Redistribution of garbage codes to underlying causes of death: a systematic analysis on Italian data based on the Global Burden of Disease Study 2023

Abstract

Objectives: to describe the garbage codes (GCs) identified in the Global Burden of Disease Study (GBD) 2023 and their redistribution to underlying causes of death in Italy in 2021. Specifically, the study aims to: 1. compare temporal trends in the proportion of GCs in Italy with those of other Western European countries with similar population sizes; 2. identify the most frequent GC packages, analyze their geographic distribution, and determine the causes to which they are reassigned; 3. examine the relationship between the proportion of GCs and contextual factors related to death certification, including the type of certifier, place of death, and requests for autopsy.

Design: descriptive epidemiological study based on GBD 2023 estimates.

Setting and participants: the analysis focuses on the Italian population in 2021, stratified by 21 subnational units, including 19 regions and 2 autonomous provinces.

Main outcome measures: total number of deaths; number of GCs, defined as deaths attributed to causes that do not represent underlying causes of death; percentage of GCs, calculated as the number of GCs divided by total deaths and multiplied by 100.

Results: the proportion of GCs in Italy gradually decreased over time, from 34.6% in 1990 to 28.8% in 2021. In 2021, the three most frequent GC packages at the national level were ‘unspecified type of stroke’ (4.28% of total deaths), ‘unspecified type of diabetes’ (2.44%), and ‘heart failure, right or left’ (2.38%). In the same year, the proportion of GCs was positively correlated with the share of deaths occurring at home (r 0.71; p <0.001), with missing data on the type of certifying physician (r 0.54, p=0.020), on place of death (r 0.77, p <0.001), and on autopsy requests (r 0.76, p <0.001).

Conclusions: misreporting of causes of death arises from multiple mechanisms, reflecting errors of different nature and severity, with important implications for public health policies and health information systems. While redistribution methods are essential to produce comparable and policy-relevant estimates, improving data quality at the source remains a critical priority.

 Keywords: Global Burden of Disease Study, causes of death, garbage codes

Introduction

Accurate cause-of-death reporting is essential for monitoring population health, informing public health policies on disease prevention and control, and guiding the allocation of healthcare resources.¹ However, it remains a significant challenge, as mortality data are often affected by misreporting in death certification, and a substantial proportion of deaths remains poorly defined.^2,3 This issue commonly manifests in the occurrence of the so-called garbage codes (GCs), defined as causes of death reported on death certificates that do not represent a valid underlying cause.^4,5 Based on the definition provided by the World Health Organization (WHO), the underlying cause of death is the disease or injury that initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence that produced the fatal injury.⁶ In this sense, typical examples of invalid underlying causes of death are senility, which is an ill-defined condition, and heart failure, which represents a terminal mechanism.

The reliability of cause-of-death coding is crucial to ensure the comparability of mortality data at the international level. To this end, the WHO establishes standards for the classification and reporting of causes of death, primarily through the International Classification of Diseases (ICD), to enable consistent tracking of mortality worldwide.⁷ Currently, most countries still rely on the tenth revision of the ICD (ICD-10), which was originally designed for paper-based reporting systems. However, many countries are progressively transitioning, though at different speeds, to ICD-11, which is a fully digital system designed to improve clinical specificity and support automated coding and validation processes. 

Although the Italian healthcare system is organized at the regional level, with substantial autonomy in the delivery of services, cause-of-death coding and mortality statistics are centrally managed by the Italian National Institute of Statistics (Istat). Cause-of-death data are derived from death certificates completed by physicians, who are required to report the full sequence of conditions leading to death, including the immediate cause, any intermediate conditions, and the underlying disease or injury that initiated the fatal process.⁶ These data are then processed by Istat using an automated coding system based on the ICD-10, ensuring standardized identification of the underlying cause of death.

Despite these efforts, GCs remain a persistent feature of mortality data.³ This highlights the need not only to identify and quantify these codes, but also to implement strategies to mitigate their impact on mortality statistics. For example, evidence-based redistribution algorithms can help reassign deaths originally coded as GCs to the most plausible underlying causes, thereby improving the reliability of mortality estimates.⁸ In line with this, the Global Burden of Disease (GBD) Study applies evidence-based redistribution methods to produce cause-specific mortality estimates by systematically reallocating GCs.⁹ Specifically, redistribution draws on multiple approaches, including proportional reallocation, cause-specific algorithms, regression-based models, reference datasets such as verbal autopsy and clinical data, and epidemiological constraints applied within a hierarchical framework.

This study aims to describe GCs identified in GBD Study 2023 and their redistribution to underlying causes of death in Italy and its subnational locations. Specifically, it aims to: 1. compare temporal trends in GC proportions in Italy with those found in other Western European countries with similar population sizes, and across Italian regions from 1990 to 2021; 2. identify the most frequent GC packages and examine their geographical distribution, as well as the causes to which they are reassigned; 3. analyse the relationship between GC proportions and selected contextual factors related to death certification, including type of certifier, place of death, and autopsy practices.

Methods

Overview

The GBD 2023 Study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).¹⁰ Within the GBD framework, GCs include conditions that fail to identify the underlying cause of death. These arise through two broad mechanisms of misreporting. The first involves the assignment of intermediate or immediate conditions as the underlying cause. The second involves the use of non-specific or insufficiently detailed diagnoses, where the underlying cause is broadly identified, but lacks the clinical specificity required for accurate classification. 

In the GBD framework, GCs are classified into four categories based on their policy relevance and on the levels of the GBD cause hierarchy to which they can be reassigned. These categories also reflect the degree to which GCs may distort public health interpretation of cause-of-death patterns. Specifically:

• Class 1 (codes with serious policy implications) includes GCs that may be redistributed to any underlying cause of death across all three Level 1 cause groups (i.e., communicable diseases, non-communicable diseases, and injuries); these codes represent the highest level of uncertainty and have the greatest potential to mislead health policy priorities;
• Class 2 (codes with substantial policy implications) includes GCs that can be reassigned within one or two of the Level 1 cause groups; although more informative than Class 1, they still introduce considerable ambiguity in distinguishing broad cause categories;
• Class 3 (codes with important policy implications) comprises GCs for which the underlying cause is likely to fall within the same ICD chapter, therefore providing more constrained information, but still limiting precise cause attribution;
• Class 4 (codes with limited policy implications) contains GCs for which the underlying cause can typically be attributed to a single specific condition.

Further details on the methods and algorithms used for the redistribution of GCs to underlying causes of death are described elsewhere.⁹ 

Statistical analysis

To quantify the overall extent of misclassification in mortality systems, all analyses were conducted using estimates of all-age pre- and post-redistribution death counts for both sexes combined. The percentage of GCs was calculated as the number of deaths assigned to GCs divided by the total number of deaths, multiplied by 100.

To contextualize Italy within comparable settings, GC proportions were compared over time with those of Western European countries with populations exceeding 10 million (i.e., Belgium, France, Germany, Greece, the Netherlands, Portugal, Spain, Sweden, and the United Kingdom), following the methodological approach adopted in previous analyses.³ To capture long-term trends, GC percentages are presented at five-year intervals from 1990 to 2020, with 2019 included as a pre-pandemic reference and 2021 as the most recent year of analysis. The choice of 2021 reflects the latest year for which Istat cause-of-death data were available to the GBD 2023 study, allowing for the identification and redistribution of GCs.

A similar comparison was performed across 19 Italian regions and 2 autonomous provinces (AP): Piemonte, Valle d’Aosta, Lombardia, AP Bolzano, AP Trento, Veneto, Friuli Venezia Giulia, Liguria, Emilia-Romagna, Toscana, Umbria, Marche, Lazio, Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicilia, and Sardegna.

To estimate the average long-term trend in GC percentages across Italian locations, a linear mixed-effects model was used, with year included as a fixed effect. Year was included as a continuous variable and centred at 1990. Location-specific heterogeneity was modelled by including random intercepts and random slopes for year at the location level allowing for covariance between the random intercept and slope. Specifically, the random intercept for location captures between-location variability in baseline GC percentages, while the random slope for year allows the temporal trend to differ across locations. Results were presented as the fixed-effect regression coefficient for year and as estimated standard deviations (SDs) of the random intercepts and slopes, with corresponding 95% confidence intervals (95%CIs).

To explore patterns in cause-of-death misclassification, the three leading GC packages in 2021 and their proportions relative to total deaths were identified for Italy and each subnational location. The redistribution of these packages to underlying causes of death was then examined by comparing pre- and post-redistribution death counts. For the purposes of this study, results are presented for Italy only, while subnational estimates are available through a dedicated interactive tool developed within this study.¹¹

In addition, for the three leading level-4 causes of death across all ages and in the paediatric population (0-14 years) in Italy in 2021, pre- and post-redistribution deaths were reported, along with the percentage inflow, defined as the proportion of deaths assigned to each cause through redistribution. This analysis aimed to assess whether the contribution of redistribution differs between causes more prevalent in adult populations and those affecting children, given potential differences in cause-of-death attribution related to multimorbidity and more complex causal pathways in adults. COVID-19 was excluded as a pandemic-specific cause not directly comparable with others and not relevant to the objective of this analysis.

To further characterize the context of death certification in Italy, we analysed Istat data on the distribution of deaths according to:

• Type of certifier (general practitioner, hospital physician, medical examiner, other physician, unknown);
• Place of death (home, healthcare facility, residential facility, unknown);
• Autopsy status (yes, no, unknown).

Data were provided for Italy and each subnational location.

Scatter plots were used to examine the relationship between GC percentages in 2021 and the distribution of deaths by certifier type, place of death, and autopsy status in the same year. Spearman correlation coefficients (r) were reported along with corresponding p-values. Statistical significance was set at α = 0.05. All analyses were performed using R.¹² 

Results

In Italy, GCs showed a gradual decline between 1990 and 2021, yet they still represented nearly one-third of all deaths in 2021, decreasing from 34.6% in 1990 to 28.8% (Table 1). Among the ten Western European countries with populations exceeding 10 million, Italy ranked 5^th in 2019 (i.e., the most recent year with complete data available for all ten countries) with 31.0% of GCs. The lowest percentage was found in the United Kingdom (22.7%), while the highest was in Greece (43.5%).

Temporal trends at both national and regional levels are shown in Figure 1. In 2021, GC percentages ranged from 23.8% in Friuli Venezia Giulia to 34.9% in Calabria (Table 2).

Overall, the national reduction between 1990 and 2021 was -5.8%, although marked heterogeneity was observed across regions (Table S1, online Supplementary Materials). Abruzzo showed the largest decrease (-10.0%), moving from the 16th; position to the 9th; in the ranking, whereas Veneto exhibited the smallest reduction (-2.7%) and moved from being the best in the ranking, together with the AP of Trento, to the 8th; position.

Two clear patterns emerged. First, a peak in GC percentages was observed in all regions except AP Trento and AP Bolzano in 2003. Second, an additional disruption in GC trends occurred in 2015 across all regions, although in AP Trento this shift appears to have started earlier, around 2013.¹¹

The linear mixed model revealed a significant downward trend over time (β -0.19; 95%CI -0.22;-0.16) (Table 3). Substantial between-region variability was observed at baseline (SD 3.19; 95%CI 2.35;4.35), while differences in temporal trends were smaller but still significant (SD 0.08; 95%CI 0.05;0.10). The near-zero covariance between random intercepts and slopes suggests no relationship between initial levels and rates of change across regions.

The three most frequent GC packages in Italy in 2021 were ‘unspecified type of stroke’ (4.28% of total deaths), ‘diabetes unspecified type’ (2.44%), and ‘heart failure unspecified right or left’ (2.38%) (Table S2, online Supplementary Materials). ‘Unspecified type of stroke’ ranked first in all regions except the AP Trento, where ‘all, ill-defined code for causes of death’ was most frequent (3.61%), and Puglia, where ‘diabetes unspecified type’ accounted for the largest share (3.34%). 

These packages reflect distinct types of misreporting. For stroke and diabetes, the underlying cause is broadly identified, but lacks essential specification, as stroke should be classified by subtype (ischemic, intracerebral, or subarachnoid haemorrhage) and diabetes by type 1 or 2. In contrast, heart failure should not be considered an underlying cause of death; at the same time, specifying the affected side (right or left) may provide some indication of the underlying pathology, pointing to a dual limitation, both conceptual and in terms of specificity. More broadly, ill-defined codes include diagnoses that cannot be considered valid causes of death, such as ‘senility’, and group together heterogeneous conditions that do not meaningfully identify an underlying cause.

Figures from S1 to S3 (online Supplementary Materials) show the redistribution of the leading three packages in Italy into level-4 causes. Most deaths originally coded as ‘unspecified type of stroke’ were reclassified as ischemic stroke (82.3%). The ‘diabetes, unspecified type’ category was overwhelmingly reassigned to type 2 diabetes (99.3%), whereas ‘heart failure unspecified right or left’ was most frequently reclassified as ischemic heart disease (33.1%).

The ten level-4 causes showing the largest absolute increases after redistribution in 2021 are presented in Figure 2. Cardiovascular diseases accounted for half of these causes, including ischemic stroke, chronic ischemic heart disease, intracerebral haemorrhage, acute myocardial infarction, hypertensive chronic kidney disease, and other cardiomyopathies.

Marked differences in inflow percentages were observed across causes. Among all ages, ischemic stroke exhibited a particularly high inflow (74.0%), while ischemic heart disease showed a moderate contribution (24.0%) and Alzheimer’s disease was minimally affected (0.4%). Among paediatric causes, neonatal preterm birth showed a low inflow (5.1%), whereas congenital causes displayed higher values, both around 15%. While some causes more prevalent at older ages showed high inflows from redistribution, this pattern was not consistent across all adult causes. Instead, a similar pattern was observed across causes affecting both adult and paediatric populations: higher inflow was found for conditions requiring greater diagnostic specificity, such as ischemic stroke and congenital anomalies, whereas more directly identifiable causes, such as Alzheimer’s disease and neonatal preterm birth, showed lower inflow.

Regional patterns in certifier type, place of death, and autopsy practices are reported in Figures from S4 to S6 (online Supplementary Materials). Missing information on certifiers varied widely, from 1.1% in the AP of Trento to 36.4% in Calabria (SD 8.1). Missing data on place of death were more homogeneous across regions (SD 1.2). A clear geographical gradient emerged, with deaths at home more frequent in Southern regions, while deaths in healthcare facilities and residential settings were more common in the North. Autopsy rates also varied, ranging from 0.5% in Campania and Sicilia to 4.3% in Friuli Venezia Giulia. Substantial variability was observed in the proportion of deaths with missing information on autopsy requests (SD 17.0), ranging from 10.0% in Valle d’Aosta to 64.6% in Calabria.

Correlation analyses showed that GC percentages in 2021 were strongly negatively associated with the percentage of deaths certified by hospital physicians (r -0.73; p <0.001) and positively associated with the proportion of missing information on certifiers (r 0.54; p=0.020) (Figure S7, online Supplementary Materials). Similarly, higher GC proportions were observed in regions with more deaths occurring at home (r 0.71; p <0.001) or with missing data on place of death (r 0.77; p <0.001), whereas negative correlations were found for deaths in healthcare facilities (r -0.75; p <0.001) and residential structures (r −0.61; p=0.005) (Figure S8, online Supplementary Materials). A comparable pattern was observed for autopsy-related variables, with GC percentages positively correlated with missing information (r 0.76; p <0.001) and negatively correlated with performed autopsies (r -0.71; p=0.002) (Figure S9, online Supplementary Materials).

Discussion

This study provides a comprehensive assessment of the magnitude, patterns, and factors associated with GCs in Italy, highlighting both temporal improvements and persistent sources of variability across regions and causes of death. 

First, the overall decline in GC percentages from 1990 to 2021 suggests a gradual improvement in the quality of cause-of-death certification. However, Italy remains mid-ranked among comparable Western European countries, suggesting substantial room for improvement. Moreover, at the subnational level, despite overall improvements over time, marked heterogeneity persists, with a clear geographical gradient disadvantaging Southern regions and Islands. Regions with poorer initial performance did not achieve greater improvements, indicating the absence of a convergence pattern. This finding is consistent with previous analyses.³

Notably, two clear disruptions in temporal trends were identified. The peak observed in 2003 likely reflects the introduction of ICD-10 coding at the national level, which may have temporarily increased misclassification during the transition phase.¹³ This pattern may also have been influenced by the excess mortality associated with the 2003 summer heatwaves, which could have further complicated cause-of-death attribution under conditions of increased mortality.¹⁴ A second disruption in 2015 coincides with a documented increase in overall mortality, as reported by national surveillance systems, which may have influenced certification practices under conditions of heightened healthcare pressure.^15,16 Findings from these systems indicate that excess mortality in winter has been largely attributed to influenza activity, whereas the summer increase has been linked to heat waves occurring in July and August 2015. This pattern may have further complicated cause-of-death attribution, particularly for conditions requiring a higher degree of specificity to be correctly classified. Similar increases in overall mortality were observed across several European countries in the same year; however, the corresponding pattern in GCs was not consistently reported.^17,18 Therefore, the temporary rise in GCs proportion observed in Italy in 2015 may therefore reflect country-specific factors affecting cause-of-death certification and coding practices. As for most recent years, the interpretation of trends should also consider the potential impact of the COVID-19 pandemic. The emergence of COVID-19 as a well-defined underlying cause of death may have increased the number of deaths correctly assigned, thereby mechanically reducing the proportion of GCs. In particular, some deaths that might previously have been classified as GCs may have been attributed to COVID-19 during the pandemic, either appropriately or as a consequence of the exceptional circumstances affecting cause-of-death certification. As a result, part of the observed decline in GC percentages in 2021 may reflect a compositional effect rather than solely an improvement in certification quality. Under this interpretation, a partial rebound in GC proportions could be expected in the post-pandemic period as mortality patterns and certification practices return to routine conditions, although longer time series will be required to assess whether the decline observed in 2020-2021 was sustained.

The analysis of the most common GC packages indicates that misclassification arises from multiple mechanisms, reflecting errors of varying nature and severity. In some instances, such as ‘unspecified type of stroke’ and ‘diabetes, unspecified type’, the underlying cause is broadly identifiable, but lacks sufficient clinical detail for precise classification. In these cases, redistribution can be undertaken with relative confidence, drawing on evidence-based information on disease prevalence and prognosis. In contrast, for conditions such as ‘heart failure, unspecified left or right’, the problem is more fundamental, as these diagnoses reflect terminal events rather than true underlying causes of death. In such cases, redistribution is inherently more complex, and the validity of reassigned underlying causes remains uncertain.

This distinction is important, because it points to different levels of error that may require tailored corrective measures, including improvements in certification practices and training. While training is essential, evidence from other countries suggests that implementing electronic death certification systems, with built-in validation procedures, could further improve data quality.^19,20 

Nevertheless, as some degree of misclassification is likely to persist, the redistribution of GCs should not be interpreted as an attempt to produce exact estimates, but rather as an evidence-based approach to assigning deaths to the most plausible underlying causes. For example, the ‘diabetes, unspecified type’ package is not redistributed according to the population prevalence of type 1 and type 2 diabetes (approximately 5-10% vs 90-95%), but rather based on evidence showing that deaths coded as unspecified are more frequently attributable to type 2 diabetes.

Previous evidence shows that the proportion of GCs increases with age, likely reflecting the greater complexity of causal pathways and multimorbidity in older populations.⁴ The results of this study are broadly consistent with this pattern, as some causes more common at older ages, particularly cardiovascular conditions, showed substantial inflow from redistribution. However, this was not uniform: the very low inflow for Alzheimer’s disease and other dementias suggests that GCs are less frequent in conditions that require a lower level of diagnostic detail to be correctly coded. This pattern also emerged among causes affecting the paediatric population, with higher inflow for congenital anomalies than for neonatal preterm birth. Therefore, these findings support the view that the specificity required for accurate attribution plays a consistent role in shaping redistribution patterns.

Finally, the associations observed between GC proportions and characteristics of death certification provide further insight into factors that may be linked to data quality. In particular, higher GC percentages tended to occur in regions where information on the certifier, place of death, and autopsy requests was more frequently missing, as well as in settings with a larger share of deaths occurring at home. Conversely, lower GC proportions were generally observed when deaths were certified by hospital physicians or occurred in healthcare facilities or residential structures, where more complete clinical information is likely to be available.

These findings have important implications for public health policies and health information systems. First, the marked regional heterogeneity suggests the need for targeted interventions in areas with persistently higher levels of GCs, particularly in Southern regions and Islands. Second, the identification of distinct types of misclassification highlights that a one-size-fits-all approach is unlikely to be effective: errors related to lack of specificity may benefit from improved clinical documentation and clearer certification guidelines, whereas more fundamental errors, such as the use of terminal conditions, require strengthened training on the correct identification of the underlying cause. Third, the strong associations observed between GC proportions and contextual factors and completeness of information suggest that strengthening certification practices outside hospital settings and reducing missing information may contribute to more accurate cause-of-death attribution. 

An illustrative example of the potential policy impact can be drawn from the comparison of mortality-to-incidence ratios (MIRs) before and after the redistribution of GCs. Stroke provides a particularly informative case: in 2021, the relative change in MIR ranged from 0.96 in AP Trento (from 0.26 to 0.51) to 2.77 in Sicilia, where the MIR nearly tripled (from 0.19 to 0.71). These differences highlight how variation in the attribution of deaths can lead to substantially different interpretations of disease management, with potential implications for risk communication, priority setting, resource allocation, and the evaluation of health system performance.

Limitations of the study

These results should, however, be interpreted in light of several limitations. The ecological analyses exploring associations between GC proportions and contextual factors are based on a limited number of subnational units, and some variables are compositional, potentially inducing spurious inverse correlations; findings should therefore be considered exploratory rather than causal. Moreover, the study relies on GBD redistribution algorithms rather than individual-level certificate validation, and the use of all-age, both-sex combined estimates may obscure age- or sex-specific patterns. In addition, the inclusion of 2021 – a pandemic year – may have affected certification practices, place of death, hospital pressure, and diagnostic completeness, even though COVID-19 was excluded from selected cause-specific analyses. Finally, the selection of less specific codes may reflect not only clinical uncertainty, but also institutional coding behaviours and performance-assessment incentives, which should be considered when interpreting regional differences in GC patterns.

In this context, while redistribution methods are essential to produce comparable and policy-relevant estimates, these findings emphasize the importance of improving data quality at the source. In this regard, the broader adoption of electronic death certification systems, coupled with real-time validation and decision support, represents a promising strategy to reduce misclassification and enhance the reliability of mortality statistics. Improving the accuracy and specificity of cause-of-death classification is critical to ensure that health policies are grounded in robust, data-driven evidence. In this perspective, the ongoing transition towards ICD-11, a fully digital classification system designed to enable greater clinical specificity and support automated coding processes, may further contribute to improving the quality and consistency of cause-of-death data.²¹

Conflicts of interest: none declared.

Funding: GZ, LM, and LR reported support from the Italian Ministry of Health, through the contribution given to the Institute for Maternal and Child Health IRCCS Burlo Garofolo, Trieste (Italy) (RC 34/2017).

References and notes

1. Hoffman RA, Venugopalan J, Qu L, Wu H, Wang MD. Improving validity of cause of death on death certificates. ACM BCB 2018;2018:178-83. doi:10.1145/3233547.3233581
2. Mikkelson L, Richards N, Lopez AD. Redefining ‘garbage codes’ for public health policy: report on the expert group meeting. 27-28 February 2017. CRVS best-practice and advocacy. Melbourne (Australia): Bloomberg Philanthropies Data for Health Initiative, Civil Registration and Vital Statistics Improvement, University of Melbourne; 2019. Available from: https://data4healthlibrary.org/sites/default/files/resources/125_UMelbourne_Redefining%20garbage%20codes%20for%20public%20health%20policy_%20Report%20on%20the%20expert%20group%20meeting.pdf (last accessed: 19.04.2026).
3. Monasta L, Alicandro G, Pasovic M, et al. Redistribution of garbage codes to underlying causes of death: a systematic analysis on Italy and a comparison with most populous Western European countries based on the Global Burden of Disease Study 2019. Eur J Public Health 2022;32(3):456-62. doi:10.1093/eurpub/ckab194
4. Naghavi M, Makela S, Foreman K, O’Brien J, Pourmalek F, Lozano R. Algorithms for enhancing public health utility of national causes-of-death data. Popul Health Metr 2010;8:9. doi:10.1186/1478-7954-8-9
5. Erhardt CL. What is “the cause of death”? JAMA 1958;168(2):161-68. doi: 10.1001/jama.1958.03000020023005
6. World Health Organization. International statistical classification of diseases and related health problems, 10^th revision. Volume 2. Instruction Manual. Sixth Edition 2019. Available from: https://icd.who.int/browse10/Content/statichtml/ICD10Volume2_en_2019.pdf (last accessed: 05.05.2026).
7. World Health Organization. Cause of death. Available from: https://www.who.int/standards/classifications/classification-of-diseases/cause-of-death (last accessed: 19.04.2026).
8. Ng TC, Lo WC, Ku CC, Lu TH, Lin HH. Improving the use of mortality data in public health: a comparison of garbage code redistribution models. Am J Public Health 2020;110(2):222-29. doi:10.2105/AJPH.2019.305439.
9. GBD 2021 Causes of Death Collaborators. Global burden of 288 causes of death and life expectancy decomposition in 204 countries and territories and 811 subnational locations, 1990-2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet 2024;403(10440):2100-32. doi:10.1016/S0140-6736(24)00367-2
10. Stevens GA, Alkema L, Black RE, et al. Guidelines for accurate and transparent health estimates reporting: the GATHER statement. Lancet 2016;388(10062):e19-23. doi: 10.1016/S0140-6736(16)30388-9.
11. GBD Garbage Codes–Italy. Available from: https://www.italian-gbd-initiative.it/gbd-italia-redistribuzione-garbage/
12. R Core Team. R: a language and environment for statistical computing. Available from: https://www.r-project.org/ (last accessed: 19.04.2026).
13. Italian National Institute of Statistics (Istat). Cause multiple di morte. Anni 2003 e 2006. Guida alla lettura. Available from: https://www.istat.it/wp-content/uploads/2011/02/guida.pdf (last accessed: 19.04.2026).
14. Michelozzi P, de Donato F, Bisanti L, et al. The impact of the summer 2003 heat waves on mortality in four Italian cities. Euro Surveill 2005;10(7):11-12. doi: 10.2807/esm.10.07.00556-en
15. Michelozzi P, De’ Donato F, Scortichini M, et al. On the increase in mortality in Italy in 2015: analysis of seasonal mortality in the 32 municipalities included in the surveillance system of daily mortality. Epidemiol Prev 2016;40(1):22-28. doi:10.19191/EP16.1.P022.010
16. Signorelli C, Odone A. Response: Mortality rate in Italy in 2015: an increase of 9.1% to be explained. BMJ; 26.02.2026. Available from: https://www.bmj.com/content/347/bmj.f5568/rr (last accessed: 19.04.2026).
17. Rajmil L, Fernández de Sanmamed MJ. Austerity Policies and Mortality Rates in European Countries, 2011-2015. Am J Public Health 2019;109(5):768-70. doi: 10.2105/AJPH.2019.304997
18. Molbak K, Espenhain L, Nielsen J, et al. Excess mortality among the elderly in European countries, December 2014 to February 2015. Euro Surveill 2015;20(11):21065. doi: 10.2807/1560-7917.es2015.20.11.21065
19. Lefeuvre D, Pavillon G, Aouba A, et al. Quality comparison of electronic versus paper death certificates in France, 2010. Popul Health Metr 2014;12(1):3. doi:10.1186/1478-7954-12-3
20. Thanigasalam A, Ramu R, Rajarethinam K. Improving completeness and reducing errors in medical certification of cause of death: The impact of electronic mortality software in a tertiary care centre in South India. Indian J Med Res 2025;161(4):420-24. doi:10.25259/IJMR_1394_2024
21. World Health Organization. International classification of diseases and Related Health Problems (ICD). ICD-11. Available from: https://www.who.int/standards/classifications/classification-of-diseases (last accessed: 19.04.2026).