A Random Forest Model Using Flow Cytometry Data Identifies Pulmonary Infection after Thoracic Injury

BACKGROUND

Thoracic injury can cause impairment of lung function leading to respiratory complications such as pneumonia. There is increasing evidence that central memory T cells of the adaptive immune system play a key role in pulmonary immunity. We sought to explore whether assessment of cell phenotypes using flow cytometry (FCM) could be used to identify pulmonary infection after thoracic trauma.

METHODS

We prospectively studied trauma patients with thoracic injuries who survived >48 h at a Level 1 trauma center from 2014-2020. Clinical and FCM data from serum samples collected within 24 h of admission were considered as potential variables. Random forest (RF) and logistic regression (LR) models were developed to estimate the risk of hospital-acquired and ventilator-associated pneumonia (PNA). Variables were selected using backwards elimination and models were internally validated with leave-one-out.

RESULTS

Seventy patients with thoracic injuries were included (median age 35y (IQR: 25.25 - 51) 62.9% (44/70) male, 61.4% (42/70) blunt trauma). The most common injuries included rib fractures (52/70, 74.3%) and pulmonary contusions (26/70, 37%). The incidence of PNA was 14/70 (20%). Median ISS was similar for patients with and without PNA (30.5 (IQR 22.6 - 39.3) vs 26.5 (IQR 21.6 - 33.3). The final RF model selected three variables (APACHE score, highest pulse rate in first 24 h, and frequency of CD4+ central memory (CM) cells) that identified PNA with an AUC of 0.93, sensitivity of 0.91, and specificity of 0.88. A logistic regression with the same features had an AUC of 0.86, sensitivity of 0.76, and specificity of 0.85.

CONCLUSION

Clinical and FCM data have diagnostic utility in the early identification of patients at risk of nosocomial pneumonia following thoracic injury. Signs of physiologic stress and lower frequency of CM cells appear to be associated with higher rates of pneumonia after thoracic trauma.

LEVEL OF EVIDENCE

Diagnostic test/criteria, IV.

Multidimensional machine learning models predicting outcomes after trauma

BACKGROUND

An emerging body of literature supports the role of individualized prognostic tools to guide the management of patients after trauma. The aim of this study was to develop advanced modeling tools from multidimensional data sources, including immunological analytes and clinical and administrative data, to predict outcomes in trauma patients.

METHODS

This was a prospective study of trauma patients at Level 1 centers from 2015 to 2019. Clinical, flow cytometry, and serum cytokine data were collected within 48 hours of admission. Sparse logistic regression models were developed, jointly selecting predictors and estimating the risk of ventilator-associated pneumonia, acute kidney injury, complicated disposition (death, rehabilitation, or nursing facility), and return to the operating room. Model parameters (regularization controlling model sparsity) and performance estimation were obtained via nested leave-one-out cross-validation.

RESULTS

A total of 179 patients were included. The incidences of ventilator-associated pneumonia, acute kidney injury, complicated disposition, and return to the operating room were 17.7%, 28.8%, 22.5%, and 12.3%, respectively. Regarding extensive resource use, 30.7% of patients had prolonged intensive care unit stay, 73.2% had prolonged length of stay, and 23.5% had need for prolonged ventilatory support. The models were developed and cross-validated for ventilator-associated pneumonia, acute kidney injury, complicated dispositions, and return to the operating room, yielding predictive areas under the curve from 0.70 to 0.91. Each model derived its optimal predictive value by combining clinical, administrative, and immunological analyte data.

CONCLUSION

Clinical, immunological, and administrative data can be combined to predict post-traumatic outcomes and resource use. Multidimensional machine learning modeling can identify trauma patients with complicated clinical trajectories and high resource needs.

An integrative model using flow cytometry identifies nosocomial infection after trauma

BACKGROUND

Flow cytometry (FCM) is a rapid diagnostic tool for monitoring immune cell function. We sought to determine if assessment of cell phenotypes using standardized FCM could be used to identify nosocomial infection after trauma.

METHODS

Prospective study of trauma patients at a Level I center from 2014 to 2018. Clinical and FCM data were collected within 24 hours of admission. Random forest (RF) models were developed to estimate the risk of severe sepsis (SS), organ space infection (OSI), and ventilator-associated pneumonia (VAP). Variables were selected using backward elimination and models were validated with leave-one-out.

RESULTS

One hundred and thirty-eight patients were included (median age, 30 years [23-44 years]; median Injury Severity Score, 20 (14-29); 76% (105/138) Black; 60% (83/138) gunshots). The incidence of SS was 8.7% (12/138), OSI 16.7% (23/138), and VAP 18% (25/138). The final RF SS model resulted in five variables (RBCs transfused in first 24 hours; absolute counts of CD56- CD16+ lymphocytes, CD4+ T cells, and CD56 bright natural killer [NK] cells; percentage of CD16+ CD56+ NK cells) that identified SS with an AUC of 0.89, sensitivity of 0.98, and specificity of 0.78. The final RF OSI model resulted in four variables (RBC in first 24 hours, shock index, absolute CD16+ CD56+ NK cell counts, percentage of CD56 bright NK cells) that identified OSI with an AUC of 0.76, sensitivity of 0.68, and specificity of 0.82. The RF VAP model resulted in six variables (Sequential [Sepsis-related] Organ Failure Assessment score: Injury Severity Score; CD4- CD8- T cell counts; percentages of CD16- CD56- NK cells, CD16- CD56+ NK cells, and CD19+ B lymphocytes) that identified VAP with AUC of 0.86, sensitivity of 0.86, and specificity of 0.83.

CONCLUSIONS

Combined clinical and FCM data can assist with early identification of posttraumatic infections. The presence of NK cells supports the innate immune response that occurs during acute inflammation. Further research is needed to determine the functional role of these innate cell phenotypes and their value in predictive models immediately after injury.

LEVEL OF EVIDENCE

Prognostic, level III.

Predicting the need for massive transfusion: Prospective validation of a smartphone-based clinical decision support tool

BACKGROUND

Improper or delayed activation of a massive transfusion protocol may have consequences to individuals and institutions. We designed a complex predictive algorithm that was packaged within a smartphone application. We hypothesized it would accurately assess the need for massive transfusion protocol activation and assist clinicians in that decision.

METHODS

We prospectively enrolled patients at an urban, level I trauma center. The application recorded the surgeon's initial opinion for activation and then prompted inputs for the model. The application provided a prediction and recorded the surgeon's final decision on activation.

RESULTS

Three hundred and twenty-one patients were enrolled (83% male; 59% penetrating; median Injury Severity Score 9; mean base deficit -4.11). Of 36 massive transfusion protocol activations, 26 had an app prediction of "high" or "moderate" probability. Of these, 4 (15%) patients received <10 u blood as a result of early hemorrhage control. Two hundred and eighty-five patients did not have massive transfusion protocol activated by the surgeon with 27 (9%) patients having "moderate" or "high" likelihood predicted by the application. Twenty-four of these did not require massive transfusion, and all patients had acidosis that unrelated to hemorrhagic shock. For 13 (50%) of the patients with "high" probability, the surgeon correctly altered their initial decision based on this information. The algorithm demonstrated an adjusted accuracy of 0.96 (95% confidence interval [0.93-0.98); P ≤ .001]), sensitivity = 0.99, specificity 0.72, positive predictive value 0.96, negative predictive value 0.99, and area under the receiver operating curve = 0.86.

CONCLUSION

A smartphone-based clinical decision tools can aid surgeons in the decision to active massive transfusion protocol in real time, although it does not completely replace clinician judgment.