Should Your Newborn Spend An Extra Day in the Hospital?

With so much emphasis on efficiency, it is no wonder that cutting down on expensive hospital stays deemed unnecessary has been a staple of American health policy for the last 30 years. Figuring out whether (and when) hospital stays are unnecessary is difficult. Ideally, one would want to randomize patients to differing length of stays and compare outcomes across these different groups. After all, a negative correlation between hospital stays and outcomes (for example) in observational studies could simply be driven by the fact that sicker people spend more time in the hospital, rather than any negative causal effect of a longer stay. Needless to say, however, a randomized experiment like this is difficult to do and probably impossible to pass by an ethics board. As such, researchers are forced to find random sources of variation in hospital stays so as to get at causal estimates.

How does one go about recovering causal estimates then? One option is to look at the correlation between stays and outcomes conditioning on the initial health/diagnosis of the individual. One could argue that, when comparing individuals with the same health status or diagnosis, any differences in lengths of stay are essentially "random." However, even with all these controls, it is entirely possible that unmeasured attributes that are related to sickness and health also determine hospital stays.

In looking at the impact of an extra day in the hospital on newborns, Douglas Almond and Joseph Doyle, Jr, employ a different strategy. Here it is in their own words:

This paper compares the costs and benefits of extending the length of hospital stay following delivery using a discontinuity in stay length for infants born close to midnight. Third-party reimbursement rules in California entitle newborns to a minimum number of hospital "days," counted as the number of midnights in care. A newborn delivered at 12:05 a.m. will have an extra night of reimbursable care compared to an infant born minutes earlier. We use a dataset of all California births from 1991-2002, including nearly 100,000 births within 20 minutes of midnight, and find that children born just prior to midnight have significantly shorter lengths of stay than those born just after midnight, despite similar observable characteristics.

The authors find that an extra hospital day has no effect on infant and maternal mortality.

The strategy used in this paper is called "regression discontinuity" and involves using sharp cutoffs or thresholds in policies and comparing individuals that fall right around that cutoff. The main assumption that needs to hold for this to work is that those falling just around the threshold are essentially identical. In this paper, the main assumption is that newborns born at 12:03 are similar in observed and unobserved health attributes to infants born at 11:57. Where thresholds are respected and there is little scope for behavioral manipulation on the part of individuals being assigned to different policies, these assumptions are generally reasonable.

In such cases, this technique can be a powerful tool towards informing health policy. For more on regression discontinuity, see this excellent resource by Guido Imbens and Thomas Lemieux.