The quick and the dead

Abstract

In this issue of the Journal, Lee et al. describe how they observed more postoperative complications in slow people 1. Speed was measured twice: by the distance walked in six minutes; and by peak oxygen consumption, pedalling a bicycle. In this editorial I consider the following questions. Can you predict peak oxygen consumption from the distance walked in six minutes? Might walking distance be useful, whether or not the prediction holds? And does the association of survival with physical fitness make sense; should the quick die young or old? The short answer is no. Some authors have interpreted small p values for Pearson's correlation coefficient as indicating good agreement between the two measures 2-4. Others have observed that, despite this, observed peak oxygen consumptions often differ from predicted values by more than one metabolic equivalent of task (MET), or 3.5 ml O2.kg−1.min−1 5. I will use Lee et al.'s data to illustrate this, the authors having kindly forwarded to me individual patient data. I generated predicted values for distance walked 6-11 and peak oxygen consumption 12 and superimposed them in red on the values they observed (Fig. 1). Clearly, equations that generate predicted values for healthy populations work poorly for this colorectal surgical cohort. So, rather than use inappropriate predictions, what happens if we use the regression line that Lee et al. plotted through their own data to generate predicted peak oxygen consumptions from observed walking distances? Two complementary tests can provide us with information on the (dis)agreement between observed and predicted oxygen consumptions: Bland and Altman's limit of agreement 13; and Lin's concordance correlation coefficient 14, 15. Figure 2 is a plot of the limit of agreement. The mean difference between observed and predicted values was 0.5 ml.kg−1.min−1. However, half of the predicted oxygen consumptions disagreed with those observed by 3.5 ml.kg−1.min−1 (one MET) or more. This variation in peak oxygen consumption is substantially more than that one would expect from repeating the bicycle test in the same cohort 16: one would expect 95% of repeated measurements to disagree by less than 2.3 ml.kg−1.min−1, in comparison with the 18.4 ml.kg−1.min−1 95% CI envelope in Fig. 2. Lin's concordance correlation coefficient is the product of precision (correlation) and accuracy (bias). Lin's coefficient is calculated from the difference between perfect concordance and the regression line (essentially the two dashed lines in Fig. 2), with perfect agreement generating a value of 1. One descriptive scale suggests that values more than 0.95 reflect substantial agreement, between 0.90 and 0.95 moderate agreement, and less than 0.90 poor agreement. The value of Lin's concordance correlation coefficient for observed and predicted peak oxygen consumptions is 0.65. In summary, oxygen consumption predicted from walking distance is likely to disagree substantially with measured consumption. Does it matter? Although the distance walked in six minutes is unlikely to generate estimates of peak oxygen consumption within one MET of measured values, Lee et al.'s data suggest that it might nevertheless fulfil the same function, quantifying the contribution of ‘fitness’ to survival and morbidity. Lee et al. increased the power of their study to detect associations of peri-operative factors with postoperative morbidity by combining outcomes. There are a number of problems with composite outcomes: the relationship of the composite with distance walked and other explanatory variables will be dominated by the most common outcome; the overall relationship is unlikely to be shared by all outcomes, with the rate of some increasing with the explanatory variable and others decreasing; and interpretation of the results is further complicated by the consequences of the component outcomes, ranging from trivial to severe, some of which required no intervention whilst others required further surgery and critical care 17, 18. These problems are in addition to those shared by unblinded observational studies, for instance confounding by intention. In Lee et al.'s paper, the composite ‘any morbidity’ was associated with lower peak oxygen consumption and younger age. One would expect postoperative complications to increase with age, not decrease. In their paper there was no univariate relationship between age and outcome: it is only the incorporation of age with peak oxygen consumption in a multivariable analysis that generates this unexpected relationship. The reason is, probably, due to the greater variation in absolute fitness in the young than the old. The range of relative mortality risk associated with fitness in the young is greater than in the old, but the range of absolute mortality risk is greater in the old, despite the smaller fitness range. Peak oxygen consumption decreases with age: 64 − (age × 0.56) ml.kg−1.min−1 for men 19 and 51 − (age × 0.46) ml.kg−1.min−1 for women 20. The mortality rate increases with age. At any age, mortality and morbidity rates are higher with lower physical fitness, with relative mortality risk increasing by 15% for a MET shortfall (compared with expected) for sex, age, height and weight 21. The strength of the relationship between physical fitness and morbidity and mortality is much stronger than the relationship between physical fitness and age itself. The reduction in physical fitness with age cannot account for the increase in mortality (and morbidity) with age. Figure 3 illustrates the observed mortality rate (black line), compared with the mortality a 20-year-old would have, should their fitness equal the observed average fitness at that age (red line). The complex association of physical fitness, mortality, morbidity and age may partially account for the paradoxical association between age and postoperative morbidity reported by Lee et al., with substantial credit being given to the play of chance and the heterogeneity of the composite outcome. Articles that highlight the association of oxygen consumption with survival raise the fundamental question: does speed save you or kill you? We are the product of more than a billion years of genetic variation and natural selection. The characteristic of oxidative capacity – along with all our other characteristics – has been selected, but only up to the age of sexual reproduction, beyond which we can only influence successful survival of ‘our’ genes by indirect effects, for instance grandparents looking after their grandchildren. Of course, only part of oxidative capacity is coded by ‘our’ (nuclear) genes, with a substantial contribution by genes inherited through oocyte mitochondria, which therefore – until recently – could not be transmitted beyond a maternal age of about 45 years. One can imagine that quick organisms, quick to catch, to flee, to eat, to procreate, were more likely to be our ancestors than the slow. But why would speed be associated with our longevity, particularly now that survival beyond reproductive age exceeds that preceding reproduction? Older people move more slowly and consume less oxygen: do fast people burn out young, leaving the slow to plod on, or were the elderly uncommonly quick when young? We are so accustomed to evidence that exercise prolongs life that we might forget to question the assumption that physical fitness per se – independent of acquired cardiovascular pathology – makes you live longer. The observation that the product of heart rate and longevity remained fairly constant across species led to the hypothesis that mammalian cells, and by extension other eukaryotic cells, had a metabolic lifespan, equivalent to a lifetime's consumption of oxygen per unit body mass: the faster the consumption the faster the organism aged and the shorter their lifespan 22, 23. The conclusion that animals die because they have consumed their allotment of oxygen is uncertain: maximum lifespan is a poor determinant of ageing; resting metabolic rate, with which longevity was associated, is not a good measure of average oxygen consumption; increased oxygen consumption does not necessarily increase the generation of reactive oxygen species, through which ageing might be mediated; the rate of oxygen consumption varies with tissue type; and the variability in oxygen consumption, within and between classes of animals, might undermine the hypothesis – birds, for instance, consume more oxygen over their relatively long lives compared with mammals of equivalent mass. Nevertheless, there remains enough evidence to support the contention that, in the end, oxygen kills you. Speed of movement is in part determined by speed of ATP production, coupled to mitochondrial oxygen reduction, so shouldn't the fast die young, assuming their speed is largely a consequence of the speed of oxygen consumption 24? The association of speed with longevity may not be causative, with the inter-species constancy of oxygen consumption per kilogram of body mass over a lifetime being the indirect consequence of other processes. Another solution categorises metabolic processes into those necessary to maintain health at rest (the resting or basal metabolic rate) and other types of work, including locomotion. Quick people might live longer than slow people if their average oxygen consumption was lower, particularly as basal oxygen consumption accounts for 50–70% of the total: you spend a lot of time not running 25, 26. Physical fitness only ‘explains’ a small fraction of mortality risk. The discrepancy between observed mortality rate and that one would expect based upon decreasing fitness alone increases with age, noticeably above the age of 60; this is important for those of us trying to generate predictive survival and morbidity models for pre-operative assessment (Fig. 3). Any general survival model, used to generate survival trajectories for the various treatment options that a surgical candidate might wish to exercise, has to incorporate more than physical fitness; consider the lower oxygen consumption but greater longevity of women, compared with men. This conclusion can be generalised: no single variable can estimate the extent of survival and quality of life that a patient needs to make an informed decision. Our task is to find out how to combine age, sex, clinical co-morbidities, physiological and biochemical measurements into a useful generic model that serves any patient, irrespective of the surgical pathology from which they suffer. One component is likely to be physical fitness, a component that could tolerate a number of different gauges, whether corridor walking distance, power, anaerobic threshold or peak oxygen consumption. Researchers, peri-operative or otherwise, might like to measure resting metabolic rate, as well as some measure of peak rate, to determine whether the gap between the two better correlates with survival and morbidity than either alone. No external funding and no competing interests declared.