Rate per 10,000 (95% CI) is neonatal deaths, days 0 through 27, per 10,000 live births, with an exact Poisson interval.
Excess per 10,000 is the rate difference: the row's own rate less the comparison group's rate. Both are per 10,000 live births, the same denominator used everywhere else on this page, so the subtraction can be read directly off the two rates above. A negative value means mortality in the row was lower than in the comparison group.
Excess deaths 2017–2023 is that rate difference multiplied by the live births in the row. Equivalently it is the deaths actually observed in the row, less the deaths that would have been expected there at the comparison group's rate, summed over the seven birth cohorts. The percentage beside it is the excess divided by the expected deaths, that is, how far above or below the comparison group's expectation the row sits. It equals the crude relative risk minus one, so a crude RR of 3.59 and a value of +259% above expected are the same statement in two forms.
Births per one extra death is the reciprocal of the rate difference. It answers: how many births occur in this setting for each extra neonatal death compared with the comparison group. In the trial literature this quantity is called the number needed to harm (NNH), the counterpart of the number needed to treat. That name is used in the source analysis and is kept here for cross-reference, but it is avoided as the headline label because these are observational data. Women who choose one setting differ systematically from women who choose another, and the difference shown is an association between groups, not a measured causal effect of the setting. Where the row's rate is lower than the comparison group's, the same quantity is labelled births per one fewer death. The panel headed What this means in practice restates the same figures three ways: per 10,000 births, as a percentage for a single pregnancy, and as a national count per year, because the same number reads as negligible or as substantial depending on which of the three you are shown.
Crude RR is unadjusted. Adjusted RR is from modified Poisson regression with robust variance, adjusted for maternal age, parity, maternal race and Hispanic origin, maternal education, trimester prenatal care began, completed week of gestation, birth weight, plurality, fetal presentation, previous cesarean delivery and birth cohort year, less any variable that defines the population. A dash (—) means no adjusted model was fitted for that particular contrast, which is not the same as NE.
NE means not estimable because fewer than three deaths occurred in that arm. Where that applies, the live births, deaths, rate and confidence interval are still shown, but no rate difference, number needed to harm or relative risk is reported.

Birth certificates record where a birth occurred, not where it was planned. A woman who plans a community birth, transfers in labor and delivers in hospital is counted in the hospital arm together with her outcome, which makes every excess shown here conservative.

The attendant item records professional credential, not model of care, and has no separate code for certified midwife; certified midwives are counted with other midwives alongside certified professional midwives and licensed midwives.

The physician-attended hospital group contains, by construction, the complicated deliveries, the antenatal referrals and the intrapartum transfers. Comparisons against it are confounded by indication. Nothing in this tool supports an inference about the performance of physicians or of midwives.

The figures in this tool were computed from the NCHS period/cohort linked birth–infant death public-use files for birth cohorts 2017 through 2023 (reference ①), parsed from the fixed-width record layout published in reference ②. For each cohort year the denominator is every live birth registered in that year and the numerator every death before 365 days of life among those births, assembled from the birth-year and following-year numerator files. Record weights were not applied, as NCHS directs for cohort files.

Annual resident live births reproduced the published NCHS totals exactly for all seven years, pooled 25,936,466 (reference ③). The analysis population is live births at 37 or more completed weeks in the four settings shown, after exclusion of births with a congenital anomaly recorded on the birth certificate: 22,708,247 births and 45,110 linked infant deaths.

Status. The data files are official and public. The analysis of them presented here is the authors' own, is not yet published, and has not been peer reviewed. Figures may change.