Ethnic differences in life expectancy, mortality and cause of death in Colombia: analysis of census and death register data

Abstract

This thesis uses data from census and death registers to undertake a systematic investigation of mortality differentials among ethnic population groups in Colombia, comparing Indigenous, Afro-descendant, and White-Mestizo ethnic groups. The main findings of this project are summarised across three dimensions: assessment of quality of mortality data, population life expectancy, and causes of death. Additionally, the study applies machine learning methods to identify the ethnic identity of deceased people based on demographic and epidemiological information derived from mortality registers.

Social inequalities among Indigenous, Afro-descendants, and White-Mestizos in Colombia are well documented in research studies. However, there is little understanding of how those ethnic inequalities manifest in mortality statistics in terms of data collection biases and analysis. This thesis will address the aforementioned knowledge gap by analysing the ethnic differentials in mortality and causes of death data comparing the census data with mortality registration records. Four research questions are addressed in this study. (i) How do life expectancy and mortality rates vary among different ethnic groups by age and sex? (ii) To what extent under-registration of deaths influences the estimation of life expectancy and mortality? (iii) Which causes of death contribute most to the differences in life expectancy and mortality among different ethnic groups? (iv) How can we apply machine learning techniques on mortality register data to improve ethnic classification in Colombia?

The first analytical chapter of the thesis aims to address under-registration of deaths, and how they affect mortality estimations in Colombia, and particularly how they under-estimate mortality indicators of Indigenous and Afro-descendants groups. The results show that the deaths of Indigenous and Afro-descendants are registered at a considerably lower rate than those of White-Mestizos, and the level of under-registration is higher for children and residents in rural areas. The ethnic minority groups are less likely to be captured in mortality registration, attributed to various social, economic, and geographic factors such as residence in remote areas, cultural practices and inhibitions, lack of awareness and the absence of institutional systems in regions dominated by ethnic minorities.

In the second analysis, we applied statistical models to correct for mortality underestimation in the mortality registers, and revised the estimates of mortality outcomes including life expectancy and life inequality defined as the average years of life lost due to early deaths. These measurements were estimated by age, gender and ethnicity. According to Colombian vital statistics, White-Mestizos show higher mortality and lower life expectancy when compared to Indigenous and Afro-descendants. However, the results show a reversal trend when mortality data are corrected for under-registration. Life expectancy at birth for Indigenous is estimated at 66.7 years, 71.5 years for Afro-descendants, and 78.9 years for White-Mestizos. Overall, White-Mestizos live on average 12.2 years longer than Indigenous and 7.4 years longer than Afro-descendants. The higher mortality in ethnic minority groups is evident and consistent when estimated using census mortality data and corrected mortality records. Similarly, the dispersion of the age of death in Indigenous and Afro-descendants is higher than for White-Mestizos, whose age of death is narrowly concentrated in the older age groups. This suggests that life disparity within ethnic groups is higher when compared to White-Mestizos.

The third analytical chapter of the thesis identifies the main causes of death by ethnic group, age, and gender. The analysis demonstrated three different patterns based on causes of death by age, sex and ethnicity. The first is the White-Mestizo pattern in which the main causes of death are degenerative diseases such as cardiovascular diseases, cancer, and respiratory diseases. The age groups most affected by such causes are the older age groups, in which the majority of such deaths are concentrated. The second pattern is exhibited primarily by the Afro-descendants, and combines very high mortality of young men between ages 15 and 40 (principally by homicide) with an emerging ageing population with increased deaths attributed to degenerative diseases. Deaths of Afro-descendants present a bimodal distribution with high mortality at younger and older ages. The epidemiological patterns of the Indigenous group, on the other hand, present high mortality in newborns and children compared with death at other ages. Likewise, the Indigenous groups had the highest proportion of deaths in women of reproductive age due to pregnancy, childbirth,
and puerperium-related causes. The causes of death more common in this ethnic group are those related to the perinatal period, and nutritional deficiencies in the case of children and circulatory diseases, cancer, and respiratory diseases in death at older ages. This pattern suggests that women belonging to indigenous groups are more vulnerable in terms of adverse living conditions with poor access to essential maternal and infant care.

The final analytical part of the thesis applies machine learning techniques on mortality records to improve the classification of ethnic identities in Colombia. We proposed a random forest classification model which considers geographical, sociodemographic, and epidemiological information by ethnicity. The results show that random forest models predict the ethnicity of deceased individuals more accurately with probabilities over 80%. The model yielded AUC = 0.913 for Indigenous people, while for Afro-descendants and White-Mestizos AUC scores were 0.811 and 0.845 respectively, suggesting a high level of accuracy in terms of correctly identifying the ethnicity of deceased people.

This research has two key limitations. First, the 2018 Census did not account for more than four million people, of which at least 1.5 million were Afro-descendants. The census omission of Afro-descendant households could lead to potential underestimation of mortality outcomes and hence the differences between ethnic groups may be biased. Second, there is lack of adequate information of the deceased people. The census asks if reported deaths were registered, but does not collect information about the causes of death. We were able to identify the level of under-registration but we could not determine if the causes of death play a role in this phenomenon. It is also possible the internal conflicts in certain geographic regions within Colombia may have contributed to the under-registration of deaths in ethnic groups.

Three recommendations are drawn from the findings of this study. First, the ethnic minority groups would need to be empowered and included in social policies and local decision-making processes and governance structures. They should be adequately trained and informed of the relevance and benefits of vital statistics for local planning, with measures to overcome cultural sensitivities and other barriers including language and geographic isolation. Second, census and surveys need to include additional questions to probe details of possible circumstances underlying the causes of death, verbal autopsies, and deaths with no funerals (missing bodies). Finally, there is a dire need to consider health policy interventions to prevent mortality in women and children attributed to malnutrition and infectious diseases especially during the perinatal period, targeting the indigenous community. For Afro-descendants, policy efforts should focus on reducing homicide mortality in young populations.

More information

Identifiers

ORCID for John Loja Torres: orcid.org/0009-0001-1496-8384

ORCID for Sabu Padmadas: orcid.org/0000-0002-6538-9374

ORCID for Jason Hilton: orcid.org/0000-0001-9473-757X

ORCID for Peter RA Hinde: orcid.org/0000-0002-8909-9152

Catalogue record

Export record