A Cautionary Tale: The Use of Propensity Matching to Evaluate Hemorrhage-Related Trauma Mortality in the American College of Surgeons TQIP Database

Abstract

Background: Propensity-matched methods are increasingly being applied to the American College of Surgeons TQIP database to evaluate hemorrhage control interventions. We used variation in systolic blood pressure (SBP) to demonstrate flaws in this approach.

Study design: Patients were divided into groups based on initial SBP (iSBP) and SBP at 1 hour (2017 to 2019). Groups were defined as follows: iSBP 90 mmHg or less who decompensated to 60 mmHg or less (immediate decompensation [ID]), iSBP 90 mmHg or less who remained greater than 60 mmHg (stable hypotension [SH]), and iSBP greater than 90 mmHg who decompensated to 60 mmHg or less (delayed decompensation [DD]). Individuals with Head or Spine Abbreviated Injury Scale score 3 or greater were excluded. Propensity score was assigned using demographic and clinical variables. Outcomes of interest were in-hospital mortality, emergency department death, and overall length of stay.

Results: Propensity matching yielded 4,640 patients per group in analysis #1 (SH vs DD) and 5,250 patients per group in analysis #2 (SH vs ID). The DD and ID groups had 2-fold higher in-hospital mortality than the SH group (DD 30% vs 15%, p < 0.001; ID 41% vs 18%, p < 0.001). Emergency department death rate was 3 times higher in the DD group and 5 times higher in the ID group (p < 0.001), and length of stay was 4 days shorter in the DD group and 1 day shorter in the ID group (p < 0.001). Odds of death were 2.6 times higher for the DD vs SH group and 3.2 times higher for the ID vs SH group (p < 0.001).

Conclusions: Differences in mortality rate by SBP variation underscore the difficulty of identifying individuals with a similar degree of hemorrhagic shock using the American College of Surgeons TQIP database despite propensity matching. Large databases lack the detailed data needed to rigorously evaluate hemorrhage control interventions.