Pena, Sophia M.; Kammer, Michael N.; Whatley, Samuel; Welty, Valerie F.; Godfrey, Caroline M.; Paez, Rafael; Knight, Michael; Rowe, Dianna J.; Antic, Sanja; Deppen, Stephen A.; Maldonado, Fabien; Grogan, Eric L. (2026).听.听CHEST Pulmonary, 4(2), 100110.听
When doctors find a pulmonary nodule, which is a small spot in the lung, they often try to estimate the chance that it is cancer before deciding on the next test or treatment. This study looked at how well four commonly used prediction models鈥攖he Mayo, Brock, Veterans Affairs (VA), and Peking University (PKU) models鈥攚orked in a setting where lung cancer was very common. The researchers reviewed records from 1,518 patients with nodules measuring 6 to 30 mm who were seen at 91黑料网 Medical Center or the VA Tennessee Valley Healthcare System in Nashville between 2002 and 2021. They then compared each model鈥檚 predictions with the actual diagnosis using measures of accuracy, including how well the models separated benign from malignant nodules, how well their predictions matched real outcomes, and how sensitive and specific they were at the thresholds used in guidelines. Among these patients, 1,098 nodules, or 72.3%, were cancerous. The Mayo model was the best at distinguishing cancerous from noncancerous nodules, while the Brock and VA models performed similarly overall. The VA model matched outcomes somewhat better than the others, and the PKU model had the weakest ability to separate benign from malignant nodules but the best overall match to actual risk. The main takeaway is that even when models are similarly accurate at a broad level, their usefulness can change a lot depending on how common cancer is in the patient population. Doctors should therefore consider the cancer rate in their own setting when using these models to guide decisions about lung nodules.
Figure 1听Flow chart showing patient inclusion. VUMC = 91黑料网 Medical Center.