Split any gap in cesarean rates into case mix and practice · states, the United States, 2016–2024 · observed rates, not predictions
C = observed cesarean rate. e = expected rate: the United States 2016–2024 cesarean rate of every cell applied to this population’s own mix of cells. SCR = C / e, the practice index: 1.00 means United States average practice for that case mix. Because C = e × SCR exactly, any difference between two populations splits exactly into a case-mix term and a practice term, with nothing left over.
Data. CDC WONDER, Natality 2016–2024 expanded, by mother’s state of legal residence (the natality public use file carries no geography). Hospital births to residents of the 50 states and the District of Columbia with a known delivery method. The unit is the delivery: a twin pregnancy is one record, taken at the delivery of the second-born, so it counts as a cesarean if that baby was delivered by cesarean; WONDER collapses set order at “3rd or higher”, so a quadruplet or higher-order set can contribute more than one record.
Cells. Robson groups 1 and 2, and 3 and 4, are combined, so that how labour began (spontaneous, induced, or cesarean before labour) counts as practice rather than case mix: 8 strata plus Unclassified. The full model crosses the strata with maternal age (6 bands), pre-pregnancy body mass index (7 bands including unknown) and race and Hispanic origin (Hispanic, non-Hispanic White, Black and Asian, and other or unknown): 1,680 cells. Reference rates are the United States cell rates for 2016–2024 combined, held fixed for every state and every year, so 2016 and 2024 are directly comparable.
Splitting the case-mix term. The case-mix term is divided among Robson stratum, age, BMI and race by Shapley averaging over raked distributions; “other” holds changes in how the factors occur together and in the unclassified share. The parts add exactly to the case-mix term. This split is shown for the full model only.
Suppression. CDC WONDER suppresses counts of one to nine and none is shown here. For the detailed cells, suppressed state cells were filled with values of one to nine that add up to the unsuppressed totals; this affects expected values only and moves a state’s practice index by less than 0.01. The step-by-step table uses pooled counts, all of which exceed nine.
What the practice index is not. It means cesareans beyond what this case mix predicts, not unnecessary cesareans. The model does not see hypertension, diabetes, fetal size, payer, hospital level or the medicolegal climate, and birth-certificate items (induction, trial of labour, presentation) are completed differently from state to state. Adjusting for race and Hispanic origin answers “is the gap explained by who delivers there?”, but it also treats as expected any excess that falls on one group: nationally, non-Hispanic Black women had 7.7% more cesareans than women of the same stratum, age and BMI. Compare the model with and without race before drawing conclusions.
Methods references. Kitagawa EM. Components of a difference between two rates. J Am Stat Assoc. 1955;50(272):1168–94. Spiegelhalter DJ. Funnel plots for comparing institutional performance. Stat Med. 2005;24(8):1185–202. Robson M, Murphy M, Byrne F. Int J Gynaecol Obstet. 2015;131(Suppl 1):S23–S27.
Suggested citation. Centers for Disease Control and Prevention, National Center for Health Statistics. National Vital Statistics System, Natality on CDC WONDER Online Database. Data are from the Natality Records 2016–2024. Accessed 28 September 2026.