BACKGROUND: The efficacy trial of diaspirin cross-linked hemoglobin (DCLHb) in traumatic hemorrhagic shock demonstrated an unexpected mortality imbalance, prompting a three-step review to better understand the cause of this finding. METHODS: Patients were enrolled in this DCLHb hemorrhagic shock study using 28-day mortality as the primary endpoint. Mortality data were primarily analyzed using the TRISS method and a nonblinded clinical review, followed by an independent Pennsylvania Trauma Outcome Study (PTOS)-derived probability of survival analyses. Finally, a trauma expert conducted a blinded clinical review of cases incorrectly predicted by these PTOS analyses. RESULTS: More of the DCLHb patients predicted to survive using TRISS actually died than in the control subgroup (24% vs. 3%, p < 0.002). Nonblinded clinical review noted that 72% of the patients who died had prior traumatic arrest, a presenting Glasgow Coma Scale score of 3, or a base deficit > 15 mEq/L. DCLHb patients predicted to survive using PTOS also more often died than did control patients (30% vs. 8%, p < 0.04). Blinded clinical review determined that 94% of the deaths were clinically justified. Both the TRISS and the PTOS models gave an adjusted mortality relative risk of 2.3, similar to the unadjusted risk data. CONCLUSION: Mortality analysis in this shock study involved both clinical case reviews and mortality prediction models. Despite the observation that nearly all of the deaths were clinically justified, the TRISS and PTOS models demonstrated excess unpredicted deaths in the DCLHb subgroup. A combined process, using both mortality prediction models and clinical case reviews, is useful in trauma studies that use a mortality endpoint.
BACKGROUND: The efficacy trial of diaspirin cross-linked hemoglobin (DCLHb) in traumatic hemorrhagic shock demonstrated an unexpected mortality imbalance, prompting a three-step review to better understand the cause of this finding. METHODS:Patients were enrolled in this DCLHb hemorrhagic shock study using 28-day mortality as the primary endpoint. Mortality data were primarily analyzed using the TRISS method and a nonblinded clinical review, followed by an independent Pennsylvania Trauma Outcome Study (PTOS)-derived probability of survival analyses. Finally, a trauma expert conducted a blinded clinical review of cases incorrectly predicted by these PTOS analyses. RESULTS: More of the DCLHb patients predicted to survive using TRISS actually died than in the control subgroup (24% vs. 3%, p < 0.002). Nonblinded clinical review noted that 72% of the patients who died had prior traumatic arrest, a presenting Glasgow Coma Scale score of 3, or a base deficit > 15 mEq/L. DCLHb patients predicted to survive using PTOS also more often died than did control patients (30% vs. 8%, p < 0.04). Blinded clinical review determined that 94% of the deaths were clinically justified. Both the TRISS and the PTOS models gave an adjusted mortality relative risk of 2.3, similar to the unadjusted risk data. CONCLUSION: Mortality analysis in this shock study involved both clinical case reviews and mortality prediction models. Despite the observation that nearly all of the deaths were clinically justified, the TRISS and PTOS models demonstrated excess unpredicted deaths in the DCLHb subgroup. A combined process, using both mortality prediction models and clinical case reviews, is useful in trauma studies that use a mortality endpoint.