Bridging modelling estimates and primary data: diabetes burden in Italy through GBD 2023 estimates and the Piedmont Diabetes Registry

Introduction

Diabetes mellitus represents one of the leading non-communi cable diseases (NCDs) worldwide and constitutes a major and growing public health challenge. Over recent decades, the global burden of diabetes has substantially increased mainly due to population ageing, urbanisation and associated lifestyle changes, including changing dietary patterns, increasing overweight and obesity prevalence, and reduced physical activity.¹ Although type 1 diabetes mellitus (T1DM) and type 2 diabetes mellitus (T2DM) share the feature of hyperglycaemia, they are distinct conditions with different pathophysiology, population groups affected, therapeutic approaches, and outcomes. Both T1DM and T2DM contribute significantly to morbidity, premature mortality, disability, and healthcare expenditure.² The World Health Organization (WHO) has repeatedly highlighted the urgent need to strengthen diabetes prevention, surveillance, and integrated care strategies.^3,4

In Italy, diabetes represents a critical public health issue within the context of one of the oldest populations in Europe. An estimated 3.5-4 million individuals live with diabetes, corresponding to approximately 6-7% of the total population, with prevalence exceeding 20% among people aged 75 years and older.⁵ The Italian National Health Service (SSN), characterised by a highly decentralised regional organisation, presents substantial geographical heterogeneity in healthcare delivery, access to services, prevention strategies, and health outcomes, which are partly linked to individual and contextual socioeconomic inequalities.⁶

Diabetes surveillance in Italy currently relies on multiple sources, including administrative healthcare databases, regional disease registries, data from the Italian National Institute of Statistics (Istat), as well as national surveillance such as PASSI.⁵ Despite these multiple sources, important limitations remain. Surveillance sources are often fragmented and heterogeneous in methodology, limiting comparability across regions and over time.⁷ Self-reported surveys provide valuable information on diagnosed diabetes and population trends; however, like administrative data sources, they can only identify individuals who have already received a clinical diagnosis. Health examination surveys complement self-reported and administrative data sources by providing objective measurements of glycaemia and identifying undiagnosed diabetes. Nevertheless, they rely on representative sampling and participant engagement, which may limit the granularity and continuity of surveillance compared with routine administrative data systems.^7,8 The Annals of the Italian Association of Medical Diabetologists (Annali AMD), collecting data from voluntarily participating specialist centres, provide valuable information on quality of care, but are limited by selective participation and incomplete primary care coverage.^7,9

Italy has been a pioneer in diabetes legislation since Law No. 115/1987, considered the first national law specifically dedicated to diabetes worldwide.¹⁰ Subsequently, the Italian National Plan on Diabetic Disease (2012) further emphasised integrated care,¹¹ while Law 130/2023 introduced a voluntary screening programme for T1DM and coeliac disease in children¹². Nevertheless, despite the Decree of the President of the Council of Ministers (DPCM) of 3 March 2017 formally establishing the legal basis for a national diabetes registry, this has not yet been operationalised.¹³ This structural gap limits the possibility of obtaining a comprehensive, timely, and nationally comparable picture of diabetes epidemiology.

Among existing regional experiences, the Piedmont Diabetes Regional Registry (DRR), established in 1989, is one of the oldest and most comprehensive population-based diabetes registries in Italy.¹⁴ The DRR is based on clinical certification of diagnosis by a diabetologist, who specifies the diabetes type, granting patients exemption from co-payments for diabetes management. While the DRR has progressively integrated multiple data sources, including hospital discharge records and pharmaceutical prescriptions, allowing sustained monitoring of prevalence and incidence by diabetes type, sex, and age group, and provides high-quality, clinician-validated information, it reflects the diagnosed and specialist-registered population. Other regional experiences, including those in in Reggio Emilia and the Autonomous Province of Bolzano,^15,16 similarly demonstrate the feasibility of sustained regional surveillance, but these remain exceptions within a broader landscape of fragmented monitoring.

The Global Burden of Disease (GBD) Study, developed by the Institute for Health Metrics and Evaluation (IHME), provides internationally comparable epidemiological estimates across countries and subnational locations.¹⁷ GBD estimates are increasingly used to inform policy-making and resource allocation. However, their robustness depends on the quality and availability of primary data inputs, and continuous comparison with primary data is essential for improving model calibration and surveillance design.^18,19

Rather than treating modelled estimates and primary data as competing approaches, there is growing recognition that both should be integrated within an iterative process of mutual validation.¹⁹ This bidirectional relationship is particularly relevant in the Italian context, where a new administrative case-finding approach has recently been developed alongside the established clinical DRR.

The Joint Action on Cardiovascular Diseases and Diabetes (JACARDI), funded under the EU4Health Programme, is actively promoting collaboration on diabetes surveillance, data harmonisation, and interoperability across 21 European countries and Italian regions.²⁰ In Italy, JACARDI developed a preliminary stepwise administrative algorithm for the identification of T1DM, T2DM, and gestational diabetes mellitus (GDM) cases from routinely collected regional health data. This JACARDI preliminary algorithm, unlike the DRR, which is based on clinician-certified specialist registration, applies standardised criteria to multiple administrative data flows (hospital discharge records, pharmaceutical dispensing, and disease-related exemptions). In Piedmont, both the DRR and the JACARDI preliminary algorithm are available for the same population and period, allowing – together with GBD modelled estimates – an unprecedented three-source comparison.

To the Authors’ knowledge, no previous Italian study has systematically compared modelled estimates, clinical registry data, and administrative case-finding for the same population. This study aims to describe the burden of T1DM and T2DM in Italy using GBD 2023 estimates at national and macro-regional levels, both all-age and age-standardised, and to explore concordance and divergence in prevalence rate between GBD, the DRR, and the JACARDI preliminary algorithm for Piedmont, discussing the implications for future diabetes surveillance and modelling.

Methods

Study design and data sources

This study is based on three data sources:

1. GBD 2023 estimates for Italy and its macro-regions, extracted from the IHME GBD Results Tool, for the period 2010-2023;

2. data from the Piedmont DRR for the period 2010-2019;

3. estimates derived by applying the JACARDI preliminary algorithm to Piedmont regional health administrative databases for 2015-2019.

Five macro-regions were defined following the GBD 2023 classification: North-West (Piedmont, Aosta Valley, Liguria, Lombardy), North-East (Autonomous Province of Bolzano, Autonomous Province of Trento, Veneto, Friuli Venezia Giulia, Emilia-Romagna), Centre (Tuscany, Marche, Umbria, Lazio), South (Abruzzo, Molise, Campania, Apulia, Basilicata, Calabria), and Islands (Sicily, Sardinia). Detailed sex- and age-stratified subnational GBD 2023 estimates for Italy are published in a companion paper within the Italian GBD Initiative.⁶

GBD 2023 methodology for diabetes estimates

The GBD Study uses a comprehensive modelling framework to generate epidemiological estimates across countries and subnational locations.¹⁷ All methods follow the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) statement.²¹ Input data sources are publicly available on the Global Health Data Exchange (GHDx) catalogue.

Crucially, GBD does not model T2DM directly. T1DM and total diabetes are modelled separately; T2DM estimates are then derived by subtracting T1DM from total diabetes. This approach introduces additional methodological uncertainty for T2DM, particularly in settings with sparse or heterogeneous primary data.^17,19 The GBD defines T1DM as people who are diagnosed by physicians and identified through a diabetic registry or hospital records, and T2DM as fasting plasma glucose of at least 7 mmol/L (126 mg/dL) or those currently treated with drugs or insulin. It is essential to report that the GBD excludes population <15 years old for T2DM.¹⁹

Non-fatal estimates are generated using DisMod-MR, a Bayesian meta-regression tool ensuring internal consistency among epidemiological parameters.¹⁷ For each measure, 95% uncertainty intervals (UIs) were calculated with the 2.5^th and 97.5^th percentile ordered values from a 250-draw distribution.¹⁷ For this study, all-age and age-standardised prevalence and incidence rates per 100,000 population were extracted for T1DM and T2DM, both sexes combined, for Italy and the five macro-regions, for 2010 and 2023, together with the percentage change over this period.

For diabetes estimates in Italy, GBD 2023 integrates multiple epidemiological data sources within a standardized modelling framework. The GBD 2023 source repository for non-fatal diabetes outcomes in Italy includes 44 citations and 1,783 source metadata records,²² encompassing population-representative studies, surveys, administrative data, registries, and other epidemiological evidence.

Piedmont Diabetes Regional Registry (DRR)

The DRR was established in 1989 by regional law.¹⁴ Registration provides exemption from co-payments for medical devices and therapies required for diabetes management and is conditional on diagnosis being certified by a diabetologist who specifies the diabetes type. While the DRR represents a clinician-validated reference, registration requires active specialist engagement and may not fully capture individuals diagnosed and managed exclusively in primary care. DRR-based T1DM and T2DM prevalence (percentage of the general population, both sexes) for 2010-2019 was used as one of the two primary-data comparators.

JACARDI preliminary algorithm

Within JACARDI,²⁰ a preliminary stepwise hierarchical algorithm was developed by the Italian pilot projects for the identification of prevalent T1DM, T2DM, and GDM cases from routinely collected regional health data, applied to the resident population as of 31st December of each reference year. The algorithm integrates multiple administrative sources through a sequential decision process and is described in detail in Figure S1 (see online Supplementary Materials). In brief, it integrates data through four sequential steps:

• Step 1 (hospital discharge records) applies ICD-9-CM diagnostic codes recorded in any diagnostic field of hospital discharge records (SDO) in the preceding five years: codes 250.x1/250.x3 indicate T1DM and codes 250.x0/250.x2 indicate T2DM; when this criterion is unambiguous, classification is assigned directly;
• Step 2 (disease-related exemption) considers the presence of an active diabetes-related co-payment exemption in the reference year as a supporting criterion for prevalent diabetes, without distinguishing type;
• Step 3 (pharmaceutical dispensing pattern) evaluates the 12-month dispensing record from direct and convention pharmacies: T1DM classification is supported by ≥2 dispensations of insulin (ATC class A10A) in the estimation year, combined with no dispensation of non-insulin antidiabetic drugs (ATC class A10B) in the preceding five years. T2DM classification is supported by ≥2 dispensations of A10B agents. In the current preliminary version of the algorithm, individuals with dispensations of SGLT2 inhibitors (gliflozins) in the absence of other diabetes-related evidence (SDO codes, exemption, or other antidiabetic drug classes beyond gliflozins) are classified within T2DM; the potential misclassification of individuals receiving gliflozins for non-diabetic indications is acknowledged as a limitation of this preliminary approach;
• Step 4 (age at first administrative event) uses age at the earliest administrative event (hospital discharge, pharmaceutical dispensing, or exemption) as a supporting criterion: age <15 years strongly supports T1DM classification, age ≥30 years supports T2DM, while ages 15-29 years are non-discriminant. Cases with discordant or insufficient criteria after all steps are classified as ‘undetermined type’. Emergency department records were not included as a classification criterion in this preliminary version of the algorithm. For GDM identification, a separate algorithm was applied to female residents of reproductive age, using pregnancy-related ICD-9-CM codes (648.8x) recorded during hospitalisation for delivery or in the 90 days preceding delivery (identified through ICD-9-CM 630-679 and/or DRG 370-375), combined with the absence of pre-existing diabetes evidence in the five years preceding the estimated conception date. For this study, JACARDI preliminary algorithm-based estimates of T1DM and T2DM prevalence in Piedmont, expressed as a percentage of the total resident population (both sexes combined), for the period 2015-2019 were used as the second primary-data comparator. The algorithm period starts in 2015, because the pharmaceutical dispensing lookback window requires five years of administrative data, making 2015 the earliest year for which complete lookback from 2010 is available. 

Statistical analyses and comparison

National and macro-regional GBD 2023 prevalence and incidence estimates are presented for 2010 and 2023, as all-age and age-standardised rates per 100,000 population, with percentage change and 95%UI. The comparison between GBD and primary-data sources was restricted to all-age prevalence (expressed as a percentage of the general population), both sexes. GBD was compared with the DRR for 2010-2019, and with the JACARDI preliminary algorithm for 2015-2019. Years 2020-2021 were excluded from all primary-data comparisons due to known disruptions in healthcare utilisation and data completeness during the COVID-19 pandemic. To facilitate interpretation across different approaches to diabetes surveillance and estimation, GBD estimates were used as a common reference against which the JACARDI and DRR estimates were compared. This approach was intended to explore how different data sources and methodological frameworks capture the same epidemiological phenomenon rather than to establish the relative accuracy or completeness of any of the three sources. Absolute differences in percentage points are presented as the primary comparison metric in Tables 3a and 3b, as they directly quantify population-level case-count discrepancy relevant for surveillance planning. Relative differences (%), which are more directly comparable across conditions with different background prevalences, are provided in Table S1a ans S1b (see online Supplementary Materials). Conventional 95% confidence intervals were not estimated for DRR and JACARDI, as both approaches derive estimates from population-based registry and administrative data rather than probability-based samples. In this setting, sampling-based measures of uncertainty would capture only a limited component of the overall uncertainty, while leaving unquantified relevant sources related to case ascertainment, data linkage, and case definition. Accordingly, differences between sources were interpreted descriptively and not subjected to formal statistical inference. The comparison was not intended as a formal validation exercise, but as an exploratory analysis to assess concordance and identify opportunities for mutual improvement across modelling, clinical registry, and administrative case-finding approaches.

Results

National burden of diabetes in Italy: GBD 2023 estimates

GBD 2023 estimates confirm that diabetes represents a substantial and growing public health burden in Italy (Figures 1 and 2).

Despite relatively wide 95% UIs, T1DM prevalence showed a relative increase of 0.18 in both all-age and age-standardised terms, with the corresponding 95% UIs excluding zero. National all-age T1DM prevalence was estimated to increase from 288.8 (95% UI 251.8-334.9) per 100,000 in 2010 to 340.5 (95% UI 299.9-384.7) in 2023 (relative change 0.18, 95%UI 0.12-0.27), and age-standardised prevalence showed an almost identical relative increase (245.7 to 289.4 per 100,000; relative change 0.18, 95% UI 0.08-0.32) (Table 1a and 1b). This parallel increase in both all-age and age-standardised T1DM prevalence, in the context of stable incidence across all macro-regions (see paragraph “Macro-regional differences”) is consistent with progressive improvements in survival and life expectancy among people living with T1DM.

T1DM incidence remained comparatively stable over time, with wide uncertainty intervals compatible with both modest increases and decreases (all-age: 8.7 to 9.0 per 100,000, relative change 0.04; 95%UI -0.24 to 0.40; age-standardised: 10.9 to 11.5, relative change 0.06) (Table 1c and 1d). At national level, all-age T2DM prevalence increased from 6,362.4 per 100,000 (95%UI 5,668.8-7,010.9) in 2010 to 7,082.5 (95%UI 6,275.5-7,900.5) in 2023, a relative increase of 0.11 (95%UI 0.08-0.14). Age-standardised T2DM prevalence showed little change over the same period (3,763.4 in 2010 vs 3,706.2 in 2023; relative change -0.02; 95%UI -0.04 to 0.01). All-age T2DM incidence increased from 257.1 to 272.0 per 100,000 (relative change 0.06; 95%UI 0.01-0.10), whereas age-standardised incidence remained broadly stable, with a slight decline in the point estimate from 186.6 to 182.9 per 100,000 (relative change -0.02; 95%UI -0.04 to 0.01]) (Tables 2a and 2d).

Macro-regional differences

The geographical patterns differ between the two diabetes types (Tables 1 and 2; Figures S2 and S3, online Supplementary Materials). For T1DM, prevalence and incidence are broadly comparable across most macro-regions, with the Islands representing a clear outlier due to the well-documented exceptionally high T1DM incidence in Sardinia, among the highest in Europe. For T2DM, a continuous gradient of increasing burden is observed from North-West to Islands.

For T1DM in 2023, all-age T1DM prevalence in the Islands reached 526.7 per 100,000 (95%UI 463.4-583.9), compared with 282.5 (95%UI 252.3-314.3) in the North-West, the lowest among macro-regions (Table 1a). T1DM incidence showed limited variation across macro-regions and modest relative changes, with wide uncertainty intervals reflecting limited precision at the subnational level (Table 1c and 1d). Unlike T2DM, the relative increase in T1DM prevalence since 2010 was broadly similar across macro-regions (ranging from 0.17 in the North-West and Islands to 0.20 in the South and Islands), indicating a more geographically homogeneous upward trend, despite persistent differences in absolute prevalence levels (Table 1a). For T2DM, in 2023 all-age prevalence ranged from 5,583.6 per 100,000 (95%UI 4,920.5-6,282.7) in the North-West to 10,197.0 (95% UI 9,187.0-11,394.7) in the Islands (Table 2a). The South (8,129.6; 95%UI 7,220.6-9,106.4) and the Islands also showed the largest relative increases in all-age T2DM prevalence since 2010 (+0.15 and +0.22, respectively), compared with +0.06 in the North-West (Table 2a). After age standardisation, absolute geographical differences persisted (Islands: 5,334.6 vs North-West: 2,876.6 per 100,000 in 2023). Age-standardised T2DM prevalence showed a modest point increase in the Islands (+0.03; 95%UI 0.0-0.07) (Table 2b), though uncertainty intervals largely overlap with those of other macro-regions. Age-standardised T2DM incidence showed no statistically significant change in any macro-region, including the Islands (+0.03; 95% UI -0.02 to 0.07) (Table 2d).

Comparison between GBD 2023, the Piedmont DRR, and the JACARDI preliminary algorithm

Table 3a and 3b presents the comparison between GBD 2023, the DRR (2010-2019), and the JACARDI preliminary algorithm (2015-2019) for all-age T1DM and T2DM prevalence (percentage of general population, both sexes) in Piedmont. Relative differences are provided in Table S1a and S1b (see online Supplementary Materials). Figure 3 shows the corresponding trends graphically.

The DRR provided T1DM prevalence estimates below those of GBD across the entire 2010-2019 period. DRR-based T1DM prevalence remained essentially stable at 0.21-0.22% throughout the decade, while GBD estimates increased from 0.27% (95%UI 0.24-0.30) to 0.30% (95%UI 0.26-0.32).

The JACARDI preliminary algorithm, available for 2015-2019, yielded T1DM estimates above those of GBD in all available years: 0.35-0.36% (algorithm) vs 0.28-0.30% (GBD).

The DRR provided T2DM prevalence estimates consistently below those of GBD throughout 2010-2019. DRR-based T2DM prevalence increased gradually from 4.53% in 2010 to 5.06% in 2019, remaining below both GBD and JACARDI preliminary algorithm estimates by approximately 1.0-1.3 percentage points throughout the period.

The JACARDI preliminary algorithm, available for 2015-2019, provided T2DM estimates slightly below but close to GBD: 5.72% (2015) to 6.03% (2019) vs GBD 5.91-6.40%. Critically, all algorithm-derived T2DM estimates fell within the corresponding GBD 95% uncertainty intervals throughout the 2015-2019 period, showing descriptive concordance with the GBD estimates. Overall, the JACARDI preliminary algorithm and GBD estimates displayed highly similar T2DM prevalence trajectories throughout the 2015–2019 period, whereas the DRR consistently yielded lower estimates.

Discussion

This study presents the GBD 2023 burden of T1DM and T2DM in Italy at national and macro-regional levels, and a novel three-source comparison for Piedmont between GBD, the DRR, and the JACARDI preliminary algorithm. Four key findings emerge. First, GBD 2023 data confirm the well-documented North-South-Islands gradient in Italy’s diabetes burden. Second, the observed increase in all-age T2DM prevalence at the national level is likely influenced by population ageing, whereas age-standardised estimates suggest little change over time. Third, the JACARDI preliminary algorithm identifies a number of T2DM cases broadly comparable with GBD estimates, while the DRR captures a lower proportion of the estimated burden; however, no source is treated as a gold standard and the observed differences may reflect case definitions and ascertainment mechanisms. Fourth, estimates of T1DM differ across data sources in both magnitude and temporal patterns, reflecting the methodological challenges of distinguishing diabetes type using routinely collected administrative data, particularly in the context of small numbers and different case definitions. These findings highlight the uncertainty inherent in diabetes burden estimation and the value of triangulating complementary data sources for diabetes surveillance.

The persistence and, for T2DM, the widening of the North-South-Islands gradient is broadly consistent with evidence on socioeconomic and lifestyle-related determinants of T2DM risk in Italy and with European trends.^1,2,6 The divergence between all-age and age-standardised temporal trends at the national level suggests that population ageing may substantially contribute to the observed increase in all-age T2DM prevalence over time. In contrast, the marked cross-regional differences in age-standardised T2DM prevalence persist after accounting for differences in population age structure, indicating geographical heterogeneity that cannot be explained by demographic composition alone. These differences may reflect a combination of variations in underlying risk factors, socioeconomic conditions, healthcare access, and other contextual determinants, which merit further investigation. However, given the uncertainty in the estimates, these patterns should be interpreted cautiously and monitored over time. These persistent geographical inequalities have direct policy implications. Meeting the WHO Global Diabetes Compact targets, including at least 80% diagnostic coverage by 2030, and monitoring progress towards Sustainable Development Goal target 3.4³ require surveillance systems capable of capturing subnational heterogeneity and identifying inequalities in disease burden, diagnosis, and care. The WHO European NCD Roadmap (2022-2027) similarly emphasises the importance of strengthening surveillance systems and improving data disaggregation, including by geography, to enable the monitoring of health inequalities across populations.²³ In this context, Italy’s decentralised health system and the documented North-South-Islands gradient underscore the need for proportionate and regionally responsive investments in surveillance and prevention.

An important observation regarding temporal trends is that age-standardised T2DM prevalence and incidence remained broadly stable between 2010 and 2023 at the national level, with similar patterns across most Italian macro-regions. Over a longer time horizon, previous GBD analyses have reported an increase in age-standardised T2DM prevalence across European Union countries, although national trajectories have varied in both magnitude and direction across settings.^1,24 Differences in the time periods considered and GBD rounds, however, limit direct comparison with the present estimates. The relative stability observed in Italy may reflect the interplay of changes in underlying risk factors, diagnosis, and clinical management, although these mechanisms cannot be disentangled within the present analysis. In the Islands, age-standardised T2DM burden showed a modest increase (0.03; 95%UI 0.00-0.07), although interpretation should be cautious given overlapping uncertainty intervals. Further monitoring is warranted to assess whether this pattern persists over time. From a public health perspective, the broadly stable age-standardised incidence and prevalence of T2DM suggest that population-level risk and disease burden, after accounting for changes in population age structure, have not substantially increased over the study period. However, the concurrent increase in all-age prevalence indicates that demographic ageing is likely to translate into a growing absolute number of people living with T2DM, even in the absence of a marked increase in age-specific risk. This has important implications for both prevention and health-system planning, underscoring the need to sustain efforts to reduce modifiable risk factors while strengthening health-system capacity to manage an increasingly ageing population with diabetes.

The three-source comparison for Piedmont provides methodologically informative insights. For T2DM, the observed proximity between estimates from the JACARDI preliminary algorithm and GBD (absolute differences of +0.17 to +0.37 percentage points, with algorithm estimates falling within GBD uncertainty intervals) indicates broadly comparable estimates between the two approaches. This should not be interpreted as independent validation of GBD, as GBD estimates are generated through a modelling framework that synthesises multiple heterogeneous data sources, whereas JACARDI preliminary algorithm applies a specific deterministic case-finding algorithm to regional administrative data. The two approaches are therefore methodologically distinct, even where some underlying administrative information may overlap. By contrast, the DRR, which relies on active enrolment by diabetologists, yields lower estimates (approximately 1.0-1.5 percentage points below both GBD and the algorithm across the study period). This difference may reflect variation in case ascertainment, but other explanations, including differences in case definitions, inclusion criteria, and registry enrolment, cannot be excluded. The historical DRR evaluation reported incomplete capture of people identified through hospital or pharmaceutical records, but those estimates predate the present study period and therefore cannot be used to quantify current completeness.¹⁴ Overall, the lower DRR estimates should be interpreted as reflecting differences between data sources in population coverage and case ascertainment, rather than as evidence that DRR underestimates the burden.

For T1DM, the pattern is more heterogeneous, as the three data sources differ in both magnitude and direction. The DRR yields estimates lower than GBD (absolute difference +0.06 to +0.08), while the JACARDI preliminary algorithm yields estimate higher than GBD (absolute difference -0.05 to -0.07). In relative terms, these differences are larger because T1DM is substantially less prevalent than T2DM. The divergence between approaches likely reflects differences in case ascertainment rather than a single source representing a definitive benchmark.

The current T1DM definition used in the preliminary algorithm (≥2 A10A dispensations in the index year and no A10B dispensations in the preceding five years) is intended to approximate insulin dependence. However, this approach may also capture insulin-treated patients with type 2 diabetes who have no recorded history of oral antidiabetic use. This represents a recognised limitation of administrative case-finding algorithms, which may have limited specificity when distinguishing T1DM from insulin-treated T2DM using routinely collected administrative data. Conversely, the lower DRR estimates compared with both GBD and the algorithm are unlikely to be explained solely by incomplete clinical registration, given the high level of T1DM ascertainment achieved by the DRR in Piedmont.²⁵ Instead, they may reflect differences in case definitions, methodological approaches, and the geographic composition of GBD national estimates. In addition, GBD national estimates incorporate regional heterogeneity, including areas with particularly high T1DM prevalence (such as Sardinia), which may influence national-level comparisons. Overall, these findings reinforce the value and need for the formal validation process that JACARDI is actively conducting, comparing algorithm-identified T1DM cases with DRR-certified cases at the individual level will be essential to refine classification criteria and improve algorithm accuracy.

More broadly, this three-source comparison illustrates the value of triangulating modelled estimates with complementary primary-data approaches to improve the interpretability of diabetes surveillance. Persistent discrepancies across sources are informative and may reflect differences in case definitions, data completeness, care pathways, and population coverage. In this sense, divergence between systems can be used as a signal for methodological refinement and surveillance strengthening.¹⁹

For T2DM, the overall proximity between the JACARDI preliminary algorithm and GBD estimates suggests a broadly comparable magnitude of the estimated burden, although the absence of a gold standard precludes any definitive assessment of accuracy. In settings where clinical registries are incomplete, administrative data algorithms may represent a useful complementary source for population-level surveillance, particularly when based on multiple data streams. It should be noted that the small absolute differences observed for T1DM primarily reflect the low overall prevalence of this condition rather than better methodological agreement across sources. In relative terms, the divergences between DRR, JACARDI preliminary algorithm, and GBD for T1DM are larger than, those observed for T2DM (Table S1, online Supplementary Materials). Rather, the opposing directions of divergence (DRR below GBD, algorithm above) point to structurally different case ascertainment mechanisms. This pattern generates testable hypotheses on specialist follow-up completeness, pharmaceutical proxies for insulin-dependent disease, and the contribution of Sardinia’s disproportionately high T1DM rates to national GBD estimates, hypotheses that the individual-level validation studies currently underway within JACARDI are designed to address.

From a public health perspective, these findings reinforce the importance of integrated surveillance systems that combine modelling approaches with routinely collected administrative and clinical data.¹⁹ The legal and organisational foundations for a national diabetes surveillance system are already in place,¹³ but their effectiveness depends on interoperability, standardised case definitions, and consistent implementation across regions. Within this context, JACARDI is supporting the development and harmonisation of regional diabetes surveillance systems across pilot projects from different regions, using a shared methodological framework that includes the administrative algorithm evaluated in this study. Rather than providing definitive estimates, this algorithm should be viewed as part of an iterative refinement process aimed at progressively improving completeness, comparability, and interpretability of diabetes burden estimates across data sources. Finally, the persistent North-South-Islands gradient underscores the need to read all surveillance findings through an equity lens, ensuring that prevention, diagnostic capacity, and healthcare resources are proportionate to population needs across regions.

Strengthening diabetes surveillance is essential for progress toward the WHO Global Diabetes Compact targets, including achieving at least 80% diagnostic coverage by 2030, and for monitoring implementation of the WHO NCD roadmap.^3,23 Meeting these targets requires interoperable, equity-oriented surveillance systems capable of identifying undiagnosed populations, geographic disparities, and gaps in care quality, of the kind that the combination of GBD modelling, clinical registries, and validated administrative algorithms can provide when used together.

Limitations

This study has several limitations. First, DRR and JACARDI preliminary algorithm are presented without uncertainty intervals, as only aggregate estimates were available for this analysis. Although both sources cover the entire population residing in Piedmont, even complete-population estimates are subject to stochastic process variability; the absence of uncertainty intervals, therefore, limits direct statistical comparability with GBD uncertainty intervals. Uncertainty quantification for primary data sources will be addressed in the forthcoming individual-level validation phase. Second, primary-data comparisons are restricted to Piedmont, a single region of approximately 4.3 million inhabitants. While Piedmont provides a uniquely data-rich setting for three-source comparison, its epidemiological profile and the heterogeneity of surveillance infrastructure across regions may limit generalisability to other Italian settings, particularly in the South and Islands where T2DM burden is higher. Third, the comparison was limited to all-age prevalence for both sexes combined; sex- and age-stratified analyses were not conducted, readers are directed to Zamagni et al. (2026)⁶ for detailed stratified subnational GBD 2023 data. Fourth, the study does not include an independent external gold standard and therefore cannot determine which data source most accurately reflects the true prevalence of diabetes. Fifth, GBD 2023 subnational estimates carry wider uncertainty intervals than national estimates. Sixth, the JACARDI preliminary algorithm has not yet undergone formal validation, and the T1DM results, in particular, should be interpreted as preliminary findings from an ongoing refinement process. Seventh, the pharmaceutical lookback window of five years required by the algorithm means that the comparison with GBD is restricted to 2015–2019, limiting comparability over time for this specific analysis. Eighth, in the current preliminary version of the algorithm, individuals dispensed SGLT2 inhibitors(gliflozins) or GLP-1s without additional diabetes-related evidence (hospital discharge diagnoses, exemptions, or other antidiabetic drug classes) are classified within T2DM; this may introduce potential misclassification within T2DM classification. Ninth, differences in case definitions and ascertainment mechanisms across GBD, the DRR and the administrative algorithm may affect the comparability of estimates, as each source relies on different inclusion criteria and methodological approaches. Finally, individual-level linkage between DRR- and algorithm-identified cohorts, currently ongoing within JACARDI, will enable the assessment of overlap and discordant cases. In the absence of a gold standard, the analysis will focus on agreement between the two approaches and characterization of discordant cases, rather than quantifying classification accuracy. 

Conclusions

Diabetes mellitus remains a major and growing public health challenge in Italy, with a persistent and, for T2DM, widening North-South-Islands gradient. GBD 2023 estimates suggest that the increase in all-age T2DM prevalence at the national level is largely consistent with population ageing, while in the Islands a slight rise in age-standardised estimates may suggest higher underlying burden, warranting continued attention, monitoring and warrant cautious interpretation.

The three-source comparison in Piedmont illustrates the value of integrating modelled estimates, clinical registries, and administrative algorithms to better characterise diabetes burden. For T2DM, the multisource administrative algorithm developed within JACARDI yields estimates broadly comparable in magnitude to GBD. Rather than indicating that one source is superior to another, the observed differences highlight how alternative data sources and methodological approaches may capture different dimensions of the same underlying burden, reflecting differences in case definitions, case ascertainment, and population coverage. For T1DM, divergence across the three sources, although modest in absolute terms, highlights the methodological challenges of distinguishing diabetes type using administrative data and underscores the importance of ongoing formal validation, currently being undertaken. 

More broadly, the central message of this study is that modelled estimates and primary data sources should be interpreted as complementary rather than competing tools. Their systematic and reciprocal comparison provides an opportunity to identify inconsistencies, improve interpretability, and iteratively strengthen surveillance systems. Divergences across sources should therefore be viewed not as evidence of failure of any single system, but as signals that can inform methodological refinement and data system improvement. The ongoing work within JACARDI to develop, pilot, and formally validate harmonised, interoperable, and multisource approaches to diabetes surveillance across Italian regions and local authorities represents a concrete step toward this goal. When successfully validated and implemented across different regional settings, this approach could provide a reproducible basis for monitoring territorial inequalities, supporting need-based resource allocation, evaluating care pathways, and informing the future development of a national diabetes surveillance system. The comparisons presented in this study provide an empirical basis for that iterative validation process and for the future development of more integrated diabetes surveillance frameworks in Italy and Europe.

Conflicts of interest: none declared.

Funding: this work was supported by the Italian Ministry of Health, through the contribution given to the Institute for Maternal and Child Health IRCCS Burlo Garofolo, Trieste (Italy). The JACARDI project has received funding from the EU4Health Program 2021-2027 under Grant Agreement No. 101126953. The views expressed are those of the authors and do not necessarily reflect those of the European Union or the European Executive Agency for Health and Digital (HaDEA).

Ethic approval: the study is based on publicly available aggregated data and on anonymized and aggregated registry and administrative data. Approval by an ethics committee is not required.

AI use: artificial intelligence tools were used to assist with the linguistic review and structuring of the manuscript. All data analysis, interpretations, and scientific responsibility remain solely with the authors.

References

1. GBD 2021 Diabetes Collaborators. Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. Lancet 2023;402(10397):203-34. doi:10.1016/S0140-6736(23)01301-6
2. International Diabetes Federation. The Diabetes Atlas, 10th edition. Brussels: IDF; 2021. Available from: www.diabetesatlas.org
3. World Health Organization. Global action plan for the prevention and control of noncommunicable diseases 2013-2030. Geneva: WHO; 2013. Available from: https://apps.who.int/iris/handle/10665/94384
4. World Health Organization. First-ever global coverage targets for diabetes adopted at the 75th World Health Assembly. Geneva: WHO; 2022. Available from: https://www.who.int/news-room/feature-stories/detail/first-ever-global-coverage-targets-for-diabetes-adopted-at-the-75-th-world-health-assembly
5. Italian Ministry of Health. Relazione 2024: stato delle conoscenze e delle nuove acquisizioni in tema di diabete mellito. Rome: Ministero della Salute; 2025. Disponibile all’indirizzo: https://www.salute.gov.it/new/sites/default/files/2025-10/Relazione%20diabete%202024.pdf
6. Zamagni G, Armocida B, Castelpietra G, et al. The burden of disease in Italy, 1990-2023: a subnational analysis from the Global Burden of Disease Study 2023. Epidemiol Prev 2026;50(1):67-77. doi:10.19191/EP26.1.A991.020
7. WHO Regional Office for Europe. Measuring the global diabetes targets in the WHO European Region: Italy case study. Copenhagen: WHO Regional Office for Europe; 2024. Available at: https://iris.who.int/server/api/core/bitstreams/0edfea7c-fd09-4b5f-b3cf-20f37afd6c2f/content
8. Tolonen H, Koponen P, Mindell J, Männistö S, Kuulasmaa K. European Health Examination Survey – towards a sustainable monitoring system. Eur J Public Health 2014;24(2):338-44. doi:10.1093/eurpub/ckt107
9. Russo G, Di Bartolo P, Candido R, et al. The AMD Annals: a continuous initiative for the improvement of type 2 diabetes care. Diabetes Res Clin Pract 2023;199:110672. doi:10.1016/j.diabres.2023.110672. Erratum in: Diabetes Res Clin Pract 2024;210:111619. doi: 10.1016/j.diabres.2024.111619
10. Italian Presidency of the Council of Ministers. Italian Law No. 115 of 16.03.1987. Disposizioni per la prevenzione e la cura del diabete mellito. Gazzetta Ufficiale No. 71 of 26.03.1987. Available from: https://www.normattiva.it/uri-res/N2Ls?urn:nir:stato:legge:1987;11
11. Italian Ministry of Health. Piano sulla malattia diabetica. Rome: Italian Ministry of Health; 2012. Available from: https://www.salute.gov.it/imgs/C_17_pubblicazioni_1885_allegato.pdf
12. Italian Law No. 130 of 15.09.2023. Gazzetta Ufficiale No. 226 of 27.09.2023. Available from: https://www.gazzettaufficiale.it/eli/id/2023/09/27/23G00140/sg
13. Italian Prime Minister Decree (DPCM) of 03.03.2017. ‘Identificazione dei sistemi di sorveglianza e dei registri di mortalità, di tumori e di altre patologie’. Gazzetta Ufficiale No. 109 of 12.05.2017. Available from: https://www.gazzettaufficiale.it/eli/id/2017/05/12/17A03142/sg
14. Gnavi R, Karaghiosoff L, Dalmasso M, Bruno G. The Piedmont Diabetes Register: opportunities and limits for epidemiological purposes. Epidemiol Prev 2006;30(1):59-64.
15. Local Health Unit of Reggio Emilia. Registro Diabete (Diabetes Registry). Available from: https://www.ausl.re.it/Luogo.jsp?id=7099
16. Autonomous Province of Bolzano-South Tyrol. Diabete (Diabetes). Available from: https://osservatorio-salute.provincia.bz.it/it/registro-diabete
17. GBD 2023 Disease and Injury and Risk Factor Collaborators. Burden of 375 diseases and injuries, risk-attributable burden of 88 risk factors, and healthy life expectancy in 204 countries and territories, including 660 subnational locations, 1990-2023: a systematic analysis for the Global Burden of Disease Study 2023. Lancet 2025;406(10513):1873-922. doi:10.1016/S0140-6736(25)01637-X
18. Bak JCG, Serné EH, Kramer MHH, Nieuwdorp M, Verheugt CL. National diabetes registries: do they make a difference? Acta Diabetol 2021;58(3):267-78. doi:10.1007/s00592-020-01576-8
19. Armocida B, Sawyer SM, Monasta L, et al. Enhancing Global Burden of Disease Estimates with Collaborative Data Insights: A Case Study of Type 1 Diabetes in Finland. J Adolesc Health 2026;78(4):662-68. doi:10.1016/j.jadohealth.2025.11.023
20. Armocida B, Formenti B, Silano M, et al. Tackling the challenge of cardiovascular diseases and diabetes across Europe: a joint action by more than 300 public health professionals. Commentary. Ann Ist Super Sanita 2024;60(1):4-7. doi:10.4415/ANN_24_01_02
21. Stevens GA, Alkema L, Black RE, et al. Guidelines for Accurate and Transparent Health Estimates Reporting: the GATHER Statement. PLoS Med 2016;13(6):e1002056. doi: 10.1371/journal.pmed.1002056
22. Global Burden of Disease Study 2023 (GBD 2023) Sources. Available from: https://sources.healthdata.org/collection/sources-2023?components=5&causes=587&impairments=0&locations=86
23. WHO Regional Office for Europe. Reducing noncommunicable diseases: a signature roadmap for the WHO European Region. Copenhagen: WHO Europe; 2022. Available from: https://www.who.int/europe/publications/i/item/WHO-EURO-2022-6620-46386-67147
24. Goodall R, Alazawi A, Hughes W, et al. Trends in type 2 diabetes mellitus disease burden in European Union countries between 1990 and 2019. Sci Rep 2021;11(1):15356. Published 2021 Jul 28. doi:10.1038/s41598-021-94807-z
25. Gesuita R, Rabbone I, Marconi V, et al. Trends and cyclic variation in the incidence of childhood type 1 diabetes in two Italian regions over 33 years and during the COVID-19 pandemic. Diabetes Obes Metab 2023;25(6):1698-703. doi:10.1111/dom.15024