Dosing Renally Cleared Drugs in Children) and you can add this where you add the kidney transporter. Once this is settled, when you create a child, the GFR part will be extrapolated using the Rhodin et al function (default in PK-Sim) and the Hayton function for the tubular secretion part.

All the best! Andrea

Aedginto commented 6 years ago

Sorry...I meant efflux transporter on the apical side of the kidney.

Aedginto commented 6 years ago

FYI: Violeta and I have gone 'off-line' on this project as it involves much more than what can be discussed on a forum. We'll update you later! Andrea

vbalbas commented 6 years ago

Yes indeed! We will update you and we will share the project! Thanks Andrea

msevestre commented 6 years ago

@vbalbas I am going to close this issue for now. Please don't hesitate to create a new issue with your updates when you are ready

vbalbas commented 5 years ago

Dear all,

I would like to update this entry! The project was finished and published: "Physiologically-Based Pharmacokinetic model for Ciprofloxacin in children with complicated Urinary Tract Infection" https://doi.org/10.1016/j.ejps.2018.11.033

A summary of the main problem we had encountered when developing the model was that we realized that the children from our study were not behaving as healthy children. Therefore, the V and CL differed from a healthy paediatric population. As we did not have data for healthy children, we proved our hypothesis by generating in silico data (with NONMEM® v.7.4 software) with a population pharmacokinetic model built using data from different clinical trials with children without renal impairment(1). Once our model was validated with the simulated data, it was then ready to be extrapolated into a renally impaired paediatric population.

Thank you very much to the people from this community and specially to @Aedginto for all the provided help and teaching in PBPK modelling and PK-Sim.

(1) P. Rajagopalan, M.R. Gastonguay. Population pharmacokinetics of ciprofloxacin in pediatric patients J. Clin. Pharmacol., 43 (2003), pp. 698-710

StephanSchaller commented 5 years ago

Thanks for the update and congratulations!

JanSchlender commented 5 years ago

Congrats to you and the other authors! Great effort