Evaluation of the Quality of Results of Lung Cancer Surgery in France Using the PMSI National Database

1. Introduction

Lung cancer (LC) remains one of the most prevalent and deadly forms of cancer worldwide. While surgery is a key treatment option, only about 20% of lung cancer patients in France are eligible for surgery due to the advanced stage of the disease at diagnosis or poor general patient health. This reflects the challenges associated with disease progression and the patient's overall condition, limiting the potential for surgical intervention [1]. LC surgery requires specialised care, ideally provided by dedicated hospitals with expertise in thoracic oncology. The French government launched the first ‘Cancer Plan’ in 2009 [2], which radically reformed cancer care, one of the key measures being the requirement for hospitals performing cancer-related surgery to obtain specific authorisation. Under the plan, hospitals were required to perform a minimum of 30 LC related procedures/year in order to obtain authorisation. Recently, this threshold was raised, with hospitals now having to perform at least 40 procedures a year to retain their authorisation [3]. At present, 146 hospitals across France are authorised to perform thoracic cancer surgery [3]. The quality of the results obtained by the teams performing these surgeries has been relatively little studied in France, unlike in other European countries and the United States [4,5]. Given the complexity of LC surgery, it has become imperative to carry out an in-depth assessment of the current state of these surgical practices throughout France, in order to improve the quality of care and in particular the frequency of severe post-operative complications. This study aims to provide an overview of hospitals authorised to perform LC surgery, and to assess their performance on the basis of key outcomes. The specific objectives of this work are as follows. Firstly, we aim to describe the characteristics of all hospitals in France that perform LC surgery, including their public or private status, their distribution across the 13 French regions, and the characteristics of the patients they admit. Secondly, we aim to assess the surgical performance of these hospitals, focusing on the Clavien Dindo classification (grade>2) as the main outcome indicator, including severe post-operative complications and 30-day in-hospital mortality.

2. Materials and Methods

Database and Inclusion

This study utilised the French national hospital database from the Programme de Médicalisation des Systèmes d’Information (PMSI), which includes discharge abstracts for all inpatient admissions to both public and private hospitals across France. Diagnoses recorded during the hospital stay are classified according to the 10th edition of the International Classification of Diseases (ICD-10) [6,7], while medical and surgical procedures performed during hospitalisation are coded according to the French Common Classification of Medical Procedures (CCAM). We included patients from this database who underwent pulmonary resection for LC in France between 2019 and 2023. Specifically, we selected patients with a primary diagnosis of LC (ICD-10 codes C34), along with a corresponding surgical procedure for LC (CCAM codes) performed during the same hospital admission [8,9].

Patient Characteristics

For each patient, we collected data on age, gender, and surgery-related factors including the type of surgical approach (thoracotomy, video assisted thoracic surgery (VATS) or robot-assisted surgery) and the nature of the resection (limited resection, lobectomy, bilobectomy or pneumonectomy). Additionally, we considered comorbidities such as various diseases (pulmonary, cardiovascular, peripheral vascular, liver, neurological, kidney, hematologic, metabolic, infectious), but also cerebrovascular events, anemia and other therapies (preoperative chemotherapy and steroids). We also calculated a modified Charlson Comorbidity Index (CCI) to assess the overall comorbidity burden [10].

Region and Hospital Characteristics

In France, hospitals are classified into several categories, including academic (teaching) hospital, non-academic hospital, non-profit private hospital, or private hospital. For each hospital, we also determined the number of pulmonary resection performed per year.

The hospitals were located in one of the 13 regions in metropolitan France: Auvergne-Rhones-Alpes (ARA), Bourgogne-Franche-Comté (BFC), Bretagne (BRE), Centre Val-de-Loire (CVL), Corse (COR), Grand-Est (GE), Hauts-de-France (HdF), Ile-de-France (IdF), Normandie (NOR), Nouvelle Aquitaine (NA), Occitanie (OC), Pays de la Loire (PdL), and Provence-Alpes-Côte d’Azur (PACA).

Outcome (Quality Indicator)

Our main outcome was based on the Clavien-Dindo classification [11], which was transformed into a binary variable. The variable was equal to 1 if the Clavien-Dindo classification was higher than 2. This grade > 2 rating included severe post-operative complications and 30-day in-hospital mortality.

Severe post-operative complications were defined as the presence of one or more of the following postoperative conditions: pain, parietal complications, tracheostomy, reintubation, adult respiratory distress syndrome, bronchopleural fistula, empyema respiratory failure, arrhythmia, malnutrition, phlebitis, pleural effusion, pulmonary embolism, pneumonia, bleeding requiring re-operation, myocardial infarction, stroke, ischemia of the lower limbs, septicemia, and heart failure.

Thirty-day in-hospital mortality was defined as all deaths occurring during the same hospital stay as the operation or within 30 days of the operation.

Hereafter, we will refer to this outcome (Clavien-Dindo>2) as severe complication.

Statistical Analyses

The number of annual procedures for each establishment was categorised into 3 quantiles: < 100, 101 to 250 and > 250 procedures/year.

To estimate the standardised outcome rate (Clavien-Dindo>2), we used an indirect method by including the following variables in the logistic regression model: age, sex, comorbidities, Charlson score, type of pulmonary resection, surgical approach and year. To describe regional variations in the number of procedures and the standardised outcome rate, we used the median, the 1st quartile and the 3rd quartile. To quantify variations within regions, we used the ratio of the 90th to the 10th decile of the standardised outcome rate of the hospitals. We used a hierarchical logistic regression model to estimate the adjusted odds ratio (aOR) for each of the three classes of number of annual procedures and type of establishment, with their 95% confidence interval. We also included patient characteristics, type of procedure and year in the model. These models are used to estimate unexplained heterogeneity between regions, indicated by the inter-regional variance [12,13].

Finally, we used the results of this modeling to see how the standardised rate estimate might evolve after simulating a new organisation of hospitals authorised to perform this surgery. We used the estimated standardised rate using the logistic regression model explained above to obtain the number of events expected in each region if only those centers performing more than 100 procedures/year were selected. Finally, the number of avoidable severe complications or deaths was calculated as the difference between the observed number of complications and the expected number of events.

The calculations were carried out using STATA V.18 statistical software (StataCorp, College Station, Texas, USA).

Ethics, Patient and Public Involvement

Patients and the public were not involved in this study. As this was a national retrospective analysis using pseudonymised data (no personally identifiable information), patient consent was not applicable. The French national hospital database does not contain any identifying details: patient identities are pseudonymised, allowing data linkage for the same individual without knowing their identity. Ethical approval for the use of this database was granted by the French National Commission for Data Protection. (declaration of compliance with reference methodology 05 obtained on 7/08/2018 under number 2204633 v0) and this study thus adhered to the tenets of the declaration of Helsinki

3. Results

From 2019 to 2023, 64,304 patients underwent lung resection for cancer. The number of surgeries increased over time (2019: 12,367 patients, 2023: 14,227 patients, Table 1). A total of 18,151 patients (28%) had a severe complication (Clavien-Dindo>2), which was used as a quality indicator. Over the study period, we observed a decrease in severe post-operative complications, as well as in mortality (Clavien-Dindo=5), which fell from 2.3% in 2019 to 1.5% in 2023 (Table 1).

Regions and Characteristics of Hospitals

A total of 171 hospitals were performing lung resections for cancer during the study period. The median number of procedures/year in France was 53, with interquartile ranges of 16 and 101. By region and according to the size of the region, the number of hospitals ranged from 2 in Corsica to 24 in Rhône-Alpes-Auvergne (Table 2). However, the number of hospitals is not proportional to the size of the population in the region. For example, 24 hospitals performed lung resections in Rhône-Alpes-Auvergne, whereas only 19 hospitals were counted in the more populous Ile-de-France region (Table 2). The variation in the number of annual procedures performed by hospitals in the different metropolitan regions is shown in Table 2. The medians range from 20 for Corsica to 121 for Grand-Est. In most regions, 25% of hospitals carry out fewer than 20 procedures/year.

Standardised Rate of Severe Complications (Clavien-Dindo>2) in the Current Situation: Estimation and Modelling

Supplement Figure S1 shows the distribution of the median standardised complication rates for hospitals in the different regions, with some regions exceeding the national rate of 28%, such as Centre-Val-de-Loire, Corsica, Grand-Est and Ile-de-France. The 90/10 interdecile ratios for the regions are shown in Figure 1.

This ratio is greater than 3, meaning severe complication rates up to three times higher between hospitals, for the following regions: Normandy, Centre-Val-de-Loire, Pays de la Loire and Rhône-Alpes-Auvergne. For the other regions, the inequalities are less significant, with ratios ranging from 2 to 1.55. The classification of hospitals into 4 quantiles of standardised complication rates is shown in Table 3. In the regions of Centre Val-de-Loire, Corsica, Grand-Est, Ile-de-France, Nouvelle Aquitaine and Provence-Alpes-Cote d'Azur, most hospitals fell into the classes with a high rate of severe complications. In the other regions, the distribution was fairly balanced between the 4 classes, except for Pays de la Loire where most hospitals were in the 1st class with a low rate of severe complications.

The number of annual procedures, transformed into 3 classes, was included in the hierarchical logistic regression model used to estimate inter-regional variance (Table 4). Teams from hospitals performing more than 100 lung resections per year reduced the risk of having a severe complication or post-operative death by 20% compared with hospitals performing less than 100 procedures/year (aOR=0.83[ 0.77-0.89] for hospitals performing between 101 and 250 lung resections per year and AOR=0.85[0.77-0.93] for hospitals performing more than 250 procedures/year). Private not-for-profit and private for-profit hospitals had respective aORs of 1.35[1.19-1.52] and 1.10[1.01- 1.19] compared with non-academic hospitals (Table 4). The median OR was 1.132, showing moderate heterogeneity between regions, taking into account the same patient and hospital characteristics.

Estimate of the Standardised Rate of Severe Complications (Clavien-Dindo>2) After Simulation of a Reorganisation Of Hospitals

Following the results of our model, we simulated a reorganisation of the hospitals, taking 100 procedures/year as a threshold value to be authorised to perform LC surgery. If this were the case, only 44 hospitals would be performing lung resections for cancer. This hypothetical organisation would have made it possible to avoid 477 severe complications or deaths.

4. Discussion

This study, conducted from 2019 to 2023, included all French hospitals (171 hospitals) performing lung cancer surgery hospitals. While the differences in severe complication (Clavien-Dindo>2) rates between regions are minor, the variations between hospitals within those regions are significant with complication rates up to three times higher between hospitals. We identified that the primary factor explaining this variation is low surgical volume and that the results are much better for hospitals carrying out more than 100 procedures/year. After observing that many hospitals had a low surgical volume (in most regions, 25% of hospitals perform fewer than 20 procedures/year), we simulated a new organisation in which a threshold of 100 procedures/year would be required to be authorised to perform this surgery. This simulated organisation, with a number of authorised hospitals reduced to 44, enabled us to estimate that 477 severe post-operative complications or deaths could have been avoided over the period 2019-2023.

The limitations of our study primarily relate to the use of ICD-10 codes for patient selection and outcome assessment. Misclassification or underreporting of biases, especially regarding comorbidities, is a potential limitation of this study. Coding practices may differ between hospitals, as diagnoses can be recorded by either clinicians or information system technicians. However, the quality of coding is routinely audited in a standardized manner by medical information professionals at each hospital to ensure the accuracy of diagnoses and enhance the documentation of comorbidities.

It is also conceivable that there could be significant differences between hospitals regarding the severity of illness, which could impact the analyses of mortality and severe postoperative complications. Nevertheless, we used a multilevel analysis model to account for this variability between hospitals. In addition, over-reporting of milder forms of complications would not have a significant impact on our results since we focused on severe complications.

Another limitation concerns the tumor, node, metastasis (TNM) stage, which can influence mortality but cannot be recorded in the PMSI. However, metastatic stages were excluded from the analyses. Additionally, it is the type of lung resection that influences 30-day mortality and severe postoperative complications, independent of TNM stage [9].

Moreover, there is a lack of certain variables, including the American Society of Anesthesiologists (ASA) score, ECOG performance status, smoking history and forced expiratory volume in one second (FEV1). However, we accounted for pre-existing pulmonary disease, which usually correlates with impaired FEV1. We also utilised the modified CCI, which has been validated as a preoperative risk score. The body mass index (BMI) is not included in our database, but we included obesity in the metabolic disease variable.

The PMSI database did not provide information on the number of beds in the surgical and intensive care unit, nurses, medical practitioners and surgeries performed by residents. This database did not contain data on the organisation of the surgical team, such as tumor board meetings, adherence to guidelines and surgeon experience. However, these factors could potentially impact the quality of care. Patient quality of life is another essential indicator that cannot be measured by the PMSI.

One of the strength of our findings is the large sample size comprising 64,304 patients. Our national administrative database is an invaluable resource for evaluating the quality of care, as it collects detailed patient information from all hospitals in France, with nationwide recruitment over a five-year period. It offers a comprehensive source of epidemiological data on hospitalised patients in France [14,15]. Furthermore, the data related to pulmonary resection for LC are sufficiently reliable, allowing for precise identification of these patients, as demonstrated in previous studies [9].

The aim of this study was to assess the results of LC surgery, showing the variability in quality of care between hospitals. Our results demonstrate that there is a need to change how this type of surgery is currently offered in France. In particular, centralisation would be an appropriate means to ensure that hospitals performing LC surgery to have sufficient volumes of activity. This type of study has not been carried out in France so far [16,17,18,19].

According to a literature review previously published by our team, the standardised 30-day mortality rate for LC surgery is higher in France than in most other European countries. In light of our findings, it is worth considering whether these differences can be partly explained by the dispersion of surgical teams in France, some of which have low annual case volumes. For instance, when we compare England and France, which have similar populations, only 27 hospitals in England have been authorised to perform lung resections versus 146 in France [20].

Many countries have already recognised the need to consolidate the hospitals performing these surgeries, a process known as regionalisation. For example, Denmark, Sweden, Finland, England, Norway, Portugal, and Ireland have implemented regionalisation of surgical facilities for complex operations. Other countries, like Austria, Germany, Switzerland, the Netherlands, and Belgium, have opted to increase the minimum threshold required to obtain authorisation.

In all these countries, the current minimum threshold is higher than in France [21,22,23], and even if the authorisation threshold for cancer surgery has recently been raised from 30 to 40 procedures/year [3], this threshold remains lower.

This study appears to confirm the detrimental effect of the dispersion of LC surgery on quality of care, especially since this dispersion is not evenly distributed.

The principle of consolidating hospitals to provide high-performing surgical facilities that benefit patients is supported by numerous publications, which have shown that the volume of surgical activity correlates with the quality of care. Furthermore, it has been shown that the dispersion of hospitals creates inequalities in patient access to innovative technologies, such as robotic surgery. For instance, teams that perform few surgical procedures struggle to implement robotic surgery programs. These programs require surgeons to undergo adequate training, which is difficult to achieve effectively if the surgical activity is too low.

We believe that the consolidation of surgical facilities can only be successful if it is carried out on a regional scale, in consultation with relevant stakeholders, and taking into account the specific needs of the population, particularly their socio-demographic characteristics. The designated hospital should include high-performing surgical facilities, both in terms of operating rooms and intensive care units, with access to innovative technologies. Another crucial criterion is the competence of the medical and surgical teams, as well as multidisciplinary collaboration. Consolidation could make selected hospitals more attractive to practitioners, offering them the opportunity to train and work within competent multidisciplinary teams.

5. Conclusions

This study shows that there are differences in practice between hospitals in each French region. However, variations in the standardised complication rate between regions remain moderate. The significant influence of the volume of activity on the occurrence of severe post-operative complications and mortality raises questions about the need to restructure the offer of care for complex surgeries such as LC surgery.

While the authorisation threshold for cancer surgery was recently raised from 30 to 40 procedures/year, this work suggests that to improve the results of LC surgery, this threshold should be raised to at least 100 procedures/year, if France wants to catch up with other European countries in terms of quality of results.