Development of a hemoptysis risk prediction model for patients following CT-guided transthoracic lung biopsy

Currently, low-dose computed tomography (CT) is recommended for lung cancer screening in routine clinical practice at many certified medical centers [1, 2]. This recommendation has contributed to increment the number of CT-detectable pulmonary lesions, including asymptomatic pulmonary nodules and masses [3]. Generally, in the presence of lesions larger than 10 mm or smaller lesions with a rapid growth rate, additional diagnostic procedures would be needed, such as transthoracic lung biopsy guided by CT, transbronchial lung biopsy guided by endobronchial ultrasound or virtual bronchoscopic navigation, or surgery [1, 4]. Among these procedures, CT-guided transthoracic needle biopsy (CT-TNB) is usually the preferred method because it is minimally invasive, has higher diagnostic accuracy and a lower cost [4‐6]. However CT-TNB also has complications, such as hemoptysis, which occurs with a frequency of 0.5–14.4%, and intrapulmonary hemorrhage, with an even higher rate of 2.9–54.5% [4, 5].

Nowadays, CT-TNB is routinely performed on an outpatient basis in many medical units. In this scenario, an important issue concerning outpatient management is not the occurrence of hemoptysis per se, but hemoptysis requiring hemostatic therapy and patient hospitalization. Therefore, it would be helpful if clinicians could predict the risk of post-CT-TNB hemoptysis in clinical practice. Unfortunately, to our knowledge, there is no recommended model for the prediction of post-CT-TNB hemoptysis.

In the present study, we retrospectively investigated patients’ clinical characteristics, lesion features, and CT-TNB parameters, in order to establish a predictive model based on valuable predictors of post-CT-TNB hemoptysis.

In the derivation cohort, 43.7% (134/307, 95% confidence interval [CI], 38.1–49.2%) of patients experienced hemoptysis following CT-TNB, and 2 (1.5% of the hemoptysis) cases required hemostasis treatment. In the validation cohort, the incidence of hemoptysis was 41.1% (53/129, 95% CI, 32.6–49.6%), and 1 (1.9% of the hemoptysis) patient required hemostasis treatment. One patient died from severe postoperative blood loss. This patient was a 53-year-old man with hypertension and uremia. A solid mass with a diameter of 3.3 cm in the right middle lobe was found on CT scan, and the shortest distance from the center of the lesion to the body surface was 1.5 cm. This patient received CT-TNB in supine position and only one biopsy was performed. The postoperative pathological result of the lesion was pulmonary inflammatory pseudotumor. Patient’s clinical characteristics, laboratory tests, and biopsy parameters are shown in Table 1.

After analyzing the 309 patients in the derivation cohort, 29 variables were reduced to 5 potential predictors based on nonzero coefficients in the LASSO regression analysis (Fig. 1). These variables were: lesion diagnosis, lesion characteristics, lesion diameter, procedure time, and puncture distance.

We established a risk prediction model for CT-TNB-induced hemoptysis based on the aforementioned 5 predictors, which independently associated with the risk of post-CT-TNB hemoptysis as assessed by logistic regression analysis (Table 2). As shown in Fig. 2, the AUC for the predictive model (black line) was 0.850 (95% CI, 0.808–0.893), while the AUC for the validation (red line) was 0.767 (95% CI, 0.684–0.851). The optimal cut-off value of ROC curve was 0.45 based on the maximum principle of Youden index, where it yielded an accuracy of 79.2%, a sensitivity of 76.3%, a specificity of 81.3%, a positive predictive value of 75.2%, and a negative predictive value of 82.2%.

To provide clinicians with a quantitative tool to predict the risk of post-CT-TNB hemoptysis, a nomogram was constructed based on the predictive model (Fig. 3). The calibration curve of the model for the risk of post-CT-TNB hemoptysis shows good consistency between prediction and observation in the derivation cohort. The unreliability test yielded a p-value of 0.994, with a Emax of 0.038 and a Eavg of 0.010, which indicated that there was no departure from a perfect fit (Fig. 4).

The decision curve for the model and that for the validation is presented in Fig. 5. DCA revealed that when the threshold probability of an individual was ≥10% (in the derivation cohort, Fig. 5a) or between 5 and 90% (in the validation cohort, Fig. 5b), application of this model to predict the risk of post-CT-TNB hemoptysis would add net benefit than applying either the treat-all or treat-none strategies.

In the present study, we developed and validated a risk prediction model for hemoptysis following CT-TNB. This predictive model incorporates five variables: diagnosis of the lesion, lesion characteristics, lesion diameter, procedure time, and puncture distance. The model showed good discrimination predictive ability (AUC: 0.850). We also constructed a nomogram based on these five variables to facilitate the individualized prediction of post-CT-TNB hemoptysis.

In the last few decades, the detection rate of lung lesions has increased, especially in the case of asymptomatic pulmonary nodules and masses [9]. To obtain a pathological diagnosis, several techniques (e.g. CT, endobronchial ultrasound, and virtual bronchoscopic navigation) have been applied for guidance during the biopsy procedure. However, CT-TNB currently remains a frequently used method in clinical practice [6, 10]. Reportedly, the main complications of CT-TNB include hemoptysis, pneumothorax, hemothorax, air embolism, and infection [4, 11, 12]. The incidence of post-CT-TNB hemoptysis, based on different study populations and the type of needle used, varies between 0.5 and 14.4%, and the rate of post-CT-TNB intrapulmonary hemorrhage is 2.9–54.5% [4, 5]. In our study, the incidence of hemoptysis was as high as 42.9%. We speculate that the following factors may have contributed to the high hemoptysis rate. First, we used an 18-gauge coaxial needle rather than fine needle aspiration to perform the biopsies, and it has been reported that the rate of hemoptysis is higher for core biopsies than when fine needle aspiration is used [5]. Second, 96% (418/436) of the patients in our study were subjected to two or more biopsies. Third, 43% (187/436) of the lesions were benign and 39% (172/436) were non-nodular lesions, and more biopsies and larger-gauge needles are recommended when trying to confirm a benign diagnosis [13]. Finally, but most importantly, we defined blood in sputum as hemoptysis in the present study.

There are a number of studies focusing on post-CT-TNB hemoptysis complications [14‐16]. However, to our knowledge, no prediction models for CT-TNB-induced hemoptysis have been previously reported. Severe hemoptysis can be life-threatening, and this complication in outpatients would be even more significant. Therefore, it would be helpful for predicting the risk of post-CT-TNB hemoptysis. In this study, we established a risk prediction model based on 5 variables determined by LASSO regression analysis. Regression shrinkage and selection via the LASSO method were first reported by Robert in 1996 [17], and this approach is considered superior to the method of choosing predictors according to the strength of their univariable association with the outcome, especially when there are a large number of variables [18, 19]. All these 5 predictors are easily available clinically. In addition, our prediction model showed both good discrimination ability and calibration.

With regard to its clinical usefulness, we performed DCA to assess whether clinical decisions taken based on this proposed model would improve patient outcomes. DCA, based on different threshold probabilities, could provide insight into the consequences of clinical decisions for clinicians [8, 20, 21]. The DCA based on our model is excellent. It shows that if the threshold probability of an individual was ≥10%, applying the model to predict post-CT-TNB hemoptysis would add a net benefit. Therefore, this model would be useful in the management of post-CT-TNB outpatients. On the one hand, it could reduce patient postoperative anxiety, which induced by post-CT-TNB hemoptysis. On the other hand, it will help physicians screen patients who have a high risk of postoperative hemoptysis, extend their postoperative observation, and provide timely treatment when necessary.

Our study had several major limitations. Firstly, it was a single-center and retrospective evaluation. Although we carried out a validation test of the model, additional independent external verification is warranted to confirm its utility in clinical practice. Secondly, we did not include quantitative measurement of the volume of hemoptysis; therefore, this model can only be used to predict post-CT-TNB hemoptysis, not to distinguish the severity of the hemoptysis. Thirdly, some variables with potential predictive value may help to improve the discrimination ability of the model, such as morphological characteristics of blood vessels, blood flow directions, vessel diameters, local blood pressure and shear stress distribution of all the vessels that involved along the punching pass; however, they were not available in the retrospective data.

We established a predictive post-CT-TNB hemoptysis model based on 5 predictors. This model not only showed good discrimination ability and calibration characteristics but also demonstrated excellent clinical application potential, as determined by DCA. Therefore, this predictive model may be of great value to facilitate the individualized preoperative prediction of post-CT-TNB hemoptysis.