Long-term efficacy and safety of lorlatinib versus alectinib in anaplastic lymphoma kinase-positive advanced/metastatic non-small cell lung cancer: matching-adjusted indirect comparison

The ALEX trial [6] was selected to provide long-term efficacy and safety data for alectinib, while individual patient level data (IPD) for lorlatinib were sourced from the CROWN trial [9,10]. The analyses considered both the intention-to-treat (ITT) population and subpopulations with and without baseline brain/CNS metastases. The subgroup analysis of patients with baseline brain/CNS metastases was planned and conducted as an exploratory analysis due to the relatively small sample size in the CROWN trial. Data from CROWN patients enrolled at Chinese sites (n = 20, 6.8%) were excluded from the MAIC due to country-specific privacy restrictions.

IPD from the CROWN trial were re-weighted such that the summary statistics of the baseline characteristics of patients in the lorlatinib arm matched those in the alectinib arm of the ALEX trial and the baseline characteristics of patients in the crizotinib arm in the CROWN trial matched those in the crizotinib arm in the ALEX trial. A single matching set was used based on the primary matching set from Garcia et al. [18], which included the most clinically important effect modifiers (EMs), as determined by a targeted literature review, assessment of quantitative evidence and consultation with clinical experts. These EMs included Eastern Cooperative Oncology Group Performance Score of 0 versus 1+, presence of brain/CNS metastases at baseline per independent review committee, and race (Asian vs non-Asian). In the CROWN trial, ethnicity was used as a proxy for missing race data. Only Asian versus non-Asian categories were considered.

We updated the MAIC with the aim to generate robust long-term comparative data and explore new effect measures with clinically meaningful interpretations. In addition to quantifying the effect of treatment in terms of the hazard ratio (HR) for PFS assessed by investigator (INV), the differences and ratios in restricted mean survival time (RMST) up to a prespecified timepoint and annual probabilities were considered to relax the proportional hazards assumption and provide additional comparative evidence that is more easily understood than the HR. AEs were compared based on rate ratios (RRs) and rate differences (RDs).

In the MAICs comparing PFS, observed Kaplan–Meier (KM) data were used where possible. PFS-INV with alectinib in ALEX was extrapolated to facilitate an MAIC using 5-year PFS data. The last observed PFS probability at year 4 (and its associated standard error [SE]) with alectinib was conservatively assumed to be constant until year 5 given the observed plateau in the KM curve. Parametric survival models were required to extrapolate PFS from years 2–4 for crizotinib (CROWN) in the brain metastases subgroup, as the observed KM curve reached 0% at approximately 22 months. Seven parametric distributions commonly recommended by health technology assessment bodies, including the National Institute for Health and Care Excellence (NICE) [19] and the Canadian Agency for Drugs and Technologies in Health (CADTH) [20] (i.e., exponential, Weibull, Gompertz, log-normal, log-logistic, gamma and generalized gamma) were tested as candidate models. The parametric model that predicted year 2–4 PFS probabilities closest to zero (without truly reaching zero) was selected for the primary analysis.

Population matching was successful, and additional results, including the baseline characteristics and sample sizes before and after matching, are presented by Garcia et al. The effective sample sizes (ESS) after matching on EMs across studies were 108.62 (of n = 139 [78%]) and 125.46 (of n = 137 [92%]) for lorlatinib and crizotinib, respectively.

Lorlatinib demonstrated superior PFS-INV compared with alectinib (HR: 0.55, 95% CI: 0.34, 0.88), with a reduction in the risk of progression or death of 45%. Additionally, the mean PFS time (RMST up to 4 and 5.5 years [66 months]) in the lorlatinib arm was statistically superior to that in the alectinib arm. The difference in the RMST ranged from 8.5 months (RMST up to 4 years) to 11.2 months (RMST up to 5.5 years), resulting in an RMST approximately 1.5–1.6-times higher with lorlatinib compared with alectinib. Specifically, lorlatinib extended the mean PFS time by 8.5 months up to year 4 (RMST difference: 8.53, 95% CI: 2.76, 14.30; RMST ratio: 1.52, 95% CI: 1.15, 2.00) and by an even greater margin by year 5 (RMST difference: 11.18, 95% CI: 3.02, 19.34; RMST ratio: 1.57, 95% CI: 1.14, 2.17) compared with alectinib (Figure 1A). Furthermore, the PFS probabilities at years 1–5 were also significantly improved in terms of RDs with lorlatinib versus alectinib, ranging from 0.15 to 0.26 increases in annual PFS probabilities. The greatest difference in the annual PFS probability was observed at year 1 (0.26 [0.10, 0.43]), and by years 4–5, the estimated annual difference in PFS was 0.15 (>0.00, 0.30) (Figure 1B). The KM curves for PFS-INV before and after adjustment are presented in Figure 2A & B. The PFS-INV curve for lorlatinib was higher (i.e., more favorable) than that of alectinib both before and after the population adjustment, although KM curve for lorlatinib was higher before the adjustment compared with after the adjustment.

Lorlatinib had a higher rate of grade ≥3 AEs (RD: 0.31, 95% CI: 0.15, 0.47; RR: 1.58, 95% CI: 1.20, 2.06) compared with alectinib. However, the rates of AEs leading to treatment discontinuation (RD: 0.02, 95% CI: -0.09, 0.13; RR: 1.18, 95% CI: 0.48, 2.94), dose interruption (RD: 0.10, 95% CI: -0.07, 0.26; RR: 1.21, 95% CI: 0.77, 1.89), and dose reduction (RD: 0.11, 95% CI: -0.03, 0.24; RR: 1.73, 95% CI: 0.86, 3.49) were similar between the two treatments (Figure 1C).

Among patients with baseline brain/CNS metastases (ESS = 64.41 of n = 72), lorlatinib demonstrated a benefit in PFS-INV compared with alectinib (HR: 0.47, 95% CI: 0.19, 1.20), although this result was not statistically significant given the small sample size in this cohort. Among patients without baseline brain/CNS metastases (ESS = 195.19 of n = 204), lorlatinib demonstrated superior PFS-INV compared with alectinib (HR: 0.51, 95% CI: 0.27, 0.94) (Figure 3A).

Among patients with baseline brain/CNS metastases, at year 1, lorlatinib significantly improved PFS (RD: 0.30, 95% CI: 0.03, 0.57). A numerical benefit in the annual PFS probability was observed in the remaining time periods, although these results were not statistically significant, presumably due to the small sample sizes and event counts. Meanwhile, among patients without baseline brain/CNS metastases, at year 1, lorlatinib was associated with a higher PFS compared with alectinib (RD: 0.19, 95% CI: -0.003, 0.39), although this result was not statistically significant. Over time, lorlatinib continued to show a consistent improvement in PFS, maintaining its advantage over alectinib and demonstrating a significant increase in PFS probabilities at years 2–4 (year 4 difference: 0.26, 95% CI: 0.06, 0.45) (Figure 3B).

As the treatment landscape continues to evolve, future research exploring the biological mechanisms underlying the efficacy of ALK+ mNSCLC treatments has the potential to provide a more comprehensive understanding of how these mechanisms influence clinical outcomes and help refine treatment strategies. Further research with larger sample sizes of patients with baseline brain/CNS metastases and longer follow-up periods is needed to confirm the long-term benefits of lorlatinib in this specific subgroup and to better understand the dynamics of its efficacy over time.Our study primarily focused on PFS and did not consider overall survival (OS) because at the time of the analysis, the required number of OS events per the protocol-specified second interim analysis in the CROWN trial had not been reached. OS follow-up research is ongoing, and additional OS analyses are planned. Future research should continue to monitor OS to provide a more complete picture of the long-term benefits of ALK+ mNSCLC treatments and to validate these findings. This ongoing OS analysis will be crucial for understanding the full impact of these treatments on patient survival and guiding future treatment protocols.

Additionally, understanding economic implications is crucial for optimizing treatment decisions and ensuring that advancements in clinical efficacy are accessible and sustainable for healthcare systems globally. The literature on the cost-effectiveness of ALK+ mNSCLC treatments reveals significant variability in outcomes across different countries due to differences in model assumptions, pricing and willingness-to-pay thresholds [27–29], emphasizing the need for ongoing research to inform personalized and cost-effective treatment strategies [30,31].

The results of this analysis should be interpreted in the context of the study limitations. First, while MAICs provide valuable insights, they cannot fully replace head-to-head RCTs, which offer a more robust direct comparison of treatments under controlled conditions. The robustness of anchored MAICs relies on the successful adjustment for differences in the distribution of all treatment EMs between trials. Our MAIC adjusted for imbalances in all suspected EMs according to clinical experts and the literature, minimizing potential bias. However, although the analysis attempted to adjust for all identified EMs, which were selected based on the literature and clinical expert recommendations, other differences that are not reported in the trials may exist, potentially leading to residual confounding. Second, the study relied on data from different trials (CROWN vs ALEX) with potentially heterogeneous patient populations, treatment protocols, and follow-up procedures. Despite matching, these differences may have impacted the comparability of the PFS and AE rates. Third, Chinese patients from the CROWN trial were not included in the MAIC analyses because of country-specific privacy restrictions. Given the small proportion of excluded patients (n = 20, 6.8%), the exclusion of these patients is unlikely to significantly impact the results’ generalizability. Fourth, the MAIC analyses of the subgroup of patients with baseline brain/CNS metastasis are limited by the relatively small sample size in the CROWN trial (n = 72) and are likely underpowered. This subgroup was investigated as an exploratory analysis and should be interpreted with caution. Fifth, although the outcome definitions were generally similar between the trials, the AE severity grading was conducted according to different versions of National Cancer Institute criteria. Since the comparisons were anchored, these differences are not expected to introduce significant bias. Sixth, the MAIC analysis adjusted for only four baseline variables, which were selected based on the literature and clinical expert recommendations. Other important prognostic factors affecting survival may have been overlooked, potentially leading to residual confounding and unaddressed differences between the study groups, affecting the comparability of the results. The crizotinib survival curves still appear different across studies because not all prognostic factors may be captured; however, this should not impact the findings from the anchored ITC, which only requires adjustment for EMs. Finally, the generalizability of these results to other diseases using targeted therapies in first-line therapy is unclear. Lorlatinib is approved specifically for ALK-positive NSCLC, and therefore, it is uncertain whether these findings can be applied more broadly, which may limit their applicability beyond ALK-positive NSCLC.

Despite these limitations, the analysis also has several notable strengths. The use of long-term data offered a more comprehensive understanding of lorlatinib’s relative efficacy and safety, which is crucial for making informed treatment decisions over both short-term and extended periods. The robust adjustment for EMs ensured a balanced and fair comparison between lorlatinib and alectinib, addressing potential population imbalances that could have otherwise distorted the results. Furthermore, the use of an anchored approach enhances the reliability and validity of the findings by leveraging indirect comparisons through a common comparator. These strengths collectively provided a solid foundation for supporting lorlatinib as an effective first-line treatment option for patients with ALK+ mNSCLC, offering both short-term and long-term benefits.