Evaluating the accuracy of lung-RADS score extraction from radiology reports: Manual entry versus natural language processing.

INTRODUCTION: Radiology scoring systems are critical to the success of lung cancer screening (LCS) programs, impacting patient care, adherence to follow-up, data management and reporting, and program evaluation. LungCT ScreeningReporting and Data System (Lung-RADS) is a structured radiology scoring system that provides recommendations for LCS follow-up that are utilized (a) in clinical care and (b) by LCS programs monitoring rates of adherence to follow-up. Thus, accurate reporting and reliable collection of Lung-RADS scores are fundamental components of LCS program evaluation and improvement. Unfortunately, due to variability in radiology reports, extraction of Lung-RADS scores is non-trivial, and best practices do not exist. The purpose of this project is to compare mechanisms to extract Lung-RADS scores from free-text radiology reports. METHODS: We retrospectively analyzed reports of LCS low-dose computed tomography (LDCT) examinations performed at a multihospital integrated healthcare network in New York State between January 2016 and July 2023. We compared three methods of Lung-RADS score extraction: manual physician entry at time of report creation, manual LCS specialist entry after report creation, and an internally developed, rule-based natural language processing (NLP) algorithm. Accuracy, recall, precision, and completeness (i.e., the proportion of LCS exams to which a Lung-RADS score has been assigned) were compared between the three methods. RESULTS: The dataset includes 24,060 LCS examinations on 14,243 unique patients. The mean patient age was 65 years, and most patients were male (54 %) and white (75 %). Completeness rate was 65 %, 68 %, and 99 % for radiologists' manual entry, LCS specialists' entry, and NLP algorithm, respectively. Accuracy, recall, and precision were high across all extraction methods (>94 %), though the NLP-based approach was consistently higher than both manual entries in all metrics. DISCUSSION: An NLP-based method of LCS score determination is an efficient and more accurate means of extracting Lung-RADS scores than manual review and data entry. NLP-based methods should be considered best practice for extracting structured Lung-RADS scores from free-text radiology reports.

2024 Update of The Society of Thoracic Surgeons Short-term Esophagectomy Risk Model: More Inclusive and Improved Calibration.

BACKGROUND: The Society of Thoracic Surgeons General Thoracic Surgery Database (STS-GTSD) previously reported short-term risk models for esophagectomy for esophageal cancer. We sought to update existing models using more inclusive contemporary cohorts, with consideration of additional risk factors based on clinical evidence. METHODS: The study population consisted of adult patients in the STS-GTSD who underwent esophagectomy for esophageal cancer between January 2015 and December 2022. Separate esophagectomy risk models were derived for 3 primary end points: operative mortality, major morbidity, and composite morbidity or mortality. Logistic regression with backward selection was used, with predictors retained in models if P < .10. All derived models were validated using 9-fold cross-validation. Model discrimination and calibration were assessed for the overall cohort and specified subgroups. RESULTS: A total of 18,503 patients from 254 centers underwent esophagectomy for esophageal cancer. Operative mortality, morbidity, and composite morbidity or mortality rates were 3.4%, 30.5%, and 30.9%, respectively. Novel predictors of short-term outcomes in the updated models included body surface area and insurance payor type. Overall discrimination was similar or superior to previous STS-GTSD models for operative mortality (C statistic = 0.72) and for composite morbidity or mortality (C statistic = 0.62), Model discrimination was comparable across procedure- and demographic-specific subcohorts. Model calibration was excellent in all patient subgroups. CONCLUSIONS: The newly derived esophagectomy risk models showed similar or superior performance compared with previous models, with broader applicability and clinical face validity. These models provide robust preoperative risk estimation and can be used for shared decision making, assessment of provider performance, and quality improvement.

Ultralow-Dose Dynamic Expiratory CT and Repeated Imaging Enhance Evaluation for Tracheomalacia.

OBJECTIVE: This study aims to determine if a novel imaging protocol (ultralow-dose dynamic expiratory computed tomography [CT] with repeated imaging) identifies tracheomalacia (TM) more reliably than traditional dynamic tracheal CT. METHODS: We performed a retrospective evaluation of 184 consecutive ultralow-dose dynamic CTs for TM during 2017. The protocol obtains images during 1 inspiration and 2 forced expirations. Tracheal narrowing during both expirations (airway narrowing [percentage] during first dynamic expiration CT [DE1], airway narrowing [percentage] during second dynamic expiration CT [DE2]) was reported as a percentage of inspiratory area. We identified maximum narrowing of each patient's sequence (maximum narrowing [percentage] on either dynamic expiration CT [DEmax] = greatest narrowing of DE1 or DE2) and compared DE1, DE2, and DEmax in individual studies and between patients. Outcomes included frequency of TM, tracheal narrowing, and severity. Reliability was assessed by comparing tracheal area narrowing and TM grade. RESULTS: There was significantly more airway narrowing using 2 expiratory image acquisitions. Average DEmax tracheal area was 12% narrower than DE1 alone and 21% worse than DE2 alone (both P < 0.001). Using DEmax, TM was diagnosed 35% more often than DE1 alone and 31% more often than DE2 alone ( P < 0.001). DEmax identified more severe distribution of TM compared with DE1 or DE2 alone ( P < 0.001). Reliability between DE1 and DE2 was good for tracheal narrowing and moderate for TM grade. The mean effective radiation dose was 2.41 millisievert (mSv) for routine inspiration CT and 0.07 mSv for each dynamic expiration CT (total effective radiation, 2.55 mSv). CONCLUSIONS: Dynamic expiration CT with 2 expiratory image acquisitions enhanced evaluation of TM, minimally increased radiation dose, and should be considered as a noninvasive screening option.

Tracheobronchoplasty: Indications and Best Approaches.

Tracheobronchomalacia (TBM) is an increasingly recognized abnormality of the central airways in patients with respiratory symptoms. Severe TBM in symptomatic patients warrants screening dynamic CT of the chest and/or awake dynamic bronchoscopy. The goal of surgical repair is to restore the C-shaped configuration of the airway lumen and splint or secure the lax posterior membrane to the mesh to ameliorate symptoms. Robotic tracheobronchoplasty is safe and associated with improvements in pulmonary function and subjective improvement in quality of life.

Improvement in postoperative lung function in patients with moderate to severe airway obstruction after robotic-assisted thoracoscopic tracheobronchoplasty.

OBJECTIVE: The study objective was to examine pulmonary function and quality of life improvement after robotic-assisted thoracoscopic tracheobronchoplasty for patients with different degrees of obstructive airway disease. METHODS: We performed a retrospective review of a prospective database of patients who underwent robotic-assisted thoracoscopic tracheobronchoplasty between 2013 and 2020. RESULTS: A total of 118 patients underwent robotic-assisted thoracoscopic tracheobronchoplasty. Preoperative and postoperative pulmonary function tests were available for 108 patients. Postoperative pulmonary function tests at a median of 16 months demonstrated a significant increase in percent predicted forced expiratory volume in 1 second (preoperative median: 76.76% predicted, postoperative: 83% predicted, P = .002). Preoperative and postoperative St George Respiratory Questionnaires were available for 64 patients with a significant decrease in postoperative score at a median of 7 months (preoperative median: 61, postoperative: 41.60, P < .001). When stratified by preoperative degree of obstruction, robotic-assisted thoracoscopic tracheobronchoplasty improved forced expiratory volume in 1 second in moderate to very severe obstruction with a statistically significant improvement in moderate (preoperative median: 63.91% predicted, postoperative median: 73% predicted, P = .001) and severe (preoperative median: 44% predicted, postoperative median: 57% predicted, P = .007) obstruction. St George Respiratory Questionnaire scores improved for all patients. Improvement for mild (preoperative median: 61.27, postoperative median: 36.71, P < .001) and moderate (preoperative median: 57.15, postoperative median: 47.52, P = .03) obstruction was statistically significant. CONCLUSIONS: Robotic-assisted thoracoscopic tracheobronchoplasty improves obstruction and symptoms. With limited follow-up, subgroup analysis showed forced expiratory volume in 1 second improved in severe preoperative obstruction and quality of life improved in moderate obstruction. Future follow-up is required to determine robotic-assisted thoracoscopic tracheobronchoplasty effects on the most severe group, but we cannot conclude that increased degree of preoperative obstruction precludes surgery.

Robotic-assisted tracheobronchoplasty: Quality of life and pulmonary function assessment on intermediate follow-up.

OBJECTIVE: The initial description of robotic tracheobronchoplasty for the treatment of tracheobronchomalacia demonstrated feasibility, safety, and short-term symptomatic and functional improvement. The purpose of the current study was to demonstrate intermediate outcomes in postoperative pulmonary function and quality of life after robotic tracheobronchoplasty. METHODS: We retrospectively reviewed prospectively collected clinical data from 42 patients who underwent robotic tracheobronchoplasty from May 2016 to December 2017. The Institutional Review Board or equivalent ethics committee of the Northwell Health approved the study protocol and publication of data. Patient written consent for the publication of the study data was waived by the Institutional Review Board. RESULTS: A total of 42 patients underwent robotic tracheobronchoplasty during the study period. Median total follow-up is 40 months. There was 1 death since surgery from an unrelated disease. Significant decreases in St George's Respiratory Questionnaire total score (preoperative mean: 64.01, postoperative mean: 38.91, P = .002), St George's Respiratory Questionnaire symptom score (preoperative median: 82.6, postoperative median: 43.99, P < .001), and St George's Respiratory Questionnaire impact score (preoperative median: 55.78, postoperative median: 25.95, P < .001) were apparent at a median follow-up of 13 months. Comparison of preoperative and postoperative pulmonary function tests revealed a significant increase in percent predicted forced expiratory volume in 1 second (preoperative median: 74% vs postoperative median: 82%, P = .001), forced vital capacity (preoperative median: 68.5% vs postoperative median: 80.63%, P < .001), and peak expiratory flow (preoperative median: 61.5% vs postoperative median: 75%, P = .02) measured at a median follow-up of 29 months. CONCLUSIONS: Robotic tracheobronchoplasty is associated with low intermediate-term mortality. Robotic tracheobronchoplasty results in significant improvement in quality of life and postoperative pulmonary function. Longer-term follow-up is necessary to continue to elucidate the effect of robotic tracheobronchoplasty on halting pathologic progression of tracheobronchomalacia and to determine the long-term impact of tracheobronchoplasty on symptomatic and functional improvement.