A text-mining approach to study the real-world effectiveness and potentially fatal immune-related adverse events of PD-1 and PD-L1 inhibitors in older patients with stage III/IV non-small cell lung cancer

Lung cancer is the leading cause of cancer-related mortality worldwide, with 2.21 million new cases and approximately 1.8 million deaths reported globally in 2020 [1]. The poor prognosis of this disease is largely due to its often late diagnosis at advanced or metastatic stage. Incidence increases significantly with age, with a median age of 70 years old at diagnosis [2]. Non–small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases [3].

The introduction of immune checkpoint inhibitors (ICIs), in particular anti-programmed cell death-protein 1 (anti-PD-1, nivolumab and pembrolizumab) and anti-programmed death ligand-1 (anti-PD-L1, atezolizumab and durvalumab) antibodies, has revolutionised the management of NSCLC in the last decade. PD-(L)1 inhibitors, as monotherapy or in combination with chemotherapy, are currently indicated for the first-line treatment or as consolidation therapy of stage III and IV NSCLC.

Several randomised clinical trials (RCTs) and observational studies have shown improved patient outcomes in terms of progression-free survival (PFS) and overall survival (OS) compared to conventional cancer treatment options [4‐8]. Additionally, the safety profile of PD-(L)1 inhibitors appears to be more favourable compared to cytotoxic chemotherapy agents [9]. However, due to the activation of autoreactive T cells in a variety of host tissues, ICIs are associated with a considerable risk of immune-related adverse events (irAEs). These irAEs can affect any organ, but are mainly detected in colon, liver, lungs, pituitary, thyroid and skin [10]. Additionally, several reports have been made about uncommon, potentially fatal irAEs during the past years [11‐14]. In fact, ICIs have been associated with potentially fatal toxicities in 0.4% to 1.2% of the patients [15]. Due to the scarcity of these fatal irAEs, the exact estimation of their incidence is challenging. According to an analysis from the World Health Organization (WHO) pharmacovigilance database, colitis, pneumonitis, hepatitis, myocarditis, myositis, nephritis, myasthenia gravis, encephalitis and meningitis are among the most common fatal toxic effects associated with PD-(L)1 inhibitors [16]. All the aforementioned adverse events are listed as potential irADRs in the summary of product characteristics (SmPC) of PD-(L)1 inhibitors.

While older adults constitute the majority of NSCLC patients in clinical practice, our knowledge on effectiveness and safety of many novel treatment options in elderly patients has remained limited due to their underrepresentation in clinical trials. Despite the fact that over 60% of patients with NSCLC receiving ICIs in clinical practice are older than 65 years, this group only made up 8% of the clinical trial participants in the CheckMate 017 and 057 trials [4, 5] and 10–15% in the Keynote 042 and 024 trials [16, 17]. It is worth mentioning that the results of a small subgroup analysis (n = 29) in CheckMate 057 [5] were suggestive of a decreased efficacy of nivolumab in patients aged 75 years or older (HR: 1.76;95% CI = 0.77–4.05). Moreover, older adults who are enrolled in clinical trials are typically fitter than the general geriatric patient population in clinical practice. In addition, aging has been associated with structural and functional deterioration of the immune system, known as ‘immunosenescence’. Immunosenescence could alter the effectiveness and degree of toxicities of immunotherapy in elderly patients [18]. Although, traditionally, older people are defined as those aged 65 years or over, within pharmacoepidemiologic studies, there is no formal cut-off age determined to define the “older patients”. Previous effectiveness and safety studies of “elderly” or “geriatric” patients receiving immunotherapy have used age cut-offs ranging from 60 to 80 years [8, 19, 20]. This variation may reflect the fact that immunosenescence is a gradual process. Clearly, the arbitrary definitions of chronological age for studying older patients do not show a true picture of the reality. However, certain cut-off points have to be defined to be able to conduct a comparative study.

The disparity between the trial participants and patients receiving ICIs in clinical practice reflects a potential knowledge gap which can largely be bridged by utilizing real-world evidence (RWE). Electronic health record (EHR) platforms are rich sources of RWE as ample information on therapies, diagnoses, laboratory test results, physicians’ notes, et cetera are easily and reliably accessible. One key challenge with the EHR data is the utilization of the free-text notes.

Recently, the introduction of Natural Language Processing (NLP) and text mining techniques has eased the extraction of free-text data. The rule-based text-mining software (IQVIA Patient Finder Solution-CTcue B.V., Amsterdam, the Netherlands) is an NLP-based tool, which can be used to extract structured as well as unstructured information from the EHRs on, among others, comorbidities, cancer-related variables, clinical outcomes and adverse events. A validation study using the CTcue showed an accuracy of 88.1–100% for the retrieval of main treatment outcomes such as PFS and OS from the EHRs when compared to manual reviewing process [21].

The current study aimed to investigate the effect of chronological age on the real-world effectiveness and immune-related safety of PD-(L)1 inhibitors in stage III and IV NSCLC using text-mining techniques.

This real-world retrospective study investigated the effectiveness and safety of PD-(L)1 inhibitors in older patients with stage III and IV NSCLC using text mining in the EHR. To the best of our knowledge, this is one of the few studies that has utilized text-mining to collect real-world data, in real-time, of such a large cohort of patients. Median age at the start of immunotherapy was 66 years, which was higher than in pivotal studies [4, 5, 8]. Although, the median PFS and median OS of patients aged 65–74 were four and five months shorter than the youngest patient group respectively, no statistically significant association was observed between age and the effectiveness of anti-PD-(L)1 therapy. The median PFS and median OS were highly comparable between the youngest and the oldest patients. Similarly, older age did not negatively impact the safety of PD-(L)1 inhibitors. In fact, the cumulative incidence of potentially fatal irAEs and the relative risk of corticosteroid prescription to treat irAEs seemed to be lower in patients aged 75 and older.

Similar to our findings, subgroup analyses of pivotal trials and real-world studies have shown no association between age and clinical outcomes (i.e. PFS and OS) [23‐27]. However, most large studies have defined “older patients” as those aged ≥ 65 years or ≥ 70 years[28, 29], reducing the certainty of the findings in patients older than 75 years [26, 30]. A pooled analysis of phase III trials for pembrolizumab demonstrated that the median OS was comparable between patients older than 75 years and their younger counterparts, which is in line with the findings of this study [16]. However, other studies have shown either longer [31] or shorter [30, 32] PFS among patients ≥ 75 receiving PD-(L)1 inhibitors as a treatment for NSCLC. Most of these studies are performed in small cohorts and need to be further validated in larger cohorts. The proportion of patients with a ECOG PS ≥ 2 is generally lower in our study compared to the abovementioned studies; this trend is similar among all the different age groups.

The association between age and effectiveness outcomes remained statistically insignificant even when age was analyzed as a linear variable. Additionally, a higher proportion of elderly patients received PD-(L)1 inhibitors as first line treatment in our cohort. This group also had a higher PD-L1 expression. Although both of these factors could potentially improve response rates, subgroup analysis by treatment line and PD-(L)1 expression showed no significant difference in effectiveness outcomes for the older patients compared to younger patients.

With regards to safety of anti-PD-(L)1 antibodies, the most commonly occurring potentially fatal irAE in our study was pneumonitis (5.2%), which lies within the range of the incidence of any-grade pneumonitis in pivotal RCTs (1–10%) [4, 5, 8, 33]. The incidence of pneumonitis was highest among patients aged 65–74 years in this cohort. As for the other studied irAEs, we observed a higher incidence compared to the reported incidences in clinical trials [4, 5, 8, 34, 35]. This was to be expected as RCTs are generally not adequately powered to detect rare events. Our findings were consistent with the reported incidences in other real-world studies [35‐37].

Similar to the results of a number of retrospective studies, we found no difference in the incidence of potentially fatal irAEs among different age groups. There are studies suggesting a higher incidence of grade ≥ 2 [24] and ≥ 3 [37] irAEs in patients aged ≥ 70 years. However, the findings of a meta-analysis performed by the Food and Drug Administration (FDA) [38] on the safety of nivolumab in older patients suggested that certain irAEs (i.e. pneumonitis and hepatitis) are more common among younger patients (< 65 years), while other irAEs (i.e. colitis and nephritis) occur more frequently in older patients (≥ 65 years). In our study a decrease in systemic corticosteroid use was observed during immunotherapy in patients aged 75 and older, indirectly suggesting a possible lower incidence of moderate-severe irAEs in older patients. Alternatively, this finding could be the result of the reluctance of the physicians in prescribing corticosteroids for older patients owing to their unfavorable safety profile. This finding must therefore be interpreted with caution.

Generally speaking, older adults have a lower life expectancy. In other words, mortality rate among older patients must be inherently higher than in younger patients irrespective of the intervention that is being investigated. Although the magnitude of this effect cannot be determined in patients with advanced stages of cancer, one could argue that a similar survival regardless of age suggests a higher effectiveness of PD-(L)1 inhibitors in older patients. On the other hand, one possible explanation for the improved safety of anti PD-(L)1 antibodies, in terms of reduced corticosteroid use, in older patients is the reduced sensitivity of the immune system with increased age. However, this theory could only be true if PD-(L)1 inhibitors also showed a lower effectiveness in older patients, which contradicts our findings as well as the abovementioned theory. In addition, preclinical studies suggest that immunosenescence results in higher concentrations of inflammatory cytokines and autoantibodies which may paradoxically increase the risk of irAEs in the elderly.

The contradictions in our findings can best be interpreted in two ways. First, the disease itself is the primary determinant of life expectancy in stage III/IV NSCLC due to the low survival rates, thus making the pure effect of age negligible. Second, oncologists may be more cautious when prescribing immunotherapy to older patients as they may have concerns about comorbidities, reduced tolerance and potential immunotherapy-related toxicities, introducing an inherent selection bias in patient population. Therefore, it may seem plausible to think that the older patients in this study would be relatively fitter (with a longer life expectancy) than the general elderly population with stage III/IV NSCLC. Be that as it may, more older patients had a poor performance score (≥ 2) at the beginning of therapy compared to younger counterparts. We do appreciate that ECOG PS is not the only indicator of frailty and other, unmeasured, factors also play a role in determining it.

This study has several limitations. First, we included a heterogeneous patient population receiving different types of PD-(L)1 inhibitors in various treatment regiments. The effectiveness and toxicity profiles of ICIs may depend on the drug type and the concurrent use of other anti-cancer therapies. Due to the limited number of patients in each treatment subgroup it was not possible to perform stratified analysis with a high enough power. Consequently, a direct comparison between our findings and the outcomes of RCTs was also not possible. Nevertheless, our results are generally comparable to those previously reported in the literature and add valuable information on the effectiveness and toxicity of ICIs in older patients with NSCLC. Second, we used the need for corticosteroid treatment as an indirect measure of the toxicity profile of PD-(L)1 inhibitors. Although the majority of moderate/serious irAEs are commonly treated with corticosteroids, this is not always the case. In addition, patients receiving the corticosteroid dexamethasone for an antiemetic indication were not excluded from the analysis. This is only relevant for patients receiving chemo-immunotherapy with a moderate-high emetic risk regimen (e.g. cisplatin and carboplatin) as dexamethasone is not indicated for low emetic risk treatments such as PD-(L)1 inhibitors. The proportion of patients receiving chemo-immunotherapy is comparable between the three age groups. Furthermore, the incidence of TNF-alpha inhibitor use to treat irAEs could not be determined dur to the insignificant number of patients.

Other limitations include the retrospective nature, the limited data availability on factors such as the genetic markers and comorbidities, and the considerable rate of missing data for ECOG PS and BMI. In order to handle the missing data on the aforementioned factors, multiple imputation was used during analysis. It seems plausible to assume that less severe or absent outcomes are more likely to be overlooked and be left unrecorded in the EHR. The multiple imputation of the missing data could therefore potentially lead to bias and overestimate the number of cases with less favorable patient parameters.

In this study we investigated the impact of age on the effectiveness and irAEs of PD-(L)1 inhibitors in patients with NSCLC. Aging is often associated with frailty. However, it goes without saying that although frailty is common in older adults, it is certainly not always a manifestation of old age. Above chronological age, frailty depends on factors like comorbidity burden, physical status, unhealthy life style and polypharmacy. Studying frailty may therefore provide more valuable data. However, much of the data on frailty was either absent in patient files or not feasible to extract using CTcue.

In spite of these limitations, this study provides important “real-world” information about the impact of age on the effectiveness and the safety of ICIs in a very large cohort of patients with stage III and IV NSCLC. Additionally, the real-world data for this study was obtained using a novel text-mining method.

In this large cohort real-world study, we used a novel text-mining technique to investigate the effect of age on the long-term effectiveness and irAEs of PD-(L)1 inhibitors in patients with stage III and IV NSCLC. All patients, regardless of age, had similar effectiveness outcomes and comparable risk of moderate-severe and potentially fatal irAEs. These findings suggest that anti-PD-(L)1 antibodies could effectively and safely be prescribed to older patients with stage III/IV NSCLC.

The successful utilization of a text-mining tool to obtain all the data for this real-world study demonstrates the capacity of this efficient technique to extract valuable pieces of information from structured and unstructured fields of the EHR.

Future studies should be performed in larger patient populations in order to be able to investigate the effect of individual immunotherapy agents and treatment regimens with high power. Additionally, a better way to characterise “older” adults in such studies may be to investigate the degree of frailty (based on geriatric assessments) instead of chronological age, as frailty may be a limiting factor for prognosis of patients receiving immunotherapy.