Identification of possible determinants of Life Expectancy using Classification and Regression Tree (CART) analysis

Results

Among world countries available in this study sample, 149 countries (75%) had a Life expectancy at birth greater than 65 years while 50 countries (25%) had a Life expectancy at birth less/equal to 65 years.

Table 1 presents the bivariate associations between country characteristics and Life expectancy at birth. Bonferroni adjustment of the p value for the evaluation of these multiple comparisons is p < .007.

Table 1 Country characteristics by Life expectancy at birth

	Life expectancy at birth greater than 65 years (N=149)		Life expectancy at birth less/equal to 65 years (N=50)
Country characteristics	Mean	standard deviation	Mean	standard deviation	F value	p value
GDP PER CAPITA (CURRENT US$)	19258.6	23322.0	2243.7	3600.3	25.7	<.0001
ACCESS TO ELECTRICITY (% OF POPULATION)	91.7	16.3	35.5	22.5	361.4	<.0001
HEALTH EXPENDITURE PER CAPITA (CURRENT US$)	1397.8	2040.8	115.2	152.7	18.8	<.0001
IMPROVED SANITATION FACILITIES  (% OF POPULATION WITH ACCESS)	86.4	16.9	31.6	18.2	351.2	<.0001
IMPROVED WATER SOURCE (% OF POPULATION WITH ACCESS)	94.6	7.7	69.6	14.7	217.8	<.0001
LABOR FORCE, FEMALE (% OF TOTAL LABOR FORCE)	40.0	9.2	44.0	8.0	7.4	0.0070
PM2.5 AIR POLLUTION, POPULATION EXPOSED TO LEVELS EXCEEDING WHO GUIDELINE VALUE (% OF TOTAL)	70.6	37.3	78.0	32.8	1.5	0.2134

Analysis of Variance for continuous variables yielded the following differences:

•  countries with Life expectancy at birth greater or equal than 65.65 years compared with countries with Life expectancy at birth less/equal to 65 years have a higher Gross domestic product (GDP), 7.425.7, p < .0001
• . and exhibited more health expenditure per capita, F=18.8, p < .0001
•  and showed a higher % of population with access to electricity, F=361.4, p < .0001
•  and were found to have more population with access to improved sanitation facilities, F=351.2, p < .0001
•   and a greater % of population with access to improved water source, F=217.8, p < .0001
• and reported a lower % of female labor force, F=7.4, p < .007
•   and finally didn’t present a statistically significantly difference in term of % of population exposed to PM2.5 levels of air pollution exceeding World Health Organization guideline value

Next, countries variables were included as possible contributors to a CART model evaluating Life expectancy at birth (Figure 1).


Figure 1 Classification tree.

Each pentagon represents a decision point. For the decision point, the predictor variable and cut point are presented. Final groups with high and low outcome probability are represented by rectangles in the figure and include outcome frequencies and percentages. Shaded rectangle represent subgroups with relatively high rates of countries with high Life expectancy at birth, nonshaded rectangles, relatively low rates.

The % of population with access to electricity was the first variable to separate the sample into two subgroups. Countries with a % of population with access to electricity greater or equal than 56.3  were more likely to have high Life expectancy at birth compared to countries not meeting this cutoff (95.3% vs. 12.2%).

Of the countries with % of population with access to electricity greater than 56.3 , a further subdivision was made again with the % of population with access to electricity. Countries who reported having a % of population with access to electricity greater or equal than 89.6 were more likely to have high Life expectancy at birth compared to countries not meeting this cutoff (99.2% vs. 78.6%).

After splitting twice the data based on % of population with access to electricity, the other variables don’t impact on Life expectancy at birth.

The model classified 92% of the sample correctly, 95% of countries with Life expectancy at birth greater or equal than 65 years (sensitivity) and 82% of countries with Life expectancy at birth lower than 65 years (specificity).