Cause of death is determined by establishing the disease or injury that started the fatal sequence, then tracing the unbroken chain of events to the final mechanism—typically through external examination, medical record review, autopsy, toxicology, and postmortem imaging, with the certifying physician or medical examiner applying international coding rules. The death certificate has two parts: Part I records the causal sequence from immediate to underlying cause, and Part II records conditions that contributed but did not cause the death.

That structure sounds administrative. It is actually a compressed scientific argument. Every line on a death certificate is a claim about causation that a physician or medical examiner must support with evidence, and that a nosologist later translates into an ICD-10 code and a statistical entry (World Health Organization). Understanding how cause of death is determined means understanding the evidence hierarchy, the failure modes, and the places where the system quietly guesses.

The causal-chain mechanism

Certifiers are trained to think in sequences, not labels. A typical Part I reads: immediate cause, then the condition that led to it, then the underlying cause. For example: pulmonary edema, due to congestive heart failure, due to hypertensive heart disease. The underlying cause—the condition that initiated the fatal sequence—is what public health statistics count, because counting immediate mechanisms produces useless rankings where "cardiac arrest" tops every list.

The causal chain must be plausible and unbroken. Certifiers are instructed not to list two conditions on the same line unless one clearly caused the other, and not to skip steps that break the logic. The mechanism matters because each link is a testable claim about physiology or pathology. Heart failure causes pulmonary edema through elevated filling pressures and transudation into alveoli; a certifier asserting that link is asserting a mechanism, not a correlation.

The manner of death is a separate axis from the cause. Cause is the disease or injury; manner is the classification—natural, accident, homicide, suicide, or undetermined. A gunshot wound is a cause; homicide is a manner. Confusing the two is one of the most common errors in public understanding of mortality data. The manner determination is legal as well as medical, which is why medical examiners, not treating physicians, handle it.

Postmortem investigation fills gaps the clinical record leaves. External examination documents injuries and identifiers. Internal autopsy examines organs grossly and microscopically. Toxicology quantifies drugs and metabolites in blood, vitreous, and tissue. Postmortem computed tomography detects fractures, hemorrhages, and air emboli noninvasively. Each modality answers a different question: what was present, what was functioning, and what was circulating.

What the numbers show about accuracy

Certification accuracy is measurable, and the measurements are sobering in specific ways. Studies comparing clinical diagnoses to autopsy findings have repeatedly found major discrepancy rates in the range of roughly 10 to 25 percent, depending on case mix and era, with the highest discordance in intensive care and elderly patients where multiple competing diseases coexist. Autopsy rates in United States hospitals have fallen below 10 percent of deaths and in many centers below 5 percent, meaning most certified causes are never verified by the reference standard (CDC).

Coding adds another layer. ICD-10 assigns underlying-cause codes from certificate text, and nosologists apply selection rules when certificates are ambiguous or internally inconsistent. When a certificate lists an implausible sequence, the coding system may select a different underlying cause than the certifier intended. Deaths coded to ill-defined categories—symptoms, signs, and abnormal findings—represent a persistent fraction of mortality data and are concentrated in older decedents and in jurisdictions with limited investigation resources.

The practical consequence is that cause-of-death statistics are excellent for population patterns and weaker for individual attribution. A death certificate is a legal document with a medical opinion inside it. That opinion is usually right about the broad category and less reliable about the specific initiating condition. Treating certificate data as ground truth overstates precision, particularly for deaths involving multimorbidity, dementia, and falls in the very old.

Where determination gets hard

Some deaths resist classification because the evidence supports competing narratives. Falls in older adults are a canonical example: a fall may be the injury that killed the person, or a syncopal episode or stroke may have caused the fall, changing the underlying cause and the manner classification. Without a witness or autopsy, the certificate records the certifier's best reconstruction, and the data carry that uncertainty silently.

Drug-related deaths illustrate the reconciliation problem. Deaths involving multiple substances require the certifier to decide which drug, if any, initiated the fatal sequence, while toxicology quantifies concentrations that may reflect tolerance, postmortem redistribution, or therapeutic use. Postmortem redistribution—drug movement from organs into central blood after death—can raise measured concentrations substantially, so toxicology values are interpreted with site-of-draw and interval corrections rather than treated as simple thresholds.

Deaths without a body or with advanced decomposition force inference from circumstance, medical history, and partial remains. Infant deaths require scene investigation and history review because the same findings can support natural, accidental, or inflicted causes. Mass-casualty events overwhelm normal investigation capacity, and preliminary cause assignments are frequently revised after full examination. Each of these situations produces certificates that look definitive but rest on graded evidence.

Timeline: how a death investigation unfolds

Stage Typical timing What is established
Pronouncement Minutes Death occurred; identity not yet certified
External examination Within hours Injuries, identifiers, initial manner impression
Medical record review Hours to days Antecedent conditions and clinical trajectory
Autopsy Within 24 to 72 hours Organ pathology, injury tracks, natural disease
Toxicology Days to weeks Drug and metabolite concentrations
Histology and ancillary tests Weeks Microscopic disease, infection, metabolic clues
Certificate completion Days to weeks Cause and manner entered into the record
Amendment or correction Months to years Revised cause after new evidence

The timeline matters because provisional determinations become public before final ones. A death initially reported as natural may later be certified as an overdose after toxicology returns, and the statistics lag the revision. Systems that snapshot provisional data produce different counts than systems that wait for final certificates, which is one reason mortality estimates for drug deaths are routinely revised upward.

Misconceptions worth correcting

The most durable misconception is that cause of death is a single fact discovered by testing. It is a judgment constructed from evidence of varying quality, and different examiners can reach different conclusions from the same record. Inter-rater variability in cause-of-death certification is documented and is a known limitation, not a scandal.

A second misconception is that autopsy always reveals the cause. Autopsy has its own blind spots: metabolic deaths, arrhythmias, and some poisonings leave no structural trace. A negative autopsy means the cause was not anatomically visible, not that no cause existed. Certifiers then rely on history, scene, and toxicology, and sometimes must record an undetermined cause.

A death certificate is a medical opinion, not a laboratory result; do not treat it as verified fact. "Cardiac arrest" is a mechanism, not a cause; certificates listing it alone are incomplete. Manner of death (natural, accident, homicide, suicide, undetermined) is separate from cause. Toxicology concentrations are not directly comparable to clinical thresholds because of postmortem redistribution. Autopsy rates below 10 percent of hospital deaths mean most causes are never confirmed at autopsy (CDC). Ill-defined cause codes concentrate in the elderly and in under-resourced jurisdictions, so comparisons across regions carry coding bias.

Practical actions for families and researchers

Families can request the death certificate and, where the law allows, the medical examiner's report, which contains the investigation narrative that the certificate compresses. Discrepancies between the two are informative: the report often records uncertainty the certificate omits. Requesting an autopsy is time-sensitive and jurisdiction-dependent, and in some states the family must ask before the body is released.

Researchers should treat cause-of-death data as a measurement with known error. Link certificate data to autopsy registries where available, use multiple-cause files rather than underlying-cause fields alone, and stratify analyses by whether a death was medically examined. For drug deaths, expect revision between provisional and final counts and avoid comparing jurisdictions with different investigation standards. For policy work, the honest reading is that cause-of-death statistics support population inference and weak individual inference—a limit that should be stated, not hidden.

The unifying rule is reconciliation: when a certificate, an autopsy report, and a toxicology result disagree, the disagreement is data, not noise. Each threshold—autopsy positivity, toxicology concentration, cause-coding selection—is applied within a different evidentiary frame, and a certifier who ignores that produces false certainty. The system's output is a best reconstruction under explicit rules, and knowing the rules is what makes the number usable.