"Can I use older years of the KID that include State identifiers to do State-level estimates?"

AHRQ strongly advises researchers against using the KID to estimate State-specific statistics. Prior to 2012, State is available as a KID data element. However, these KID samples were not designed to yield a representative sample of hospitals at the State level. AHRQ recommends that researchers employ the SID for State-level estimates.

The KID is designed to represent hospitals and discharges nationally, including those outside the sampling frame. Systematic random sampling is used to select 10% of normal newborns born in the hospital and 80% of other newborns and pediatric cases from each frame hospital. Each KID sample is drawn from the sampling frame consisting of discharge data submitted by HCUP Partners—statewide data organizations that agree to participate in the KID. Data from non-Partner States are missing completely from the sampling frame, and data from Partner States are sometimes incomplete because of different State reporting requirements, different State restrictions, or other data omissions. As a result, the proportion of KID discharges in a stratum that are from a given State is unlikely to equal the State's actual proportion of discharges in that stratum. Consequently, the sample of KID discharges is unlikely to be representative of discharges in the State, and the KID sample weights will not be appropriate at the State level.

The level of this "misrepresentation" varies across the States in any given year of the KID, which further confounds State-to-State comparisons based on State-specific estimates from the KID. Moreover, for a given State the level of misrepresentation changes from year to year as States (and hospitals) enter and exit the sampling frame over time. This further confounds State-specific trends based on State-specific estimates from the KID.

Finally, because the KID was not designed to be representative at the State level, design-based estimates of standard errors are not possible, which severely hampers State-level inferences. In summary, KID State-level estimates would be very imprecise at best and biased at worst.