Issues With Big Data: Variability in Reported Demographics and Complications Associated With Posterior Spinal Fusion in Pediatric Patients.

Background: Clinical and administrative registries provide large volumes of data that can be used for clinical research. However, there are several limitations relating to the quality, consistency, and generalizability of big data. In this study, we aim to compare reported demographics and certain outcomes in patients undergoing posterior spinal fusion (PSF) for adolescent idiopathic scoliosis (AIS), neuromuscular scoliosis (NS), and Scheuermann kyphosis (SK) between 3 commonly utilized databases in pediatric orthopaedic research.

Methods: We used International Classification of Diseases, Ninth Revision (ICD-9), International Classification of Diseases, 10th Revision (ICD-10), and Current Procedural Terminology (CPT) codes to identify patients in the National Surgical Quality Improvement Program (NSQIP), Healthcare Cost and Utilization Project (HCUP), and Pediatric Health Information System (PHIS) between the ages of 10 to 18 that underwent PSF for AIS, SK, and NS from 2012 to 2015. We compared various demographic factors, such as sex, race/ethnicity, age, and rates of postsurgical infection and 30-day readmissions. Data was analyzed with descriptive and univariate statistics.

Results: We identified 9891 patients that underwent PSF in NSQIP, 10,771 patients in PHIS, and 4335 patients in HCUP over the study period. There were significant differences in patient demographics, readmission rates, and infection rates between all patients that underwent PSF across the databases (P<0.01), as well as specifically in patients with AIS (P<0.01). HCUP had the highest proportion of Hispanic patients that underwent PSF (13.5%), as well as patients who had AIS (13.3%) or NS (17.9%). The PHIS database had the highest proportion of patients undergoing PSF for SK. Among patients with NS, there were significant differences in race across the databases (P<0.01), but no significant differences in sex, ethnicity, or readmission (P>0.05). In addition, there were significant differences in race (P=0.04) and readmission (P=0.01) across databases for patients with SK, but no differences in sex or ethnicity (P>0.05). NSQIP reported the highest rate of 30-day readmissions for patients undergoing PSF (17.9%) compared with other databases (HCUP 4.1%, PHIS 12.1%).

Conclusions: There are significant differences in patient demographics, sample sizes, and rates of complications for pediatric patients undergoing PSF across 3 commonly utilized US administrative databases. Given the variability in reported outcomes and demographics, generalizability is difficult to extrapolate from these large data sources. In addition, certain databases should be selected to appropriately power studies focusing on particular patient populations or outcomes.