Predictors of Hospital Readmission and Surgical Site Infection in the United States, Denmark, and Japan: Is Risk Stratification a Universal Language?

Study Design: Retrospective review of three spine surgery databases.
Objectives: The purpose of the present study is to determine whether predictors of hospital readmission and surgical site infection (SSI) after lumbar fusion will be the same in United States, Denmark, and Japan.
Summary of Background Data: Because clinical decision making becomes more data driven, risk stratification will be crucial to minimize complications. Spine surgeons worldwide face this issue, leading to parallel efforts to address risk stratification. This raises the question as to whether pooled data would be valuable and whether models generated in one country would be applicable to other populations.
Methods: Predictors of SSI and 30-day readmission from three prospective databases (National Neurosurgery Quality and Outcomes Database [N2QOD] N = 2653, DaneSpine N = 1993, Japan Multicenter Spine Database [JAMSD] N = 3798) were determined and compared to identify common or divergent predictive risks.
Results: Predictive variables differed in the three databases, for both readmission and SSI. Factors predictive for hospital readmission were American Society of Anesthesiologists (ASA) grade in N2QOD (P = 0.013, odds ratio [OR] 2.08), fusion levels in DaneSpine (P = 0.005, OR 1.67), and sex in JAMSD (P = 0.001, OR = 2.81). Associated differences in demographics and procedural factors included mean ASA grade (N2QOD = 2.45, JAMSD = 1.72) and fusion levels (N2QOD = 1.39, DaneSpine = 1.52, JAMSD = 1.34). For SSI, sex (P = 0.000, OR = 3.30), diabetes (P = 0.000, OR = 2.90), and length of stay (P = 0.000, OR = 1.02) were predictive in JAMSD. No predictors were identified in N2QOD or DaneSpine.
Conclusion: Predictors of SSI and hospital readmission differ in the United States, Denmark, and Japan, suggesting that risk stratification models may need to be population specific or adjusted. Some differences in measured parameters exist in the three databases analyzed; however, patient and procedure selection also appear to differ and may limit the ability to directly pool data from different regions. Therefore, risk stratification models developed in one country may not be directly applicable to other countries.
Level of Evidence: 2.