TY - JOUR
T1 - K-Means Clustering to Elucidate Vulnerable Subpopulations Among Medicare Patients Undergoing Total Joint Arthroplasty
AU - Ranti, Daniel
AU - Warburton, Andrew J.
AU - Hanss, Kaitlin
AU - Katz, Daniel
AU - Poeran, Jashvant
AU - Moucha, Calin
N1 - Publisher Copyright:
© 2020 Elsevier Inc.
PY - 2020/12
Y1 - 2020/12
N2 - Background: The role of preoperative laboratory values for risk stratification following joint arthroplasty is currently ambiguous. In order to improve upon existing risk stratification within joint arthroplasty, this study sought to define novel phenotypes of total hip or total knee arthroplasty patients based entirely on preoperative laboratory measures. These phenotypes (“clusters”) were compared to elucidate statistically and clinically significant differences in outcomes. Methods: A total of 134,252 patients were gathered from the National Surgical Quality Improvement Program database between 2005 and 2015. “K-means” with 3 clusters was applied using 9 preoperative laboratory values: sodium, blood urea nitrogen (BUN), creatinine, albumin, bilirubin, white blood cell count, hematocrit, platelet count, and international normalized ratio of prothrombin values (INR). Outcome measures included 30-day readmissions, severe adverse events, and discharge to nonhome. Results: Cluster 2 was characterized by elevated preoperative BUN, creatinine, and INR and demonstrated almost twice the rate of adverse events (3.52% vs 2.20% and 2.22%), 30-day readmissions (6.39% vs 3.31% and 3.71%), and discharge to nonhome (47.97% vs 30.50% and 35.85%). Cluster 3 was characterized by a slightly higher risk of discharge to nonhome than cluster 1 and was overwhelmingly female (79.5% female, 35.8% discharge to nonhome). Cluster 1 represents the lowest-risk subgroup, experiencing the lowest rates of readmissions, adverse events, and discharge to nonhome. Conclusion: Preoperative laboratory values, namely BUN, creatinine, and INR, are useful in identifying patients at risk of adverse outcomes. This analysis supports the existing surgical literature pushing for preoperative hydration as a targeted intervention to expedite recovery.
AB - Background: The role of preoperative laboratory values for risk stratification following joint arthroplasty is currently ambiguous. In order to improve upon existing risk stratification within joint arthroplasty, this study sought to define novel phenotypes of total hip or total knee arthroplasty patients based entirely on preoperative laboratory measures. These phenotypes (“clusters”) were compared to elucidate statistically and clinically significant differences in outcomes. Methods: A total of 134,252 patients were gathered from the National Surgical Quality Improvement Program database between 2005 and 2015. “K-means” with 3 clusters was applied using 9 preoperative laboratory values: sodium, blood urea nitrogen (BUN), creatinine, albumin, bilirubin, white blood cell count, hematocrit, platelet count, and international normalized ratio of prothrombin values (INR). Outcome measures included 30-day readmissions, severe adverse events, and discharge to nonhome. Results: Cluster 2 was characterized by elevated preoperative BUN, creatinine, and INR and demonstrated almost twice the rate of adverse events (3.52% vs 2.20% and 2.22%), 30-day readmissions (6.39% vs 3.31% and 3.71%), and discharge to nonhome (47.97% vs 30.50% and 35.85%). Cluster 3 was characterized by a slightly higher risk of discharge to nonhome than cluster 1 and was overwhelmingly female (79.5% female, 35.8% discharge to nonhome). Cluster 1 represents the lowest-risk subgroup, experiencing the lowest rates of readmissions, adverse events, and discharge to nonhome. Conclusion: Preoperative laboratory values, namely BUN, creatinine, and INR, are useful in identifying patients at risk of adverse outcomes. This analysis supports the existing surgical literature pushing for preoperative hydration as a targeted intervention to expedite recovery.
KW - AKI
KW - K-means clustering
KW - arthroplasty
KW - dehydration
KW - readmissions
KW - unsupervised machine learning
UR - http://www.scopus.com/inward/record.url?scp=85088931955&partnerID=8YFLogxK
U2 - 10.1016/j.arth.2020.06.063
DO - 10.1016/j.arth.2020.06.063
M3 - Article
C2 - 32739081
AN - SCOPUS:85088931955
SN - 0883-5403
VL - 35
SP - 3488
EP - 3497
JO - Journal of Arthroplasty
JF - Journal of Arthroplasty
IS - 12
ER -