DosOpt : Bayesian in-treatment vancomycin dose adjustment of neonates



Input data

Sample ID

Covariate entry

Paste concentration and time data input

Dose episode parameters


Data view
Model training
Model diagnostics
Dose episode visualisation
Dose optimisation
Report download
Legal notices
Sample ID

DosOpt is a therapeutic drug monitoring based (TDM) dose optimisation tool. It combines published pharmacokinetic models with patient data by applying Bayesian methods.

DosOpt can
  • estimate patient pharmacokinetics
  • predict individual drug concentration time-courses in custom treatment scenarios
  • recommend doses to patients undergoing active treatment and before first dose

DosOpt is implemented in R and applies libraries such as shiny, useshinyjs, ggmcmc, R2jags, ggplot2 and others. Reports are compiled using knitr and rmarkdown. Models are fitted using JAGS - a Gibbs sampler for Markov Chain Monte Carlo (MCMC) parameter estimation. The tool has been developed in collaboration of BIIT research group and Institute of Microbiology of UT. Bug reports and feature requests can be submitted to ttasa [at]

Concentration measurements

Dosage regimen


Population model parameters

Individual model parameters

Goodness of model fit

Histograms of parameter values

Density plots of MCMC chains

Parameter traceplots

Autocorrelation plots

Potential Scale Reduction Factor

Geweke diagnostics

Simulated concentration episode

Download analysis report



A pharmacokinetic analysis in DosOpt consists of a number of sequential operations. A standard analysis would follow DosOpt line of tabs from left to right. At a minimum an informative analysis requires the selection of a population PK model from a list of included models and user uploading patient treatment data. As of Oct. 2018 DosOpt includes 12 vancomycin PK models for neonates. Full analysis consists of uploading data, fitting a Bayesian model, checking its posterior fit and diagnostic parameters. Then, individual kinetics estimates would be used in time-concentration curve simulations under user-defined conditions. Dose recommendations rely on optimised doses combined from in-treatment concentrations and PK priors against user defined pharmacodynamic (PD) target control values. All results can be downloaded in a compiled PDF formatted report. For legal notices and disclaimer please see section Legal notices.

Data view

Library of pharmacokinetic models

DosOpt currently includes two PK models to be used as priors in estimating individual kinetics.
List of available models in the interface are listed in 'Model selection'. Model named 'Test model' serves illustrative and testing purposes and has no clinical basis. Test model itself and values in example data are completely arbitrary. Validated models adapted from the academic literature and their characteristics can be found in the listing below.

Allegaert et al. [1]
Anderson et al. [2]
Bhongsatiern et al. [3]
De Cock et al. [4]
Frymoyer et al. [5]
Grimsley and Thomson [6]
Kimura et al. [7]
Lo et al. [8]
Marques-Minana et al. [9]
Oudin et al. [10]
Seay et al. [11]
Zhao et al. [12]

Data input

Data can be included via 4 different input methods. In case PK models include covariates, these would always be provided through the interface. This enables the upload format staying the same regardless of PK model specifications.
  • Example data - an example dataset for the selected model.
  • Upload data - user uploads a csv formatted data file. Format is specified in section 'CSV data format'.
  • Paste data - user textually enters individual specific concentration and dose episodes. Format is specified in section 'Paste data' format
  • Prior estimation - no data upload needed. Only covariates need to be specified. Individual estimates are derived based on population parameters and individual covariates
Required fields

TIME - event time (minutes). Specified in minutes after first dose. First dose episode is expected to happen at 0.

EVID - type of event. 1 - dose episode, 0 - concentration measurement

CONC - concentration measurement (mg/L). Ignored for event types other than 0.

AMT - dose amount (mg). Ignored for event types other than 1.

DUR - duration of infusion (minuts). Ignored for event types other than 1.

ID - patient ID (Currently redundant!! Read ID in from SampleID textfield in Data View tab).

CSV data
Concentration and dose episodes are laid out in 6 columns separated by commas. Each row represents a single event.
Paste data
Concentration and dose episodes are laid out in 6 rows separated by commas.

Main data panel

Successful data insertion/upload displays concentration measurements and dose information in 3 separate columns. Covariates are entered through graphical interface. In cases of some observations not showing up in the panel in expected ways it is most recommended to check whether the uploaded data follows the required format.

Model training

'Train model with input data' starts Bayesian pharmacokinetic parameter estimation. Population estimates only apply population PK model estimates and covariate data. Individual estimates also include information about the on-going treatment. This includes both the serum concentrations and dosing times/amounts. Both sets along with confidence intervals show in the main panel. Only population parameters are displayed in 'Prior estimation' input mode.
Execution of 'Simulate population/individual posterior fits' creates posterior model goodness fit plots based on population and individual model estimate simulations. Bayesian model fitting produces numerous conditionally independent PK parameter sets. A full time-concentration curve is constructed in case of each parameter set. Medians and 95% empirical credible intervals of simulated concentration-time courses are visualised in the outputted posterior fit plots. Actual concentration measurements are visualised as separate points for visual validation. It is possible for the user to toggle between model types and interactively change the time-range for the figure.

Model diagnostics

Model diagnostic plots provide several visuals for the fitted Bayesian model convergence and its attributes. These are designed using R library 'ggmcmc'. Diagnostic plots include parameter distributions, traceplots, autocorrelations and some statistics on convergence. More detailed explanations can be found here .

Dose episode visualisation

This set of methods enables the user to visualise and extrapolate concentration changes under custom dosing scenarios. Users specify the following parameters
  • Time of next dose - start time of the next dose administration (hours after first dose).
  • Duration of infusion - length of constant rate drug administration (hours).
  • Length of dosing interval - endpoint of the next dose episode (hours).
  • Next dose amount - dose amount administered at next administration (mg).
'Simulate a dose episode' button executes the simulations and concludes with a visual of patient predicted concentration changes. The median estimates are surrounded by 95% confidence. These are constructed similarly to the posterior fit models described in 'Model training'. Checking 'Full plot' visualises the whole course of the treatment from the beginning. Otherwise, only concentration changes in time frame of the next dose administration is displayed.
These visuals can be selected for the selection in final report by ticking the 'Add simulation to report' box after executing the simulation procedure. Selected plots can later be found in the downloaded report.

Dose optimisation

This subsection aims to identify the best steady-state dose achieving a user-specified target associated with therapeutic success. Similarly to subsection 'Dose episode visualisation' the user specifies the start time of the next dose. Unlike before, the user provides a range of possible doses instead of a fixed amount and a list of between-dose intervals that are to be tested for optimal PTA. Best corresponding dose is searched from the given dose range. This excludes unsuitably high and low doses and restricts the potential search domain.

The user also needs to select the target control method against which optimisation is performed. These targets are mostly dependent on the drug and population cohort under investigation. We have provided suggested target controls and respective values for models in the library but it is up to the user to choose an appropriate one.

Implemented target control methods

Trough concentration - desired/accepted range of concentrations in the end of the dosing episode. User selects the range that the concentration should fall at their lowest level. Suggested dose is the one maximising the number of individual simulations within the specified range. Outputted percentage can be interpreted as a probability that an individual will achieve trough concentrations within this target range.

The results output the dose optimisation plot for the highest PTA as well as full table of PTAs and optimal doses for all test intervals. Result PTAs are accompanied with safety estimates for AUC24h<400 and AUC24h>700 with the best dose

AUC/MIC - 24h normalised area-under-the curve to be attained with a given probability. The user specifies the desired probability of target attainment (PTA) for reaching the patient AUC and minimal-inhibitory-concentration (MIC). The procedure optimises for the lowest dose that is closest to the specified PTA. The outputted number can be interpreted as the probability that the patient reaches the desired PTA.

Report download

User can download a single PDF formatted report that includes all individual elements of the analysis. From simulated dose episodes only those selected by checking the appropriate checkbox in 'Dose episode visualisation' would be included.


Bayesian statistics allow using existing knowledge about the domain in statistical analyses. Informative priors decrease the required amount of data for precise estimate inferral. Combination of prior knowledge and observed data with Bayesian computation returns a probabilistic posterior distribution. In this application, we use population PK models and patient concentration measurements with dosage info to derive individual patient kinetics. with JAGS: a Gibbs sampler for Markov Chain Monte Carlo (MCMC) parameter estimation. DosOpt aims to streamline the analysis process from model fitting to concrete dose suggestions.

Model references

[1] Allegaert K, Anderson BJ, van den Anker JN, Vanhaesebrouck S, de Zegher F. Renal drug clearance in preterm neonates: relation to prenatal growth. Ther Drug Monit. 2007 Jun;29(3):284???91. doi: 10.1097/FTD.0b013e31806db3f5

[2] Anderson BJ, Allegaert K, Van den Anker JN, Cossey V, Holford NHG. Vancomycin pharmacokinetics in preterm neonates and the prediction of adult clearance. Br J Clin Pharmacol. 2007 Jan;63(1):75???84. doi: 10.1111/j.1365-2125.2006.02725.x

[3] Bhongsatiern J, Stockmann C, Roberts JK, Yu T, Korgenski KE, Spigarelli MG, et al. Evaluation of Vancomycin Use in Late-Onset Neonatal Sepsis Using the Area Under the Concentration-Time Curve to the Minimum Inhibitory Concentration >400 Target. Ther Drug Monit. 2015 Dec;37(6):756-65. doi: 10.1097/FTD.0000000000000216.

[4] De Cock RFW, Allegaert K, Sherwin CMT, Nielsen EI, de Hoog M, van den Anker JN, et al. A neonatal amikacin covariate model can be used to predict ontogeny of other drugs eliminated through glomerular filtration in neonates. Pharm Res. 2014 Mar;31(3):754???67. doi: 10.1007/s11095-013-1197-y

[5] Frymoyer A, Hersh AL, El-Komy MH, Gaskari S, Su F, Drover DR, et al. Association between vancomycin trough concentration and area under the concentration-time curve in neonates. Antimicrob Agents Chemother. 2014 Nov;58(11):6454???61. doi: 10.1128/AAC.03620-14

[6] Grimsley C, Thomson A. Pharmacokinetics and dose requirements of vancomycin in neonates. Arch Dis Child Fetal Neonatal Ed. 1999 Nov;81(3):F221???7. doi: 10.1136/fn.81.3.F221

[7] Kimura T, Sunakawa K, Matsuura N, Kubo H, Shimada S, Yago K. Population Pharmacokinetics of Arbekacin, Vancomycin, and Panipenem in Neonates. Antimicrob Agents Chemother. 2004 Apr;48(4):1159???67. doi: 10.1128/AAC.48.4.1159-1167.2004

[8] Lo Y-L, van Hasselt JGC, Heng S-C, Lim C-T, Lee T-C, Charles BG. Population pharmacokinetics of vancomycin in premature Malaysian neonates: identification of predictors for dosing determination. Antimicrob Agents Chemother. 2010 Jun;54(6):2626???32. doi: 10.1128/AAC.01370-09

[9] Marques-Minana M-R, Saadeddin A, Peris J-E. Population pharmacokinetic analysis of vancomycin in neonates. A new proposal of initial dosage guideline. Br J Clin Pharmacol. 2010 Nov;70(5):713???20. doi: 10.1111/j.1365-2125.2010.03736.x

[10] Oudin C, Vialet R, Boulamery A, Martin C, Simon N. Vancomycin prescription in neonates and young infants: toward a simplified dosage. Arch Dis Child Fetal Neonatal Ed. 2011 Sep;96(5):F365-370. doi: 10.1136/adc.2010.196402

[11] Seay RE, Brundage RC, Jensen PD, Schilling CG, Edgren BE. Population pharmacokinetics of vancomycin in neonates. Clin Pharmacol Ther. 1994 Aug;56(2):169???75. doi: 10.1038/clpt.1994.120

[12] Zhao W, Lopez E, Biran V, Durrmeyer X, Fakhoury M, Jacqz-Aigrain E. Vancomycin continuous infusion in neonates: dosing optimisation and therapeutic drug monitoring. Arch Dis Child. 2013 Jun;98(6):449???53. doi: 10.1136/archdischild-2012-302765

Copyright: 2015-2018 T. Tasa. All Rights Reserved.


DosOpt has been created for personal use only. The use of any result generated by DosOpt is in any case the sole risk and responsibility of the DosOpt user. Therapeutic decision should not solely rely on DosOpt as information provided by DosOpt does not replace clinical judgement. Although DosOpt has been validated carefully, there is no guarantee for the accuracy of the provided results. When using DosOpt you automatically agree with this disclaimer and the legal notices.


The author reserves the right not to be responsible for the topicality, correctness, completeness or quality of the information provided. Liability claims regarding damage caused by the use of any information provided, including any kind of information which is incomplete or incorrect, will therefore be rejected. All offers are not-binding and without obligation. Parts of the pages or the complete publication including all offers and information might be extended, changed or partly or completely deleted by the author without separate announcement.

Referrals and links

The author is not responsible for any contents linked or referred to from his pages - unless he has full knowledge of illegal contents and would be able to prevent the visitors of his site fromviewing those pages. If any damage occurs by the use of information presented there, only the author of the respective pages might be liable, not the one who has linked to these pages. Furthermore the author is not liable for any postings or messages published by users of discussion boards, guestbooks or mailinglists provided on his page.


The author reserves the right not to be responsible for the topicality, correctness, completeness or quality of the information provided. Liability claims regarding damage caused by the use of any information provided, including any kind of information which is incomplete or incorrect, will therefore be rejected. All offers are not-binding and without obligation. Parts of the pages or the complete publication including all offers and information might be extended, changed or partly or completely deleted by the author without separate announcement.

Privacy policy

If the opportunity for the input of personal or business data (email addresses, name, addresses) is given, the input of these data takes place voluntarily. The use and payment of all offered services are permitted - if and so far technically possible and reasonable - without specification of any personal data or under specification of anonymised data or an alias. The use of published postal addresses, telephone or fax numbers and email addresses for marketing purposes is prohibited, offenders sending unwanted spam messages will be punished.

Legal validity of this disclaimer

This disclaimer is to be regarded as part of the internet publication which you were referred from. If sections or individual terms of this statement are not legal or correct, the content or validity of the other parts remain uninfluenced by this fact. The text of this disclaimer is based on the services of

Version history

28th Oct 2018 - Interval optimization. Steady state optimisation. Interface flexibility. Minor bug fixes. Safety evaluations.

04th Apr 2018 - 11 new PK models, faster load-up, improved report

11th Jan 2017 - Episode visualisations/optimisations 5x faster

06th Oct 2016 - Help page updated. Units and report output standardised. Optimisation output fixed.

20th May 2016 - Public link created. AUC/MIC target calculation modified.

30th Apr 2016 - Model diagnostics improved. Prediction errors incorporated in probability calculations.

08th Jan 2016 - Help page added. Visualisations in ggplot2.

30th Dec 2015 - Core version available for testing.