Median Prediction Error Research Articles

Effective lens position and pseudophakic refraction prediction error at 10½ years of age in the Infant Aphakia Treatment Study

The refraction prediction error (PE) for infants with intraocular lens (IOL) implantation is large, possibly related to an effective lens position (ELP) that is different than in adult eyes. If these eyes still have nonadult ELPs as they age, this could result in persistently large PE. We aimed to determine whether ELP or biometry at age 10½ years correlated with PE in children enrolled in the Infant Aphakia Treatment Study (IATS). We compared the measured refraction of eyes randomized to primary IOL implantation to the "predicted refraction" calculated by the Holladay 1 formula, based on biometry at age 10½ years. Eyes with incomplete data or IOL exchange were excluded. The PE (predicted - measured refraction) and absolute PE were calculated. Measured anterior chamber depth (ACD) was used to assess the effect of ELP on PE. Multiple regression analysis was performed on absolute PE versus axial length, corneal power, rate of refractive growth, refractive error, and best-corrected visual acuity. Forty-three eyes were included. The PE was 0.63 ± 1.68 D; median absolute PE, 0.85 D (IQR, 1.83 D). The median absolute PE was greater when the measured ACD was used to calculate predicted refraction instead of the standard A-constant (1.88 D [IQR, 1.72] D vs 0.85 D [IQR, 1.83], resp. [P = 0.03]). Absolute PE was not significantly correlated with any other parameter. Variations in ELP did not contribute significantly to PE 10 years after infant cataract surgery.

Intraocular Lens Calculation Using 8 Formulas in Silicone Oil–Filled Eyes Undergoing Silicone Oil Removal and Phacoemulsification After Retinal Detachment

To evaluate formulas for intraocular lens (IOL) calculation in silicone oil (SO)-filled eyes. Retrospective, consecutive case series. We conducted our study at the Department of Ophthalmology, Goethe University, Frankfurt, Germany, and included SO-filled eyes that received SO removal combined with phacoemulsification and IOL implantation. Preoperative assessments included biometry (IOLMaster 700; Carl Zeiss Meditec). To evaluate the measurements, we compared the mean prediction error, and mean and median absolute prediction error of 8 different formulas. A total of 90 eyes matched our inclusion criteria. The median absolute error was lowest in the Barrett Universal II formula (0.43 diopters [D] ± 0.75) followed by Kane (0.44 D ± 0.75), Hill-radial basis function (0.47 D ± 0.74), Holladay II (0.47 D ± 0.77), Sanders Retzlaff Kraff/theoretical (0.51 D ± 0.74), Holladay I (0.51 D ± 0.76), and Haigis and Hoffer Q (0.52 D ± 0.74 each). Regarding eyes within ±0.5 D Barrett Universal II (57.8%, 52 eyes) performed best, again followed by Kane (56.7%, 51 eyes) and Hill-radial basis function (54.4%, 49 eyes). Using modern formulas for IOL calculation in oil-filled eyes improves predictability but still not as good as in unoperated eyes. This issue is created by the change in refractive index due to the SO fill and therefore a lower precision of axial length measurement and effective lens position prediction.

External Validation of a Vancomycin Population Pharmacokinetic Model and Developing a New Dosage Regimen in Neonates.

Vancomycin is the drug of choice in the treatment of MRSA infections. In a published vancomycin population pharmacokinetic study on neonates in Singapore healthcare institutions, it was found that vancomycin clearance was predicted by weight, postmenstrual age, and serum creatinine. The aim of this study was to externally validate the vancomycin population pharmacokinetic model to develop a new dosage regimen in neonates, and to compare this regimen with the existing institutional and NeoFax® dosage regimens. A retrospective chart review of neonates who received vancomycin therapy and therapeutic drug monitoring was conducted. The median prediction error percentage was calculated to assess bias, while the median absolute prediction error percentage and the root mean squared error percentage were calculated to assess precision. The new dosage regimen was developed using Monte Carlo simulation. A total of 20 neonates were included in the external validation dataset. Eighteen of them were premature, with a median gestational age of 27.7 (25.9-31.5) weeks and postmenstrual age of 30.5 (27.3-34.3) weeks at the point of vancomycin initiation. No apparent systematic bias was found in the predictions of the model. The external validation performed in the current study found the model to be generally unbiased. Our new vancomycin dosage regimen was able to achieve target trough concentrations and area under the curve (AUC24) at a greater proportion as compared to existing institutional and NeoFax® dosage regimens. The pharmacokinetic model built in the previous study can be used to conduct reliable population simulations of our Asian neonatal population in Singapore. The new dosage regimen was able to achieve target trough concentrations and AUC24 better than existing institutional and NeoFax® dosage regimens.

Reliability of the Minto model for target-controlled infusion of remifentanil during cardiac surgery with cardiopulmonary bypass

BackgroundThe Minto pharmacokinetic model is used for target-controlled infusion of remifentanil. The reliability of this model has never been evaluated during normothermic cardiac surgery with cardiopulmonary bypass (CPB). The aim of this study was to assess the predictive performance of the model during CPB to determine its reliability during cardiac surgery. MethodsThis was a single-centre observational study. Arterial blood samples were drawn at five time points: T1, after tracheal intubation; T2, immediately before CPB; T3, 10 min after starting CPB; T4, 45 min after starting CPB; T5, 10 min after weaning off CPB. Prediction error (PE) and absolute prediction error (APE) were calculated for each sample and used to determine median prediction error (MDPE) and median absolute prediction error (MDAPE) per patient. Risk factors for APE >30% were assessed using multivariable analysis. Results are presented as medians with inter-quartile ranges. ResultsFifty-eight patients with 283 blood samples (110 during CPB) were included. In the pre-CPB period, MDPE and MDAPE were –17.3 [–32.9 to 2.3] and 24.6 [12–37.7]%, whereas during CPB, they were –1.8 [–15.6 to 11.1] and 14.0 [6.74–27.1]%, respectively. There was no statistically significant difference between measured and predicted remifentanil plasma concentrations during CPB. Age, preoperative albumin concentrations, temperature, and haemodilution were not independently associated with MDAPE >30%. ConclusionsThe Minto model accurately predicts plasma remifentanil concentrations during cardiac surgery with CPB. Clinical trial registration2017-A03153-50.

Development and validation of a model to calculate anesthetic agent consumption from inspired and end-expired concentrations, minute ventilation, fresh gas flow and dead space ventilation.

Anesthetic agent consumption is often calculated as the product of fresh gas flow (FGF) and vaporizer dial setting (FVAP). Because FVAP of conventional vaporizers is not registered in automated anesthesia records, retrospective agent consumption studies are hampered. The current study examines how FVAP can be retrospectively calculated from the agent's inspired (FIN) and end-expired concentration (FET), FGF, and minute ventilation (MV). Theoretical analysis of agent mass balances in the circle breathing reveals FVAP = [FIN - (dead space fraction * FIN + (1 - dead space fraction) * FET) * (1 - FGF/MV)]/(1-(1 - FGF/MV)). FIN, FET, FGF and MV are routinely monitored, but dead space fraction is unknown. Dead space fraction for sevoflurane, desflurane, and isoflurane was therefore determined empirically from an unpublished data set of 161 patient containing FVAP, FIN, FET, MV and FGF ranging from 0.25 to 8 L/min delivered via an ADU® (GE, Madison, WI, USA). Dead space fraction for each agent was determined empirically by having Excel's solver function calculate the value of dead space fraction that minimized the sum of the squared differences between dialed FVAP and predicted FVAP. With dead space fraction known, the model was then prospectively tested for sevoflurane in O2/air using data collected over the course of two weeks with one FLOW-i (Getinge, Solna, Sweden) and one Zeus workstation (Dräger, Lübeck, Germany). Because both workstations use an electronically controlled vaporizer/injector, the dialed FVAP were available to allow the calculation of median performance error (MDPE) and median absolute performance error (MDAPE). MDPE and MDAP are reported as median and interquartiles. The empirical dead space fraction for isoflurane, sevoflurane, and desflurane were 0.59, 0.49, and 0.66, respectively. For prospective testing, a total of 149.4h of useful data were collected from 78 patient with the Zeus and Flow-i combined, with FGF ranging from 0.18 to 8 L/min. The model predicted dialed FVAP well, with a MDPE of -1 (-11, 6) % and MDAPE of 8 (4, 17) %. FVAP can be retrospectively calculated from FIN, FET, FGF, and MV plus an agent specific dead space fraction factor with a degree of error that we believe suffices for retrospective sevoflurane consumption analyses. Performance with other agents and N2O awaits further validation.

A framework to model global, regional, and national estimates of intimate partner violence

BackgroundAccurate and reliable estimates of violence against women form the backbone of global and regional monitoring efforts to eliminate this human right violation and public health problem. Estimating the prevalence of intimate partner violence (IPV) is challenging due to variations in case definition and recall period, surveyed populations, partner definition, level of age disaggregation, and survey representativeness, among others. In this paper, we aim to develop a sound and flexible statistical modeling framework for global, regional, and national IPV statistics.MethodsWe modeled IPV within a Bayesian multilevel modeling framework, accounting for heterogeneity of age groups using age-standardization, and age patterns and time trends using splines functions. Survey comparability is achieved using adjustment factors which are estimated using exact matching and their uncertainty accounted for. Both in-sample and out-of-sample comparisons are used for model validation, including posterior predictive checks. Post-processing of models’ outputs is performed to aggregate estimates at different geographic levels and age groups.ResultsA total of 307 unique studies conducted between 2000–2018, from 154 countries/areas, and totaling nearly 1.8 million unique women responses informed lifetime IPV. Past year IPV had a similar number of studies (n = 332), countries/areas represented (n = 159), and individual responses (n = 1.8 million). Roughly half of IPV observations required some adjustments. Posterior predictive checks suggest good model fit to data and out-of-sample comparisons provided reassuring results with small median prediction errors and appropriate coverage of predictions’ intervals.ConclusionsThe proposed modeling framework can pool both national and sub-national surveys, account for heterogeneous age groups and age trends, accommodate different surveyed populations, adjust for differences in survey instruments, and efficiently propagate uncertainty to model outputs. Describing this model to reproducible levels of detail enables the accurate interpretation and responsible use of estimates to inform effective violence against women prevention policy and programs, and global monitoring of elimination efforts as part of the Sustainable Development Goals.

External validation of vancomycin population pharmacokinetic models in ten cohorts of infected Chinese patients

Appropriate dosage of vancomycin is critical for safety and efficacy in treating infectious diseases. Various population pharmacokinetic (PPK) models have been established. To lay a foundation for the clinical application, it is important to perform external validation of the published model. The aim of this study was to find the most suitable vancomycin PPK model for treatment of infection in China amongst all studied models developed for adults. A systematic literature search of published population vancomycin pharmacokinetic analyses was performed using the PubMed database. The identified models were evaluated in ten independent cohorts from China. Nonlinear mixed-effects modeling (NONMEM) was used for data analysis. The median prediction error (MPE), the distribution of prediction error (PE), and normalized prediction distribution errors (NPDE) were calculated and described. Predictive performance was evaluated by bias and accuracy. A total of 449 vancomycin concentrations from 397 infected participants were used for external validation. The MPE of the three models from the Chinese population was less than 10%. Half of the models had an acceptable MPE. For patients with different site infections and different renal functions, each model differed in its predictive ability to some extent. The NPDE results showed that all models had an obvious bias. None of the models had good performance in both prediction- and simulation-based diagnostics. Carrying out sufficient external validation of the PPK model before clinical application is of utmost importance. In clinical practice, the model's population closest to the application population should be used.

In Silico Predictions of the Gastrointestinal Uptake of Macrocycles in Man Using Conformal Prediction Methodology

The gastrointestinal uptake of macrocyclic compounds is not fully understood. Here we applied our previously validated integrated system based on machine learning and conformal prediction to predict the passive fraction absorbed (fa), maximum fraction dissolved (fdiss), substrate specificities for major efflux transporters and total fraction absorbed (fa,tot) for a selected set of designed macrocyclic compounds (n = 37; MW 407-889 g/mol) and macrocyclic drugs (n = 16; MW 734-1203 g/mole) in vivo in man. Major aims were to increase the understanding of oral absorption of macrocycles and further validate our methodology. We predicted designed macrocycles to have high fa and low to high fdiss and fa,tot, and average estimates were higher than for the larger macrocyclic drugs. With few exceptions, compounds were predicted to be effluxed and well absorbed. A 2-fold median prediction error for fa,tot was achieved for macrocycles (validation set). Advantages with our methodology include that it enables predictions for macrocycles with low permeability, Caco-2 recovery and solubility (BCS IV), and provides prediction intervals and guides optimization of absorption. The understanding of oral absorption of macrocycles was increased and the methodology was validated for prediction of the uptake of macrocycles in man.

Strategies for formula constant optimisation for intraocular lens power calculation

BackgroundTo investigate modern nonlinear iterative strategies for formula constant optimisation and show the application and results from a large dataset using a set of disclosed theoretical-optical lens power calculation concepts.MethodsNonlinear iterative optimisation algorithms were implemented for optimising the root mean squared (SoSPE), the mean absolute (SoAPE), the mean (MPE), the standard deviation (SDPE), the median (MEDPE), as well as the 90% confidence interval (CLPE) of the prediction error (PE), defined as the difference between postoperative achieved and formula predicted spherical equivalent power of refraction. Optimisation was performed using the Levenberg-Marquardt algorithm (SoSPE and SoAPE) or the interior point method (MPE, SDPE, MEDPE, CLPE) for the SRKT, Hoffer Q, Holladay 1, Haigis, and Castrop formulae. The results were based on a dataset of measurements made on 888 eyes after implantation of an aspherical hydrophobic monofocal intraocular lens (Vivinex, Hoya).ResultsFor all formulae and all optimisation metrics, the iterative algorithms showed a fast and stable convergence after a couple of iterations. The results prove that with optimisation for SoSPE, SoAPE, MPE, SDPE, MEDPE, and CLPE the root mean squared PE, mean absolute PE, mean PE, standard deviation of PE, median PE, and confidence interval of PE could be minimised in all situations. The results in terms of cumulative distribution function are quite coherent with optimisation for SoSPE, SoAPE, MPE and MEDPE, whereas with optimisation for SDPE and CLPE the standard deviation and confidence interval of the PE distribution could only be minimised at the cost of a systematic offset in mean and median PE.ConclusionNonlinear iterative techniques are capable of minimising any statistical metrics (e.g. root mean squared or mean absolute error) of any target parameter (e.g. PE). These optimisation strategies are an important step towards optimising for the target parameters which are used for evaluating the performance of lens power calculation formulae.

Project Hyperopic Power Prediction II: The Effects of Second Eye Refinement Methods on Prediction Error in Hyperopic Eyes

Purpose The purpose of the study was to evaluate the potential accuracy of different second eye refinement methods in a patient cohort with short axial eye length to assess the performance of intraocular lens (IOL) power calculation schemes in high hyperopes. Methods The study design was a single-center, single-surgeon retrospective consecutive case series. The setting of the study was in Augen- und Laserklinik, Castrop-Rauxel, Germany. Patients were assessed after uneventful bilateral cataract surgery implanting either spherical (SA60AT) or aspheric (ZCB00) IOLs. Inclusion criteria were an axial eye length of ≤21.5 mm and/or emmetropizing IOL power of >28.5 dpt. Outcome measures were the mean absolute prediction error (MAE), median absolute prediction error, mean prediction error with standard deviation, median prediction error, and the percentage of eyes with an absolute prediction error (absPE) within 0.25 dpt, 0.5 dpt, 0.75 dpt, or 1.0 dpt. Second eye refinement was performed using the first eye prediction error, either with a correction coefficient of 0.50 (SER1), or an individual coefficient optimized for MAE. Results A total of 55 patients were assessed. A statistically significant reduction in the absPE after the application of SER1 was observed in 9 of 13 formulae. The SER1 refined Hoffer Q, refined Holladay I, refined Holladay II, refined Kane, refined Okulix, and refined PEARL-DGS provided a smaller absPE than other methods. Conclusion In this patient cohort with a short axial eye length, the second eye refinement led to a lower MAE in almost all formulae. The use of refinement in Kane, Okulix, PEARL-DGS, and Castrop formulae exhibited the lowest MAE.

Assessment of microalgae species, biomass, and distribution from spectral images using a convolution neural network

Effective monitoring of microalgae growth is crucial for environmental observation, while the applications of this monitoring could also be expanded to commercial and research-focused microalgae cultivation. Currently, the distinctive optical properties of different microalgae groups are targeted for monitoring. Since different microalgae can grow together, their spectral signals are mixed with ambient properties, making estimations of species biomasses a challenging task. In this study, we cultured five different microalgae and monitored their growth with a mobile spectral imager in three separate experiments. We trained and validated a one-dimensional convolution neural network by introducing absorbance spectra of the cultured microalgae and simulated pairwise mixtures of them. We then tested the model with samples of microalgae (monocultures and their pairwise mixtures) that were not part of the training or validation data. The convolution neural network classified microalgae accurately in the monocultures (test accuracy = 95%, SD = 4) and in the pairwise mixtures (test accuracy = 100%, SD = 0). Median prediction errors for biomasses were 17% (mean = 22%, SD = 18) for the monocultures and 17% (mean 24%, SD = 28) for the pairwise mixtures. As the spectral camera produced spatial information of the imaged target, we also demonstrated here the spatial distribution of microalgae biomass by applying the model across 5 × 5 pixel areas of the spectral images. The results of this study encourage the application of a one-dimensional convolution neural network to solve classification, regression, and distribution problems related to microalgae observation, simultaneously.

External Evaluation of a Single Center Precision Tacrolimus Dosing Model for Pediatric Heart Transplant

<h3>Purpose</h3> Extensive variability in tacrolimus pharmacokinetics (PK) influences attainment of therapeutic immunosuppression following pediatric heart transplant. Population PK (popPK) model-based personalized dosing of tacrolimus can reduce the need for tacrolimus titration and improve long-term pediatric heart transplant outcomes. The objective of this study was to externally evaluate a previously described single-center popPK model for personalized tacrolimus therapy. <h3>Methods</h3> A multicenter, retrospective analysis utilized demographics, tacrolimus dosing and concentration data, and laboratory values collected from pediatric inpatients during the first 6 weeks post-transplant, or until discharge, from three sites. PopPK analysis used the modeling software NONMEM v7.2. Observed concentration data were compared to data simulated with our single-center popPK model, using median prediction error (MPE, a measure of bias, <15% acceptable) and median absolute prediction error (MAPE, a measure of accuracy, <30% acceptable). <h3>Results</h3> A total of 285 children: 115 female, 196 white, 32 black, were included. Median age and creatinine clearance were 9.1 years and 101 mL/min/1.73m², respectively. Fluconazole use was rare, occurring in only 6 subjects. When utilizing all data the MPE and MAPE (interquartile range) were acceptable, at -3% (-25%, 10%) and 18% (5%, 49%), respectively, and not different by institution. However, when allowed only 3 concentrations (i.e. the minimum number to adequately predict future concentrations in our prior single-center study) the model poorly predicted future concentrations at all sites (MPE=31%; MAPE=47%). MPE and MAPE varied by site, ranging from 7 to 61% and 30 to 63%, respectively. Including site as a covariate in the popPK model improved predictions (MPE=21%; MAPE=37%) and decreased variability in predictive ability between sites (MPE: 11 to 34%; MAPE: 32 to 45%). <h3>Conclusion</h3> Our single-center popPK model did not adequately predict tacrolimus concentrations from external institutions. Study site appears to drive some of the poor model performance. Future directions include identifying study site differences that influence model performance, as well as identifying additional patient-specific factors affecting tacrolimus dosing requirements.

A Free Open-Source Bayesian Vancomycin Dosing App for Adults: Design and Evaluation Study.

BackgroundIt has been suggested that Bayesian dosing apps can assist in the therapeutic drug monitoring of patients receiving vancomycin. Unfortunately, Bayesian dosing tools are often unaffordable to resource-limited hospitals. Our aim was to improve vancomycin dosing in adults. We created a free and open-source dose adjustment app, VancoCalc, which uses Bayesian inference to aid clinicians in dosing and monitoring of vancomycin.ObjectiveThe aim of this paper is to describe the design, development, usability, and evaluation of a free open-source Bayesian vancomycin dosing app, VancoCalc.MethodsThe app build and model fitting process were described. Previously published pharmacokinetic models were used as priors. The ability of the app to predict vancomycin concentrations was performed using a small data set comprising of 52 patients, aged 18 years and over, who received at least 1 dose of intravenous vancomycin and had at least 2 vancomycin concentrations drawn between July 2018 and January 2021 at Lakeridge Health Corporation Ontario, Canada. With these estimated and actual concentrations, median prediction error (bias), median absolute error (accuracy), and root mean square error (precision) were calculated to evaluate the accuracy of the Bayesian estimated pharmacokinetic parameters.ResultsA total of 52 unique patients’ initial vancomycin concentrations were used to predict subsequent concentration; 104 total vancomycin concentrations were assessed. The median prediction error was –0.600 ug/mL (IQR –3.06, 2.95), the median absolute error was 3.05 ug/mL (IQR 1.44, 4.50), and the root mean square error was 5.34.ConclusionsWe described a free, open-source Bayesian vancomycin dosing calculator based on revisions of currently available calculators. Based on this small retrospective preliminary sample of patients, the app offers reasonable accuracy and bias, which may be used in everyday practice. By offering this free, open-source app, further prospective validation could be implemented in the near future.

Prospective validation of gas man simulations of sevoflurane in O2/air over a wide fresh gas flow range.

The use of inhaled anesthetics has come under increased scrutiny because of their environmental effects. This has led to a shift where sevoflurane in O2/air has become the predominant gas mixture to maintain anesthesia. To further reduce environmental impact, lower fresh gas flows (FGF) should be used. An accurate model of sevoflurane consumption allows us to assess and quantify the impact of the effects of lowering FGFs. This study therefore tested the accuracy of the Gas Man® model by determining its ability to predict end-expired sevoflurane concentrations (FETsevo) in patients using a protocol spanning a wide range of FGF and vaporizer settings. After IRB approval, 28 ASA I-II patients undergoing a gynecologic or urologic procedure under general endotracheal anesthesia were enrolled. Anesthesia was maintained with sevoflurane in O2/air, delivered via a Zeus or FLOW-i workstation (14 patients each). Every fifteen min, FGF was changed to randomly selected values ranging from 0.2 to 6 L/min while the sevoflurane vaporizer setting was left at the discretion of the anesthesiologist. The FETsevo was collected every min for 1h. For each patient, a Gas Man® simulation was run using patient weight and the same FGF, vaporizer and minute ventilation settings used during the procedure. For cardiac output, the Gas Man default setting was used (= Brody formula). Gas Man®'s performance was assessed by comparing measured with Gas Man® predicted FETsevo using linear regression and Varvel's criteria [median performance error (MDPE), median absolute performance error (MDAPE), and divergence]. Additional analysis included separating performance for the wash-in (0-15min) and maintenance phase (15-60min). For the FLOW-i, MDPE, MDAPE and divergence were 1% [-6, 8], 7% [3, 15] and -0.96%/h [-1.14, -0.88], respectively. During the first 15min, MDPE and MDAPE were 18% [1, 51] and 21% [8, 51], respectively, and during the last 45min 0% [-7, 5] and 6% [2, 10], respectively. For the Zeus, MDPE, MDAPE and divergence were 0% [-5, 8], 6% [3, 12] and -0.57%/h [-0.85, -0.16], respectively. During the first 15min, MDPE and MDAPE were 7% [-6, 28] and 13% [6, 32], respectively, and during the last 45min -1% [-5, 5] and 5% [2, 9], respectively. In conclusion, Gas Man® predicts FETsevo in O2/air in adults over a wide range of FGF and vaporizer settings using different workstations with both MDPE and MDAPE < 10% during the first hour of anesthesia, with better relative performance for simulating maintenance than wash-in. In the authors' opinion, this degree of performance suffices for Gas Man® to be used to quantify the environmental impact of FGF reduction in real life practice of the wash-in and maintenance period combined.

Evaluation and Application of Population Pharmacokinetic Models for Identifying Delayed Methotrexate Elimination in Patients With Primary Central Nervous System Lymphoma

Objective: Several population pharmacokinetic (popPK) models have been developed to determine the sources of methotrexate (MTX) PK variability. It remains unknown if these published models are precise enough for use or if a new model needs to be built. The aims of this study were to 1) assess the predictability of published models and 2) analyze the potential risk factors for delayed MTX elimination. Methods: A total of 1458 MTX plasma concentrations, including 377 courses (1–17 per patient), were collected from 77 patients who were receiving high-dose MTX for the treatment of primary central nervous system lymphoma in Huashan Hospital. PopPK analysis was performed using the NONMEM® software package. Previously published popPK models were selected and rebuilt. A new popPK model was then constructed to screen potential covariates using a stepwise approach. The covariates were included based on the combination of theoretical mechanisms and data properties. Goodness-of-fit plots, bootstrap, and prediction- and simulation-based diagnostics were used to determine the stability and predictive performance of both the published and newly built models. Monte Carlo simulations were conducted to qualify the influence of risk factors on the incidence of delayed elimination. Results: Among the eight evaluated published models, none presented acceptable values of bias or inaccuracy. A two-compartment model was employed in the newly built model to describe the PK of MTX. The estimated mean clearance (CL/F) was 4.91 L h−1 (relative standard error: 3.7%). Creatinine clearance, albumin, and age were identified as covariates of MTX CL/F. The median and median absolute prediction errors of the final model were -10.2 and 36.4%, respectively. Results of goodness-of-fit plots, bootstrap, and prediction-corrected visual predictive checks indicated the high predictability of the final model. Conclusions: Current published models are not sufficiently reliable for cross-center use. The elderly patients and those with renal dysfunction, hypoalbuminemia are at higher risk of delayed elimination.

Development of a Generic Physiologically-Based Pharmacokinetic Model for Lactation and Prediction of Maternal and Infant Exposure to Ondansetron via Breast Milk.

Ondansetron is commonly used in breastfeeding mothers to treat nausea and vomiting. There is limited information in humans regarding safety of ondansetron exposure to nursing infants and no adequate study looking at ondansetron pharmacokinetics during lactation. We developed a generic physiologically-based pharmacokinetic lactation model for small molecule drugs and applied this model to predict ondansetron transfer into breast milk and characterize infant exposure. Drug-specific model inputs were parameterized using data from the literature. Population-specific inputs were derived from a previously conducted systematic literature review of anatomic and physiologic changes in postpartum women. Model predictions were evaluated using ondansetron plasma and breast milk concentration data collected prospectively from 78 women in the Commonly Used Drugs During Lactation and infant Exposure (CUDDLE) study. The final model predicted breast milk and plasma exposures following a single 4mg dose of intravenous ondansetron in 1,000 simulated women who were 2days postpartum. Model predictions showed good agreement with observed data. Breast milk median prediction error (MPE) was 18.4% and median absolute prediction error (MAPE) was 53.0%. Plasma MPE was 32.5% and MAPE was 43.2%. The model-predicted daily and relative infant doses were 0.005mg/kg/day and 3.0%, respectively. This model adequately predicted ondansetron passage into breast milk. The calculated low relative infant dose indicates that mothers receiving ondansetron can safely breastfeed. The model building blocks and population database are open-source and can be adapted to other drugs.

Pharmacokinetics of Cefepime in Children on Extracorporeal Membrane Oxygenation: External Model Validation, Model Improvement and Dose Optimization.

Cefepime is a first-line therapy for Gram-negative infections in children on extracorporeal membrane oxygenation. Cefepime pharmacokinetics (PK) in children on extracorporeal membrane oxygenation still needs to be better established. This was a prospective single-center PK study. A maximum of 12 PK samples per patient were collected in children <18 years old on extracorporeal membrane oxygenation who received clinically indicated cefepime. External validation of a previously published population PK model was performed by applying the model in a new data set. The predictive performance of the model was determined by calculating prediction errors. Because of poor predictive performance, a revised model was developed using NONMEM and a combined data set that included data from both studies. Dose-exposure simulations were performed using the final model. Optimal dosing was judged based on the ability to maintain free cefepime concentrations above the minimal inhibitory concentration (MIC) for 68% and 100% of the dosing interval. Seventeen children contributed 105 PK samples. The mean (95% CI) and median (interquartile range) prediction errors were 33.7% (19.8-47.7) and 17.5% (-22.6 to 74.4). A combined data set was created, which included 33 children contributing 310 PK samples. The final improved 2-compartment model included weight and serum creatinine on clearance and oxygenator day and blood transfusion on volume of the central compartment. At an MIC of 8 mg/L, 50 mg/kg/dose every 8 hours reached target concentrations. Dosing intervals of 8 hours were needed to reach adequate concentrations at an MIC of 8 mg/L. Longer dosing intervals were adequate with higher serum creatinine and lower MICs.

In silico predictions of the human pharmacokinetics/toxicokinetics of 65 chemicals from various classes using conformal prediction methodology

Pharmacokinetic/toxicokinetic (PK/TK) information for chemicals in humans is generally lacking. Here we applied machine learning, conformal prediction and a new physiologically-based PK/TK model for prediction of the human PK/TK of 65 chemicals from different classes, including carcinogens, food constituents and preservatives, vitamins, sweeteners, dyes and colours, pesticides, alternative medicines, flame retardants, psychoactive drugs, dioxins, poisons, UV-absorbents, surfactants, solvents and cosmetics. About 80% of the main human PK/TK (fraction absorbed, oral bioavailability, half-life, unbound fraction in plasma, clearance, volume of distribution, fraction excreted) for the selected chemicals was missing in the literature. This information was now added (from in silico predictions). Median and mean prediction errors for these parameters were 1.3- to 2.7-fold and 1.4- to 4.8-fold, respectively. In total, 59 and 86% of predictions had errors <2- and <5-fold, respectively. Predicted and observed PK/TK for the chemicals was generally within the range for pharmaceutical drugs. The results validated the new integrated system for prediction of the human PK/TK for different chemicals and added important missing information. No general difference in PK/TK-characteristics was found between the selected chemicals and pharmaceutical drugs.

Accuracy of intraocular lens power calculation for cataract surgery after deep anterior lamellar keratoplasty.

The purpose of this study was to assess the accuracy of intraocular lens power (IOL) formulas for cataract surgery after deep anterior lamellar keratoplasty (DALK). This retrospective study included eyes which had previously undergone DALK and underwent standard phacoemulsification with monofocal IOL implantation between January 2012 and January 2021 at Ospedali Privati Forlì "Villa Igea" (Forlì, Italy). The predicted spherical equivalent (SE) was calculated using the Barrett Universal II, Emmetropia Verifying Optical (EVO), Haigis, Hoffer Q, Hoffer QST, Holladay 1, Holladay II, Kane and SRK/T formulas. Prediction error (PE) was calculated as the actual postoperative SE refraction minus the SE predicted refraction. Eighty-two eyes of 82 patients were included. The mean PE was negative using all formulas. Friedman test revealed a statistically significant difference of the median absolute PE (MedAE) among the different IOL formulas (P=0.005). On the basis of the MedAE, the formulas were ranked as follows: SRK/T (0.805 D), Kane (0.810 D), EVO (0.845 D), Hoffer QST (0.847 D), Barrett (0.895 D), Holladay 1 (0.915 D), Haigis (1.010 D) and Hoffer Q (1.070 D) formulas. All formulas had a tendency towards a myopic refractive shift in post-DALK eyes. Although the SRK/T, Kane, EVO and Hoffer QST formulas were more accurate, predictability of refractive outcomes was lower than in virgin eyes.

Continuous action deep reinforcement learning for propofol dosing during general anesthesia

PurposeAnesthesiologists simultaneously manage several aspects of patient care during general anesthesia. Automating administration of hypnotic agents could enable more precise control of a patient's level of unconsciousness and enable anesthesiologists to focus on the most critical aspects of patient care. Reinforcement learning (RL) algorithms can be used to fit a mapping from patient state to a medication regimen. These algorithms can learn complex control policies that, when paired with modern techniques for promoting model interpretability, offer a promising approach for developing a clinically viable system for automated anesthestic drug delivery. MethodsWe expand on our prior work applying deep RL to automated anesthetic dosing by now using a continuous-action model based on the actor-critic RL paradigm. The proposed RL agent is composed of a policy network that maps observed anesthetic states to a continuous probability density over propofol-infusion rates and a value network that estimates the favorability of observed states. We train and test three versions of the RL agent using varied reward functions. The agent is trained using simulated pharmacokinetic/pharmacodynamic models with randomized parameters to ensure robustness to patient variability. The model is tested on simulations and retrospectively on nine general anesthesia cases collected in the operating room. We utilize Shapley additive explanations to gain an understanding of the factors with the greatest influence over the agent's decision-making. ResultsThe deep RL agent significantly outperformed a proportional-integral-derivative controller (median episode median absolute performance error 1.9% ± 1.8 and 3.1% ± 1.1). The model that was rewarded for minimizing total doses performed the best across simulated patient demographics (median episode median performance error 1.1% ± 0.5). When run on real-world clinical datasets, the agent recommended doses that were consistent with those administered by the anesthesiologist. ConclusionsThe proposed approach marks the first fully continuous deep RL algorithm for automating anesthestic drug dosing. The reward function used by the RL training algorithm can be flexibly designed for desirable practices (e.g. use less anesthetic) and bolstered performances. Through careful analysis of the learned policies, techniques for interpreting dosing decisions, and testing on clinical data, we confirm that the agent's anesthetic dosing is consistent with our understanding of best-practices in anesthesia care.