A cause of death is the disease, injury, or event that initiates the chain of bodily failures leading to a person’s death, and pinning it down is far less straightforward than most people assume. Globally, cardiovascular diseases account for roughly a third of all deaths each year, making heart attacks and strokes the single largest category. But the cause written on a death certificate depends on who fills it out, what tools they have, and how they interpret often ambiguous evidence. Studies consistently find that the majority of death certificates contain at least one error, and in many parts of the world, no formal death registration system exists at all.
What Actually Goes on a Death Certificate
In most countries with civil registration systems, the death certificate is the official document that records cause of death. It follows a standardized format developed by the World Health Organization, with a causal sequence that works backward from the immediate cause (say, pulmonary embolism) through intermediate causes (deep vein thrombosis) to an underlying cause (hip fracture from a fall). The underlying cause of death is the one that matters most for public health statistics, because it represents the condition that set the fatal chain in motion. A separate field records the manner of death: natural, accident, suicide, homicide, or undetermined.
This sounds tidy on paper, but in practice the system is riddled with mistakes. A systematic review and meta-analysis of death certificate errors found that the most common problems were missing time intervals between conditions (present in about 81% of flawed certificates), inappropriate listing of comorbidities (around 45%), and incorrect identification of the underlying cause (roughly 39%). Other frequent errors included listing conditions in the wrong causal sequence, writing down a mechanism of death (like “cardiac arrest”) instead of an actual disease, and using abbreviations that coding clerks couldn’t interpret.
1PubMed. Common errors in reporting cause-of-death statement on death certificates: A systematic review and meta-analysisA separate study that audited actual certificates found that 85% contained at least one error, and more than half had multiple errors.2PubMed Central. Death Certification: Errors and Interventions Many of these are formatting mistakes that trained coders can work around, but a meaningful fraction involve genuinely wrong diagnoses that distort mortality statistics. When thousands of death certificates in a country list “cardiac arrest” as a cause of death, for instance, that tells public health researchers nothing useful. Everyone’s heart stops when they die. The question is why it stopped.
The World’s Leading Killers
Despite these data quality issues, the broad picture of what kills people globally is well established. Cardiovascular diseases are the dominant cause, responsible for about a third of all deaths worldwide in 2019. Ischemic heart disease alone killed an estimated 9.1 million people that year, and stroke killed another 6.6 million. Together, those two conditions accounted for 85% of all cardiovascular deaths.3PubMed Central. The Heart of the World Cancer, respiratory diseases, and diabetes fill out most of the remaining top spots in high-income countries, while infectious diseases like lower respiratory infections, diarrheal diseases, and tuberculosis still claim millions of lives annually in lower-income settings.
The pattern of leading causes of death has shifted dramatically over the past century, a phenomenon researchers call the epidemiologic transition. Societies that industrialize tend to see infectious disease deaths fall and chronic disease deaths rise, driven by improved sanitation, antibiotics, and vaccines on one side and aging populations, sedentary lifestyles, and processed diets on the other.4PubMed Central. The Epidemiologic Transition: Changing Patterns of Mortality and Population Dynamics This transition is incomplete in many parts of the world, which is why the global picture includes both heart disease and malaria as major killers simultaneously.
Child mortality tells a particularly striking version of this story. A study tracking 50 years of data across countries found that all-cause mortality dropped by 85 to 93% in children aged one to four years old and by 80 to 87% in children aged five to nine. But the declines were much smaller for young men aged 15 to 24, where mortality fell only 41 to 48% over the same period, largely because injuries, violence, and suicide replaced the infectious diseases that had been conquered in younger age groups.5The Lancet. 50-year trends in mortality among children and young people: a global population-based study
The Gap Between Listed Causes and Actual Causes
There is a useful distinction between what kills you on a medical level and what actually caused the conditions that killed you. A death certificate might list coronary artery disease as the underlying cause, but the real driver could be decades of smoking, a sedentary lifestyle, or an unhealthy diet. Researchers have tried to quantify these upstream “actual causes” to guide prevention efforts.
A landmark analysis of U.S. deaths in 2000 estimated that tobacco was responsible for roughly 435,000 deaths (about 18% of all deaths), while poor diet combined with physical inactivity accounted for another 400,000 (nearly 17%). Alcohol consumption was third at about 85,000 deaths. Further down the list were microbial agents, toxic agents, motor vehicle crashes, firearms, risky sexual behaviors, and illicit drugs.6JAMA. Actual Causes of Death in the United States, 2000 A later analysis using 2005 data confirmed that tobacco smoking was responsible for an estimated 467,000 deaths and high blood pressure for about 395,000. Overweight and obesity, physical inactivity, and high blood glucose each contributed roughly 190,000 to 216,000 deaths.7PubMed Central. The Preventable Causes of Death in the United States: Comparative Risk Assessment of Dietary, Lifestyle, and Metabolic Risk Factors
Air pollution is another upstream killer that rarely appears on death certificates. Research linking fine particulate matter (PM2.5) exposure to mortality in the United States estimated that it contributed to over 56,000 cardiovascular deaths and more than 40,000 cerebrovascular deaths annually, along with thousands of deaths from lung cancer, dementia, chronic kidney disease, and pneumonia.8JAMA Network Open. Burden of Cause-Specific Mortality Associated With PM2.5 Air Pollution in the United States Globally, one model estimated that outdoor fine particulate matter exposure was associated with 8.9 million deaths in 2015, a figure substantially larger than previous estimates that had looked at only five specific disease categories.9PubMed Central. Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter None of those deaths would have “air pollution” written anywhere on the certificate.
Socioeconomic status further complicates the picture. A study examining how race and socioeconomic status interact with cause-specific mortality found that for women, socioeconomic factors accounted for roughly 37 to 67% of the excess mortality among Black Americans for conditions like accidents, ischemic heart disease (in those aged 35 to 54), diabetes, and homicide. But socioeconomic status explained far less of the disparity for hypertension, infections, and stomach cancers. For men, the pattern was similar but not identical, with socioeconomic status explaining 30 to 55% of the excess risk for accidents, lung cancer, stroke, and homicide, but almost none of the disparity for prostate cancer or hypertension.10PubMed. Race, socioeconomic status, and cause-specific mortality Poverty does not appear on death certificates either, but it shapes who dies from what.
How Forensic Pathologists Figure It Out
When the cause of death is not obvious from a person’s medical history, or when the circumstances are suspicious, the case goes to a forensic pathologist. The traditional autopsy remains the gold standard: an external examination followed by internal dissection of the body’s organs, supplemented by toxicology screens, histology (examining tissue slices under a microscope), and sometimes microbiology or genetic testing. In many jurisdictions, a medical examiner or coroner decides which deaths require an autopsy, generally those involving violence, accidents, unexplained circumstances, or deaths in custody.
Even with a full autopsy, the initial working diagnosis can change. One study of forensic autopsy cases found that about 17% of cases had a true change in their final diagnosis after all test results came back, meaning the cause or manner of death assigned during the initial examination was revised. The likelihood of a diagnostic change was related to the person’s age.11PubMed Central. Accuracy and validity of determined cause of death and manner of death following forensic autopsy prosection Toxicology results are a common reason for revision; a death that initially looks like a natural cardiac event can turn out to be an overdose once the drug screen comes back weeks later.
Determining manner of death adds another layer of difficulty, especially for suicides. Research on death classification has shown that suicide determination is not standardized across medical examiners, and many suspected suicides end up classified as accidental or undetermined.12PubMed Central. Comparative analysis of suicide, accidental, and undetermined cause of death classification This inconsistency means that suicide statistics almost certainly undercount the true number of deaths, which has real consequences for funding, research, and prevention programs.
Estimating Time of Death
Pinpointing when someone died matters enormously in criminal investigations, and it’s one of the hardest things forensic pathologists do. The current gold standard is a method based on body cooling: measuring the body’s temperature and working backward along a mathematical curve to estimate when normal body temperature was lost. Additional clues come from the stiffness of the muscles (rigor mortis), the settling of blood in the tissues (lividity), and the electrical and mechanical responses of skeletal muscle to stimulation.13PubMed. Methods for determining time of death
For bodies that have been dead longer, investigators turn to different tools entirely. Insect colonization of the body (forensic entomology), the succession of microbial communities in decomposing tissue, and chemical changes in the fluid of the eye (vitreous humor) all follow somewhat predictable timelines. More experimental methods include RNA and DNA degradation patterns and various imaging techniques.14PubMed Central. Current Understanding and Future Research Direction for Estimating the Postmortem Interval: A Systematic Review All of these methods carry substantial margins of error, and factors like ambient temperature, the person’s body composition, clothing, and whether the body was in water can throw estimates off considerably.
Virtual Autopsy and Imaging-Based Approaches
Traditional autopsy rates have been declining in many countries for decades, driven by cost, religious objections, and family reluctance. This has fueled interest in “virtual autopsy,” which uses CT scans and sometimes MRI to examine the body without making incisions. The approach has clear strengths and equally clear limitations.
In a study of 25 cases in Verona, virtual autopsy correctly identified the cause of death in about 65% of cases where traditional autopsy found one, and the overall agreement between the two methods was 64%. The technology performed substantially better in traumatic deaths, where it achieved 81% sensitivity and 84% overall accuracy, compared to nontraumatic deaths.15Journal of Pathology Informatics. Virtual Autopsy as a Screening Test Before Traditional Autopsy: The Verona Experience on 25 Cases A broader systematic review of pediatric cases confirmed that virtual autopsy was superior to conventional autopsy for detecting skeletal injuries and bullet trajectories, and was useful for identifying foreign body aspiration, drowning, and blood clots in the lungs. However, for natural deaths without trauma, non-contrast imaging did not offer more information than traditional autopsy, and pathologists sometimes misinterpreted normal postmortem changes in the images as disease.16PubMed Central. Usefulness of virtual autopsy in diagnosing pathologies in the paediatric population: A systematic review
The takeaway is that virtual autopsy is a useful screening tool and a strong complement to traditional methods, especially in trauma cases, but it is not yet a reliable replacement when the death appears natural and the question is which internal disease process was responsible.
Molecular Autopsy for Unexplained Sudden Deaths
When a young, apparently healthy person drops dead and the traditional autopsy finds nothing structurally wrong with the heart or any other organ, the case may be labeled “sudden unexplained death.” In many of these cases, the underlying problem was an inherited abnormality in the heart’s electrical system that left no visible trace in the tissue. This is where molecular autopsy comes in: genetic testing performed on samples from the deceased to look for mutations associated with arrhythmia disorders.
The value of this approach goes beyond solving a single death. If a genetic variant linked to dangerous heart rhythms is identified in the deceased, their living blood relatives can be screened for the same variant. Those who carry it can then be monitored or treated before they suffer the same fate.17PubMed Central. Molecular autopsy in sudden cardiac death Molecular autopsy is still not routine in most places, but it has become standard practice at specialized cardiac pathology centers and is increasingly recommended by professional guidelines when conventional autopsy fails to find a cause in young sudden-death cases.
Brain Death as a Legal Cause of Death
The concept of death itself has a definition problem. Most deaths are obvious: the heart stops, breathing ceases, and the body begins to cool. But modern intensive care can keep blood circulating and lungs ventilating long after the brain has permanently ceased to function. Countries and jurisdictions vary in how they handle this, but most now accept brain death, defined as the complete and irreversible loss of all brain function, as a legal form of death.
Determining brain death requires ruling out conditions that can mimic it, such as severe hypothermia, drug intoxication, or certain metabolic disturbances. The clinical examination checks for coma, the absence of all brainstem reflexes, and the inability to breathe independently (the apnea test).18JAMA. Determination of Brain Death/Death by Neurologic Criteria: The World Brain Death Project A 2023 consensus guideline unified the criteria for both adults and children, aiming to standardize a process that had previously varied between institutions and between pediatric and adult practice.19PubMed Central. Pediatric and Adult Brain Death/Death by Neurologic Criteria Consensus Guideline When brain death is the finding, the underlying cause, such as a traumatic brain injury or massive stroke, is what goes on the death certificate.
When No Doctor Is Present
About half the world’s deaths occur in places without routine death registration. In rural communities across much of sub-Saharan Africa, South Asia, and parts of Southeast Asia, many people die at home without ever seeing a physician. In these settings, researchers use a method called verbal autopsy: a trained interviewer asks the deceased person’s family members a structured set of questions about the symptoms, timeline, and circumstances of the death, and a physician or algorithm then assigns a probable cause.20Epidemiologic Reviews. Verbal Autopsy: Methods in Transition
Verbal autopsy is imperfect by design. Family members may not have noticed key symptoms, may recall them inaccurately, or may describe them using local terminology that doesn’t map neatly onto biomedical categories. The method works reasonably well for deaths with distinctive symptom profiles, like measles in a child or injuries from a road accident, but it struggles with conditions that look similar from the outside, like distinguishing one type of heart disease from another. Still, it is the only practical tool for generating cause-of-death data in populations where the alternative is no data at all.
Excess Mortality and Counting Deaths That Are Hard to Count
Even in countries with functioning vital registration, the official cause-of-death data can miss the full impact of a crisis. During a pandemic, a war, or a natural disaster, deaths increase from both the direct event and from secondary effects like overwhelmed hospitals, disrupted supply chains, and delayed treatments for other conditions. Excess mortality estimation tries to capture this total toll by comparing the number of deaths that actually occurred to the number that would have been expected in a normal period.
This approach gained enormous public visibility during the COVID-19 pandemic. In many countries, the number of excess deaths exceeded the official COVID-19 death toll, sometimes dramatically. The gap reflected deaths that were never tested and therefore never counted, deaths caused indirectly by the pandemic’s disruption of healthcare, and deaths in countries without the capacity to test or report accurately. The WHO’s estimates of pandemic-associated excess mortality used statistical models that predicted expected deaths and then quantified the difference with uncertainty intervals, even for countries lacking complete data.21Nature. The WHO estimates of excess mortality associated with the COVID-19 pandemic
Excess mortality estimation is not just a pandemic tool. Researchers have developed flexible statistical frameworks that can detect concerning increases in death rates over time and distinguish sudden spikes from gradual trends.22PubMed Central. A Flexible Statistical Framework for Estimating Excess Mortality The method has been applied to heat waves, hurricanes, armed conflicts, and famine. Its strength is that it sidesteps the cause-of-death classification problem entirely: it does not ask what killed each person, only whether more people died than expected. Its weakness is the mirror image of that strength: it tells you the size of the disaster’s impact, not which specific diseases or injuries drove it.
Religious and Cultural Objections to Autopsy
For many families, the idea of an autopsy is deeply distressing, and for some religious communities the objection is theological. Research exploring Muslim and Jewish perspectives on autopsy found that both traditions emphasize prompt burial and keeping the body intact. Participants described their understanding that the body should not be disturbed unnecessarily after death and should be returned to the ground in the condition it arrived in.23PLoS ONE. “We might get a lot more families who will agree”: Muslim and Jewish perspectives on less invasive perinatal and paediatric autopsy These concerns apply across age groups but are felt with particular intensity when the deceased is a baby or a child.
Less invasive alternatives like virtual autopsy and needle biopsy have been developed partly in response to these objections. The same study found that families who would refuse a traditional autopsy were substantially more open to imaging-based examination, which could provide at least some diagnostic information without the perceived violation of the body. Striking a balance between the public health need for accurate cause-of-death data and the rights and beliefs of grieving families is an ongoing challenge for coroners and medical examiners worldwide.
AI and Automated Cause-of-Death Coding
With millions of death certificates generated each year worldwide, the manual process of reading each certificate and assigning a standardized cause-of-death code is slow, expensive, and inconsistent between coders. Several research groups have been developing machine learning systems to automate this step.
An Australian system trained to classify diseases from free-text death certificates achieved high accuracy for common conditions like diabetes, influenza, pneumonia, and HIV, with performance that matched or approached human coders for those categories. More detailed classification was more variable but still reached reasonable accuracy.24PubMed Central. Automatic classification of diseases from free-text death certificates for real-time surveillance A Korean system tested on over 300,000 certificates from 2022 achieved roughly 63% accuracy for final cause-of-death classification and about 95% accuracy for a broader “tentative cause” classification. Error analysis showed that the system struggled with rare diseases and certain disease categories that required additional context beyond what was written on the certificate.25PubMed Central. Machine learning for automated cause-of-death classification from 2021 to 2022 in Korea: development and validation of an ICD-10 prediction model
These systems are unlikely to replace human review entirely, at least not soon. The certificates themselves are often messy: handwritten, abbreviated, and sometimes internally contradictory. But automated coding could speed up the process enough to make real-time mortality surveillance practical, flagging sudden increases in deaths from particular causes days or weeks earlier than traditional reporting allows. For public health agencies trying to detect outbreaks or environmental hazards, that kind of speed matters a great deal.