Median Prediction Error Research Articles

Corneal power estimation for intraocular lens power calculation after corneal laser refractive surgery in Chinese eyes

To develop and evaluate an algorithm for corneal power estimation in intraocular lens (IOL) power calculation after corneal laser refractive surgery in Chinese eyes. Guangdong Eye Institute, Guangdong Academy of Medical Sciences, Guangzhou, China. Prospective comparative case series. Corneal parameters in Chinese eyes and German eyes were measured using a rotating Scheimpflug camera. Corneal power was simplified as a corrective algorithm: K(C) = 1.114 × K(M) + K(2) (K(M) = measured K reading; K(2) = K(P) - K(A) × K(P) × CT/1.376; K(A) = anterior corneal power; K(P) = posterior corneal power; CCT = central corneal thickness). The variation and change in K(2) induced by refractive surgery were analyzed in Chinese eyes. The corrective algorithm was identified as K(C) = 1.114 × K(M) - 6.20. The method was evaluated in Chinese cataract cases after refractive surgery using the Haigis formula. No difference in anterior corneal radius (R(A)) or CCT between Chinese eyes and German eyes were found; however, the posterior corneal radius (R(P)), R(A)/R(P) ratio, keratometric index, and K(2) were different. The mean K(2) was -6.23 diopters (D) ± 0.24 (SD) in Chinese eyes and -6.12 ± 0.23 D in German eyes (P<.01). The mean change in K(2) induced by refractive surgery was -0.02 ± 0.06 D. The median absolute prediction error in IOL power calculation was 0.43 D (range 0.01 to 1.80 D). The algorithm was a relatively reliable method in IOL power calculation after corneal refractive surgery in Chinese eyes.

Design and implementation of a closed‐loop control system for infusion of propofol guided by bispectral index (BIS)

This study describes the design of a hypnosis closed-loop control system with propofol. The controller used a proportional-integral (PI) algorithm with the bispectral index (BIS) as the feedback signal. Our hypothesis was that a PI closed-loop control could be applied in clinical practice safely keeping the BIS within a pre-determined target range. The adjustment of the PI parameters was based on simulation. The procedure had three steps: obtaining a patient model using data from 12 patients, designing and adjusting the controller in simulation, and fine tuning the PI parameters in a pilot study (10 patients). The resulting controller was tested in 24 American Society of Anesthesiology (ASA) I-II patients. The controller directly decides the infusion rate of propofol, and no model is necessary in its online operation. The BIS target was set to 50. Remifentanil was used for analgesia. We evaluated the efficiency and safety of the automatic feedback system. It worked properly in all the patients. The median performance error was -1.62, and the median absolute performance error was 11.03. Average propofol-normalized consumption was 5.3 ± 1.8 mg/kg/h. Mean percentage of BIS in the range 40-60 was 83%. Mean time to open eyes was 8 ± 4 min. Time to extubation was 9 ± 5 min. Hemodynamic adverse event or intraoperative awareness were not recorded. The closed-loop system was able to maintain the BIS within an acceptable range of levels. The control of a propofol infusion guided by the BIS is feasible without hemodynamic instability in ASA I/II patients.

Real-Time Prediction of Waiting Time in the Emergency Department, Using Quantile Regression

Emergency department (ED) waiting times can affect patient satisfaction and quality of care. We develop and validate a model that predicts an individual patient's median and 95th percentile waiting time by using only data available at triage. From the existing ED information system, we extracted date and time of triage completion, start time of emergency physician consultation, and patient acuity category (1=most urgent, 3=least urgent). Quantile regression was applied for model development and parameter estimation by using visits from January 2011. We assessed absolute prediction error, defined as the median difference between the 50th percentile (median) predicted waiting time and actual waiting time, and the proportion of underestimated prediction, defined as the percentage of patients whose actual waiting time exceeded the 95th percentile prediction. The model was validated retrospectively with June 2010 data and prospectively with data from April to June 2011 after integration with the existing ED information system. The derivation set included 13,200 ED visits; 903 (6.8%) were patient acuity category 1, 5,530 (41.9%) were patient acuity category 2, and 6,767 (51.3%) were patient acuity category 3. The median and 95th percentile waiting times were 17 and 57 minutes for patient acuity category 2 and 21 and 89 minutes for patient acuity category 3, respectively. The final model used predictors of patient acuity category, patient queue sizes, and flow rates only. In the retrospective validation, 5.9% of patient acuity category 2 and 5.4% of category 3 waiting times were underestimated. The median absolute prediction error was 11.9 minutes (interquantile range [IQR] 5.9 to 22.1 minutes) for patient acuity category 2 and 15.7 minutes (IQR 7.5 to 30.1 minutes) for category 3. In prospective validation, 4.3% of patient acuity category 2 and 5.8% of category 3 waiting times were underestimated. The median absolute prediction error was 9.2 minutes (IQR 4.4 to 15.1 minutes) for patient acuity category 2 and 12.9 minutes (IQR 6.5 to 22.5 minutes) for category 3. Using only a few data elements available at triage, the model predicts individual patients' waiting time with good accuracy.

Evaluation of the predictive performance of a pharmacokinetic model for propofol in Japanese macaques (Macaca fuscata fuscata)

Propofol is a short-acting intravenous anesthetic used for induction/maintenance anesthesia. The objective of this study was to assess a population pharmacokinetic (PPK) model for Japanese macaques during a step-down infusion of propofol. Five male Japanese macaques were immobilized with ketamine (10 mg/kg) and atropine (0.02 mg/kg). A bolus dose of propofol (5 mg/kg) was administrated intravenously (360 mg/kg/h) followed by step-down infusion at 40 mg/kg/h for 10 min, 20 mg/kg/h for 10 min, and then 15 mg/kg/h for 100 min. Venous blood samples were repeatedly collected following the administration. The plasma concentration of propofol (Cp) was measured by high-speed LC-FL. PPK analyses were performed using NONMEM VII. Median absolute prediction error and median prediction error (MDPE), the indices of prediction inaccuracy and bias, respectively, were calculated, and PE - individual MDPE vs. time was depicted to show the variability of prediction errors. In addition, we developed another population pharmacokinetic model using previous and current datasets. The previous PK model achieved stable prediction of propofol Cp throughout the study period, although it underestimates Cp. The step-down infusion regimen described in this study would be feasible in macaques during noninvasive procedures.

External Validation of Aminoglycoside Models Used in Web Calculators and Clinical Decision Support Systems After Laboratory Conversion to Serum Creatinine Isotope Dilution Mass Spectrometry Assay

BackgroundModels to predict gentamicin t½ from serum creatinine (SCr) estimated creatinine clearance (CrCl) are currently being incorporated into smart-device applications and clinical decision support modules without external validation. ObjectiveThe aim of this study was to determine whether such models remain viable after conversion to isotope dilution mass spectrometry (IDMS) SCr assay. MethodsThis study analyzed data from retrospective reviews of the medical records of nonobese adults receiving the aminoglycoside gentamicin and having ≥2 evaluable serum gentamicin concentrations after laboratory IDMS SCr conversion, from January 2008 to August 2009, at a tertiary care hospital in Florida. A literature search found a number of cited aminoglycoside models. This group of models was classified as group 1. The World Wide Web was also searched for the term aminoglycoside dosing calculators, with 6 models found and referred to as group 2. Predictive performance measures were used to compare the model results with the t½ calculated from gentamicin concentrations using the Nelder-Mead algorithm. ResultsThe records of 39 patients met the inclusion criteria (23 men, 16 women; age range, 18–86 years; range of estimated CrCl, 55–115 mL/min) and provided the “gold standard” aminoglycoside t½. A gentamicin t½ was predicted from several published models (group 1) and from other models used in online smart-device applications (group 2) and clinical decision modules. The median (interquartile range) root mean square errors were 0.48 (0.44 to 0.65) and 0.48 (0.45 to 0.70) hours from group-1 and -2 models, respectively. The median mean relative prediction errors were 9% (−14% to +13%) and 11% (+1% to +21%) from groups 1 and 2. The median mean absolute prediction errors were 21% (19% to 28%) and 21% (20% to 30%) from groups 1 and 2. Adjusting SCr by +20% improved the predictive ability in 3 of 12 cited models and in 5 of 6 models used in applications. ConclusionsModels to predict gentamicin t½ should be externally validated at one's institution before use. The findings from the present study provide a framework for conducting external validation.

Modeling virtualized applications using machine learning techniques

With the growing adoption of virtualized datacenters and cloud hosting services, the allocation and sizing of resources such as CPU, memory, and I/O bandwidth for virtual machines (VMs) is becoming increasingly important. Accurate performance modeling of an application would help users in better VM sizing, thus reducing costs. It can also benefit cloud service providers who can offer a new charging model based on the VMs' performance instead of their configured sizes. In this paper, we present techniques to model the performance of a VM-hosted application as a function of the resources allocated to the VM and the resource contention it experiences. To address this multi-dimensional modeling problem, we propose and refine the use of two machine learning techniques: artificial neural network (ANN) and support vector machine (SVM). We evaluate these modeling techniques using five virtualized applications from the RUBiS and Filebench suite of benchmarks and demonstrate that their median and 90th percentile prediction errors are within 4.36% and 29.17% respectively. These results are substantially better than regression based approaches as well as direct applications of machine learning techniques without our refinements. We also present a simple and effective approach to VM sizing and empirically demonstrate that it can deliver optimal results for 65% of the sizing problems that we studied and produces close-to-optimal sizes for the remaining 35%.

A limited sampling strategy for the simultaneous estimation of tacrolimus, mycophenolic acid and unbound prednisolone exposure in adult kidney transplant recipients

The aim of this study was to develop a limited sampling strategy (LSS) for the simultaneous estimation of exposure to tacrolimus, mycophenolic acid and unbound prednisolone in adult kidney transplant recipients. Tacrolimus, mycophenolic acid and unbound prednisolone area under the concentration-time curve profiles from 0 to 12 h post dose (AUC(0-12)) were collected from 20 subjects. Multiple linear regression analyses were performed to develop a LSS enabling the simultaneous estimation of exposure to all three drugs. Median percentage prediction error and median absolute percentage prediction error were calculated via jackknife analysis to evaluate bias and imprecision. LSS showed superior ability to predict exposure compared with single concentration-time points. A LSS incorporating concentration measurements at 0.5 h (C(0.5)), 2 h (C(2)) and 4 h (C(4)) post dose displayed acceptable predictive ability for all three drugs. This LSS may serve as a useful research tool for further investigation of the utility of concentration monitoring of these medications.

Optimal mapping of terrestrial gamma dose rates using geological parent material and aerogeophysical survey data

Regulatory authorities need ways to estimate natural terrestrial gamma radiation dose rates (nGy h⁻¹) across the landscape accurately, to assess its potential deleterious health effects. The primary method for estimating outdoor dose rate is to use an in situ detector supported 1 m above the ground, but such measurements are costly and cannot capture the landscape-scale variation in dose rates which are associated with changes in soil and parent material mineralogy. We investigate the potential for improving estimates of terrestrial gamma dose rates across Northern Ireland (13,542 km²) using measurements from 168 sites and two sources of ancillary data: (i) a map based on a simplified classification of soil parent material, and (ii) dose estimates from a national-scale, airborne radiometric survey. We used the linear mixed modelling framework in which the two ancillary variables were included in separate models as fixed effects, plus a correlation structure which captures the spatially correlated variance component. We used a cross-validation procedure to determine the magnitude of the prediction errors for the different models. We removed a random subset of 10 terrestrial measurements and formed the model from the remainder (n = 158), and then used the model to predict values at the other 10 sites. We repeated this procedure 50 times. The measurements of terrestrial dose vary between 1 and 103 (nGy h⁻¹). The median absolute model prediction errors (nGy h⁻¹) for the three models declined in the following order: no ancillary data (10.8) > simple geological classification (8.3) > airborne radiometric dose (5.4) as a single fixed effect. Estimates of airborne radiometric gamma dose rate can significantly improve the spatial prediction of terrestrial dose rate.

Evaluation of SLOG/TCI-III pediatric system on target control infusion of propofol

BackgroundThe target-controlled infusion-III (SLOG/TCI-III) system was derived from a model set up by the local pediatric population for target control infusion of propofol.MethodsThe current study aimed at evaluating the difference between target concentrations of propofol and performance, which was measured using the SLOG/TCI-III system in children. Thirty children fulfilling the I-II criteria according to American Society of Anesthesiology were enrolled in the study. The target plasma concentration of propofol was fed into the SLOG/TCI-III system and compared with the measured concentrations of propofol. Blood samples were collected and analyzed by high performance liquid chromatography with fluorescence detector. The performance error (PE) was determined for each measured blood propofol concentration. The performances of the TCI-III system were determined by the median performance error (MDPE), the median absolute performance error (MDAPE), and Wobble (the median absolute deviation of each PE from the MDPE), respectively.ResultsConcentration against target concentration showed good linear correlation: concentration = 1.3428 target concentration - 0.2633 (r = 0.8667). The MDPE and MDAPE of the pediatric system were 10 and 22%, respectively, and the median value for Wobble was 24%. MDPE and MDAPE were less than 15 and 30%, respectively.ConclusionsThe performance of TCI-III system seems to be in the accepted limits for clinical practice in children.

Abstract P224: Determinants of Quality of Life Following Myocardial Infarction

Introduction: Optimizing quality of life (QOL) after acute myocardial infarction (AMI) is a major goal of treatment. Unlike mortality, for which numerous prediction models are available, no risk stratification models predicting QOL after AMI exist. Methods: In the prospective 24-center US TRIUMPH registry, 2693 AMI patients were assessed at baseline and 1 year after AMI for disease-specific QOL with the Seattle Angina Questionnaire (SAQ). Median regression (to account for skewed outcome distribution) was used to build a prediction model for 1-year QOL. Using 60 candidate sociodemographic, economic, and clinical variables, a linear regression model was built that fit all covariates against the predicted value of SAQ QOL as the dependent variable. The least significant covariates were sequentially removed until 90% of the variance of the full model was explained. Models accounted for clustering within site. Non-linear spline terms were considered for all continuous variables. Results: After variable selection, poor QOL at 1 year was associated with younger age, female sex, prior heart failure, prior CABG, depressive symptoms, perceived stress, low monthly financial reserves, self-reported financial difficulty getting medical care, and poor baseline SAQ QOL (Table). The R 2 of the median regression model was 11.9%, and the median absolute prediction error (the expected value of the absolute difference between the future and predicted responses) was 8.44 (interquartile range: 3.81 to 16.28). Conclusions: This is the first predictive model for QOL after AMI ever reported and demonstrates that traditional demographics and medical comorbidities only partly predict QOL. This model identifies important patient factors associated with QOL that may be targets for further intervention (such as financial difficulties and stress), can be used to identify high-risk patients, and may permit comparisons among hospitals in risk-adjusted QOL outcomes. Table Predictors of Long-term Disease-Specific QOL After Myocardial Infarction Predictor Estimate (95% Confidence Interval) of 1-Year Change in Median SAQ QOL Scores P-Value Age (per 10 years) 3.10 (2.58, 3.62) &lt;0.001 Female sex -2.61 (-4.03, -1.18) &lt;0.001 Prior heart failure -2.48 (-5.15, 0.19) 0.069 Prior bypass surgery -3.69 (-6.23, -1.14) 0.005 Baseline disease-specific QOL score (per 5 points) 0.43 (0.27, 0.60) &lt;0.001 Baseline Patient Health Questionnaire depression score (per 5 points) -2.37 (-3.25, -1.48) &lt;0.001 Baseline Perceived Stress Score (per 4 points) -1.15 (-2.02, -0.28) 0.010 Self-reported financial difficulty getting medical care -7.99 (-10.99, -4.99) &lt;0.001 Monthly financial reserves Just enough to make ends meet vs. Some money left over -0.08 (-1.34, 1.18) 0.90 Not enough to make ends meet vs. Some money left over -3.77 (-6.35, -1.19) 0.004

Performance evaluation of paediatric propofol pharmacokinetic models in healthy young children

The performance of eight currently available paediatric propofol pharmacokinetic models in target-controlled infusions (TCIs) was assessed, in healthy children from 3 to 26 months of age. Forty-one, ASA I-II children, aged 3-26 months were studied. After the induction of general anaesthesia with sevoflurane and remifentanil, a propofol bolus dose of 2.5 mg kg(-1) followed by an infusion of 8 mg kg(-1) h(-1) was given. Arterial blood samples were collected at 1, 2, 3, 5, 10, 20, 40, and 60 min post-bolus, at the end of surgery, and at 1, 3, 5, 30, 60, and 120 min after stopping the infusion. Model performance was visually inspected with measured/predicted plots. Median performance error (MDPE) and the median absolute performance error (MDAPE) were calculated to measure bias and accuracy of each model. Performance of the eight models tested differed markedly during the different stages of propofol administration. Most models underestimated propofol concentration 1 min after the bolus dose, suggesting an overestimation of the initial volume of distribution. Six of the eight models tested were within the accepted limits of performance (MDPE<20% and MDAPE<30%). The model derived by Short and colleagues performed best. Our results suggest that six of the eight models tested perform well in young children. Since most models overestimate the initial volume of distribution, the use for TCI might result in the administration of larger bolus doses than necessary.

Clinical evaluation of a simultaneous closed-loop anaesthesia control system for depth of anaesthesia and neuromuscular blockade*

We developed a closed-loop system to control the depth of anaesthesia and neuromuscular blockade using the bispectral index and the electromyogram simultaneously and evaluated the clinical performance of this combined system for general anaesthesia. Twenty-two adult patients were included in this study. Anaesthesia was induced by a continuous infusion of remifentanil at 0.4 μg.kg(-1) .min(-1) (induction dose) and then 0.25 μg.kg(-1) .min(-1) (maintenance dose) and propofol at 2 mg.kg(-1) 3 min later. The combined automatic control was started 2 min after tracheal intubation. The depth of anaesthesia was recorded using bispectral index monitoring using a target value of 40. The target value of neuromuscular blockade, using mivacurium, was a T1/T1(0) twitch height of 10%. The precision of the system was calculated using internationally defined performance parameters. Twenty patients were included in the data analysis. The mean (SD) duration of simultaneous control was 129 (69) min. No human intervention was necessary during the computer-controlled administration of propofol and mivacurium. All patients assessed the quality of anaesthesia as 'good' to 'very good'; there were no episodes of awareness. The mean (SD) median performance error, median absolute performance error and wobble for the control of depth of anaesthesia and for neuromuscular blockade were -0.31 (1.78), 6.76 (3.45), 6.32 (2.93) and -0.38 (1.68), 3.75 (4.83), 3.63 (4.69), respectively. The simultaneous closed-loop system using propofol and mivacurium was able to maintain the target values with a high level of precision in a clinical setting.

Clinical evaluation of an end-tidal target-controlled infusion closed-loop system for isoflurane administration in horses undergoing surgical procedures

A new volatile anaesthetic agent delivery system was tested in 15 horses undergoing scheduled surgical procedures. The delivery system consisted of a laptop computer (with dedicated software), a computer-controlled syringe driver (loaded with liquid isoflurane) connected to the inspiratory arm of a large-animal circle breathing system and a respiratory gas monitor, providing isoflurane end-tidal concentrations (ET(measured)) every 20 s to the computer. Following induction and connection to the breathing system, mechanical ventilation was started. The bodyweight (BW), fresh gas flow, breathing system and ventilator volume, and end-tidal isoflurane target (ET(target)) were entered into the computer. Using Lowe's equation, the software calculated the prime dose to be delivered by the syringe driver over 2 min. After this, the system delivered each minute the amount of isoflurane as determined by the following equation: Isoflurane (mL) = {2 × λ(B/G) × (200 × BW(0.75)) × (ET(target) - ET(measured)) + (fresh gas flow - (BW(0.75) × 0.07)) × (ET(measured))}/206. A fresh gas flow of 4 L oxygen min(-1) was administered until the inspired fraction of oxygen reached 0.7, and was then decreased. A target of 1.5% end-tidal isoflurane was initially used and subsequently adjusted to the clinical requirements. The system performance was evaluated using the median prediction error (MDPE) and the median absolute performance error (MDAPE), which were -3.6% and 5.29%, respectively. It was concluded that this system was useful to achieve end-tidal target-controlled infusion of isoflurane during equine anaesthesia.

Mathematical method to build an empirical model for inhaled anesthetic agent wash-in

BackgroundThe wide range of fresh gas flow - vaporizer setting (FGF - FD) combinations used by different anesthesiologists during the wash-in period of inhaled anesthetics indicates that the selection of FGF and FD is based on habit and personal experience. An empirical model could rationalize FGF - FD selection during wash-in.MethodsDuring model derivation, 50 ASA PS I-II patients received desflurane in O2 with an ADU® anesthesia machine with a random combination of a fixed FGF - FD setting. The resulting course of the end-expired desflurane concentration (FA) was modeled with Excel Solver, with patient age, height, and weight as covariates; NONMEM was used to check for parsimony. The resulting equation was solved for FD, and prospectively tested by having the formula calculate FD to be used by the anesthesiologist after randomly selecting a FGF, a target FA (FAt), and a specified time interval (1 - 5 min) after turning on the vaporizer after which FAt had to be reached. The following targets were tested: desflurane FAt 3.5% after 3.5 min (n = 40), 5% after 5 min (n = 37), and 6% after 4.5 min (n = 37).ResultsSolving the equation derived during model development for FD yields FD=-(e(-FGF*-0.23+FGF*0.24)*(e(FGF*-0.23)*FAt*Ht*0.1-e(FGF*-0.23)*FGF*2.55+40.46-e(FGF*-0.23)*40.46+e(FGF*-0.23+Time/-4.08)*40.46-e(Time/-4.08)*40.46))/((-1+e(FGF*0.24))*(-1+e(Time/-4.08))*39.29). Only height (Ht) could be withheld as a significant covariate. Median performance error and median absolute performance error were -2.9 and 7.0% in the 3.5% after 3.5 min group, -3.4 and 11.4% in the 5% after 5 min group, and -16.2 and 16.2% in the 6% after 4.5 min groups, respectively.ConclusionsAn empirical model can be used to predict the FGF - FD combinations that attain a target end-expired anesthetic agent concentration with clinically acceptable accuracy within the first 5 min of the start of administration. The sequences are easily calculated in an Excel file and simple to use (one fixed FGF - FD setting), and will minimize agent consumption and reduce pollution by allowing to determine the lowest possible FGF that can be used. Different anesthesia machines will likely have different equations for different agents.

Evaluation of limited sampling strategies for total and free prednisolone in adult kidney transplant recipients

The aims of this study were to evaluate the predictive performances of all previously derived limited sampling strategies (LSSs) for total prednisolone, and to derive LSSs for free prednisolone in an independent cohort of adult kidney transplant recipients. Total and free prednisolone area under the concentration-time curve profiles from 0 to 12 hours post-dose (AUC(0-12)) were collected from 20 subjects. All previously published total prednisolone LSSs were identified from the literature. Free prednisolone LSSs were developed using multiple linear regression analyses. AUC predicted by each of the LSSs was compared with AUC(0-12). Median percentage prediction error (MPPE) and median absolute percentage prediction error (MAPE) were calculated to evaluate bias and imprecision. Total dose-adjusted prednisolone exposure varied 5-fold among study participants, while free dose-adjusted prednisolone exposure varied 3-fold. Correlation (r²) between total and free prednisolone AUC(0-12) was 0.79 (p = <0.0001) for the entire study cohort. This correlation was poorer in those early compared with late post-transplant (r² = 0.42 (p = 0.04) versus r² = 0.59 (p = 0.009) respectively). Ten previously published LSSs for total prednisolone and 15 derived LSSs for free prednisolone performed with acceptable levels of bias and imprecision (<15%). Of the free prednisolone LSSs, an equation incorporating 0.25-, 2- and 4-h concentrations showed the highest predictive power (AUC₀-₁₂ = -17.20 + 1.18 × C₀.₂₅ + 2.75 × C₂ + 4.45 × C₄; MPPE = 0.1%, MAPE = 4.6%). Wide between-subject variability in drug exposure suggests a role for TDM. LSSs can accurately estimate both total and free prednisolone AUC(0-12). However, given the poor correlation observed between the two parameters, our data suggest that free prednisolone LSSs may be preferable.

Vancomycin Dosing: Assessment of Time to Therapeutic Concentration and Predictive Accuracy of Pharmacokinetic Modeling Software

Therapeutic drug monitoring is usually required for safe and effective administration of vancomycin. However, dosing recommendations from published guidelines are not suitable in achieving therapeutic vancomycin concentrations in a timely manner in patients with normal renal function. To audit vancomycin dosing and concentrations at our institution and evaluate the predictive accuracy of a pharmacokinetic simulation program, with a view to implementing a pharmacy-based pharmacokinetic service for vancomycin monitoring. Patients receiving vancomycin were identified prospectively through the therapeutic drug monitoring archives. Patient information was obtained from medication charts and medical records that were located on wards. Data were entered into the MM-USC*Pack program (Jelliffe R, University of Southern California, 2008, version 12.10). This software was used to predict initial and subsequent concentrations of vancomycin based on patient parameters. The predictive accuracy of this software was evaluated by comparing the predicted concentrations to the observed concentrations. During a 6-week period, 204 concentrations were measured in 77 patients. The most common dosing regimen was 1 g every 12 hours. Overall, initial trough concentrations were subtherapeutic (<10 mg/L) in 58% of patients and trough concentrations did not become therapeutic at any stage throughout therapy in 25% of patients. The pharmacokinetic modeling software demonstrated little systematic bias (-3.1%), but the precision (median prediction error) was 23% (interquartile range, 11-45%). Predictions were poorer in obese patients (body mass index >35 kg/m(2)) and in patients with unstable renal function. A delay in attaining target trough concentrations was observed in a significant proportion of patients. Pharmacokinetic modeling software is a potential tool to improve the timeliness of achieving adequate dosing by allowing concentrations to be determined prior to steady-state. The program was able to predict vancomycin concentrations across a heterogeneous patient population with little systematic bias, but only moderate precision.

DCE-MRI biomarkers of tumour heterogeneity predict CRC liver metastasis shrinkage following bevacizumab and FOLFOX-6.

Background:There is limited evidence that imaging biomarkers can predict subsequent response to therapy. Such prognostic and/or predictive biomarkers would facilitate development of personalised medicine. We hypothesised that pre-treatment measurement of the heterogeneity of tumour vascular enhancement could predict clinical outcome following combination anti-angiogenic and cytotoxic chemotherapy in colorectal cancer (CRC) liver metastases.Methods:Ten patients with 26 CRC liver metastases had two dynamic contrast-enhanced MRI (DCE-MRI) examinations before starting first-line bevacizumab and FOLFOX-6. Pre-treatment biomarkers of tumour microvasculature were computed and a regression analysis was performed against the post-treatment change in tumour volume after five cycles of therapy. The ability of the resulting linear model to predict tumour shrinkage was evaluated using leave-one-out validation. Robustness to inter-visit variation was investigated using data from a second baseline scan.Results:In all, 86% of the variance in post-treatment tumour shrinkage was explained by the median extravascular extracellular volume (ve), tumour enhancing fraction (EF), and microvascular uniformity (assessed with the fractal measure box dimension, d0) (R2=0.86, P<0.00005). Other variables, including baseline volume were not statistically significant. Median prediction error was 12%. Equivalent results were obtained from the second scan.Conclusion:Traditional image analyses may over-simplify tumour biology. Measuring microvascular heterogeneity may yield important prognostic and/or predictive biomarkers.

Diagnostic efficiency of truncated area under the curve from 0 to 2 h (AUC0–2) of mycophenolic acid in kidney transplant recipients receiving mycophenolate mofetil and concomitant tacrolimus

Recently, the use of the truncated area under the curve from 0 to 2 h (AUC(0-2)) of mycophenolic acid (MPA) has been proposed for therapeutic monitoring in liver transplant recipients. The aim of our study was the evaluation of the clinical usefulness of truncated AUC(0-2) in kidney transplant patients. Plasma MPA was measured in samples taken before the morning dose of mycophenolate mofetil, and one-half and 2 h post-dose, completing 63 MPA concentration-time profiles from 40 adult kidney transplant recipients. The AUC from 0 to 12 h (AUC(0-12)) was calculated using the validated algorithm of Pawinski et al. The truncated AUC(0-2) was calculated using the linear trapezoidal rule, and extrapolated to 0-12 h (trapezoidal extrapolated AUC(0-12)) as previously described. Algorithm calculated and trapezoidal extrapolated AUC(0-12) values showed high correlation (r=0.995) and acceptable dispersion (ma68=0.71 μg·h/mL), median prediction error (6.6%) and median absolute prediction error (12.6%). The truncated AUC(0-2) had acceptable diagnostic efficiency (87%) in the classification of subtherapeutic, therapeutic or supratherapeutic values with respect to AUC(0-12). However, due to the high inter-individual variation of the drug absorption-rate, the dispersion between both pharmacokinetic variables (ma68=6.9 μg·h/mL) was unacceptable. The substantial dispersion between truncated AUC(0-2) and AUC(0-12) values may be a serious objection for the routine use of MPA AUC(0-2) in clinical practice.

Evaluation of predictive performance of propofol target-controlled infusion system incorporating the Schnider pharmacokinetic parameters

Objective To evaluate the predictive performance of propofol target-controlled infusion (TCI) system incorporating the Schnider pharmacokinetic parameters in Chinese patients. Methods Forty ASA Ⅰ or Ⅱ patients, aged 25-45 yr, with body mass index 20-25 kg/m2 , scheduled for gynecological laparoscopic surgery un der general anesthesia, were enrolled in this study. Anesthesia was induced with TCI of propofol (target plasma concentration (Cp) 3 μg/ml) and remifentanil (Cp 4 ng/ml) . Propofol was infused by Orchestra TCI system incorporating the Schnider pharmacokinetic parameters. Tracheal intubation was facilitated with rocuronium 0.6 mg/kgafter the patients lost consciousness. The patients were mechanically ventilated. PETCO2 was maintained at 30-40 mm Hg. Anesthesia was maintained with TCI of remifentanil (Cp 4 ng/ml) and propofol (Cp 3-5 μg/ml) and intermittent iv boluses of atracurium 0.2 mg/kg. BIS value was maintained at 40-45. Venous blood samples were obtained at 15, 30, 45 and 60 min after pneumoperitoneum for measurement of blood propofol concentrations by high performance liquid chromatography with fluorescence detector. Performance error, median prediction performance error, median absolute performance error, wobble and divergence of propofol TCI system were calculated. Results The value for performance error was 21 % (13%), for median prediction performance error 6.7 % (37.4%),for median absolute performance error 19% (18%), for divergence - 0.65%/h (0.82%/h) and for wobble 16.3% (15.2% ) . Conclusion The accuracy of propofol TCI system incorporating the Schnider pharmacokinetic parameters is high in Chinese patients and its predictive performance is acceptable clinically. Key words: Propofol; Pharmacokinetics; Drug deliver systems

Evaluation of limited sampling methods for estimation of tacrolimus exposure in adult kidney transplant recipients

To examine the predictive performance of limited sampling methods for estimation of tacrolimus exposure in adult kidney transplant recipients. Twenty full tacrolimus area under the concentration-time curve from 0 to 12 h post-dose (AUC(0-12)) profiles (AUCf) were collected from 20 subjects. Predicted tacrolimus AUC(0-12) (AUCp) was calculated using the following: (i) 42 multiple regression-derived limited sampling strategies (LSSs); (ii) five population pharmacokinetic (PK) models in the Bayesian forecasting program TCIWorks; and (iii) a Web-based consultancy service. Correlations (r(2)) between C(0) and AUCf and between AUCp and AUCf were examined. Median percentage prediction error (MPPE) and median absolute percentage prediction error (MAPE) were calculated. Correlation between C(0) and AUCf was 0.53. Using the 42 LSS equations, correlation between AUCp and AUCf ranged from 0.54 to 0.99. The MPPE and MAPE were <15% for 29 of 42 equations (62%), including five of eight equations based on sampling taken ≤2 h post-dose. Using the PK models in TCIWorks, AUCp derived from only C(0) values showed poor correlation with AUCf (r(2)=0.27-0.54) and unacceptable imprecision (MAPE 17.5-31.6%). In most cases, correlation, bias and imprecision estimates progressively improved with inclusion of a greater number of concentration time points. When concentration measurements at 0, 1, 2 and 4 h post-dose were applied, correlation between AUCp and AUCf ranged from 0.75 to 0.93, and MPPE and MAPE were <15% for all models examined. Using the Web-based consultancy service, correlation between AUCp and AUCf was 0.74, and MPPE and MAPE were 6.6 and 9.6%, respectively. Limited sampling methods better predict tacrolimus exposure compared with C(0) measurement. Several LSSs based on sampling taken 2 h or less post-dose predicted exposure with acceptable bias and imprecision. Generally, Bayesian forecasting methods required inclusion of a concentration measurement from >2 h post-dose to adequately predict exposure.