Immune checkpoint inhibitors of the PD-1/PD-L1-axis in non-small cell lung cancer: promise, controversies and ambiguities in the novel treatment paradigm

Abstract

Immune checkpoint inhibitors (ICIs) have received much attention not least for melanoma since the award of the Nobel prize in 2018. Here, we review the current state of knowledge about the use of these monoclonal antibodies (mAbs) in non-small cell lung cancer (NSCLC). These drugs have generally been conditionally approved on limited early data and there are few long-term follow-up data from randomized clinical trials. The effect observed for NSCLC thus far is, on average, moderately better than that obtained with chemotherapy. Severe side-effects are more common than might have been expected. The drugs themselves are expensive and are associated with time-consuming histopathologic testing even though the predictive value of these tests can be discussed. In addition, monitoring for side-effects involves increased workload and budgetary expense for clinical chemistry laboratories. Here, we review and summarize the current knowledge, controversies and ambiguities of ICIs for the treatment of NSCLC.

Keywords:

• Non-small-cell lung carcinoma
• biomarkers
• immune checkpoint inhibition
• PD-1/PD-L1
• pharmaceutical economics

Background

The non-surgical, non-radiological, non-endocrinological treatment of cancer has developed by leaps and bounds. A number of increasingly powerful but also toxic chemotherapeutic agents like the taxanes and platinums have been available for more than 25 years. About 20 years ago, the monoclonal antibody (mAbs) rituximab and trastuzumab changed the treatment of lymphoma and breast cancer. Today a whole host of mAbs is on the market. The first immune checkpoint inhibitor (ICI) ipilimumab was approved in 2011 for metastatic melanoma and since then, the ICIs have opened up a new field of treatment options. Some of the expected benefits at that time, in addition to better survival, were less acute toxicity and no mutagenic side-effects as with chemotherapy.

The ICIs essentially fall into 2 classes: inhibitors of cytotoxic T-lymphocyte–associated protein 4 (CTLA-4) and inhibitors of the PD-1/PD-L1-axis, which in turn can target programmed cell death receptor 1 (PD-1) or programmed cell death ligand 1 (PD-L1) [1–4]. All ICIs work by preventing immune evasion by cancer cells. Ipilimumab is still the only CTLA-4-inhibitor approved. There now are 3 PD-1-inhibitors – pembrolizumab (Keytruda), nivolumab (Opdivo), and cemiplimab (Libtayo) as well as 3 PD-L1 inhibitors – atezolizumab (Tecentriq), avelumab (Bavencio) and durvalumab (Imfinzi). In all, they have been approved for 14 indications during the last five years. Melanoma has received most attention as this seems to be the disease in which the ICIs work the best and for which most ICIs are approved. The 4 PD-1/PD-L1-inhibitors that have been on the market the longest are summarized in Table 1. These four (pembrolizumab, nivolumab, atezolizumab and durvalumab) are all approved for non-small-cell lung cancer (NSCLC).

Table 1. Summary of the 4 most commonly used immune checkpoint inhibitors in non-small-cell lung cancer (NSCLC).

Molecular target of ICI	PD-1 inhibitors		PD-L1 inhibitors
Generic name	Pembrolizumab	Nivolumab	Durvalumab	Atezolizumab
Trade name	Keytruda	Opdivo	Imfinzi	Tecentriq
mAbs used to assess TPS	SP263 & 22C3	SP263 & 28-8	SP263	SP142
PD-L1 cells scored by ICH	TC	TC	TC	TC and/or IC
Approval for 1st line	Monotherapy: advanced, expressing PD-L1 with tumor proportion score (TPS) ≥50% with no EGFR or ALK genomic tumor aberrations Combination therapy: in metastatic squamous and non-squamous provided no EGFR or ALK genomic tumor aberrations	Not approved	Locally advanced Stage III; tumor expressing PD-L1 TPS ≥1% without progression after chemoradiotherapy	Monotherapy: not approved Combination therapy: All allowed in combination with various chemotherapies for metastatic non-squamous; not EGFR mutant or ALK positive
Approval for 2nd line	Monotherapy: metastatic, expressing PD-L1 TPS ≥ 1%; Patients with EGFR or ALK genomic tumor aberrations should first have had disease progression on approved therapy for these aberrations	Monotherapy: locally advanced or metastatic	Not approved	Monotherapy: locally advanced or metastatic; Patients with EGFR mutant or ALK positive NSCLC should also have received targeted therapies before Combination therapy: not mentioned
TPS required for 1st line therapy	≥50% for monotherapy, not required for combination with chemotherapy	Not approved	≥1%	Not approved
TPS required for 2nd line therapy	≥1%	None required	Not approved	None required
TPS investigated in trials	TC ≥1% and ≥50%	TC ≥1%, ≥5% and ≥10%	35% were TC ≥ 25%, 32% TPS 1–24% and 33% TC < 1%.	TC ≥50%; IC ≥10% TC or IC ≥ 5% TC or IC ≥ 1% TC and IC < 1%
Assay antibody clone	22C3	28-8	SP263	SP142
Test platform	Dako Autostainer Link 48		Ventana Benchmark Ultra

mAbs: monoclonal antibody; TPS: tumor proportion score; ICH: immune cytohistochemistry; TC: tumor cells; IC: immune cells.

NB. Approval descriptions based on EMA approvals. The exact approval can vary between the EU (EMA) and USA (FDA).

Lung cancer has a global mortality rate of 18.6/100,000 and this is by far the highest worldwide [5]. The pattern is similar in Sweden [6]. In this review, we focus on (NSCLC), which embraces about 80% of all lung cancer and includes both adenocarcinoma and squamous cell carcinoma. The 5-year age-standardized net survival in high-income countries in 2010–2014 was about 20%, almost double compared to the 1990s [7]. In the UK, it increased from 7% to almost 15%.

Effect of ICIs in NSCLC according to recently published studies

A major problem in assessing how well ICIs work in NSCLC is that there are precious few results from mature clinical trials. Nearly all data are from interim follow-ups, and the time period is often quite short. The exceptions are a 5-year follow-up that has recently been published for a study of pembrolizumab (Keynote-001) [8]. There have also been extended follow-up of Keynote-024 [9] and Keynote-042 studies (2 years) [10], and also a recent 2-year follow-up for nivolumab [11].

Keynote-001 [8] was a non-randomized, uncontrolled study of 101 treatment-naïve and 449 previously treated patients with advanced-stage IV NSCLC. This study formed the basis for the accelerated approval of pembrolizumab. Median follow-up was 60.6 months (51.8 − 77.9) during which 450 (82%) patients died. The median overall survival (OS) in treatment-naive patients was 22.3 months and 5-year OS was 23.2%, while among the previously treated it was 10.5 months and 15.5%, respectively. Progression-free survival (PFS) was longer the higher the expression of PD-L1. This cohort made up of heterogeneous patients is the only study in NSCLC with such long follow-up.

During 2019, two other updates of studies with pembrolizumab were presented, the so-called Keynote-024 and -042 studies [9,10]. Keynote-024 compared pembrolizumab with platinum doublet chemotherapy in first line (treatment-naïve) metastatic NSCLC with PD-L1 tumor proportional score (TPS) ≥50%. In the up-dated follow up (median 25 months) [9], the median OS was 30.0 for pembrolizumab vs. 14.2 months with chemotherapy (HR 0.63; 95% CI 0.47–0.86). The HR rose to 0.49 after adjustment for those given chemotherapy that on progression had crossed over to receive pembrolizumab.

Keynote-042 also studied first-line patients but with TPS ≥1% and stratified between TPS 1 − 49% vs. ≥50%, histology (squamous vs. non-squamous), performance status and geography [10]. Keynote-042 did not allow cross-over to pembrolizumab. The OS and HR were significantly higher in three TPS groups analyzed – ≥50%, ≥20% and ≥1% – but the overall effect was driven by the ≥50% group. However, when one looks at the Kaplan–Meier curves for OS early on the chemotherapy arm performs better and then they cross – at ca 6.5 months for ≥50%, ca 10 months for 1−49%. There was, in fact, no improvement in OS for the 1–49% group, whose HR for OS was 0.92 (CI 0.77–1.11) and median OS were 13.4 (10.7–18.2) vs. 12.1 (11.0–14.0) months. Another curious observation was that there was no clinical benefit in never-smokers (EGFR- and ALK-mutations were exclusion criteria).

The success of pembrolizumab is in contrast to nivolumab, which in its study in first-line vs. platinum doublet (study CM-026) [12] did not show any advantage for nivolumab. Thus, one PD1-inhibtor worked and the other did not. The reason for nivolumab’s failure has been dissected in retrospective (post-hoc) analyses one of which reported that patients with high tumor mutation burden benefitted significantly from nivolumab independent of PD-L1 status [13]. Another reason given was that CM-026 enrolled patients with TPS ≥5% but Keynote-042 allowed ≥1%, even lower than CM-026. CM-026 allowed cross-over from chemotherapy to nivolumab (but so did Keynote-024, but not -042).

Another noteworthy recent study on pembrolizumab is a retrospective analysis of first-line treatment in 187 patients with NSCLC [14]. It found that patients with 90–100% expression of PD-L1 do better than those with 50–89% expression. Median OS was 15.9 months for the 50–89% group (n = 107) but it had not been reached for the 90–100% group (n = 80) (p = .002). Thus, this indicates that pembrolizumab monotherapy may be preferable to chemotherapy-based regimens in patients with tumors showing a very high expression of PD-L1. The authors note that there was a similar finding in a study of avelumab, a PD-L1-inhibitor, compared to docetaxel in previously treated NSCLC, that is the clinical response to ICI seemed to improve with higher PD-L1 cut-off ≥1%, ≥50% and ≥80%, but in the end, there was no benefit of OS [15].

Nivolumab has been investigated in an interesting study adopting a complex design. A preliminary analysis based on tumor mutational burden (TMB) was reported in 2018 [13], and a larger analysis with 2-year follow-up data was published in 2019 [11]. The main analysis concerns the combination of the CTLA4-inhibitor ipilimumab with nivolumab compared with chemotherapy but data on the single-agent nivolumab was mainly found in a 42-page appendix. The study population included patients with advanced-stage IV or recurrent NSCLC but they must not have received any previous treatment for this disease, and no relevant EGFR- or ALK-mutations must be present in their tumor specimen. The most interesting summary of the study results is found in an inconspicuous figure, which showed that patients with PD-L1 TPS <1% did better than patients TPS ≥50%, and much better than patients in the group 1–49%. This applied to all patients, and also those with high TMB. In the overall analysis for patients with TPS ≥1%, OS for nivolumab + ipilimumab was 17.1 vs. 14.9 months for chemotherapy (p = .007), a difference of 2.2 months (66 d). For those with TPS <1% it was 17.2 vs. 12.2 months but no p value was reported. This is in contrast to the results for pembrolizumab which also targets PD-1 (see above).

Outcome of ICIs in NSCLC according to recent systematic reviews

During 2019, so far three systematic reviews with meta-analyses and slightly different designs have been published [16–18] with a focus on ICIs with or without chemotherapy in first-line treatment of NSCLC. All three concluded that adding ICIs significantly improved the treatment of NSCLC but cautioned that longer follow-up and more mature data were required for firm conclusions.

Modeling efficacy of ICI treatment

In the absence of mature data, a few investigators have attempted to handle the situation by modeling based on their systematic review with meta-analysis. Three very recent publications were found and the authors did not restrict themselves to NSCLC.

Shen and Zhao [19] questioned the need for testing for expression of PD-L1 by immunohistochemistry (IHC) prior to ICI treatment. In their systematic review of published data from eight randomized controlled trials, they concluded that, based on the overall response rate, although 34% of PD-L1 positive tumors responded to ICI treatment, also 20% of PD-L1 negative tumors responded as well. Their threshold for PD-L1 positivity or negativity was that PD-L1 stained cell accounted for 1% of tumor cells, or tumor and immune cells (IC), assayed by IHC staining methods.

Haslam and Prasad used a different approach, previously adopted when they assessed the benefit of genetic treatments [20–22]. Statistics from the American Cancer Society were used to estimate the number of patients with advanced cancer who would be eligible for treatment with ICIs. The researchers then did a year-by-year comparison with the ICIs with their indications as approved by the FDA and from this modeling, they concluded that less than 13% of those treatable would benefit from ICIs, while almost 44% were eligible for the treatment. They also contended that the percentage of patients with modeled benefit had peaked in 2014 and then decreased. The conclusion was that more recent indications diluted the strength of any treatment benefit since then.

A French group performed an inventive meta-analysis with modeling of Phase 3 studies, first presented in 2018 and then published in 2019 [23]. Because there were no 5-year follow-up data, they defined ‘durable response’ as the percentage of patients with more than either 3 times the median PFS and/or 2 times the median OS. There were 11,640 patients distributed between 26 study arms treated with ICIs and 16 without. A statistically significant benefit in favor of ICI treatment was found for all tumors, including NSCLC, both in terms of PFS and OS. In their modeling based on OS, both in all indications and in NSCLC specifically, ICI treatment achieved 30% compared to 23% without ICI, while their comparison of the PD-1/PD-L1 class with CTLA-4 it was 31 vs. 29%.

It is interesting in this context to note that two groups [24,25] have independently concluded that so-called real world data (RWD) validate efficacy-based outcomes from RCTs. One group found that in the USA only about 15% of studies could be replicated using data from insurance claims and/or electronic health records [24].

Side-effects; how toxic are the ICI agents?

In general, about 10–20% of patients treated with ICIs get severe (grade 3 & 4) side-effects and even deaths (grade 5) have been described. CTLA-4 inhibitors appear to be associated with an increased rate of and more difficult side-effects than PD-1/PD-L1-inhibitors. It also seems that inhibitors of PD-L1 are slightly less toxic than PD-1. All bodily systems can be affected by a form of ‘inflammation’ (they are ICIs after all). An overall term, immune-related adverse events (irAE), has been coined. This involves the skin – mainly rash; the GI-tract – diarrhea and colitis; the thyroid – mostly hypo – but also hyperthyroidism; pneumonitis, a special problem in patients with NSCLC; type1 diabetes (T1D).

Pillai et al. [26] looked at 3284 patients treated with PD-1 inhibitors and 2460 who received a PD-L1 inhibitor (Table 2). They found a statistically significant difference only for pneumonitis (4% for PD-1 inhibitor vs. 2% for PD-L1 inhibitor; p = .01). Hypothyroidism (at 7 vs. 4%) almost achieved significant difference (p = .07). For other main items, they found general side-effects were more commonly associated with PD-L1 inhibitors (about 20 vs. 10%) but irAEs were more commonly related to PD-1 inhibitors. In a review of the literature on pneumonitis, Petri et al. [27] concluded that the incidence varied considerably, 3–4% was considered a reasonable average, but when it occurred it was severe in 30–60% of the cases.

Table 2. Side-effects from monotherapy with immune checkpoint inhibitors (ICI) of PD-1 vs. PD-L1 in NSCLC.

	ICI of PD-1 (n = 3284) (%)	ICI of PD-L1 (n = 2460) (%)	p Value
Objective response	19	19	.17
All AE	64	66	.8
Grade 3–5 toxicity	13	21	.15
Fatigue	19	21	.4
irAE	16	11	.07
Hypothyroidism (all grades)	6.7	4.2	.07
Pneumonitis (all grades)	4	2	.01

Grade 5 = death; AE: adverse event; irAE: immune-related AE [26].

In a summary of the more serious side-effects in the big study comparing nivolumab + ipilimumab vs. nivolumab vs. chemotherapy (1:1:1) [11], treatment-related side-effects were twice as common with the ICI combination and chemotherapy compared with nivolumab alone. Nevertheless, stopping treatment was more common with both the single and two ICI treatments as with chemotherapy (13 vs. 7 vs. 3%). The grade 3 and/or 4 toxicities were 32.8 and 36.0%, respectively.

Chen et al. [16] summarized that the incidence of toxicity for PD-1/PD-L1-inhibitors and chemotherapy was about the same for any grade AE, any high-grade AE and treatment-related deaths but treatment discontinuation was statistically significantly more common for the ICIs than chemotherapy (18.5 vs. 12.3%, p = .01). The irAEs were more common in the ICI group than in chemotherapy, with 9% hypothyroid, 5% hyperthyroid, 1.5% thyroiditis, 5.5% pneumonitis, 3% hepatitis, severe skin reactions and colitis 2% each, 1% nephritis, and finally, hypophysitis and diabetes 0.5% each. Tun et al. [18], who cite Chen et al., reported slightly higher figures for these irAEs but also specifically mentioned ‘rash’ at 25%.

Importantly, as far as we can tell, there are no population-based reports on side-effects (serious or more trivial) and no clear data on the length of hospitalization because of side-effects (hospitalization and prolonged hospitalization are part of the definition of SAE).

The irAEs are not trivial [11,16,18,26,27]. In many cases, not least in the case of pneumonitis [27], they may require hospitalization for a week and high-dose corticosteroids for 3 weeks, and subsequently lower doses.

Another issue is patient reticence to report side-effects to their oncologists as this may take them off ICI treatment, or difficulties in recognizing side-effects, as they may not be as well-known as for chemotherapy. The consequence may be that when side-effects get too severe to be ignored, they may also be more difficult to manage. This is much less likely to occur in a clinical trial. While tumor hyper-progression during immunotherapy has been described as a rare AE [28], an opposite aspect is the suggested association of side effects and better outcomes [29].

Demands on the clinical chemistry and diagnostic hematology services

This recommendation for clinical chemistry laboratory monitoring is not that much more comprehensively onerous than for standard chemotherapy. A check-up for thyroid dysfunction (TSH and free T4) and diabetes (glucose) must be done before start of treatment and then subsequently monitored every fortnight during treatment and follow-up. HbA1c is not part of such protocols, though some advices that HbA1c be measured as well, nor are plasma protein electrophoresis to screen for paraproteins. Obviously, in the case of major toxicities, which are more common than generally thought, extra monitoring will be necessary. Thus, ICI treatment will involve some increase in the workload for Clinical Chemistry Departments.

But the main burden will fall on the histopathology and/or molecular genetics laboratories. Precision medicine, including ICIs, increase their workload significantly, and thereby their expenses. They need to keep expensive reagents available as well as dedicated equipment. The mAbs used in assays are generally tied to certain analytical platforms. Not only should PD-L1 expression be assessed (even though some of the agents do not require this) but mutations in genes like EGFR and BRAF, and rearrangements in ALK and ROS1 must also be checked for.

Assessment of expression of PD-L1

IHC with monoclonal antibodies (mAbs) is currently used to assess the expression of PD-L1. There is an excellent up-to-date review of the status of these tests [3]. The mAbs included in commercial kits are SP263, 22C3, 28-8 and SP142. In many countries (incl. Sweden), commercial kits are primarily, if not exclusively used. E1L3N is a common non-commercial mAbs in lab developed tests (LDT) in many countries, which is cheaper than commercial kits but in the EU the new EU IVD-regulation will make LDT more difficult to use. mAbs SP263, 22C3 and 28-8 detect epitopes in the PD-L1 extracellular domain, while SP142 is directed against epitopes on the intracellular portion [3,30–32].

SP263 has been used with most ICIs and in NSCLC; it can be used for both pembrolizumab (Keytruda), nivolumab (Opdivo) and durvalumab (Imfinzi) (Table 1). SP142 has been used with atezolizumab (Tecentriq), which is also approved for NSCLC. 22C3 was used in the pembrolizumab registration studies, and 28-8 was used in the corresponding study for nivolumab.

A survey of 344 pathologists from 310 institutions in 64 countries found that the method of PD-L1 testing varied around the world [33]. Briefly, 22C3 was the most often used mAb in North-America (and also Middle-East and Africa) while SP263 dominated in Europe and the rest of the world. Nevertheless, many labs used several mAbs. This seemed to reflect the regulatory approval for various indications and also which automation (DAKO/Agilent or Ventana/Roche). The use of the LDT with E1L3N varies between regions.

SP142 gives a significantly lower signal at IHC than SP263, 22C3 and 28-8 but it is not known how to interpret this difference, while SP263, 22C3 and 28-8 are overall considered to result in similar staining [2–4,30,31,34–43], though not all observers agree [32]. There is also general agreement that there is poor concordance between the mAbs for PD-L1 expression on IC. Each commercial antibody is used with the platform of that company, i.e. Ventana (Roche, Basel, Switzerland) for Ventana’s Mabs SP263 and SP142 and Dako (Agilent, Santa Clara, CA) for Dako’s 22C3 and 22-8. The platform-specificity increases costs because it means that labs must run both platforms, if they are to cover all mAbs.

In an unusual study Munari et al. [44] investigated 198 NSCLC mainly early stage cases to compare 22C3 on its Dako platform with 22C3 on a Ventana platform and also with SP263 (Ventana) scored by 2 pathologists. In their hands, SP263 was the most reliable assay to interpret. They noted that the pathologist trained on 22C3 consistently scored more positive cases than the one trained on SP263 both at the 1 and 50% cut-offs. There was less difference between 22C3 run on the two platforms than between 22C3 and SP263. This is further supported by Ilie et al. [45] where the results for 223 C were almost identical for Dako and Ventana platforms. Hendry et al. [46] evaluated all 4 mAbs in 355 cases of NSCLC using 1 single pathologist trained on 22C3. Like others, they found lower scores with SP142 and poor concordance of ICs. They also found that SP263 scored more positive than 22C3 and 28-8, both at the 1 and 50% level, but concluded there was better concordance between 22C3 and 28-8. Brunnström et al. [35] had 7 pathologists look at 55 samples stained with all 4 mAbs and found a significantly better agreement between pathologists when using ≥50% as cutoff. They concluded that the concordance between the PD-L1 antibodies 22C3, 28-8 and SP263 was relatively good (for lung cancers) and any one of these assays was sufficient for decision on treatment with nivolumab, pembrolizumab and durvalumab, but that the scoring of the pathologist presents an intrinsic source of error that should be considered especially at low PD-L1 scores.

The Blueprint 1 study [38] is sometimes quoted as arguing against the interchangeability of testing mAbs. This study used commercial (i.e. not real world) samples and compared 22C3 ≥ 1% vs. 28-8 ≥ 1% vs. SP263 ≥ 25% while Blueprint 2 [42] made a real world comparison. In the Blueprint 2 study [42] by the same researchers, they state the ‘three PD-L1 assays (22C3, 28-8 and SP263) showed comparable analytical performances on TCs but SP142 stained fewer’, whereas IC staining was in poor concordance. In Blueprint 2, which compared the same 4 mAbs plus a new mAb 73-10 developed for avelumab, they reported highly comparable TC staining with 22C3, 28-8 and SP263, less with SP142, and more with 73-10; and poor correlation for IC. Some of these authors in an overview [32] nevertheless question the equivalence of the different diagnostic mAbs.

In countries where LDT with E1L3N are used, screening might be done with this antibody [31]. It should, however, be noted that in the USA, the FDA assignations of ‘companion’ test may result in US clinicians keeping faith with the mAb as assigned by the FDA, based on the clinical trials. In large practices, all mAbs may be used in the histopathology department anyway, in which they might as well be used as they were in the clinical trials that led to approval of the treatment. It is worth noting though, that in the survey [33], 25% of respondents from North-America and 15% from Central/South America outsourced their samples (no figures given for the EU or the ROW).

The variations in staining and its assessment are probably owed to a mixture of reasons, such as variation in the proportion of adenocarcinoma vs. squamous cell carcinoma, frequency of smokers, or ex-smokers among patients and other factors. Also, it seems easier to determine if there is no staining (<1%) or if staining is strong (≥50%) while the group of 1–49% staining seems harder to adjudicate, though studies disagree on this matter [47]. There may also be heterogeneity of the expression throughout a tumor, and a quite variable concordance in PD-L1 expression between matched biopsies and resected tumors has been observed [48–50].

Though there have been calls for TPS for PD-L1 to be reported in a continuous manner from 0 to 100% and let the requesting physician categorize the level, only some national pathology societies do this while many these days often report in the cut-off values ​​(<1%, 1–49%, ≥50%) [3,51,52].

Both in the USA and Europe, the clinically interesting values ​​are still 1% + and 50% + as there are requirements for pembrolizumab as monotherapy in second and first line, respectively. Atezolizumab and nivolumab may be used even in the case of negative PD-L1 (both in the second line, azetolizumab also in first line when in combination), but some (though not all) national guidelines advise only give nivolumab at PD-L1 1% +. There is also a consensus that all NSCLC should be PD-L1 tested. In Europe (but not the United States), there is also a requirement for PD-L1 ≥ 1% for durvalumab as ‘maintenance therapy’ after curative chemoradiotherapy. Thus, the requirement of the pathologist is to (at least) report <1%, 1–49% or ≥50%.

Sweden has a national consensus and quality document/care program that states that (if possible) we should report on the scale <1%, 1–4%, 5–9%, 10–24%, 25–49%, 50–74% and ≥75% [51]. Other countries/departments have different scales (e.g. Denmark recommends reporting exact percentages) but there is no international consensus and there are certainly many countries without national consensus.

Regarding inclusion in clinical trials, things have looked different. Most have not had a TPS requirement for PD-L1, but patients have been included regardless and then one has tried to obtain PD-L1 analysis (on current or archived biopsies or – to a lesser extent – resections) in as many cases as possible. Subsequently, PD-L1 levels have been analyzed in relation to treatment response/survival. For example, pembrolizumab in the first line was inclusion criterion PD-L1 ≥ 50% (Keynote-024) [53].

Health economics

Health economic evaluations have assessed the cost-effectiveness and budget impact of PD-L1-guided therapy for NSCLC in diverse country settings with mixed results [54–63]. Findings of treatment cost-effectiveness in Brazil, France, Hong Kong and the USA [54–56,59,60,62] have been questioned by opposing findings in Australia, China, Switzerland, the UK and the USA [57,58,61,63,64]. Analyses have been marked by high treatment costs. As evidenced in Table 3, in Sweden such costs have been calculated in excess of SEK 60,000 per month and have been associated with variable but positive quality-adjusted life-year (QALY) gains. Authorities have assessed the severity of NSCLC as very high thereby indicating a willingness to accept ICERs in excess of SEK 500,000, the generally accepted cut-off limit.

Table 3. Monthly treatment costs and cost-effectiveness of ICIs for NSCLC, Sweden 2019.

Treatment	Treatment cost (SEK, per month)	QALY gain	ICER (SEK/QALY)	Source
Atezolizumab	63,000	0.52	880,000	[73]
Durvalumab	96,000	1.09	857,440	[67]
Nivolumab	62,000	0.6	890,136	[74]
Pembrolizumab	66,320	1.15	745,585	[72]

QALY: quality-adjusted life-year; ICER: incremental cost-effectiveness ratio.

The modeling of the costs of PD-L1 testing has varied to reflect company preferences as well as the treatment indication(s) achieved. In the UK submission for pembrolizumab, PD-L1 testing was modeled with a unit cost of £40.50 (2019 SEK 546)¹ [65]. The analysis assumed that in order to identify one patient eligible for treatment, 8.39 patients would need to be tested thereby giving rise to a cost of £337.51 (2019 SEK 4566) per patient. In the submission for atezolizumab, PD-L1 testing costs were not included in the atezolizumab treatment arm due to the license achieved that included all patients irrespective of PD-L1 expression [66]. In Sweden, the cost of PD-L1 testing has been calculated in the durvalumab submission at SEK 2323 exercising modest impact on overall treatment cost-effectiveness [67]. Why the Swedish cost-estimate should be almost half the English one, is an interesting question that speaks to the transferability challenges of health economic analysis across national borders.

Existing economic evaluations of PD-L1-dependent treatment strategies have incorporated adverse event (AE) information from the underlying clinical trials with anemia, fatigue, nausea, thromboembolism, pneumonitis and other irAEs, and peripheral neuropathy comprising the most expensive event categories [65,66,68–71]. Reimbursement submissions in Sweden have indicated that the impact of such events may be marginal in the determination of treatment cost-effectiveness [67,72–74]. This finding contrasts with evidence that existing evaluations of antineoplastic drugs, and of NSCLC drugs in particular, tend to underestimate AE incidence and the associated costs and effects that may arise [75,76]. AE selection, the simultaneous experience of multiple AEs, dose modifications and the actual quality of life impact on NSCLC patients and their caregivers [77] comprise factors whose complexity may not have been adequately captured thereby injecting uncertainty in existing cost-effectiveness estimates.

The benefits of NSCLC treatment selection and patient response are challenged by the budgetary consequences arising from the use of PD-L1 assays and test platforms with specific therapeutic agents as shown in Table 1. Studies have highlighted that while molecular testing may enable better treatment selection and savings in the longer term, it is also associated with significant impact on hospital laboratory budgets [78]. As indicated by Sheppard et al. [79] utilization of PD-L1 testing decreased treatment costs by 47% but also increased diagnostic costs by €354,783 (2019 SEK 4,4 million)². Chabrol et al. [55] analyzed the budgetary impact of PD-L1 testing in second-line therapy for NSCLC from a Brazilian private payer perspective. Although the analysis indicated considerable treatment cost savings, any assessment of study findings needs to take into account the fact that study testing costs were fully subsidized by manufacturers thereby limiting the generalizability or transferability of results into other country settings. In Sweden, PD-L1 testing would involve an addition to existing diagnostic costs whose magnitude would be determined by the frequency of the test’s utilization in the national care pathway [51]. According to treatment guidelines, PD-L1 testing should be performed at the time of initial diagnosis and then repeated at progress, relapse or metastasis for the initiation of immunotherapy subject to the indication of administered treatment(s), the presence (or not) of mutations and disease stage. It should be noted that the field is undergoing change making an extension of ICI indications to first-line treatment not unlikely. This would in turn involve an adaptation of clinical practice and require additional resources for the delivery of the required testing in the not-too distant future.

New compounds and principles in development

Not surprisingly given the success of the ICIs launched so far, there is a lot of activity in developing more agents acting on targets other than the PD-1/PD-L1-axis [80,81]. Most of these are mAbs but there are also binding-proteins and fusion-proteins. There is also a drive to develop dual-target compounds that is a single compound that attacks two pathways. The two compounds most advanced target lymphocyte-associated gene 3 (LAG3) and TIM3 judging by the number of Phase 2 and even Phase 3 studies.

LAG3 is a transmembrane protein, which binds the major histocompatibility complex (MHC) class II [81]. LAG3 has mainly been found on tumor infiltrating regulatory T-cells (Tregs). LAG3 is cleaved for optimal function of T-cells. This releases soluble LAG3 (sLAG3). In animal models, dual blockade with anti- LAG3 and -PD-1 more than doubled the tumor clearance. A large number of compounds targeting LAG3 are being studied in Phase 1−3 but only 2 studies specifically mentioned NSCLC.

Another advanced target is antibodies against TIM3 (T-cell immunoglobulin and mucin domain-3) stimulates immune attack on cancer cells by a dual blockade [80]. These mAbs block both the Tregs from down regulating the T-cells that attack cancer cells and also the cancer cells from inhibiting the T-cells. There are at least 10 ongoing Phase I trials with compounds directed at TIM3.

The ICIs in use have in common that they encourage killer T-cells to attack cancer cells but only about 20% have lasting clinical benefit [82]. Three articles in Nature this year (2020) have shown that B-cells in tertiary lymphoid structure (TLS) are associated with favorable responses to ICIs, at least in myeloma, sarcoma and renal cell [83–85]. TLS are collections of B and T cells induced by immunological stimulation. They nurture B-cells. The studies showed that TLS was frequently more marked and conspicuous in those that responded than did not. One problem for future studies is that because TLS-formation is not seen in animals like rodents (they are a feature of humans) there are no good experimental models.

Closing remarks

The ICIs have generally been approved through the conditional (accelerated) pathway and then expanded into other diseases. Institutions like NICE in England and other national bodies have unusually speedily agreed to recommend funding of these treatments. Prasad’s group [21] has shown that, though 8 ICI indications granted Accelerated Approval proved to improve OS in confirmatory trials, the FDA has allowed 5 agents that were granted Accelerated Approval to remain on the market in spite of subsequent trials failing to show statistically significant improved OS in confirmatory trials.

Now an intense development and extension of new indications are ongoing. Although ICIs have been approved for use with as low TPS as ≥1% or without any positive tumor cells (atezolizumab), it could be argued that till further data are available, treatment should be restricted to patients with high TPS ≥50% or even higher. Therefore, national oncology societies must keep detailed registers of patients treated in the real world setting so that a more complete picture can be painted.

Reviews of the current status of IHC testing for PD-L1 [3,4,31,32,86,87] seem somewhat contradictory and it is difficult to recommend a common strategy for antibody testing. Treatment decisions based on SP142 analysis appear to need separate consideration and should not be the standard test for NSCLC today, while screening with SP263 (or 22C3) may be a reasonable strategy though sending samples to lab with appropriate analyses set-up might be just as simple.

Given the moderate predictive value of current testing, there is a need to develop methods beyond IHC to determine PD-L1 expression, such as mRNA, microarray and massive parallel sequencing analysis [88–91]. Better predictive markers are urgently needed to identify patients that are likely to respond and, importantly, patients unlikely to respond must also be identified. The ethical balance between the risk of AE vs. effect of treatment must be clarified.

The concept of ‘pseudoprogression’ has become part of the treatment with ICI. Essentially, this is a concern that what looks like progression on imaging (e.g. on a CT scan), may be an inflammatory reaction to treatment. It is an approach that came in from the use of temozolomide and bevacizumab in glioblastoma. Instead of declaring progression, treatment continues for another couple of cycles to be certain. Although regulatory approvals recommend treatment continuation for clinically stable patients with initial evidence of disease progression until disease progression is confirmed, it is desirable that pseudoprogression is kept also under review and evaluated to see if it truly matters for long-term survival.

The multiplicity of commercially available test platforms as well as the variability in staining protocols and test reference points indicate the technical and other challenges that need to be overcome to make these health technologies successfully and sustainably integrated into national reimbursement schedules and clinical praxis [86,92–94].

Although it is easy to be critical about how the ICI has been introduced and used, nevertheless the studies with pembrolizumab can be seen as consistent with a not insignificant benefit. The longest, 5-year follow-up is only from the non-randomized, non-controlled study (Keynote-001) but both Keynote-024 and -042 have had some time to mature.

Conclusion

There is currently a lack of long-term follow-up data as regulatory approval of ICI for NSCLC has generally been conditional and based on early, immature data. The need for population-based, real-world data on the effects, incidence of AEs, severe AEs and associated health economic analyses of ICI treatment in NSCLC cannot be overstated. The long-term benefit in NSCLC remains to be established; the jury is still out.

Disclosure statement

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article. This work was written independently; no company or institution supported the authors financially or by providing a professional writer.

Notes

Notes

1 Inflation and conversion to 2019 SEK prices based on CCEMG – EPPI-Centre Cost Converter’ (v.1.6 last update: 29 April 2019) 13.5 SEK/£available at https://eppi.ioe.ac.uk/costconversion/ (accessed 2019-10-30)

2 Inflation and conversion to 2019 SEK prices based on CCEMG – EPPI-Centre Cost Converter’ (v.1.6 last update: 29 April 2019) 12.45 SEK/€available at https://eppi.ioe.ac.uk/costconversion/ (accessed 2019-10-30)