Objectives: Classifying radiologic pulmonary lesions as malignant is challenging. Scoring systems like the Mayo model lack precision in predicting the probability of malignancy. We developed the logistic scoring system ‘LIONS PREY’ (Lung lesION Score PREdicts malignancY), which is superior to existing models in its precision in determining the likelihood of malignancy. Methods: We evaluated all patients that were presented to our multidisciplinary team between January 2013 and December 2020. Availability of pathological results after resection or CT-/EBUS-guided sampling was mandatory for study inclusion. Two groups were formed: Group A (malignant nodule; n = 238) and Group B (benign nodule; n = 148). Initially, 22 potential score parameters were derived from the patients’ medical histories. Results: After uni- and multivariate analysis, we identified the following eight parameters that were integrated into a scoring system: (1) age (Group A: 64.5 ± 10.2 years vs. Group B: 61.6 ± 13.8 years; multivariatep-value: 0.054); (2) nodule size (21.8 ± 7.5 mm vs. 18.3 ± 7.9 mm;p= 0.051); (3) spiculation (73.1% vs. 41.9%;p= 0.024); (4) solidity (84.9% vs. 62.8%;p= 0.004); (5) size dynamics (6.4 ± 7.7 mm/3 months vs. 0.2 ± 0.9 mm/3 months;p< 0.0001); (6) smoking history (92.0% vs. 43.9%;p< 0.0001); (7) pack years (35.1 ± 19.1 vs. 21.3 ± 18.8;p= 0.079); and (8) cancer history (34.9% vs. 24.3%;p= 0.052). Our model demonstrated superior precision to that of the Mayo score (p= 0.013) with an overall correct classification of 96.0%, a calibration (observed/expected-ratio) of 1.1, and a discrimination (ROC analysis) of AUC (95% CI) 0.94 (0.92–0.97). Conclusions: Focusing on essential parameters, LIONS PREY can be easily and reproducibly applied based on computed tomography (CT) scans. Multidisciplinary team members could use it to facilitate decision making. Patients may find it easier to consent to surgery knowing the likelihood of pulmonary malignancy. The LIONS PREY app is available for free on Android and iOS devices.
目的：将影像学肺部病变准确分类为恶性具有挑战性。现有预测模型（如Mayo模型）在评估恶性概率方面精确度不足。本研究开发的逻辑评分系统"LIONS PREY"（肺部病灶评分预测恶性概率）在判断恶性可能性方面展现出优于现有模型的精确度。 方法：纳入2013年1月至2020年12月期间提交多学科团队讨论的所有病例，要求具备手术切除或CT/EBUS引导穿刺后的病理结果。设立两组：A组（恶性结节，n=238）和B组（良性结节，n=148）。从患者病史中初步提取22项潜在评分参数。 结果：经单变量与多变量分析，最终确定纳入评分系统的八项参数：（1）年龄（A组64.5±10.2岁 vs B组61.6±13.8岁，多变量p值0.054）；（2）结节大小（21.8±7.5 mm vs 18.3±7.9 mm，p=0.051）；（3）毛刺征（73.1% vs 41.9%，p=0.024）；（4）实性成分（84.9% vs 62.8%，p=0.004）；（5）尺寸动态变化（6.4±7.7 mm/3月 vs 0.2±0.9 mm/3月，p<0.0001）；（6）吸烟史（92.0% vs 43.9%，p<0.0001）；（7）吸烟包年数（35.1±19.1 vs 21.3±18.8，p=0.079）；（8）癌症病史（34.9% vs 24.3%，p=0.052）。本模型精确度显著优于Mayo评分（p=0.013），总体正确分类率达96.0%，校准度（观测值/预期值比）为1.1，区分度（ROC分析）AUC（95% CI）达0.94（0.92-0.97）。 结论：LIONS PREY系统聚焦核心参数，可基于CT影像实现简便、可重复的应用。多学科团队成员可借助该系统辅助临床决策。患者通过了解肺部恶性病变概率，可更从容地决定是否接受手术治疗。LIONS PREY应用程序已在Android和iOS平台免费发布。
LIONS PREY: A New Logistic Scoring System for the Prediction of Malignant Pulmonary Nodules
肿瘤(癌症)患者之家