Nationwide Trends in Type 1 and Type 2 Diabetes in France (2010–2019): A Population-Based Study Using a Machine Learning Classification Algorithm

Abstract

Introduction

Diabetes represents an increasing public health challenge in France, yet national data distinguishing type 1 from type 2 diabetes and insulin use remain limited. This study aimed to describe trends in the epidemiology, care pathways and health outcomes of adult individuals living with type 1 or type 2 diabetes in France from 2010 to 2019. It focused on individuals treated or not with insulin and applied a predictive classification algorithm to accurately distinguish between diabetes types using real-world data.

Methods

A 10-year retrospective population-based cohort study was conducted from a representative one-tenth sample of the French national healthcare database (i.e. SNDS, Système National des données de Santé), covering nearly the entire French population. Adults (≥ 18 years) affiliated with the general insurance scheme were included. A machine learning algorithm, trained on clinical data from general practitioners, was applied to classify diabetes type. Annual trends in prevalence, incidence, comorbidities, treatments, outpatient care, complications and mortality were assessed.

Results

Among an extrapolated 5.5 million individuals with diabetes in 2019, 3.5% had type 1 diabetes and 96.5% had type 2 diabetes. The prevalence of type 2 diabetes increased from 6.2% in 2010 to 8.0% in 2019, while type 1 diabetes remained stable. Comorbidity rates were high and increasing in insulin-treated individuals with type 2 diabetes. In 2019, 15.3% of insulin-treated individuals with type 2 diabetes had at least one complication-related hospitalisation. Specialist consultations were underused, especially in type 2 diabetes. The mortality rate in individuals with type 1 diabetes declined from 2.6% to 1.5%, with an increase in mean age at death.

Conclusion

This national study provides updated insights into diabetes in France and highlights the need to improve access to specialised care and reinforce long-term surveillance strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13300-025-01781-0.

Key Summary Points

Why carry out this study?

Diabetes is a growing public health burden in France, but national epidemiological data distinguishing type 1 from type 2 diabetes and insulin use are scarce.

Previous studies lacked robust methods to accurately classify diabetes type and treatment pathways in administrative health databases.

What was the hypothesis of the study?

Applying a machine learning classification algorithm to nationwide healthcare data would enable precise identification of type 1 and type 2 diabetes and reveal evolving trends in prevalence, complications and care.

What was learned from the study?

From 2010 to 2019, the prevalence of type 2 diabetes increased markedly while type 1 diabetes remained stable. Insulin-treated individuals with type 2 diabetes showed the highest burden of comorbidities and complications.

Despite rising disease burden, access to specialist care remained suboptimal, highlighting gaps in diabetes management. The study provides validated methods and updated data to support public health strategies and international comparisons.

Introduction

Over the last few decades, diabetes has remained a growing global burden, with a significant increase in its prevalence in both developed and developing countries, making diabetes an international public health priority [1]. Approximately 537 million adults were living with diabetes in 2021 worldwide; projections estimate that this number will rise to 643 million adults by 2030 [2]. Similarly, in European countries, about 61 million adults are living with diabetes (expected to increase to 67 million by 2030) [3]. Diabetes is associated with an increased risk of serious complications [4, 5], and is considered the eighth leading cause of death and disability in the world [6]. Moreover, as a result of its high prevalence and the associated morbidity and mortality, diabetes represents a growing and significant burden to healthcare systems [6].

To date, the global epidemiology of childhood-onset type 1 diabetes has been well studied. However, epidemiological data on adults with type 1 diabetes are lacking at the international level [7]. Assessing the trends in the epidemiology of type 1 diabetes is paramount because of the frequent misclassification as type 2 diabetes, and prevention and management strategies differ between the two types of diabetes [7]. In addition, most of the studies reported epidemiology data in the overall population with type 2 diabetes, without distinguishing those treated with insulin from those treated with other antidiabetic treatments while the management and disease evolution of these individuals may differ. Furthermore, the potential misclassification of individuals with insulin-treated type 2 diabetes as individuals with type 1 diabetes highlights the need for accurate identification and characterization of this distinct subgroup.

In France, the French national healthcare data system (SNDS) allows the study of outcomes in a large cohort of individuals with diabetes (around four million people live with diabetes in France) [8, 9]. However, as the SNDS is a medico-administrative database, one of its limitations is the lack of clinical data, which may be useful and relevant for distinguishing type 1 and 2 diabetes. To date, previous French population-based studies on the epidemiology of diabetes have been conducted over a short-term follow-up, with smaller sample sizes, potential participation bias (non-response to questionnaires) or self-reported complications and may have some limitations in terms of the distinction between diabetes types in SNDS (e.g. identification of adults with type 2 diabetes based on age ≥ 45 years and/or delivery of medications), even though differentiating between type 1 and type 2 diabetes is essential in diabetes surveillance [10–15]. Therefore, the use of a classification algorithm developed through the linkage of SNDS data with clinical data is of great importance to accurately depict both type 1 and type 2 diabetes. Indeed, from a public health perspective, it is paramount to provide updated and consistent data on the epidemiology and health outcomes of both type 1 and type 2 diabetes in a large cohort of individuals in order to inform decision-makers and implement large-scale targeted interventions to reduce the clinical and economic burden of the disease in France, and to allow the comparison of trends in different countries.

This study aimed to estimate (1) the trend in epidemiology of both type 1 and type 2 diabetes (treated and untreated with insulin) in adult individuals from 2010 to 2019 in France, using a predictive type 1/type 2 classification algorithm, and (2) to characterize the profile of individuals with diabetes in terms of comorbidities, therapeutic management, complications and mortality.

Methods

Study Design and Data Source

A 10-year retrospective cross-sectional population-based cohort study was conducted in adults with type 1 and type 2 diabetes from 1 January 2010 to 31 December 2019, using a 1/10th sample of the French nationwide claims and hospital database (SNDS, Système National des données de Santé). The SNDS contains individual pseudonymized data on all outpatient and inpatient reimbursed healthcare services for beneficiaries covered by the different healthcare insurance schemes (approximately 99% of the French population, i.e. 67 million persons) [8]. This database includes general sociodemographic characteristics (age, sex, date of death in outpatient and inpatient settings (excluding the cause of death)), outpatient healthcare consumption including drugs and medical devices dispensed, date and nature of medical visits and procedures (medical acts and biology tests) and all discharge summaries from private and public hospitals (diagnoses, date and duration of hospital stays, medical procedures). It also contains long-term disease (LTD) registration (30 specific and most expensive chronic diseases) for which full coverage for all medical expenses was insured [8].

Study Population

The source population was a 1/10th sample of adult individuals identified with diabetes in the SNDS database, and continuously insured by the main French healthcare insurance system for active and retired salaried workers (and their relatives), that is, the General Scheme (covering 76% of the French general population, excluding individuals with local mutualist sections (LMS)) from 1 January 2010 to 31 December 2019 [16]. Individuals with diabetes were identified according to the published criteria routinely used by the national health insurance fund (CNAM) [17]. A 1/10th sample random sample of individuals with diabetes was drawn to reduce the dataset size while maintaining statistical power. Stratified random sampling without replacement was applied, using strata based on sex, birth month and year, department of residence and CMUc (Couverture médicale universelle complémentaire) which indicates low-income status. From this sample, the study population included individuals aged 18 or older at the date of diabetes identification (i.e. index date), and continuously affiliated to the strict General Scheme throughout the follow-up period. Adults with gestational diabetes (having a hospitalisation with a main, related or associated diagnosis of gestational diabetes using the International Classification of Diseases, 10th revision (ICD-10 codes O240–O244 or O249) during the study period) were excluded. A machine learning algorithm, developed through a linkage between SNDS data and primary clinical data (IQVIA’s Ambulatory Electronic Medical Records, EMR) from a network of 1200 French general practitioners (GPs) (representative of the national GP population and covering 2.8% of the French population), was applied to accurately identify the type of diabetes in the study population as individuals with chronic conditions such as diabetes are generally followed consistently by their primary care physician. The type of diabetes entered by GPs during consultations (and then coded in ICD-10 codes) in EMR data was used as the “true” diabetes diagnosis, and then compared with the prediction of the model. Decision rules were applied for selecting individuals for whom the type of diabetes was considered “highly likely” (e.g. individuals with both T1D and T2D diagnoses, or individuals with T1D who had prescriptions for oral antidiabetic drugs or glucagon-like peptide 1 (GLP-1) receptor agonists, as well as those with gestational diabetes were excluded from the final EMR population). The algorithm was developed on a training set (for model derivation) and a test set (for validation of model performances). This process assessed the algorithm’s ability to accurately identify individuals with T1D (considered as true positives for the calculation of model performances), reject individuals with T2D (considered as true negatives) and minimize classification errors such as false positives and false negatives. Thus, the best model was selected for its ability to predict T1D on the test set, via the measure of the F1-score metric (harmonic mean of precision and sensitivity). The selection of model predictors considered for the development of the algorithm was guided by clinical relevance and data reliability. Different look-back periods (1, 2 or 5 years prior to the end of an individual’s follow-up) were used depending on the nature of the predictor. This flexible approach improved both the clinical significance and predictive performance of the model and enhanced the robustness of case identification. In addition, to account for the imbalance in the type of diabetes (96% of individuals with T2D, and 4% with T1D in the linked cohort), which can be an issue in machine learning model training, the imbalance was managed during model training in two ways: (1) the F1-score was used as the optimization metric (instead of accuracy, for instance, which is not adapted to unbalanced designs); (2) two strategies were tested during model training, using both the original unbalanced dataset and a balanced development set created by down-sampling the majority class (T2D) to match the number of minority class cases (T1D). This allowed for the assessment of the impact of class distribution on model training and performance. Further details on methodology for the development of the predictive algorithm are provided in Supplementary Online Resource 1. Once an individual was classified by the predictive algorithm as having a specific type of diabetes, they were considered as having that diagnosis consistently throughout the duration of the study. The inclusion period was from 1 January 2010 to 31 December 2019, with a 1-year historical period preceding the index date (total extraction period from 1 January 2009 to 31 December 2019). Individuals were followed from index date until date of death, end of the study period (31 December 2019) or loss to follow-up (defined as no healthcare consumptions over a 12-month period, ensuring the inclusion of individuals continuously covered by the General Scheme over the study period (i.e. exclusion of individuals who changed their insurance scheme, moved abroad or emigrated), whichever occurred first.

For all calendar analyses, people living with diabetes (PwD) were classified as follows: individuals with type 1 diabetes (T1D), individuals with type 2 diabetes (T2D) (including individuals treated and not receiving any antidiabetic medications (i.e. untreated individuals) within the year), individuals with type 2 diabetes receiving insulin in the year (i.e. insulin-treated individuals with type 2 diabetes, T2Di) either as monotherapy by multiple daily injections (MDI) or insulin pump therapy or insulin in combination with other antidiabetic treatments) and individuals with type 2 diabetes receiving only antidiabetic treatments within the year (excluding insulin and/or insulin pump therapy) (i.e. non-insulin-treated individuals with type 2 diabetes, T2Dni).

This study was approved by the French Data Protection Supervisory Authority “Commission Nationale Informatique et Libertés” (CNIL) (authorization no. 919119). As this study was a retrospective analysis of data from the SNDS, no informed consent was required and therefore not obtained. Retrospective analysis of pseudonymized data and therefore consent for publication is not required.

Epidemiology of Type 1 and Type 2 Diabetes

The prevalence and the incidence of diabetes were extrapolated to the total French population and reported in the four populations of interest mentioned in Sect. “Study Population”. The extrapolated number of adults with diabetes was obtained by dividing the number of individuals in the study population by 76% (to compensate for the General Scheme selection) [16] and by multiplying by 10 (to account for the 1/10 sample).

More specifically, the prevalence was reported as the total number of adult individuals identified as living with diabetes during a given year (i.e. the numerator), over the total number of French inhabitants ≥ 18 years in a given year (i.e. the denominator), based on the yearly estimates published by the National Institute for Statistics and Economic Studies (INSEE) [18]. Individuals with no healthcare consumption during a full calendar year were not considered in the prevalence count, even if they had been previously identified.

The incidence rate was reported as the number of incident individuals living with diabetes in a given year divided by the total French population in that year. For a given year, the denominator was estimated as the mean total population (≥ 18 years) free of diabetes at the beginning and at the end of the same given year [18, 19]. Incidence rates were expressed per 100,000 persons.

Characteristics, Treatment Regimens and Outpatient Consultations for Diabetes

Main descriptive characteristics of PwD were reported at the beginning of each calendar year of interest (1 January) and described among prevalent PwD in the four populations of interest defined above. For each calendar year, the descriptive variables included sociodemographic characteristics (age, sex), comorbidities assessed in the preceding year, treatment regimens used in PwD and concomitant treatments received during the year of follow-up. Outpatient consultations carried out by the specialists during the year were also described (Supplementary material Table 1).

Diabetes-Related Complications and All-Cause Mortality

Serious diabetes-related complications were defined as the presence of at least one hospital stay with a main or related diagnosis in short-stay institutions (medicine, surgery, obstetrics and odontology) in a given year (cf. Supplementary material Table 1). All-cause mortality was estimated as the percentage of individuals dying during the year (including in-hospital deaths and deaths occurring in ambulatory settings). The average age of deceased individuals was also reported.

Statistical Analysis

Cross-sectional descriptive analyses were performed annually to describe study outcomes (characteristics, treatments, diabetes-related complications and mortality). For each of these calendar years, the overall extrapolated prevalence and prevalence by age group were reported for the four populations of interest, as well as the overall extrapolated incidence rates. All analyses were performed with SAS^® software (SAS Institute, version 9.4, NC, USA).

Results

Study Population Selection

Among the 524,423 adults of the initial 1/10th sample of PwD from the SNDS database, 420,587 PwD were included during the 2010–2019 study inclusion period, corresponding to an extrapolated number of 5,534,039 PwD. The algorithm identified 14,543 adults with T1D (3.5%) and 406,044 with T2D (96.5%) (Fig. 1). The main reason for the end of follow-up was the end of the study period (T1D, 83.1%; T2D, 76.7%), followed by death (T1D, 13.4%; T2D, 18.5%) and loss to follow-up (T1D, 3.6%; T2D, 4.8%) (data not shown).

Fig. 1.

Cohort flowchart diagram. T1D type 1 diabetes, T2D type 2 diabetes

Population Characteristics of Type 1 and Type 2 Diabetes

Sociodemographic Characteristics and Comorbidities

In the most recent year (i.e. 2019), individuals with T1D were more frequently men (59.9%) with a mean age of 46.1 years (standard deviation, SD 18.35), while 51.1% of individuals with T2D were women [mean age 67.1 years (SD 12.82)] (Table 1). Some differences were observed regarding the mean age, number and type of comorbidities, according to the group of PwD (Table 1). Insulin-treated individuals with type 2 diabetes (T2Di) notably had a high proportion of comorbidities e.g. hypertension (82.1% in 2019), myocardial infarction/chronic ischemic heart diseases (15.6%), heart failure/rhythm disorders (13.7%), chronic renal failure (not restricted to end-stage renal disease) (11.1%) and stroke (6.4%). In comparison to 2010, individuals with T1D appeared younger in 2019 (mean age 46.1 (SD 18.35) vs. 48.1 (16.26)) (Table 1, Supplementary material Table 2). In individuals with T2D, the mean age seemed to have increased over the years, notably among non-insulin-treated individuals (T2Dni) [mean age 64.7 years (SD 12.01) vs. 66.5 years (12.03)] (Table 1, Supplementary material Table 3). The proportions of individual comorbidities remained globally stable over the study period in individuals with T1D, except for arteriopathies, heart failure/rhythm disorders, dyslipidemia and arterial hypertension, where a downward trend was observed (Supplementary material Table 2). In individuals with T2Di and T2Dni, a rising trend was reported for most comorbidities (except arterial hypertension or dyslipidemia) (Supplementary material Tables 4 and 5).

Table 1.

Characteristics and treatments of prevalent individuals with T1D and T2D (overall, insulin-treated (T2Di) and non-insulin-treated (T2Dni)), 2019

	Total T1D N = 12,102	Total T2Da N = 320,362	T2Di N = 62,479	T2Dni N = 215,905
Age (years)
Mean (SD)	46.1 (18.35)	67.1 (12.82)	69.1 (12.76)	66.5 (12.03)
Median (Q1–Q3)	46.0 (30.0–60.0)	68.0 (59.0–76.0)	70.0 (61.0–79.0)	68.0 (59.0–76.0)
Age groups (years), n (%)
18–29	2915 (24.1%)	2214 (0.7%)	301 (0.5%)	1059 (0.5%)
30–39	1793 (14.8%)	5753 (1.8%)	803 (1.3%)	3118 (1.4%)
40–49	2010 (16.6%)	19,714 (6.2%)	3041 (4.9%)	13,443 (6.2%)
50–59	2221 (18.4%)	55,551 (17.3%)	9391 (15.0%)	39,863 (18.5%)
60–69	1795 (14.8%)	95,669 (29.9%)	17,579 (28.1%)	68,619 (31.8%)
70–79	984 (8.1%)	84,872 (26.5%)	17,091 (27.4%)	58,407 (27.1%)
≥ 80	384 (3.2%)	56,589 (17.7%)	14,273 (22.8%)	31,396 (14.5%)
Gender, n (%)
Male	7251 (59.9%)	156,529 (48.9%)	29,489 (47.2%)	108,957 (50.5%)
Number of comorbidities in the preceding year
Mean (SD)	1.2 (1.52)	1.9 (1.34)	2.4 (1.54)	1.8 (1.22)
Comorbidities in the preceding year, n (%)
Myocardial infarction/chronic ischemic heart diseases	769 (6.4%)	34,088 (10.6%)	9722 (15.6%)	20,532 (9.5%)
Stroke	335 (2.8%)	14,075 (4.4%)	4005 (6.4%)	7819 (3.6%)
Lower limb amputation	93 (0.8%)	1149 (0.4%)	727 (1.2%)	311 (0.1%)
Retinopathies	303 (2.5%)	12,038 (3.8%)	2994 (4.8%)	7690 (3.6%)
Neuropathies/neuritis associated with diabetes	725 (6.0%)	12,177 (3.8%)	4503 (7.2%)	6068 (2.8%)
Arteriopathies	550 (4.5%)	15,816 (4.9%)	5355 (8.6%)	8351 (3.9%)
Chronic renal failure	761 (6.3%)	13,070 (4.1%)	6939 (11.1%)	3971 (1.8%)
Psychiatric disorders	1229 (10.2%)	25,948 (8.1%)	6907 (11.1%)	14,381 (6.7%)
Malignancies	571 (4.7%)	32,460 (10.1%)	7322 (11.7%)	20,160 (9.3%)
Endocrine disorders (dysthyroidism)	359 (3.0%)	5776 (1.8%)	2164 (3.5%)	2658 (1.2%)
Heart failure/rhythm disorders	502 (4.1%)	28,802 (9.0%)	8576 (13.7%)	15,133 (7.0%)
Arthropathies	11 (0.1%)	110 (0.0%)	64 (0.1%)	36 (0.0%)
Dyslipidemia	3381 (27.9%)	165,569 (51.7%)	38,089 (61.0%)	115,131 (53.3%)
Arterial hypertension	4415 (36.5%)	232,913 (72.7%)	51,278 (82.1%)	156,988 (72.7%)
Exposure to antidiabetic treatments in the year, n (%)
Monotherapy by insulin in multiple daily injections	8522 (70.4%)	20,698 (6.5%)	20,698 (33.1%)	NA
Insulin pump therapy	3227 (26.7%)	1272 (0.4%)	1272 (2.0%)	NA
Insulin with other treatments	353 (2.9%)	40,509 (12.6%)	40,509 (64.8%)	NA
Monotherapy (excluding insulin and GLP-1 analogues)	NA	121,449 (37.9%)	NA	121,449 (56.3%)
Metformin	NA	93,116 (29.1%)	NA	93,116 (43.1%)
Sulfonamides	NA	19,994 (6.2%)	NA	19,994 (9.3%)
DDP4i	NA	7108 (2.2%)	NA	7108 (3.3%)
Other	NA	1231 (0.4%)	NA	1231 (0.6%)
Dual therapy (excluding insulin and GLP-1 analogues)	NA	55,355 (17.3%)	NA	55,355 (25.6%)
Metformin and sulfonamides	NA	20,510 (6.4%)	NA	20,510 (9.5%)
Metformin and DPP4i	NA	28,964 (9.0%)	NA	28,964 (13.4%)
Sulfonamides and DPP4i	NA	4999 (1.6%)	NA	4999 (2.3%)
Other	NA	882 (0.3%)	NA	882 (0.4%)
Triple or more therapy (excluding insulin and GLP-1 analogues)	NA	26,305 (8.2%)	NA	26,305 (12.2%)
Meftormin, sulfonamides and DPP4i	NA	24,487 (7.6%)	NA	24,487 (11.3%)
Meftormin, DPP4i and other treatments	NA	415 (0.1%)	NA	415 (0.2%)
Other	NA	1403 (0.4%)	NA	1403 (0.6%)
GLP-1 analogue alone or in combination (except with insulin)	NA	12,796 (4.0%)	NA	12,796 (5.9%)
Untreated within the year	NA	41,978 (13.1%)	NA	NA
Concomitant treatments in the year, n (%)
Antiplatelet agents	2828 (23.4%)	122,057 (38.1%)	32,016 (51.2%)	80,314 (37.2%)
Lipid-lowering agents	4019 (33.2%)	180,482 (56.3%)	40,465 (64.8%)	127,327 (59.0%)
Antihypertensive treatments	4812 (39.8%)	242,091 (75.6%)	52,426 (83.9%)	165,633 (76.7%)
Psychotropic drugs	2713 (22.4%)	95,107 (29.7%)	21,963 (35.2%)	60,282 (27.9%)

NA not applicable, T1D type 1 diabetes, T2D type 2 diabetes, T2Di type 2 diabetes treated with insulins, T2Dni type 2 diabetes treated with other antidiabetic treatments than insulins

^aT2D includes non-insulin-treated individuals, insulin-treated individuals and untreated T2D individuals within the given year of interest

Therapeutic Management of Diabetes

Among individuals with T1D, more than half were treated with insulin (MDI) in monotherapy (70.4%), followed by insulin pump therapy (26.7%) and other antidiabetic treatments with insulin (2.9%) in 2019 (Table 1). From 2010 to 2019, the proportion of individuals with T1D using insulin pump therapy increased from 12.5% to 26.7% (Table 1, Supplementary material Table 2). In individuals with T2D, 67.4% were non-insulin-treated (T2Dni), while 19.5% were treated with insulin (T2Di) in 2019. Individuals with T2Dni were more frequently treated with monotherapy (56.3% of them, mainly with metformin (76.7%), the use of which has increased over the years compared to sulfonamides) (Table 1, Supplementary material Table 4). This was followed by 25.6% of them receiving dual therapy and 12.2% having triple or more therapies. A low proportion of individuals with T2Dni were exposed to GLP-1 analogues alone or in combination (2010, 1.4%; 2019, 5.9%). In individuals with T2Di, a high proportion were treated with insulin in combination with other antidiabetic treatments (64.8%), followed by insulin MDI in monotherapy (33.1%) (Table 1, Supplementary material Table 5). Few of them were exposed to insulin pump therapy, but an increase in its use was observed (2010, 0.8%; 2019, 2.0%).

Outpatient Consultations

Around 41.3% of individuals with T1D had at least one consultation with an endocrinologist during the 2019 year, followed by only 26.6% and 8.5% of individuals with T2Di and T2Dni, respectively; less than half of individuals had a consultation with a cardiologist and ophthalmologist (Fig. 2). The majority of individuals had at least one visit with a GP (more than 90% of individuals), and a non-negligible proportion of them had paramedical appointments with a nurse for procedural purposes (such as blood sample collection) (60.6% and 71.1% for individuals with T1D and T2D, respectively). An increase of the percentages of individuals visiting a nurse and a pedicure/podiatry was observed, whatever the type of diabetes. Evolution of outpatient consultations from 2011 to 2019 is further described in Supplementary material Tables 2–5.

Fig. 2.

Outpatient consultations in individuals with T1D and T2D (overall, insulin-treated (T2Di) and non-insulin-treated (T2Dni)), in 2019. T1D type 1 diabetes, T2D type 2 diabetes (treated with insulins or other antidiabetic treatments than insulins and untreated individuals), T2Di type 2 diabetes treated with insulins, T2Dni type 2 diabetes treated with other antidiabetic treatments than insulins

Occurrence of Serious Diabetes-Related Complications and All-Cause Mortality in Type 1 and Type 2 Diabetes

Serious Diabetes-Related Complications

In total, 13.7% of individuals with T1D had at least one hospitalisation stay for diabetes-related complications in 2019; the most frequent complications were microangiopathies (9.7%), notably retinopathies (4.5%), followed by acute metabolic complications (3.7%) (Table 2). Among individuals with T2D, 8.5% were hospitalised for diabetes-related complications; this proportion reached 15.3% in individuals with T2Di, with frequent admissions for microangiopathies (12.3%). A smaller proportion of hospitalisations was found for individuals with T2Dni (6.9%). Overall, few individuals were hospitalised for serious macrovascular complications. Over the study years, the frequency of serious complications seemed globally unchanged regardless the type of diabetes (Supplementary material Tables 2–5).

Table 2.

Diabetes-related complications in prevalent individuals with T1D and T2D (overall, insulin-treated and non-insulin-treated T2D), 2019

	2019
	Total T1D N = 12,102	Total T2Da N = 320,362	T2Di N = 62,479	T2Dni N = 215,905
At least one hospital stay for diabetes-related complicationsb, n (%)	1662 (13.7%)	27,263 (8.5%)	9542 (15.3%)	14,876 (6.9%)
Microangiopathies	1171 (9.7%)	22,469 (7.0%)	7667 (12.3%)	12,436 (5.8%)
Renal failure	279 (2.3%)	3810 (1.2%)	2199 (3.5%)	1098 (0.5%)
Retinopathy	544 (4.5%)	11,920 (3.7%)	3264 (5.2%)	7558 (3.5%)
Neuropathy	336 (2.8%)	6449 (2.0%)	2039 (3.3%)	3665 (1.7%)
Low limb damage	80 (0.7%)	1056 (0.3%)	603 (1.0%)	358 (0.2%)
Peripheral angiopathy	< 10	63 (0.0%)	30 (0.0%)	28 (0.0%)
Macrovascular complications	136 (1.1%)	4257 (1.3%)	1595 (2.6%)	2248 (1.0%)
Myocardial infarction	64 (0.5%)	1830 (0.6%)	670 (1.1%)	993 (0.5%)
Stroke	34 (0.3%)	1944 (0.6%)	619 (1.0%)	1082 (0.5%)
Vascular disorders	43 (0.4%)	529 (0.2%)	333 (0.5%)	190 (0.1%)
Rheumatological complications	< 10	19 (0.0%)	< 10	< 10
Acute metabolic complications	446 (3.7%)	1404 (0.4%)	766 (1.2%)	507 (0.2%)

T1D type 1 diabetes, T2D type 2 diabetes, T2Di type 2 diabetes treated with insulins, T2Dni type 2 diabetes treated with other antidiabetic treatments than insulins

^aT2D includes non-insulin-treated individuals, insulin-treated individuals and untreated T2D individuals within the given year of interest

^bHospitalisations with a main or related diagnosis for a complication of diabetes in short-stay institutions (medicine, surgery, obstetrics and odontology: French acronym MCO)

All-Cause Mortality from 2010 to 2019

The proportion of deaths in individuals with T1D decreased from 2.5% in 2010 to 1.5% in 2019, coupled with an increase in the average age of deceased individuals from 64.1 years in 2010 to 66.8 years in 2019 (Supplementary material Table 2). The mortality rate in individuals with T2D (treated and untreated individuals) remained stable across the years (2.3% and 2.8%, respectively), while an increase in the average age at death was observed (76.4 years to 78.8 years) (Supplementary material Table 3).

Annual Prevalence and Incidence of Type 1 and Type 2 Diabetes

Over a 10-year period, the prevalence of T1D remained stable with around 2.6‰ and 3.0‰ of individuals with T1D in France from 2010 to 2019, with a slight tendency to increase (Fig. 3, Supplementary material Table 2). According to age group, a net increase of the prevalence of T1D in the 18–29 years was observed in comparison to other age groups (2010, 2.1‰; 2011, 4.2‰) (Supplementary material Fig. 1A). In contrast, the prevalence of T2D increased over the years from 6.2% in 2010 to 8.0% in 2019, notably in the older age groups (70–79 years, from 16.9% to 20.5%; 80 years, from 12.4% to 18.2%) (Fig. 3, Supplementary material Fig. 1B). Individuals with T2Dni accounted for a large part of the total type 2 diabetes prevalence (4.7% and 5.4% of the total French adult population in 2010 and 2019, respectively) while individuals with T2Di represented around 1.1% and 1.6% of the French population (Fig. 3, Supplementary material Tables 4–5 and Fig. 1). Overall, the prevalence of treated individuals with T2D among the French adult population increased from 2010 (5.8%) to 2019 (7.0%) while the prevalence of untreated individuals with T2D was estimated at 0.8% in 2010 and 1.0% in 2019 (data not shown). The incidence rate (IR) of T1D has increased over the years from 9.5 new cases per 100,000 persons (≥ 18 years old) in 2011 to 19.5 new cases per 100,000 persons in 2019 (Supplementary material Table 2 and Fig. 2). On the contrary, the IR of T2D has slightly decreased over the years (2011, 573.2 per 100,000 persons; 2019, 487.7 per 100,000 persons), mainly for individuals with T2Dni (Supplementary material Tables 3–5, Fig. 2).

Fig. 3.

Annual prevalence of T1D and T2D (overall, insulin-treated (T2Di) and non-insulin-treated (T2Dni)), by calendar year, France. T1D type 1 diabetes, T2D type 2 diabetes (treated with insulins or other antidiabetic treatments than insulins and untreated individuals), T2Di type 2 diabetes treated with insulins, T2Dni type 2 diabetes treated with other antidiabetic treatments than insulins. Annual prevalence for each subgroup of interest was calculated among the total French population (≥ 18 years) in a given year (INSEE data)

Discussion

To our knowledge, this study provides for the first time robust population-based data and an updated complete picture of the trend in epidemiology and health outcomes with high accuracy for adults with T1D and T2D (treated or not treated with insulin) at the population level in France over a 10-year period, using a high-performing type 1/type 2 predictive classification algorithm developed as part of this study. Over the years, the frequency of comorbidities and diabetes-related complications presented no major changes in individuals with T1D. In individuals with T2D, a more important upward trend of the disease burden was found in both individuals treated and not treated with insulin, with notably a high rate of comorbidities in TD2i; this latter subgroup had also a high rate of complications. In line with these findings, recommendations for monitoring diabetes appear to be insufficiently followed with low recourse to medical consultations, in a context of a constantly increasing number of individuals living with T1D and T2D in France.

The present study provides comprehensive data on the profile of both individuals with T1D and T2D. In accordance with the literature, cardiovascular diseases were more frequent in individuals with T2D, while renal disorders or neuropathies were mainly reported in individuals with T1D [11, 13]. It should be noted that cardiovascular risk factors were also frequent in individuals with T1D (e.g. dyslipidemia, arterial hypertension). This is consistent with the well-established increase of cardiovascular risk in individuals with chronic renal failure [20]. Individuals with T2Di presented a non-negligible frequency of microangiopathies (notably renal failure). This finding should be interpreted in the context of the contraindication of certain oral antidiabetic agents (e.g. metformin) in advanced renal disease, which necessitates a transition to insulin therapy. Over the years, a greater increase in most comorbidities was found in individuals with T2D than in the general adult population (e.g. coronary diseases, chronic renal failure) [21, 22]. This increase may be also associated in part with the higher prevalence of multimorbidity as the population ages, in line with increased life expectancy and the diabetes duration [23, 24]. Additionally, the results of the present study showed no obvious reduction in the occurrence of serious complications in individuals with T2D. Projections from CNAM estimate that complications specific to diabetes (excluding terminal insufficiency and lower-limb amputations) would increase by 5% in individuals with T2D from 2015 to 2027 [10]. These findings could partly be explained by the low use of consultations by individuals which is inconsistent with recommendations for the management of PwD. Indeed, less than half of individuals with T1D or T2Di visited an endocrinologist in 2019 in our study. The same finding was observed for cardiologist or ophthalmologist consultations, while cardiovascular diseases are the most frequent cause of excess mortality in PwD [25, 26], and diabetic retinopathy is known as the most frequent microvascular complication of diabetes [11, 13, 27]. However, these findings should be interpreted with caution, as the frequency of such specialist visits is likely influenced by patients’ individual risk profiles and the presence or absence of diabetes-related complications.

Therefore, despite the large implementation of combined measures aimed at reducing the diabetes epidemic (e.g. prevention activities targeting lifestyle changes, population-based interventions such as awareness and public health campaigns including environmental changes), our study underlines the constant importance of reinforcing the implementation of targeted interventions to effectively reduce and prevent diabetes-related complications or associated comorbidities in PwD. This is of great importance considering the constant increase of PwD, particularly the population with T2D, which remains high in France (8.0% in individuals with T2D ≥ 18 years in 2019). A recent French study estimated the prevalence of T2D at 10.4% among individuals aged 45 and over in 2019 [14]. This finding is consistent with international data, where diabetes prevalence (type 2 diabetes representing around 90% of overall PwD) increases over the years or at the best remains stable in some countries with still high prevalence [28, 29]. However, although the prevalence of T2D has been increasing, international studies converged toward a downward or stable trend in the incidence of T2D since 2010 in several countries [1]; our results outlined this trend for recent years. The recent French study reported 216,117 incident-treated individuals with T2D in 2019; this is consistent with incident cases found in our study including both treated and untreated individuals with T2D (236,316 incident in 2019) [14].

This increase in prevalence may partly be explained by the progressive ageing of the population and may reflect in part improvements in diabetes management leading to increased life expectancy. Indeed, regarding the evolution of the management of PwD, this study highlighted that more than a quarter of individuals with T1D (26.7%) used insulin pump therapy in 2019; their use has doubled since 2010 (12.5%). Furthermore, the present study showed that the mean age at death increased by 3 years over the years in individuals with T1D; for the same period, mean age at death of the general adult population increased by about 2 years (2010, 76.7 years; 2019, 79.5 years) [23, 24]. The development of new innovative technologies (e.g. continuous glucose monitoring devices (CGM), smart insulin pens, insulins pumps) in the treatment of individuals with T1D could in part participate in this increase, as outlined elsewhere [30, 31]. However, the mean age at death of individuals with T1D remains lower than the general adult population (66.8 years vs. 79.5 years in 2019). For individuals with T2D, mean age at death (78.8 years in 2019) was relatively close to the general adult population. In addition, for this population, it is worth noting that prevalent individuals with T2D who died in 2019 were, on average, older and they are assumed to have a longer duration of diabetes compared to prevalent individuals from previous years. While these two factors may be associated with an increased risk of both microvascular and macrovascular complications, as well as higher mortality, no apparent increase in the percentage of deaths was observed. This trend was also outlined in a prior French study [15]. This observation may suggest an improvement in the management of individuals with T2D, including a more comprehensive cardiovascular risk, contributing to stabilizing mortality outcomes in this population. Supporting this, results of a prior UK BioBank study have shown that comprehensive risk factor control can substantially mitigate the excess mortality typically associated with T2D, even in older individuals or those with longer disease duration [32].

From a public health perspective, it is essential to effectively monitor the epidemiology of diabetes to guide policymakers in managing diabetes [26, 30]. For this purpose, the methodology used to build and train the predictive type 1/type 2 classification algorithm developed in the present study may be reused by the scientific community as an epidemiologic tool to monitor diabetes, and to conduct future epidemiological and pharmacoeconomic studies on each type of diabetes in France. Indeed, the machine learning algorithm (LASSO regression) displayed higher performance on our population (in terms of sensitivity, specificity and F1-score) than other models using SNDS data published in the literature that were reapplied to our data [12, 33]. In particular, we observed that previous models lacked sensitivity in our sample (many TD1 cases were classified as TD2) [12, 33]. In addition, the algorithm considered 66 variables to discriminate between types of diabetes, including CGM devices available since 2017 in France (unlike prior algorithms). This could partly explain the high performance of our model, and the observed increase in TD1 incidence notably from 2017, when considering CGM devices allowing for better differentiation of TD1 and TD2 in adults. Finally, from a research point of view, these recent and exhaustive data describing the profile of individuals living with TD1 or TD2 will be very useful for investigators wishing to have baseline data to carry out a headcount calculation, for example.

This study has some limitations. First, we cannot exclude the possibility of classification bias by the model (presence of false positives or negatives), especially for individuals who are incident in 2019, as their diabetes is recent, and the algorithm has only 1 year of data available to classify their type of diabetes. Additionally, the algorithm was developed on the basis of EMR data, considering the diabetes type indicated by the GP, and it cannot be ruled out that some misclassifications may have been made by the physicians. In the same way, we cannot exclude the possibility that some individuals consulted multiple physicians, which could potentially lead to divergent diagnoses regarding the type of diabetes. However, this bias appears to be limited in our study, as the data were collected from a panel of 1200 GPs, and an individual would have to consult at least two physicians within this same panel for such inconsistencies to arise. Further details on strengths and limits of the algorithm are provided in Supplementary material 1. Furthermore, maturity-onset diabetes of the young (MODY) cases were not excluded from the study population, while this population could potentially be misclassified as type 1 diabetes, particularly in younger individuals. The diagnosis of MODY cannot be identified in the SNDS database as no specific ICD codes exists for such disease and genetic data are unavailable. Nevertheless, we can assume that this classification bias has limited impact on the study results because of the low prevalence of the disease (MODY cases represent between 2% and 3% of diabetes cases in France) [34]. Second, this study was conducted only in adult individuals, and thus does not provide an updated picture of TD1 management in children. Non-serious diabetes-related complications were not available in the present study, underestimating the overall burden due to the complications, as well as some comorbidities that are primarily managed in primary care settings, which are not captured through LTD status and/or are not easily approximated using algorithms based on diagnostic codes, procedures or medication dispensing data (e.g. early-stage renal impairment abnormality, arthropathies). Third, only individuals affiliated with the General Scheme were included in the study because of the non-exhaustiveness of medico-administrative data from other health insurance schemes at the beginning of the study period. According to prior French studies such as the ENTRED survey, data were extrapolated to the entire French population under the assumption that, for equivalent age and sex, the prevalence of diabetes is similar between individuals covered by the general health insurance scheme (covering nearly 80% of the French population) and those affiliated with other schemes [35]. The method used for extrapolation has tended to compensate for this by applying a correction factor. However, it should be noted that this extrapolation is based on the assumption that the percentage of individuals affiliated to the General Scheme among PwD is the same as that observed in the general population (76%). Fourth, there is no picture of the use of recent treatments such as glifozines or limited data on GLP-1 agonists as a result of the study periods considered (availability of extraction). Fifth, as we used the SNDS, a medico-administrative database, there is a lack of clinical data (e.g. family history of diabetes, body mass index scores, lab test results), and the assumption that all reimbursed medications were correctly taken by individuals must be made. In addition, the SNDS does not include any information regarding individuals’ ethnic origin in accordance with French data protection regulation or migratory trajectories [36]. Finally, we cannot exclude that the study design may have partially contributed to an artificial increase in the number of individuals identified as living with diabetes over the years. Indeed, the population for 2019 includes individuals identified with diabetes from 2010 to 2019, providing a longer look-back period than that available that for the 2010 population (i.e. 1 year).

Conclusion

This study provides an updated and complete picture of the trend in epidemiology and health outcomes of type 1 diabetes and type 2 diabetes (treated or not treated with insulin) at the national level in France for the first time, using a robust predictive type 1/type 2 classification algorithm. From 2010 to 2019, few apparent reductions of diabetes-related complications or comorbidities for both types of diabetes have been observed in the context of insufficient monitoring of PwD in regards to medical guidelines, as reflected by insufficient use of specialised consultations. These findings emphasise the need to constantly increase awareness of healthcare professionals and health authorities to effectively prevent and manage (in a multidisciplinary manner) the most at-risk complications or comorbidities for each type of diabetes (especially in insulin-treated individuals with type 2 diabetes where care support for this population represents a major public health issue) and reduce the global burden of the disease, in collaboration with PwD (e.g. implementation of tailored patient-centred care). This is especially important as the population living with diabetes increases, becomes older and requires more complex treatments. From a public health perspective, it is also crucial for further research to estimate the economic burden of type 2 and type 1 diabetes in real-life settings to enhance both the clinical and economic impact of diabetes on the healthcare system.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Guy Fagherazzi, Pierre Serusclat, Cécile Berteau, Antoine Pouyet, Michael Joubert, Barbara Roux, Oriane Bretin, Emilie Casarotto, Pascaline Rabiéga and Yolaine Rabat contributed to the conception and study design. Oriane Bretin, Emilie Casarotto, Yolaine Rabat, Pascaline Rabiéga and Barbara Roux contributed to data acquisition and statistical analysis. All named authors were involved in the interpretation of the data and critically revised the manuscript critically for intellectual content, with the first draft being written by Oriane Bretin, Emilie Casarotto, Pascaline Rabiéga and Barbara Roux. All named authors reviewed and approved the final version of the manuscript for publications. All named authors are the guarantors of this work.

Funding

This work was funded by Roche Diabetes Care France, including payment of the journal’s Rapid Service fee.

Data Availability

The datasets generated during and/or analyzed during the current study are not publicly available due to legal and policy restrictions regarding their use. According to the principles of data protection and French regulations, the authors cannot publicly release the data from the Système National des Données de Santé (SNDS) (French law to access SNDS https://www.snds.gouv.fr). The script codes used for the development of the predictive algorithm are publicly available (Nouvelle ressource en open source dans la BOAS: EPICODIAB—Annonces—Communauté d'entraide SNDS).

Declarations

Conflicts of Interest

Guy Fagherazzi has provided advisory/speaking services and/or has received research grants and/or speaker honoraria from MSD, MSDAvenir, Eli Lilly, Roche Diabetes Care, Sanofi, AstraZeneca, Danone Research, Diabeloop, Bristol Myers Squibb, L'Oréal R&D, Abbvie Pharmaceutical, Pfizer, Vitalaire and Akuity Care. Michael Joubert declares consultant and/or speaker fees and/or research support from Abbott, Air Liquide Santé International, Amgen, Asdia, AstraZeneca, Bayer, BMS, Boehringer-Ingelheim, Dexcom, Dinno Santé, Glooko, Insulet, Lifescan, Lilly, LVL médical, Medtronic, MSD, Nestle HomeCare, Novonordisk, Organon, Orkyn, Roche Diabetes, Sanofi, Tandem, Vitalaire, Voluntis, Ypsomed. Michael Joubert is an Editorial Board Member of Diabetes Therapy. Michael Joubert was not involved in the selection of peer reviewers for the manuscript nor any of the subsequent editorial decisions. Pierre Serusclat has provided advisory/speaking services and/or research support from Roche Diabetes Care as an expert on the scientific committee of the present study. Cécile Berteau and Antoine Pouyet are employees of Roche Diagnostics France and Timkl France, respectively. Oriane Bretin, Emilie Casarotto, Yolaine Rabat, Pascaline Rabiéga and Barbara Roux are employees of IQVIA RWS France.

Ethical Approval

The study was approved by the French Data Protection Supervisory Authority “Commission Nationale Informatique et Libertés” (CNIL) (authorization no. 919119). As this study was a retrospective analysis of data from the SNDS, no informed consent was required and therefore not obtained. Retrospective analysis of pseudonymized data and therefore consent for publication is not required.