key: cord-0928764-o38tfjh7
authors: Strain, Tessa; Brage, Søren; Sharp, Stephen J; Richards, Justin; Tainio, Marko; Ding, Ding; Benichou, Jacques; Kelly, Paul
title: Use of the prevented fraction for the population to determine deaths averted by existing prevalence of physical activity: a descriptive study
date: 2020-06-17
journal: Lancet Glob Health
DOI: 10.1016/s2214-109x(20)30211-4
sha: d1f13fa1296f65490deb1d51d48344a1364ba8cd
doc_id: 928764
cord_uid: o38tfjh7

BACKGROUND: Disease and mortality burdens of unhealthy lifestyle behaviours are often reported. In contrast, the positive narrative around the burdens that an existing behaviour have averted is rarely acknowledged. We aimed to estimate the prevented fraction for the population (PFP) for premature mortality averted by physical activity on a global scale. METHODS: In this descriptive study, we obtained previously published data on physical activity prevalence (2001–16) and relative risks of all-cause mortality for 168 countries. We combined the data in Monte-Carlo simulations to estimate country-specific, mean PFP values, corresponding to percentage of mortality averted, and their 95% CIs. High prevented fractions indicated an increased proportion of deaths averted due to physical activity. Using mortality data for all people in a country aged 40–74 years, we estimated the number of premature deaths averted for all adults and by gender. We present the median and range of the prevented fractions globally, by WHO region, and by World Bank income classification. FINDINGS: The global median PFP was 15·0% (range 6·6–20·5), conservatively equating to 3·9 million (95% CI 2·5–5·6) premature deaths averted annually. The African region had the highest median prevented fraction (16·6% [range 12·1–20·5]) and the Americas had the lowest (13·1% [10·8–16·6]). Low-income countries tended to have higher prevented fractions (group median 17·9% [12·3–20·5]) than high-income countries (14·1% [6·6–17·8]). Globally, the median prevented fraction was higher for men (16·0% [7·8–20·7] than women (14·1% [5·0–20·4]). INTERPRETATION: Existing physical activity prevalence has contributed to averting premature mortality across all countries. PFP has utility as an advocacy tool to promote healthy lifestyle behaviours. By making the case of what has been achieved, the prevented fraction can show the value of current investment and services, which might be conducive to political support. FUNDING: UK Medical Research Council, British Heart Foundation, Cancer Research UK, Economic and Social Research Council, National Institute for Health Research, Wellcome Trust, Heart Foundation Australia.

Fraction for the Population using an example of physical activity. Figure 1A . The Prevented Fraction for the Population.

The observed disease cases (the dark grey circles) occur in both the active and the inactive, but the incidence is higher amongst the inactive. The number of cases that are averted by activity levels (the light grey circles) will be determined by the difference in incidence rates and by the prevalence of activity. The sum of the dark and light grey circles represents the hypothetical total number of cases that would be expected if the whole population was inactive. The Prevented Fraction for the Population expresses the number of averted cases (the light grey circles) as a proportion of the hypothetical total number of cases if the whole population were inactive (light + dark grey circles).

All circles in this figure represent observed disease cases. The black circles represent cases that occur in both the active and inactive (a 'baseline' level of disease incidence). The dark grey circles represent cases that occur in the inactive group, above this 'baseline' level. As such, they are considered attributable to inactivity levels. The Population Attributable Fraction and the Population Preventable Fraction express the attributable disease cases (dark grey circles) as a proportion of all observed disease cases (dark grey + black circles), i.e. the potential proportion of cases that could be eliminated if everyone was active.

Figures 1A and 1B adapted from Koepsell, Zatsick, & Rivara (2011) and Spasoff (1999) .

Koepsell TD, Zatzick DF, Rivara FP. Estimating the Population Impact of Preventive Interventions from Randomized Trials. Am J Prev Med 2011; 40: 191-8. Spasoff RA. Epidemiologic Methods for Health Policy. New York: Oxford University Press; 1999.

Our aim was to identify examples of use of the Prevented Fraction for the Population relating to lifestyle behaviours, such as physical activity, diet and nutrition, alcohol, and smoking. Because of the variety of terms used for the concept, we searched PubMed and WebofScience search engines for the terms "prevented fraction", "preventive fraction" and "preventable fraction". We performed this search on the 12 th March 2019 and updated it on 12 th June 2019 with no language restrictions.

We obtained the following number of hits Other non-behavioural 1 10 25

No title 0 0 10

We read the full texts of all those identified in the 'lifestyle behaviours' category to ascertain whether they used the Prevented Fraction for the Population as defined by the International Dictionary of Epidemiology and described in our paper. i.e. they were estimating the proportion of a disease burden that had been averted due to current levels of a risk factor, or that could have been averted due to a given level of a risk factor. These 10 papers are highlighted in grey in the table below along with two further articles, identified through further reading that met these criteria. The other papers generally estimated the proportion of a disease burden that could be averted due to future changes in a risk factor. They estimated the percentage of cancer reduction that could be prevented by combined oral contraception use. The report summarizes the method behind the Health Economic Assessment tool (HEAT) for walking and cycling that can be used to quantify the reduction in mortality risk of current levels of walking or cycling. *This paper uses the terminology of 'the proportion of injuries that have been prevented' due to a given level of cycling helmet use, but further inspection of the formulae presented and the text description confirmed it did not meet the definition of the Prevented Fraction for the Population described in the paper. **This paper is ambiguous in its language and formulae referred to.

We also selected a random sample of papers in the other categories, across the search terms used, to investigate whether the concept in question was being used. We identified examples across all categories e.g. Formulae 1 and 2, both presented in Box 1, are equivalent. Prevented Fraction for the Population is abbreviated to PFP in the following formulae.

Ip can be expressed as

Pu can be expressed as

And Ie can be expressed as

Substituting Formulae C and D into B, Ip can be expressed as

Substituting Formula E into Formula 1, the PFP can be defined as:

This can be rearranged to the following Formulae G and H, eventually simplifying to Formula 2:

There are two stages to the derivation of this formula: calculating Pd from Pe and then substituting that into Formula 2. Prevented Fraction for the Population is abbreviated to PFP in the following formulae.

The prevalence of activity amongst cases (Pd) is defined as the number of cases that were active over the total number of cases:

Divide numerator and denominator by Iu, cancel out n, replace Pu as 1-Pe

Rearrange this to solve for Pe

Now substitute into Formula 2: -Add in a non-zero counterfactual because of concerns that 0% of the population meeting the physical activity guidelines is not realistic. -To be conservative in the estimates of the benefits of physical activity, we considered a 10% level as a counterfactual. -We looked at using adjusted and unadjusted non-zero counterfactual formulae Version 3:

-We could not find justification for a 10% counterfactual, or a reasonable justification for any absolute level. -We therefore decided to use the minimum observed levels for a country, and term this the minimum plausible risk. -Guthold et al. (2018) published more up to date, harmonised, prevalence data for more nations so we changed our data source -We removed the coronary heart disease analyses to streamline the paper's results -We removed the unadjusted non-zero counterfactual results to streamline the paper -We decided to estimate the number of deaths corresponding to the estimated Prevented Fractions for the Population to aid interpretation Edits made after Review 1 from Lancet Global Health (Feb 2020):

-We added in sex-specific results.

-We confirmed that the mortality data from the United Nations was the most appropriate to use. Although there are other sources of non-communicable disease deaths, they are not by age group Appendix 6. Overview of method.

When estimating the 'unadjusted' Prevented Fraction for the Population, the prevalence distribution was the prevalence of activity (Pe).

We did not model any variance around the prevalence of activity (Pe) or prevalence of activity amongst cases (Pd) for the counterfactual scenarios.

Obtained data Generated distribution RR (95% CI) of inactivity obtained from Lee et al. Population (2015) for the 168 countries with physical activity data to be included in the present study:

7,022,019,600 (7.0 billion; 95.2% of the world population)

Population (2015) for the 14 countries for which there were no mortality data: 527,820 (<0.01% of total world population). We imputed mortality data for these countries by using the median ratio of deaths between the ages of 40-74 years to total population of countries in the same WHO region and income group. We multiplied this median value by the population of the country for which there was no mortality data. These 14 countries were: The Potential Impact Fraction formula relates closely to that of the Population Attributable Fraction:

Note: these two formulae define the exposure as harmful and unexposure as protective.

The Potential Impact Fraction is an extension of the Population Attributable Fraction formula that replaces the incidence rate in the unexposed with a counterfactual incidence rate.

The same can be done for the PFP.

The proportion of active in the counterfactual population is represented by Pe'. Formula A can be rewritten as:

This can be rearranged as:

Formulae F and B both compare the difference in disease burden between two scenarios, and express that as a proportion of the scenario with the higher disease burden. Therefore, they can be considered equivalent but differing in terms of exposure definition (harmful or protective).

To extend this formula to use Pd rather than Pe, substitute in a variant of Formula 4 (Pe defined in terms of Pd and RR):

(a variant of Formula 4, see Appendix 4, p9)

The numerator of formula G can be rearranged:

Next, multiply the numerator by the denominator:

This can be expanded out and cancelled to:

Multiply out to get both fractions on the same denominator

The numerator of formula K expands and cancels to:

The denominator of formula K can be partially expanded out to:

Together,

This cancels and simplifies to Number of deaths averted for males and females may not add up to the estimate for both sexes. This is because the distributions of deaths across males and females are not even, and differences are magnified when multiplied by a derivative of activity prevalence which also varies considerably by sex.

a: The main estimates calculated using Formula 3 (see Box 1), the adjusted relative risk, and 0% activity counterfactual.

b: The unadjusted estimates calculated using Formula 2 (see Box 1), the unadjusted relative risk, and 0% activity counterfactual.

c: The partially adjusted estimates calculated using Formula 2 (see Box 1), the adjusted relative risk, and 0% activity counterfactual.

d: The adjusted estimates calculated using Formula 5 (see Appendix 10, pp23-25), the adjusted relative risk, and 54.3% activity counterfactual.

e: Deaths averted based on an imputed estimate of total deaths between the ages of 40-79 years. This was derived using the median ratio of deaths to total population for countries in the same region/income categories (Appendix 9, p22).

Prevented Fractions for the Population (95% confidence intervals) 

This is because the distributions of deaths across males and females are not even, and differences are magnified when multiplied by a derivative of activity prevalence which also varies considerably by sex. a: The main estimates calculated using Formula 3 (see Box 1), the adjusted relative risk, and 0% activity counterfactual. b: The unadjusted estimates calculated using Formula 2 (see Box 1), the unadjusted relative risk, and 0% activity counterfactual. c: The partially adjusted estimates calculated using Formula 2 (see Box 1), the adjusted relative risk, and 0% activity counterfactual. d: The adjusted estimates calculated using Formula 5 (see Appendix 10, pp23-25), the adjusted relative risk