Date of Award
Open Access Thesis
Medical Doctor (MD)
Jonathan N. Grauer
“Big data” is increasingly being used for orthopaedic research. However, the acute and complex nature of orthopaedic trauma makes data collection and data analysis difficult. This thesis presents three different clinical studies, which together illustrate how databases may best be used to answer clinical questions in orthopaedic trauma. Specifically the studies aim to show (1) how different databases capture trauma populations, (2) how databases may be used for hypothesis discovery studies, and (3) how databases may be used for hypothesis testing studies.
(1) How do populations of femoral shaft fracture patients differ in three commonly used national databases, specifically in regards to age and preexisting comorbidities? (2) What risk factors are associated with delayed surgery after elderly hip fractures in a national cohort and subsequently an institutional cohort? (3) Does hospital resource utilization differ between subpopulations of patients in Medicare Diagnosis Related Group 536 (fractures of the hip and pelvis), despite equal Medicare hospital reimbursement?
(1) Patients with surgically managed femoral shaft fractures were identified in the Nationwide Inpatient Sample (NIS), National Surgical Quality Improvement Program (NSQIP) and National Trauma Data Bank (NTDB). The distributions of age and Charleston Comorbidity Index were compared between populations. (2) A retrospective cohort study was conducted of all elderly hip fracture patients receiving surgical management from 2011-2012 in the NTDB and from 2009-2015 at a single academic trauma center. Multivariate analysis was used to identify the independent effect of various risk factors on surgical timing. (3) Patients with hip fractures, non-operative pelvic fractures, acetabulum fractures, and operative pelvic fractures were identified in the 2011 – 2012 NTDB. Total inpatient length of stay, intensive care unit (ICU) stay, and ventilator time were compared across groups using multivariate analysis that controlled for patient and hospital factors.
(1) A predominantly older population with more preexisting comorbidities was found in NSQIP (age = 71.5, CCI = 4.9), while a substantially younger population with fewer preexisting comorbidities was fond in NTDB (age = 45.2, CCI = 2.1). Bimodal distributions in the NIS population indicate a more mixed population (age = 56.9, CCI = 3.2). Differences in age were all statistically significant (p < 0.001). (2) In the national cohort, mean time to surgery was 31.3 hours (standard deviation: 31.6 hours). The risk factors with largest association with delays were total arthroplasty surgery (coefficient, in hours [95% confidence interval]: 7.7 [6.1 – 9.3]) coagulopathy, including chronic anticoagulation (7.1 [6.1 – 8.0]), and congestive heart failure (6.9 [6.0 – 7.9]). In the institutional cohort, mean time to surgery was 32.4 hours (standard deviation: 29.0 hours). In this cohort, the only statistically significant risk factors associated with surgical timing were total arthroplasty surgery (24.5 [13.7 – 35.4]), transfer from outside hospital (22.1 [15.1 – 29.1]), warfarin anticoagulation (13.7 [8.5 – 18.8]), other anticoagulation (10.5 [2.4 – 18.5]), and preoperative hematocrit < 35% (5.5 [2.0 – 9.0]). (3) After controlling for patient and hospital factors, the difference in inpatient length of stay compared to hip fracture patients was -0.2 days (95% C.I.: -0.4 to -0.1 days; P = 0.001) for non-operative pelvis fractures, 1.7 days (95% C.I.: 1.4 to 1.9 days; P < 0.001) for acetabulum fractures, and 7.7 days (95% C.I.: 7.0 to 8.4 days; P < 0.001) for operative pelvic fractures. Similar differences were also noted for IVU stay and ventilator time.
(1) While these three national databases have been commonly used for orthopaedic trauma research, differences in the populations they contain are not always readily apparent. Care must be taken to fully understand these populations before performing or evaluating database research, as these differences clearly affect observed outcomes. (2) Of all risk factors identified, access to arthroplasty and management of chronic anticoagulation may be the most modifiable in order to reduce delayed hip fracture surgeries. Physician call coverage and algorithms for more rapid reversal of anticoagulation, namely warfarin anticoagulation, warrant further investigation.
(3) Because hospitals are reimbursed equally for these subgroups of Medicare DRG 536, those centers that care for a greater proportion of more-complex pelvic trauma will experience lower financial margins per trauma patient, limiting their potential for growth and investment compared with competing institutions that may not routinely see high-energy trauma.
Samuel, Andre Michael, "Using “big Data” To Study Orthopaedic Trauma Populations: Looking At Fractures From A Bird’s Eye View As Illustrated By Three Studies" (2016). Yale Medicine Thesis Digital Library. 2075.