Trends in Comorbidities and Complications Among Patients Undergoing Hip Fracture Repair

Janis Bekeris, MD; Lauren A. Wilson, MPH; Dace Bekere, MD; Jiabin Liu, MD, PhD; Jashvant Poeran, MD, PhD; Nicole Zubizarreta, MPH; Megan Fiasconaro, MS; Stavros G. Memtsoudis, MD, PhD, MBA


Anesth Analg. 2021;132(2):475-484. 

In This Article

Materials and Methods

Study Design and Cohort

Given the observational nature of this retrospective study using deidentified data, the Institutional Review Board (IRB) of Hospital for Special Surgery (IRB #2012-050) waived the requirement for informed consent.

Patient information was derived from the Premier Healthcare database (Premier Healthcare Solutions, Inc, Charlotte, NC).[15] Individuals who underwent a hip fracture repair surgery from 2006 to 2016 were identified and classified into procedure groups according to International Classification of Diseases, Ninth Revision(ICD-9) procedure codes: internal fixation (78.55, 79.15, 79.25, 79.35, 79.55, 79.65, 79.85, 79.95), hemiarthroplasty (81.52), or total hip arthroplasty (THA) (00.7X, 00.85, 00.86, 00.87, 81.40, 81.51, 81.53). Patients were further classified based on ICD-9 diagnosis codes for the location of their fracture: femoral neck (820.0x, 820.1X, 820.8, 820.9), intertrochanteric (820.21, 820.31), subtrochanteric (820.22, 820.32), or multiple locations. Patients with missing sex information were excluded.

Study Variables

Variables of interest were patient and health care characteristics and complications. Patient characteristics of interest included age (continuous), sex, race (African American, Caucasian, or other), and insurance provider (commercial, Medicaid, Medicare, uninsured, or unknown). Health care characteristics analyzed included location (urban versus rural), size (<300 beds, 300–500 beds, >500 beds), and teaching status. Type of anesthesia was defined using billing descriptions as well as codes and was classified as general, neuraxial, or both. Fracture type and type of repair surgery were classified as mentioned above. All Elixhauser comorbidities[16] were evaluated as well as obstructive sleep apnea (ICD-9 diagnosis codes: 372.X, 780.51, 780.53, 780.57, and 786.03).

Complications analyzed were acute myocardial infarction, other cardiovascular complications, hemorrhage/hematoma, stroke, pulmonary complications, pulmonary embolism, venous thromboembolism, pneumonia, sepsis/septic shock, wound complications, acute renal failure, other genitourinary complications, delirium, inpatient falls, and 30-day mortality. All adverse events were defined based on ICD-9 diagnosis codes that were defined as not "present on admission" (Supplemental Digital Content 1, Appendix 1, Procedures requiring blood transfusions were defined according to ICD-9 procedure codes (99.0X), current procedural terminology (CPT) codes (within the range from P9010 to P9040), and/or relevant billing codes.

Statistical Analysis

Annual frequencies and percentages were reported for all patient and health care characteristics, with age reported as median and interquartile range (IQR). Complication rates were reported as the number of complications per 1000 inpatient days to adjust for variance in length of stay. The number of hospitals participating in the Premier database increased from 373 in 2006 to 612 in 2016.[15] Therefore, we elected to report annual incidence rather than exclusively reporting frequencies.

Trends in categorical outcomes over time were analyzed using Cochran–Armitage trend tests. Nonbinary categorical outcomes were analyzed as mutually exclusive variables. While traditionally this may not be an appropriate approach, for our purposes, it allowed us to identify instances where, for example, there may be significant trends in 2 levels of a categorical variable, but not the third. Linear regression analyses with year as the sole predictor were used to identify significant trends over time for continuous outcomes. To adjust for our analysis of trends across 76 different outcomes, after applying a Bonferroni correction to the overall α level of .05, results with a P value <.0006 were considered statistically significant; significance trends were also evaluated graphically. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC).