Raphael Veil
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Raphael Veil has a total of 156 co-authors. Here are the top co-authors by number of shared publications:
- Anne-Sophie Jannot
Biomedical Informatics and Public Health Department UPL Univ Paris Nanterre F92000 Nanterre France
2 shared publications - Brigitte Ranque
Service de médecine interne, hôpital européen Georges-Pompidou, université Paris-Cité, Assistance publique-Hôpitaux de Paris, 20, rue Leblanc, 75015 Paris, France
2 shared publications Léo Delbos
AP-HP, Bicêtre Hospital, Epidemiology and Public Health Department, 94270 Le Kremlin Bicêtre, France
2 shared publications- Michel Dupuis
AP-HP, Bicêtre Hospital, Pediatric Imaging Department, 94270 Le Kremlin Bicêtre, France
2 shared publications - Simon Bussy
Modal'X UMR 8001 CNRS Sorbonne University Paris France
2 shared publications A. John Henderson
Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, United Kingdom
1 shared publication
Latest Publications
- Open Access
- 3 April 2026
- Published by Wiley in Liver International
- Vol. 46 (5) , e70596
Echocardiography, Bile Acids and Spleen Stiffness to Rule Out High‐Risk Varices in Children With Biliary Atresia
Background and Aims: Detecting gastroesophageal varices with a high risk of bleeding (HRV) in children with biliary atresia requires endoscopies from an early age to apply primary prophylaxis of bleeding. However, repeated endoscopies are invasive and costly. This study was performed to define tools to rule out HRV, with the objective of reducing the number of endoscopies. Methods: Children with biliary atresia and portal hypertension were prospectively enrolled. Laboratory tests, elastography, echocardiography, and endoscopy were performed. Variables were subjected to principal component analysis, followed by multivariate logistic regression analyses for HRV prediction. The effectiveness of composite models and criteria based on liver and spleen stiffness measurements was evaluated using the missed HRV ratio and the spared endoscopy rate. Results: Among the 100 children enrolled, a higher indexed right atrium area, higher left ventricle end‐diastolic diameter, younger age, higher serum bile acid concentration, and higher spleen stiffness were independently associated with HRV. In children under 3 years old, a model combining indexed right atrium area from echocardiography and serum bile acid concentration achieved a 37% spared endoscopy rate with no missed HRV. For children over 3 years old, spleen stiffness of ≤ 30 kPa was associated with a 41% spared endoscopy rate and no missed HRV. Conclusion: Combining an echocardiography‐based model for children under 3 years old with spleen stiffness criteria for older children may reduce the need for screening endoscopies by 40% in children with biliary atresia and portal hypertension.
- 3 April 2026
- Published by Wiley in Liver International
- Vol. 46 (5) , e70596
Open AccessEchocardiography, Bile Acids and Spleen Stiffness to Rule Out High‐Risk Varices in Children With Biliary Atresia
Background and Aims: Detecting gastroesophageal varices with a high risk of bleeding (HRV) in children with biliary atresia requires endoscopies from an early age to apply primary prophylaxis of bleeding. However, repeated endoscopies are invasive and costly. This study was performed to define tools to rule out HRV, with the objective of reducing the number of endoscopies. Methods: Children with biliary atresia and portal hypertension were prospectively enrolled. Laboratory tests, elastography, echocardiography, and endoscopy were performed. Variables were subjected to principal component analysis, followed by multivariate logistic regression analyses for HRV prediction. The effectiveness of composite models and criteria based on liver and spleen stiffness measurements was evaluated using the missed HRV ratio and the spared endoscopy rate. Results: Among the 100 children enrolled, a higher indexed right atrium area, higher left ventricle end‐diastolic diameter, younger age, higher serum bile acid concentration, and higher spleen stiffness were independently associated with HRV. In children under 3 years old, a model combining indexed right atrium area from echocardiography and serum bile acid concentration achieved a 37% spared endoscopy rate with no missed HRV. For children over 3 years old, spleen stiffness of ≤ 30 kPa was associated with a 41% spared endoscopy rate and no missed HRV. Conclusion: Combining an echocardiography‐based model for children under 3 years old with spleen stiffness criteria for older children may reduce the need for screening endoscopies by 40% in children with biliary atresia and portal hypertension.
- Open Access
The Prognostic Value of Olfactory Dysfunction in Patients with COVID-19: The COVIDORA Study
Background: Among all studies describing COVID-19 clinical features during the first wave of the pandemic, only a few retrospective studies have assessed the correlation between olfac-tory dysfunction (OD) and the evolution of disease severity. The main aim was to assess whether OD is a predictive factor of COVID-19 severity based on the patient’s medical management (outpa-tient care, standard hospital admission, and ICU admission). Methods: A national, prospective, mul-ticenter cohort study was conducted in 20 public hospitals and a public center for COVID-19 screen-ing. During the first wave of the pandemic, from 6 April to 11 May 2020, all patients tested positive for COVID-19 confirmed by RT-PCR underwent two follow-up ENT consultations within 10 days of symptom onset. The main outcome measures were the evolution of medical management (out-patient care, standard hospital admission, and ICU admission) at diagnosis and along the clinical course of COVID-19 disease. Results: Among 481 patients included, the prevalence of OD was 60.7%, and it affected mostly female patients (74.3%) under 65 years old (92.5%), with fewer comor-bidities than patients with normal olfactory function. Here, 99.3% (290/292) of patients with OD presented with non-severe COVID-19 disease. Patients reporting OD were significantly less hospi-talized than the ones managed as outpatients, in either a standard medical unit or an ICU. Conclu-sions: As regards the clinical course of COVID-19 disease, OD could predict a decreased risk of hospitalization during the first wave of the pandemic.
Open AccessThe Prognostic Value of Olfactory Dysfunction in Patients with COVID-19: The COVIDORA Study
Background: Among all studies describing COVID-19 clinical features during the first wave of the pandemic, only a few retrospective studies have assessed the correlation between olfac-tory dysfunction (OD) and the evolution of disease severity. The main aim was to assess whether OD is a predictive factor of COVID-19 severity based on the patient’s medical management (outpa-tient care, standard hospital admission, and ICU admission). Methods: A national, prospective, mul-ticenter cohort study was conducted in 20 public hospitals and a public center for COVID-19 screen-ing. During the first wave of the pandemic, from 6 April to 11 May 2020, all patients tested positive for COVID-19 confirmed by RT-PCR underwent two follow-up ENT consultations within 10 days of symptom onset. The main outcome measures were the evolution of medical management (out-patient care, standard hospital admission, and ICU admission) at diagnosis and along the clinical course of COVID-19 disease. Results: Among 481 patients included, the prevalence of OD was 60.7%, and it affected mostly female patients (74.3%) under 65 years old (92.5%), with fewer comor-bidities than patients with normal olfactory function. Here, 99.3% (290/292) of patients with OD presented with non-severe COVID-19 disease. Patients reporting OD were significantly less hospi-talized than the ones managed as outpatients, in either a standard medical unit or an ICU. Conclu-sions: As regards the clinical course of COVID-19 disease, OD could predict a decreased risk of hospitalization during the first wave of the pandemic.
- 16 October 2023
- Published by Elsevier in Diagnostic and Interventional Imaging
- Vol. 105 (3) , 104-109
External validation of an artificial intelligence solution for the detection of elbow fractures and joint effusions in children
- 16 October 2023
- Published by Elsevier in Diagnostic and Interventional Imaging
- Vol. 105 (3) , 104-109
External validation of an artificial intelligence solution for the detection of elbow fractures and joint effusions in children
- 12 November 2022
- Published by Springer Nature in International Orthopaedics
- Vol. 47 (2) , 467-477
Deconstructing forearm casting task by videos with step-by-step simulation teaching improved performance of medical students: is making working student’s memory work better similar to a process of artificial intelligence or just an improvement of the prefrontal cortex homunculus?
Purpose To compare two teaching methods of a forearm cast in medical students through simulation, the traditional method (Trad) based on a continuous demonstration of the procedure and the task deconstruction method (Decon) with the procedure fragmenting into its constituent parts using videos. Methods During simulation training of the below elbow casting technique, 64 medical students were randomized in two groups. Trad group demonstrated the entire procedure without pausing. Decon group received step-wise teaching with educational videos emphasizing key components of the procedure. Direct and video evaluations were performed immediately after training (day 0) and at six months. Performance in casting was assessed using a 25-item checklist, a seven item global rating scale (GRS Performance), and a one item GRS (GRS Final Product). Results Fifty-two students (Trad n = 24; Decon n = 28) underwent both day zero and six month assessments. At day zero, the Decon group showed higher performance via video evaluation for OSATS (p = 0.035); GRS performance (p < 0.001); GRS final product (p < 0.001), and for GRS performance (p < 0.001) and GRS final product (p = 0.011) via direct evaluation. After six months, performance was decreased in both groups with ultimately no difference in performance between groups via both direct and video evaluation. Having done a rotation in orthopaedic surgery was the only independent factor associated to higher performance. Conclusions The modified video-based version simulation led to a higher performance than the traditional method immediately after the course and could be the preferred method for teaching complex skills.
- 12 November 2022
- Published by Springer Nature in International Orthopaedics
- Vol. 47 (2) , 467-477
Deconstructing forearm casting task by videos with step-by-step simulation teaching improved performance of medical students: is making working student’s memory work better similar to a process of artificial intelligence or just an improvement of the prefrontal cortex homunculus?
Purpose To compare two teaching methods of a forearm cast in medical students through simulation, the traditional method (Trad) based on a continuous demonstration of the procedure and the task deconstruction method (Decon) with the procedure fragmenting into its constituent parts using videos. Methods During simulation training of the below elbow casting technique, 64 medical students were randomized in two groups. Trad group demonstrated the entire procedure without pausing. Decon group received step-wise teaching with educational videos emphasizing key components of the procedure. Direct and video evaluations were performed immediately after training (day 0) and at six months. Performance in casting was assessed using a 25-item checklist, a seven item global rating scale (GRS Performance), and a one item GRS (GRS Final Product). Results Fifty-two students (Trad n = 24; Decon n = 28) underwent both day zero and six month assessments. At day zero, the Decon group showed higher performance via video evaluation for OSATS (p = 0.035); GRS performance (p < 0.001); GRS final product (p < 0.001), and for GRS performance (p < 0.001) and GRS final product (p = 0.011) via direct evaluation. After six months, performance was decreased in both groups with ultimately no difference in performance between groups via both direct and video evaluation. Having done a rotation in orthopaedic surgery was the only independent factor associated to higher performance. Conclusions The modified video-based version simulation led to a higher performance than the traditional method immediately after the course and could be the preferred method for teaching complex skills.
- Open Access
- 18 August 2022
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 11 (16) , 4850
Impact of the 1st Wave of the COVID-19 Pandemic and Lockdown on In Utero Transfer Activity in the Paris Area, France
Background: To assess changes in the number and profile of in utero transfer requests during the first lockdown. Methods: An observational, retrospective, cohort study. All pregnant women, from the Paris area (France), for whom a request for in utero transfer to the transfer unit was made during the first lockdown in France (from 17 March to 10 May 2020) or during a mirror period (years 2016 to 2019) were included. We compared the numbers and proportions of various indications for in utero transfer, the rates of in utero transfer acceptance and the proportion of outborn deliveries. Results: 206 transfer requests were made during the lockdown versus 227, 236, 204 and 228 in 2016, 2017, 2018 and 2019, respectively. The relative proportion of requests for threatened preterm births and for fetal growth restriction decreased from 45% in the mirror period to 37% and from 8 to 3%, respectively. The transfer acceptance rates and outborn deliveries did not differ between time periods. Conclusions: Although a reduction in in utero transfer requests was observed for certain indications, the first lockdown was not associated with a decrease in acceptance rates nor in an increase in outborn births of pregnancies with a high risk of prematurity in the Paris area.
- 18 August 2022
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 11 (16) , 4850
Open AccessImpact of the 1st Wave of the COVID-19 Pandemic and Lockdown on In Utero Transfer Activity in the Paris Area, France
Background: To assess changes in the number and profile of in utero transfer requests during the first lockdown. Methods: An observational, retrospective, cohort study. All pregnant women, from the Paris area (France), for whom a request for in utero transfer to the transfer unit was made during the first lockdown in France (from 17 March to 10 May 2020) or during a mirror period (years 2016 to 2019) were included. We compared the numbers and proportions of various indications for in utero transfer, the rates of in utero transfer acceptance and the proportion of outborn deliveries. Results: 206 transfer requests were made during the lockdown versus 227, 236, 204 and 228 in 2016, 2017, 2018 and 2019, respectively. The relative proportion of requests for threatened preterm births and for fetal growth restriction decreased from 45% in the mirror period to 37% and from 8 to 3%, respectively. The transfer acceptance rates and outborn deliveries did not differ between time periods. Conclusions: Although a reduction in in utero transfer requests was observed for certain indications, the first lockdown was not associated with a decrease in acceptance rates nor in an increase in outborn births of pregnancies with a high risk of prematurity in the Paris area.
- Top Cited Papers
- 1 March 2022
- Published by Elsevier in Diagnostic and Interventional Imaging
- Vol. 103 (3) , 151-159
External validation of a commercially available deep learning algorithm for fracture detection in children
- 1 March 2022
- Published by Elsevier in Diagnostic and Interventional Imaging
- Vol. 103 (3) , 151-159
Top Cited PapersExternal validation of a commercially available deep learning algorithm for fracture detection in children
- Open Access
- 11 April 2020
- Published by Oxford University Press (OUP) in Clinical Infectious Diseases
- Vol. 71 (15) , 825-832
Cluster of Coronavirus Disease 2019 (COVID-19) in the French Alps, February 2020
Background: On 7 February 2020, French Health authorities were informed of a confirmed case of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in an Englishman infected in Singapore who had recently stayed in a chalet in the French Alps. We conducted an investigation to identify secondary cases and interrupt transmission.Methods: We defined as a confirmed case a person linked to the chalet with a positive reverse-transcription polymerase chain reaction sample for SARS-CoV-2.Results: The index case stayed 4 days in the chalet with 10 English tourists and a family of 5 French residents; SARS-CoV-2 was detected in 5 individuals in France, 6 in England (including the index case), and 1 in Spain (overall attack rate in the chalet: 75%). One pediatric case, with picornavirus and influenza A coinfection, visited 3 different schools while symptomatic. One case was asymptomatic, with similar viral load as that of a symptomatic case. Seven days after the first cases were diagnosed, 1 tertiary case was detected in a symptomatic patient with from the chalet a positive endotracheal aspirate; all previous and concurrent nasopharyngeal specimens were negative. Additionally, 172 contacts were monitored; all contacts tested for SARS-CoV-2 (N = 73) were negative.Conclusions: The occurrence in this cluster of 1 asymptomatic case with similar viral load as a symptomatic patient suggests transmission potential of asymptomatic individuals. The fact that an infected child did not transmit the disease despite close interactions within schools suggests potential different transmission dynamics in children. Finally, the dissociation between upper and lower respiratory tract results underscores the need for close monitoring of the clinical evolution of suspected cases of coronavirus disease 2019.
- 11 April 2020
- Published by Oxford University Press (OUP) in Clinical Infectious Diseases
- Vol. 71 (15) , 825-832
Open AccessCluster of Coronavirus Disease 2019 (COVID-19) in the French Alps, February 2020
Background: On 7 February 2020, French Health authorities were informed of a confirmed case of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in an Englishman infected in Singapore who had recently stayed in a chalet in the French Alps. We conducted an investigation to identify secondary cases and interrupt transmission.Methods: We defined as a confirmed case a person linked to the chalet with a positive reverse-transcription polymerase chain reaction sample for SARS-CoV-2.Results: The index case stayed 4 days in the chalet with 10 English tourists and a family of 5 French residents; SARS-CoV-2 was detected in 5 individuals in France, 6 in England (including the index case), and 1 in Spain (overall attack rate in the chalet: 75%). One pediatric case, with picornavirus and influenza A coinfection, visited 3 different schools while symptomatic. One case was asymptomatic, with similar viral load as that of a symptomatic case. Seven days after the first cases were diagnosed, 1 tertiary case was detected in a symptomatic patient with from the chalet a positive endotracheal aspirate; all previous and concurrent nasopharyngeal specimens were negative. Additionally, 172 contacts were monitored; all contacts tested for SARS-CoV-2 (N = 73) were negative.Conclusions: The occurrence in this cluster of 1 asymptomatic case with similar viral load as a symptomatic patient suggests transmission potential of asymptomatic individuals. The fact that an infected child did not transmit the disease despite close interactions within schools suggests potential different transmission dynamics in children. Finally, the dissociation between upper and lower respiratory tract results underscores the need for close monitoring of the clinical evolution of suspected cases of coronavirus disease 2019.
- 15 March 2020
- Published by Wolters Kluwer Health in AIDS
- Vol. 34 (4) , 493-500
Virological and immunological impact of integrase inhibitor-based regimens initiated during primary HIV-1 infection
Current international guidelines recommend either boosted protease inhibitor (PI/r)–based or integrase inhibitors (INSTI)–based regimens during primary HIV infection (PHI), even though the latter have only demonstrated their superiority at the chronic stage. We compared the effectiveness of INSTI- versus PI/r-based combined antiretroviral therapy (cART) initiated during PHI. This study was conducted among patients who initiated cART between 2013 and 2017, using data from the ANRS-PRIMO cohort and the Dat’AIDS study. Cumulative proportions of patients reaching viral suppression (HIV-1 RNA <50 cp/mL) were calculated using Turnbull's estimator for interval-censored data. CD4 cells and CD4/CD8 ratio increases were estimated using mixed linear models. Results were adjusted for the data source. Among the 712 study patients, 299 received an INSTI-based cART. Patients baseline characteristics were similar between groups. Viral suppression was reached more rapidly in INSTI- versus PI/r-treated patients (p < 0.01), with cumulative proportions of 37% vs 6% at 4 weeks, 77% vs 38% at 12 weeks, 92% vs 79% at 24 weeks and 93% in both groups at 48 weeks. At 4 weeks, INSTI-treated patients had gained on average 40 CD4 cells/μl (p=0.05) over PI/r-treated ones; mean CD4 counts were similar in the two groups at 48 weeks. The CD4/CD8 ratio followed the same pattern. Results were similar when restricted to a comparison between Dolutegravir- vs Darunavir-based cART. Based on this study and available literature, we recommend the use of INSTI-based cART for treatment initiation during PHI, since it leads to faster viral suppression and immune restoration.
- 15 March 2020
- Published by Wolters Kluwer Health in AIDS
- Vol. 34 (4) , 493-500
Virological and immunological impact of integrase inhibitor-based regimens initiated during primary HIV-1 infection
Current international guidelines recommend either boosted protease inhibitor (PI/r)–based or integrase inhibitors (INSTI)–based regimens during primary HIV infection (PHI), even though the latter have only demonstrated their superiority at the chronic stage. We compared the effectiveness of INSTI- versus PI/r-based combined antiretroviral therapy (cART) initiated during PHI. This study was conducted among patients who initiated cART between 2013 and 2017, using data from the ANRS-PRIMO cohort and the Dat’AIDS study. Cumulative proportions of patients reaching viral suppression (HIV-1 RNA <50 cp/mL) were calculated using Turnbull's estimator for interval-censored data. CD4 cells and CD4/CD8 ratio increases were estimated using mixed linear models. Results were adjusted for the data source. Among the 712 study patients, 299 received an INSTI-based cART. Patients baseline characteristics were similar between groups. Viral suppression was reached more rapidly in INSTI- versus PI/r-treated patients (p < 0.01), with cumulative proportions of 37% vs 6% at 4 weeks, 77% vs 38% at 12 weeks, 92% vs 79% at 24 weeks and 93% in both groups at 48 weeks. At 4 weeks, INSTI-treated patients had gained on average 40 CD4 cells/μl (p=0.05) over PI/r-treated ones; mean CD4 counts were similar in the two groups at 48 weeks. The CD4/CD8 ratio followed the same pattern. Results were similar when restricted to a comparison between Dolutegravir- vs Darunavir-based cART. Based on this study and available literature, we recommend the use of INSTI-based cART for treatment initiation during PHI, since it leads to faster viral suppression and immune restoration.
- Open Access
- 19 September 2019
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 8 (9) , 1502
Trajectories of Biological Values and Vital Parameters: An Observational Cohort Study of Adult Patients with Sickle Cell Disease Hospitalized for a Non-Complicated Vaso-Occlusive Crisis
Hospital admission of patients with sickle-cell disease (SCD) presenting with a vaso-occlusive crisis (VOC) can be justified by pain refractory to usual outpatient care and/or the occurrence of a complication. Yet, the trajectories of vital parameters and standard biomarkers throughout a non-complicated VOC has not been established. In this observational cohort study, we describe the course of routine parameters throughout 329 hospital stays for non-complicated VOC. We used a new spline-based approach to study and visualize non-specific time-dependent variables extracted from the hospital clinical data warehouse. We identified distinct trends during the VOC for hemoglobin level, leukocytes count, C-Reactive Protein (CRP) level and temperature. Hemoglobin decreased after admission and rarely returned to steady state levels before discharge. White blood cell counts were elevated at admission before immediately decreasing, whereas eosinophils increased slowly throughout the first five days of the stay. In over 95% of non-complicated VOC-related stays, the CRP value was below 100 mg/L within the first day following admission and above normal after 48 hours, and the temperature was below 38 °C throughout the entire stay. Knowing the typical trajectories of these routine parameters during non-complicated VOC may urge the clinicians to be more vigilant in case of deviation from these patterns.
- 19 September 2019
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 8 (9) , 1502
Open AccessTrajectories of Biological Values and Vital Parameters: An Observational Cohort Study of Adult Patients with Sickle Cell Disease Hospitalized for a Non-Complicated Vaso-Occlusive Crisis
Hospital admission of patients with sickle-cell disease (SCD) presenting with a vaso-occlusive crisis (VOC) can be justified by pain refractory to usual outpatient care and/or the occurrence of a complication. Yet, the trajectories of vital parameters and standard biomarkers throughout a non-complicated VOC has not been established. In this observational cohort study, we describe the course of routine parameters throughout 329 hospital stays for non-complicated VOC. We used a new spline-based approach to study and visualize non-specific time-dependent variables extracted from the hospital clinical data warehouse. We identified distinct trends during the VOC for hemoglobin level, leukocytes count, C-Reactive Protein (CRP) level and temperature. Hemoglobin decreased after admission and rarely returned to steady state levels before discharge. White blood cell counts were elevated at admission before immediately decreasing, whereas eosinophils increased slowly throughout the first five days of the stay. In over 95% of non-complicated VOC-related stays, the CRP value was below 100 mg/L within the first day following admission and above normal after 48 hours, and the temperature was below 38 °C throughout the entire stay. Knowing the typical trajectories of these routine parameters during non-complicated VOC may urge the clinicians to be more vigilant in case of deviation from these patterns.
- Open Access
- 6 March 2019
- Published by Springer Nature in BMC Medical Research Methodology
- Vol. 19 (1) , 50
Comparison of methods for early-readmission prediction in a high-dimensional heterogeneous covariates and time-to-event outcome framework
Choosing the most performing method in terms of outcome prediction or variables selection is a recurring problem in prognosis studies, leading to many publications on methods comparison. But some aspects have received little attention. First, most comparison studies treat prediction performance and variable selection aspects separately. Second, methods are either compared within a binary outcome setting (where we want to predict whether the readmission will occur within an arbitrarily chosen delay or not) or within a survival analysis setting (where the outcomes are directly the censored times), but not both. In this paper, we propose a comparison methodology to weight up those different settings both in terms of prediction and variables selection, while incorporating advanced machine learning strategies. Using a high-dimensional case study on a sickle-cell disease (SCD) cohort, we compare 8 statistical methods. In the binary outcome setting, we consider logistic regression (LR), support vector machine (SVM), random forest (RF), gradient boosting (GB) and neural network (NN); while on the survival analysis setting, we consider the Cox Proportional Hazards (PH), the CURE and the C-mix models. We also propose a method using Gaussian Processes to extract meaningfull structured covariates from longitudinal data. Among all assessed statistical methods, the survival analysis ones obtain the best results. In particular the C-mix model yields the better performances in both the two considered settings (AUC =0.94 in the binary outcome setting), as well as interesting interpretation aspects. There is some consistency in selected covariates across methods within a setting, but not much across the two settings. It appears that learning withing the survival analysis setting first (so using all the temporal information), and then going back to a binary prediction using the survival estimates gives significantly better prediction performances than the ones obtained by models trained “directly” within the binary outcome setting.
- 6 March 2019
- Published by Springer Nature in BMC Medical Research Methodology
- Vol. 19 (1) , 50
Open AccessComparison of methods for early-readmission prediction in a high-dimensional heterogeneous covariates and time-to-event outcome framework
Choosing the most performing method in terms of outcome prediction or variables selection is a recurring problem in prognosis studies, leading to many publications on methods comparison. But some aspects have received little attention. First, most comparison studies treat prediction performance and variable selection aspects separately. Second, methods are either compared within a binary outcome setting (where we want to predict whether the readmission will occur within an arbitrarily chosen delay or not) or within a survival analysis setting (where the outcomes are directly the censored times), but not both. In this paper, we propose a comparison methodology to weight up those different settings both in terms of prediction and variables selection, while incorporating advanced machine learning strategies. Using a high-dimensional case study on a sickle-cell disease (SCD) cohort, we compare 8 statistical methods. In the binary outcome setting, we consider logistic regression (LR), support vector machine (SVM), random forest (RF), gradient boosting (GB) and neural network (NN); while on the survival analysis setting, we consider the Cox Proportional Hazards (PH), the CURE and the C-mix models. We also propose a method using Gaussian Processes to extract meaningfull structured covariates from longitudinal data. Among all assessed statistical methods, the survival analysis ones obtain the best results. In particular the C-mix model yields the better performances in both the two considered settings (AUC =0.94 in the binary outcome setting), as well as interesting interpretation aspects. There is some consistency in selected covariates across methods within a setting, but not much across the two settings. It appears that learning withing the survival analysis setting first (so using all the temporal information), and then going back to a binary prediction using the survival estimates gives significantly better prediction performances than the ones obtained by models trained “directly” within the binary outcome setting.