Young Adult Mortality Hump

Definition

The young adult mortality hump is a transient elevation in the force of mortality occurring roughly between ages $10$ and $35$ , superimposed on the underlying senescent trend. It is best defined as excess mortality beyond the prevailing senescent level — not all deaths in a fixed age range (e.g., $10$ – $34$ ). This distinction matters because some deaths in young adulthood are senescent (driven by early-onset Gompertz forces) while some hump mortality extends beyond age 30. Treating the two as interchangeable misstates both the hump's trend and its causal composition. Recognized since Thiele (1871), the hump received little systematic analytic attention until Remund, Camarda, and Riffe (2018) proposed a formal cause-of-death decomposition method. See Remund et al. 2018 — A Cause-of-Death Decomposition of Young Adult Excess Mortality.

Key Ideas

Three overlapping mortality phases: Ontogenescence (declining mortality in infancy/childhood), the hump (excess in young adulthood $\approx 10$ – $35$ ), and senescence (exponentially rising mortality from $\approx 30$ + onward, with possible late-life deceleration). These phases overlap — some senescent mortality occurs before age 30, and some hump mortality extends beyond it. A death at age $25$ may be senescent (area $A_1$ ) or part of the hump (area $A_2$ ).
Remund-Camarda-Riffe (2018) method: Sum of Smooth Exponentials (SSE) nonparametric P-spline model separates the force of mortality into hump ( $H$ ) and senescence ( $S$ ) components with shape constraints (log-concave for $H$ , monotonically increasing for $S$ ). Cause-of-death decomposition uses cause deletion: remove each cause, refit SSE simultaneously, and measure the drop in the estimated hump component. Implemented in the MortHump R package (CRAN).
Seven hump-contributing causes (U.S.): Motor vehicle accidents, suicides, homicides, other accidents, poisonings (mainly opioid/drug overdoses), human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS), maternal mortality (females). Selected via principal component analysis (PCA) on the shape (first differences over age) of cause-specific vs. all-cause mortality.
Hump $\neq$ deaths at ages $10$ – $34$ : Hump LEL (years of life expectancy lost to the hump) is smaller than LEL for all deaths at ages $10$ – $34$ , because senescent deaths overlap with that range. Critically, their trends can diverge: 1971–1984 U.S. males saw all-cause LEL at $10$ – $34$ fall $18\%$ while hump LEL rose $7\%$ .
Cause rankings differ: The relative ranking of causes contributing to hump LEL differs substantially from their ranking by absolute deaths at ages $10$ – $34$ (mean rank difference $1.47$ for males, $4.25$ for females). HIV/AIDS and traffic accidents are both more important in hump LEL than in absolute rates; in some years no cause occupies the same rank under both methods.

How It Works

The SSE model fits the all-cause force of mortality $\mu$ as the sum of two additive smooth components over ages $10$ – $90$ :

$\mu_H(x)$ : hump component (P-spline with log-concave shape — rises then falls during young adulthood)
$\mu_S(x)$ : senescence component (P-spline with monotonically increasing shape — Gompertz-like through old age)

Cause decomposition: simultaneously fit SSE on cause-deleted data with constraints that cause-specific contributions to $H$ and $S$ sum to the all-cause $H$ and $S$ . The constrained system is estimated via data augmentation (van Dyk and Meng 2001) with a composite link model. Cause-specific hump contribution $= \mu_H(\text{all-cause}) - \mu_H(\text{cause-deleted})$ . Life expectancy loss attributable to the hump and to each cause is computed via standard Arriaga decomposition.

U.S. Results (1959–2015)

Period	Male hump pattern	Female hump pattern
1960s	Traffic + other accidents $\approx 80\%$ of hump LEL	Traffic-dominant
1970s–1980s	Progressive widening into 30s (cohort effect)	Widening from ~1970
Late 1990s	Sharp collapse (HIV antiretrovirals 1996; homicide drop)	Homicide decline
2000–2015	Renewed widening from opioid poisonings	Poisoning surge

Male cause shift: traffic/other accidents ( $\approx 80\%$ in 1960s) $\to$ suicides + homicides ( $\approx 50\%$ ) + poisonings ( $\approx 25\%$ recent).

Age/period/cohort patterns:

Age: Hump onset stable at $\approx 15$ – $20$ across all causes (possibly reflecting adolescent neurological development — mismatch between prefrontal cortex and limbic system maturation)
Period: Sudden all-cause hump collapse ~1997 from HIV antiretrovirals (males) and homicide decline (females); homicide spike early 1990s
Cohort: Generations born after 1950 carry higher excess mortality in suicides, homicides, and poisonings, progressively widening the tail of the hump from the 1970s–1980s onward

Cross-National Comparative Evidence (Heuveline 2002)

Heuveline (2002) provides an early large-scale descriptive analysis of young adult violent mortality across 26 countries (1950–1996) using World Health Organization (WHO) Mortality Database data and age-sex standardization. This is a conceptual precursor to the SSE-based hump decomposition in that it documents the dominant contribution of behavior-related causes to young adult mortality, but it uses a fixed age range ( $15$ – $34$ ) and does not attempt to separate hump from senescent components — the methodological limitation later addressed by Remund et al.

Key comparative findings relevant to the hump concept:

The three violent causes (suicide, homicide, motor vehicle injury) rose from $26\%$ to $43\%$ of all-cause mortality at ages $15$ – $34$ between the late 1950s and mid-1970s; remained above $40\%$ thereafter.
The U.S. is a sharp outlier among industrialized nations: highest combined violent mortality, driven by exceptionally high homicide ( $14.06$ per $100{,}000$ vs. Finland's $2.88$ as the European maximum).
Russia has the highest post-1980 violent mortality of all 30 countries studied when Eastern European data are included.
A 5-component PCA on the 32-variable age-sex-cause matrix captures $95\%$ of variance; PC1 ( $35\%$ ) separates suicide from homicide/other unintentional; PC2 ( $24\%$ ) aggregates violent vs. accidental causes.

Heuveline's analysis illustrates what the hump looks like when measured via fixed age ranges and standardized rates rather than via excess-over-senescent decomposition. The two approaches can diverge in trend and cause-composition analyses, as Remund et al. demonstrated. See Patrick Heuveline and Heuveline 2002 — An International Comparison of Adolescent and Young Adult Mortality.

Secular Shift in Hump Timing (Goldstein 2011)

Goldstein (2011) uses the accident hump peak age as an indirect proxy for male sexual maturity timing, arguing that risk-taking behavior peaks at a biologically fixed lag after pubertal onset. Using the Gompertz residual method — fit Gompertz $M(x) = a \cdot \exp(b \cdot x)$ to ages $40$ – $80$ (senescent regime), compute residuals at $10$ – $40$ , identify peak via cubic spline — he extracts hump timing from Human Mortality Database (HMD) period data for Sweden (1751+), Denmark, Norway, England/Wales, and Italy.

Key findings:

Secular decline: $\approx 0.19$ y/decade from mid-18th century through mid-20th century; hump peak fell from $\approx 22$ – $24$ in 1750 to $\approx 17$ – $20$ by 1950.
Post-1950 halt: trend ceased in all five countries simultaneously, paralleling the slowdown in female menarche decline.
England/Wales exception: delayed change tracks delayed secular height gain from urbanization, supporting the nutritional-environment mechanism.
Corroboration: Voice-change age in Bach's Leipzig choir ( $\approx 18$ y, 1720s–1740s) vs. 20th-century London choirboys ( $\approx 13$ y) — two data points consistent with the $\approx 0.19$ y/decade rate.
Sex differential: male rate ( $\approx 0.19$ y/decade) is slower than female menarche rate ( $\approx 0.3$ y/decade), consistent with Life-History Theory (female reproductive timing more phenotypically plastic).
Social mismatch: biological maturity has moved earlier while social adulthood has moved later, widening the gap between the two.

This framework is distinct from the Remund-Camarda-Riffe approach (cause-of-death decomposition of hump level) — Goldstein instead tracks hump timing as a developmental clock. The assumption that the lag between puberty onset and peak risk-taking is constant across centuries and social contexts is asserted rather than tested. See Sexual Maturity Secular Trend and Joshua R. Goldstein.

Why It Matters

Measurement: Standard young adult mortality studies use fixed age ranges ( $10$ – $34$ or $10$ – $24$ ) that confound hump with senescent mortality, producing misleading trend analysis and causal inference. The hump definition isolates the demographically and socially specific process of excess young adult risk.
Policy: Identifying the causes driving the hump — period shocks (HIV epidemic, opioid crisis, homicide waves), cohort effects (post-1950 birth cohorts), age effects (adolescent risk-taking) — is essential for designing interventions targeting vulnerability during the transition to adulthood.
Opioid crisis: Poisoning contributions to both male and female humps are rising since ~2000, consistent with the opioid overdose epidemic. The hump framework isolates this as excess young adult mortality distinct from senescent poisoning deaths.
International comparisons: Cross-country comparisons of young adult mortality based on absolute rates at fixed age ranges will be confounded by differences in the senescent component and the location of the hump. The SSE method enables comparable hump-specific analysis.

Open Questions

Can the SSE+cause-deletion method be reliably extended to small subpopulations (racial/ethnic groups, states, counties) where small cells reduce P-spline stability?
How universal is the hump across high- vs. low-income country mortality schedules? In high-mortality contexts, the hump may blend with endemic infectious disease mortality.
What is the relative contribution of biological processes (adolescent neurodevelopment) vs. social structures (inequality, drug markets, policing) to the cohort and period patterns? The empirical patterns are consistent with both.
What does COVID-19 do to the hump? Preliminary evidence suggests young adult excess mortality rose sharply 2020–2021, potentially from both direct COVID deaths and indirect excess mortality from suicides, overdoses, and deferred care.