Abstract
Introduction
Despite generally excellent patient outcomes for Total Knee Arthroplasty (TKA), there remains a contingent of patients, up to 20%, who are not satisfied with the outcome of their procedure. (Beswick, 2012) There has been a large amount of research into identifying the factors driving these poor patient outcomes, with increasing recognition of the role of non-surgical factors in predicting achieved outcomes. However, most of this research has been based on single database or registry sources and so has inherited the limitations of its source data. The aim of this work is to develop a predictive model that uses expert knowledge modelling in conjunction with data sources to build a predictive model of TKR patient outcomes.
Method
The preliminary Bayesian Belief Network (BBN) developed and presented here uses data from the Osteoarthritis Initiative, a National Institute of Health funded observational study targeting improved diagnosis and monitoring of osteoarthritis. From this data set, a pared down subset of patient outcome relevant preoperative questionnaire sets has been extracted. The BBN structure provides a flexible platform that handles missing data and varying data collection preferences between surgeons, in addition to temporally updating its predictions as the patient progresses through pre and postoperative milestones in their recovery. In addition, data collected using wearable activity monitoring devices has been integrated. An expert knowledge modelling process relying on the experience of the practicing surgical authors has been used to handle missing cross-correlation observations between the two sources of data.
Results
The model presented here has been internally cross validated and has some interesting facets, including the strongest single predictive question of bad outcome for the patient being the presence of lower back pain. Clinical implementation and long term predictive accuracy result collection is ongoing.
Discussion
Unsatisfied patients represent a significant minority of TKR recipients, with multiple, multifaceted causal factors both in surgery and out implicated. Historically, focus has been on the role of management and improvement of the surgical factors, which is linked to the fact that surgical factors can often lead to far more disastrous consequences for the patient and the basic principle that “you only improve what you measure.” Growing collection of Patient Reported Outcome Measures by registries around the world has exposed the fact that management of patient factors has lagged behind. (Judge, 2012) Increasingly, the pivotal role of unmet expectations in determining patient satisfaction (Noble, 2006) and the “expectation gap” (Ghomrawi, 2012) between surgeons and patients has been exposed as an opportunity to improve patient outcomes. By developing a model that uses existing surgical expert knowledge to integrate research identified preoperative factors that can be accurately and practically gathered in a clinical setting, a workflow that manages patient expectations in order to optimize outcomes could reduce dissatisfaction rates in TKR recipients. Future work should focus on improving clinical integration and, in the absence of sufficiently wide, deep and complete patient response and predictor datasets, ways of harnessing existing expert knowledge into an evolving predictive tool of patient outcomes.
