AIMS:Pharmacokinetic interaction studies typically focus on oral administration, but intravenous (IV) administration bypasses intestinal degradation and hepatic first-pass metabolism, leading to distinct drug-drug interaction (DDI) magnitude. This study aimed to develop a predictive model for DDIs involving IV-administered drugs. METHODS:Drugs metabolized by the five major cytochrome P450 enzymes were analysed. Data from IV DDI studies, including area under the concentration-time curve (AUC) ratios of object drugs with and without precipitants, were collected. A Bayesian approach estimated IV contribution ratios (CR) from published oral CR values. Predictions were validated by assessing whether predicted-to-observed ratios fell within 50-200% of observed values. RESULTS:Data for 33 drugs and 87 AUC ratios involving 26 precipitants were analysed. In the training set, 95% of AUC ratios were within the acceptability range, with a mean bias of 0.09 mg·h/L and 16% imprecision. In the validation set, 85% of predictions were within range, with a bias of -0.012 mg·h/L and 27.3% imprecision. Mean AUC ratios for inhibitors were 2.28 (oral) and 1.78 (IV). For inducers, mean AUC ratios were 0.209 (oral) and 0.62 (IV). CONCLUSIONS:Despite the limited dataset, the model demonstrated robust performance with 85% of predictions validated. It expands the DDI-predictor framework for quantitative prediction of cytochrome-mediated drug-drug interactions affecting drug exposure to intravenously administered object drugs, thereby providing a tool to predict and understand IV-specific DDIs effectively.

INTRODUCTION:Tramadol is widely used in post-operative pain management. Concentrations of tramadol and its active metabolite O-demethyl tramadol (O-dT) are influenced by cytochrome P450 2D6 (CYP2D6) genotype. OBJECTIVES:The objectives of this study were to investigate the relationships between CYP2D6 phenotype and tramadol pharmacokinetics/ pharmacodynamics (PK/PD) in real life setting. METHODS:CYTRAM was a prospective multicenter study conducted in patients who received post-operative intravenous (IV) tramadol. Patients underwent CYP2D6 genotyping and phenotyping as well as tramadol and O-dT concentration measurements at 24 h and 48 h of therapy. A PK model was used to estimate tramadol and O-dT area under the concentration-time curve (AUC). Pain rating was performed at 24 h and 48 h and evaluated by Numeric Rating Scale (NRS). Patients with NRS < 4 were classified as responders. Variables associated with analgesic response were examined. RESULTS:PK data from 285 patients and pain scores from 273 and 204 patients were available at 24 h and 48 h, respectively. Mean AUC of O-dT was two-fold lower in CYP2D6 poor metabolizers (PM) compared with subjects with other phenotypes (non-PM). At 48 h, pain scores were significantly higher in PM than in non-PM patients. In addition, PM subjects exhibited a significantly lower rate of response (50% versus 78%) and a higher rate of morphine use (47% versus 17%) at 48 h compared with non-PM subjects. However, CYP2D6 phenotype was not an independent predictor of response in the entire population. CONCLUSION:Our results support current pharmacogenetic guidelines and suggest avoiding tramadol for postoperative management in patients with known CYP2D6 PM phenotype. REGISTRATION:ClinicalTrials.gov (NCT00952159).

We investigated whether machine learning (ML) could enhance exposure and survival prediction, refine target exposures and customise the initial dosage of high-dose (HD) busulfan in patients undergoing a conditioning regimen for bone marrow transplantation. Using retrospective data from 71 adult patients with HD busulfan, ML models were developed to predict drug clearance, area under the curve (AUC), optimal starting dose and one-year survival. Model’s performance was evaluated using the area under the curve (AUC)-ROC, the area under the curve (AUC)-PR, sensitivity, specificity, RMSE, and SHAP value interpretation. Random Forest models integrating albuminemia, GFR, bilirubin, cumulative AUC, and peak concentrations improved exposure prediction over pharmacokinetic modelling (AUC-ROC: 0.80 vs. 0.56). Patients with low albumin levels and Busulfan AUC above 66 mg/L·h had shorter survival times, whereas other could tolerate exposure levels of up to 90.5 mg/L·h. A busulfan clearance model was developed to estimate the AUC prior to the initial dose, and this could outperform standard mg/kg dosing. ML-guided dosing would have reduced the dose for patients overexposed with mg/kg dosing and conversely increased starting dose for those underexposed with standard dosing. Therefore, implementing such ML models to support PK-guided dosing could improve both target exposure and dosing with HD Busulfan.

ABSTRACT In our institution, therapeutic drug monitoring of daptomycin is performed routinely and cases of high trough concentrations have been observed in patients without known risk factors. The aim of this study was to identify risk factors of daptomycin overexposure. We performed a case-control study of daptomycin overexposure in patients who received daptomycin between 2013 and 2021. Cases and controls were defined as patients with trough concentration (Cmin) ≥60 mg/L and Cmin <60 mg/L, respectively. Univariate and multivariate analyses were performed with logistic regression models. Retained variables were further analyzed by subgroup analysis and comparison of the pharmacokinetic parameters of daptomycin. We analyzed data from 78 and 26 patients in the control and case groups, respectively. The male-to-female ratio was 1.5. The median (interquartile range) of age, body weight, and creatinine clearance was 66.5 (55–77) years, 77 (65–96) kg, and 98.5 (53–124) mL/min, respectively. Increasing body mass index (BMI) and co-administration of irbesartan were identified as risk factors of daptomycin overexposure with odds ratio (OR) (95% confidence interval [CI]) of 2.9 [1.4–6.2], and 6.1 [1.1–40.8], respectively, whereas increasing creatinine clearance was associated with decreasing risk, with OR of 0.16 [0.05–0.35]. The influence of BMI was attributed to the non-linear relationship between body weight and daptomycin PK parameters and the use of weight-based dosing in patients with high BMI. In addition to renal impairment, high BMI and irbesartan co-administration may be associated with an augmented risk of daptomycin overexposure. Dosing based on actual body weight should be avoided in obese patients.

This study investigates the pharmacokinetic variability and exposure-response relationships of Venetoclax (VEN) in adult patients with acute myeloid leukemia (AML) who are ineligible for intensive chemotherapy. The study was conducted in a real-world clinical setting and included 48 patients who were treated with VEN in combination with azacytidine. We found significant inter-individual variability of 68% and intra-individual variability of 39% in plasma VEN concentrations. In addition, higher VEN concentrations were associated with better hematological responses. Median overall survival for the entire cohort was 17.8 months, with 1- and 2-year survival rates of 51% and 36.4%, respectively. A comparison of the 14-day VEN and 28-day VEN protocols showed that patients benefited from longer treatment durations, which resulted in more courses being administered. Plasma concentrations in the 14-day VEN protocol were higher than in the 28-day VEN protocol (2330 + 1675 ng/mL vs. 1503 + 966 ng/mL, respectively) without an increase in toxicities. The optimal protocol would be 400 mg/14d if we consider survival as a function of the 1818ng/mL cutoff. These results highlight the importance of considering therapeutic drug monitoring (TDM) as a strategy to optimize treatment outcomes by balancing efficacy and safety.

Monitoring of drug use in athletes is of interest both for health and competition-related issues. Considering the advantages of Dried Blood Sampling (low invasiveness, easy sampling, long term storage), we have validated a quantitative LC-MS/HRMS method for the screening of 16 nonsteroidal anti-inflammatory drugs. For all drugs, accuracy and imprecision were within 15% for the 3 levels of quality control and lower than 20% for the lower limit of quantification. Application was performed from samples obtained for Ultra-Trail du Mont-Blanc® 2021 and 2022. A focus on ibuprofen and its metabolites (hydroxyibuprofen, carboxyibuprofen, ibuprofen glucuronide and hydroxyibuprofen glucuronide) was made because the results showed that it was the most detected nonsteroidal anti-inflammatory drug. Further, an interpretation of the ibuprofen concentrations was proposed either from experimental data obtained after an intake of ibuprofen by 10 control subjects, or from a pharmacokinetic modelling and simulations. Depending on the analytical performances of the method, we proposed possible detection windows for ibuprofen in runners. The pharmacokinetic model made it possible to consider two scenarios with and without modification of the total clearance of ibuprofen linked to a modification of the pharmacokinetics of the drugs due to the practice of a long and intense physical activity.

Background The indications of daptomycin have been extended to off-label indications including prosthesis-related infection, and bone and joint infection (BJI). However, efficacy and safety have not been thoroughly demonstrated compared with the standard of care. This systematic review and meta-analysis aimed to compare the treatment effect of daptomycin and glycopeptides for complicated infections.Materials and methods MEDLINE, Embase and Web of Science were searched for randomized controlled trials (RCTs) comparing daptomycin and standard of care for Gram-positive infections, published until 30 June 2021. The primary outcome was defined as all-cause mortality. Secondary outcomes were clinical and microbiological success. The main safety outcome was any severe adverse event (SAE) (grade >= 3).Results Overall, eight RCTs were included in the meta-analysis, totalling 1095 patients. Six (75%) were in complicated skin and soft-structure infections, one (12.5%) in bacteraemia and one (12.5%) in a BJI setting. Six RCTs used vancomycin as a comparator and two used either vancomycin or teicoplanin. All-cause mortality and clinical cure were not different between groups. The microbiological cure rate was superior in patients who received daptomycin [risk ratio (RR) = 1.17 (95% CI: 1.01-1.35)]. The risk of SAEs [RR = 0.57 (95% CI: 0.36-0.90)] was lower in the daptomycin arm.Conclusions While daptomycin is associated with a significantly lower risk of SAEs and a better microbiological eradication, substantial uncertainty remains about the best treatment strategy in the absence of good-quality evidence, especially in bacteraemia and endocarditis where further RCTs should be conducted.

Journal Article Corrected proof Reply to Tannous et al. Get access Romain Garreau, Romain Garreau Groupement Hospitalier Nord, Service de Pharmacie, Hospices Civils de Lyon, Lyon, FranceLBBE—Laboratoire de Biométrie et Biologie Evolutive, CNRS, UMR 5558, Université Lyon 1, Villeurbanne, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Truong-Thanh Pham, Truong-Thanh Pham Division of Infectious Diseases, Geneva University Hospitals, Geneva, SwitzerlandGroupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service des Maladies Infectieuses et Tropicales, Centre de Référence pour la prise en charge des Infections Ostéo-Articulaires complexes (CRIOAc Lyon), Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Laurent Bourguignon, Laurent Bourguignon Groupement Hospitalier Nord, Service de Pharmacie, Hospices Civils de Lyon, Lyon, FranceLBBE—Laboratoire de Biométrie et Biologie Evolutive, CNRS, UMR 5558, Université Lyon 1, Villeurbanne, FranceISPB, Facultés de Médecine et de Pharmacie de Lyon, University of Lyon, Université Lyon 1, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Aurélien Millet, Aurélien Millet Groupement Hospitalier Sud, Service de Biochimie et Biologie Moléculaire, UM Pharmacologie - Toxicologie, Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar François Parant, François Parant Groupement Hospitalier Sud, Service de Biochimie et Biologie Moléculaire, UM Pharmacologie - Toxicologie, Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar David Bussy, David Bussy Groupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service des Maladies Infectieuses et Tropicales, Centre de Référence pour la prise en charge des Infections Ostéo-Articulaires complexes (CRIOAc Lyon), Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Marine Desevre, Marine Desevre Groupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service des Maladies Infectieuses et Tropicales, Centre de Référence pour la prise en charge des Infections Ostéo-Articulaires complexes (CRIOAc Lyon), Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Victor Franchi, Victor Franchi Groupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service des Maladies Infectieuses et Tropicales, Centre de Référence pour la prise en charge des Infections Ostéo-Articulaires complexes (CRIOAc Lyon), Hospices Civils de Lyon, Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Tristan Ferry, Tristan Ferry Groupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service des Maladies Infectieuses et Tropicales, Centre de Référence pour la prise en charge des Infections Ostéo-Articulaires complexes (CRIOAc Lyon), Hospices Civils de Lyon, Lyon, FranceISPB, Facultés de Médecine et de Pharmacie de Lyon, University of Lyon, Université Lyon 1, Lyon, FranceCIRI—Centre International de Recherche en Infectiologie, Inserm, U1111, CNRS, UMR5308, Ecole Normale Supérieure de Lyon, Université́ Claude Bernard Lyon 1, University of Lyon Lyon, France Search for other works by this author on: Oxford Academic PubMed Google Scholar Sylvain Goutelle Sylvain Goutelle Groupement Hospitalier Nord, Service de Pharmacie, Hospices Civils de Lyon, Lyon, FranceLBBE—Laboratoire de Biométrie et Biologie Evolutive, CNRS, UMR 5558, Université Lyon 1, Villeurbanne, FranceISPB, Facultés de Médecine et de Pharmacie de Lyon, University of Lyon, Université Lyon 1, Lyon, France Correspondence: S. Goutelle, Groupement Hospitalier Nord, Hôpital de la Croix-Rousse, Service de pharmacie, Hospices Civils de Lyon, 104 grande rue de la Croix-Rousse, 69004 Lyon, France (sylvain.goutelle@chu-lyon.fr). https://orcid.org/0000-0002-1853-2932 Search for other works by this author on: Oxford Academic PubMed Google Scholar Clinical Infectious Diseases, ciae018, https://doi.org/10.1093/cid/ciae018 Published: 17 January 2024 Article history Received: 10 January 2024 Editorial decision: 12 January 2024 Published: 17 January 2024 Corrected and typeset: 07 February 2024

BACKGROUND AND OBJECTIVE:Chronic kidney disease (CKD) may alter drug renal elimination but is also known for interacting with hepatic metabolism via multiple uremic components. However, few global models, considering the five major cytochromes, have been published, and none specifically address the decrease in cytochrome P450 (CYP450) activity. The aim of our study was to estimate the possibility of quantifying residual cytochrome activity as a function of filtration rate, according to the data available in the literature.METHODS:For each drug in the DDI-predictor database, we collected available pharmacokinetic data comparing drug exposition in the healthy patient and in various stages of CKD, before building a model capable of predicting the variation of exposure according to the degree of renal damage. We followed an In vivo Mechanistic Static Model (IMSM) approach, previously validated for predicting change in liver clearance. We estimated the remaining fraction parameters at glomerular filtration rate (GFR) = 0 and the alpha value of GFR to 50% impairment for the 5 major cytochromes using a non-linear constrained regression using Matlab software.RESULTS:Thirty-one compounds had usable pharmacokinetic data, with 51 AUC ratios between healthy and renal impaired patients. The remaining CYP3A4 activity was estimated to be 0.4 when CYP2D6, 2C9, 2C19 and 1A2 activity was estimated to be 0.43; 1; 0.73 and 0.7, respectively. The alpha value was estimated to be at 6.62; 25; 9.8; 1.38 and 11.04 for each cytochrome. In comparison with published data, all estimates but one were correctly predicted in the range of 0.5-2.CONCLUSION:Our approach was able to describe the impact of CKD on metabolic elimination. Modelling this process makes it possible to anticipate changes in clearance and drug exposure in CKD patients, with the advantage of greater simplicity than approaches based on physiologically-based pharmacokinetic modelling. However, a precise estimation of the impact of renal failure is not possible with an IMSM approach due to the large variability of the published data, and thus should rely on specific pharmacokinetic modelling for narrow therapeutic margin drugs.

BACKGROUND:High-dose daptomycin is increasingly used in patients with bone and joint infection (BJI). This raises concerns about a higher risk of adverse events (AEs), including daptomycin-induced eosinophilic pneumonia (DIEP) and myotoxicity. We aimed to examine pharmacokinetic and other potential determinants of DIEP and myotoxicity in patients with BJI receiving daptomycin. METHODS:All patients receiving daptomycin for BJI were identified in a prospective cohort study. Cases were matched at a 1:3 ratio, with controls randomly selected from the same cohort. Bayesian estimation of the daptomycin daily area under the concentration-time curve over 24 hours (AUC24h) was performed with the Monolix software based on therapeutic drug monitoring (TDM) data. Demographic and biological data were also collected. Risk factors of AEs were analyzed using Cox proportional hazards model. RESULTS:From 1130 patients followed over 7 years, 9 with DIEP, 26 with myotoxicity, and 106 controls were included in the final analysis. Daptomycin AUC24h, C-reactive protein, and serum protein levels were associated with the risk of AEs. The adjusted hazard ratio of DIEP or myotoxicity was 3.1 (95% confidence interval [CI], 1.48-6.5; P < .001) for daptomycin AUC24h > 939 mg/h/L, 9.8 (95% CI, 3.94-24.5; P < .001) for C-reactive protein > 21.6 mg/L, and 2.4 (95% CI, 1.02-5.65; P = .04) for serum protein <72 g/L. CONCLUSIONS:We identified common determinants of DIEP and myotoxicity in patients with BJI. Because the risk of AEs was associated with daptomycin exposure, daptomycin TDM and model-informed precision dosing may help optimize the efficacy and safety of daptomycin treatment in this setting. A target AUC24h range of 666 to 939 mg/h/L is suggested.

La dalbavancine est désormais une option dans le traitement suppressif des infections ostéo-articulaires (IOA). Cependant, cet usage étant hors-AMM, il n'y a pas de posologie définie dans ce contexte. L'objectif de cette étude est de décrire la mise en oeuvre et les premiers résultats d'une méthode d'adaptation posologique de la dalbavancine basée sur un modèle pharmacocinétique (PK) chez les patients traités pour une IOA. Il s'agit d'une étude monocentrique prospective menée au sein d'un centre de référence des IOA complexes de mars 2022 à janvier 2023. Dans une première étape, un modèle PK de population a été construit à partir de données de la littérature [1-2], puis importé dans le logiciel BestDose. Ce programme a été utilisé par le service de pharmacie pour calculer les posologies de dalbavancine. Les dosages plasmatiques de dalbavancine ont été effectués par chromatographie liquide haute performance couplée à de la spectrométrie de masse par un laboratoire externe de référence. Un ratio AUC/CMI quotidien compris entre 500 et 1500 a été choisi comme cible PK/PD à atteindre pour chaque patient, sur la base de données expérimentales publiées. La qualité d'ajustement du modèle aux données a été évaluée par la comparaison des concentrations prédites et mesurées. Les doses et intervalles posologiques initiaux et personnalisés ont été analysés. La cohorte étudiée comportait 16 patients (5 femmes, 11 hommes) avec un âge médian de 70 ans (min, 30; max, 88). Les valeurs médianes (min, max) de poids et de clairance de la créatinine étaient respectivement de 78 kg (51, 110) et de 86 ml/min (30, 135). Parmi les pathogènes responsables d'IOA, Staphylococcus epidermidis a été retrouvé dans 11 cas sur 16 (69 %). Les CMI étaient souvent basses, ≤ 0,03 mg/L chez 10 patients. La posologie initiale de dalbavancine était de 1500 mg pour tous les patients. Au total, 81 dosages plasmatiques de dalbavancine ont été modélisées. L'ajustement du modèle PK aux concentrations mesurées était très satisfaisant avec une erreur relative moyenne de -6 % et un coefficient de détermination de 0.97 entre prédictions et observations. En moyenne, trois modélisations posologiques par patient ont été réalisées sur la période. A la dernière évaluation, la dose médiane était de 1000 mg (min, 500, max, 1500) et l'intervalle posologique médian de 36 jours (min, 19; max, 96). La modélisation a permis d'allonger l'intervalle posologique de plus de 50% chez 5 patients sur 16. La demi-vie d'élimination estimée variait de 7 à 37 jours selon les patients. Une forte variabilité PK de la dalbavancine a été observée dans cette étude, justifiant une personnalisation des doses pour les traitements prolongés des IOA. Les dosages et la modélisation PK permettent de réduire les doses et/ou d'espacer les administrations de dalbavancine chez certains patients, ce qui devrait réduire les contraintes et le coût du traitement. S.G. Participation à un symposium Correvio en 2019

Abstract Background:To inform rational dosing of antibiotics, traditional Therapeutic Drug Monitoring (TDM) with empiric dose adjustment of has been increasingly supplanted by the use of Model-Informed Precision Dosing (MIPD) software.Our objective was to evaluate a model-informed precision dosing approach specifically designed to individualize empiric tobramycin dosing in adults with cystic fibrosis, and to compare target attainment between both the model-based and the empiric dosing approaches. Methods:The BestDose MIPD-software has been used with a published population pharmacokinetic model of tobramycin. To evaluate the MIPD strategy, we used retrospective data from CF adults treated with tobramycin at our local CF center. Empiric dose adjustments from the clinical staff were examined.Using a simulation-based methodology, individualized tobramycin doses that maximize the probability of attaining a Cmax/MIC ratio of 32 mg/L were retrospectively calculated, and compared with empiric dose adjustments. Results:Overall, 101 CF adults were evaluated. Tobramycin Cmax in patients were low (mean of 25.7 ± 5.6 mg/L). The percentage of patients predicted to achieve a Cmax/MIC of ≥ 32 mg/L was low (12.0 %, mean dose of 8.8 ± 1.3 mg/kg) with empiric dose adjustment. TDM with retrospective PK/PD modelling suggest increasing tobramycin dosing in a much larger number of patients (88.0 %, mean dose of 11.6 ± 2.5 mg/kg). Conclusions:Tobramycin doses empirically corrected by clinicians after TDM were predicted to be still insufficient to achieve the efficacy target in most CF patients. Meanwhile a model-based dosing approach that individualizes the tobramycin dosing led to significantly improved achievement of expected target exposure levels in CF adults. Prospective clinical evaluation is warranted.

The pharmacokinetics of ceftolozane-tazobactam (TOL-TAZ) and ceftazidime-avibactam (CEF-AVI) is influenced by renal function. Application of recommended dosages in patients with renal impairment requires the use of fractions of the full dose, as only one dosage is available for both antibiotics.

Population pharmacokinetic (PK) modeling is a widely used approach to analyze PK data obtained from groups of individuals, in both industry and academic research. The approach can also be used to analyze pharmacodynamic (PD) data and pooled PK/PD data. There are 2 main families of population PK methods: parametric and nonparametric. The objectives of this article are to present an overview of nonparametric methods used in population pharmacokinetic modeling and to explain their specific characteristics to inform scientists and clinicians about their potential value for data analysis, simulation, dosage design, and therapeutic drug monitoring (TDM). Nonparametric methods have several interesting characteristics for population PK analysis, including computation of exact likelihoods, the ability to accommodate parameter probability distributions of any shape (eg, non-Gaussian), and to detect subpopulations and outliers. Nonparametric population methods are also highly relevant for model-based TDM and design of individualized drug dosage regimens. Several algorithms have been developed to estimate model parameter values within an individual and compute that individual's dosage to achieve target drug exposure with maximum precision and accuracy. Nonparametric modeling methods for both population and individual PK analysis are available under user-friendly packages.

The evolution of functional autonomy loss leads to institutionalization of people affected by Alzheimer’s disease (AD), to an alteration of their quality of life and that of their caregivers. To predict loss of functional autonomy could optimize prevention strategies, aids and cost of care. The aim of this study was to develop and to cross-validate a model to predict loss of functional autonomy as assessed by Instrumental Activities of Daily Living (IADL) score. Outpatients with probable AD and with 2 or more visits to the Clinical and Research Memory Centre of the University Hospital were included. Four Tree-Augmented Naïve bayesian networks (6, 12, 18 and 24 months of follow-up) were built. Variables included in the model were demographic data, IADL score, MMSE score, comorbidities, drug prescription (psychotropics and AD-specific drugs). A 10-fold cross-validation was conducted to evaluate robustness of models. The study initially included 485 patients in the prospective cohort. The best performance after 10-fold cross-validation was obtained with the model able to predict loss of functional autonomy at 18 months (area under the curve of the receiving operator characteristic curve = 0.741, 27% of patients misclassified, positive predictive value = 77% and negative predictive value = 73%). The 13 variables used explain 41.6% of the evolution of functional autonomy at 18 months. A high-performing predictive model of AD evolution of functional autonomy was obtained. An external validation is needed to use the model in clinical routine so as to optimize the patient care.