Infant Deaths and Infant Mortality Rates

Limitations

Infant deaths (under 1 year) by year of death and county of residence; rates are per 1,000 live births. OASIS withholds rates on 1-4 deaths (NULL here). Race-group rates divide deaths by the infant's race (death record) by births by the mother's race (birth record); race series break before 2015; race-group deaths or births are NULL where over 20% of the place-year's lack race, rates with either (our mask). Sex rows have no births count. OASIS says 2010's rate is low from under-reporting. Age at death nests: all = neonatal + postneonatal and neonatal = early_neonatal + late_neonatal, so never add neonatal to its parts. Causes nest: each of the 4 groups contains the causes listed in it and all is every infant death, so never add causes together; the listed causes never add up to their group, and the groups leave 11%-19% of Georgia's infant deaths a year in none of them (105 of 817 in 2024: infections, respiratory and circulatory diseases, ill-defined causes other than SIDS and others). num_births is OASIS's births module's count for the place, year and group (all, and the seven race groups by the mother's race and ethnicity), the denominator of every rate (num_deaths x 1,000 / num_births reproduces each published rate), repeated on every cause and age row. A period rate: deaths and births of the same calendar year, so a death in January may be of a baby born the year before. Male and female rates are OASIS's, per 1,000 births of that sex, but OASIS publishes no births by sex (its births module has no sex breakdown), so num_births is NULL there and male or female rates cannot be pooled across places. race_unknown is our calculation, all minus the seven race groups (unknown ethnicity, or non-Hispanic of unknown race): deaths only, no births or rate. Deaths carry the infant's race and ethnicity from the death certificate (OASIS's infant deaths equal its mortality module's deaths under age 1 in every race, ethnicity and sex cell, checked every year), while births, the denominator, carry the mother's (OASIS's known data issues: imputing the mother's race in birth data lowered the white infant mortality rate), so a race group's rate can be biased where the two differ, as for multiracial infants; 118 rows (2000-2025) have deaths but no births in the group (all with 1-4 deaths, so no rate). Race groups follow the platform convention from OASIS's race x ethnicity cross: hispanic is Hispanic of any race; white, black, asian, native_american (American Indian or Alaska Native), pacific_islander (Native Hawaiian or Other Pacific Islander) and multiracial are non-Hispanic. The smaller groups' series break in ways that look like recording changes (our reading): native_american deaths were 4-13 a year in 2000-2005 on about 200 births (a rate of 69.9 in 2004), and at most 2 since 2010, the break the mortality module shows; births to multiracial mothers were 237-648 a year in 2000-2006, then 3,538 in 2007 and 6,961 in 2008, while births to white mothers fell from 66,940 to 57,568 (probably the revised birth certificate Georgia introduced in mid-2007, which OASIS links to more births of unknown race; our reading); and 42-64 infant deaths a year in 2000-2006 are race_unknown, against at most 28 since. OASIS also cautions that Hispanic ethnicity in birth data from 2008 may be misreported. Don't read the small groups' changes before 2015 as trends. Counties break harder and for longer: in 84 county-years of 2000-2012 (none since; 17 counties, most often Dade, Catoosa and Walker on the Tennessee border, 12 years each, and Tift and nearby counties from 2007) more than 25% of births have no recorded race or ethnicity of the mother, up to 78%. Those births drop out of every race group's num_births while the infant's death record keeps its ethnicity, so those counties' race-group rates run far high: Tift's black rate in 2008 is 263.2 (5 deaths on 19 births to non-Hispanic black mothers, when 70% of its births had no recorded race or ethnicity), and Catoosa's white rate in 2009 is 31.4 (6 deaths, 191 births), as OASIS publishes them; those county-years' race-group births and rates are NULL here, their deaths served (our mask, below). Compare a county's race-group rates over time from 2013 only (its share of births without race or ethnicity has been at most 17% since). The rate counts two recorded events, each with its own race, so our mask (OASIS's own 20% threshold for missing items) measures the unknown share on each, on the place-year's all-causes, all-ages rows: where race or ethnicity is unknown for more than 20% of a place-year's infant deaths (the infant's), every race group's deaths for that place and year, every cause and age, are NULL; where the mother's is unknown for more than 20% of its live births (the county-years the births topic masks), every race group's births are; and the rate is NULL where either count is. Each group's count would miss its share of the unknowns, so its rate follows the recording. A place-year with no infant deaths has no deaths' share: its 0 deaths stay. The all, male, female and race_unknown rows are kept. That covers 182 place-years (2000-2018, 57 counties, 4 of them from 2013; never Georgia): 109 on the deaths' share (often a county's few deaths; 85 of them keep their births) and 97 on the births' (73 keep their deaths). OASIS's known data issues say 2010's infant mortality rate (6.3) is low from under-reporting and should be about 6.5: Georgia's infant deaths were 1,060 in 2009, 836 in 2010 and 904 in 2011. Rates on few births swing widely: the highest county infant mortality rate (all groups) is 58.3 (county 13065, 2015: 6 deaths, 103 births), and the highest county rate of any row is 333.3 (county 13135, 2004, native_american: 7 deaths, 21 births). 86% of rows with at least one death have no rate (1-4 deaths in small counties, groups and causes, or a race group's births our mask NULLs). OASIS publishes every count, however small, and withholds rates on 1-4 deaths (NULL here); the platform adds no small-cell suppression of its own (its race mask is a quality mask) and never recomputes a withheld rate. A place and group with no live births (num_births 0) has no rate (NULL): OASIS shows 0 there, but a rate per 1,000 births with no births is undefined and would rank as the lowest (our mask; deaths stay as published). Deaths are placed by county of residence (not where the death occurred). Counties add up to Georgia, in deaths and births, and the race groups (with race_unknown) and the sexes to all in deaths, except a race group's in the place-years our race mask NULLs it: there a race group's counties fall short of Georgia's row. Not served: OASIS's SES-vulnerability rows (in this module its SES rates divide by the SES group's own births, but the platform's demographics have no SES category yet), and the percent measures; this module has no payer list. 2025 is the newest year. tract_geoid is always NULL.