Predicting the risk of becoming eligible for the disability pension: Machine learning methods applied to French health data / Prédire le passage en invalidité : les méthodes d'apprentissage automatique appliquées aux données de santé françaises

Introduction: Benefiting from the disability pension implies morbid (physical and psychological) and social (fall in income) implications for the person. It also has economic consequences for society, with increasing expenses since 2011 (+4.9% on average per year). Investing in preventive actions against the loss of the ability to work should limit these consequences, but it requires targeting people at risk. The development of artificial intelligence opens up prospects in this regard. Purpose of the Research: To target, using supervised machine learning methods, those people with a high probability of becoming eligible for the disability pension over the course of the year based on their socio-demographic and medical characteristics (pathologies, work stoppages, drugs taken, and medical procedures). Method: Among the beneficiaries of the French public welfare system aged 20­64 in 2017, we compared the socio-demographic and medical characteristics between 2014 and 2016 of those who received a disability pension in 2017 and not before, and those who did not receive a disability pension from 2014 to 2017. The determination of the boundary between these two groups was tested using logistic regression, decision trees, random forests, naive Bayes classifiers, and support vector machines. The models' performance was compared with respect to accuracy, precision, sensitivity, specificity, and AUC (area under the curve). Finally, the predictive power of each factor was measured by AUC too. Results: The boosted logistic regression had the best performance for three of the five criteria, but low sensitivity. The best sensitivity was obtained with the support vector machines, with an accuracy close to that of the boosted logistic regression, but a lower precision and specificity. Random forests offered the best discriminatory ability. The naive Bayes classifier had the worst performance. The most predictive factors in becoming eligible for the disability pension were having 30 days or more off sick in 2014, 2015, and 2016 and being aged 55 to 64. Conclusion: Supervised learning methods have appeared relevant for identifying people with the highest probability of becoming eligible for the disability pension and, more broadly, for steering public and social policies.

Introduction: Le recours à la pension d'invalidité a des implications morbides (physiques ou psychiques) et sociales (baisse du revenu). Il a aussi des conséquences économiques pour la société, avec des dépenses croissantes depuis 2011 (+4,9 % en moyenne par année). Prévenir la perte de la capacité à travailler devrait permettre de limiter ces conséquences, mais nécessite de cibler les personnes à risque. Le développement des méthodes d'intelligence artificielle ouvre des perspectives en ce sens. But de l'étude: Cibler les personnes ayant une « forte ¼ probabilité de devenir bénéficiaires d'une pension d'invalidité dans l'année au regard de leurs caractéristiques sociodémographiques et médicales (pathologies, arrêts de travail, médicaments et actes médicaux) à partir de méthodes d'apprentissage automatique supervisé. Méthodes: Parmi les bénéficiaires du régime général âgés de 21 à 64 ans en 2017, comparaison des caractéristiques de 2014 à 2016 entre les nouveaux bénéficiaires d'une pension d'invalidité en 2017 et ceux n'en bénéficiant pas. La détermination de la frontière entre ces deux groupes a été testée à l'aide de la régression logistique, des arbres de décision, des forêts aléatoires, de la classification naïve bayésienne et des séparateurs à vaste marge. Les performances des modèles ont été comparées au regard de la justesse, la précision, la sensibilité, la spécificité et l'AUC (Area Under the Curve). Le pouvoir prédictif de chaque facteur est estimé à partir de l'AUC. Résultats: La régression logistique boostée avait les meilleures performances sur trois des cinq critères retenus, mais une faible sensibilité. La meilleure sensibilité était obtenue avec les séparateurs à vaste marge, avec une justesse proche de la régression logistique boostée mais une précision et une spécificité inférieures. Les forêts aléatoires offraient la meilleure capacité discriminatoire. Les facteurs les plus prédictifs du risque de passer en invalidité étaient le bénéfice d'au moins 30 jours d'indemnités journalières pour maladie en 2014, 2015 et 2016 et le fait d'être âgé de 55 à 64 ans. Conclusion: Les méthodes d'apprentissage supervisé sont apparues pertinentes pour le ciblage des personnes les plus à risque de recourir à la pension d'invalidité et, plus largement, pour le pilotage d'autres prestations sociales.

A French nationwide study compared various conditions and healthcare use of individuals < 65 years with a Down's syndrome to those without.

Few regular national clinical data are available for individuals with Down's syndrome (IDS) bearing in mind that they are subject to countries variations in medical termination of pregnancy and screening. Individuals < 65 in 2019 were selected in view of the low number of older IDS. Thus, 98% of 52.4 million people with correct data were included from the national health data system. IDS (35,342) were identified on the basis of the International Classification of Diseases 10th revision code (Q90). Risk ratios (RR) were calculated to compare the frequencies in 2019 between IDS and individual without Down's syndrome (IWDS) of use of health care. The prevalence of IDS was 0.07% (48% women), comorbidities were more frequent, especially in younger patients (24% < 1 year had another comorbidity, RR = 20), as was the percentage of deaths (4.6%, RR = 10). Overall, tumours were less frequent in IDS compared with IWDS (1.2%, RR = 0.7) except for certain leukaemias and testicular tumours (0.3%, RR = 4). Cardiac malformations (5.2%, RR = 52), dementia (1.2%, RR = 29), mental retardation (5%, RR = 21) and epilepsy (4%, RR = 9) were also more frequent in IDS. The most frequent hospital diagnoses for IDS were: aspiration pneumonia (0.7%, RR = 89), respiratory failure (0.4%, RR = 17), sleep apnoea (1.1%, RR = 8), cryptorchidism (0.3%, RR = 5.9), protein-energy malnutrition (0.1%, RR = 7), type 1 diabetes (0.2%, RR = 2.8) and hypothyroidism (0.1%, RR = 72). IDS were more likely to use emergency services (9%, RR = 2.4), short hospital stay (24%, RR = 1.6) or hospitalisation at home (0.6%, RR = 6). They consulted certain specialists two to three times more frequently than IWDS, for example cardiologists (17%, RR = 2.6). This study is the first detailed national study comparing IDS and non-IDS by age group. These results could help to optimize prenatal healthcare, medical and social support.

Factors associated with hospital admission and 30-day readmission for children less than 18 years of age in 2018 in France: a one-year nationwide observational study.

BACKGROUND: Nationwide data for children for short-stay hospitalisation (SSH) and associated factors are scarce. This retrospective study of children in France < 18 years of age followed after their birth or birthday in 2018 focused on at least one annual SSH, stay < 1 night or ≥ 1 night, or 30-day readmission ≥ 1 night. METHODS: Children were selected from the national health data system (SNDS), which includes data on long-term chronic disease (LTD) status with full reimbursement and complementary universal coverage based on low household income (CMUC). Uni and multivariate quasi-Poisson regression were applied for each outcome. RESULTS: Among 13.211 million children (94.4% population, 51.2% boys), CMUC was identified for 17.5% and at least one LTD for 4% (0-<1 year: 1.5%; 14-<18 year: 5.2%). The most frequent LTDs were pervasive developmental diseases (0.53%), asthma (0.24%), epilepsy (0.17%), and type 1 diabetes (0.15%). At least one SSH was found for 8.8%: SSH < 1 night (4.9%), SSH ≥ 1 night (4.5%), readmission (0.4%). Children with at least one SSH were younger (median 6 vs. 9 years) and more often had CMUC (21%), a LTD (12%), an emergency department (ED) visit (56%), or various primary healthcare visits than all children. Those with a SSH ≥1 night vs. < 1 night were older (median: 9 vs. 4 years). They had the same frequency of LTD (13.4%) but more often an ED visit (78% vs. 42%). Children with readmissions were younger (median 3 years). They had the highest levels of CMUC (29.3%), LTD (34%), EDs in their municipality (35% vs. 29% for the whole population) and ED visits (87%). In adjusted analysis, each outcome was significantly less frequent among girls than boys and more frequent for children with CMUC. LTDs with the largest association with SSH < 1 night were cystic fibrosis, sickle cell diseases (SCD), diabetes type 1, those with SSH ≥1 night type 1 diabetes epilepsy and SCD, and those for readmissions lymphoid leukaemia, malignant neoplasm of the brain, and SCD. Among all SSH admissions of children < 10 years, 25.8% were potentially preventable. CONCLUSION: Higher SSH and readmission rates were found for children with certain LTD living in low-income households, suggesting the need or increase of specific policy actions and research.

Underuse of primary healthcare in France during the COVID-19 epidemic in 2020 according to individual characteristics: a national observational study.

BACKGROUND: The organization of healthcare systems changed significantly during the COVID-19 pandemic. The impact on the use of primary care during various key periods in 2020 has been little studied. METHODS: Using individual data from the national health database, we compared the numbers of people with at least one consultation, deaths, the total number of consultations for the population of mainland France (64.3 million) and the mean number of consultations per person (differentiating between teleconsultations and consultations in person) between 2019 and 2020. We performed analyses by week, by lockdown period (March 17 to May 10, and October 30 to December 14 [less strict]), and for the entire year. Analyses were stratified for age, sex, deprivation index, epidemic level, and disease. RESULTS: During the first lockdown, 26% of the population consulted a general practitioner (GP) at least once (-34% relative to 2019), 7.4% consulted a nurse (-28%), 1.6% a physiotherapist (-80%), and 5% a dentist (-95%). For specialists, consultations were down 82% for ophthalmologists and 37% for psychiatrists. The deficit was smaller for specialties making significant use of teleconsultations. During the second lockdown, the number of consultations was close to that in 2019, except for GPs (-7%), pediatricians (-8%), and nurses (+ 39%). Nurses had already seen a smaller increase in weekly consultations during the summer, following their authorization to perform COVID-19 screening tests. The decrease in the annual number of consultations was largest for dentists (-17%), physiotherapists (-14%), and many specialists (approximately 10%). The mean number of consultations per person was slightly lower for the various specialties, particularly for nurses (15.1 vs. 18.6). The decrease in the number of consultations was largest for children and adolescents (GPs: -10%, dentists: -13%). A smaller decrease was observed for patients with chronic diseases and with increasing age. There were 9% excess deaths, mostly in individuals over 60 years of age. CONCLUSIONS: There was a marked decrease in primary care consultations in France, especially during the first lockdown, despite strong teleconsultation activity, with differences according to age and healthcare profession. The impact of this decrease in care on morbidity and mortality merits further investigation.

Homeopathy in France in 2011-2012 according to reimbursements in the French national health insurance database (SNIIRAM).

BACKGROUND: The use of homeopathic medicine is poorly described and the frequency of combined allopathic and homeopathic prescriptions is unknown. OBJECTIVE: To analyse data on medicines, prescribers and patients for homeopathic prescriptions that are reimbursed by French national health insurance. METHODS: The French national health insurance databases (SNIIRAM) were used to analyse prescriptions of reimbursed homeopathic drugs or preparations in the overall French population, during the period July 2011-June 2012. RESULTS: A total of 6,705,420 patients received at least one reimbursement for a homeopathic preparation during the 12-month period, i.e. 10.2% of the overall population, with a predominance in females (68%) and a peak frequency observed in children aged 0-4 years (18%). About one third of patients had only one reimbursement, and one half of patients had three or more reimbursements. A total of 120,110 healthcare professionals (HCPs) prescribed at least one homeopathic drug or preparation. They represented 43.5% of the overall population of HCPs, nearly 95% of general practitioners, dermatologists and pediatricians, and 75% of midwives. Homeopathy accounted for 5% of the total number of drug units prescribed by HCPs. Allopathic medicines were coprescribed with 55% of homeopathic prescriptions. CONCLUSION: Many HCPs occasionally prescribe reimbursed homeopathic preparations, representing however a small percentage of reimbursements compared to allopathic medicines. About 10% of the French population, particularly young children and women, received at least one homeopathic preparation during the year. In more than one half of cases, reimbursed homeopathic preparations are prescribed in combination with allopathic medicines.