The bubble in predicting bubbles?

‘Predicting bubbles’ is an activity that one intuitively does not associate with thoracic surgeons. While the study of famous historical market bubbles such as the Dutch Tulipmania in 1634 make fine reading and while the Roaring Twenties seem to go for a remake, other specialties focus on the prediction models of this phenomenon in an ever increasing way. This was even labelled ‘The bubble in predicting bubbles’ on Twitter^®.

Prolonged air leak (PAL) after anatomical lung resections is significantly related to longer hospital stay and increased hospitalization cost [1]. The most common definition of PAL is an air leak lasting for >5 days postoperatively, although some studies also use 7 days as a cut-off [2, 3].

The relative importance of PAL increased over recent years. Before the implementation of minimally invasive techniques and enhanced recovery protocols, PAL could be concealed behind an expected longer recovery after thoracotomy and/or more frequent complications. Today, PAL will often be the main reason of prolonged chest drain duration and/or hospital stay.

JCOG0802, a randomized controlled trial including 1106 patients with non-small cell lung cancer (NSCLC) of ≤2 cm in size, demonstrated that airleak grade 2 (defined as ≥7 days or requiring pleural adhesion therapy) was more common after segmentectomy compared to lobectomy (6.5% vs 3.8%) [3]. In this trial, segmentectomies were subdivided into simple and complex procedures. Simple segmentectomy was defined as a segmental resection of the right or left segment 6, left superior, or lingular segment. Complex segmentectomy was resection of any other segment, in fact a segment with >1 intersegmental plane. Grade 2 air leak was more common after complex (7.6%, n = 23/300) compared to simple (4.8%, n = 11/229) segmentectomies, suggesting the obvious relationship between the amount of parenchymal dissection to obtain an anatomical resection and the occurrence of prolonged airleak. CALGB 14503, a randomized controlled trial between lobectomy and sublobar resections, did not show significant differences in prolonged airleak, but the sublobar group contained non-anatomical (wedge) resections in almost 60% [4].

Previous studies have tried to develop a scoring system to predict the chance of PAL, but these were built on databases containing mainly lobectomies. The results of recent RCTs, the increasing proportion of smaller lesions and the rising numbers in pulmonary metastases surgery are thought to lead to relatively more segmentectomies in the future.

In this issue of the journal, authors from the Thoracic Center of Western Switzerland report their findings of a four-centre retrospective study between January 2016 and October 2020 [5]. It includes 453 consecutive segmentectomies and looks at the predictive factors for postoperative airleak. In a multivariable analysis, tobacco use, diffusion capacity (DLCO) <80%, lower body mass index (BMI) and presence of pleural adhesions were identified as risk factors for PAL, while obesity was protective. The paper is unique in focusing specifically on segmentectomies when comparing 3 existing European scoring systems for the prediction of PAL. The receiver operating characteristic (ROC) areas under the curve did not differ significantly between the 3 investigated models. The minimal and maximum value of the 95% confidence intervals were 0.599 and 0.729, respectively.

Several findings are interesting to discuss. The paper adds data to the current literature that there is an important risk for PAL after segmentectomy. During segmentectomies, a large amount of intersegmental trans-parenchymal division has to be performed, even if well-developed fissures are present. Furthermore, deeper, more peripheral, dissection is necessary along the bronchovascular tree, with a higher risk of parenchymal injury. With 9% at 7 days and 14% at 5 days, the PAL ratio in this paper is higher than reported by JCOG0802, but this difficult to compare due to demographics and characteristics of the population. However, the difference between 5 and 7 days illustrates the importance of a single definition. In fact, one of the investigated models was constructed with the 7 days limit, but 5 days was used in the present comparison.

In addition, the authors did not find a difference between simple and complex segmentectomies (12%, n = 22/181 vs 15%, n = 42/272) as observed in JCOG 0802. One can hypothesize that the liberal use of stapling devices for the intersegmental plane diminishes the difference in PAL between complex and simple segmentectomies. Complex segmentectomies need more division of intersegmental planes. But, both simple and complex segmentectomies need deep hilar dissection with risk of parenchymal disruption.

The discrimination capacity of the different tested prediction models was moderate with area’s under the curve lower than 0.7. This should not come as a surprise as the used parameters in the models are rather non-specific, such as gender, extent of resection, pulmonary function, age and BMI. One should wonder whether modern techniques such as computed tomography (CT) 3D reconstruction highlighting areas of lung destruction and completeness of fissures as now used in lung reduction surgery or bronchial valve therapy planning could be useful and incorporated [6].

Besides, what can the clinical implication be of these PAL prediction models, if performant? Will surgeons adopt their technique because of the results of the model, change the type of resection and/or use more preventive sealants? Future studies should be prospective and randomized, ideally demonstrating that the use of a prediction model changes the outcome for the patient. At this time, the most important argument in favour of the use of these models is patient counselling. It is therefore necessary to only incorporate preoperative known parameters, and not presence of ‘pleural adhesions’ as was done on 2 of the current investigated models.

The Thoracic Center of Western Switzerland is to be congratulated for their contribution. It reminds us to include the discussion on the risk of PAL when informing patients, exercise caution when handling lung parenchyma and use countermeasures to minimize air leak. The authors demonstrated that the current models are not highly discriminatory and further work is needed. In other words, one can expect that the ‘bubble in predicting bubbles’ is not (yet) reached.

REFERENCES