Perioperative Risk Adjustment for Total Shoulder Arthroplasty: Are Simple Clinically Driven Models Sufficient?

Background There is growing interest in value-based health care in the United States. Statistical analysis of large databases can inform us of the factors associated with and the probability of adverse events and unplanned readmissions that diminish quality and add expense. For example, increased operating time and high blood urea nitrogen (BUN) are associated with adverse events, whereas patients on antihypertensive medications were more likely to have an unplanned readmission. Many surgeons rely on their knowledge and intuition when assessing the risk of a procedure. Comparing clinically driven with statistically derived risk models of total shoulder arthroplasty(TSA) offers insight into potential gaps between common practice and evidence-based medicine.

Questions/Purposes (1) Does a statistically driven model better explain the variation in unplanned readmission within 30 days of discharge when compared with an a priori five-variable model selected based on expert orthopaedic surgeon opinion? (2) Does a statistically driven model better explain the variation in adverse events within 30 days of discharge when compared with an a priori five-variable model selected based on expert orthopaedic surgeon opinion?

Methods Current Procedural Terminology codes were used to identify 4030 individuals older than 17 years of age who had TSA in which osteoarthritis was the primary etiology. A logistic regression model for adverse event and unplanned readmission within 30 days was constructed using (1) five variables chosen a priori based on clinic expertise (age, American Society of Anesthesiologists classification ≥ 3, body mass index, smoking status, and diabetes mellitus); and (2) by entering all variables with p < 0.10 in bivariate analysis. We then excluded 870 patients (22%) based on preoperative factors felt to make large discretionary surgery unwise to focus our research on appropriate procedures. Infirm patients have more pressing needs than alleviation of shoulder pain and stiffness. Among the remaining 3160 patients, logistic regression models for adverse event and unplanned readmission within 30 days were constructed in a similar manner to the complete models. The five a priori risk factors used in each model based on clinical expertise were selected by consensus of an expert orthopaedic surgeon panel.

Results When patients unfit for discretionary surgery were excluded, the clinically driven model found no risk factors and accounted for 1.4% of the variation in unplanned readmission. In contrast, the statistically driven model explained 4.6% of the variation and found operating time (hours) (odds ratio [OR], 1.26; 95% confidence interval [CI], 1.04-1.53) and hypertension requiring medications (OR, 1.95; 95% CI, 1.01-3.76) were associated with unplanned readmission accounting for all other factors. However, neither the clinically driven model (pseudo R², 1.4%) nor statistically driven model (pseudo R², 4.6%) provided much explanatory power. When patients unfit for discretionary surgery were excluded, no factors in the clinically driven model were significant and the model accounted for 0.95% of the variation in adverse events. In the statistically driven model, age (OR, 1.03; 95% CI, 1.01-1.06), men (OR, 1.64; 95% CI, 1.05-2.57), operating time (hours) (OR, 1.27; 95% CI, 1.07-1.52), and high BUN (OR, 3.12; 95% CI, 1.35-7.21) were associated with adverse events when accounting for all other factors, explaining 3.3% of the variation. However, neither the clinically driven model (pseudo R², 0.95%) nor the statistically driven model (pseudo R², 3.3%) provided much explanatory power.

Conclusions The observation that a statistically derived risk model performs better than a clinically driven model affirms the value of research based on large databases, although the models derived need to be tested prospectively.

Clinical Relevance Clinicians can utilize our results to understand that clinician intuition may not always offer the best risk adjustment and that factors impacting TSA unplanned readmission and adverse events may be best derived from large data sets. However, because current analyses explain limited variation in outcomes, future studies might look to better define what factors drive the variation in unplanned readmission and adverse events.