Hello. My name is Mark Johnson, and I'm a neurosurgery resident at the University of Cincinnati. I would like to thank the Scientific Program Committee for the invitation to present our work on validating the Surviving Penetrating Injury to the Brain, or SPIN, score. Hello. I have nothing to disclose. Developed in 2016, the SPIN score is a logistic regression-based model that utilizes clinical variables to estimate in-hospital and six-month mortality following cranial gunshot wounds. The components of the SPIN score include motor GCS as a categorical variable, pupillary function, self-inflicted injury motive, transfer status, gender, injury severity score, and initial INR. As we can see in the figure in the middle, the SPIN model had an excellent discrimination with an area under the curve of 0.96, where model discrimination refers to the ability of a model to separate individuals who will develop an event of interest, mortality in this case, from those who will not. Importantly, as we see in this receiver-operator curve, motor GCS and pupillary function were the dominant predictors of mortality in this model, with an area under the curve of 0.93. Last year, the SPIN score was validated in the multicenter series of 257 patients, again finding excellent model discrimination, as seen in figure A on the right side of the screen. An interesting observation from this study is that of the 362 eligible patients, 105 were missing an INR in admission. Excluding INR resulted in a marginal decrease in model discrimination, shown on the left in panel B. Given the need for a generalizable tool for mortality prediction in cranial gunshot wounds, we set out to validate the SPIN score's performance in our institutional data set. Additionally, we attempted to distill the SPIN score to its core components without sacrificing model discrimination. In order to accomplish this goal, we retrospectively analyzed data from our trauma registry over a six-year period, applying the initial inclusion and exclusion criteria from the derivation model. 108 patients were eligible for the validation, with seven missing an initial INR. We first calculated the area under the curve for the full SPIN model in the 101 eligible patients. We then performed multivariable logistic regression to identify SPIN components associated with in-hospital mortality. Starting with the SPIN component, and it categorized our continuous format that was both significantly associated with in-hospital mortality and had the highest area under the curve, we sequentially added the next SPIN component with the next highest area under the curve and tested whether this significantly improved the overall area under the curve. Addition of components was continued if significant improvement was observed. Here we have patient demographics from all three studies. There was relative uniformity in age, gender, total and motor GCS, injury severity score, pupillary findings, and incidence of withdrawal of care. Self-inflicted injuries were twice as common in our series, possibly explaining the slightly higher mortality rates observed. Here we see a receiver-operator curve for a full SPIN model denoted by the gray line with Xs with an area under the curve of 0.96. Given this excellent discrimination, we were able to validate the SPIN model in predicting inpatient mortality in our cohort. Superimposed on this graph are receiver-operator curves for motor GCS as a categorical and continuous variable alone. Motor GCS as a categorical variable denoted by the dashed line had a significantly lower model discrimination than the full SPIN model. However, when motor GCS was treated as a continuous variable, the solid black line, discrimination did not differ significantly from the full SPIN model. Here we show additional permutations of our attempt to improve model discrimination. However, no additional variable added to continuous motor GCS alone improved model discrimination significantly. In conclusion, we were able to successfully validate the SPIN score for inpatient mortality. However, discrimination did not differ significantly from continuous motor GCS alone. Given these findings, we propose the continuous six-point motor GCS may be sufficient as a generalizable predictor of inpatient mortality with reduced bias due to missing data. Following submission to this meeting, we have additionally validated this finding for six-month mortality. Hello. I would like to thank the employees of the UC Medical Center Trauma Registry and Center for Clinical and Translational Sciences, as well as the members of the Crane Collaborative for their support, and thank you again for the opportunity to share our work.

Mark Johnson

neurosurgery resident

University of Cincinnati

Surviving Penetrating Injury to the Brain (SPIN) score

cranial gunshot wounds