Bayesian forecasting for intravenous tobramycin dosing in adults with Cystic Fibrosis using one versus two serum concentrations in a dosing interval ==================================================================================================================================================== * Philip G. Drennan * Yann Thoma * Lucinda Barry * Johan Matthey * Sheila Sivam * Sebastiaan J. van Hal ## Abstract **Background** Intravenous tobramycin requires therapeutic drug monitoring (TDM) to ensure safety and efficacy when used for prolonged treatment, as in infective exacerbations of Cystic Fibrosis (CF). The 24 hour area under the concentration time curve (AUC24) is widely used to guide dosing, however there remains variability in practice around methods for its estimation. **Objectives** To determine the potential for a sparse sampling strategy using a single post-infusion tobramycin concentration and Bayesian forecasting, to assess the AUC24 in routine practice. **Methods** Adults with CF receiving once daily tobramycin had paired concentrations measured 2 hours (c1) and 6 hours (c2) following end of infusion as routine monitoring. We estimated AUC24 exposures using Tucuxi, a Bayesian forecasting application incorporating a validated population pharmacokinetic model. We performed simulations to estimate AUC24 using the full dataset using c1 and c2, compared to estimates using depleted datasets (c1 or c2 only), with and without concentration data from earlier in the course. We assessed agreement between each simulation condition and the reference graphically, and numerically using median difference (Δ) AUC24, and (relative) root mean square error (rRMSE) as measures of bias and accuracy respectively. **Results** 55 patients contributed 512 concentrations from 95 tobramycin courses and 256 TDM episodes. Single concentration methods performed well, with median ΔAUC24 <2 mg.h.l-1 and rRMSE of <15% for sequential c1 and c2 conditions. **Conclusions** Bayesian forecasting, using single post-infusion concentrations taken 2-6 hours following tobramycin administration can adequately estimate true exposure in this patient group and are suitable for routine TDM practice. **Key Points** * - In stable adult patients with Cystic fibrosis without significant renal impairment, Bayesian forecasting allows accurate estimation of tobramycin AUC24 using a single blood sample taken 2-6 hours post-infusion with acceptable accuracy, especially when including prior measured concentrations. * - A single sample approach with Bayesian forecasting is logistically less complicated than a two-sample approach, and could facilitate best-practice TDM in the outpatient setting. * - A more intensive sampling strategy with Bayesian forecasting using two tobramycin concentrations in a dosing interval should be considered in unstable patients, or where observed concentrations deviate significantly from model predictions. ## 1. Introduction Life expectancy for patients with Cystic Fibrosis (CF) continues to improve, owing to a range of strategies to maintain lung function, including aggressive treatment of infective pulmonary exacerbations due to *Pseudomonas aeruginosa*.(1) Intravenous tobramycin is commonly prescribed for this purpose.(2) Well-known toxicities of tobramycin include nephrotoxicity and ototoxicity, and there is increasing interest in minimising these long-term sequelae, which are at least partly attributable to cumulative lifetime exposure.(2–4) Once-daily tobramycin dosing is increasingly becoming standard practice in many CF centres, however centres differ with regard to therapeutic drug monitoring (TDM) practices, which range from simple trough concentration and nomogram methods, to more direct monitoring of the 24 hour area under the concentration time curve (AUC24).(5–7) At our centre (Royal Prince Alfred Hospital, Sydney, Australia), we perform AUC24 based monitoring (target AUC24 100 mg.h.l-1) using a log-linear regression (LLR) method, which requires two-concentrations within a dosing interval and simple pharmacokinetic calculations coded into a Microsoft Excel spreadsheet.(8) Although simple and accurate, significant attention is required to coordinate correct application of the TDM protocol, including obtaining two blood samples at the appropriate times, and ensuring the dosing and monitoring information is recorded accurately.(9,10) Thereafter, clinicians performing the AUC24 estimation must perform the required calculations in a timely manner in order to return dosing advice prior to the next dose. There is increasing interest in the ambulatory management of pulmonary exacerbations, however it is difficult and impractical to collect two post-infusion blood samples in the outpatient setting. A simple, accurate method of AUC24 estimation to inform ongoing dosing using a single blood sample may therefore greatly increase the efficiency of this TDM practice, broaden its utility to outpatient settings, and improve patient acceptability. Bayesian forecasting, which incorporates information from population pharmacokinetic models with one or two measured post-infusion concentrations has been shown to accurately characterise tobramycin pharmacokinetics in adults with CF, compared to results derived from intensive sampling.(11,12) In a small study of 12 adults receiving once-daily tobramycin, Bayesian methods resulted in less biased and more precise AUC24 estimation than a LLR method, and one- and two-sample Bayesian strategies were similarly accurate.(11) The authors of this study considered all three methods potentially suitable for routine TDM. The aim of the current study was to examine the potential generalisability of this finding by studying a larger group of adult patients, representative of those typically managed in CF centres. We sought to determine the agreement in AUC24 estimation between a one-sample and a two-sample approach using Bayesian forecasting, in order to determine the potential suitability of a one-sample Bayesian forecasting approach for routine TDM in this group of patients. ## 2. Methods ### 2.1. Patients and data We collected data retrospectively from adults (>18 years) with CF who received once-daily intravenous tobramycin for pulmonary exacerbations, as part of routine practice between January and December 2018. Clinical staff were instructed to take blood samples for the measurement of tobramycin concentrations at two time points (hereafter designated t1 and t2) at approximately 2 hours (c1) and 6-8 hours (c2) following the end of the infusion (0.5 hours). Dosing and drug concentration information, in addition to a patient’s height, weight, and the most recent creatinine measurement, was submitted to the TDM monitoring service for estimation of pharmacokinetic parameters and subsequent dosing advice, in order to achieve a target AUC24 of 100mg/L (acceptable range 80-120mg.h.l-1). In routine practice our TDM service currently uses a log linear regression method for AUC24 estimation, with proportional dose adjustment for subsequent dose recommendations.(8) We recruited all patients with CF contributing at least one pair of tobramycin concentrations, which constituted a single TDM episode. We excluded episodes where c2 was below the lower limit of quantification, and where any of the concentration measurements was considered unreliable based on clinician assessment at the time of the TDM episode—for example where c1 was higher than c2. These cases were identified by the clinician performing the TDM, and prompted a request to repeat the measurements the following day. The study was approved by the Sydney Local Health District Ethics Committee (ref X19-0168) with a waiver of informed consent due to retrospective collection of routinely collected data. ### 2.2. Tobramycin assay Plasma tobramycin concentrations were determined using immune spectrophotometric assay on a Roche/Hitachi cobas c system platform with LLOQ of 0.33 mg l-1 and intra- and inter-day assay coefficients of variation <10%. ### 2.3. Simulation of sampling regimens We simulated the potential effect of different sparse-sampling strategies on PK parameter and AUC24 estimates using the Tucuxi Bayesian forecasting software ([www.tucuxi.ch](http://www.tucuxi.ch)), using a previously validated population pharmacokinetic model for adults with CF by Hennig et al.(12–14) Patient age, height, weight, and serum creatinine were covariates for the underlying population pharmacokinetic model used to generate the *a priori* pharmacokinetic parameter estimates. We used the full set of sequential concentration pairs (Sc1c2) available at the time of each TDM episode as the reference (‘REF’), and compared this to one of five simulated sampling conditions (‘SIM’) which incorporated less information: 1. c1 and c2 without sequential prior concentration measurements from the same TDM episode (*NS**c1c2*), 2. c1 measurements only, using sequential information from previous c1 measurements (*S**c1*); c) c1 measurements only, without prior sequential measurements (NSc1); d) c2 measurements only, with sequential measurements (Sc2; e) c2 measurements only, without sequential measurements (NSc2). We used customised scripts to perform the simulations using the Tucuxi engine, to reduce the risk of errors and to ensure reproducibility of the results. ### 2.4. Statistical analysis and measures of agreement We performed statistical analysis using R (version 3.3.2, R Foundation for Statistical Computing, Vienna, Austria) implemented in the RStudio environment (version 1.0.136, RStudio Team (2016), RStudio, Inc., Boston, MA). As a measure of bias we estimated agreement of the PK parameters, and AUC24 between the reference and each of the simulated conditions for each TDM episode. The main parameter of interest was the estimated AUC24, calculated from the estimated clearance and the administered dose according to equation (1): ![Formula][1] We calculated absolute differences in AUC24 estimates between each of the simulation conditions and the reference method according to equation (2): ![Formula][2] The relative difference in exposure estimates was calculated as per equation (3): ![Formula][3] Where ![Graphic][4] denotes the estimated AUC24 calculated using data from simulation m (*NS**c1c2*, NSc1, Sc1, NSc2, Sc2) for the *i*th individual on the *j*th TDM episode, and ![Graphic][5] denotes the estimated AUC24 calculated using the all prior concentration data available for individual *i* at the time of the *j*th TDM episode. For the purposes of this study separate admissions were treated as independent cases. As a measure of accuracy of the simulated method with the reference we calculated the root mean square error (RMSE, *mg. h. l*−1) and the relative RMSE (rRMSE, %) according to equations (4), and (5): ![Formula][6] ![Formula][7] Where n is the total number of TDM episodes. We explored agreement of simulated conditions with the reference overall, and according to number of previous TDM episodes graphically, using boxplots and relative difference plots.(15) ## Results ### 2.5. Patients and data Between January 1st and December 21st 2018, 54 patients were admitted on 95 occasions for an infective exacerbation of their CF and received a course of tobramycin as part of their antimicrobial therapy. There were 256 TDM episodes (range 1-5 per course) and 512 concentrations. The median age was 31 years (IQR 22-40.25), and 23 (43%) patients were male. Tobramycin was given for a median 13 days (IQR 11-14). Demographics are summarised in table 1. Dosing and monitoring data, and pharmacokinetic parameter estimates for each TDM episode are shown in table 2 and figure 1. The median dose of tobramycin was 8.6 mg.kg-1 (IQR 7.1-10). Within a dosing interval, c1 was measured at a median of 2 h (IQR 2 – 2.6), and c2 at a median of 6.2 h (IQR 6-6.7). Across all TDM episodes the median estimated AUC24 using the Bayesian forecasting approach with all available data was 94.2 mg.h.L-1 (IQR 75.3-112.9). View this table: [Table 1:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/T1) Table 1: Baseline demographics and admissions View this table: [Table 2:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/T2) Table 2: Tobramycin concentrations and parameter estimates for each TDM episode (n=256) ![Figure 1:](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/01/04/2020.12.31.20249095/F1.medium.gif) [Figure 1:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/F1) Figure 1: Time versus tobramycin concentration (n=512) ### 2.6. Comparison of simulation methods Figure 2 shows the relative difference plot of ΔAUC24 (%) for each of the simulation conditions. There was no clear trend in agreement according to estimated AUC24. Table 3 and figure 3 show the estimated tobramycin exposure for the reference condition based on the full data set (Sc1c2) and each of the simulation conditions, and differences between the methods, pooled across all TDM episodes (n=256). AUC24 estimates were on average modestly greater than the reference condition, with median ΔAUC24 <5% compared to Sc1c2 for all conditions except NSc1. rRMSE of all methods was <15% of the reference AUC24. Large differences in estimated exposure, defined arbitrarily as ΔAUC24>30% were rare in all conditions but occurred least frequently in the Sc1 condition (1.6% of episodes). Table 4 and figure 4 show comparisons of NSc1c2, Sc1, and Sc2 with the reference condition, stratified by TDM episode. As expected for NSc1c2, both ΔAUC24 and rRMSE increased for subsequent TDM episodes, as this condition only considered the most recent available data. For Sc1 and Sc2, rRMSE improved modestly for second and subsequent TDM episodes, and was <15% for all TDM episodes View this table: [Table 3:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/T3) Table 3: Comparison of simulation conditions View this table: [Table 4:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/T4) Table 4: Comparison of simulation conditions according to number of preceding TDM episodes ![Figure 2:](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/01/04/2020.12.31.20249095/F2.medium.gif) [Figure 2:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/F2) Figure 2: Relative difference plot of mean estimated AUC24 versus difference in AUC24 (%) for the reference method (full dataset) versus each simulated condition, with median difference and 2.5th and 97.5th percentiles (dashed lines) ![Figure 3:](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/01/04/2020.12.31.20249095/F3.medium.gif) [Figure 3:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/F3) Figure 3: Distribution of differences in estimated AUC24 (%) for full data set (reference condition) versus each simulation condition ![Figure 4:](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/01/04/2020.12.31.20249095/F4.medium.gif) [Figure 4:](http://medrxiv.org/content/early/2021/01/04/2020.12.31.20249095/F4) Figure 4: Distribution of differences in estimated AUC24 for full dataset (reference condition) versus selected simulation conditions according to TDM episode ## Discussion Our study suggests that in stable adult patients with CF pulmonary exacerbations, the tobramycin AUC24 can be reliably estimated using Bayesian forecasting and a single post infusion concentration. In a small proportion of cases the estimates differed more significantly. The reasons for these outliers are unclear but could relate to erroneous information (e.g. time of dose or blood sample recorded incorrectly), or physiological instability. This observations illustrates the importance of attention to situations where observed concentrations are significantly different from those predicted by the model, in order to apply clinical judgement in when unexpected results occur. The graphical interface provided by modern Bayesian forecasting software allows a visual representation of this discrepancy, which may improve the identification of unusual results. Where observed concentrations deviate substantially from those predicted by the model, patients may benefit from repeat or more intensive sampling to confirm observed concentrations and to ensure an accurate estimation of their pharmacokinetic parameters. This approach would be particularly applicable to critically unwell patients, given known physiological fluctuations in these patients where a single sample approach may be less appropriate.(16–19). Similarly, in stable patients, significantly discrepant concentrations from those expected either relative to previous estimates or predicted on the population pharmacokinetics should prompt careful consideration of a pre-analytical source of error.(20) Although a more limited sampling approach resulted in less accurate pharmacokinetic parameter estimation relative to the reference, the effect was usually small (rRMSE <15%). Although there is no established standard for acceptable accuracy in this context, it is of comparable magnitude to accepted target ranges used in clinical practice.(12) In specific situations, e.g. those at higher risk of renal dysfunction, or unstable patients with sepsis or requiring ICU admission, a more stringent approach may be considered appropriate. In these situations, clinicians may choose to monitor the patient more closely, e.g. by continuing with two-concentration monitoring until the patient is judged stable, and then switching to a single concentration approach, e.g. to reduce phlebotomy or line access, or to facilitate ambulatory management. Accuracy and bias of estimates were similar whether using a sample taken approximately 2 hours after the end of the infusion or approximately 6 hours after the end of infusion. These results largely agree with the results of Gao et al, who found that a single concentration taken 70 – 640 minutes yielded sufficiently accurate results.(10) For practical purposes a timed blood sample taken any time between 2-6 hours post end of infusion is likely to be sufficient for routine monitoring. Numerous software applications for performing Bayesian forecasting are now available, however have not yet been widely-adopted, possibly owing to access to and cost of specialist software, required expertise, and the switching cost of new TDM workflows.(7,21–23) The simplicity of a single-sample Bayesian forecasting method may therefore help to offset some of these barriers for adoption. Patients with CF have generally stable tobramycin pharmacokinetics over time, and as such may be considered ideal candidates for Bayesian forecasting methods, which allow accumulation of data for an individual patient over time.(24) Patients with CF are frequently managed in CF centres and may have multiple admissions to the same institution, thus in theory clinicians managing these patients should be able to derive an accurate individualised dosing strategy for each patient. Bayesian forecasting software which allows automated population of results from an electronic medical record may therefore be of great benefit. ## 3. Conclusions Bayesian forecasting, using a single post infusion sample between 2-6 hours post infusion and prior concentration measurements can precisely and accurately estimate tobramycin exposure. This simple approach increases the flexibility of a Bayesian TDM workflow, and may facilitate efficient outpatient monitoring for ambulatory therapy. ## Supporting information STROBE checklist [[supplements/249095_file02.docx]](pending:yes) ## Data Availability Data requests should be submitted to the corresponding author. Access to de-identified data stored on the synapse.org repository may be granted to suitably qualified researchers following review. ## Declarations ### Authors’ contributions PD conceived the study, assisted with data collection and analysed the data and drafted the manuscript. LB collected the data and reviewed the manuscript. YT and JM performed the simulations and reviewed the manuscript. SS supported development of the TDM service and data collection, and reviewed the manuscript. SVH conceived the study and reviewed the manuscript. ### Funding No sources of funding were used in the preparation of this article ### Conflicts of interest The authors have no conflicts of interest to declare ### Availability of data and material Data requests should be submitted to the corresponding author. Access to de-identified data stored on the synapse.org repository may be granted to suitably qualified researchers following review. ## Footnotes * Oxford University Hospitals NHS Foundation Trust, Oxford, UK * Received December 31, 2020. * Revision received December 31, 2020. * Accepted January 4, 2021. * © 2021, Posted by Cold Spring Harbor Laboratory This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/) ## 4. References 1. 1.Keogh RH, Szczesniak R, Taylor-Robinson D, Bilton D. Up-to-date and projected estimates of survival for people with cystic fibrosis using baseline characteristics: A longitudinal study using UK patient registry data. J Cyst Fibros Off J Eur Cyst Fibros Soc. 2018;17(2):218–27. 2. 2.Prayle A, Smyth AR. Aminoglycoside use in cystic fibrosis: therapeutic strategies and toxicity. Curr Opin Pulm Med. 2010 Nov;16(6):604–10. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1097/MCP.0b013e32833eebfd&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20814306&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 3. 3.Al-Aloul M, Miller H, Alapati S, Stockton PA, Ledson MJ, Walshaw MJ. Renal impairment in cystic fibrosis patients due to repeated intravenous aminoglycoside use. Pediatr Pulmonol. 2005 Jan;39(1):15–20. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/ppul.20138&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=15521084&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000225722200003&link_type=ISI) 4. 4.Garinis AC, Cross CP, Srikanth P, Carroll K, Feeney MP, Keefe DH, et al. The cumulative effects of intravenous antibiotic treatments on hearing in patients with cystic fibrosis. J Cyst Fibros. 2017 May 1;16(3):401–9. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jcf.2017.01.006&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 5. 5.Prescott WA. A survey of extended-interval aminoglycoside dosing practices in United States adult cystic fibrosis programs. Respir Care. 2014 Sep;59(9):1353–9. [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6ODoicmVzcGNhcmUiO3M6NToicmVzaWQiO3M6OToiNTkvOS8xMzUzIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjEvMDEvMDQvMjAyMC4xMi4zMS4yMDI0OTA5NS5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 6. 6.Soulsby N, Bell S, Greville H, Doecke C. Intravenous aminoglycoside usage and monitoring of patients with cystic fibrosis in Australia. What’s new? Intern Med J. 2009 Aug;39(8):527–31. [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19220547&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 7. 7.Paviour S, Hennig S, Staatz CE. Usage and monitoring of intravenous tobramycin in cystic fibrosis in Australia and the UK. J Pharm Pract Res. 46(1):15–21. 8. 8.Begg EJ, Barclay ML, Duffull SB. A suggested approach to once-daily aminoglycoside dosing. Br J Clin Pharmacol. 1995 Jun 1;39(6):605–9. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/j.1365-2125.1995.tb05719.x&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=7654477&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=A1995RC78600004&link_type=ISI) 9. 9.Nezic L, Derungs A, Bruggisser M, Tschudin-Sutter S, Krähenbühl S, Haschke M. Therapeutic drug monitoring of once daily aminoglycoside dosing: comparison of two methods and investigation of the optimal blood sampling strategy. Eur J Clin Pharmacol. 2014 Jul;70(7):829–37. 10. 10.Gao Y, Hennig S, Barras M. Monitoring of Tobramycin Exposure: What is the Best Estimation Method and Sampling Time for Clinical Practice? Clin Pharmacokinet. 2018 Aug 24; 11. 11.Barras MA, Serisier D, Hennig S, Jess K, Norris RLG. Bayesian Estimation of Tobramycin Exposure in Patients with Cystic Fibrosis. Antimicrob Agents Chemother. 2016 Oct 21;60(11):6698–702. [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MzoiYWFjIjtzOjU6InJlc2lkIjtzOjEwOiI2MC8xMS82Njk4IjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjEvMDEvMDQvMjAyMC4xMi4zMS4yMDI0OTA5NS5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 12. 12.Burgard M, Sandaradura I, van Hal SJ, Stacey S, Hennig S. Evaluation of Tobramycin Exposure Predictions in Three Bayesian Forecasting Programmes Compared with Current Clinical Practice in Children and Adults with Cystic Fibrosis. Clin Pharmacokinet. 2017 Nov 13; 13. 13.Hennig S, Standing JF, Staatz CE, Thomson AH. Population pharmacokinetics of tobramycin in patients with and without cystic fibrosis. Clin Pharmacokinet. 2013 Apr;52(4):289–301. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1007/s40262-013-0036-y&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23420517&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 14. 14.Dubovitskaya A, Buclin T, Schumacher M, Aberer K, Thoma Y. TUCUXI: An Intelligent System for Personalized Medicine from Individualization of Treatments to Research Databases and Back. In ACM Press; 2017 [cited 2018 Aug 8]. p. 223–32. Available from: [http://dl.acm.org/citation.cfm?doid=3107411.3107439](http://dl.acm.org/citation.cfm?doid=3107411.3107439) 15. 15.Pollock MA, Jefferson SG, Kane JW, Lomax K, MacKinnon G, Winnard CB. Method comparison--a different approach. Ann Clin Biochem. 1992 Sep;29 (Pt 5):556–60. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1177/000456329202900512&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=1444169&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 16. 16.Reimann IR, Meier-Hellmann A, Traut T, Reinhart K, Hoffmann A. Monitoring of a single post-infusion blood sample to estimate the actual peak and trough concentration of tobramycin in critically ill patients. Exp Toxicol Pathol Off J Ges Toxikol Pathol. 2003 Jun;54(5–6):493–8. 17. 17.Denaro CP, Ravenscroft PJ. Comparison of Sawchuk-Zaske and Bayesian forecasting for aminoglycosides in seriously ill patients. Br J Clin Pharmacol. 1989 Jul;28(1):37–44. [PubMed](http://medrxiv.org/lookup/external-ref?access_num=2775614&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 18. 18.Rodvold KA, Blum RA. Predictive performance of Sawchuk-Zaske and Bayesian dosing methods for tobramycin. J Clin Pharmacol. 1987 Jun;27(5):419–24. [PubMed](http://medrxiv.org/lookup/external-ref?access_num=3693587&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=A1987L204800016&link_type=ISI) 19. 19.Conil J-M, Georges B, Ruiz S, Rival T, Seguin T, Cougot P, et al. Tobramycin disposition in ICU patients receiving a once daily regimen: population approach and dosage simulations. Br J Clin Pharmacol. 2011 Jan;71(1):61–71. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/j.1365-2125.2010.03793.x&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21143502&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 20. 20.Prescott WA, Mancuso MA. Clinical impact of laboratory error on therapeutic drug monitoring of once-daily tobramycin in cystic fibrosis: Case series. SAGE Open Med Case Rep. 2014;2:2050313X14521158. 21. 21.Drennan P, Doogue M, van Hal SJ, Chin P. Bayesian therapeutic drug monitoring software: past, present and future. Int J Pharmacokinet [Internet]. 2018 Dec 12 [cited 2018 Dec 13]; Available from: [https://www.future-science.com/doi/10.4155/ipk-2018-0005](https://www.future-science.com/doi/10.4155/ipk-2018-0005) 22. 22.Fuchs A, Csajka C, Thoma Y, Buclin T, Widmer N. Benchmarking therapeutic drug monitoring software: a review of available computer tools. Clin Pharmacokinet. 2013 Jan;52(1):9–22. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1007/s40262-012-0020-y&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23196713&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) 23. 23.Kumar AA, Burgard M, Stacey S, Sandaradura I, Lai T, Coorey C, et al. An evaluation of the user-friendliness of Bayesian forecasting programs in a clinical setting. Br J Clin Pharmacol. 2019;85(10):2436–41. 24. 24.Alghanem S, Paterson I, Touw DJ, Thomson AH. Influence of multiple courses of therapy on aminoglycoside clearance in adult patients with cystic fibrosis. J Antimicrob Chemother. 2013 Jun 1;68(6):1338–47. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/jac/dkt035&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23435691&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F01%2F04%2F2020.12.31.20249095.atom) [1]: /embed/graphic-1.gif [2]: /embed/graphic-2.gif [3]: /embed/graphic-3.gif [4]: /embed/inline-graphic-1.gif [5]: /embed/inline-graphic-2.gif [6]: /embed/graphic-4.gif [7]: /embed/graphic-5.gif