ERS5748. A Study of Patient Outcomes Using a Hybrid Human and Artificial Intelligence Workflow for Follow-up Recommendations from Radiology
Authors * Denotes Presenting Author
  1. Kayla Nakashima; University of California - Irvine
  2. Riya Bansal; University of California - Irvine
  3. Chang Shu; University of California - Irvine
  4. Erwin Ho; University of California - Irvine
  5. Star Lopez *; University of California - Irvine
  6. River Wolf; University of California - Irvine
  7. Roozbeh Houshyar; University of California - Irvine
Radiologists frequently include follow-up recommendations in their reports. However, physician-adherence to these recommendations often falls short, leading to suboptimal clinical care and medical litigation. To address this issue, our institution implemented a hybrid system involving both natural language processing (NLP) software to identify missed follow-up imaging recommendations and a quality and patient safety nurse to coordinate care between providers and patients. The revenue generated from completion of recommended imaging surpasses the operating costs, underscoring the potential profitability of this hybrid system. It is well known there are economic and emotional burdens for a patient and caregiver when a patient is undergoing assessment and/or ultimately diagnosed with cancer. Early detection through timely follow-up imaging has the potential to lessen this burden. In this study, we aim to assess the system’s impact on health and welfare outcomes for patients who receive timely follow-up imaging exams.

Materials and Methods:
A total of 250 consecutive missed radiology follow-up exams over the course of 1 year that were addressed using our hybrid follow-up system were assessed for results and patient outcomes. Clinical diagnoses for these exams were established by reviewing the history, imaging findings, and lab results of each patient. Next, diagnoses were categorized into “exclude cancer/stable benign finding(s)” and “de novo malignancy.” Non-cancer related diagnostic concerns were excluded from the study. This left us with 211 exams for cost-benefit analysis. Cancer type was assigned to these 211 exams, and morbidity was analyzed using a systematic analysis for the Global Burden of Disease study.

Of the 211 cancer-related cases, the 3 most common cancer-types included 71 thyroid, 44 lung, and 24 ovarian. According to the National Cancer Institute, the average cost of care for these cancers during the last year of life is $107,437, $110,248, and $112,018 respectively. The absolute DALYs (Disability-Adjusted Life Years) in millions for 2019 were 1.23, 45.9, and 5.36, respectively. The use of this hybrid follow-up system has the potential to detect cancer early, preventing an even greater financial burden and lessening morbidity.

Because of this hybrid system, communication between patient and ordering physician improved along with adherence to radiologist follow-up within the national guidelines. The ability to detect early-stage malignancy improves clinical outcomes and increases treatment options, affording patients time to make relevant life decisions. Given the significance of early detection and improved outcomes, systematic follow-up and proactive communication have the potential to positively impact overall patient satisfaction. Limitations of this study include lack of patient feedback on the significance of follow-up communication and no comparative analysis of a patient population for which our follow-up system was not employed.