A Prediction Model to Optimize Invasive Mediastinal Staging Procedures for Non-small Cell Lung Cancer in Patients With a Radiologically Normal Mediastinum The Quebec Prediction Model

BACKGROUND: Current guideline-recommended criteria for invasive mediastinal staging in patients with a radiologically normal mediastinum fail to identify a significant proportion of patients with occult mediastinal disease (OMD), despite it leading to a large number of invasive staging procedures. RESEARCH QUESTION: Which variables available before surgery predict the probability of OMD in patients with a radiologically normal mediastinum? STUDY DESIGN AND METHODS: We identified all cTxN0/N1M0 non-small cell lung cancer tumors staged by CT imaging and PET with CT imaging in our institution between 2014 and 2018 who underwent gold standard surgical lymph node dissection or were demonstrated to have OMD before surgery by invasive mediastinal staging techniques and divided them into a derivation and an independent validation cohort to create the Quebec Prediction Model (QPM), which allows calculation of the probability of OMD. RESULTS: Eight hundred three patients were identified (development set, n = 502; validation set, n = 301) with a prevalence of OMD of 9.1%. The developed prediction model included largest mediastinal lymph node size (P < .001), tumor centrality (P = .23), presence of cN1 disease (P = .29), and lesion standardized uptake value (P = .09). Using a calculated probability of more than 10% as a threshold to identify OMD, this model had a sensitivity, specificity, positive predictive value, and negative predictive value in the derivation cohort of 73.9% (95% CI, 58.9%-85.7%), 81.1% (95% CI, 77.2%-84.6%), 28.3% (95% CI, 23.4%-33.8%), and 96.8% (95% CI, 95.0%-98.1%), respectively. It performed similarly in the validation cohort (P = .77, Hosmer-Lemeshow test; P = .5163, Pearson c2 and unweighted sum-of squares statistics; and P = .0750, Stukel score test) and outperformed current guideline recommended criteria in identifying patients with OMD (area under the receiver operating characteristic curve [AUC] for American College of Chest Physicians guidelines criteria, 0.65 [95% CI, 0.59-0.71]; AUC for European Society of Thoracic Surgeons guidelines criteria, 0.60 [95% CI, 0.54-0.67]; and AUC for the QPM, 0.85 [95% CI, 0.80-0.90]). INTERPRETATION: The QPM allows the clinician to integrate available information from CT and PET imaging to minimize invasive staging procedures that will not modify management, while also minimizing the risk of unforeseen mediastinal disease found at surgery.

Automated summary

The study developed a prediction model to identify the probability of occult mediastinal disease (OMD) in patients with a radiologically normal mediastinum. The model included variables such as largest mediastinal lymph node size, tumor centrality, presence of cN1 disease, and lesion standardized uptake value. This model outperformed current guideline-recommended criteria in identifying patients with OMD.