Life expectancy has risen sharply in the last 50 years. We applied the classic Michaelis–Menten enzyme kinetics to demonstrate a novel mathematical relationship of income to childhood (aged 0–5 years) and adult (aged 15–60 years) survival. We treat income as a substrate that is catalyzed to increase survival (from technologies that income buys) for 180 countries from 1970 and 2007. Michaelis–Menten kinetics permit estimates of maximal survival and, uniquely, the critical income needed to achieve half of the period-specific maximum. Maximum child and adult survival rose by about 1% per year. Critical incomes fell by half for children, but doubled for men. HIV infection and smoking account for some, but not all, of the rising critical incomes for adult survival. Altering the future cost curve for adult survival will require more widespread use of current interventions, most notably tobacco control, but also research to identify practicable low-cost drugs, diagnostics, and strategies.https://doi.org/10.7554/eLife.00051.001
In 1975 Samuel Preston published a classic paper that showed life expectancy was related to national income. When plotted as a graph, with national income on the horizontal axis and life expectancy on the vertical axis, the Preston curve shows that an increase in national income leads to an increase in life expectancy, with the increases in life expectancy becoming proportionally smaller as income increases. Moreover, Preston showed that innovations in healthcare (such as vaccinations, public health education, and sanitation systems) were increasing the maximum life expectancy that can be achieved for any given national income (defined as GDP per capita): this can be seen by comparing the Preston curves from 1960 and 2000 shown in Figure 1A. Indeed, global life expectancy increased by about 25 years over the course of the 20th century, which suggests that the level of daily income needed to achieve a certain life expectancy should be falling over time.
To explore this in greater detail, Hum et al. have constructed a mathematical model to investigate the relationship between health and income across different age groups and income levels. They found that most of the gains in life expectancy for low- and middle-income countries have been achieved by reducing child mortality, with gains in life expectancy for adults being restricted mostly to high-income countries. The model, which is based on the mathematical equations used to describe the kinetics of enzymatic reactions, makes it possible to estimate the improvements of health that can be made over time, and also the level of income that is needed to achieve these improvements.
In particular, Hum et al. have established a new parameter, the critical income, which is the level of income needed to achieve half of the maximal health found in high-income countries for the year in question. Based on available data from over 150 countries, they found that critical incomes fell by half for children between 1970 and 2007, but doubled for adult males during the same period. The rise in critical income for adults was due partly to the HIV epidemic and increases in smoking in low- and middle-income countries, reflecting the growing problems presented by noncommunicable diseases. Hum et al. conclude that increasing the survival among adults will require increased use of proven cost-effective interventions, most notably tobacco control, plus new research to identify low-cost drugs, diagnostics, and other public health strategies.https://doi.org/10.7554/eLife.00051.002
In most low-income countries, death in middle age is increasing in relative importance as other causes decrease and the effects of smoking increase (Jha, 2009). Most adult deaths are due to noncommunicable diseases, such as vascular, respiratory, or neoplastic diseases, with a significant minority due to HIV/AIDS, malaria, and tuberculosis (Beaglehole et al., 2011b). This has spurred calls for global efforts to tackle chronic diseases of adults akin to the already established global efforts to reduce childhood, maternal, and infectious disease deaths through the United Nations (UN) 2015 Millennium Development Goals (MDGs) (Beaglehole et al., 2011a).
In a classic paper, Preston (1975) offered an elegant analysis to establish a cross-sectional relationship between mean national life expectancy and national income per capita. The ‘Preston curve’ shows a positive relationship between national income levels and life expectancy in poorer countries but with smaller marginal returns at higher incomes. Preston noted an upward rise in the curve for three decades over the 20th century, leading to higher life expectancies for any given national income (defined as GDP per capita) level over time (Figure 1A is a reproduced graph of the Preston curve for the years 1960 and 2000). This upward rise has been attributed to the adoption of low-cost health technologies for child survival (Mathers et al., 2008; United Nations, 2011), such as immunization and oral rehydration, and to improvements in nutrition (Fogel, 1997), sanitation (Cutler and Miller, 2004), primary education for women (Caldwell, 1979; Cutler et al., 2006), or a combination of these variables (Rodgers, 2002). Indeed, some countries such as Costa Rica, Cuba, and Sri Lanka all achieve life expectancy levels greater than expected by their income (Caldwell, 1986).
Life expectancy has continued to increase globally, and this has led to the belief that less daily income is required to achieve a certain life expectancy level (Casabonne and Kenny, 2012). But these trends mask heterogeneity across age groups and income levels. In this article, we argue that the majority of gains in life expectancy for low- and middle-income countries have been achieved through reductions in child mortality while improvements in adult mortality have been restricted mostly to high-income countries. This has reduced inequality between countries for child survival, however, increased inequality for adult survival. The central objective of this article is to construct a mathematical model derived from the field of biochemistry/enzymology to analyze upward and lateral movements in the Preston curve. More specifically, we establish a new parameter ‘critical income’ (analogous to the Michaelis constant [Km]) to investigate the trends in income levels associated with reductions in mortality over the past 40 years; this model is further applied to age-specific populations to elucidate child and adult contributions to improving life expectancy. We also estimate the impact of HIV/AIDS and smoking on adult critical income values.
The relationship between health and income has been described using a variety of empirical models, generally of the form (Wagstaff and Van Doorslaer, 2000)
where hi is a health indicator (such as life expectancy or survival rates) and yi is income, for country or unit i.
Preston, originally, characterized the relationship between life expectancy (e0) at birth and gross domestic product per capita (‘GDP’) as a logistic function in the form
where GDP′ is GDP linearized on a scale of 0 to 1. Figure 1A is a reproduced plot of the original ‘Preston curve’, with an upward shift between the years 1960 and 2000 (Preston, 1975). Although the fit of this model to data is good, the model parameters are harder to interpret. It is suggested that identifying an inflection point (the point of diminishing returns) could clarify the analysis of these parameters (Rogers and Crimmins, 2011), for a generalized logistic function of the form
where ‘c’ is the maximal life expectancy and the inflection point is at a GDP value of ‘ln(a)/b’.
More commonly, however, a log-linear relationship is used as a simpler alternative to the logistic function
The Preston model explains the majority of the observed variance in life expectancy, but the coefficients (‘a’ and ‘b’) are not interpretable as no meaning exists for each coefficient nor is a theoretical relationship between them stipulated. The inadequacy of the Preston models to methodologically identify a single point (such as an inflection point; Rogers and Crimmins, 2011), which describes the changing curvature at lower income levels, has led to greater emphasis on the upward rise in the Preston curve, and less attention is devoted to lateral movements (along the income axis). Nonparametric regressions have been used to identify a ‘hinge’ on the Preston curve (Deaton and Case, 2009); however, we know of no known study that has explicitly quantified shifts across income levels.
Here, we identify a new construct called ‘critical income’ through the novel application of a mathematical model by Michaelis and Menten (MM) to empirically track child and adult mortality at different incomes. The Michaelis–Menten mathematical model first described in 1913 (Michaelis and Menten, 1913) was initially developed to analyze enzyme kinetics. Enzymes are biomacromolecules that act as catalysts, agents that accelerate the rate of a chemical reaction without being consumed in the process. In the absence of these enzymes, some thermodynamically favorable reactions may be kinetically hindered from occurring. For a single reaction, enzyme E binds to a substrate S to form an intermediate complex ES, which is converted into a product P and the original enzyme
Figure 1B is a plot of the MM equation for a hypothetical reaction. The classic MM equation describes the dependence of the enzyme velocity ν on substrate concentration [S]. ν asymptotically approaches a maximum value (νmax) at high [S] when enzymatic sites are saturated. Km is the half saturation constant—substrate concentration at which ν = 0.5νmax. Km is a function of the forward and reverse reaction rate constants, where a lower Km value indicates a more efficient catalyst
We extend the application of the Michaelis–Menten kinetics model to describe life expectancy, child, and adult survival. We use the analogy that GDP is a substrate, and health determinants and widespread applications of public health research, treatments and interventions are catalysts that increase health and survival (Ad Hoc Committee on Health Research Relating to Future Intervention Options, 1996). Indeed, infrastructure (such as water sanitation and education systems), vaccinations (leading to long-term immunity), and public health knowledge can be viewed as catalysts that are not consumed in entirety during the process.
where for a given year, the mean life expectancy at birth (e0) in a country is related to its GDP per capita (per day, at constant 2005 international dollars, adjusted for purchasing power parity and inflation). The MM model is characterized by two parameters: the life expectancy of the highest income countries (e0,max), and critical income (Kinc). We introduce a new parameter called ‘critical income’ as a meaningful construct that can be estimated to assess the relationship between income and mortality. This construct is defined as the level of daily income associated with the achievement of 50% of the maximum life expectancy (i.e., GDP per capita at e0 = 0.5 × e0,max); ‘maximum life expectancy’ is empiric, approximating the observed average life expectancy in high-income countries (Rodgers, 2002). While it is a biochemical convention to report the 50% mark, the critical income value is adaptable to determine higher fractional level of maximum life expectancy. In particular, two, four, and nine times the critical income yields the income required to achieve 66.7%, 80%, and 90% of the maximal life expectancy, respectively.
The asymptotic leveling of life expectancy at high income seen in the Preston or MM curves is analogous to the saturation of enzymatic sites at high substrate concentration, thus, further increases in GDP leads to only marginal increases in life expectancy.
This model establishes a systematic and empirical method for monitoring not only previously documented upward rises in the Preston curve but also potential shifts along the income axis. Whereas an upward rise can increase the maximum achievable survival for all countries, lateral shifts indicate an increase or decrease of income required to achieve this survival. These lateral shifts are particularly significant for more resource-limited countries that have yet to reach asymptotic leveling in the health and income relationship.
Age-specific contributions to changes in life expectancy are quantified using child mortality rate as the probability of a child born in a specific year dying before reaching the age of 5 years, referred to as 5q0, and the gender-specific adult mortality rate, representing the probability (for a given year) that an individual who has just turned 15 years will not reach the age of 60 years, referred to as 45q15. We transform these values to survival rates, where a mortality rate of 5 per 1000 corresponds to a 99.5% survival rate. The corresponding MM functions for child survival from age 0 to 5 years (5p0) and for adults from the age of 15 to 60 years (45p15), by gender, are
where in a given year, 5p0max and 45p15max are the maximum survival rates in high income countries and Kinc is the critical income associated with that age-specific group and gender.
Two large and widespread public health factors that have influenced mortality over the last four decades have been the HIV/AIDS pandemic and smoking, which makes more common most vascular, respiratory, and neoplastic diseases as well as tuberculosis (Gajalakshmi et al., 2003; Jha, 2009). We test if adjustment for the marked heterogeneity of HIV prevalence (between the ages of 15 and 49 years, as proxy for general population infection levels) and cigarette consumption (at ages 15 years or older) make less efficient the relationship between GDP per capita and adult mortality—as indicated by critical income values.
Figure 2A shows the graphical similarity between the logistic, log-linear, and MM model fits of life expectancy for the year 1990. For the years 1970 to 2007, there was no significant difference in the coefficient of determination (R2) for the logistic (M = 0.666, SD = 0.064), log-linear (M = 0.645, SD = 0.073), and MM regressions (M = 0.622, SD = 0.083; F(2,26) = 0.807, p=0.458, not statistically significant), indicating that the MM model is statistically as valid as the Preston log-linear and logistic models. See Table 1 for the regression coefficient for all years.
Although the fits of the models were comparable, the parameters of the logistic and log-linear models have less obvious explanatory value. For the logistic function, all inflection points were determined to be negative (from −$2.69 in 1970 to −$8.47 in 2007) and are impossible income values for any country. This suggests that the logistic function is overly complex and not necessary to model the data. For the log-linear function, the model parameters did not allow for a means to intuitively track lateral movements in the Preston curve. In addition, when using annual income (rather than daily income), the parameter ‘a’ was not statistically significant for any year from 1970 to 2000. Table 2 is a comparison of all three models for the year 1990.
For each year, the maximum life expectancy approximated the 5% trimmed mean life expectancy observed in countries with annual incomes greater than $12,276 (t(8) = −1.596, p=0.149, not statistically significant; Table 1). In addition, we performed a sensitivity analysis on the maximum life expectancy and critical income parameters and determined that both were not sensitive to the random removal of 5% of the data (t(8) = 1.07, p=0.314, not statistically significant; t(8) = 1.46, p=0.178, not statistically significant, respectively; Table 1).
An upward rise is observed for life expectancy and income from 1970 to 2007 (Figure 2B). The maximal life expectancy rose from 67.8 years (95% uncertainty interval 65.4–70.1 years) in 1970 to 75.5 years (95% uncertainty interval 73.9–77.1 years) in 2007. This change in maximal life expectancy represents a linear increase of 75 (95% uncertainty interval 48–99 years) days per calendar year (R2 = 0.875) over the last 40 years, and is comparable to the life expectancy increase of almost 90 days per calendar year for the 20th century (Oeppen, 2002).
In addition to maximal life expectancy increase, a lower national income is associated with a higher life expectancy now than it was 40 years ago. The critical income (in constant 2005 international dollars) needed to achieve half of maximal overall life expectancy declined from $1.48 ($1.18–$1.78) in 1970 to $1.21 ($0.98–$1.44) in 2007, equivalent in 2007 to the extreme poverty line of $1.25 per day; this represents an 18% decrease in critical income to gain almost 4 additional years of life expectancy. Critical income declined linearly at a rate of −$0.09 per decade (−$0.06 to −$0.13, R2 = 0.839). Table 1 also reports the incomes required to achieve 66.7%, 80%, and 90% of the maximal life expectancies.
Maximum survival for all age groups rose at statistically the same rate between 1970 and 2007 (Figure 3A, Table 3). Maximum child survival to the age of 5 years increased from 94.5% (92.7–96.2%) to 98.0% (97.3–98.6%), an increase of 0.8% (0.3–1.3%) per decade (R2 = 0.665); maximum adult female survival at ages 15–59 years rose from 85.9% (84.1–87.7%) to 90.1% (87.8–92.3%), an increase of 1.1% (0.5–1.6%) per decade (R2 = 0.762); and maximum adult male survival at ages 15–59 years rose from 77.7% (75.6–79.7%) to 82.1% (79.0–85.0%), an increase of 1.1% (0.6–1.7%) per decade (R2 = 0.749).
However, critical incomes diverged dramatically for children and adults (Figure 3B). From 1970 to 1980, the critical income values for children, adult males, and adult females were statistically equivalent, between $0.54–$0.58 per day. For children, the critical income values declined gradually, however, with a large drop in 1990–1995. Over the 40-year period from 1970 to 2010, the critical income for child survival fell by over half from $0.58 ($0.46–$0.70) to $0.24 ($0.20–$0.28). In contrast, critical income more than doubled for adult male survival from $0.54 ($0.38–$0.71) to $1.20 ($0.81–$1.58) and rose over 50% for adult female survival from $0.57 ($0.43–$0.70) to $0.89 ($0.65–$1.12) (see Table 3 for all years and Supplementary file 1 for each country-specific critical income). These percentage increases were similar even if critical income was defined differently, for example, income needed to achieve 66.7%, 80%, or 90% of maximum survival (not shown).
The 1970 and 2007 survival curves for adult men crossed over at a value for $10.95 GDP per capita per day. There are 58 countries with a total adult male population of approximately 780 million (or 35% of the world adult male population) below this income value. For these countries, adult male survival was lower in 2007 than in 1970 (Figure 4A). For women, the comparable 1970 and 2007 survival curves crossed at a value of $5.93 GDP per capita per day, corresponding to 32 countries, with approximately 150 million women (or 7% of the world adult female population), where female survival in 2007 was worse than in 1970 (Figure 4B). By contrast, between 1970 and 2007, for the 58 countries with an income under $10.95 per capita per day in 2007, child mortality improved by 11.8%, adult male survival fell by 2.7%, and incomes rose by 8.9% (or an absolute increase of $0.98).
Particularly given the marked increase in critical incomes needed to achieve maximal adult male survival, we tested if HIV infection and deaths and smoking (both, greater in males than in females) explained the increasing critical income values for adult males (Figure 5). For adult males in the year 2000, HIV prevalence (range 0.06–26% aged 15–49 years) and cigarette consumption (range 54.6–3385.2 per year at ages 15 years or older) were shown to influence critical income. Selecting the 86 countries with complete survival, income, HIV, and cigarette data, the critical income value was $2.02 ($1.32–$2.73). Adjusting for HIV and smoking prevalence to the average across countries reduced the critical income to $1.22 ($0.76–$1.69), meaning that HIV and smoking could explain about half of the increases in critical income. In 1970, the critical income was $0.54 ($0.38–$0.71), thus, HIV infection and smoking do not explain all of the worsening of critical income observed by the year 2000. HIV infection and smoking had no statistical impact on the maximum survival.
These results are further supported by a first differences analysis (Table 4). Over the 10-year period from 1990 to 2000, a rise of one cigarette consumed per day per person and 1% in HIV prevalence led, for adult men, to a $1.70 ($0.96–$2.43) and $0.70 ($0.27–$1.13) increase in critical income, respectively. For adult females, the impact of a 1% rise in HIV prevalence per capita was associated with a $0.40 ($0.31–$0.49) increase in critical income; however, the impact of smoking was lower, with a change of $0.38 ($0.23–$0.54). Moreover, adding a covariate for HIV prevalence also improved the goodness of fit dramatically for all years and genders (Table 3).
Our establishment of a new parameter ‘critical income’ provides novel insights into the relationship of global mortality changes with income. It builds on the well-established Preston curve functions by quantifying the rise in survival. More profoundly, our analyses reveal that while less and less income is required to improve childhood survival, the opposite is true to improve adult survival—particularly in low- and middle-income countries.
Unlike for adults, child survival replicates the upward and lateral trend in the original Preston curve for life expectancy. This suggests that for low- and middle-income countries, the majority of the past gains in life expectancy have been achieved via declining child mortality. This is consistent with the UN Population Division trends given the large impact of child mortality (compared to adult mortality) on overall life expectancy (United Nations, 2011). The greatest decline in child critical income was achieved after 1990, coinciding with actions following the UN's World Summit for Children. With justification, cost-effective interventions have been disproportionately devoted to child and maternal health (Daar et al., 2007), and more recently to control of infectious diseases such as HIV/AIDS, malaria, and tuberculosis. Increasing coverage of inexpensive health interventions such as immunization, insecticide-treated nets, and case management of childhood infections could be contributing to the decline in critical income for child survival (Jamison et al., 2006; Mathers et al., 2008). Moreover, there might be complementary benefits of education in reducing child mortality (Gakidou et al., 2010). Our study does not address any causal relationship between such interventions and reductions in child mortality; however, our results imply that the achievement of the UN MDG 4 (to reduce child mortality by two-thirds from 1990 levels) might be due to the falling levels of income needed to increase child survival.
For adult survival, however, there is a reversal of fortune. While achievable adult survival rates have improved, improvements are only associated with those countries at higher income levels. The rise in critical income suggests that the marginal costs of increasing adult longevity are rising; this may explain the lower rate of decline in adult mortality in countries with low income (Rajaratnam et al., 2010). The emergence of HIV/AIDS in the 1980s and the rise in global smoking prevalence in low- and middle-income countries (Guindon and Boisclair, 2003) can explain much, but not the entire rise in critical income for adult males. For males, the rise in smoking accounts for over 40% more of the variance in critical income compared to 30% for HIV/AIDS. For adult women, the impact of cigarettes on critical income is much lower, which reflects the five times lower prevalence of smoking among women compared to men (Guindon and Boisclair, 2003). Indeed, previous studies have already highlighted the impact of smoking and HIV/AIDS on adult survival in developing countries. Even low levels (4%) of HIV prevalence in rural Tanzania can increase overall adult mortality by more than 50% (Todd et al., 1997). Our findings also show that smoking increases critical income but has no statistical impact on maximum survival. This is in line with trends in global smoking, where prevalence of smoking (and subsequently the deaths attributed to smoking) are rising in low- and middle-income countries but declining in high-income countries (Jha, 2009).
Under the current conditions, an approximate national income per capita of $2.20 per day would be required in 2007 to attain the same achievable adult male survival rate with $1.25 per day in 1970. Moreover, should the critical income costs for adults continue to rise (in line with current trends), survival rates for low- and middle-income countries might well deteriorate into the future. In contrast, high-income countries have benefited from the rise in maximum survival among adults. This is likely due to more widespread availability of secondary treatments for chronic diseases, most notably for vascular disease, and in particular from sharp reductions in smoking (Jha, 2009).
The probabilities of premature adult deaths before the age of 70 from vascular, respiratory, and neoplastic diseases are remarkably similar in low-, middle-, countries and high-income countries (Strong et al., 2005). Low-cost, effective, and feasible interventions (Jha et al., 2002) exist against these diseases, most importantly tobacco control (Jha, 2009), but also increased reduction in hazardous alcohol intake (particularly in former Soviet states; Zaridze et al., 2009), and low-cost drugs for secondary management of existing disease. However, these interventions are still not widely used in low-income countries (Jha et al., 2012). Over the past few decades, research and development of new technologies (drugs, vaccines, and policies) have focused mostly on childhood and infectious disease, with fewer worldwide investments for adult chronic diseases (Ad Hoc Committee on Health Research Relating to Future Intervention Options, 1996). Thus, the longer-term trajectory of critical incomes for adult survival might well depend on the development of newer interventions, as well as more widespread application of interventions already proven to be cost effective (Jamison et al., 2006).
This is, as far as we know, the first application of enzyme kinetics to mortality changes. There is no a priori reason to exclude a relationship of mortality and income being comparable to biological reactions. Indeed, since its initial formulation in 1913, the MM equation has been sufficiently adaptable to explain various levels of biological complexity from thousands of single enzyme reactions (English et al., 2005), reversible and quasi-steady state systems (Briggs, 1925), the growth of microbial cultures on a nutrient substrate (Monod, 1949), the growth in size of a variety of different animal species (López et al., 2000), and the ability of organisms to acclimate to changes in environmental conditions (Bonachela et al., 2011).
Like all models, ours faces certain limitations. The relationship of income to mortality is comparable using the log-linear, logistic, MM, or indeed other methods. But none of the models can directly elucidate the mechanism(s) that converts income into better survival. The novel insight from the MM method is that we can model income as a substrate that catalyzes further changes that more directly impact survival. National income is quite large in relation to health spending for most countries (the median national spending on health is 5% of the GDP, and the United States at about 18% of GDP represents an outlier). Thus, income fulfills, partially, the MM definition of a substrate that is an input converted into health interventions and their use. We also reason that income directly enables certain technologies, immunization programs, epidemiological knowledge, education, and sanitation systems and other areas, which may themselves be interpreted as ‘catalysts’—agents that accelerate the rate of a reaction without being fully consumed in the process.
Moreover, the MM method advances some understanding of mechanisms by defining critical income levels that measure the efficiency of particular countries using available income to keep up with maximal achievable survival by other countries during that narrow 5-year time period. The new variable critical income (akin to Km) provides a measure to observe a ‘lateral shift’ that was not possible mathematically with the log-linear or logistic models. Additionally, the obvious impacts of two large risk factors (HIV and tobacco) on adult survival suggest some mechanistic insights between income and survival (and indeed the fact that male survival was more unequal from greater male smoking, and female and male survival were affected equally by the more even spread of HIV infection, further strengthens this case).
We caution however that full explanation of the links between specific technologies catalyzed by income need more research. For example, the Monod equation (an adaptation of the MM equation to fit the growth of bacterial cultures; Monod, 1949), which was first proposed in 1949, was only interpreted thermodynamically 50 years later (Liu et al., 2003). Nonetheless, this equation was (and continues to be) used extensively in the pharmaceutical and food industries as well as in waste treatment systems. Additionally, our analyses could be considerably strengthened by examining trends in age, gender, and cause-specific mortality, but cause-specific mortality data are simply unavailable for most countries; for instance, notwithstanding global efforts to identify data for child mortality, less than 3% of all child deaths worldwide were certified according to cause of death (Liu et al., 2012). Indeed, expanded efforts to measure causes of death is a big global priority (Jha, 2012; Vogel, 2012) and the development of new physically or biologically inspired analytical models can help elucidate better understanding of global mortality trends.
Data for 180 countries at 5-year intervals from 1970 to 2007 were included in the analysis, with definitions of income as per those from the World Bank (Supplementary file 1); the Institute for Health Metrics and Evaluation (Institute for Health Metrics and Evaluation, 2012), UN Population Division (United Nations, 2011), UNAIDS online database (UNAIDS, 2010), and Penn World Table 6.3 databases (Heston et al., 2009) were used for child and adult mortality, life expectancy and country population by age-groups, prevalence of HIV among adults aged 15–49 years, and GDP per capita at constant 2005 international dollars and purchasing price parity, respectively. Cigarette consumption per capita is from the American Cancer Society (Guindon and Boisclair, 2003). Availability of data for all 180 countries was not consistent across all years and all variables. Data from missing countries (by individual year and population group and disease and risk burden) were removed from that specific regression analysis. A list of countries is found in Supplementary file 1. The UN Population Division database for child (UN Inter-agency for Child Mortality Estimates, 2012) and adult (United Nations, 2011) mortality were used to confirm results. To confirm the completeness and accuracy of the Institute for Health Metrics and Evaluation database, statistical analysis was duplicated using 5-year rolling averages from UN Inter-agency Group for Child Mortality Estimates (UN Inter-agency for Child Mortality Estimates, 2012) and UN Population Division database adult mortality (United Nations, 2011), with general consistency in the trend lines (Figure 6A,B). 5-year rolling averages also increased the R2 values compared to single-year regressions; for regressions using UN datasets, the average R2 rose to above 0.5 for all age groups (including high HIV-prevalent countries)—compared with lower correlation coefficients for single-year regressions using the Institute for Health Metrics and Evaluation dataset (Table 3).
For the Michaelis–Menten adapted model and the logistic model, a nonlinear regression analysis (using iterative parameter estimation algorithms; Greco and Hakala, 1979) was used to calculate the parameters. The logistic and log-linear regressions were analyzed for life expectancy to compare the goodness of fit. To study progress over time, we calculated these eo,max, 5p0max and 45p15max, and Kinc values for each age group and gender at 5-year intervals from 1970 to 2007. Countries with missing mortality or income estimates were eliminated from the regression analysis on a year-by-year basis. We also computed estimates of critical income for each specific country, year, age group, and gender, assuming that the maximum survival was a constant across all countries. Additional analyses were completed by adding covariates for HIV prevalence in a given year (from 1990 to 2007) and cigarette consumption per capita (in the year 2000). For the year 2000, analysis was restricted to countries with complete data for both HIV prevalence and cigarette consumption (n = 86). First differences analysis was used to confirm the impact of HIV/AIDS prevalence and cigarette consumption on critical income values over the 10-year period from 1990 to 2000, such that
We used SPSS (version 19) to conduct all regression analyses.
Investing in Health Research and Development: Report of the AD Hoc Committee on Health Research Relating to Future Intervention Options, Convened Under the Auspices of the World Health OrganizationGeneva: World Health Organization.
UN High-Level Meeting on Non-Communicable Diseases: addressing four questionsLancet 378:449–455.
Priority actions for the non-communicable disease crisisLancet 377:1438–1447.
Dynamic model of flexible phytoplankton nutrient uptakeProc Natl Acad Sci USA 108:20633–20638.
A further note on the kinetics of enzyme actionBiochem J 19:1037–1038.
Education as a factor in mortality decline an examination of Nigerian dataPopulation Stud 33:395–413.
Routes to low mortality in poor countriesPopulation and Development Review 12:171–220.
Developments in the Economics of Aging317–349, Developments in the Economics of Aging, Chicago, University of Chicago Press, p..
The best things in life are (nearly) free: technology, knowledge, and global healthWorld Dev 40:21–35.
The role of public health improvements in health advances: the twentieth-century United States1–22, Demography, 42, NBER Working Paper Series.
The determinants of mortality (No. w11963)Cambridge, National Bureau of Economic Research.
Grand challenges in chronic non-communicable diseasesNature 450:494–496.
Ever-fluctuating single enzyme molecules: Michaelis-Menten equation revisitedNat Chem Biol 2:87–94.
New findings on secular trends in nutrition and mortality: Some implications for population theoryIn: MR Rosenzweig, O Stark, editors. Handbook of Population and Family Economics, Volume 1. New York: Elsevier. pp. 433–481.
Smoking and mortality from tuberculosis and other diseases in India: retrospective study of 43000 adult male deaths and 35000 controlsLancet 362:507–515.
Increased educational attainment and its effect on child mortality in 175 countries between 1970 and 2009: a systematic analysisLancet 376:959–974.
Evaluation of methods for estimating the dissociation constant of tight binding enzyme inhibitorsJ Biol Chem 254:12104–12109.
Past, Current and Future Trends in Tobacco UseSan Francisco: Center for Tobacco Control Research and Education.
Disease control priorities in developing countriesDT Jamison, JG Breman, AR Measham, G Alleyne, M Claeson et al., editors. New York: Oxford University Press.
Avoidable global cancer deaths and total deaths from smokingNat Rev Cancer 9:655–664.
Counting the dead is one of the world’s best investments to reduce premature mortalityHypothesis 10.
Improving the health of the global poorScience 295:2036–2039.
A thermodynamic interpretation of the Monod equationCurr Microbiol 46:233–234.
Global, regional, and national causes of child mortality: an updated systematic analysis for 2010 with time trends since 2000Lancet 379:2151–2161.
A generalized Michaelis-Menten equation for the analysis of growthJ Anim Sci 78:1816–1828.
The global burden of disease: 2004 updateGeneva: World Health Organization.
Die fermente und ihre wirkungenBiochem z 49:333–369.
The growth of bacterial culturesAnnu Rev Microbiol 3:371–394.
Demography. Broken limits to life expectancyScience 296:1029–1031.
The changing relation between mortality and level of economic developmentPopulation studies 29:231–248.
Worldwide mortality in men and women aged 15–59 years from 1970 to 2010: a systematic analysisLancet 375:1704–1720.
Income and inequality as determinants of mortality: an international cross-section analysisInt J Epidemiol 31:533–538.
International handbook of adult mortality, Vol. 2RG Rogers, EM Crimmins, editors. New York: Springer.
Preventing chronic diseases: how many lives can we save?Lancet 366:1578–1582.
HIV-associated adult mortality in a rural Tanzanian populationAIDS 11:801–807.
World Mortality Report 2011New York: United Nations.
World Population Prospects: The 2010 Revision, Highlights and Advance TablesNew York: United Nations.
The segmented trend line of highest life expectanciesPopul Dev Rev 35:159–187.
Global health. How do you count the dead?Science 336:1372–1374.
Income inequality and health: what does the literature tell us?Annu Rev Public Health 21:543–567.
Alcohol and cause-specific mortality in Russia: a retrospective case-control study of 48,557 adult deathsLancet 373:2201–2214.
Eduardo FrancoReviewing Editor; McGill University, Canada
eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.
Thank you for choosing to send your work entitled “Global divergence in critical income for adult and childhood survival: analyses of mortality using Michaelis-Menten” for consideration at eLife. Your article has been evaluated by a Senior Editor, a Reviewing Editor, and by 3 external peer reviewers. The Reviewing Editor has assembled the following comments based on the reviewers' reports.
The reviewers' two key concerns are about your use of the Michaelis-Menten analogue and whether your conclusion about the potential role of HIV infection and smoking on the final impact of income on health is really justified. These concerns are articulated below. We expect you will be able to address them in your revision.
Comments relating to the Michaelis-Menten function:
* The use of the MM function to model the relationship between national income and health needs to be better justified. In what way is national wealth converted into better health by the action of an “enzyme”, i.e. something (measured by v-max) limited in quantity but not itself used up in the process?
* If the MM function is not used for mechanistic reasons, and is just a convenient form of function to explain the relationship between GDP and life expectancy, then it needs to be justified statistically. This is problematic because it appears to fit just as well (or badly) as the log-linear model. There is no attempt to show that the log-linear model is a worse fit to countries at low-income levels.
* K_inc is intuitively easy to explain but is a problematic indicator because virtually all the data points are above it. Hence it may not be a very robust parameter to estimate, and is also difficult to visualise what a country with such a low-income level may look like. It might be better to take the daily level of income associated with achievement of the median life expectancy (for example). It would be good also to explore how sensitive to data the conclusions are about this critical income level, e.g., by removing certain data points.
* The authors should explain more clearly why the more classic logistic function (or one of the many S-curve variations are either inappropriate or do not have the adequate properties.
* Since the model generated by Michaelis-Menten dynamics clearly reaches an asymptote, one needs to fit it to a function that converges to an asymptote. There are essentially two ways to do this: either 1) use a converging exponential a*(1-exp(-b*x)) where a would be the asymptote, be the rate of increase and x would be GDP, or something similar with an extra parameter such as a logistic curve, or 2) use a polynomial function a*x/(1+b*x) where a, b, and x would play more or less the same role, or again something similar with an extra-parameter which could involve powers. What would make the study more interesting would be if one could either a) choose between the several possible functions on the basis of the data or b) choose between the several possible functions on the basis of a theoretical understanding of what is going on.
Comments relating to health significance:
* The important influence of HIV and smoking sounds very plausible, but is not incorporated into the model in a systematic way. A better way would be to take the list of top 5 or 10 (say) childhood mortality causes for which there are adequate data, and incorporate them all into the model, then remove the ones that are less important by some variable selection technique. Currently, it isn't clear whether HIV and/or smoking are just proxy variables for some other causes of death, or even the general amount of effort placed on public health issues in a given country.
* There have been dramatic changes in the three leading risk factors for the global disease burden over the last 20 years. Currently these are high blood pressure, tobacco smoke, and alcohol use compared to 1990, when the top 3 risk factors were childhood underweight, household air pollution from solid fuel use, and tobacco smoke. This has been very well described in a recent analysis in the Lancet. Income alone does not adequately explain these changes – rather it is a reflection of a range of complex interactions of which GDP and family income are two factors. It is unclear to us how this analysis deals with the complex web of interactions that impact on life expectancy or the burden of disease and our read of the manuscript did not leave us with a clear indication of how this enzyme mathematical model adds value as a new approach to the problem if it does not deal with the multitude of variables impacting life expectancy.https://doi.org/10.7554/eLife.00051.014
We believe our paper has benefited from the reviewer comments. To address the specific comments:
1. Comments relating to the Michaelis-Menten function:
a. The use of the MM function to model the relationship between national income and health needs to be better justified. In what way is national wealth converted into better health by the action of an “enzyme”, i.e. something (measured by v-max) limited in quantity but not itself used up in the process?
Response: We have expanded the discussion of limitations of applying the MM to model the relationship between national income and health. First, we cannot attribute a theoretical mechanism to our model; but we reason that income directly enables certain technologies, immunization schemes, public health knowledge, education, and sanitation systems, which may be interpreted as “catalysts”, whose effects are long lasting, are not consumed in the reaction (or on minimally consumed as they cost only a very small % of GDP), and can increase the efficiency of the transformation of income to health. Second, the main contribution of our finding is to identify a new variable (akin to Km), which provides a measure of efficiency or “lateral shift” that simply was not possible mathematically from the earlier adaptations of the Preston curve. This critical income value is also directly policy relevant – it shows marginal costs are falling for children but rising for adult survival.
Historically, the literature has often identified a relationship but determined the mechanism much later. For example, Jacques Monod who in 1949 adapted the MM equation from single enzymes to fit the growth of bacterial cultures: his work was only interpreted thermodynamically 50 years afterwards.
These points are now added to the methods and discussion.
b. If the MM function is not used for mechanistic reasons, and is just a convenient form of function to explain the relationship between GDP and life expectancy, then it needs to be justified statistically. This is problematic because it appears to fit just as well (or badly) as the log-linear model. There is no attempt to show that the log-linear model is a worse fit to countries at low-income levels.
Response: Notwithstanding the merits of the analogy, we agree that the Preston curve can be modeled empirically using a variety of mathematical functions with a converging asymptote. To that end, we have expanded our analysis in the manuscript to include regressions using a logistic, log-linear and MM model. (See subsequent comment replies.) We have removed the earlier confusing citations regarding a worse fit for low-income countries.
We have also clarified that the MM is not simply a different ways of showing the relationship between GDP and survival (and that the usual tests of statistical fitness are as robust as alternatives), but that it provides a new metric to capture lateral shifts or efficiency – that is critical income.
c. K_inc is intuitively easy to explain but is a problematic indicator because virtually all the data points are above it. Hence it may not be a very robust parameter to estimate, and is also difficult to visualize what a country with such a low-income level may look like. It might be better to take the daily level of income associated with achievement of the median life expectancy (for example). It would be good also to explore how sensitive to data the conclusions are about this critical income level, e.g., by removing certain data points.
Response: We agree that the 50% value is a relatively low value and that not many countries have such a low income level. However, we believe strongly in retaining this as: (a) it’s conceptually simpler; (b) it makes the mathematics and their future use, say applied to specific diseases, easier. That being said, the critical income value is adaptable to determine higher fractional level of maximal life expectancy. For example, two-, four- and nine-times the critical income would yield the incomes associated with 66.7%, 80%, and 90% of the maximal life expectancy, respectively. The proof that four times the critical income yields the income (GDP) required to achieve 80% of the maximal life expectancy is below:
In addition, we show that our MM model asymptote (LEmax) approximates the 5% trimmed life expectancy of high-income countries. We also perform a sensitivity analysis on the model parameters by randomly removing 5% of the countries in the dataset; we show that the critical income and maximal life expectancy are not significantly impacted.
d. The authors should explain more clearly why the more classic logistic function (or one of the many S-curve variations) are either inappropriate or do not have the adequate properties.
Response: See response above. For the logistic model, we find the fitness to be statistically equivalent to that of the MM analogue; however, the interpretability of the model parameters is less intuitive given that all the inflection points were irrational (negative) income values. For the log-linear regressions, without transforming the income from annual to “per day” values, many of the model parameters were not statistically significant.
Given the fitness of the model and the additional tests, we believe this model generates a unique, easy to generate and extremely useful parameter in critical income that is not found in the other log-linear and logistic models.
e. Since the model generated by Michaelis-Menten dynamics clearly reaches an asymptote one needs to fit it to a function that converges to an asymptote. There are essentially two ways to do this: either 1) use a converging exponential a*(1-exp(-b*x)) where a would be the asymptote, be the rate of increase and x would be GDP, or something similar with an extra parameter such as a logistic curve, or 2) use a polynomial function a*x/(1+b*x) where a, b, and x would play more or less the same role, or again something similar with an extra-parameter which could involve powers. What would make the study more interesting would be if one could either a) choose between the several possible functions on the basis of the data or b) choose between the several possible function on the basis of a theoretical understanding of what is going on.
Response: Thank you. We have included further analysis in our manuscript regarding the logistic, log-linear and MM functions. We did not however add the exponential function (a*(1-exp(-b*x), as this function resulted in a statistically worse fit (R2 mean =0.452, SD=.106) for the data compared to the MM function (t(8)=-14.5, p<0.0001). In addition, the polynomial function a*x/(1+b*x) is algebraically identical to the MM function, where a=LEmax/Kinc and b=1/Kinc.
2. Comments relating to health significance:
a. The important influence of HIV and smoking sounds very plausible, but is not incorporated into the model in a systematic way. A better way would be to take the list of top 5 or 10 (say) childhood mortality causes for which there are adequate data, and incorporate them all into the model, then remove the ones that are less important by some variable selection technique. Currently, it isn't clear whether HIV and/or smoking are just proxy variables for some other causes of death, or even the general amount of effort placed on public health issues in a given country.
Response: Our objective was two-fold. First, to provide the MM construct and show it is empirically robust. The second was to test which are the determinants of Critical Income. The choice of HIV/AIDS and tobacco is obvious as these are large and still growing causes of death for adults (Todd et al. HIV-associated adult mortality in a rural Tanzanian population. AIDS. 1997May11;11(6):801.
Jha P. Avoidable global cancer deaths and total deaths from smoking. Nat Rev Cancer. 2009 Aug.20;9(9):655–64). Moreover, there are gender differences, which are shown below.
|Smoking prevalence is rising in low and middle income countries and declining in high-income countries||Smoking increases critical income and has no statistical impact on maximum survival|
|Smoking prevalence is higher in men than women||Smoking has a greater impact on critical income for adult males compared to adult females|
|HIV prevalence is highest in low and middle income countries||HIV prevalence increases critical income and has no statistical impact on maximum survival|
|HIV prevalence impacts men and women equally||HIV prevalence decreases critical income values in both men and women.|
We see future extension of the MM testing alternative interventions, and these will be subject to better cross sectional data for countries as well as better information not only on overall survival but death rates from specific conditions (say cancers, heart disease, etc).
b. There have been dramatic changes in the three leading risk factors for the global disease burden over the last 20 years. Currently these are high blood pressure, tobacco smoke and alcohol use compared to 1990, when the top 3 risk factors were childhood underweight, household air pollution from solid fuel use and tobacco smoke. This has been very well described in a recent analysis in the Lancet. Income alone does not adequately explain these changes – rather it is a reflection of a range of complex interactions of which GDP and family income are two factors. It is unclear to us how this analysis deals with the complex web of interactions that impact on life expectancy or the burden of disease and our read of the manuscript did not leave us with a clear indication of how this enzyme mathematical model adds value as a new approach to the problem if it does not deal with the multitude of variables impacting life expectancy.
Response: While we believe our model could be amended to include other risk factors, many of the existing databases with time series (such as fasting plasma glucose level, blood pressure, body-mass index, and cholesterol), are relatively new and all explicitly use income as a covariate to create their estimates. Moreover, the data available are only of all cause deaths. Cause-specific time series data for children or adults are not yet available to enable the kind of analyses proposed by the reviewer. We welcome further efforts to “count the dead”, given for example only 3% of all child deaths worldwide are certified according to cause of death (Liu et al. 2012).
Danaei et al., 2011. National, regional, and global trends in fasting plasma glucose and diabetes prevalence since 1980: systematic analysis of health examination surveys and epidemiological studies with 370 country-years and 2·7 million participants. The Lancet - 2 July 2011 (Vol. 378, Issue 9785, Pages 31–40)
Danaei et al., 2011. National, regional, and global trends in systolic blood pressure since 1980: systematic analysis of health examination surveys and epidemiological studies with 786 country-years and 5·4 million participants. The Lancet - 12 February 2011 (Vol. 377, Issue 9765, Pages 568–577)
Finucane et al. 2011. National, regional, and global trends in body-mass index since 1980: systematic analysis of health examination surveys and epidemiological studies with 960 country-years and 9·1 million participants. The Lancet - 12 February 2011 (Vol. 377, Issue 9765, Pages 557–567)
Farzadfar et al. 2011. National, regional, and global trends in serum total cholesterol since 1980: systematic analysis of health examination surveys and epidemiological studies with 321 country-years and 3·0 million participants. The Lancet - 12 February 2011 (Vol. 377, Issue 9765, Pages 578–586)
Liu et al. 2012. Global, regional, and national causes of child mortality: an updated systematic analysis for 2010 with time trends since 2000. The Lancet - 9 June 2012 (Vol. 379, Issue 9832, Pages 2151–2161)https://doi.org/10.7554/eLife.00051.015
- Ryan J Hum
- Ryan J Hum
- Yu-Ling Cheng
- Anita M McGahan
- Prabhat Jha
- Ryan J Hum
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank Beverly Bradley for comments.
- Eduardo Franco, McGill University, Canada
- Received: July 10, 2012
- Accepted: September 25, 2012
- Version of Record published: December 13, 2012 (version 1)
© 2012, Hum et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
Case fatality rates in severe falciparum malaria depend on the pattern and degree of vital organ dysfunction. Recent large-scale case-control analyses of pooled severe malaria data reported that glucose-6-phosphate dehydrogenase deficiency (G6PDd) was protective against cerebral malaria but increased the risk of severe malarial anaemia. A novel formulation of the balancing selection hypothesis was proposed as an explanation for these findings, whereby the selective advantage is driven by the competing risks of death from cerebral malaria and death from severe malarial anaemia. We re-analysed these claims using causal diagrams and showed that they are subject to collider bias. A simulation based sensitivity analysis, varying the strength of the known effect of G6PDd on anaemia, showed that this bias is sufficient to explain all of the observed association. Future genetic epidemiology studies in severe malaria would benefit from the use of causal reasoning.
Poor diet and lifestyle exposures are implicated in substantial global increases in non-communicable disease burden in low-income, remote, and Indigenous communities. This observational study investigated the contribution of the fecal microbiome to influence host physiology in two Indigenous communities in the Torres Strait Islands: Mer, a remote island where a traditional diet predominates, and Waiben a more accessible island with greater access to takeaway food and alcohol. Counterintuitively, disease markers were more pronounced in Mer residents. However, island-specific differences in disease risk were explained, in part, by microbiome traits. The absence of Alistipes onderdonkii, for example, significantly (p=0.014) moderated island-specific patterns of systolic blood pressure in multivariate-adjusted models. We also report mediatory relationships between traits of the fecal metagenome, disease markers, and risk exposures. Understanding how intestinal microbiome traits influence response to disease risk exposures is critical for the development of strategies that mitigate the growing burden of cardiometabolic disease in these communities.