Next, we used the effect sizes on mortality found from the literature review to estimate the number of lives potentially saved per year among the poorest billion by each evidence-based intervention.11x11Bukhman, G, Mocumbi, AO, and Horton, R. Reframing NCDs and injuries for the poorest billion: a Lancet Commission. Lancet. 2015;
Summary | Full Text | Full Text PDF | PubMed | Scopus (10) | Google ScholarSee all References We used a population attributable fraction (PAF) estimation and a Monte Carlo simulation technique, similar to previous research work measuring population-level interventions.15x15Pratt, M, Sarmiento, OL, Montes, F et al. The implications of megatrends in information and communication technology and transportation for changes in global physical activity. Lancet. 2012;
Summary | Full Text | Full Text PDF | PubMed | Scopus (119) | Google ScholarSee all References,16x16Lee, I-M, Shiroma, EJ, Lobelo, F et al. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet. 2012;
Summary | Full Text | Full Text PDF | PubMed | Scopus (1907) | Google ScholarSee all References
For the lives saved estimation-modelling exercise, we selected only interventions for which the outcome was mortality change. We excluded interventions for falls for institutionalised individuals because this issue mostly takes place in high-income countries, and “raising the minimum legal drinking age by 3 years” for road traffic injuries, because we only have data on the age of the casualty, not the driver, which prevents us from estimating the effect of this intervention. The eligible interventions were then subcategorised into conceptually related interventions, similar to what was done in DCP-3, to get ranges for their effect sizes.17x17Jamison, DT, Gelband, H, Horton, S et al. Disease control priorities: improving health and reducing poverty. in: DT Jamison, H Gelband, S Horton,
Disease control priorities. vol 9. 3rd edn. World Bank Group,
Washington DC; 2018
Google ScholarSee all References
For mortality estimations, we used data from the Global Burden of Disease 2015 (GBD 2015) study for total number of deaths, by age for both sexes, per country and injury cause (road injury, two-wheel and four-wheel vehicle road injury, and drowning).18x18GBD 2015 Mortality and Causes of Death Collaborators. Global, regional, and national life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death, 1980–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet. 2016;
Summary | Full Text | Full Text PDF | PubMed | Scopus (603) | Google ScholarSee all References Since the GBD 2015 study only provides estimates for the mean, minimum, and maximum values (instead of the mean and some type of dispersion metric, such as standard deviation), use of a normal distribution was not possible. The most appropriate distribution in this case was a triangular distribution, which allowed us to set a minimum, mean, and maximum value for both the effect sizes and deaths. Estimates on the potential annual lives saved per country and by intervention are available in the appendixappendix. As a robustness check for this distributional assumption, we did sensitivity analyses using a uniform distribution. The results of these additional analyses are available on request from the corresponding author. Additionally, given that not all countries have the same level of baseline enforcement of interventions (specifically for road safety), we used data from the Global Status Report on Road Safety 2015 to account for baseline conditions of enforcement for road safety interventions.19x19WHO. Global status report on road safety 2015. World Health Organization,
Google ScholarSee all References
To estimate the PAF of the risk factor avoided through the evidence-based intervention, we matched the effect sizes found for each intervention from the literature review with the mortality data of the corresponding target population on the basis of country, age category, and cause of death, as shown in equation 1:
where RR is relative risk, and the estimated number of lives saved is a function of the number of deaths of each country (i), cause of death (k), and the mortality prevented by intervention (j).
Our PAF estimations are similar to those used by Lee and colleagues16x16Lee, I-M, Shiroma, EJ, Lobelo, F et al. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet. 2012;
Summary | Full Text | Full Text PDF | PubMed | Scopus (1907) | Google ScholarSee all References as a method to evaluate the PAF for the population as a whole by use of unadjusted RRs. Using Monte Carlo simulations with 10 000 iterations, we obtained uncertainty estimates by intervention and country.
Specifically, the intervention of compulsory personal flotation device wearing prevents drowning when boating. Statistics on drowning when boating are not systematically reported for every country; however, some evidence shows that boating is the activity before drowning in 16–30% of cases of drowning.20x20Royal Life Saving. National drowning report 2016. Royal Life Saving,
Floreat Forum WA; 2016https://lifesavingwa.com.au/news/community/national-drowning-report-2016. ()
Google ScholarSee all References,21x21Minnesota Department of Natural Resources. Boat & water safety statistics. Saint Paul, MN: Department of Natural Resources. http://www.dnr.state.mn.us/safety/boatwater/statistics.html. ()
Google ScholarSee all References Therefore, we conservatively estimated the effect of this intervention to be effective for 16% of all drowning deaths, as shown in equation 2:
For road traffic interventions, we included into equation 1 a variable indicating baseline enforcement for each risk factor that is affected by road safety interventions based on data from the Global Status Report on Road Safety 2015 (equation 3).19x19WHO. Global status report on road safety 2015. World Health Organization,
Google ScholarSee all References
For countries with no helmet laws, we made separate estimations of the effect of the enactment of the helmet law and the effect of helmet enforcement. We did not estimate the effect of enacting helmet laws in countries that already had one.
For all the estimations, we made several assumptions. First, we assumed independence of the interventions, which implies that each intervention is assessed independently of other concurrent interventions. This assumption is important because the effectiveness parameters come from studies that assess the independent effect of each intervention. In this study, we are not able to account for either the potential synergy (multiplicative effect) or any reductions in the marginal benefit obtained from a single intervention when combined with another. Second, we assumed that the interventions are based on national policies that affect the entire population at risk, and do not only focus on individuals living in poverty. Therefore, our study assumes that the probability of any intervention saving a life is the same for a person who is or is not in the poorest billion. This assumption is likely to be conservative since individuals belonging to the poorest billion are likely to have a higher-than-average exposure to risk factors for injuries—eg, speed reduction interventions reduce the risk of injury more for vulnerable people who live or work close to unsafe roads than for those living in areas where urban characteristics protect them from speeding vehicles; and swimming lessons would reduce the risk of drowning to a greater extent among poorer children who are more likely to live closer to bodies of water than those of a higher socioeconomic background. Third, we assumed that the effect sizes of the interventions found in the literature would be transferable to the countries targeted by the Lancet NCDI Poverty Commission if the interventions were actually implemented. Assuming an appropriate implementation of these interventions, we estimated the potential lives saved. We cannot be certain about whether the effectiveness of these interventions is systematically larger or smaller in low-income and middle-income countries than in high-income coun-tries. On the one hand, higher-income countries have more resources and an enabling context to implement these strategies in a more effective manner than low-income and middle-income countries. On the other hand, if the implementation is properly done, the marginal effect of the intervention is likely to be higher in low-income and middle-income countries than in high-income countries, since the proportion of the population at high risk is larger and other interventions might not be in place. For this reason, whether the full implementation of these interventions would have larger or smaller effectiveness parameters in low-income and middle-income countries than in high-income countries is unclear. Finally, we assumed that countries reporting unknown enforcement for drink driving (Benin, Maldives, and São Tomé and Príncipe), unknown helmet-use enforcement (Afghanistan, The Gambia, Guinea-Bissau, and Liberia), unknown seat-belt-use enforcement (Afghanistan, Benin, Bangladesh, Guinea-Bissau, Liberia, Mali, Niger, Somalia, and Vanuatu), and unknown enforcement for speed (Benin, Guinea, and Vanuatu) had inputted the global average baseline enforcement values from the Global Status Report on Road Safety 2015 (ie, on a scale of 1–10, 5 for speed, 5 for helmets, 4 for drink driving, and 6 for seat-belts).