Natural Language Processing of Patient-Experience Comments After Primary Total Knee Arthroplasty

Abstract

BackgroundThere is interest in improving patient experience after total knee arthroplasty (TKA) due to recent shifts toward value-based medicine. Patient narratives are a valuable but unexplored source of information. MethodsRecords of 319 patients who had undergone primary TKA between August 2016 and August 2019 were linked with vendor-supplied patient satisfaction data, which included patient comments and the Press Ganey satisfaction survey. Using machine-learning-based natural language processing, 1048 patient comments were analyzed for sentiment and classified into themes. Postoperative outcomes, patient-reported outcome measures, and traditional measures of satisfaction were compared between patients who provided a negative comment vs those who did not (positive, neutral, mixed grouped together). Multivariable regression was used to determine perioperative variables associated with providing a negative comment. ResultsOf the 1048 patient comments, 25% were negative, 58% were positive, 8% were mixed, and 9% were neutral. Top 2 themes of negative comments were room condition (25%) and inefficient communication (23%). There were no differences in most of the studied outcomes (eg, peak pain intensity, length of stay, or Knee Injury and Osteoarthritis Outcome Score Junior and pain scores at 6-week follow-up) between the 2 cohorts (P > .05). However, patients who made negative comments were less likely to highly recommend their hospital care to peers (P < .001). Finally, patients who had higher American Society of Anesthesiologists Score and those who received a scopolamine patch were more likely to provide negative comments (P < .05). ConclusionAlthough the current study showed that patient satisfaction might not be a proxy for traditional objective perioperative outcomes, efforts to improve the nontechnical aspects of medicine are still crucial in providing patient-centered care.