Real-life effectiveness of first-line anticancer treatments in stage IIIB/IV NSCLC patients: Data from the Czech TULUNG Registry

Introduction

Treatment of advanced-stage (IIIB/IV) non-small cell lung cancer (NSCLC) has undergone notable evolution during the last 20 years. Antifolates, antiangiogenics, tyrosine kinase inhibitors (TKIs), anaplastic lymphoma kinase (ALK) inhibitors and immunotherapy have been introduced, with patients gaining a clear advance in quality of life, progression-free survival (PFS) and overall survival (OS).¹ In our previous study using data from the TULUNG Registry, we demonstrated that probability of two-year survival of stage IIIB/IV patients with NSCLC doubled between 2011–12 and 2015–16.² This observation may be explained by the introduction of several classes of new drugs in recent decades, together with the opportunity to introduce treatment multilinearity/sequencing in individual patients.² Importantly, a positive response in patients after first-line anticancer treatment appears to be the strongest predictor to continue therapy to subsequent lines, eventually resulting in longer overall patient survival.^{3, 4}

Real-life effectiveness of anticancer drugs (as well as of any other drugs) used in clinical practice should be routinely evaluated as it may significantly differ from drug efficacy observed in clinical trials. This is very important, especially in the advanced stages of lung cancer where any treatment failure may lead to rapid disease progression and early death. In a Dutch study by Cramer-van der Welle et al., real-life effectiveness of first-line chemotherapeutic regimens and gefinib and erlotinib was assessed.⁵ The authors concluded that a significant gap between clinical efficacy and real-life effectiveness existed with EE factors (efficacy/effectiveness) ranging from 0.52 to 0.87 for various treatment regimens.⁵ However, in this study, the reported N for gefitinib and erlotinib reached only 35 and 24 patients, respectively, leaving the validity of the results for TKIs disputable.⁵ In a different study, the authors estimated that the RCTs overestimate OS and PFS by ca 6% and 18%, respectively.⁶

Another issue is that real-life patients frequently have various comorbidities, unstable brain metastases, polypharmacy or may be older, resulting in noninclusion in clinical trials. This may leave the majority of patients and their caregivers with uncertainity about the effectiveness of newly started treatments. For example, the study by Halpin et al. showed that approximately 27% of patients with chronic obstructive pulmonary disease (COPD) were found eligible to participate in a clinical trial on COPD treatment.⁷ For patients with NSCLC, Kawachi and colleagues found that around 60% of patients were ineligible for a clinical trial, mainly due to brian metastases, poor performance status or a respiratory disease.⁸

Exploiting the opportunities given by the existence of the TULUNG Registry, in this study we aimed to analyze real-life effectiveness of TKIs, bevacizumab and pemetrexed as first-line anticancer treatments in patients with stage IIIB/IV NSCLC. We hypothesized that real-life effectiveness of first-line treatments may be different (probably lower) than efficacy reported in previous clinical trials.

Methods

Statistical analysis

Each patient group with a specific treatment regimen was considered a separate subgroup for statistical analysis. Basic characteristics of each specific treatment cohort are described by descriptive statistics. Continuous parameters are described by means and 95% confidence intervals (CI) and by median with minimum and maximum values. Categorical parameters are presented as absolute and relative frequencies. Relative frequencies were calculated based on numbers of patients in the relevant subgroups.

Overall survival (OS) is defined as the time from initiation of first-line anticancer treatment to the date of death (due to any cause). Progression-free survival (PFS) is defined as the time from initiation of first-line anticancer treatment to the date of first documented progression or death due to any cause. Treatment responses were assessed using the RECIST 1.1 criteria.¹⁰ OS and PFS were estimated using Kaplan-Meier method and all the point estimates include a 95% CI.

The pooled estimates of median PFS and median OS from clinical trials were calculated as a weighted estimate of population medians, using this formula¹¹:

where m_i represents the median survival within each study population (i = 1,2,…k), w_i refers to the weight of each study and is equivalent to the sample size of each study divided by the total sample size.

A new term – “index of real-life effectiveness” (IRE) was proposed for a calculated ratio of real-life OS (or PFS) of an anticancer drug/drug regimen to its comparator pooled OS (or PFS) calculated from clinical trials.

Our results of the IRE analyses showed that a significant proportion of long-term survivors existed in most of the studied treatment subgroups. We also decided to analyze the factors associated with exceptionally longer OS. For this purpose, data of patients surviving ≥110% of calculated reference OS (from clinical trials) were used, cutoffs for entering regression analyses were set as follows: afatinib ≥26.2 months (calculated OS from trials was 24.7 months), bevacizumab ≥14.7 months (calculated OS from trials was 13.4 months), bevacizumab maintenance ≥14.6 months (calculated OS from trials was 13.3 months), erlotinib ≥24.9 months (calculated OS from trials was 22.7 months), gefitinib ≥29.3 months (calculated OS from trials was 26.6 months), pemetrexed ≥11.2 months (calculated OS from trials was 10.2 months) and pemetrexed maintenance ≥14.9 months (calculated OS from trials was 13.5 months). Univariate and multivariate logistic regression models were constructed for all types of treatment regimens separately and for these variables: age, sex, ECOG PS, clinical stage at treatment initiation, adverse events and presence of comorbidities.

Statistical analyses were performed using IBM SPSS, Statistics (version 25.0) and R software (version 3.5.1).

Results

Data of 2157 patients from the TULUNG Registry fulfilled the inclusion criteria and were eligible for further statistical analyses. Of these, 62 patients were treated with erlotinib, 147 with afatinib, 325 with gefitinib, 466 with bevacizumab (186 patients continued on bevacizumab maintenance), and 1157 with pemetrexed (255 patients continued on pemetrexed maintenance). Basic characteristics of each treatment subgroup are presented in Table 1.

Survival analysis from the TULUNG cohort showed the following results for the studied treatment subgroups (displayed as median): erlotinib – PFS 12.2 months and OS 23 months, afatinib – PFS 13.1 months and OS 29.3 months, gefitinib – PFS 10.4 months and OS 19.6 months, pemetrexed – PFS 4.9 months and OS 12.2 months, pemetrexed maintenance – PFS 5.5 months and OS 17.5 months, bevacizumab – PFS 5.6 months and OS 15.8 months, bevacizumab maintenance – PFS 4.5 months and OS 15.8 months. Complete survival data including 95% confidence intervals are presented in Tables 2 and 3.

Calculated (pooled) OS and PFS from clinical trials for each treatment regimen are presented in Tables 2 and 3. Calculated IREs for particular treatment regimens were: erlotinib – 1.123 for PFS and 1.013 for OS, afatinib – 1.189 for PFS and 1.184 for OS, gefitinib – 1.314 for PFS and 0.736 for OS, pemetrexed – 0.972 for PFS and 1.188 for OS, pemetrexed maintenance – 1.252 for PFS and 1.294 for OS, bevacizumab – 0.676 for PFS and 1.178 for OS, and bevacizumab maintenance – 0.99 for PFS and 1.189 for OS (Tables 2 and 3).

Multivariate logistic regression models showed these factors independently and significantly associated with longer OS (110% of pooled reference OS from clinical trials for each regimen separately): lower PS for afatinib; lower PS, absence of adverse events and female sex for bevacizumab; lower PS and female sex for pemetrexed. Complete results of univariate and multivariate regression models are presented in Table 4.

Discussion

We present the results of one of the largest studies on real-life effectiveness of modern-era first-line anticancer treatments in patients with stage IIIB/IV NSCLC. The study cohort is robust (2157 patients) and reflects the therapeutic impact of various treatment regimens in an era that preceded the routine clinical use of immunotherapy, ALK inhibitors and combined multiclass drug regimens. Our data analysis ended with patients who were being treated with first-line treatment initiated up to 30 June 2018. Since then, several newer drugs including ALK inhibitors: crizotinib, alectinib; TKIs: osimertinib; immunotherapy: nivolumab, pembrolizumab, and durvalumab maintenance have been approved for clinical use in Czechia. With regard to this study, the number of patients in these new groups were small and treatment duration was too short for a meaningful survival analysis.

For comparison of real-life effectiveness of an anticancer drug (or a drug regimen) with its efficacy known from clinical trials, we proposed the term “index of real-life effectiveness (IRE)”, a ratio of real-life OS (or PFS) and a pooled OS (or PFS) calculated from clinical trials. IRE represents a simple mode to express how treatment efficacy from clinical trials translates into daily clinical practice. Obviously, this parameter may reach different values depending on the study population and period. In turn, the values of IRE (particularly those lower than 1.0) may be a very important feedback for clinicians taking care of lung cancer patients.

A similar attempt to express real-life effectiveness was done by Dutch authors from the Santeon group.⁵ In their study, a pooled efficacy/effectiveness factor of 0.77 for all the studied treatments on a cohort of 1122 metastatic-stage NSCLC patients was observed.⁵ Cramer-van der Welle and her colleagues assumed that an approximate 25% lower effectiveness of anticancer treatments (compared to trials) should be expected during clinical decision-making. The authors of that study concluded that the reduced effectiveness is a constant pattern irrespective of type of treatment regimen. These observations are, however, in contrast with our findings where real-life effectiveness for OS of most of the studied regimens turned out to be even higher (ie, with IRE >1) than comparator efficacy from trials. This is clear evidence that parameters such as the IRE (or EE factor) perform unequally in different populations and the results and conclusions of the Dutch study cannot be extrapolated to all countries and their healthcare systems. Moreover, IRE or EE factor is not rigid and may be subject to changes over time. Clinical decision-making with the aid of similar parameters should be performed with caution and respect to local healthcare systems.

We propose several explanations for the discrepancy between our findings and the observations made by the Dutch authors in the Cramer-van der Welle study. First, in the Dutch study, some of the specific treatment subgroups consisted of only a few patients (ie, n = 35 for gefitinib and n = 24 for the erlotinib subgroups). Second, differences may exist between the Czech and the Dutch health care systems regarding care for lung cancer patients. Third, there is a liberal attitude towards euthanasia in the Netherlands. Deaths from lung cancer annually account for approximately 10–11 000 of all deaths in the Netherlands.¹² Of these, ~10% have been reported to be attributable to euthanasia.¹³ Replacement of further treatment lines or best supportive care after disease progression by the act of euthanasia may result in shorter survival. Importantly, in their study, Pardon et al. found that of the 13 lung cancer patients that first requested euthanasia but finally did not end up with it, eight patients subsequently had “alleviated symptoms”.¹⁴ We speculate that the euthanasia issue might also bias the survival results in the Cramer-van der Welle study (median OS for the complete cohort was only 8.02 months).⁵ This issue has not been specifically addressed by the authors of that study.⁵ In contrast, our results showed a pooled median OS of 15.3 months.

Another potential contributor to better survival in our study was the increased chance of treatment multilinearity/sequencing. In our previous study from the TULUNG Registry, treatment multilinearity was more frequent in recent years (1.54 line per patient in 2015–16 compared to 1.48 line in 2011–12) and associated with longer OS in patients with lung adenocarcinoma.² The effect was more prominent with increasing number of treatment lines.² Better survival outcomes in more recent patients (compared to previous years) were also observed in a Swedish study.¹⁵ Importantly, three studies showed that a success within first-line therapy also positively affected outcomes in subsequent treatment lines.^{3, 4, 16}

Both bevacizumab- and pemetrexed-based regimens had good clinical effect on OS (IREs were ≥1.17) in our patients not harboring any clinically relevant mutation. Both these regimens present effective tools for improving survival measures.^17-30 In an Italian real-life study performed on only 62 NSCLC patients, median OS for first-line bevacizumab maintenance treatment was 10.5 months.³¹ Median PFS was 7.4 months for bevacizumab-based therapy while 6.4 months for pemetrexed in a Taiwanese study.³² Another interesting finding was that patients completing maintenance-type regimens had higher IRE for OS than those receiving conventional regimens of treatment. These data may simply reflect the proportion of well-responders in the cohort, in our case it was around 20%–24%.

In our EGFR-positive patients, afatinib was the most effective drug regarding effect on PFS and OS. This is partly due to treatment restriction in the first years of its use (afatinib was approved only for patients with ECOG PS 0-1) and younger age of patients. Our data showed that gefinitib had lowest IRE for OS (0.736) of the three studied TKIs, while IRE for PFS was 1.314. The main reason for this discrepancy might be that gefitinib has a more favorable safety profile and is given to fragile, older patients with NSCLC.³³ In our cohort, patients treated by gefitinib were oldest (mean age: 67.4 years) of all TKIs groups but had lowest overall adverse event rate (27.7%) compared to erlotinib and afatinib (43.6% vs. 42.9%). However, the frailty of these patients may result in lower rates of subsequent treatment lines and shorter OS. Our comparator OS was also calculated on more trials´ data; recently, longer OS for gefitinib was reported than previously.^34-36 The real-life OS for gefitinib was about similar in our (19.6 months) and in the Dutch study (21.19 months).⁵ In a real-life Taiwanese study, the observed PFS and OS were 11.9 and 26.9 months, respectively.³⁷ Comparable to our results, in a Californian study, TKIs had the most prominent effect on OS among all the studied treatments in stage IV nonsquamous NSCLC patients.³⁸

We observed certain discrepancies between IREs for OS and PFS in various studied regimens (IRE for PFS was <1 but IRE for OS was >1 for pemetrexed, bevacizumab and bevacizumab maintenance). We speculate that the PFSs might be subject to bias by different intervals between CT scans performed during real-life treatment and clinical trials. On the other hand, OS data are more precise and as such, should be considered an objective measure of treatment success in our cohort. In Czechia, approximately 80% of center care for lung cancer patients is concentrated in the hands of pneumologists/pneumooncologists. We speculate that a pulmonologist's concentrated and timely management of not only lung cancer, but also pulmonary comorbidities (COPD, pleural effusions, hemoptysis, endobronchial procedures, pulmonary embolism, pulmonary adverse events of treatments etc) and other supportive care may preserve good ECOG PS enabling the patient to enter subsequent lines of treatment, which will eventually result in longer OS. Although treatment selection and initiation should be guided by scientific evidence, in the real-life setting, selection of patients for certain treatment may slightly differ. Therefore, we cannot completely rule out selection bias in our OS results.

OS data were surprisingly good in our cohort. With special emphasis on long-term survivors, the results of the regression models showed constant patterns across all treatment subgroups: lower PS, lower disease stage, female sex, younger age and less comorbidity tended to be associated with longer OS (≥110% of calculated pooled OS from trials). However, most of these findings did not reach statistical significance due to low N in the specific subgroup. Of note, the presence of adverse events in bevacizumab-treated patients tended to result in a worse prognosis (OR 0.39; 95% CI: 0.17–0.84; P = 0.02), while for TKIs-based regimens, the presence of adverse events was associated with higher probability of longer OS. This observation was more pronounced in the erlotinib subgroup (OR 3.31; 95% CI: 0.73–16.84; P = 0.127). In accordance with our results, younger age, less advanced disease stage and less comorbidity were associated with a more favorable outcome in TKI-treated NSCLC patients in a Swedish observational study.¹⁵

Disclosure

The authors of this manuscript report no conflicts of interest with regard to this study.