A wide range of health, lifestyle and cognitive variables influence mortality [1]. Factors in old age that impact on survival are frequently related to life-long tendencies. Childhood intelligence would be an example of this with regard to cognitive variables in adulthood [2], though its effects may lessen for cardiovascular causes of mortality over the age of 65 [3]. In addition, there is a distorting factor that affects mortality of older Scottish men between 1980 and 2002: World War II. Paradoxically, men in the highest quartile for childhood IQ were significantly more likely to die during the war [2]. This means that there are relatively fewer men from this upper IQ quartile available to survive beyond 75, so that within the population the overall protective effect of higher IQ on survival would be expected to be less.

In Scotland cardiovascular disease is the greatest contributor to all-cause mortality [4]. Reductions in cardiovascular mortality are attributable both to modern treatments (medical and surgical) and to improvements in population risk factors (blood pressure, cholesterol and smoking) [4]. The relatively smaller improvements in mortality in those aged over 75 years may reflect a lack of evidence-based practice in this age group that has been historically poorly represented in heart disease trials [5]. Who receives such survival-improving treatment may reflect a combination of current behaviours (e.g. smoking) and life-long traits (e.g. education and social class). This may partly explain how socio-economic disadvantages persist into old age [6]. Moreover, effects of socio-economic disadvantage may be amplified by new population behaviours. For example, the increased prevalence of smoking among women that coincides with the epochs of recent mortality studies, has had a major effect [7]. On the other hand, discrete exposures may have age-dependent effects. An example is that non-cancer mortality related to the atomic bombings of Hiroshima and Nagasaki was highest in those aged 30–49 years in 1945 [8]. Since most people die in old age, determining the risk factors for mortality in older people is important, but requires a life-course approach to identify key influences on survival.

In this cohort of Scots born in 1921 survival beyond 76 years related to gender, cognition and physical health measures. The very narrow age range meant that chronological age, itself, did not predict survival. Furthermore, early-life influences, such as childhood IQ, did not have any significant effect in contrast to younger cohorts [2, 3]. Neither did variables relevant to earlier adult life such as education and occupation. Lower scores on several of the Wave 1 mental ability variables were associated with poorer survival: in a multivariate model UCO, a measure of executive function, was the best cognitive predictor. Amongst health variables, multivariate models showed that conventional measures of health status, such as diagnosed disease and medication use, were poor predictors of survival. Key health status predictors were respiratory function, mood score and smoking habit. The survival of participants who had given up smoking was no worse than those who had never smoked. Those who continued to smoke were more than twice as likely to die during the nine years follow-up. Lower mood was an important predictor of mortality even within the 'normal' range. A combined model showed that the major predictor of mortality at this age remained gender, even after other socio-demographic, cognitive and health status variables were adjusted for.

A study of 114 people aged 70 years and older from a Dutch population found that dyspnoea scores were significant predictors of death over an eight year follow-up period [17]. A systematic review of the associations between pulmonary function and cardiovascular mortality comprising a combined sample of 83,880 adults of all ages found that those in the lowest quintile of FEV₁ had nearly double the risk of dying compared with those in the highest quintile [18]. Our data are consistent with these findings and, in addition, show that respiratory function is a significant predictor independent of smoking habit, socio-environmental or early life influences. Smoking was related to survival in older men in the Western Collaborative Study [19], though this was unadjusted for pulmonary function. Smoking was a significant predictor of mortality independent of lung disease in the Asset and Health Dynamics Among the Oldest Old Survey [20]. Neither of these studies differentiated between current and ex-smokers. Our data suggest that smoking beyond 76 years of age continues to incur an increased risk of dying. Impaired cognition is also well recognised as a predictor of death, even when a wide range of other covariables are adjusted for [21]. Within the range of mental abilities, non-verbal tests were better predictors of survival in 546 non-institutionalised Texans aged over 70 years, in particular an executive clock-drawing task [22]. We also found a range of mental abilities associated with increased mortality, with executive function most strongly associated in multivariable models. It is possible that these findings reflect the influence of a key underpinning cognitive function, such as represented by reaction time that, itself, predicts survival [23].

The study is limited in several ways. First, although the narrow age range ensures that exposures are very similar for all participants (e.g. occupation, education, childhood experiences), it makes generalisation more difficult. For example, the majority of participants had outdoor sanitation at age 11, a situation very different in Aberdeen today. Secondly, the sample size means that important effects in small sub-groups could be missed. This is particularly relevant for multivariate models that included UCO where power was substantially reduced and consequently type 2 statistical error more likely. For this reason we did not investigate potential interactions (e.g. between sex and smoking). Thirdly, some variables that might be important in other populations were not considered (e.g. ethnic group – the sample is uniformly Caucasian). Fourthly, we chose backward elimination based on likelihood ratio change. Alternative strategies may have identified different predictors. Moreover, we did not include a time-dependent term in the models for much the same reason as that for interactions not being considered. Fifthly, the vital status for just over 1% of the participants was unknown at the census date because they had moved, though we had no evidence for their deaths being registered in Scotland. It is possible that they had emigrated from Scotland and died elsewhere. However, even if all five had died, it is unlikely to have altered the results substantially.

One advantage of a narrow age cohort is that key predictors of mortality can also be considered as measures of biological age. That is, although all the participants are almost identical in chronological age, some are at greater risk of dying than others. These individual differences in survival can be used as one criterion for biological age. In this cohort conventional health status items, such as diagnoses, were not good measures of biological age by this criterion. Respiratory function and cognition were better measures. But even after accounting for physical and mental health status, women were significantly biologically younger than men. Further investigations are required to elucidate the mechanisms underpinning this observation.

Mortality was common beyond 76 years of age in the ABC1921 cohort. Many socio-environmental variables associated with premature mortality no longer predicted death at this age. However, continuing to smoke still more than doubled the risk of dying. Both physical and psychological variables independently predicted survival in old age: respiratory function and executive function in particular. Male gender conferred increased risk of mortality and this was not explained by the broad range of socio-environmental, mental ability and health status variables examined in the study.