Impact of Big Data Analytics for Assessing a Common Orthopaedic Procedure: Commentary on an article by Kevin Bozic, MD, MBA, et al.: "Quality Measure Public Reporting Is Associated with Improved Outcomes Following Hip and Knee Replacement".

Abstract

Commentary The study by Bozic et al. is an important contribution to the field of orthopaedic surgery because it demonstrates the potential relationship between surgical outcomes and national projects that are aimed at improving overall patient care among the public. The study shows a reduction in readmission and complication rates for total knee arthroplasty and total hip arthroplasty from 2010 to 2017. I actively performed total joint replacement procedures in a small, rural hospital during the study period of the article, and was made acutely aware of the reimbursement hazards to hospitals if they could not perform to standards for readmission or rates of complications. I felt the same pressure to perform in that setting as I did when working in a very competitive urban medical center. The study by Bozic et al. demonstrated that the effect created by the Centers for Medicare & Medicaid Services (CMS) management was clearly in the right direction. Logistic regression is the principal statistical method used in the study. Regression analysis is a highly efficient method for statistical analysis because it relies solely on a binary outcome: the patient either was readmitted or had a complication or not. There is little chance for interpretation, and as a result, the outcomes of the study were as reliable as the financial data for the same study cohort. Although the patient information comes only from the CMS Medicare database, one questions whether the data were generalizable to the total population of patients undergoing total joint replacement. I would suggest that most—if not all—patients over 65 years old would belong to this patient group, and the group would categorically be of greater age with higher risks than younger patients. The authors advance a statistical measure, the standardized risk percentage difference, to assess changes that occur year to year in the risk variables1,2. This is done by dividing the percentage change (e.g., the increase in patients who experienced renal failure) by the standard deviation calculated for that variable from baseline data, which clearly makes an adjustment for sample size and statistical variance. For the renal failure complication factor, the absolute increase was 6.5% from 2010 to 2017, which was a relative increase of 85.3%, and the standardized percentage difference was 20.9%. Similarly, the risk factor of obesity increased by 18.6% in the standardized difference percentage calculation. This is an important finding and methodology, and one that we will see in future publications. The authors did not suggest or infer the possibility for causation of the outcome. There was concern by the authors that this study would limit access of groups of patients on the basis of risks or other factors that could adversely predict outcome. Surgeons may game the system to get better outcomes. Interestingly, most of the comorbidity risk variables in the study remained consistent over the duration of the study, and many actually increased; however, the variations in the individual comorbidities did not decrease to the extent that was seen for readmission rates. Similarly, my observation would be that complications are less under the control of the surgeon than readmission rates. Although overall complication rates go down, some actually increase, with clear examples such as renal failure and obesity, which are clearly increasing over the time of study. The study by Bozic et al. clearly qualifies as a “big data” experiment, and to state that such experiments give you the answer is a non sequitur3. Although the technology of assembling literally millions of data points into a database that is somewhat usable is remarkable, the key is how we as readers and practicing physicians found it to be useful information. The authors simplify the process by using data analytic tools that may be found in a high school advanced-placement statistics textbook or Wikipedia, and then compile the important take-aways into tables that enable data visualization or “viz.” This a key information tool to provide insights for the end user. Most of what we need to understand and take away from this article may be found in the pair of figures. The rest is up to us, and this entails applying the data to our experience, observations, future expectations, biases, and orthopaedic gestalt. We need to get into the information technology game.