Evaluation of the predictive performance of Bayesian dosing for warfarin in Chinese patients

Improved stacking ensemble learning based on feature selection to accurately predict warfarin dose

Background

With the rapid development of artificial intelligence, prediction of warfarin dose via machine learning has received more and more attention. Since the dose prediction involve both linear and nonlinear problems, traditional machine learning algorithms are ineffective to solve such problems at one time.

Objective

Based on the characteristics of clinical data of Chinese warfarin patients, an improved stacking ensemble learning can achieve higher prediction accuracy.

Methods

Information of 641 patients from southern China who had reached a steady state on warfarin was collected, including demographic information, medical history, genotype, and co-medication status. The dataset was randomly divided into a training set (90%) and a test set (10%). The predictive capability is evaluated on a new test set generated by stacking ensemble learning. Additional factors associated with warfarin dose were discovered by feature selection methods.

Results

A newly proposed heuristic-stacking ensemble learning performs better than traditional-stacking ensemble learning in key metrics such as accuracy of ideal dose (73.44%, 71.88%), mean absolute errors (0.11 mg/day, 0.13 mg/day), root mean square errors (0.18 mg/day, 0.20 mg/day) and R2 (0.87, 0.82).

Conclusions

The developed heuristic-stacking ensemble learning can satisfactorily predict warfarin dose with high accuracy. A relationship between hypertension, a history of severe preoperative embolism, and warfarin dose is found, which provides a useful reference for the warfarin dose administration in the future.

Application of a warfarin dosing calculator to guide individualized dosing versus empirical adjustment after fixed dosing: a pilot study

Background: Warfarin has a narrow therapeutic window and individual variation, and patients require regular follow-up and monitoring of the International Normalized Ratio (INR) for dose adjustment. The calculation method of Warfarin Dosing Calculator (WDC) software is based on the European and American populations, and its accuracy in the Chinese population is yet to be verified. Objective: This study was to evaluate the feasibility of applying Warfarin Dosing Calculator software intervention in a real-world clinical research setting in China. Methods: The pilot study divided the included patients after valve replacement into an experimental group and a control group, with 38 cases in each group. In the control group, the initial dose was fixed at 2.5 mg/d and the dose was adjusted empirically during the study period; in the experimental group, the Warfarin Dosing Calculator software was applied to guide the dosing, and patients in both groups were followed up for 3 months. Analysis of the incidence anticoagulation outcomes and excessive anticoagulation events in both groups. Kaplan-Meier survival curves were used to analyze the correlation between different dosing regimens and first International Normalized Ratio attainment, and Logrank tests were performed. Results: The mean time required for first International Normalized Ratio compliance in the experimental group was 4.38 days less than in the control group, and the mean number of tests was 1.43 less (p < 05). Time in therapeutic range (TTR) was significantly higher in the experimental group than in the control group (p < 05). Kaplan-Meier survival curve analysis showed that the first International Normalized Ratio attainment rate was significantly higher in the experimental group than in the control group (p = 01). No major bleeding events occurred in either group, but other excessive anticoagulation events (INR>3.5 and minor bleeding) were significantly reduced in the experimental group compared with the control group (p < 05). Conclusion: Application of Warfarin Dosing Calculator software to guide individualized warfarin dosing may be better than a fixed dose of 2.5 mg/d. It may be shorten the time to first International Normalized Ratio attainment, and the attainment rate in the same time, and can better improve the mean Time in therapeutic range level value and reduce excessive anticoagulation events, which improves the safety of warfarin anticoagulation therapy in clinical practice. Clinical Trial Registration: https://www.chictr.org.cn/showproj.html?proj=52793, ChiCTR2000032393.

Long Short-Term Memory Network for Development and Simulation of Warfarin Dosing Model Based on Time Series Anticoagulant Data

Background

Warfarin is an effective treatment for thromboembolic disease but has a narrow therapeutic index, and dosage can differ tremendously among individuals. The study aimed to develop an individualized international normalized ratio (INR) model based on time series anticoagulant data and simulate individualized warfarin dosing.

Methods

We used a long short-term memory (LSTM) network to develop an individualized INR model based on data from 4,578 follow-up visits, including clinical and genetic factors from 624 patients whom we enrolled in our previous randomized controlled trial. The data of 158 patients who underwent valvular surgery and were included in a prospective registry study were used for external validation in the real world.

Results

The prediction accuracy of LSTM_INR was 70.0%, which was much higher than that of MAPB_INR (maximum posterior Bayesian, 53.9%). Temporal variables were significant for LSTM_INR performance (51.7 vs. 70.0%, P < 0.05). Genetic factors played an important role in predicting INR at the onset of therapy, while after 15 days of treatment, we found that it might unnecessary to detect genotypes for warfarin dosing. Using LSTM_INR, we successfully simulated individualized warfarin dosing and developed an application (AI-WAR) for individualized warfarin therapy.

Conclusion

The results indicate that temporal variables are necessary to be considered in warfarin therapy, except for clinical factors and genetic factors. LSTM network may have great potential for long-term drug individualized therapy.

Trial Registration

NCT02211326; www.chictr.org.cn:ChiCTR2100052089.

External Evaluation of a Bayesian Warfarin Dose Optimization Based on a Kinetic-Pharmacodynamic Model

Warfarin is a representative anticoagulant with large interindividual variability. The published kinetic-pharmacodynamic (K-PD) model allows the prediction of warfarin dose requirement in Swedish patients; however, its applicability in Japanese patients is not known. We evaluated the model's predictive performance in Japanese patients with various backgrounds and relationships using Bayesian parameter estimation and sampling times. A single-center retrospective observational study was conducted at Tokyo Women's Medical University, Medical Center East. The study population consisted of adult patients aged >20 years who commenced warfarin with a prothrombin time-international normalized ratio (PT-INR) from June 2015 to June 2019. The published K-PD model modified by Wright and Duffull was assessed using prediction-corrected visual predictive checks, focusing on clinical characteristics, including age, renal function, and individual prediction error. The external dataset included 232 patients who received an initial warfarin daily dose of 3.2 ± 1.28 mg with 2278 PT-INR points (median [range] follow-up period of 23 d [7-28]). Prediction-corrected visual predictive checks carried a propensity for underprediction. Additionally, age >60 years, body mass index ≤25 kg/m², and estimated glomerular filtration rate ≤60 mL/min/1.73 m² had a pronounced tendency to underpredict PT-INR. However, Bayesian prediction using four prior observations reduced underprediction. To improve the prediction performance of these special populations, further studies are required to construct a model to predict warfarin dose requirements in Japanese patients.