Methods: De-identified data on patient visits between 2009 and 2014 were exported from the EMR data warehouse of an urban tertiary-care hospital. CDI presence was defined by a positive toxin assay or an assigned ICD9 diagnosis code. The MatchIt R package was used to perform nonparametric case-control matching (1:2 ratio) between CDI patients and a weighted cohort of comparable non-CDI patients via logistic regression of age, gender, diagnosis codes, and surgery procedures against CDI status; this simulates a dataset from a randomized experiment suitable for parametric modeling. A Cox proportional-hazards model was then used to estimate the effect of CDI on length of stay, which is a proxy for attributable cost, since the marginal cost of one inpatient-day in a given hospital unit is relatively stable.
Results: On a set of 30,000 matched EMR records, we found an attributable median increase of 10 inpatient-days per CDI patient, higher than the national average of 3.3 (Zimlichman et al. JAMA Intern Med. 2013). One limitation is that the model does not correct for reverse causation (longer stays increasing CDI risk).
Conclusion: Despite the lack of randomization inherent to EMR data, matching strategies can help compare a CDI patient cohort against control patients with equal modeled propensity for CDI, allowing estimation of cost attributable to CDI within a single hospital. This estimate may justify the scope of spending for infection control measures specifically targeting CDI.
T. O'donnell, None
S. Huprikar, None
H. Van Bakel, None
A. Kasarskis, None
E. Scott, None
See more of: Poster Abstract Session