The strong lung/non-lung cancer death rate associations observed among Massachusetts males suggest that tobacco smoke load is a potential cause of most prematurely fatal cancers in this population. These associations suggest that all-sites cancer death rate SAFs are 73% (SR 61–82%) for males over all ages and 74% (SR 61–82%) for males aged 30–74 years.
SAFs of age-adjusted cancer death rates calculated by this methodology are substantially higher than the cancer death rate SAFs calculated based on the CPS II relative risks. The present method incorporates all non-lung cancers; sizeable, representative, recent male populations; and age-adjusted death rates, which are the most reliable measure of progress in global action against cancer [16
]. In contrast, previous calculations of SAF were generally based on sometimes outdated lists of smoking-related cancer sites [1
]; tobacco smoke exposure based on smoking status, which extensively underestimates or misclassifies due to brief, unrecognized, disregarded, unrecalled or secondhand smoke exposure [4
]; as well as select smaller, racially homogeneous populations, that are unrepresentative of the general population or even of Whites only [5
The limitations of this study are noted. Extrapolation from population-level associations has previously resulted in inaccurate estimates of individual relative risk due to the ecologic fallacy [20
]. The results of these analyses may be unrepresentative of the association among females, in other US states, or of longer-term trends prior to ICD-9. The associations observed do not distinguish between types of smoking exposure due to in-utero, other secondhand smoke, sensitive stage (teenage) [21
], or active smoking since smoke load reflects cumulative lifetime damage [22
The validity of the present findings is reinforced by the strong and consistently positive smoke load/cancer death associations seen over time among both males of all ages and aged 30–74 years. In an earlier analysis among Black US males, lung cancer death rates predicted approximately 98% and 97% of the variances in non-lung cancer death rates throughout the 34% rise from 1969–1990 and the subsequent decline of 11% from 1990–2000, respectively [8
]. The present analysis of more recent, and more representative data, including comprehensive smoke load estimation, concurrent with outcome assessments is in contrast to the use of less recent and less-representative data reported in most cohort studies to date [4
] and less reliable and comprehensive point exposure measures. Those measures are often based on single smoking self-report observations that were assessed as long as decades before the measured outcomes [22
SAFs can be inaccurate and misleading under various circumstances such as omission of certain smoking-attributable cancers, use of unrepresentative relative risks, misclassified exposure status, and incomplete disclosure of assumptions concerning sensitivity. SAFs that are based on site-specific cancer death relative risks assume that smoking causes no cancer deaths at other cancer sites. The error in this assumption has been shown repeatedly in subsequent studies linking smoking to additional cancer sites [23
]. Artificially low relative risk estimates might result from high smoke exposures and death rates among "never smokers" in the cohort or from unrepresentatively low exposures and death rates among "smokers" due to low prior smoking or accelerated quitting among smokers informed of the risk of premature death.
Each of the above issues appears to be present in the CPS II cohort and the smoke effect estimates that are based on it. The CPS II definition of "lifelong never smoker" likely included many irregular, brief, or forgetful smokers and only excluded persistent, regular smokers (persons had who smoked at least one cigarette per day for one year) [24
]. Most CPS II "smokers" resurveyed at twelve years denied current smoking [4
]. This might help to explain the reason that concurrent national average male death rates greatly exceeded the average CPS II male death rates for all causes [4
] and exceeded even the CPS II "smoker" lung cancer death rate at ages 35–49 years [7
The smoke load/non-lung cancer death rate ratios observed in this study are remarkably consistent across time. Smoke load variation provides a likely and possibly causal explanation for a large majority of the cancer death rate disparities studied to date (see results) [7
]. Lung cancer death rates (smoke load) can explain 88% of the variance in non-lung-stomach-uterine corpus rates from 1985 to 2004 among Korean females [9
]. The estimated Korea female all-sites cancer death rate SAF in 2004 was 43% (sensitivity range 29–56%) [9
]. Residual confounding from smoke load variations [22
] may be an explanation for associations found in prior studies between cancer deaths and other epidemiological risk factors, such as cooking or outdoor air pollution, diets, industrial exposures, or medical treatments, made independently of smoking status or other poor proxies for smoke load [25
]. Alternatively, those non-tobacco exposures may theoretically cause premature lung and non-lung cancer deaths in the same ratio as does smoke exposure.
Nonetheless, the temporal trends of lung cancer epidemics in most countries seem to be more compatible with smoking patterns than with cooking, other air pollution, oral tobacco, or nutrition-driven epidemics. Other than declining death rates attributable to effective treatments for uterine, breast and cervical cancer, most currently available cancer treatments may have as little effect on smoke loads and premature mortality as treatments for premature mortality due to HIV, Kaposi's sarcoma, and pneumocystosis had before anti retrovirals were found that reduced HIV loads [30