Prediction of all-cause mortality for chronic kidney disease patients using four models of machine learning.

BACKGROUND: The prediction tools developed from general population data to predict all-cause mortality are not adapted to chronic kidney disease (CKD) patients, because this population displays a higher mortality risk. This study aimed to create a clinical prediction tool with good predictive performance to predict the 2-year all-cause mortality of stage 4 or stage 5 CKD patients. METHODS: The performance of four different models (deep learning, random forest, Bayesian network, logistic regression) to create four prediction tools was compared using a 10-fold cross validation. The model that offered the best performance for predicting mortality in the Photo-Graphe 3 cohort was selected and then optimized using synthetic data and a selected number of explanatory variables. The performance of the optimized prediction tool to correctly predict the 2-year mortality of the patients included in the Photo-Graphe 3 database were then assessed. RESULTS: Prediction tools developed using the Bayesian network and logistic regression tended to have the best performances. Although not significantly different from logistic regression, the prediction tool developed using the Bayesian network was chosen because of its advantages and then optimized. The optimized prediction tool that was developed using synthetic data and the seven variables with the best predictive value (age, erythropoietin-stimulating agent, cardiovascular history, smoking status, 25-hydroxy vitamin D, parathyroid hormone and ferritin levels) had satisfactory internal performance. CONCLUSIONS: A Bayesian network was used to create a seven-variable prediction tool to predict the 2-year all-cause mortality in patients with stage 4-5 CKD. Prior to external validation, the proposed prediction tool can be used at: https://dev.hed.cc/?a=jpfauvel&n=2022-05%20Modele%20Bayesien%2020000%20Mortalite%207%20variables%20Naif%20Zou%20online(1).neta for research purposes.

Prediction of all-cause mortality in haemodialysis patients using a Bayesian network.

BACKGROUND: All-cause mortality in haemodialysis (HD) is high, reaching 15.6% in the first year according to the European Renal Association. METHODS: A new clinical tool to predict all-cause mortality in HD patients is proposed. It uses a post hoc analysis of data from the prospective cohort study Photo-Graph V3. A total of 35 variables related to patient characteristics, laboratory values and treatments were used as predictors of all-cause mortality. The first step was to compare the results obtained using a logistic regression to those obtained by a Bayesian network. The second step aimed to increase the performance of the best prediction model using synthetic data. Finally, a compromise between performance and ergonomics was proposed by reducing the number of variables to be entered in the prediction tool. RESULTS: Among the 9010 HD patients included in the Photo-Graph V3 study, 4915 incident patients with known medical status at 2 years were analysed. All-cause mortality at 2 years was 34.1%. The Bayesian network provided the most reliable prediction. The final optimized models that used 14 variables had areas under the receiver operating characteristic curves of 0.78 ± 0.01, sensitivity of 72 ± 2%, specificity of 69 ± 2%, predictive positive value of 70 ± 1% and negative predictive value of 71 ± 2% for the prediction of all-cause mortality. CONCLUSIONS: Using artificial intelligence methods, a new clinical tool to predict all-cause mortality in incident HD patients is proposed. The latter can be used for research purposes before its external validation at: https://www.hed.cc/? a=twoyearsallcausemortalityhemod&n=2-years%20All-cause%20Mortality%20Hemodialysis.neta.

A decision support tool to find the best cyclosporine dose when switching from intravenous to oral route in pediatric stem cell transplant patients.

PURPOSE: Managing the pharmacokinetic variability of immunosuppressive drugs after pediatric hematopoietic stem cell transplantation (HSCT) is a clinical challenge. Thus, the aim of our study was to design and validate a decision support tool predicting the best first cyclosporine oral dose to give when switching from intravenous route. METHODS: We used 10-years pediatric HSCT patients' dataset from 2008 to 2018. A tree-augmented naïve Bayesian network model (method belonging to artificial intelligence) was built with data from the first eight-years, and validated with data from the last two. RESULTS: The Bayesian network model obtained showed good prediction performances, both after a 10-fold cross-validation and external validation, with respectively an AUC-ROC of 0.89 and 0.86, a percentage of misclassified patients of 28.7% and 35.2%, a true positive rate of 0.71 and 0.65, and a false positive rate of 0.12 and 0.14 respectively. CONCLUSION: The final model allows the prediction of the most likely cyclosporine oral dose to reach the therapeutic target specified by the clinician. The clinical impact of using this model needs to be prospectively warranted. Respecting the decision support tool terms of use is necessary as well as remaining critical about the prediction by confronting it with the clinical context.

Population pharmacokinetics of teicoplanin administered by subcutaneous or intravenous route and simulation of optimal loading dose regimen.

Objectives: To investigate the population pharmacokinetics of teicoplanin in patients treated by the subcutaneous (sc) and/or intravenous (iv) route. Patients and methods: Non-linear mixed-effects modelling described teicoplanin concentrations from 98 patients with infection caused by Gram-positive cocci. Monte Carlo simulations were performed to evaluate the probability of target attainment (PTA) of various dosage regimens. Results: Teicoplanin concentrations were best described by a two-compartment model with clearance predicted by estimated glomerular filtration rate. Estimated absorption rate constant (between-subject variability) was 0.039 h-1 (77%), clearance was 0.305 L/h (28%), central volume was 10.3 L (49%), inter-compartmental clearance was 4.42 L/h (66%) and peripheral volume was 97.4 L (51%). The sc route was associated with lower initial Cmin and AUC (day 3: loading phase) compared with the iv route. This difference appeared to vanish after 14 days, with comparable simulated PTAs based on the Cmin and AUC for all tested dosages (400, 600, 800 and 1000 mg every 12 h). However, a loading dose regimen with five administrations of either 400 or 600 mg was not sufficient to achieve the target Cmin (≥15 mg/L) for both routes. Also, PTAs for higher MIC (≥1.0 mg/L) were poor with all regimens for both routes. Conclusions: This is the first study examining the pharmacokinetic/pharmacodynamic implications of using the sc route for teicoplanin. Subcutaneous administration is associated with lower Cmin and AUC values after the loading phase compared with iv administration. Therefore, iv administration should be preferred in the first few days of therapy. This study also shows that loading doses of teicoplanin higher than currently recommended should be used to improve PTA.

Is Trough Concentration of Vancomycin Predictive of the Area Under the Curve? A Clinical Study in Elderly Patients.

BACKGROUND: Current guidelines suggest that vancomycin trough concentrations (Cmin) between 15 and 20 mg/L should be achieved to optimize vancomycin exposure and effect. The objective of this study was to analyze the correlation between vancomycin Cmin and the area under the concentration-time curve (AUC) and assess the ability to predict an AUC target of 400 mg·h/L based on Cmin. METHODS: A retrospective analysis of vancomycin therapeutic drug monitoring data collected in 95 elderly patients treated with intermittent intravenous vancomycin was performed. For each patient, individual pharmacokinetic parameters of vancomycin and AUC24 were estimated from concentration measurements using a Bayesian approach. The relationship between vancomycin Cmin and AUC was studied using global and local correlation analysis as well as logistic regression with Receiver Operating Characteristic curve analysis. RESULTS: The overall correlation between AUC24 and Cmin was significant but moderate (R = 0.51). When vancomycin Cmin was greater than 15 mg/L, the corresponding AUC24 was >400 mg·h/L in 95% of cases. However, AUC24 values >400 mg·h/L were obtained with Cmin < 15 mg/L in more than 30% of the cases. The logistic regression analysis identified a Cmin value of 10.8 mg/L as the optimal predictor of AUC24 > 400 mg·h/L. CONCLUSIONS: The results of this study indicate that the recommended target range of 15-20 mg/L for vancomycin Cmin seems acceptable for controlling vancomycin exposure, although a value of approximately 11 mg/L appears to be optimal and may be safer.

Pharmacokinetics of Vancomycin in Elderly Patients Aged over 80 Years.

Since the 1950s, vancomycin has remained a reference treatment for severe infections caused by Gram-positive bacteria, including methicillin-resistant Staphylococcus aureus Vancomycin is a nephrotoxic and ototoxic drug mainly eliminated through the kidneys. It has a large interindividual pharmacokinetic variability, which justifies monitoring its plasma concentrations in patients. This is especially important in patients aged over 80 years, who frequently have renal impairment. However, the pharmacokinetics of vancomycin in this population is very poorly described in the literature. The objective of this work was to propose a model able to predict the pharmacokinetics of vancomycin in very elderly people. First, a population pharmacokinetic model was carried out using the algorithm NPAG (nonparametric adaptive grid) on a database of 70 hospitalized patients aged over 80 years and treated with vancomycin. An external validation then was performed on 41 patients, and the predictive capabilities of the model were assessed. The model had two compartments and six parameters. Body weight and creatinine clearance significantly influenced vancomycin volume of distribution and body clearance, respectively. The means (± standard deviations) of vancomycin volume of distribution and clearance were 36.3 ± 15.2 liter and 2.0 ± 0.9 liter/h, respectively. In the validation group, the bias and precision were -0.75 mg/liter and 8.76 mg/liter for population predictions and -0.39 mg/liter and 2.68 mg/liter for individual predictions. In conclusion, a pharmacokinetic model of vancomycin in a very elderly population has been created and validated for predicting plasma concentrations of vancomycin.

Impact of vitamin D supplementation on health-care use in a 25-hydroxyvitamin D-tested population in France: a population-based descriptive cohort study.

OBJECTIVE: Chronic vitamin D deficiency has been associated in some patients with diffuse musculoskeletal pain. These unspecific symptoms may partly explain why vitamin D deficiency is often diagnosed late. Our aim was to analyse health-care claims after vitamin D supplementation in patients likely to have vitamin D deficiency. DESIGN: Ambulatory health-care claims were compared before and after a vitamin D supplementation prescribed following a 25-hydroxyvitamin D assay. SETTING: Health Insurance Fund (FHIF) database of the Rhône-Alpes area, France. SUBJECTS: Among patients reimbursed for a 25-hydroxyvitamin D assay between 1 December 2008 and 31 January 2009, those supplemented with vitamin D after the assay were matched on the date of assay to patients who did not receive vitamin D. RESULTS: Among the 3023 patients who had a 25-hydroxyvitamin D assay, 935 were consequently supplemented and matched to 935 patients not supplemented. Their median age was 50·0 and 49·5 years, respectively. Patients supplemented decreased their muscle relaxant consumption whereas no change was observed in the reference group, the difference between the two groups was significant (P=0·03). Second and third Pain Relief Ladder prescriptions decreased in both groups but not significantly differently between groups (P=0·58). There was a decrease in prescriptions of biological examination in both groups with no significant difference. CONCLUSIONS: Besides a decrease in muscle relaxant prescriptions in the supplemented group, it was difficult to assess the impact of vitamin D supplementation in patients likely to have vitamin D deficiency. Prospective cohort studies and randomized trials are needed to assess the efficiency of screening and supplementing vitamin D deficiency.

External validation of a 2-year all-cause mortality prediction tool developed using machine learning in patients with stage 4-5 chronic kidney disease.

BACKGROUND: Chronic kidney disease (CKD) is associated with increased mortality. Individual mortality prediction could be of interest to improve individual clinical outcomes. Using an independent regional dataset, the aim of the present study was to externally validate the recently published 2-year all-cause mortality prediction tool developed using machine learning. METHODS: A validation dataset of stage 4 or 5 CKD outpatients was used. External validation performance of the prediction tool at the optimal cutoff-point was assessed by the area under the receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, and specificity. A survival analysis was then performed using the Kaplan-Meier method. RESULTS: Data of 527 outpatients with stage 4 or 5 CKD were analyzed. During the 2 years of follow-up, 91 patients died and 436 survived. Compared to the learning dataset, patients in the validation dataset were significantly younger, and the ratio of deceased patients in the validation dataset was significantly lower. The performance of the prediction tool at the optimal cutoff-point was: AUC-ROC = 0.72, accuracy = 63.6%, sensitivity = 72.5%, and specificity = 61.7%. The survival curves of the predicted survived and the predicted deceased groups were significantly different (p < 0.001). CONCLUSION: The 2-year all-cause mortality prediction tool for patients with stage 4 or 5 CKD showed satisfactory discriminatory capacity with emphasis on sensitivity. The proposed prediction tool appears to be of clinical interest for further development.

Performance of estimated glomerular filtration rates to monitor change in renal function in kidney transplant recipients.

BACKGROUND: Glomerular filtration rate estimates (e-GFR) are often used to evaluate the changes in renal function, but have not been validated for this purpose in kidney transplant recipients (KTRs). The aim of this study was to evaluate the validity of e-GFR for monitoring serial changes in renal function in KTR using directly measured GFR by inulin clearance (I-GFR) as the reference standard. METHODS: Performances of inverse serum creatinine (1/creat) and Cockcroft and Gault, Modification of Diet in Renal Disease, and Chronic Kidney Disease Epidemiology Collaboration formulas were assessed to estimate the changes in I-GFR. RESULTS: A total of 1935 I-GFR clearance procedures were performed in 631 KTRs who underwent serial measurements between 2003 and 2009. The baseline median I-GFR were 51.0 mL/min/1.73 m(2) (confidence interval 95%: 23-84 mL/min/1.73 m(2)]. The performances of 1/creat and formulas for detecting the I-GFR variations between two consecutive measurements (n = 1304) were similar. To detect the variations of <20% (increase or decrease), sensitivities ranged between 50 and 56%, and specificities between 64 and 69%. To detect the variations >20% (increase or decrease), sensitivities ranged between 27% and 39%, and specificities between 88 and 97%. Bland-Altman plots confirmed the scattering of values for individual patients. CONCLUSIONS: In a population of Caucasian KTRs, the mean changes in GFR are correctly estimated whatever the formula used in the range of 23-84 mL/min/1.73 m(2) and can thus be applied in population studies. However, in clinical practice, individual changes in GFR evaluated by formulas should be interpreted with caution in KTRs.

Are triplicate urine samples necessary to assess albuminuria?

BACKGROUND/AIMS: Urinary albumin excretion is subject to intra-individual variability. Thus, for research purposes, it is recommended to test three urine samples collected over a short period of time. The objective of our analysis was to check the usefulness of triplicate samples to determine the albuminuric status of diabetic patients. METHODS: We present the results of the non-planned retrospective analysis of 246 triplicate morning urine samples obtained from 95 type 2 diabetics included in three multinational, randomized, double-blind studies. Albuminuria was determined by immunoturbidimetry on fresh samples in the same central laboratory. Microalbuminuria was defined by a urine albumin to creatinine ratio (UACR) between 2.5 and 25 mg/mmol in males and between 3.5 and 35 mg/mmol in females. Concordance was obtained when the second and/or third sample (UACR2 and UACR3) confirmed the albuminuric status obtained from the first sample (UACR1). RESULTS: Considering the first samples, 9% were within the normal range, 35% showed microalbuminuria and 56% showed macroalbuminuria. The overall concordance rate was 95%. The log of UACR was highly correlated between samples. Bland-Altman plots expressed in percent variations between two samples confirmed that the mean variation was low (around 8%) but revealed the scattering of values, 95% being between -60 and +77% of variation between samples. CONCLUSIONS: There is no benefit in repeating morning UACR determination in diabetic patients to accurately categorize a subject as having normo-, micro- or macroalbuminuria. However, in order to accurately quantify albuminuria, repeated determinations are required.

Prediction Tool to Estimate Potassium Diet in Chronic Kidney Disease Patients Developed Using a Machine Learning Tool: The UniverSel Study.

There is a need for a reliable and validated method to estimate dietary potassium intake in chronic kidney disease (CKD) patients to improve prevention of cardiovascular complications. This study aimed to develop a clinical tool to estimate potassium intake using 24-h urinary potassium excretion as a surrogate of dietary potassium intake in this high-risk population. Data of 375 adult CKD-patients routinely collecting their 24-h urine were included to develop a prediction tool to estimate potassium diet. The prediction tool was built from a random sample of 80% of patients and validated on the remaining 20%. The accuracy of the prediction tool to classify potassium diet in the three classes of potassium excretion was 74%. Surprisingly, the variables related to potassium consumption were more related to clinical characteristics and renal pathology than to the potassium content of the ingested food. Artificial intelligence allowed to develop an easy-to-use tool for estimating patients' diets in clinical practice. After external validation, this tool could be extended to all CKD-patients for a better clinical and therapeutic management for the prevention of cardiovascular complications.

The best way to detect elevated albuminuria.

BACKGROUND: It has been reported that first morning specimens are more reliable than random spot specimens to assess 24-hour urinary albumin excretion rate (UAER), especially if albuminuria is expressed as albuminuria to creatininuria ratio. We aimed to investigate the influence of (a) posture and activity and (b) the units to best estimate 24-hour albuminuria. METHODS: In this retrospective study, 24-hour UAER was compared to 60 min 'supine' and 90 min 'activity' albuminuria in 124 patients tested for resistant hypertension. The ability to adjust urinary albumin concentration (UAC) to creatininuria (ACR) or to collection duration (tAER) values in order to increase the reliability of albuminuria values was also analyzed. RESULTS: Compared to 24-hour UAER, UAC (mg/l), tAER (µg/min) and ACR (mg/mmol) during the supine period had a similar concordance rate in normo-, micro- and macroalbuminuric patients. The UAC in the supine period was well related to 24-hour UAER. However, UAC almost doubled during activity. Adjustment to creatininuria improved the correlation between albuminuria during both periods and 24-hour UAER, but mainly during the activity period. CONCLUSIONS: Our results confirm that UAC is dependent on physical activity. Correction of UAC by creatininuria (ACR) provides a satisfactory estimation of 24-hour UAER. Thus, for practical reasons, it is advisable to use ACR, where no differences appear to exist, whether a supine urine sample or an activity urine sample is obtained.

A Clinical Decision Support Tool to Find the Best Initial Intravenous Cyclosporine Regimen in Pediatric Hematopoietic Stem Cell Transplantation.

To optimize cyclosporine A (CsA) dosing regimen in pediatric patients undergoing hematopoietic stem cell transplantation (HSCT), we aimed to provide clinicians with a validated decision support tool for determining the most suitable first dose of intravenous CsA. We used a 10-year monocentric data set of pediatric patients undergoing HSCT. Discretization of all variables was performed according to literature or thanks to algorithms using Shannon entropy (from information theory) or equal width intervals. The first 8 years were used to build the Bayesian network model. This model underwent a 10-fold cross-validation, and then a prospective validation with data of the last 2 years. There were 3.3% and 4.1% of missing values in the training and the validation data set, respectively. After prospective validation, the Tree-Augmented Naïve Bayesian network shows interesting prediction performances with an average area under the receiver operating characteristic curve of 0.804, 32.8% of misclassified patients, a true-positive rate of 0.672, and a false-positive rate of 0.285. This validated model allows good predictions to propose an optimized and personalized initial CsA dose for pediatric patients undergoing HSCT. The clinical impact of its use should be further evaluated.

Intra-individual Pharmacokinetic Variability of Intravenous Busulfan in Hematopoietic Stem Cell-Transplanted Children.

BACKGROUND: Busulfan therapeutic drug monitoring (TDM) is necessary to better achieve the target exposure in children before hematopoietic stem cell transplantation (HSCT). However, TDM-based dosing may be challenging if intra-individual pharmacokinetic variability (also denoted inter-occasion variability [IOV]) occurs during therapy. OBJECTIVES: The objectives of this study were to describe and quantify busulfan IOV in children, and to investigate its potential determinants. METHODS: We performed a new analysis of published data from children who received intravenous busulfan over 4 days before HSCT. We calculated individual pharmacokinetic parameters on each day of therapy using a published population pharmacokinetic model of busulfan and analyzed their changes. Population estimation of IOV was also performed with non-linear mixed effects (NLME) modeling. Potential predictors of significant decrease in busulfan clearance (CL) were assessed by using machine learning approaches. RESULTS: IOV could be assessed in 136 children. Between day (D) 1 and D2, most patients (80%) experienced a decrease in busulfan CL, with a median change of - 7.9%. However, both large decreases (minimum, - 48.5%) and increases in CL (maximum, + 44%) were observed. Over D1-D3 of therapy, mean CL significantly decreased (- 15%), with a decrease of ≥ 20% in 22% of patients. Some patients also showed unstable CL from day to day. NLME modeling of IOV provided a coefficient of variation of 10.6% and 13.1% for volume of distribution (Vd) and CL, respectively. Some determinants of significant decreases in busulfan CL were identified, but predictive performance of the models was limited. CONCLUSIONS: Significant busulfan intra-individual variability may occur in children who receive a HSCT and is hardly predictable. The main risk is busulfan overexposure. Performing TDM repeatedly over therapy appears to be the best way to accurately estimate busulfan exposure and perform precision dosing.

Blood pressure control in patients with chronic kidney disease according to office and home blood pressures.

OBJECTIVES: The aim of this study was to assess blood pressure (BP) control in patients with chronic kidney disease (CKD) according to office and home BP and to assess the prevalence of normal BP, white-coat uncontrolled hypertension (WUCH), masked uncontrolled hypertension (MUCH) and elevated BP. METHODS: Patients with renal failure with or without proteinuria were included in this multicenter observational study. Office BP was first measured by the physician using a self-monitoring BP device (three automatic readings), then by the patient at home (morning and evening) over 3 consecutive days. WUCH was defined as a systolic BP (SBP)/diastolic BP (DBP) ≥140/90 mmHg in the clinic and SBP/DBP<135/85 mmHg at home. MUCH was defined as SBP/DBP <140/90 mmHg in the clinic and SBP/DBP ≥135/85 mmHg at home. RESULTS: Among the 243 included subjects, data of 225 patients were analyzed. Mean estimated glomerular filtration rate was 37.7 ± 15.7 mL/min/1.73 m and mean office SBP/DBP was 154 ± 19/83 ± 13 mmHg. Mean office SBP/DBP was significantly higher than home SBP/DBP (+9.0 ± 15.1/+7.0 ± 10.0 mmHg, P < 0.01). Normal BP (office and home BP), WUCH, MUCH and elevated BP (office and home BP) rates were 12.0, 14.2, 6.7 and 67.1%, respectively. The patients were taking, on average, 2.8 ± 1.5 antihypertensive drugs/day. CONCLUSION: BP control in patients with CKD was poor. Routine use of 'out-of-office' BP measurement, in addition to office BP by which we can identify patients with WUCH or MUCH, should be recommended based on the current findings.

[Amikacin pharmacokinetics in adults: a variability that question the dose calculation based on weight]. / Pharmacocinétique de l'amikacine chez l'adulte : une hétérogénéité qui remet en cause le calcul de la dose basé sur le poids.

The use of amikacin is difficult because of its toxicity and its pharmacokinetic variability. This variability is almost ignored in adult standard dosage regimens since only the weight is used in the dose calculation. Our objective is to test if the pharmacokinetic of amikacin can be regarded as homogenous, and if the method for calculating the dose according to patients' weight is appropriate. From a cohort of 580 patients, five groups of patients were created by statistical data partitioning. A population pharmacokinetic analysis was performed in each group. The adult population is not homogeneous in term of pharmacokinetics. The doses required to achieve a maximum concentration of 60 mg/L are strongly different (585 to 1507 mg) between groups. The exclusive use of the weight to calculate the dose of amikacine appears inappropriate for 80% of the patients, showing the limits of the formulae for calculating doses of aminoglycosides.

[Not Available]. / Pharmacocinétique de l'amikacine chez l'adulte : une hétérogénéité qui remet en cause le calcul de la dose basé sur le poids.

The use of amikacin is difficult because of its toxicity and its pharmacokinetic variability. This variability is almost ignored in adult standard dosage regimens since only the weight is used in the dose calculation. Our objective is to test if the pharmacokinetic of amikacin can be regarded as homogenous, and if the method for calculating the dose according to patients' weight is appropriate. From a cohort of 580 patients, five groups of patients were created by statistical data partitioning. A population pharmacokinetic analysis was performed in each group. The adult population is not homogeneous in term of pharmacokinetics. The doses required to achieve a maximum concentration of 60mg/L are strongly different (585 to 1507mg) between groups. The exclusive use of the weight to calculate the dose of amikacine appears inappropriate for 80% of the patients, showing the limits of the formulae for calculating doses of aminoglycosides.

Variables linked to hepatitis B vaccination success in non-dialyzed chronic kidney disease patients: Use of a bayesian model.

BACKGROUND: Hepatitis B vaccination is recommended for chronic kidney disease (CKD) patients before starting dialysis. We performed an analyis aimed to describe the clinical and biological parameters related to the success of vaccination in CKD patients before starting dialysis. METHODS: We extracted data of 170 non-dialyzed patients who were offered hepatitis B vaccination from a register. They received a first vaccination of 40µg followed by boosters after one, two and six months. Patients were considered protected if their hepatitis B antibody level was >10IU/L, three months apart. A logistic regression and a Bayesian model were used to describe the relationships between variables and the success of vaccination. RESULTS: Vaccination protected 50.6% of the patients. Model adjustment to the data was higher using the Bayesian model compared to the logistic regression (with area under the ROC curve of 0.955±0.007 vs 0.775±0.066 respectively). The Bayesian model's robustness studied using a 10 fold cross validation showed a percentage of misclassified subjects of 12.4±1.8%, a sensitivity of 87.7±0.3%, a specificity of 87.5±0.3%, a positive predictive value of 87.8±0.3% and negative predictive value of 87.4±0.2%. As classified by the Bayesian model, the variables most related to successful vaccination were, in descending order: age, eGFR, protidemia, albuminemia, cause of renal failure, gender, previous vaccination and weight. CONCLUSION: The Bayesian network confirmed that both kidney function and nutritional status of patients are important factors to explain the success of vaccination against hepatitis B in CKD patients before dialysis. For research purposes, before an external validation, the network can be used online at www.hed.cc/?s=Bhepatitis&n=ReseauhepatiteBsup10.neta.

Prediction of docetaxel toxicity in older cancer patients: a Bayesian network approach.

Chemotherapy is an essential therapy in the fight against cancer. Polypathology and polymedication are often encountered in elderly patients, making this population especially at risk for adverse drug reactions, and particularly with cytotoxic drugs. The objective of this study was to build a model to predict high-grade toxicity in elderly patients treated with docetaxel. Data from the trial TAX-108 have been used to create the model. The variable to predict was the occurrence of grade 3 or 4 toxicity. The explanatory variables entered in the model were anthropometric and biological characteristics of patients at inclusion; fragility criteria (SMAF, CIRS-G, performance status); location of the primary tumor; chemotherapy history, radiotherapy or surgery; weekly dose of docetaxel, cumulative dose administered. A Bayesian network model was developed using a global search procedure and an Expectation-Maximization algorithm. A 10-fold cross-validation was performed. A toxicity of grade 3 or higher was observed in 54% of patients. The variables providing the most information were the primary site (19.4%), the dose per course (17.5%), and albuminemia (13.1%). The area under the curve of the model obtained after cross-validation was 74 ± 1.4%. The model built allows classifying correctly 71.21 ± 0.9% of patients in our sample in the cross-validation procedure. The sensitivity and specificity of the model were 75 and 67%, respectively, and the positive and negative predictive values were 73 and 69%. The encouraging results from this first study show that Bayesian networks could help assess the benefit-risk ratio of chemotherapy in elderly patients.

Bayesian Networks: A New Approach to Predict Therapeutic Range Achievement of Initial Cyclosporine Blood Concentration After Pediatric Hematopoietic Stem Cell Transplantation.

BACKGROUND: Pediatric hematopoietic stem cell transplantation (HSCT) allows the treatment of numerous diseases, both malignant and non-malignant. Cyclosporine, a narrow therapeutic index drug, is the major immunosuppressant used to prevent graft-versus-host disease (GVHD), but may also cause severe adverse effects in case of overdosing. OBJECTIVE: The objective of this study is to predict the initial cyclosporine residual blood concentration value after pediatric HSCT, and consequently the dose necessary to reach the therapeutic range, using a mathematical individual predictive model. METHODS: Clinical and biological data collected from the graft infusion for 2 months after transplantation in 155 pediatric patients undergoing HSCT between 2008 and 2016 were used to generate synthetic data for 1000 subjects which were used to build a Bayesian network model. We compared the characteristics and sensitivity to clinical or biological missing data of this model with four other methods. RESULTS: The tree-augmented Naïve Bayesian network showed the best characteristics, with no missing data (area under the curve of the receiving operator characteristics curve [AUC-ROC] of 0.89 ± 0.02), 18.9 ± 2.6% of patients misclassified, and positive and negative predictive values of 85.9 ± 3.4% and 74.2 ± 5.1%, respectively, and this trend is found in the synthetic dataset from no to 10% missing data. The most relevant variables that could influence whether the initial residual cyclosporine concentration is in the therapeutic range are the last dose before measurement and the mean dose before measurement. CONCLUSIONS: We developed and cross-validated an online Bayesian network to predict the first cyclosporine concentration after pediatric HSCT. This model allows simulation of different dosing regimens, and enables the best dosing regimen to reach the therapeutic range immediately after transplantation to be found, minimizing the risk of adverse effects and GVHD occurrence.