Trauma risk score matching for observational studies in orthopedic trauma dataset and code

The dataset presented was collected via retrospective review from an orthopedic trauma database approved by the institutional review board at the author's institution from patients treated at any of the four hospitals serviced by the academic orthopedic surgery department. Femoral neck and intertrochanteric hip fracture patients from low energy mechanisms admitted between October 2014 and February 2020, were selected if they were age 55 or older and had recorded sex, body mass index (BMI), Charlson Comorbidity Index (CCI), American Society of Anaesthesiologists (ASA) physical status classification, Glasgow Coma Score, Abbreviated Injury Severity score for the chest, head and neck, and extremities, and ambulation status prior to injury. The resultant 1,590 subject dataset may be analysed via the supplied R statistical code to determine the frequency of equipoise in baseline and outcome variables from propensity matching via three matching schemes. The code implements three matching schemes including matching by (1) The Score for Trauma Triage in Geriatric and Middle-Aged (STTGMA) (2) CCI alone, or (3) a combination of sex, age, CCI and BMI. The code selects a subset of ten percent of hip fracture patients by a pseudorandom number generator (PRNG). The code matches the remaining patients 1:1 to the selected patients by propensity score generated by logistic regression of STTGMA, CCI, or a combination of sex, age, CCI and BMI using greedy nearest neighbor matching without replacement by the MatchIt package for R software. The code then compares matched cohorts by Chi-square, Fisher, or Mann-Whitney U test with significance level of 0.05 representing a 5% chance of significant differences due to random sampling of subjects. The supplied code repeats the random selection, matching and testing process 100,000 times for each matching method. The resultant code output is the frequency of significantly different demographic or outcome parameters among matched cohorts by matching method. This data and statistical code have reuse potential to explore alternative matching schemes. The supplied baseline variables should be robust enough to derive alternative risk scores for each patient which may be included as a matching variable for comparison. The authors also look forward to unexpected ways that this data may be used by readers.