Summary
Published in Demography 55: 957–978 (2018). Proposes and applies a method for decomposing the young adult mortality hump — defined as excess mortality beyond the prevailing senescent level, not all deaths in a fixed age range — into cause-of-death contributions. Uses the nonparametric Sum of Smooth Exponentials (SSE) P-spline model and cause deletion to estimate each cause's contribution to the hump component. Applied to U.S. mortality by sex from 1959 to 2015, showing that trends in hump mortality diverge qualitatively from trends in observed death rates at the same ages, and that cause-of-death contributions to the hump can differ sharply in rank and magnitude from contributions to all deaths at ages 10–34.
Key Claims
- The hump should be defined as excess mortality relative to the prevailing senescent level. Some deaths at young adult ages belong to the senescent component; some hump mortality extends beyond age 30. Ignoring this overlapping structure misstates both the hump's trend and its causal composition.
- Method: SSE model (Camarda et al. 2016) decomposes the force of mortality into additive hump (H) and senescence (S) components using P-splines with shape constraints (log-concave for H, monotonically increasing for S). Cause decomposition by simultaneously refitting SSE on cause-deleted data with constraints that cause-specific contributions sum to all-cause hump and senescence. Implemented in the MortHump R package (CRAN).
- Seven causes identified as hump contributors via principal component analysis (PCA) on shape (first differences over age) of cause-specific vs. all-cause mortality: motor vehicle accidents, suicides, homicides, other accidents, poisonings (mainly drug overdoses), HIV/AIDS, maternal mortality (females).
- Divergent trends: 1971–1984 for U.S. males, all-cause life-expectancy loss (LEL) at ages 10–34 fell 18%, while hump LEL rose 7%. For females 1969–1985: all-cause LEL fell 27%, hump LEL rose 87%. Standard absolute-rate analysis would have missed this growing excess.
- Male hump cause composition shifted dramatically: traffic + other accidents
80% of hump LEL in the 1960s → less than one-third today, replaced by suicides + homicides (50%) and poisonings (~25% in recent decades).
- Female hump: more stable domination by traffic accidents, with recent surge in poisonings.
- HIV/AIDS played a much larger relative role in hump LEL than its share of absolute deaths at ages 10–34 would suggest. Antiretroviral therapies introduced in 1996 produced the sharp period decline in the male hump around 1997 (and coincided with homicide decline for females).
- Progressive widening of the hump from ~1980 (males) and ~1970 (females): a cohort effect from generations born after 1950 with elevated excess mortality in suicides, homicides, and poisonings extending the hump tail into the 30s. Abruptly reversed in the late 1990s; renewed widening since 2000 driven by opioid poisonings.
- Cause rankings differ substantially between hump-only and absolute 10–34 mortality: mean rank differences of 1.47 for males, 4.25 for females; in some years (1993–1996 for females) no cause occupies the same rank under both methods.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"We conceive of the young adult mortality hump as excess mortality beyond the prevailing senescent level of mortality. Research on young adult mortality should consider this difference when focusing on this specific phase of the life course."
"When isolated, trends in the U.S. hump differ qualitatively from trends in observed mortality rates in the same age range."
My Take
A methodologically innovative paper that solves a real measurement problem: standard young adult mortality studies use arbitrary age ranges that mix hump and senescent deaths, producing misleading trend and cause decompositions. The SSE+cause-deletion approach is elegant and well-implemented, and the MortHump R package makes it reproducible. The finding that hump trends and observed rates trend in opposite directions during key periods (1970s–1980s) is striking and has clear policy relevance. Limitation: the nonparametric P-spline approach requires substantial data; it may not be reliable for small subpopulations or countries with poor vital registration. The opioid result at the tail end (2000–2015) is timely; a follow-up through 2020 would capture the full opioid crisis and COVID-19's impact on young adult excess mortality.