A New Algorithm Optimized for Initial Dose Settings of Vancomycin Using Machine Learning

Machine Learning Methods for Precision Dosing in Anticancer Drug Therapy: A Scoping Review

INTRODUCTION

In the last decade, various Machine Learning techniques have been proposed aiming to individualise the dose of anticancer drugs mostly based on a presumed drug effect or measured effect biomarkers. The aim of this scoping review was to comprehensively summarise the research status on the use of Machine Learning for precision dosing in anticancer drug therapy.

METHODS

This scoping review was conducted in accordance with the interim guidance by Cochrane and the Joanna Briggs Institute. We systematically searched the databases Medline (via PubMed), Embase and the Cochrane Library for research articles and reviews including results published after 2016. Results were reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) checklist.

RESULTS

A total of 17 relevant studies was identified. In 12 of the included studies, Reinforcement Learning methods were used, including Classical, Deep, Double Deep and Conservative Q-Learning and Fuzzy Reinforcement Learning. Furthermore, classical Machine Learning methods were compared in terms of their performance and an artificial intelligence platform based on parabolic equations was used to guide dosing prospectively and retrospectively, albeit only in a limited number of patients. Due to the significantly different algorithm structures, a meaningful comparison between the various Machine Learning approaches was not possible.

CONCLUSION

Overall, this review emphasises the clinical relevance of Machine Learning methods for anticancer drug dose optimisation, as many algorithms have shown promising results enabling model-free predictions with the potential to maximise efficacy and minimise toxicity when compared to standard protocols.

Determining steady-state trough range in vancomycin drug dosing using machine learning

BACKGROUND

Vancomycin is a renally eliminated, nephrotoxic, glycopeptide antibiotic with a narrow therapeutic window, widely used in intensive care units (ICU). We aimed to predict the risk of inappropriate vancomycin trough levels and appropriate dosing for each ICU patient.

METHODS

Observed vancomycin trough levels were categorized into sub-therapeutic, therapeutic, and supra-therapeutic levels to train and compare different classification models. We included adult ICU patients (≥ 18 years) with at least one vancomycin concentration measurement during hospitalization at Mayo Clinic, Rochester, MN, from January 2007 to December 2017.

RESULT

The final cohort consisted of 5337 vancomycin courses. The XGBoost models outperformed other machine learning models with the AUC-ROC of 0.85 and 0.83, specificity of 53% and 47%, and sensitivity of 94% and 94% for sub- and supra-therapeutic categories, respectively. Kinetic estimated glomerular filtration rate and other creatinine-based measurements, vancomycin regimen (dose and interval), comorbidities, body mass index, age, sex, and blood pressure were among the most important variables in the models.

CONCLUSION

We developed models to assess the risk of sub- and supra-therapeutic vancomycin trough levels to improve the accuracy of drug dosing in critically ill patients.

Optimizing vancomycin dosing in pediatrics: a machine learning approach to predict trough concentrations in children under four years of age

BACKGROUND

Vancomycin trough concentration is closely associated with clinical efficacy and toxicity. Predicting vancomycin trough concentrations in pediatric patients is challenging due to significant inter-individual variability and rapid physiological changes during maturation.

AIM

This study aimed to develop a machine learning model to predict vancomycin trough concentrations and determine optimal dosing regimens for pediatric patients < 4 years of age using ML algorithms.

METHOD

A single-center retrospective observational study was conducted from January 2017 to March 2020. Pediatric patients who received intravenous vancomycin and underwent therapeutic drug monitoring were enrolled. Seven ML models [linear regression, gradient boosted decision trees, support vector machine, decision tree, random forest, Bagging, and extreme gradient boosting (XGBoost)] were developed using 31 variables. Performance metrics including R-squared (R2), mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) were compared, and important features were ranked.

RESULTS

The study included 120 eligible trough concentration measurements from 112 patients. Of these, 84 measurements were used for training and 36 for testing. Among the seven algorithms tested, XGBoost showed the best performance, with a low prediction error and high goodness of fit (MAE = 2.55, RMSE = 4.13, MSE = 17.12, and R2 = 0.59). Blood urea nitrogen, serum creatinine, and creatinine clearance rate were identified as the most important predictors of vancomycin trough concentration.

CONCLUSION

An XGBoost ML model was developed to predict vancomycin trough concentrations and aid in drug treatment predictions as a decision-support technology.

Machine Learning: A New Approach for Dose Individualization

The application of machine learning (ML) has shown promising results in precision medicine due to its exceptional performance in dealing with complex multidimensional data. However, using ML for individualized dosing of medicines is still in its early stage, meriting further exploration. A systematic review of study designs and modeling details of using ML for individualized dosing of different drugs was performed. We have summarized the status of the study populations, predictive targets, and data sources for ML modeling, the selection of ML algorithms and features, and the evaluation and validation of their predictive performance. We also used the Prediction model Risk of Bias Assessment Tool (PROBAST) to assess the risk of bias of included studies. Currently, ML can be used for both a priori and a posteriori dose selection and optimization, and it can also assist the implementation of therapeutic drug monitoring. However, studies are mainly focused on drugs with narrow therapeutic windows, predominantly immunosuppressants (N = 23, 35.9%) and anti-infectives (N = 21, 32.8%), and there is currently only very limited attention for special populations, such as children (N = 22, 34.4%). Most studies showed poor methodological quality and a high risk of bias. The lack of external validation and clinical utility evaluation currently limits the further clinical implementation of ML for dose individualization. We therefore have proposed several ways to improve the clinical relevance of the studies and facilitate the translation of ML models into clinical practice.

A Deep Learning-Based Approach for Prediction of Vancomycin Treatment Monitoring: Retrospective Study Among Patients With Critical Illness

BACKGROUND

Vancomycin pharmacokinetics are highly variable in patients with critical illnesses, and clinicians commonly use population pharmacokinetic (PPK) models based on a Bayesian approach to dose. However, these models are population-dependent, may only sometimes meet the needs of individual patients, and are only used by experienced clinicians as a reference for making treatment decisions. To assist real-world clinicians, we developed a deep learning-based decision-making system that predicts vancomycin therapeutic drug monitoring (TDM) levels in patients in intensive care unit.

OBJECTIVE

This study aimed to establish joint multilayer perceptron (JointMLP), a new deep-learning model for predicting vancomycin TDM levels, and compare its performance with the PPK models, extreme gradient boosting (XGBoost), and TabNet.

METHODS

We used a 977-case data set split into training and testing groups in a 9:1 ratio. We performed external validation of the model using 1429 cases from Kangwon National University Hospital and 2394 cases from the Medical Information Mart for Intensive Care-IV (MIMIC-IV). In addition, we performed 10-fold cross-validation on the internal training data set and calculated the 95% CIs using the metric. Finally, we evaluated the generalization ability of the JointMLP model using the MIMIC-IV data set.

RESULTS

Our JointMLP model outperformed other models in predicting vancomycin TDM levels in internal and external data sets. Compared to PPK, the JointMLP model improved predictive power by up to 31% (mean absolute error [MAE] 6.68 vs 5.11) on the internal data set and 81% (MAE 11.87 vs 6.56) on the external data set. In addition, the JointMLP model significantly outperforms XGBoost and TabNet, with a 13% (MAE 5.75 vs 5.11) and 14% (MAE 5.85 vs 5.11) improvement in predictive accuracy on the inner data set, respectively. On both the internal and external data sets, our JointMLP model performed well compared to XGBoost and TabNet, achieving prediction accuracy improvements of 34% and 14%, respectively. Additionally, our JointMLP model showed higher robustness to outlier data than the other models, as evidenced by its higher root mean squared error performance across all data sets. The mean errors and variances of the JointMLP model were close to zero and smaller than those of the PPK model in internal and external data sets.

CONCLUSIONS

Our JointMLP approach can help optimize treatment outcomes in patients with critical illnesses in an intensive care unit setting, reducing side effects associated with suboptimal vancomycin administration. These include increased risk of bacterial resistance, extended hospital stays, and increased health care costs. In addition, the superior performance of our model compared to existing models highlights its potential to help real-world clinicians.

Machine learning in medication prescription: A systematic review

BACKGROUND

Medication prescription is a complex process that could benefit from current research and development in machine learning through decision support systems. Particularly pediatricians are forced to prescribe medications "off-label" as children are still underrepresented in clinical studies, which leads to a high risk of an incorrect dose and adverse drug effects.

METHODS

PubMed, IEEE Xplore and PROSPERO were searched for relevant studies that developed and evaluated well-performing machine learning algorithms following the PRISMA statement. Quality assessment was conducted in accordance with the IJMEDI checklist. Identified studies were reviewed in detail, including the required variables for predicting the correct dose, especially of pediatric medication prescription.

RESULTS

The search identified 656 studies, of which 64 were reviewed in detail and 36 met the inclusion criteria. According to the IJMEDI checklist, five studies were considered to be of high quality. 19 of the 36 studies dealt with the active substance warfarin. Overall, machine learning algorithms based on decision trees or regression methods performed superior regarding their predictive power than algorithms based on neural networks, support vector machines or other methods. The use of ensemble methods like bagging or boosting generally enhanced the accuracy of the dose predictions. The required input and output variables of the algorithms were considerably heterogeneous and differ strongly among the respective substance.

CONCLUSIONS

By using machine learning algorithms, the prescription process could be simplified and dosing correctness could be enhanced. Despite the heterogenous results among the different substances and cases and the lack of pediatric use cases, the identified approaches and required variables can serve as an excellent starting point for further development of algorithms predicting drug doses, particularly for children. Especially the combination of physiologically-based pharmacokinetic models with machine learning algorithms represents a great opportunity to enhance the predictive power and accuracy of the developed algorithms.

Discrepancy between perceptions and acceptance of clinical decision support Systems: implementation of artificial intelligence for vancomycin dosing

BACKGROUND

Artificial intelligence (AI) tools are more effective if accepted by clinicians. We developed an AI-based clinical decision support system (CDSS) to facilitate vancomycin dosing. This qualitative study assesses clinicians' perceptions regarding CDSS implementation.

METHODS

Thirteen semi-structured interviews were conducted with critical care pharmacists, at Mayo Clinic (Rochester, MN), from March through April 2020. Eight clinical cases were discussed with each pharmacist (N = 104). Following initial responses, we revealed the CDSS recommendations to assess participants' reactions and feedback. Interviews were audio-recorded, transcribed, and summarized.

RESULTS

The participants reported considerable time and effort invested daily in individualizing vancomycin therapy for hospitalized patients. Most pharmacists agreed that such a CDSS could favorably affect (N = 8, 62%) or enhance (9, 69%) their ability to make vancomycin dosing decisions. In case-based evaluations, pharmacists' empiric doses differed from the CDSS recommendation in most cases (88/104, 85%). Following revealing the CDSS recommendations, we noted 78% (69/88) discrepant doses. In discrepant cases, pharmacists indicated they would not alter their recommendations. The reasons for declining the CDSS recommendation were general distrust of CDSS, lack of dynamic evaluation and in-depth analysis, inability to integrate all clinical data, and lack of a risk index.

CONCLUSION

While pharmacists acknowledged enthusiasm about the advantages of AI-based models to improve drug dosing, they were reluctant to integrate the tool into clinical practice. Additional research is necessary to determine the optimal approach to implementing CDSS at the point of care acceptable to clinicians and effective at improving patient outcomes.

Risk evaluation of carbapenem-induced liver injury based on machine learning analysis

INTRODUCTION

Information regarding carbapenem-induced liver injury is limited, and the rate of liver injury caused by meropenem (MEPM) and doripenem (DRPM) remains unknown. Decision tree (DT) analysis, a machine learning method, has a flowchart-like model where users can easily predict the risk of liver injury. Thus, we aimed to compare the rate of liver injury between MEPM and DRPM and construct a flowchart that can be used to predict carbapenem-induced liver injury.

METHODS

We investigated patients treated with MEPM (n = 310) or DRPM (n = 320) and confirmed liver injury as the primary outcome. We used a chi-square automatic interaction detection algorithm to construct DT models. The dependent variable was set as liver injury from a carbapenem (MEPM or DRPM), and factors including alanine aminotransferase (ALT), albumin-bilirubin (ALBI) score, and concomitant use of acetaminophen were used as explanatory variables.

RESULTS

The rates of liver injury were 22.9% (71/310) and 17.5% (56/320) in the MEPM and DRPM groups, respectively; no significant differences in the rate were observed (95% confidence interval: 0.710-1.017). Although the DT model of MEPM could not be constructed, DT analysis showed that the incidence of introducing DRPM in patients with ALT >22 IU/L and ALBI scores > -1.87 might be high-risk.

CONCLUSIONS

The risk of developing liver injury did not differ significantly between the MEPM and DRPM groups. Since ALT and ALBI score are evaluated in clinical settings, this DT model is convenient and potentially useful for medical staff in assessing liver injury before DRPM administration.

Therapeutic Drug Monitoring of Vancomycin Concentrations for the Management of Bone and Joint Infections: An Urgent Need

Vancomycin is used for the treatment of bone and joint infections (BJI), but scarce information is available about its pharmacokinetic/pharmacodynamic (PK/PD) characteristics. We aimed to identify the risk factors associated with the non-achievement of an optimal PK/PD target in the first therapeutic drug monitoring (TDM). Methods: A retrospective study was conducted in a tertiary hospital from January 2020 to January 2022. Patients with BJI and TDM of vancomycin on day 2 of treatment were included. Initial vancomycin fixed doses (1 g every 8 h or 12 h) was decided by the responsible doctors. According to TDM results, dosage adjustments were performed. An AUC_24h/MIC < 400 mg × h/L, between 400 and 600 mg × h/L and >600 mg × h/L, were defined as suboptimal, optimal and supratherapeutic, respectively. Patients were grouped into these three categories. Demographic, clinical and PK characteristics were compared between groups. Nephrotoxicity at the end of treatment was assessed. Results: A total of 94 patients were included: 22 (23.4%), 42 (44.7%) and 30 (31.9%) presented an infratherapeutic, optimal and supratherapeutic PK/PD targets, respectively. A younger age and initial vancomycin dose <40 mg/kg/day were predictive factors for achieving a suboptimal PK/PD target, while older age, higher serum-creatinine and dose >40 mg/kg/day were associated with overexposure. The nephrotoxicity rate was 22.7%. More than 50% of patients did not achieve an optimal PK/PD. Considering age, baseline serum-creatinine and body weight, TDM is required to readily achieve an optimal and safe exposure.

AI Models to Assist Vancomycin Dosage Titration

Background: Effective treatment using antibiotic vancomycin requires close monitoring of serum drug levels due to its narrow therapeutic index. In the current practice, physicians use various dosing algorithms for dosage titration, but these algorithms reported low success in achieving therapeutic targets. We explored using artificial intelligent to assist vancomycin dosage titration. Methods: We used a novel method to generate the label for each record and only included records with appropriate label data to generate a clean cohort with 2,282 patients and 7,912 injection records. Among them, 64% of patients were used to train two machine learning models, one for initial dose recommendation and another for subsequent dose recommendation. The model performance was evaluated using two metrics: PAR, a pharmacology meaningful metric defined by us, and Mean Absolute Error (MAE), a commonly used regression metric. Results: In our 3-year data, only a small portion (34.1%) of current injection doses could reach the desired vancomycin trough level (14-20 mcg/ml). Both PAR and MAE of our machine learning models were better than the classical pharmacokinetic models. Our model also showed better performance than the other previously developed machine learning models in our test data. Conclusion: We developed machine learning models to recommend vancomycin dosage. Our results show that the new AI-assisted dosage titration approach has the potential to improve the traditional approaches. This is especially useful to guide decision making for inexperienced doctors in making consistent and safe dosing recommendations for high-risk medications like vancomycin.

Development of a decision flowchart to identify the patients need high-dose vancomycin in early phase of treatment

Background

The standard dose of vancomycin (VCM, 2 g/day) sometimes fails to achieve therapeutic concentration in patients with normal renal function. In this study, we aimed to identify factors to predict patients who require high-dose vancomycin (> 2 g/day) to achieve a therapeutic concentration and to develop a decision flowchart to select these patients prior to VCM administration.

Methods

Patients who had an estimated creatinine clearance using the Cockcroft–Gault equation (eCCr) of ≥50 mL/min and received intravenous VCM were divided into 2 cohorts: an estimation set (n = 146, from April to September 2016) and a validation set (n = 126, from October 2016 to March 2017). In each set, patients requiring ≤2 g/day of VCM to maintain the therapeutic trough concentration (10–20 μg/mL) were defined as standard-dose patients, while those who needed > 2 g/day were defined as high-dose patients. Univariate and multivariate logistic regression analysis was performed to identify the predictive factors for high-dose patients and decision tree analysis was performed to develop decision flowchart to identify high-dose patients.

Results

Among the covariates analyzed, age and eCCr were identified as independent predictors for high-dose patients. Further, the decision tree analysis revealed that eCCr (cut off value = 81.3 mL/min) is the top predictive factor and is followed by age (cut off value = 58 years). Based on these findings, a decision flowchart was constructed, in which patients with eCCr ≥81.3 mL/min and age < 58 years were designated as high-dose patients and other patients were designated as standard-dose patients. Subsequently, we applied this decision flowchart to the validation set and obtained good predictive performance (positive and negative predictive values are 77.6 and 84.4%, respectively).

Conclusion

These results suggest that the decision flowchart constructed in this study provides an important contribution for avoiding underdosing of VCM in patients with eCCr of ≥50 mL/min.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40780-021-00231-w.

Artificial intelligence to deep learning: machine intelligence approach for drug discovery

Drug designing and development is an important area of research for pharmaceutical companies and chemical scientists. However, low efficacy, off-target delivery, time consumption, and high cost impose a hurdle and challenges that impact drug design and discovery. Further, complex and big data from genomics, proteomics, microarray data, and clinical trials also impose an obstacle in the drug discovery pipeline. Artificial intelligence and machine learning technology play a crucial role in drug discovery and development. In other words, artificial neural networks and deep learning algorithms have modernized the area. Machine learning and deep learning algorithms have been implemented in several drug discovery processes such as peptide synthesis, structure-based virtual screening, ligand-based virtual screening, toxicity prediction, drug monitoring and release, pharmacophore modeling, quantitative structure-activity relationship, drug repositioning, polypharmacology, and physiochemical activity. Evidence from the past strengthens the implementation of artificial intelligence and deep learning in this field. Moreover, novel data mining, curation, and management techniques provided critical support to recently developed modeling algorithms. In summary, artificial intelligence and deep learning advancements provide an excellent opportunity for rational drug design and discovery process, which will eventually impact mankind. The primary concern associated with drug design and development is time consumption and production cost. Further, inefficiency, inaccurate target delivery, and inappropriate dosage are other hurdles that inhibit the process of drug delivery and development. With advancements in technology, computer-aided drug design integrating artificial intelligence algorithms can eliminate the challenges and hurdles of traditional drug design and development. Artificial intelligence is referred to as superset comprising machine learning, whereas machine learning comprises supervised learning, unsupervised learning, and reinforcement learning. Further, deep learning, a subset of machine learning, has been extensively implemented in drug design and development. The artificial neural network, deep neural network, support vector machines, classification and regression, generative adversarial networks, symbolic learning, and meta-learning are examples of the algorithms applied to the drug design and discovery process. Artificial intelligence has been applied to different areas of drug design and development process, such as from peptide synthesis to molecule design, virtual screening to molecular docking, quantitative structure-activity relationship to drug repositioning, protein misfolding to protein-protein interactions, and molecular pathway identification to polypharmacology. Artificial intelligence principles have been applied to the classification of active and inactive, monitoring drug release, pre-clinical and clinical development, primary and secondary drug screening, biomarker development, pharmaceutical manufacturing, bioactivity identification and physiochemical properties, prediction of toxicity, and identification of mode of action.

A hybrid machine learning/pharmacokinetic approach outperforms maximum a posteriori Bayesian estimation by selectively flattening model priors

Model‐informed precision dosing (MIPD) approaches typically apply maximum a posteriori (MAP) Bayesian estimation to determine individual pharmacokinetic (PK) parameters with the goal of optimizing future dosing regimens. This process combines knowledge about the individual, in the form of drug levels or pharmacodynamic biomarkers, with prior knowledge of the drug PK in the general population. Use of “flattened priors” (FPs), in which the weight of the model priors is reduced relative to observations about the patient, has been previously proposed to estimate individual PK parameters in instances where the patient is poorly described by the PK model. However, little is known about the predictive performance of FPs and when to apply FPs in MIPD. Here, FP is evaluated in a data set of 4679 adult patients treated with vancomycin. Depending on the PK model, prediction error could be reduced by applying FPs in 42–55% of PK parameter estimations. Machine learning (ML) models could identify instances where FPs would outperform MAPs with a specificity of 81–86%, reducing overall root mean squared error (RMSE) of PK model predictions by 12–22% (0.5–1.2 mg/L) relative to MAP alone. The factors most indicative of the use of FPs were past prediction residuals and bias in past PK predictions. A more clinically practical minimal model was developed using only these two features, reducing RMSE by 5–18% (0.20–0.93 mg/L) relative to MAP. This hybrid ML/PK approach advances the precision dosing toolkit by leveraging the power of ML while maintaining the mechanistic insight and interpretability of PK models.