Real-World Clinical and Economic Consequences of Next-Generation Anaplastic Lymphoma Kinase-Positive (ALK +) Tyrosine Kinase Inhibitors (TKIs) in Patients with TKI-Naïve ALK + Metastatic Non-Small Cell Lung Cancer in the United States

Abstract

Introduction

This study evaluated the real-world utilization, effectiveness, occurrence of adverse events (AEs), and healthcare costs in patients treated with next-generation anaplastic lymphoma kinase (ALK) tyrosine kinase inhibitors (TKI), alectinib, brigatinib, and lorlatinib, among ALK TKI-naïve patients with ALK rearrangement-positive metastatic non-small cell lung cancer (ALK + mNSCLC) in the US.

Methods

This retrospective, observational cohort study used Veradigm electronic health records linked with Komodo closed claims. Eligible patients were adults with ALK + mNSCLC diagnosis who received their first ALK TKI (alectinib, brigatinib, or lorlatinib) between 5/22/2020 (US FDA approval of first-line brigatinib) and 12/31/2024. Descriptive analyses included baseline patient demographics, clinical characteristics, occurrence of AEs, and treatment patterns. Unweighted Cox proportional hazards models and propensity score-weighted Cox proportional hazards models were conducted to compare real-world treatment effectiveness [real-world time to treatment discontinuation (rwTTD), real-world time to next treatment (rwTTNT), real-world progression-free survival (rwPFS), real-world overall survival (rwOS)], and all-cause healthcare utilization and costs across ALK TKIs.

Results

A total of 481 patients (alectinib: 391; brigatinib: 32; lorlatinib: 58) were included. Mean age ranged from 58.5 to 63.7 years across the three treatment groups; more than half of the patients were female (less than half male), White, or covered by commercial insurance. In the unweighted and weighted analyses, there were no statistically significant differences in rwTTD, rwTTNT, rwPFS, or rwOS between treatment groups. Mean weighted all-cause healthcare costs per-patient-per-month during the index TKI treatment duration were not significantly different across cohorts (alectinib: US$20,873; brigatinib: $25,693; lorlatinib: $24,984).

Conclusion

This study found no statistically significant differences in real-world clinical effectiveness and total all-cause healthcare costs between alectinib, brigatinib, or lorlatinib for ALK TKI-naïve patients with ALK + mNSCLC. Future studies with larger sample sizes and longer follow-up time are needed to confirm these findings.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s12325-026-03679-0.

Key Summary Points

Why carry out this study?

Alectinib, brigatinib, and lorlatinib are next-generation anaplastic lymphoma kinase (ALK) tyrosine kinase inhibitors (TKIs) approved for first-line treatment of ALK-positive metastatic non-small cell lung cancer (ALK+ mNSCLC), but head-to-head data are lacking.

This study compared real-world clinical effectiveness (time to treatment discontinuation, time to next treatment, progression-free survival, and overall survival) and all-cause healthcare costs among patients with TKI-naïve ALK+ mNSCLC treated with alectinib, brigatinib, or lorlatinib in routine US clinical practice.

What was learned from the study?

In this analysis, no statistically significant differences were found in clinical effectiveness outcomes or mean monthly healthcare costs per patient across the three treatment groups after propensity score weighting.

These findings indicate that, in routine clinical practice, evidence does not yet support differentiating treatment choice based on real-world clinical outcomes or treatment costs. Studies with larger sample sizes and extended follow-up are needed.

Introduction

In 2025, cancers of the lung and bronchus were estimated to account for 11.1% of all new cancer diagnoses and 20.2% of all cancer deaths in the United States (US) [1]. While the overall 5-year survival rate is estimated to be 28.1%, the rate varies considerably by histology, stage at diagnosis, and the presence of sensitizing mutations [1, 2]. Anaplastic lymphoma kinase-positive (ALK+) metastatic non-small cell lung cancer (mNSCLC) is an aggressive molecular subtype of lung cancer that is characterized by rapid progression and early dissemination, with central nervous system metastases occurring in 30% of patients at the time of diagnosis and up to 70% through the course of their disease [3, 4]. For these patients, the utilization of ALK tyrosine kinase inhibitors (TKI) has significantly improved median overall survival (OS) to over 4 years [5].

In the US, the first-generation ALK TKI crizotinib and the next-generation ALK TKIs alectinib, brigatinib, and lorlatinib were approved for first-line (1L) treatment of ALK + mNSCLC by the US Food and Drug Administration (FDA) in 2011, 2017, 2020, and 2021, respectively [6–9]. First-generation ALK TKI crizotinib has limited ability to cross the blood–brain barrier, leading to reduced efficacy among patients with brain metastases [10]. The next generation ALK TKIs, alectinib, brigatinib, or lorlatinib, are more potent, have superior blood–brain barrier penetration, have demonstrated superior efficacy in clinical trials compared to crizotinib [11–13], and are now preferred 1L treatment options over crizotinib for ALK + mNSCLC [14]. However, there is a lack of head-to-head clinical trials comparing clinical outcomes between the next-generation ALK TKIs.

Recent studies have used US real-world data to demonstrate superior effectiveness in favor of the next-generation ALK TKIs compared to crizotinib to support the results observed in clinical trials [15, 16]. Other real-world evidence studies evaluating next-generation ALK TKIs have conducted pairwise comparisons [17, 18], with no study comprehensively assessing the comparative effectiveness of all three agents within a single analytic framework. This represents a critical gap in the evidence base, particularly in an era where various, highly active targeted therapies are simultaneously available and routinely considered as standard treatment options. Beyond comparative clinical outcomes, healthcare cost analysis using real-world data can help inform treatment decision-making in the context of value-based care initiatives. Despite this, no real-world evidence study to date has concurrently evaluated both the comparative effectiveness and the comparative costs of the three next-generation ALK TKIs. Consequently, decision-makers currently lack integrated evidence to fully assess the clinical and economic trade-offs among these therapies in routine practice.

In this real-world evidence study, we directly compare the clinical and economic outcomes associated with all three next-generation ALK TKIs within a unified analytic framework. By jointly evaluating real-world effectiveness and healthcare costs, this study aims to generate the first comprehensive evidence to help optimize treatment strategies for patients with ALK + mNSCLC in today’s treatment landscape.

Methods

Study Design and Data Sources

This was a retrospective, observational cohort study of patients with ALK + mNSCLC who received alectinib, brigatinib, or lorlatinib as their first ALK TKI therapy since mNSCLC diagnosis. The Veradigm Network electronic health records (EHR) linked with closed claims and mortality data from Komodo Health (linked dataset) between January 1, 2016, and December 31, 2024, were used in this study. The Veradigm Network EHR includes data from primary and specialty practices across the US [19]. The Komodo closed claims source includes individuals with commercial insurance, Medicare, Medicaid, and other insurance types [19]. Komodo’s mortality data, which are updated monthly, are aggregated from government records, public listings, private claims, and obituary data. For patients with cancer, Komodo’s mortality data captures > 90% of deaths reported by the US Centers for Disease Control and Prevention, according to the data provider [20]. The linked dataset provides comprehensive medical information on patients from all 50 states and has been shown to be representative of the US population [19].

The linked dataset only contained de-identified data as per the de-identification standard defined in Section §164.514(a) of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rule. The process by which the data were de-identified is attested to through a formal determination by a qualified expert, as defined in Section §164.514(b) (1) of the HIPAA Privacy Rule. Because this study used only de-identified patient records, it was no longer subject to the HIPAA Privacy Rule and was therefore exempt from institutional review board approval and frm obtaining informed consent according to US law. The study was designed, implemented, and reported in accordance with the International Society for Pharmacoepidemiology Guidelines for Good Pharmacoepidemiology Practice, and was conducted in compliance with the ethical principles laid down in the Declaration of Helsinki and only used de-identified data.

Study Cohort

To be eligible for this study, patients needed to have at least one medical claim with evidence of a lung cancer diagnosis (Supplementary Table 1) between January 1, 2016 (the date with the first available data), and December 31, 2024, at least one medical claim with evidence of metastatic disease between the initial lung cancer diagnosis and December 31, 2024, and started monotherapy use of alectinib, brigatinib, or lorlatinib after the initial mNSCLC diagnosis, between May 22, 2020 (the 1L approval of brigatinib by the FDA) and December 31, 2024. The index date was defined as the start date (earliest claim for a prescription fill) for alectinib, brigatinib, or lorlatinib. Because real-world datasets rarely capture molecular status, consistent with prior studies using similar data sources [16, 21–23], receipt of ALK TKIs was used as a proxy for an ALK + NSCLC diagnosis in this study. Patients were required to be at least 18 years of age on the index date, have at least 6 months of continuous enrollment with medical and pharmacy benefits or at least one record in the EHR prior to the index date, and have at least one claim and at least one EHR record after the index date. Patients were excluded if they had evidence of ALK TKI (alectinib, brigatinib, crizotinib, ceritinib, or lorlatinib) use any time prior to the index date. Since lorlatinib can also be used as a subsequent therapy for ROS1-positive NSCLC [14], patients were excluded if they received treatment indicative of a ROS1 rearrangement—defined as use of entrectinib or repotrectinib at any time after the mNSCLC diagnosis—among those who indexed on lorlatinib.

Based on index therapy, patients were grouped into three mutually exclusive groups. The baseline period was considered to be the 6 months preceding the index date. The follow-up period began on the index date and continued until the end of continuous enrollment in medical and pharmacy benefits. The index treatment period began on the index date and continued until the earliest of either discontinuation of the index therapy or the end of follow-up. The study overview diagram is outlined in Fig. 1.

Fig. 1.

Study overview diagram. ALK anaplastic lymphoma kinase-positive, mNSCLC metastatic non-small cell lung cancer, TKI tyrosine kinase inhibitors

Study Measures

Demographic characteristics, including age, sex, race, and geographic region, were captured on the index date. Sex and race were captured administratively as documented in the underlying dataset. Sex was not missing for any individuals. Where race was missing, it was documented as missing. Where race was non-missing, the values were mapped to White, Black, Asian, and Other (which included multi-racial). Clinical characteristics, such as the National Cancer Institute comorbidity index [24] and the presence of brain, bone, liver, and kidney metastases, were captured in the 6-month baseline period. The days from the first metastasis diagnosis to the index date and the duration of follow-up were also collected.

Treatment patterns were assessed using only claims data and reported based on the following line of therapy (LOT) rules for mNSCLC treated with targeted therapies published in Grady et al., using a 180-day gap period to define discontinuation based on the chart review validation conducted by Meng et al. [25, 26]. According to the algorithm, treatment patterns were classified as continuation, discontinuation, augmentation, restart, and switching. Definition details of the treatment patterns are listed in the footnotes of Supplementary Table 2. The type of next line therapy was classified as chemotherapy, immunotherapy, targeted therapy, or a combination of the previous. Treatment sequencing after the index therapy was descriptively summarized. Treatment duration was determined by days’ supply for medications captured in pharmacy claims or by the estimated duration of clinical benefit for medications captured in medical claims.

Captured clinical outcomes included real-world time to treatment discontinuation (rwTTD), real-world time to next treatment (rwTTNT), real-world progression-free survival (rwPFS), and real-world OS (rwOS). The rwTTD was calculated as the time from index to treatment discontinuation or death, whichever came first. The rwTTNT was calculated as the time from the index date until the start of the next line of therapy or death, whichever came first. The rwPFS was defined as the time from index until the earliest of the following: start of the next LOT, diagnosis of a metastasis in a new location, or death. The rwOS was defined as the length of time from the index date until death. For clinical outcome estimates, if no event was observed, patients would be censored at the end of continuous enrollment or the end of available data, whichever came first.

This analysis also descriptively assessed the occurrence of 25 prespecified adverse events (AEs) (Supplementary Table 1) during the index treatment period; two AEs (suicidal ideation and hypophosphatemia) were later excluded due to an absence of relevant records in the dataset. Codes for suicidal ideation were redacted in the underlying dataset and there were no laboratory values for phosphate levels among any patients in the study cohort during the study period. The occurrence of AEs was reported among patients who did not have them during the baseline period. Most of the conditions were assessed by diagnosis codes, while hypertriglyceridemia and increased ALT/AST were assessed by laboratory results in the EHR (Supplementary Table 1). AEs occurring in 10% or more of patients in any of the three treatment groups were reported in this study.

All-cause healthcare utilization and costs were captured for the index treatment period. Utilization and costs were calculated overall and for the following subcategories: inpatient admissions, emergency department (ED) visits, outpatient (OP) office visits, laboratory visits, radiology visits, other medical services, and pharmacy services. Other medical services include all services that do not fit into one of the other categories, such as infusion services, care at skilled nursing facilities, and durable medical equipment, etc.

Paid costs were missing for a subset of the specialty pharmacy claims in the raw Komodo dataset. To address this, we imputed paid costs for the index medications following a three-step approach: (1) when the paid cost was available, it was used as observed; (2) when the paid cost was missing but the submitted cost was available, the paid cost was imputed by multiplying the submitted cost by the median cost ratio for the corresponding ALK TKI National Drug Code (NDC); and (3) when neither paid nor submitted cost was available, the paid cost was imputed using the median paid cost across all records with the same NDC. Median cost ratios were calculated at three hierarchical levels of specificity. If available, we applied a patient-specific median cost ratio using records for the same NDC within the same patient. If fewer than three records were available, we applied a payer-specific median cost ratio based on records for the same NDC within the same payer type. If fewer than three payer-specific records were available, we applied an overall median cost ratio calculated from all study records with the same NDC. Box plots of the mean monthly pharmacy costs after imputation are provided in Supplementary Fig. 1.

All costs were annualized and adjusted for inflation using the U.S. Medical Care consumer price index (CPI) from the US Department of Labor’s Bureau of Labor Statistics and reported in 2024 US dollars [27].

Statistical Analysis

Continuous variables were captured as mean [standard deviation (SD)] and compared using Student’s t tests. Categorical variables were captured as counts and percentages and compared using chi-squared tests. Time-to-event outcomes were described using Kaplan–Meier methods and presented as Kaplan–Meier curves, and were evaluated using unadjusted Cox proportional hazards models with robust standard errors. To account for multiple comparisons, a Bonferroni correction was applied to the alpha level (alpha = 0.05/3 = 0.0167) [28], and P value < 0.0167 was considered significant in the analysis. The AE analysis was descriptive only and did not compare between groups.

In comparative analyses of treatment effectiveness and healthcare costs, inverse probability of treatment weighting (IPTW) was implemented to balance the baseline differences across treatment groups [29]. To optimize the IPTW, first a list of potential covariates was determined based on study objectives, clinical input, literature [30], and an assessment of baseline differences between cohorts. Eastern Cooperative Oncology Group (ECOG) performance status was included in the initial covariate list but was removed later because it was not available for any patients in the final analytic cohort. For the final covariate list, the assumed causal relationships among covariates, treatment, and outcomes informing variable selection are illustrated in Supplementary Fig. 2; the classification of each covariate by variable type, with rationale for inclusion, is provided in Supplementary Table 3.

In order to compare all three index treatment cohorts simultaneously, multinomial modeling was executed accounting for the following baseline patient characteristics: age (categorical < 65 vs. > 65 years), sex (male, female), race (white, black, Asian, other/unknown), time from metastasis to index date, baseline brain metastasis, and index year. Standardized mean difference values were assessed with cut-offs of ≤ 0.1 and ≤ 0.25, upon assessment variables were considered sufficiently balanced for comparing between groups [31]. Winsorizing, at the 1st and 99th percentile, was applied to handle extreme weights [32]. Implementation of IPTW resulted in slight changes to the reported size of the study cohorts because the weighting of the patient counts resulted in an effective sample size [29]. Study measures are reported for both the weighted and the unweighted populations. The weight distribution characteristics are reported in Supplementary Table 4.

Given the small cohort sizes, overlap distributions and proportional hazards (via Schoenfeld residuals) were assessed to test the proportional hazards assumptions [29]. Propensity score overlap was substantial between all three cohorts. Schoenfeld residual tests were conducted for all four endpoints in both the unweighted and weighted models and are reported in Supplementary Table 5. The proportional hazards assumption was violated in the weighted rwTTNT and rwOS models. Restricted mean survival time analysis, which does not require the proportional hazards assumption, was therefore conducted as a sensitivity analysis using the method developed by Conner et al. [33]. For each outcome, the time horizon was specified as the minimum across treatment groups of the last observed event time.

The analysis was conducted using SAS V9.4. The Kaplan–Meier curves were prepared using R version 4.5.1.

Results

Patient Characteristics

A total of 481 patients (alectinib: 391; brigatinib: 32; lorlatinib: 58) met the selection criteria and were included in the analysis (Fig. 2). Among them were 3 patients in the lorlatinib cohort (5.2%) who initiated on lorlatinib prior to approval of this medication for first-line use. As they met all other study criteria, these patients were retained in the final cohort. In the three treatment groups, the mean age ranged between 58.5 and 63.7 years; more than half of the patients were female, White, or had commercial insurance coverage. The median (IQR) time from first metastasis diagnosis to index date was 40 (23, 137) days for the alectinib cohort, 58 (32, 260) days for the brigatinib cohort, and 217 (51, 482) days for the lorlatinib cohort. Compared to patients who indexed on alectinib or brigatinib, patients who indexed on lorlatinib were identified in more recent years. Baseline brain metastases were observed in 37.1%, 50.0%, and 51.7% of the alectinib, brigatinib, and lorlatinib cohorts, respectively. More details of patient baseline demographics and clinical characteristics are reported in Table 1.

Fig. 2.

Patient selection. 1L first line, ALK anaplastic lymphoma kinase-positive, EHR electronic health records, LOT line of therapy, ROS1 ROS proto-oncogene 1, TKI tyrosine kinase inhibitors ^aDefined as either entrectinib use anytime after metastasis among patients indexing on crizotinib, or entrectinib or repotrectinib use anytime after metastasis among patients indexing on lorlatinib

Table 1.

Baseline patient characteristics

	Alectinib	Brigatinib	Lorlatinib	P value
	N = 391	N = 32	N = 58	A vs. B		B vs. L		A vs. L
Age, mean (SD)	58.5 (12.5)	59.0 (15.1)	63.7 (12.4)	0.830		0.109		0.003
Sex, n (%)				0.310		0.775		0.079
Female	208 (53.2)	20 (62.5)	38 (65.5)
Male	183 (46.8)	12 (37.5)	20 (34.5)
Race, n (%)				0.912		0.636		0.590
White	201 (51.4)	16 (50.0)	29 (50.0)
Black	26 (6.6)	2 (6.3)	1 (1.7)
Asian	41 (10.5)	5 (15.6)	6 (10.3)
Othera	63 (16.1)	4 (12.5)	12 (20.7)
Unknown/not reported	60 (15.3)	5 (15.6)	10 (17.2)
Geographic Region, n (%)				0.459		0.218		0.007
Northeast	82 (21.0)	3 (9.4)	12 (20.7)
Midwest	83 (21.2)	7 (21.9)	5 (8.6)
South	114 (29.2)	11 (34.4)	25 (43.1)
West	112 (28.6)	11 (34.4)	15 (25.9)
Other/unknown	0 (0.0)	0 (0.0)	1 (1.7)
Payer, n (%)				0.515		0.393		0.031
Commercial	226 (57.8)	17 (53.1)	29 (50.0)
Medicare Advantage	95 (24.3)	11 (34.4)	21 (36.2)
Medicare fee-for-service	10 (2.6)	0 (0.0)	4 (6.9)
Medicaid	60 (15.3)	4 (12.5)	4 (6.9)
Year of index, n (%)				0.356		0.001		0.000
2020	68 (17.4)	10 (31.3)	3 (5.2)
2021	107 (27.4)	7 (21.9)	11 (19.0)
2022	121 (30.9)	10 (31.3)	13 (22.4)
2023	72 (18.4)	4 (12.5)	22 (37.9)
2024	23 (5.9)	1 (3.1)	9 (15.5)
Days from first metastasis diagnosis to index date, median (IQR)	40 (23, 137)	58 (32, 260)	217 (51, 482)	0.079		0.020		0.000
Duration of follow-up in days, median (IQR)	506 (233, 837)	591 (283, 875)	355 (183, 681)	0.796		0.182		0.045
Anti-cancer treatment between metastasis and index date, n (%)	105 (26.9)	7 (21.9)	16 (27.6)	0.539		0.552		0.907
NCI comorbidity index, mean (SD)	0.62 (0.64)	0.68 (0.56)	0.67 (0.65)	0.612		0.986		0.530
Baseline metastases, n (%)
Brain	145 (37.1)	16 (50.0)	30 (51.7)	0.148		0.876		0.033
Bone	183 (46.8)	15 (46.9)	25 (43.1)	0.994		0.730		0.598
Liver	82 (21.0)	8 (25.0)	6 (10.3)	0.592	0.066		0.057
Kidney	7 (1.8)	0 (0.0)	0 (0.0)	0.445	0.997		0.304

Significant P values shown in italics

^aThe "other" category is a residual category for individuals with a documented race not captured in the 3 listed categories, such as Alaskan Native, Pacific Islander, or multi-racial

A alectinib, B brigatinib, L lorlatinib, N number, IQR interquartile range, SD standard deviation

Treatment Patterns and Sequencing

The median (IQR) follow-up time for alectinib, brigatinib, and lorlatinib cohorts was 506 (233, 837), 591 (283, 875), and 355 (183, 681) days, respectively (Table 1). Most patients in each group (alectinib: 59.6%; brigatinib: 56.3%; lorlatinib: 65.5%) were still receiving their index LOT on the last day of their available data (Supplementary Table 2). Of those who discontinued their index therapy, 69.0% (alectinib), 57.1% (brigatinib), and 70.0% (lorlatinib) had a next LOT during the follow-up. Figure 3 describes the treatment sequencing after the index therapies of alectinib, brigatinib, and lorlatinib. In the next LOT, lorlatinib monotherapy was most used. A full breakdown of the next LOT regimens by index medication can be found in Supplementary Table 6.

Fig. 3.

Sankey diagram of treatment sequencing after the index therapy

Adverse Events

AEs that were observed during the index LOT treatment period among patients without evidence of the condition at baseline are summarized in Table 2. Twelve AEs were observed in 10% or more of patients in at least one treatment group. Among these, cough was common across all three groups, with an occurrence of 20% or higher. Fatigue was diagnosed in 31.9% of the alectinib cohort, 23.8% of the brigatinib cohort, and 17.5% of the lorlatinib cohort.

Table 2.

Percentage of patients with adverse events^a among those who did not have the condition in the baseline period

	Alectinib	Brigatinib	Lorlatinib
	N = 391	N = 32	N = 58
Cognitive disorder/mental impairment	32/360 (8.9%)	3/28 (10.7%)	10/51 (19.6%)
Cough	51/222 (23.0%)	5/20 (25.0%)	8/40 (20.0%)
Dyspnea/Shortness of breath	57/200 (28.5%)	2/15 (13.3%)	11/35 (31.4%)
Fatigue	83/260 (31.9%)	5/21 (23.8%)	7/40 (17.5%)
Headaches	45/312 (14.4%)	3/24 (12.5%)	5/50 (10.0%)
Hyperglycemia	31/329 (9.4%)	4/30 (13.3%)	5/51 (9.8%)
Hypertension	38/202 (18.8%)	1/16 (6.3%)	8/24 (33.3%)
Insomnia	29/355 (8.2%)	3/28 (10.7%)	4/54 (7.4%)
Nausea/Vomiting	34/323 (10.5%)	6/22 (27.3%)	5/49 (10.2%)
Peripheral edema	66/341 (19.4%)	4/28 (14.3%)	17/51 (33.3%)
Rash	33/374 (8.8%)	6/32 (18.8%)	1/57 (1.8%)
Weight gain	12/387 (3.1%)	0/32 (0.0%)	6/58 (10.3%)

N number

^aOnly adverse events that occurred in at least 10% of one of the treatment groups are presented in this table

In the alectinib group, dyspnea/shortness of breath and peripheral edema were observed in 28.5% and 19.4% of patients, respectively. In the brigatinib group, nausea or vomiting was observed in 27.3% of patients, and rash in 18.8%. For lorlatinib, hypertension, dyspnea or shortness of breath, cognitive disorder or mental impairment, and weight gain were observed in of 33.3%, 31.4%, 19.6%, and 10.3% of patients, respectively.

Inverse Probability of Treatment Weighting

The descriptive analysis results of real-world treatment effectiveness in the unweighted population are available in Supplementary Fig. 3. In the comparative analysis, IPTW was employed to balance differences in baseline characteristics between cohorts. The weighted Ns were 390.2 in the alectinib cohort, 31.8 in the brigatinib cohort, and 53.2 in the lorlatinib cohort and the effective sample sizes were 381.2, 24.6, and 36.5, respectively. Covariate balance before and after IPTW is reported in Fig. 4A, B, respectively. Baseline characteristics were well balanced after weighting and are reported in Supplementary Table 7.

Fig. 4.

Covariate balance as measured by standardized mean difference A before and B after inverse probability of treatment weighting

Treatment Effectiveness

The Kaplan–Meier curves of real-world treatment effectiveness (rwTTNT, rwTTD, rwPFS, and rwOS) with the weighted study population are presented in Fig. 5. Hazard ratios and 95% confidence intervals (CIs) for the weighted and unweighted univariate Cox proportional hazard models are reported in Table 3. In the weighted analysis, there were no statistically significant differences between cohorts in rwTTNT, rwTTD, rwPFS, rwOS. However, the CIs for rwOS were large due to the low frequency of events. Findings were broadly similar in direction when outcomes were assessed by restricted mean survival time; however, using this approach, the longer rwOS observed in the brigatinib cohort compared to the alectinib cohort was statistically significant for the weighted analysis (Supplementary Table 8).

Fig. 5.

Kaplan–Meier plots of treatment effectiveness outcomes for the weighted population: A real-world time to next treatment, B real-world time to treatment discontinuation, C real-world progression free survival, and D real-world overall survival

Table 3.

Results from univariate Cox proportional hazard models [HR (95% confidence interval)] of rwTTNT, rwTTD, rwPFS, and rwOS for weighted and unweighted populations^a

	Alectinib vs. Brigatinib	Lorlatinib vs. Brigatinib	Alectinib vs. Lorlatinib
Weighted
rwTTNT	1.16 (0.66, 2.03)	1.45 (0.63, 3.31)	0.80 (0.41, 1.54)
rwTTD	1.11 (0.64, 1.92)	1.07 (0.49, 2.30)	1.04 (0.58, 1.86)
rwPFS	0.92 (0.58, 1.47)	1.18 (0.65, 2.14)	0.78 (0.52, 1.17)
rwOS	2.86 (0.66, 12.43)	3.69 (0.70, 19.59)	0.78 (0.33, 1.84)
Unweighted
rwTTNT	1.16 (0.65, 2.08)	1.36 (0.64, 2.87)	0.86 (0.50, 1.46)
rwTTD	0.93 (0.57, 1.53)	1.03 (0.54, 1.96)	0.91 (0.57, 1.45)
rwPFS	0.85 (0.54, 1.33)	0.85 (0.48, 1.49)	1.01 (0.68, 1.49)
rwOS	1.69 (0.43, 6.65)	2.45 (0.54, 11.08)	0.69 (0.32, 1.50)

HR hazard ratio, rwOS real-world overall survival, rwPFS real-world progression free survival, rwTTD real-world time to treatment discontinuation, rwTTNT real-world time to next treatment

^aAn HR > 1 indicates that the risk of an event is higher in cohort 1 compared to cohort 2. For example in the rwOS model a HR > 1 indicates that the risk of death is higher in cohort 1 than in cohort 2

Healthcare Resources Utilization and Healthcare Costs

The descriptive analysis results of HCRU and all-cause healthcare costs during the index therapy in the unweighted patient population are available in Supplementary Table 9. In the weighted analysis, patients in the three treatment groups had no significant differences in total inpatient stays, number of medical visits (ED, OP, and laboratory visits), number of other medical services, or number of pharmacy claims (Table 4). For patients in the alectinib, brigatinib, and lorlatinib cohorts, the mean (SD) weighted all-cause healthcare costs per-patient per-month (PPPM) during the index treatment period were $20,873 ($12,894), $25,693 ($29,672), and $24,984 ($11,609), respectively. The itemized healthcare costs, mean (SD), PPPM, are reported in Table 4.

Table 4.

All-cause HCRU and healthcare costs during the index therapy of the weighted population

	Alectinib	Brigatinib	Lorlatinib	P valuea
	N = 390.2	N = 31.8	N = 53.2	A vs. B	B vs. L	A vs. L
HCRU, mean (SD)
Total hospitalized days	3.1 (7.7)	1.6 (3.9)	2.8 (5.5)	0.280	0.253	0.838
Number of ED visits	0.8 (1.8)	0.7 (1.2)	0.9 (2.0)	0.831	0.661	0.698
Number of OP office visits	26.3 (26.2)	25.0 (31.3)	24.4 (23.5)	0.798	0.916	0.619
Number of laboratory visits	14.0 (12.5)	12.7 (13.8)	14.2 (17.6)	0.563	0.679	0.929
Number of radiology visits	8.6 (8.3)	6.7 (8.1)	8.2 (7.3)	0.206	0.372	0.733
Number of other OP services	22.4 (58.4)	16.1 (25.5)	35.4 (82.0)	0.549	0.201	0.149
Number of pharmacy claims	48.1 (54.0)	66.6 (93.2)	62.9 (66.6)	0.083	0.833	0.069
Total healthcare costs, mean (SD)	$209,963 ($164,288)	$219,455 ($186,019)	$238,853 ($170,409)	0.756	0.625	0.232
Total healthcare costs, PPPM, mean (SD)	$20,873 ($12,894)	$25,693 ($29,672)	$24,984 ($11,609)	0.078	0.876	0.028
Inpatient costs, PPPM	$1,529 ($4,959)	$715 ($2,275)	$1,554 ($3,107)	0.359	0.189	0.972
ED visit costs, PPPM	$272 ($793)	$192 ($498)	$340 ($939)	0.576	0.412	0.563
OP office visit costs, PPPM	$456 ($578)	$428 ($442)	$380 ($330)	0.790	0.569	0.348
Laboratory visit costs, PPPM	$229 ($436)	$365 ($799)	$305 ($598)	0.121	0.696	0.261
Radiology visit costs, PPPM	$941 ($1,330)	$1,164 ($3,148)	$977 ($1,595)	0.432	0.718	0.854
Other OP service costs, PPPM	$1,269 ($2,178)	$668 ($905)	$2,561 ($6,222)	0.123	0.092	0.003
Pharmacy costs, PPPM	$16,176 ($10,624)	$22,161 ($27,201)	$18,865 ($8,946)	0.010	0.417	0.079

A alectinib, B brigatinib, ED emergency department, HCRU healthcare resource utilization, IQR Interquartile Range, L lorlatinib, N number, OP outpatient, PPPM per-patient per-month, SD standard deviation

Discussion

To our knowledge, this is the first real-world evidence study comparing all three next-generation ALK TKIs, alectinib, brigatinib, and lorlatinib, for TKI naïve ALK + mNSCLC, in terms of clinical effectiveness and economic burden. In the unweighted descriptive analysis and the weighted analysis, differences among the three ALK TKIs in rwTTD, rwTTNT, rwPFS, and rwOS were not statistically significant in the main analysis. In the sensitivity analysis using restricted mean survival time to assess time-dependent outcomes, rwOS was longer among patients in the brigatinib cohort compared to the alectinib cohort. No statistically significant differences across treatment groups were observed in weighted healthcare resource utilization or total costs.

In the ALEX, ALTA-1L, and CROWN clinical trials, alectinib, brigatinib, and lorlatinib demonstrated superior clinical efficacy compared to crizotinib [11, 13, 34]. With a lack of head-to-head clinical trials comparing the efficacy of next-generation ALK TKIs, clinical decisions on optimal treatment options are based on the results of individual clinical trials, treatment preferences, and toxicity profiles [35]. Notably, the absence of data on the mechanism of acquired resistance or on the reliability of the molecular diagnostics of individual trials means that clinicians are recommended to consider different treatment options rather than relying on any single clinical trial [36]. Additionally, it is challenging to translate treatment efficacy observed in clinical trials to real-world settings because patients in routine clinical practice can be different from trial participants. Clinical trials may exclude patients with a higher burden of disease, those with complex comorbidities, and members of underrepresented groups [37–39]; furthermore, issues such as treatment adherence and drug toxicity may affect drug dosage intensity and treatment outcomes in the real world [40]. Therefore, studies using real-world data are critical to understanding the clinical and economic consequences of 1L ALK TKIs use in daily clinical practice.

Three recent real-world evidence studies have directly compared alectinib and brigatinib as 1L treatment of ALK + mNSCLC. Jeon et al. conducted a retrospective observational study of 176 patients treated with alectinib and 32 patients treated with brigatinib for 1L ALK + mNSCLC, using EHR data from Korea, and demonstrated that alectinib and brigatinib had similar clinical benefits in objective response rate, rwPFS, or rwTTNT [17]. Kim et al. compared alectinib (71% used in 1L) and brigatinib (44.4% used in 1L) using univariate and multivariate Cox proportional hazard models with EHR data from seven hospitals, and concluded that both ALK TKIs had similar treatment effectiveness in terms of rwTTD, duration of response, OS, and response rate [18]. Ghosh et al. evaluated and compared rwTTD, rwTTNT, and rwOS for 1L brigatinib (n = 58) versus alectinib (n = 1100) and 1L brigatinib versus lorlatinib (n = 81) in the US with weighted analysis using Komodo claims data, and demonstrated no significant differences in treatment effectiveness in rwTTD, rwTTNT, and rwOS in both comparisons [16]. In one other study, a comparative analysis among the next-generation ALK TKIs was not feasible because of the small sample size, as only one patient received 1L brigatinib and four received 1L lorlatinib [15]. Mudumba et al. conducted a descriptive (i.e., non-comparative) analysis of ALK TKIs and found that the median 1L rwTTD was 15.3, 7.8, and 2.8 months for alectinib, brigatinib, and lorlatinib, respectively [21]. In their analysis, 22 patients treated with brigatinib and 16 patients treated with lorlatinib were identified from Optum’s Clinformatics Data Mart database between 2016 and 2021, with median follow-up times of 8.5 and 3.5 months, respectively. The small sample size and short follow-up time for the brigatinib and lorlatinib cohorts may have contributed to the shorter observed rwTTD in this study compared to other studies. Compared with prior studies, our analysis offers a more direct and comprehensive comparison of the three ALK TKIs in rwTTD, rwTTNT, rwPFS, and rwOS. Using both unweighted and weighted approaches with a more recent data cutoff, the current study provides updated real-world evidence on the comparative effectiveness of alectinib, brigatinib, and lorlatinib.

The four clinical endpoints evaluated in this analysis capture distinct clinical constructs. Specifically, rwTTD reflects treatment tolerability and adherence, representing the point at which therapy was discontinued regardless of cause, while rwTTNT reflects the interval until physician-initiated treatment change, which may be driven by progression, toxicity, or clinical judgment. In addition, rwPFS captures proxy-assessed tumor control, defined here using a composite of treatment change, new metastasis diagnosis, and death, while rwOS reflects overall survival benefit inclusive of the effects of subsequent therapies. The absence of significant differences in rwTTD and rwTTNT suggests that tolerability and rates of treatment change were broadly similar across agents, while the null findings for rwPFS and rwOS suggest that disease progression trajectories and overall survival did not differ significantly during the available follow-up. Nevertheless, given the small comparator cohort sizes, wide confidence intervals, for rwOS in particular, and limited follow-up, these findings should be interpreted with caution.

In our analysis, the descriptive summary of AEs that occurred in the real world is consistent with reports from clinical trials. For example, in the CROWN study, hypercholesterolemia (70%), hypertriglyceridemia (64%), edema (55%), increased weight (38%), peripheral neuropathy (34%), cognitive effects (21%), anemia (19%), hypertension (18%), and mood effects (16%) were frequently reported in patients treated with lorlatinib [12]. In ALEX, the most common treatment-related AEs in patients receiving alectinib were anemia (22.4%), increased bilirubin level (19.1%), and peripheral edema (18.4%) [11]. In ALTA-1L, patients treated with brigatinib were found to have high incidences of diarrhea (58%), increased blood creatine phosphokinase (50%), and coughing (36%). In the observational study by Kim et al., the AE profiles for alectinib and brigatinib were comparable to each other, except for a significantly higher proportion of patients with elevated creatine phosphokinase and aspartate aminotransferase/alanine aminotransferase in the brigatinib cohort, and a significantly higher proportion of patients with elevated bilirubin in the alectinib cohort [18]. These findings suggest that treatment-related AEs and their management should be carefully considered when making treatment decisions in routine clinical practice.

Two recent studies have evaluated 1L ALK + mNSCLC treatment characteristics preferred by treating oncologists, patients, and caregivers. Mileham et al. conducted a discrete choice experiment study in the US and examined oncologists’ treatment preferences [41]. Their study showed that efficacy was the most important attribute when selecting 1L ALK + mNSCLC therapy, but there was significant heterogeneity in what aspect of efficacy was prioritized. While most oncologists (61%) prefer therapies with the longest 3-year PFS, 39% focused more on the intracranial duration of response. Beyond efficacy, treatment-related AEs (e.g., edema) and AE management were considered in the treatment selection. Oncologists valued the convenience of dosing and the availability of subsequent treatment options to a lesser extent. Another discrete choice experiment study by Culver et al. examined the treatment preference of 205 patients with stage 4 ALK + mNSCLC and 125 caregivers in the US. In their study, patients and caregivers were divided into two distinctive groups based on their preference patterns, “balanced traders” and “benefit maximizers” [42]. While "balanced traders" valued improving the chance of 3-year PFS along with lowering the risks of AEs, "benefit maximizers" chose treatments predominantly based on the chance of 3-year PFS [42]. Both studies emphasize the importance of clinical outcomes in treatment decision-making. However, when treatment alternatives yield similar clinical outcomes, as shown in our analysis, drug toxicity becomes a key factor in decision-making. In these cases, physicians, patients, and caregivers should engage in shared decision-making, accounting for potential AEs and their management when selecting the most appropriate therapy.

Some published studies have explored the healthcare costs for ALK + mNSCLC treatment. In Mudumba et al., mean total healthcare costs PPPM were $29,083, $27,292, and $35,414 in the alectinib, brigatinib, and lorlatinib cohorts, respectively [21]. Notably, the follow-up time in their analysis was from the start of 1L ALK TKI to death or loss to follow-up, which means the costs of second and later LOTs were also considered. In another study with the MarketScan claims database by Wu et al., the mean 1L all-cause healthcare costs within 12 months of the index ALK TKI therapy, PPPM, was $21,961. In their analysis, however, all patients treated with ALK TKIs were included in a single cohort and the majority (68.8%) of patients received crizotinib [22]. In addition to the previously mentioned difference in study design, the difference in payers captured by the claims data sources may also contribute to the different cost estimates. In our analysis of Komodo closed claims, 56.5% of the study patients had commercial insurance, 26.4% had Medicare Advantage, and 14.1% were covered by Medicaid, whereas the Mudumba et al., study leveraged Optum data and the Wu et al., study leveraged Marketscan data, and neither included patients with Medicaid. Costs among patients with Medicaid plans are often lower than costs among patients with commercial plans due to lower negotiated rates [43, 44].

Limitations

This study has several limitations. First, as with other real-world evidence studies, the EHR and claims data used were not originally collected for research purposes and therefore have inherent limitations. For example, the leveraged dataset, like other EHR and claims data sources, lacks genetic testing results and LOT information. In this analysis, ALK + mNSCLC was inferred based on the treatments received, consistent with other studies evaluating clinical and economics outcomes in ALK + mNSCLC [16, 21–23], and treatment patterns were characterized using a published LOT algorithm. The 180-day gap to define treatment discontinuation was based on the manual chart review validation of the EHR- and claims-based LOT algorithm; however, the use of such a long gap definition may overestimate LOT duration compared to other algorithms [26].

Because paid amounts were missing for many of the index medication pharmacy claims and had to be imputed from existing claims, real-world economic evaluations in this study are intended to be informative rather than definitive. Future studies should consider segmenting the other medical services costs to further explore the burden attributable to medication acquisition. It should be also noted that PPPM cost estimates may be inflated in patients with shorter follow-up times if diagnostic and treatment-initiation costs are concentrated early in therapy.

As a non-randomized observational study, the analysis was subject to unequal cohort sizes, differences in baseline characteristics across groups, differences in time to treatment windows, and limited and variable follow-up times. Although the imbalance in cohort size was reflective of real-world prescribing patterns during the study period, it may have reduced statistical power to detect between-group differences and increased the risk of type II error. Small sample sizes in brigatinib and lorlatinib groups also limited the feasibility of subgroup analyses (e.g., patients with brain metastasis at baseline). Next, while IPTW was applied to balance measured baseline characteristics, residual confounding from unmeasured factors may remain. Clinically important variables, such as ECOG performance status, could not be accounted for because they were not available in the dataset. Consequently, residual differences in disease severity due to patient channeling may persist and may not be fully mitigated by IPTW.

Interpretation of the study findings should also consider the possibility that the study design introduced an immortal time bias and excluded patients with more severe disease, particularly in the lorlatinib cohorts which had the longest median time from first metastasis to index. Further, given the limited follow-up by the data cutoff, most patients remained on their index ALK TKI therapy throughout the available observation period, precluding characterization of subsequent LOT and associated clinical outcomes. With these in mind, longer-term outcomes (e.g., rwOS) should be interpreted in light of the reduced opportunity to observe events, and the potential exclusion of sicker patients, which resulted in wide CIs and potential overestimates of rwOS. In addition, differences in follow-up across treatments could affect the frequency or intensity of patient monitoring and, in turn, the likelihood of a study endpoint (e.g., AE or progression criteria) being captured, introducing an unaccounted for censoring bias. However, as the number of outpatient office visits during the 1L treatment period was similar between cohorts, the potential for differential endpoint captures due to visit frequency is likely reduced.

In our descriptive AE analysis, to minimize the risk of misclassifying a recurrent condition as treatment-related toxicity, we assessed AEs only among patients without the condition of interest during the baseline period. This approach, together with reliance on diagnosis codes and a limited set of available laboratory measures, likely underestimates the true AE frequency, particularly among patients with a history of the condition who might be at higher risk of the AE. Accordingly, the AE analysis results should be interpreted as capturing incident events rather than providing estimates of overall AE frequency.

Finally, although the study included patients with Medicaid, Medicare, Medicare Advantage, and commercial coverage, the findings may not be applicable to uninsured patients or to individuals covered by plans not represented in the dataset (e.g., Veterans Health Administration). Given differences in healthcare system structures and access to care, the results should also be interpreted cautiously when applied outside the US. Furthermore, there were some limitations on the available demographics data. For instance, sex reflects the administrative field (not gender identity) and race may be incomplete or misclassified in EHR/claims, limiting subgroup inferences. Furthermore, only an aggregated “Asian” category was captured for race and this may mask regional differences.

Conclusions

In this real-world analysis of next-generation ALK TKI use among ALK TKI-naïve patients with ALK + mNSCLC, no statistically significant differences were detected in treatment effectiveness and all-cause mean healthcare costs between patients treated with alectinib, brigatinib, or lorlatinib in the main analysis. In the absence of head-to-head clinical trials or prior real-world studies directly comparing these three agents, our analysis offers the initial evidence from real-world data. By evaluating both clinical and economic outcomes, this study provides valuable, though underpowered, insights to inform treatment-decision making in routine clinical practice. Additional real-world studies with larger sample sizes and longer follow-up time are needed to support these findings.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Jyoti Malhotra, Elizabeth Marchlewicz, Yan Zhang, and Luis Hernandez contributed to the conceptualization of the study. Kenneth Pack and Isabelle Winer were responsible for the data curation, formal analysis and visualization of the data. Elizabeth Marchlewicz, Yan Zhang, Kenneth Pack and Isabelle Winer were responsible for preparing the first draft of the manuscript. Jyoti Malhotra, Elizabeth Marchlewicz, Yan Zhang, Isabelle Winer, Kenneth Pack, Christopher Danes, Hao Le, and Luis Hernandez contributed to developing the methodology and revising of the manuscript. All authors approved the final submitted version of the manuscript.

Funding

This study and the associated publication Rapid Service and Open Access fees were funded by Takeda Pharmaceuticals America, Inc.

Data Availability

The datasets analyzed during the current study are not publicly available as they were used under license from Veradigm and Komodo Health. Due to data use agreements and its proprietary nature, restrictions apply regarding the availability of the data. Further information is available from the corresponding author.

Declarations

Conflict of Interest

Elizabeth Marchlewicz, Isabelle Winer, and Kenneth Pack are employees of Veradigm, which received funding from Takeda Pharmaceuticals America, Inc. to conduct this study. Yan Zhang, Christopher Danes, Hoa Le, and Luis Hernandez and employees and own stock of Takeda. Jyoti Malhotra is paid consultant at Pfizer and Takeda, received grants or contracts from Astra Zeneca, and participated in an advisory board held by Takeda.

Ethical Approval

The linked dataset only contained de-identified data as per the de-identification standard defined in Section §164.514(a) of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rule. The process by which the data were de-identified is attested to through a formal determination by a qualified expert as defined in Section §164.514(b) (1) of the HIPAA Privacy Rule. Because this study used only de-identified patient records, it was no longer subject to the HIPAA Privacy Rule and was therefore exempt from institutional review board approval and obtaining informed consent according to US law. The study was designed, implemented, and reported in accordance with the International Society for Pharmacoepidemiology Guidelines for Good Pharmacoepidemiology Practice, and was conducted in compliance with the ethical principles laid down in the Declaration of Helsinki and only used de-identified data.