Risk Assessment of Malignancy in Solitary Pulmonary Nodules in Lung Computed Tomography: A Multivariable Predictive Model Study
Introduction
Lung cancer remains the leading cause of cancer-related deaths worldwide, accounting for 18.4% of total cancer deaths in 2018. The use of low-dose computed tomography (LDCT) for lung cancer screening has shown promise in reducing mortality rates by 20%. However, LDCT screening also presents challenges, including false positives, false negatives, overdiagnosis, and radiation exposure. Notably, 95% of positive LDCT results do not lead to a lung cancer diagnosis, potentially causing unnecessary physical and psychological distress for patients.
To enhance the effectiveness of LDCT screening, predictive models have been developed to assess the malignancy probability of pulmonary nodules detected during screening. These models aim to guide decisions regarding CT follow-up, biopsy, or surgery, ultimately improving diagnostic accuracy and reducing unnecessary interventions.
Background and Study Objectives
The complexity of computed tomography (CT) images, particularly those of solitary pulmonary nodules (SPNs), often leads to diagnostic challenges. Accurate differentiation between benign and malignant nodules is crucial for effective lung cancer treatment. While several predictive models exist, their applicability to different populations requires validation.
The Mayo model, developed by Swensen et al., was the first lung cancer risk prediction model and has been widely recommended. However, its performance in Chinese populations remains uncertain. This study aimed to externally validate and recalibrate the Mayo model using a large Chinese cohort and to develop a new predictive model tailored to this population.
Methods
Study Population and Data Collection
This retrospective study included 1,450 patients from three tertiary hospitals in China who underwent surgical resection for SPNs between 2014 and 2020. Patients were divided into training (n=849), internal validation (n=365), and external validation (n=236) sets. Clinical and CT imaging data were collected, including patient demographics, smoking history, nodule characteristics, and pathological diagnoses.
External Validation and Recalibration of the Mayo Model
The original Mayo model was applied to the training set, and its performance was evaluated using the area under the receiver operating characteristic curve (AUC). The model was then recalibrated by re-estimating the coefficients of all covariates using the training data.
Development of the New Predictive Model
A new logistic regression model was developed using backward stepwise selection with the Akaike information criterion. The model incorporated various clinical and imaging features, including nodule type, history of chronic obstructive pulmonary disease (COPD), nodule margin characteristics, and presence of specific imaging signs.
Model Performance Evaluation
The performance of the Mayo model, revised Mayo model, and new model was assessed using AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The models were compared using the DeLong test.
Results
External Validation and Recalibration of the Mayo Model
The original Mayo model showed poor performance on the training set (AUC: 0.653; 95% CI: 0.613-0.694). After recalibration, the revised Mayo model achieved a slightly improved AUC of 0.671 (95% CI: 0.635-0.706). However, both models performed poorly on the internal and external validation sets, indicating their limited applicability to the Chinese population.
Development and Performance of the New Model
The new predictive model demonstrated superior performance, achieving an AUC of 0.891 (95% CI: 0.865-0.917) on the training set. It maintained high performance on the internal validation set (AUC: 0.888; 95% CI: 0.842-0.934) and external validation set (AUC: 0.876; 95% CI: 0.831-0.920).
The new model included nine independent risk factors for malignant pulmonary nodules: nodule type, age, history of COPD, spiculation, lobulation, vacuole sign, calcification, presence of vessels, and satellite nodules. Notably, solid nodule type and calcification were identified as protective factors.
Discussion
The study’s findings highlight the limitations of the Mayo model in Chinese populations and the need for population-specific predictive models. The new model developed in this study demonstrated significantly better performance than both the original and revised Mayo models, particularly in differentiating between benign and malignant pulmonary nodules.
Key findings and implications:
-
The Mayo model showed poor predictive performance in the Chinese cohort, even after recalibration, emphasizing the importance of population-specific model development.
-
The new model incorporated novel risk factors, including nodule type, history of COPD, and specific imaging characteristics, which contributed to its improved performance.
-
The model’s high sensitivity (79.65%-92.10%) and specificity (72.97%-83.74%) suggest its potential utility in clinical practice for reducing missed diagnoses of malignant nodules.
-
The inclusion of history of COPD as a risk factor aligns with previous research showing an increased lung cancer risk in COPD patients, even among non-smokers.
-
The identification of solid nodule type and calcification as protective factors may reflect the unique characteristics of the study population, particularly the high prevalence of tuberculosis granulomas.
Limitations and Future Directions
While the study provides valuable insights, several limitations should be noted:
-
The external validation cohort was relatively small and limited to two centers, potentially limiting the generalizability of the findings.
-
The study population primarily consisted of high-risk nodules requiring surgical resection, which may have led to an overestimation of malignancy risk.
-
The model’s performance in differentiating benign nodules was slightly reduced, possibly due to the imbalance between malignant and benign cases in the training set.
Future research should focus on:
-
Expanding the external validation cohort to include more diverse populations and centers.
-
Investigating the model’s performance in different clinical settings, including primary care and screening populations.
-
Exploring the incorporation of additional biomarkers or advanced imaging features to further improve predictive accuracy.
Conclusion
This study demonstrated the limitations of the Mayo model in Chinese populations and developed a new predictive model tailored to this population. The new model showed superior performance in differentiating between benign and malignant pulmonary nodules, particularly in reducing missed diagnoses of malignant cases. The inclusion of novel risk factors and population-specific considerations contributed to the model’s improved accuracy. These findings highlight the importance of developing population-specific predictive models for lung cancer diagnosis and the potential for such models to enhance clinical decision-making and patient outcomes.
doi.org/10.1097/CM9.0000000000001507
Was this helpful?
0 / 0