Machine Learning-Based Prediction of Digoxin Toxicity in Heart Failure: A Multicenter Retrospective Study

Abstract

Digoxin toxicity (plasma digoxin concentration ≥0.9 ng/mL) is associated with worsening heart failure (HF). Decision tree (DT) analysis, a machine learning method, has a flowchart-like model where users can easily predict the risk of adverse drug reactions. The present study aimed to construct a flowchart using DT analysis that can be used by medical staff to predict digoxin toxicity. We conducted a multicenter retrospective study involving 333 adult patients with HF who received oral digoxin treatment. In this study, we employed a chi-squared automatic interaction detection algorithm to construct DT models. The dependent variable was set as the plasma digoxin concentration (≥ 0.9 ng/mL) in the trough during the steady state, and factors with p < 0.2 in the univariate analysis were set as the explanatory variables. Multivariate logistic regression analysis was conducted to validate the DT model. The accuracy and misclassification rates of the model were evaluated. In the DT analysis, patients with creatinine clearance <32 mL/min, daily digoxin dose ≥1.6 µg/kg, and left ventricular ejection fraction ≥50% showed a high incidence of digoxin toxicity (91.8%; 45/49). Multivariate logistic regression analysis revealed that creatinine clearance <32 mL/min and daily digoxin dose ≥1.6 µg/kg were independent risk factors. The accuracy and misclassification rates of the DT model were 88.2 and 46.2 ± 2.7%, respectively. Although the flowchart created in this study needs further validation, it is straightforward and potentially useful for medical staff in determining the initial dose of digoxin in patients with HF.

INTRODUCTION

Digoxin is an agent used for managing atrial fibrillation with heart failure (HF), and its efficacy is well-known as a second-line option.^1,2) A randomized, double-blind clinical trial (DIG trial) reported that the administration of digoxin reduced both hospitalization and worsening HF³⁾ and has the potential to increase all-cause mortality in some meta-analyses.^4,5) These studies suggest that digoxin should be carefully administered in patients with HF. One of the most important factors affecting digoxin toxicity is its plasma concentration. Since plasma concentrations of digoxin exceeding 0.9 ng/mL are strongly associated with worsening HF,^6,7) plasma levels of digoxin may be optimized to the range between 0.5 to 0.8 ng/mL in patients with HF.⁸⁾ Therefore, therapeutic drug monitoring is necessary to confirm the efficacy and safety of digoxin in the clinical setting.

Digoxin treatment is likely to result in inter-individual plasma concentrations because of its large distribution volume (7.3 L/kg) and long half-life (24–36 h).⁹⁾ Because the urinary excretion rate of the unchanged form of digoxin is 60–80%,⁹⁾ renal dysfunction is a significant risk factor for its increased plasma concentrations.¹⁰⁾ HF often complicates renal injury because it disturbs systemic circulation and organ perfusion, especially in patients with HF.¹¹⁾ Moreover, the clearance of digoxin in patients with HF was decreased compared to patients without HF, independent of renal function.¹²⁾ This suggests that renal and cardiac function, including left ventricular ejection fraction (LVEF), could be attributed to the plasma digoxin concentration. Therefore, it may be necessary to consider the risk of digoxin toxicity combined with LVEF and renal function in patients with HF.

Recent studies suggest that machine learning (ML) may be useful in the medical field as machines iteratively learn to predict potential risks.¹³⁾ Decision tree (DT) analysis, a type of ML, has a flowchart-like model where users can easily predict the risk of adverse drug reactions by considering the mutual relationship between multiple factors.^14–16) Therefore, DT analysis may be a useful new predictive tool for digoxin toxicity in clinical settings, but this has not yet been explored. In the present study, we conducted a multicenter retrospective study to construct a simple flowchart-based DT analysis that can be used by medical staff to predict digoxin toxicity before digoxin therapy.

MATERIALS AND METHODS

Study Design

This retrospective study was conducted at eight hospitals belonging to the Tokai-Hokuriku Group of the National Hospital Organization (Shizuoka Medical Center, Nagoya Medical Center, Kanazawa Medical Center, National Center for Geriatrics and Gerontology, Mie Chuo Medical Center, Higashinagoya National Hospital, Shizuoka Institute of Epilepsy and Neurological Disorders, and Toyohashi Medical Center).

Data Collection

Data on patients receiving oral digoxin at the eight hospitals from January 1, 2011, to December 31, 2021, were collected. Data, such as patient characteristics (sex, age, height, body weight, body mass index, renal function, LVEF, medical conditions, and concomitant drugs) and digoxin-related data (dosage, period of treatment, and plasma concentration), were collected from electronic medical records. The definition of HF in this study was as follows: the diagnosis of HF in a patient by an attending physician based on a comprehensive evaluation of symptoms and underlying medical conditions. The exclusion criteria were as follows: age <18 years, no plasma digoxin measurement, no echocardiography data, not diagnosed with HF, undergoing hemodialysis, and those with missing data. Because we evaluate the trough digoxin concentration during the steady state, patients with plasma concentrations measured less than 7 d after oral digoxin administration were excluded.¹⁷⁾ In addition, patients with digoxin concentrations below the limit of quantification (< 0.2 ng/mL) were excluded. Only the most recent data were selected for patients with multiple episodes of plasma digoxin measurement during the study period. Finally, a total of 333 patients with HF who received oral digoxin were eligible for analysis (Fig. 1).

Fig. 1. Flow Diagram Illustrating the Patient Selection Process for the DT Analysis of Digoxin Toxicity in Patients with HF

Data Calculation

As plasma digoxin concentrations of ≥0.9 ng/mL have been reported to be involved in worsening HF,^6,7) considering the prognosis of patients with HF, a plasma digoxin concentration of ≥0.9 ng/mL falls in the toxic range. Therefore, in this study, digoxin toxicity was defined as plasma digoxin concentrations of ≥0.9 ng/mL. The relationship between digoxin toxicity and patient characteristics, including sex, age, body mass index, digoxin dose, creatinine clearance (Ccr), number of concomitant drugs used, use of typical inducers of P-glycoprotein (P-gp) (phenytoin, carbamazepine, phenobarbital, or rifampicin),¹⁸⁾ use of typical inhibitors of P-gp (erythromycin, clarithromycin, protease inhibitors, or azole antifungals),¹⁹⁾ and LVEF, was investigated. Ccr was estimated using the Cockcroft–Gault formula.²⁰⁾

The concentration-to-dose (C/D) ratio of digoxin was calculated using the following equation:

  

DT Analysis

This study constructed a flowchart based on the chi-squared automatic interaction detection (CHAID) algorithm. CHAID is a multi-branch DT method that repeats the process of conducting a chi-square test to determine the association between two variables. One dependent variable and one independent variable were set, and the result was divided by the variable combination with the highest chi-square value. This procedure is repeated, and the tree branching stops when the stopping criteria are satisfied. The stop criteria of the branches were: (i) parent nodes ≤20 subjects or child nodes ≤10 subjects and (ii) no significant differences among the independent variables. Here, the dependent variable was set as the plasma digoxin concentration, and the factors with p < 0.2 in the univariate analysis were selected as explanatory variables.¹⁴⁾ DT analysis was performed using SPSS Decision Trees Version 27 (IBM Japan, Tokyo, Japan).

Model Validation

The DT validation criteria constructed in this study were based on a previous report by Miyai et al.¹⁴⁾ Multivariate logistic regression analysis was performed to determine the independent factors associated with a plasma digoxin concentration ≥0.9 ng/mL. Factors that exhibited p < 0.05 in the univariate analysis and LVEF were selected as explanatory variables. The cut-off value for age was determined using the receiver operating characteristic (ROC) curve. The Hosmer–Lemeshow test was performed to assess the goodness of fit of the multivariate logistic regression model, which was set at p > 0.05. To evaluate the misclassification rate of the flowchart constructed using the CHAID algorithm, 10-fold cross-validation was performed. This procedure divides the population data into 10 subsamples of equal sizes. Nine of these data subsamples were analyzed for training purposes, and the remaining data were evaluated based on the results of the analysis. This process was repeated for every subsample, and the mean of the 10 iterations was estimated as the misclassification risk value.

Statistical Analysis

Continuous variables between the plasma digoxin concentration ≥0.9 ng/mL or <0.9 ng/mL groups were analyzed using Student’s t-test or the Mann–Whitney U test. If the continuous variables followed a normal distribution, Student’s t-test was used for comparison. In contrast, the Mann–Whitney U test was conducted for data with non-normal distributions. Categorical variables were compared using the chi-squared test. Fisher’s exact test was used to include one cell with an expected value of <5 on a 2 × 2 contingency table. All statistical analyses were performed using SPSS version 27 (IBM Japan, Tokyo, Japan), and statistical significance was set at p < 0.05.

Ethics Approval and Consent to Participate

This study was performed in accordance with the Ethical Guidelines for Medical and Health Research Involving Human Subjects. The study protocol was approved by the ethics committee of Shizuoka Medical Center (Approval Ref. 2021-R25), Nagoya Medical Center (Approval Ref. 2021-053), Kanazawa Medical Center (Approval Ref. R03-069), the National Center for Geriatrics and Gerontology (Approval Ref. 21TB33), Mie Chuo Medical Center (Approval Ref. MCERB-202129), Higashinagoya National Hospital (Approval Ref. 3-10), Shizuoka Institute of Epilepsy and Neurological Disorders (Approval Ref. 2021-16), and Toyohashi Medical Center (Approval Ref. 3-19). Owing to the retrospective case-control study design, consent was obtained from each patient using an opt-out document posted on the websites of the respective hospitals.

RESULTS

Patients

The details of the patient characteristics are shown in Table 1. A total of 333 patients [(176 male, 157 female; median age: 79 years [range: 72–85 years]; body mass index: 20.81 kg/m² (range: 18.28–23.31 kg/m²)] were included. The median Ccr was 39.09 mL/min, which mainly included renal dysfunction. The median digoxin dose was 2.04 µg/kg/d, and the dosing period ranged from 57 to 2302 d. Sixty-seven percent of the included patients had HF with preserved LVEF. In addition, 224 patients with HF had atrial fibrillation as a comorbidity. Patients undergoing hemodialysis were excluded. As shown in Table 2, the proportion of patients with plasma concentration of digoxin ≥0.9 ng/mL was 55.9% (186/333).

Table 1. Summary of Patient Data

Factors	n = 333
Sex (Male/Female)	176/157
Age	79 (72, 85)e)
Height (m)	1.57 (1.48, 1.65)e)
Body weight (kg)	50.6 (42.0, 59.6)e)
Body mass index (kg/m2)	20.81 (18.28, 23.31)e)
Serum creatinine (mg/dL)	1.00 (0.74, 1.36)e)
Ccr (mL/min)	39.09 (27.42, 56.62)e)
Daily dose of digoxin (µg/kg body weight)	2.04 (1.37, 2.68)e)
Period of digoxin treatment (d)	642 (57, 2302)e)
Plasma digoxin concentration (ng/mL)	0.83 (0.52, 1.24)e)
LVEF (%)	59.4 (42.7, 67.0)e)
HFrEFa), n (%)	72 (21.6)
HFmrEFb), n (%)	38 (11.4)
HFpEFc), n (%)	223 (67.0)
Patient medical conditions, nd)
Coronary artery bypass graft	3
Graft replacement	11
Atrial fibrillation	224
Heart valve replacement or formation	44

Ccr, creatinine clearance; HF, heart failure; HFmrEF, heart failure with mid-range ejection fraction; HFpEF, heart failure with preserved ejection fraction; HFrEF, heart failure with reduced ejection fraction; LVEF, left ventricular ejection fraction. a) LVEF < 40%. b) 40% ≦ LVEF < 50%. c) 50% ≦ LVEF. d) Including multiple answers. e) Each value represents the median (25 to 75% percentile).

Table 2. Univariate Analysis of Factors Contributing to Plasma Digoxin Concentration (≥ 0.9 ng/mL) in Patients with HF

Factors	< 0.9 ng/mL	≥ 0.9 ng/mL	p-Value
	n = 147	n = 186
Sex (Male/Female)	92/55	84/102	0.002a)
Age	76.16 ± 12.15d)	79.05 ± 10.99d)	0.040c)
Body mass index (kg/m2)	23.36 ± 3.48d)	23.81 ± 4.08d)	0.341c)
Daily dose of dioxin (µg/kg body weight)	1.89 ± 0.90d)	2.47 ± 1.31d)	< 0.001c)
Ccr (mL/min)	72.05 ± 29.18d)	64.98 ± 29.77d)	< 0.001c)
Number of concomitant drugs	4.45 ± 3.43d)	5.42 ± 3.78d)	0.365c)
Typical inducer of P-gp, n (%)	2 (1.36)	3 (1.62)	1.000b)
Carbamazepine, n	0	1	—
Rifampicin, n	2	2	—
Typical inhibitor of P-gp, n (%)	3 (2.04)	5 (2.69)	1.000b)
Azole antifungals, n	1	0	—
Clarithromycin, n	1	3	—
Protease inhibitors, n	0	1
Unknown, n	1	1	—
LVEF (%)	53.25 ± 16.02d)	56.03 ± 15.79d)	0.138c)

Ccr, creatinine clearance; HF, heart failure; LVEF, left ventricular ejection fraction; P-gp, P-glycoprotein. a) Chi-square test. b) Fisher’s exact test. c) Mann–Whitney U test. d) Each value represents the mean ± standard deviation.

DT Analysis

According to the univariate analysis, sex (p = 0.002), age (p = 0.040), digoxin dose (p < 0.001), Ccr (p < 0.001), and LVEF (p = 0.138) were selected as the explanatory variables (Table 2). As shown in the flowchart (Fig. 2), the tree was divided into three levels, and four groups were extracted. A Ccr ≥32 mL/min was most associated with digoxin toxicity, followed by a divergence in the daily digoxin dose of ≥1.6 µg/kg. When patients had a Ccr <32 mL/min and a daily digoxin dose ≥1.6 µg/kg, their LVEF value branched with a cut-off of 50%. The incidence of digoxin toxicity was lowest in patients with Ccr ≥32 mL/min, while the incidence in those with a daily digoxin dose <1.6 µg/kg was 34.9% (22/63). In contrast, patients with Ccr <32 mL/min, daily dose ≥1.6 µg/kg, and LVEF ≥50% showed a high incidence of digoxin toxicity, at 91.8% (45/49). However, the tree was not branched by the LVEF without Ccr and a daily dosage of digoxin.

Fig. 2. DT Model for Predicting Digoxin Toxicity in Patients with HF

Ccr, creatinine clearance; HF, heart failure; LVEF, left ventricular ejection fraction.

Multivariate Logistic Regression Analysis

The cut-off value for age was 72 years (sensitivity, 0.812; specificity, 0.299; area under the curve, 0.566) based on the ROC curve. In addition, the cut-off values for daily digoxin dose, Ccr, and LVEF calculated by the CHAID algorithm were 1.6 µg/kg, 32 mL/min, and 50%, respectively. Multivariate logistic regression analysis revealed that the adjusted odds ratio for males was 1.72 (95% confidence interval (CI): 1.074–2.760, p = 0.024), for those with a daily dose of digoxin ≥1.6 µg/kg was 2.93 (95% CI: 1.756–4.902, p < 0.001), and for those with a Ccr <32 mL/min was 2.66 (95% CI: 1.530–4.633, p < 0.001) (Table 3). While the C/D ratio of digoxin in patients with Ccr <32 mL/min was significantly higher than those with Ccr ≥32 mL/min (Fig. 3A), there was no difference in the C/D ratio between males and females (Fig. 3B).

Table 3. Factors Influencing the Plasma Digoxin Concentration (≥0.9 ng/mL) in HF Patients Analyzed Using Multivariate Logistic Regression Analysis

Factors	Adjusted OR	95% CI	p-Value
Age ≥72 years	1.09	0.613–1.922	0.778
Male	1.72	1.074–2.760	0.024
Daily dose of digoxin ≥1.6 µg/kg body weight	2.93	1.756–4.902	< 0.001
Ccr <32 mL/min	2.66	1.530–4.633	< 0.001
LVEF ≥50%	1.18	0.714–1.941	0.778

Ccr, creatinine clearance; HF, heart failure; LVEF, left ventricular ejection fraction; OR, odds ratio; 95% CI, 95% coefficient interval.

Fig. 3. Effect of the Creatinine Clearance or Sex on the C/D Ratio of Digoxin in Patients with HF

Ccr, creatinine clearance. (A) Ccr and (B) sex. *** p < 0.001 compared to patients with Ccr ≥32 mL/min. NS, not significant; C/D, concentration-to-dose ratio.

Validation of the DT Model

According to the Hosmer–Lemeshow test, the p-value was determined to be 0.582. The accuracy rates for multivariate logistic regression and DT analyses were 76.9 and 88.2%, respectively. The misclassification risk of the flowchart constructed using the CHAID algorithm was estimated to be 46.2 ± 2.7%.

DISCUSSION

Although the risk factors for digoxin toxicity have been well elucidated, the rate of 30-d readmission for digoxin toxicity is still high (21.5%) in a population-level cohort study,²¹⁾ indicating that the construction of a prediction tool for digoxin toxicity may be urgently needed. Here, we constructed a flowchart using the CHAID algorithm to predict digoxin toxicity in patients with HF.

According to the flowchart used in this study, to reduce the risk of digoxin toxicity when introducing digoxin to patients with HF with severe renal dysfunction (Ccr <32 mL/min), the initial daily dose should be <1.6 µg/kg. Furthermore, if the patient has a Ccr <32 mL/min, daily dose ≥1.6 µg/kg, and LVEF ≥50%, the plasma concentration of digoxin should be monitored early and frequently.

Digoxin clearance has been reported to be correlated with Ccr in patients with HF.^22,23) Using DT analysis, Ccr ≥32 mL/min was selected as the first node (Fig. 2). In addition, the C/D ratio of Ccr <32 mL/min was higher than that of Ccr ≥32 mL/min, suggesting that a decrease in Ccr may be the most important factor that defines the plasma concentration of digoxin in a dose-independent manner.

Patients with a Ccr <32 mL/min, daily dose ≥1.6 µg/kg, and LVEF ≥50% may show an increase in plasma digoxin concentrations. HF with preserved left ventricular ejection fraction, which is defined as having an LVEF ≥50%, has been characterized primarily via left ventricular diastolic failure.²⁴⁾ Left ventricular diastolic dysfunction results in decreased cardiac output and organ perfusion owing to inadequate blood volume in the left ventricle. Renal blood flow was reported to decrease in correlation with decreased cardiac output.²⁵⁾ Our previous study revealed that left ventricular diastolic failure might be involved in hypermagnesemia induced by magnesium oxide administration because magnesium is mainly extracted by the kidney,²⁶⁾ suggesting that left ventricular diastolic failure may decrease the renal clearance of digoxin. However, the tree was not branched by the LVEF without Ccr and a daily dosage of digoxin, which is consistent with the univariate analysis showing no significant difference in LVEF (Table 2). Thus, it was speculated that the effect of LVEF on plasma digoxin concentration is limited to patients with renal dysfunction (Ccr <32 mL/min) and those who received high doses (daily dose ≥1.6 µg/kg) of digoxin. Therefore, the LVEF alone might not accurately predict digoxin toxicity in HF patients. Shimamoto et al.²⁷⁾ reported that a reduced LVEF could decrease the clearance of vancomycin, a drug typically excreted by the kidney. Therefore, further research should be conducted to determine whether left ventricular systolic or diastolic failure mainly affects the clearance of renally excreted drugs, including digoxin.

Digoxin is a substrate of P-gp, which is responsible for its intestinal absorption and urinary extraction.²⁸⁾ It has been reported that clarithromycin inhibits P-gp, resulting in a 1.7-fold increase in the area under the plasma digoxin concentration curve, which is mediated by enhanced intestinal adsorption and reduced non-glomerular renal clearance.²⁹⁾ Moreover, protease inhibitors may also contribute to the elevation of plasma concentrations of digoxin via similar mechanisms.³⁰⁾ Since the rate of the concomitant use of P-gp inhibitors was not significantly different among patients with digoxin levels ≥0.9 and <0.9 ng/mL (Table 2), P-gp inhibition might not affect plasma digoxin concentration in this study.

Multivariate logistic regression analysis was performed to evaluate the accuracy of the DT analysis (Table 3). The accuracy and fitness of the multivariate logistic regression model were good, and three factors (male sex, daily dose ≥1.6 µg/kg, and Ccr <32 mL/min) were identified as risk factors for digoxin toxicity. A previous study revealed no sex differences in the pharmacokinetics of digoxin.³¹⁾ As the C/D ratio in males was equivalent to that in females (Fig. 3B), it is reasonable that males were not selected in the CHAID algorithm. Since its accuracy was comparable to that reported in previous studies,^14,32,33) this DT analysis may be valid.

Our study has few limitations. First, although there was no consensus on the misclassification rates for predicting adverse drug reactions, the rate obtained in this study was higher than that reported in previous studies.^14,32,33) It was speculated that misclassification might be caused by the smaller sample size used in this study compared with previous studies.^14,32,33) Second, because this study was retrospective in nature, there may have been insufficient correction for confounding factors, including the severity of HF, edema, and urine volume. Third, although the timing of blood sampling was based on trough values, if blood was collected to confirm suspected digoxin toxicity, it may include blood samples that are not trough. Fourth, adherence to digoxin treatment could not be evaluated. Fifth, because 67% of the patients with HF eligible in this study had preserved ejection fraction, it is possible that the effect of reduced LVEF on plasma digoxin concentrations was not adequately assessed. Finally, due to the small sample size, we limited the assessment of the influence of P-gp inhibitors, such as clarithromycin and protease inhibitors. Considering these limitations, further prospective studies with larger sample sizes are needed to improve the accuracy of the DT model. Nonetheless, a strength of the present study was that it was a multicenter study; thus, there was a minimal bias for scientific rigor or external validity. If this flowchart is applied to clinical settings, a combination of renal function, dosage, and LVEF can be used to easily assess the risk of digoxin toxicity before its administration to patients with HF. This may lead to the creation of guidelines on the frequency of plasma digoxin concentration monitoring and dosage reduction according to several risks.

CONCLUSION

We successfully created a prediction flowchart for determining digoxin toxicity in patients with HF. As the therapeutic strategy for HF is based on LVEF, medical staff are aware of a patient’s LVEF; Ccr is the most common parameter for dose adjustment in these instances. Although the flowchart created still needs further validation, it is very simple and potentially useful for medical staff in determining the initial dose of digoxin in patients with HF.

Acknowledgments

The present study was funded by Grants from the Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (KAKENHI) (Grant No. 22K15331).

Conflict of Interest

The authors declare no conflict of interests.

REFERENCES

• 1) Ponikowski P, Voors AA, Anker SD, et al. 2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) developed with the special contribution of the Heart Failure Association (HFA) of the ESC. Eur. Heart J., 37, 2129–2200 (2016).
• 2) Tsutsui H, Isobe M, Ito H, et al. JCS 2017/JHFS 2017 guideline on diagnosis and treatment of acute and chronic heart failure-digest version. Circ. J., 83, 2084–2184 (2019).
• 3) Digitalis Investigation Group. The effect of digoxin on mortality and morbidity in patients with heart failure. N. Engl. J. Med., 336, 525–533 (1997).
• 4) Vamos M, Erath JW, Hohnloser SH. Digoxin-associated mortality: a systematic review and meta-analysis of the literature. Eur. Heart J., 36, 1831–1838 (2015).
• 5) Ziff OJ, Lane DA, Samra M, Griffith M, Kirchhof P, Lip GY, Steeds RP, Townend J, Kotecha D. Safety and efficacy of digoxin: systematic review and meta-analysis of observational and controlled trial data. BMJ, 351, h4451 (2015).
• 6) Adams KF Jr, Gheorghiade M, Uretsky BF, Patterson JH, Schwartz TA, Young JB. Clinical benefits of low serum digoxin concentrations in heart failure. J. Am. Coll. Cardiol., 39, 946–953 (2002).
• 7) Muk B, Vámos M, Bógyi P, Szabó B, Dékány M, Vágány D, Majoros Z, Borsányi T, Duray GZ, Kiss RG, Nyolczas N. The impact of serum concentration-guided digoxin therapy on mortality of heart failure patients: a long-term follow-up, propensity-matched cohort study. Clin. Cardiol., 43, 1641–1648 (2020).
• 8) Rathore SS, Curtis JP, Wang Y, Bristow MR, Krumholz HM. Association of serum digoxin concentration and outcomes in patients with heart failure. JAMA, 289, 871–878 (2003).
• 9) Mooradian AD. Digitalis. An update of clinical pharmacokinetics, therapeutic monitoring techniques and treatment recommendations. Clin. Pharmacokinet., 15, 165–179 (1988).
• 10) O’Riordan S, Ouldred E, Brice S, Jackson SH, Swift CG. Serum cystatin C is not a better marker of creatinine or digoxin clearance than serum creatinine. Br. J. Clin. Pharmacol., 53, 398–402 (2002).
• 11) Damman K, Navis G, Voors AA, Asselbergs FW, Smilde TD, Cleland JG, van Veldhuisen DJ, Hillege HL. Worsening renal function and prognosis in heart failure: systematic review and meta-analysis. J. Card. Fail., 13, 599–608 (2007).
• 12) Naafs MA, van der Hoek C, van Duin S, Koorevaar G, Schopman W, Silberbusch J. Decreased renal clearance of digoxin in chronic congestive heart failure. Eur. J. Clin. Pharmacol., 28, 249–252 (1985).
• 13) Kavakiotis I, Tsave O, Salifoglou A, Maglaveras N, Vlahavas I, Chouvarda I. Machine learning and data mining methods in diabetes research. Comput. Struct. Biotechnol. J., 15, 104–116 (2017).
• 14) Miyai T, Imai S, Kashiwagi H, Sato Y, Kadomura S, Yoshida K, Yoshimura E, Teraya T, Tsujimoto T, Kawamoto Y, Itoh T, Ueno H, Goto Y, Takekuma Y, Sugawara M. A risk prediction flowchart of vancomycin-induced acute kidney injury to use when starting vancomycin administration: a multicenter retrospective study. Antibiotics (Basel), 9, 920 (2020).
• 15) Kumai M, Imai S, Kato S, Koyanagi R, Tsuruga K, Yamada T, Takekuma Y, Sugawara M. Construction of a risk prediction model of extended release oxycodone tablet-induced nausea and clarification of predictive factors. Biol. Pharm. Bull., 44, 593–598 (2021).
• 16) Imai S, Kashiwagi H, Sato Y, Miyai T, Sugawara M, Takekuma Y. Factors affecting creatine phosphokinase elevation during daptomycin therapy using a combination of machine learning and conventional methods. Br. J. Clin. Pharmacol., 88, 1211–1222 (2022).
• 17) Iisalo E. Clinical pharmacokinetics of digoxin. Clin. Pharmacokinet., 2, 1–16 (1977).
• 18) Elmeliegy M, Vourvahis M, Guo C, Wang DD. Effect of P-glycoprotein (P-gp) inducers on exposure of P-gp substrates: review of clinical drug-drug interaction studies. Clin. Pharmacokinet., 59, 699–714 (2020).
• 19) Hanke N, Frechen S, Moj D, Britz H, Eissing T, Wendl T, Lehr T. PBPK models for CYP3A4 and P-gp DDI prediction: a modeling network of rifampicin, itraconazole, clarithromycin, midazolam, alfentanil, and digoxin. CPT. Pharmacometrics. Syst. Pharmacol., 7, 647–659 (2018).
• 20) Cockcroft DW, Gault MH. Prediction of creatinine clearance from serum creatinine. Nephron, 16, 31–41 (1976).
• 21) Angraal S, Nuti SV, Masoudi FA, Freeman JV, Murugiah K, Shah ND, Desai NR, Ranasinghe I, Wang Y, Krumholz HM. Digoxin use and associated adverse events among older adults. Am. J. Med., 132, 1191–1198 (2019).
• 22) Konishi H, Shimizu S, Chiba M, Minouchi T, Koida M, Yamaji A. Predictive performance of serum digoxin concentration in patients with congestive heart failure by a hyperbolic model based on creatinine clearance. J. Clin. Pharm. Ther., 27, 257–265 (2002).
• 23) Hirai T, Kasai H, Naganuma M, Hagiwara N, Shiga T. Population pharmacokinetic analysis and dosage recommendations for digoxin in Japanese patients with atrial fibrillation and heart failure using real-world data. BMC Pharmacol. Toxicol., 23, 14 (2022).
• 24) Pieske B, Tschöpe C, de Boer RA, et al. How to diagnose heart failure with preserved ejection fraction: the HFA-PEFF diagnostic algorithm: a consensus recommendation from the Heart Failure Association (HFA) of the European Society of Cardiology (ESC). Eur. Heart J., 40, 3297–3317 (2019).
• 25) Leithe ME, Margorien RD, Hermiller JB, Unverferth DV, Leier CV. Relationship between central hemodynamics and regional blood flow in normal subjects and in patients with congestive heart failure. Circulation, 69, 57–64 (1984).
• 26) Asai Y, Yanagawa T, Yamamoto T, Sato Y. Exploratory study of pharmacists’ monitoring methods based on left ventricular function for hypermagnesemia by magnesium oxide in heart failure. J. Clin. Pharmacol., 63, 48–56 (2023).
• 27) Shimamoto Y, Fukuda T, Tominari S, Fukumoto K, Ueno K, Dong M, Tanaka K, Shirasaka T, Komori K. Decreased vancomycin clearance in patients with congestive heart failure. Eur. J. Clin. Pharmacol., 69, 449–457 (2013).
• 28) Wessler JD, Grip LT, Mendell J, Giugliano RP. The P-glycoprotein transport system and cardiovascular drugs. J. Am. Coll. Cardiol., 61, 2495–2502 (2013).
• 29) Rengelshausen J, Göggelmann C, Burhenne J, Riedel KD, Ludwig J, Weiss J, Mikus G, Walter-Sack I, Haefeli WE. Contribution of increased oral bioavailability and reduced nonglomerular renal clearance of digoxin to the digoxin-clarithromycin interaction. Br. J. Clin. Pharmacol., 56, 32–38 (2003).
• 30) Schmitt C, Kaeser B, Riek M, Bech N, Kreuzer C. Effect of saquinavir/ritonavir on P-glycoprotein activity in healthy volunteers using digoxin as a probe. Int. J. Clin. Pharmacol. Ther., 48, 192–199 (2010).
• 31) Lee LS, Chan LN. Evaluation of a sex-based difference in the pharmacokinetics of digoxin. Pharmacotherapy, 26, 44–50 (2006).
• 32) Melillo P, Izzo R, Orrico A, Scala P, Attanasio M, Mirra M, De Luca N, Pecchia L. Automatic prediction of cardiovascular and cerebrovascular events using heart rate variability analysis. PLOS ONE, 10, e0118504 (2015).
• 33) Nakayama N, Oketani M, Kawamura Y, Inao M, Nagoshi S, Fujiwara K, Tsubouchi H, Mochida S. Algorithm to determine the outcome of patients with acute liver failure: a data-mining analysis using decision trees. J. Gastroenterol., 47, 664–677 (2012).