Data science and machine learning in anesthesiology / 대한마취과학회지

Machine learning (ML) is revolutionizing anesthesiology research. Unlike classical research methods that are largely inference-based, ML is geared more towards making accurate predictions. ML is a field of artificial intelligence concerned with developing algorithms and models to perform prediction tasks in the absence of explicit instructions. Most ML applications, despite being highly variable in the topics that they deal with, generally follow a common workflow. For classification tasks, a researcher typically tests various ML models and compares the predictive performance with the reference logistic regression model. The main advantage of ML lies in its ability to deal with many features with complex interactions and its specific focus on maximizing predictive performance. However, emphasis on data-driven prediction can sometimes neglect mechanistic understanding. This article mainly focuses on the application of supervised ML to electronic health record (EHR) data. The main limitation of EHR-based studies is in the difficulty of establishing causal relationships. However, the associated low cost and rich information content provide great potential to uncover hitherto unknown correlations. In this review, the basic concepts of ML are introduced along with important terms that any ML researcher should know. Practical tips regarding the choice of software and computing devices are also provided. Towards the end, several examples of successful ML applications in anesthesiology are discussed. The goal of this article is to provide a basic roadmap to novice ML researchers working in the field of anesthesiology.

Introduction to dynamical systems analysis in quantitative systems pharmacology: basic concepts and applications

Quantitative systems pharmacology (QSP) can be regarded as a hybrid of pharmacometrics and systems biology. Here, we introduce the basic concepts related to dynamical systems theory that are fundamental to the analysis of systems biology models. Determination of the fixed points and their local stabilities constitute the most important step. Illustration of a phase portrait further helps investigate multistability and bifurcation behavior. As a motivating example, we examine a cell circuit model that deals with tissue inflammation and fibrosis. We show how increasing the severity and duration of inflammatory stimuli divert the system trajectories towards pathological fibrosis. Simulations that involve different parameter values offer important insights into the potential bifurcations and the development of efficient therapeutic strategies. We expect that this tutorial serves as a good starting point for pharmacometricians striving to widen their scope to QSP and physiologically-oriented modeling.

Introduction to dynamical systems analysis in quantitative systems pharmacology: basic concepts and applications

Quantitative systems pharmacology (QSP) can be regarded as a hybrid of pharmacometrics and systems biology. Here, we introduce the basic concepts related to dynamical systems theory that are fundamental to the analysis of systems biology models. Determination of the fixed points and their local stabilities constitute the most important step. Illustration of a phase portrait further helps investigate multistability and bifurcation behavior. As a motivating example, we examine a cell circuit model that deals with tissue inflammation and fibrosis. We show how increasing the severity and duration of inflammatory stimuli divert the system trajectories towards pathological fibrosis. Simulations that involve different parameter values offer important insights into the potential bifurcations and the development of efficient therapeutic strategies. We expect that this tutorial serves as a good starting point for pharmacometricians striving to widen their scope to QSP and physiologically-oriented modeling.

Characterization of circadian blood pressure patterns using non-linear mixed effects modeling

Characterizing the time course of baseline or pre-drug blood pressure is important in acquiring unbiased estimates of antihypertensive drug effect. In this study, we recruited 23 healthy male volunteers and measured systolic (SBP) and diastolic blood pressure (DBP) over 24 hours on an hourly basis. Using a non-linear mixed effects model, circadian rhythm observed in blood pressure measurements was described by incorporating two cosine functions with periods 24 and 12 hours. A mixture model was applied to identify subgroups exhibiting qualitatively different circadian rhythms. Our results suggested that 78% of the study population, defined as ‘dippers’, demonstrated a typical circadian profile with a morning rise and a nocturnal dip. The remaining 22% of the subjects defined as ‘non-dippers’, however, were not adequately described using the typical profile and demonstrated an elevation of blood pressure during night-time. Covariate search identified weight as being positively correlated with mesor of SBP. Visual predictive checks using 1,000 simulated datasets were performed for model validation. Observations were in agreement with predicted values in ‘dippers’, but deviated slightly in ‘non-dippers’. Our work is expected to serve as a useful reference in assessing systematic intra-day blood pressure fluctuations and antihypertensive effects as well as assessing drug safety of incrementally modified drugs.

Population pharmacodynamics of cilostazol in healthy Korean subjects

Cilostazol is used for the treatment of intermittent claudication, ulceration and pain. This study was conducted to develop a population pharmacodynamic (PD) model for cilostazol's closure time (CT) prolongation effect in healthy Korean subjects based on a pharmacokinetic (PK) model previously developed. PD data were obtained from 29 healthy subjects who participated in a study conducted in 2009 at Severance Hospital. The PK model used was a two-compartment model with first order absorption. CT data were best described by a turnover model with a fractional turnover rate constant (K(out)) inhibited by drug effects (Eff), which were represented by a sigmoid E(max) model [Eff = E(max) · C(γ) / (EC₅₀(γ)+C(γ))] with E(max) being maximum drug effect, EC₅₀ drug plasma concentration at 50% of E(max), C drug plasma concentrations, and γ the Hill coefficient. For the selected PD model, parameter estimates were 0.613 hr⁻¹ for K(out), 0.192 for E(max), 730 ng/ml for EC₅₀ and 5.137 for γ. Sex and caffeine drinking status significantly influenced the baseline CT, which was 85.36 seconds in male non-caffeine drinkers and increased by 15.5% and 16.4% in females and caffeine drinkers, respectively. The model adequately described the time course of CT. This was the first population PD study for cilostazol's CT prolongation effect in a Korean population.

Relationship between body weight and postmenstrual age in a Korean pediatric population

Weight is a covariate representative of body size and is known to influence drug disposition. Recently, with increased use of allometric scaling, this variable has become more significant in accounting for variability in pharmacokinetic parameters. In adults, weight can be considered as a time invariant covariate because physical development is complete. As a result, when weight is missing in data, the typical or median value (say, 70 kg) could be imputed. On the contrary, weight continuously changes with age in the pediatric population. In this case, it is more appropriate to consider different median weight for each age group. We constructed a prediction model for weight using postmenstrual age (PMA) with the data consisting of 83,014 Korean pediatric patients. Weight, PMA, and gender information were collected from electronic medical records. Sigmoid models multiplied by exponential or logistic function were tested for basic model structure. Covariate effects on model parameters were then investigated using selection criteria of p < 0.001. All analyses were performed using NONMEM 7.3.0 and R3.2.0. The sigmoid model multiplied by logistic function best described the data and there was a significant difference between boys and girls in model parameters. It is expected that the results obtained in this work can be used for imputation of missing weights in pediatrics when PMA is available. In addition, the developed model can be used for clinical studies in children under 12 years old whose weight change rapidly with age and for model building in dealing with time varying body weight as a covariate.

An imputation-based method to reduce bias in model parameter estimates due to non-random censoring in oncology trials

In oncology trials, patients are withdrawn from study at the time when progressive disease (PD) is diagnosed, which is defined as 20% increase of tumor size from the minimum. Such informative censoring can lead to biased parameter estimates when nonlinear mixed effects models are fitted using NONMEM. In this work, we investigated how empirical Bayes estimates (EBE) could be exploited to impute missing tumor size observations and partially correct biases in the parameter estimates. 50 simulated datasets, each consisting of 100 patients, were generated based on the published model. From the simulated dataset, censoring due to PD diagnosis has been implemented. Using the post-hoc EBEs acquired from fitting the censored datasets using NONMEM, imputed values were generated from the tumor size model. Model fitting was carried out using censored and imputed datasets. Parameter estimates using both datasets were compared with true values. Tumor growth rate and cell kill rate were approximately 28% and 16% underestimated when fitted using the censored dataset, respectively. With the imputed datasets, relative biases of tumor growth rate and cell kill rate decreased to about 6% and 0%, respectively. Our work demonstrates that using EBEs acquired from fitting the model to the censored dataset and imputing the unknown tumor size observations with individual predictions beyond the PD time point is a viable option to solve the bias associated with structural parameter estimates. This approach, however, would not be helpful in getting better estimates of variance parameters.