| A | B | C | D | E | F | G | |
|---|---|---|---|---|---|---|---|
1 | |||||||
2 | Summary: This document summarizes the distribution of all-cause mortality over time within the first 5 years of life (u5 mortality), including breakdowns of the infant mortality period (u1 mortality) and postneonatal mortality (months 2-12). It summarizes analysis done by Romero-Prieto et al. 2021, who calculated the probability of death during different periods of time from DHS data. This summary includes data from DHS surveys from LMIC (3 from South Asia and 15 from Africa) in the period 2013-2018. | ||||||
3 | Summary by Brett Keller, data analysis by Kameron Smith | ||||||
4 | Description | Link (if applicable) | |||||
5 | Romero-Prieto 2021, "Estimating the infant mortality rate from DHS birth histories in the presence of age heaping" | PLOS ONE | Romero-Prieto 2021 | |||||
6 | Appendix S2 of Romero-Prieto et al. 2021 graphically displays the relationship between age the cumulative probability of dying, calculated from DHS microdata using a synthetic cohort approach. Methods details are in the paper links. | Romero-Prieto 2021, S2 Appendix | |||||
7 | We requested and received the summary results presented in S2 appendix in tabular format on Nov 16, 2023. The raw data received is in the tab "raw data" and includes DHS surveys through 2018. We received permission from the authors to publish this data and our analysis of the data. | Link here for internal reference only. | |||||
8 | We restricted our analysis to surveys from 2013-2018, and only included Sub-Saharan Africa and South Asia (excluding the Maldives) regions. We subtracted the previous period's mortality to convert from a cumulative probability to a probability. If more than one survey was available in the same time period, we used only the latest one. The data were reshaped into a single, compact table in the tab 'observed p in period'. We used the observed values from the DHS microdata rather than the log-quadratic model developed by Romero-Prieto et al. because they said (in our email exchange) "Perhaps, the log-quadratic model is not providing the best representation of the age patterns of under-five mortality of some countries. Particularly, in Africa and South Asia. This point is discussed in the paper–and we have the same impression from another publication that we had in 2022." | ||||||
9 | We converted "qx: observed cumulative probability of dying (at the end of the age interval)." into the probability of dying in each time period. The data were reshaped into a single, compact table in the tab 'observed p in period'. | "Observed p in period" tab | |||||
10 | The Results tab includes the results in the original Romero-Prieto time periods, plus the distribution by month for the first year of life, distribution by month for the postneonatal period, and distribution by year for the first five years. | ||||||
11 | The numbers presented are as close to the raw data collected by DHS as possible: No corrections for age-heaping were made, in part for simplicity and in part because the Romero-Prieto et al. 2021 paper argues that age heaping is not a large concern. All numbers are for all-cause mortality, which is what is measured by the DHS: any breakdown by cause-specific mortality would require substantial data collection or modeling assumptions. | ||||||
12 | I recommend using all-country average: the observed distribution of mortality over time is similar enough across regions that I think it's generally better to use the average across all countries. This is partly for simplicity, and partly to avoid overfitting to noise from specific surveys. However, we included averages for South Asia and Africa separately in case a researcher decides they want to use a specific distribution. Country specific distributions are also available in the tab 'observed p in period', but I think these are more likely to be noisy. | ||||||
13 | |||||||
14 | |||||||
15 | |||||||
16 | Variables in original data set: | ||||||
17 | ctry | country | |||||
18 | DHS_code | DHS country code | |||||
19 | surv_year | survey data collection year | |||||
20 | Region | region | |||||
21 | Subregion | subregion | |||||
22 | age_d | age at the beginning of the age interval | |||||
23 | n_d | length of interval in days | |||||
24 | x (in days) | x: is age in days at the end of the age interval. x = age_d + n_d. | |||||
25 | qx (observed) | qx: observed cumulative probability of dying (at the end of the age interval). | |||||
26 | qx (predicted) | p.qx: predicted or model-based cumulative probability of dying (at the end of the age interval). | |||||
27 | |||||||