Health care use and costs associated with clinically impactful immune-related adverse events during immune checkpoint inhibitor therapy for non–small cell lung cancer

Abstract

BACKGROUND:

Immune checkpoint inhibitors (ICIs) are widely used in advanced non–small cell lung cancer (NSCLC) but can lead to immune-related adverse events (irAEs) that may disrupt care. Although irAEs are well described clinically, there is limited evidence quantifying their downstream health care use and costs in Medicare beneficiaries with advanced NSCLC.

OBJECTIVE:

To evaluate health care use and costs associated with clinically impactful irAEs among older adults receiving ICI therapy.

METHODS:

Using Surveillance, Epidemiology, and End Results–Medicare data, we identified patients (fee-for-service beneficiaries) aged 66 to 85 years with advanced NSCLC and who initiated nivolumab, pembrolizumab, or atezolizumab, alone or in combination with each other or chemotherapy, between 2015 and 2017 (patient identification window), with follow-up through 2019. These agents represent ICIs approved for advanced NSCLC during the identification period. Clinically impactful irAEs were defined using irAE diagnosis codes accompanied by evidence of systemic immunosuppressant use and treatment interruption and were modeled as a time-varying exposure. Health care use and costs were assessed using weighted longitudinal models accounting for time-varying confounding (marginal structural models with stabilized inverse probability weights). Two-part generalized estimating equations estimated per-patient-per-month (PPPM) and cumulative outcomes over 6, 12, and 24 months, with sensitivity analyses including models incorporating ICI drug costs and alternative approaches to account for censoring.

RESULTS:

Among 4,867 patients, 1,667 (34.3%) experienced a clinically impactful irAE. Adjusted all-cause health care use was higher among patients with irAEs (mean 15.8 [SE = 0.12] vs 12.0 [SE = 0.12] visits PPPM; difference, 3.9; 95% CI = 2.9-4.8), with more than double the odds of inpatient hospitalization (odds ratio, 2.21; 95% CI = 2.15-2.29). Adjusted all-cause medical costs averaged $16,042 (SE = 14.7) vs $12,721 (SE = 13.1) PPPM (difference, $3,322; 95% CI = $3,283-$3,361), driven primarily by inpatient care ($2,922; 95% CI = $2,903-$2,941). At 24 months, cumulative costs per patients were $846,013 (SE = 12,786) for patients with irAEs compared with $751,606 (SE = 11,358) for those without (difference, $94,407; 95% CI = $91,606-$97,207). Findings were consistent in sensitivity analyses.

CONCLUSIONS:

Clinically impactful irAEs were associated with substantially higher health care use and costs among older adults with advanced NSCLC receiving ICIs, with inpatient care accounting for most excess costs. These findings have implications for managed care strategies focused on toxicity monitoring, hospitalization prevention, and value-based ICI management.

Plain language summary

Immune checkpoint inhibitors (ICIs) are a type of medication used to treat advanced non–small cell lung cancer. These medications can cause immune-related adverse events (irAEs), which may be serious. This study found that irAEs occurred in approximately 1 in 3 older patients. Patients with irAEs had more hospital stays and medical visits and had higher health care use and costs. These findings show that irAEs cause care and cost challenges in routine cancer treatment.

Implications for managed care pharmacy

This study highlights increased health care use and costs associated with clinically impactful irAEs in older adults receiving ICIs for advanced non–small cell lung cancer. Hospitalizations were the main driver of excess costs. These results can inform coverage policies, care coordination, and toxicity monitoring strategies. The findings support value-based care models and cost forecasting to balance the clinical benefits of ICIs with the economic impact of irAE-related hospital and long-term care.

Immune checkpoint inhibitors (ICIs) have transformed the treatment of advanced non–small cell lung cancer (NSCLC), improving survival and enabling durable disease control across multiple disease settings.^1–5 Because their use has expanded in routine practice, immune-related adverse events (irAEs) have emerged as a central challenge in oncology care delivery.^6–8 ICIs act by blocking inhibitory immune checkpoints, thereby enhancing T-cell–mediated antitumor activity.⁴ This immune activation can also disrupt normal immune tolerance, leading to inflammatory toxicities affecting healthy tissues across multiple organ systems, which may require immunosuppressive therapy and, in severe cases, hospitalization.^8,9 These toxicities can involve nearly any organ system and range from mild to life-threatening, often necessitating immunosuppressive therapy, specialty consultation, or hospitalization.^9,10 Underlying immune dysregulation, including preexisting autoimmune conditions, may influence susceptibility to irAEs, given the mechanism of immune checkpoint inhibition.¹¹ As a result, clinically impactful irAEs represent not only a toxicity management issue but also a potential inflection point for health care use and costs during treatment with ICIs, for which these immune-mediated toxicities are most well characterized.^9,10,12–14

Despite growing clinical experience with irAEs, evidence characterizing their downstream health care use and cost implications remains limited. Prior studies, largely based on integrated health systems or pooled administrative claims, suggest higher hospitalization rates and short-term costs among patients experiencing irAEs.^12–16 However, many analyses combine heterogeneous tumor types, include relatively small samples, and rely on short follow-up periods. In addition, few studies align person-time to irAE onset or account for time-varying exposure, raising the potential for misattribution of use and costs. Evidence derived from commercially insured or Medicare Advantage populations may also have limited generalizability to fee-for-service Medicare beneficiaries, given Medicare’s central role in financing cancer care in the United States.¹⁷

Older adults with advanced NSCLC represent a particularly relevant population in which to examine the economic consequences of irAEs. Adults aged 65 years or older account for approximately two-thirds of NSCLC diagnoses and make up a substantial proportion of patients receiving ICIs in routine clinical practice, with up to approximately 64% of treated patients aged at least 65 years in real-world studies.^1,18–20 Age-related immune changes, higher multimorbidity, and differences in toxicity recognition and management may further amplify health care use and costs following clinically significant irAEs in this group.^21,22 Despite this, population-based estimates of the incremental health care use and costs attributable to clinically impactful irAEs among Medicare beneficiaries with advanced NSCLC are lacking.

To address this gap, we conducted a retrospective cohort study using linked Surveillance, Epidemiology, and End Results (SEER)-Medicare data to quantify the all-cause health care use and costs associated with clinically impactful irAEs among older adults with advanced NSCLC treated with ICIs. By focusing on irAEs that required escalation of care and applying longitudinal analytic methods that account for time-varying exposure, this study provides population-based estimates of the economic impact of clinically impactful irAEs in fee-for-service Medicare beneficiaries with advanced NSCLC.

Methods

DATA SOURCE

We used linked data from the SEER program and Medicare fee-for-service claims. SEER provided cancer diagnosis and tumor characteristics, whereas Medicare inpatient, outpatient, carrier, durable medical equipment, home health, and part D files captured health care use and costs. Beneficiary enrollment information was obtained from the Medicare Master Beneficiary Summary File. These data sources were selected to comprehensively capture cancer care, treatment-related complications, and downstream medical management among older adults. The study was approved by the University of Houston’s institutional review board and followed the Strengthening the Reporting of Observational Studies in Epidemiology guideline.²³

STUDY DESIGN AND SAMPLE

We conducted a retrospective longitudinal cohort study (Figure 1). Adults aged 66 to 85 years with advanced or metastatic NSCLC (mNSCLC) diagnosed between 2010 and 2017 who initiated ICI therapy between March 2015 and December 2017 were identified. This identification window aligns with the initial approvals of ICIs for advanced NSCLC^20,24 and reflects the SEER-Medicare data available at the time of analysis, allowing sufficient follow-up through 2019 to assess downstream outcomes. NSCLC was defined using SEER primary site and histology codes (Supplementary Table 1 ^{(625.9KB, pdf)} , available in online article), with advanced disease defined by regional or distant stage at diagnosis. Eligible ICIs included nivolumab, pembrolizumab, and atezolizumab (Supplementary Table 2 ^{(625.9KB, pdf)} ), which represented agents approved for advanced NSCLC during the patient identification period.²⁴ Patients receiving ICIs as monotherapy or in combination with other systemic therapies, including chemotherapy or other immunotherapies, were not excluded and were captured as part of routine clinical practice.

FIGURE 1.

Study Design and Timeline

Study timeline and analytic windows. The index date (time 0) was the first ICI claim. Follow-up was organized in monthly intervals with health care visits and costs aggregated to person-months and cumulatively summarized over prespecified windows (0-6, 0-12, and 0-24 months). Follow-up ended at the earliest of death, loss of Medicare fee-for-service parts A/B or part D coverage (including Medicare Advantage enrollment), diagnosis of a second primary cancer, reaching age 85 years, last observed claim, or study end (December 31, 2019). Baseline covariates were assessed at index and controlled for using stabilized inverse probability of exposure and censoring weights; clinically impactful irAEs were modeled as a time-varying exposure during follow-up. Weighted means and 95% CIs are reported for cumulative outcomes.

HRU = health care resource use; ICI = immune checkpoint inhibitor; irAE = immune-related adverse event.

The study index date was the first observed ICI claim. Patients were required to have continuous enrollment in Medicare parts A, B, and D for at least 12 months before ICI initiation. Baseline characteristics were assessed during the 12 months before the index date. Patients aged 85 years or older were excluded to reduce incomplete capture of comorbidity and cancer care.^25,26 Study follow-up began at ICI initiation and continued until death; loss of Medicare parts A, B, and D coverage; diagnosis of a second primary cancer; or study end (December 31, 2019), whichever occurred first. Continuous fee-for-service enrollment in parts A, B, and D was required for complete capture of ICI administration (primarily part B outpatient claims), irAE-related hospitalizations (part A inpatient claims), and immunosuppressive medications (part D prescription drug claims).

EXPOSURE

The study exposure was developing a clinically impactful irAE after ICI initiation. irAEs were identified using diagnosis codes from Medicare claims based on established literature and clinical guidelines (Supplementary Table 3 ^{(625.9KB, pdf)} ).^9,27,28 Because clinical trial–based toxicity grading (eg, Common Terminology Criteria for Adverse Events) is not available in administrative claims data, we used a claims-based proxy designed to capture irAEs that are clinically consequential in routine practice.²⁹ To identify clinically impactful events, irAEs were required to meet at least 1 of the following criteria: (1) use of a systemic immunosuppressant within 14 days of an irAE diagnosis or (2) interruption of ICI therapy defined by a dosing gap exceeding 50% of the expected cycle length.^7,9,27,30 These criteria were selected to identify irAEs that prompted escalation of medical management or treatment modification, consistent with moderate-to-severe toxicities that would typically require intervention in clinical practice and are aligned with guideline-based management of irAEs,⁹ which typically involves immunosuppressive therapy and, in more severe cases, hospitalization and increased health care use.

This approach has been used in prior real-world studies to approximate clinically meaningful irAEs.^7,30 Exposure to a clinically impactful irAE (referred to as “irAEs” hereinafter unless otherwise specified) was modeled as a time-varying variable, with onset defined as the first month in which a qualifying irAE was observed. Modeling irAEs as a time-varying exposure accounts for the dynamic timing of toxicity onset during treatment and reduces bias that would arise from classifying patients as exposed before the event occurs.

OUTCOMES

The primary outcomes were all-cause health care use and costs measured on a per-patient-per-month (PPPM) basis. Use was assessed as counts of inpatient, outpatient, and physician visits as well as additional health care services, including pharmacy dispensing events, durable medical equipment claims, and home health services. These categories were included to more comprehensively capture payer-relevant resource use in older adults with advanced NSCLC, for whom frailty, functional decline, and supportive care needs may extend beyond traditional visit-based encounters.^22,31,32 Costs reflected total Medicare payments and beneficiary liabilities and were aggregated by care setting and month. Secondary outcomes included (1) cumulative mean use and costs, defined as the sum of monthly visit counts and costs from index through prespecified horizons (6, 12, and 24 months) to capture short-, intermediate-, and longer-term burden during typical ICI treatment and follow-up,³³ and (2) irAE-related use and costs, defined as services and payments associated with claims containing qualifying irAE diagnosis codes; irAE-related outcomes were analyzed separately to characterize care directly attributable to irAE management.

STATISTICAL ANALYSIS

Baseline characteristics were summarized descriptively using means (SDs) for continuous variables and frequencies (percentages) for categorical variables. Longitudinal use and cost outcomes were analyzed using marginal structural models with stabilized inverse probability of exposure and censoring weights to account for time-varying exposure (irAE), confounding (time-varying steroid use for use as a proxy for underlying clinical severity and toxicity management), and censoring (loss of eligibility, death, hospice enrollment).^34,35 Weighted 2-part models were used to estimate monthly use and costs, accounting for these weights. In the first part, weighted logistic generalized estimating equations were used to estimate the odds of any use or cost in a given month, including the odds of any inpatient hospitalization and any outpatient visit. In the second part, conditional on nonzero values, use counts were modeled using negative binomial regression, and costs were modeled using gamma regression with a log link to accommodate overdispersion and right-skewness. Marginal means were derived to estimate PPPM and cumulative outcomes. In the primary analysis, ICI treatment status was treated as a time-varying confounder to account for treatment interruption or discontinuation following irAE occurrence, which can influence subsequent health care use, costs, and censoring. ICI drug acquisition costs were excluded to isolate downstream medical spending associated with irAE management and to avoid masking these effects with mechanically reduced drug costs after treatment interruption.

Sensitivity Analyses

As sensitivity analyses, ICI drug costs were included to estimate total health care costs, and Lin’s weighted regression was used as an alternative to generalized estimating equations to estimate cost differences while accounting for right censoring due to death or Medicare disenrollment.³⁶ In an additional sensitivity analysis, we excluded patients with preexisting autoimmune diseases to assess the robustness of findings to baseline differences in immune-related risk. Descriptive trend plots were also used to illustrate adjusted PPPM ICI and non-ICI medical costs, including inpatient and outpatient costs, relative to the month of irAE onset.

All costs were inflation adjusted to 2019 US dollars using the medical care component of the Consumer Price Index, consistent with the final year of available data. Robust SEs were used to account for within-patient correlation over time, and 95% CIs were used to quantify statistical uncertainty for all model-based estimates, with statistical significance defined at the 0.05 level. Thus, 95% CIs not crossing zero indicate statistically significant differences at an α of 0.05. Statistical analyses were performed in R version 4.3.0 (R Foundation for Statistical Computing).

Results

COHORT CHARACTERISTICS

The study included 4,867 fee-for-service Medicare beneficiaries with advanced NSCLC (mean age at ICI initiation, 74.3 [SD = 4.8] years; 2,560 [SD = 52.6%] female) (Supplementary Figure 1 ^{(625.9KB, pdf)} ). The cohort was predominantly White (3,972 [SD = 81.6%]), with 342 (SD = 7.0%) Black patients, 212 (SD = 4.4%) Hispanic patients, 242 (SD = 5.0%) Asian or Pacific Islander patients, and 99 (SD = 2.0%) individuals of other race or ethnicity (Table 1). Overall, 1,667 patients (SD = 34.3%) experienced a clinically impactful irAE during follow-up.

TABLE 1.

Cohort Characteristics

Characteristics	Total (n = 4,867)	irAEa (n = 1,667)	No irAE (n = 3,200)	P valueb
Age, mean (SD), years	74.3 (4.8)	74.3 (4.9)	74.2 (4.8)	0.66
Sex, n (%)
Female	2,560 (52.6)	892 (53.5)	1,668 (52.1)	0.36
Male	2,307 (47.4)	775 (46.5)	1,532 (47.9)
Race and ethnicityc, n (%)
White	3,972 (81.6)	1,393 (83.6)	2,579 (80.6)	0.13
Black	342 (7.0)	109 (6.5)	233 (7.3)
Hispanic	212 (4.4)	65 (3.9)	147 (4.6)
Asian/Pacific Islander	242 (5.0)	73 (4.4)	169 (5.3)
Others	99 (2.0)	27 (1.6)	72 (2.3)
Medicare-Medicaid dual eligibility, n (%)
Yes	1,142 (23.5)	372 (22.3)	770 (24.1)	0.17
No	3,725 (76.5)	1,295 (77.7)	2,430 (75.9)
Original reason for enrollment, n (%)
Old age	4,186 (86.0)	1,417 (85.0)	2,769 (86.5)	0.15
Disability/end-stage renal disease	681 (14.0)	250 (15.0)	431 (13.5)
SEER region, n (%)
Midwest	394 (8.1)	152 (9.1)	242 (7.6)	0.07
Northeast	2,053 (42.2)	707 (42.4)	1,346 (42.1)
South	961 (19.8)	341 (20.5)	620 (19.4)
West	1,459 (30.0)	467 (28.0)	992 (31.0)
Metropolitan statistical area, n (%)
Metropolitan	4,173 (85.7)	1,418 (85.1)	2,755 (86.1)	0.33
Nonmetropolitan	694 (14.3)	249 (14.9)	445 (13.9)
Census-tract poverty level, n (%)
Low (0%-<5%)	1,106 (22.7)	417 (25.0)	689 (21.5)	0.08
Moderate (5%-9.9%)	1,134 (23.3)	373 (22.4)	761 (28.8)
High (10%-19.9%)	1,234 (25.4)	421 (25.3)	813 (25.4)
Very high (≥20%)	739 (15.2)	240 (14.4)	499 (15.6)
Unknown	654 (13.4)	216 (13.0)	438 (13.7)
Histology, n (%)
Adenocarcinoma	2,962 (60.9)	1,009 (60.5)	1,953 (61.0)	0.16
Squamous cell carcinoma	1,398 (28.7)	500 (30.0)	898 (28.1)
Other histologic types	507 (10.4)	158 (9.5)	349 (10.9)
Staged, n (%)
Regional	1,523 (31.3)	546 (32.8)	977 (30.5)	0.11
Distant	3,344 (68.7)	1,121 (67.3)	2,223 (69.5)
Metastatic site, n (%)
Brain	425 (8.7)	122 (7.3)	303 (9.5)	0.03
Bone	816 (16.8)	238 (14.3)	578 (18.1)	0.002
Liver	1,339 (27.5)	88 (5.3)	225 (7.0)	0.05
Charlson Comorbidity Score, mean (SD)	2.3 (1.9)	2.3 (2.0)	2.2 (1.9)	0.47
Frailty, n (%)
Nonfrail	629 (12.9)	204 (12.2)	425 (13.3)	0.59
Prefrail	2,493 (51.2)	869 (52.1)	1,624 (50.8)
Mildly frail	1,369 (28.1)	472 (28.3)	897 (28.0)
Moderate to severely frail	376 (7.7)	122 (7.3)	254 (7.9)
Preexisting autoimmune condition, n (%)	861 (17.7)	348 (20.9)	513 (16.0)	<0.0001
Prior hospitalization, n (%)	4,239 (87.1)	1,425 (85.5)	2,814 (87.9)	0.002
Surgery, n (%)	1,787 (36.7)	629 (37.7)	1,158 (36.2)	0.29
Radiation, n (%)	3,112 (63.9)	1,019 (61.1)	2,093 (65.4)	0.003
Chemotherapy, n (%)	2,735 (56.2)	952 (57.1)	1,783 (55.7)	0.35
Targeted therapy, n (%)	992 (20.4)	348 (20.9)	644 (20.1)	0.54
Immunosuppressants, n (%)	420 (8.6)	174 (10.4)	246 (7.7)	0.001
Months from diagnosis to ICI initiation, n (%)	15.3 (15.4)	15.5 (15.3)	15.2 (15.5)	0.11
Line of ICI therapy, n (%)
1L	806 (16.6)	258 (15.5)	548 (17.1)	0.42
2L	2,170 (44.6)	763 (45.8)	1,407 (44.0)
3L	1,108 (22.8)	374 (22.4)	734 (22.9)
4L+	783 (16.1)	272 (16.3)	511 (16.0)
ICI type, n (%)
Nivolumab	2,829 (58.1)	957 (57.4)	1,872 (58.5)	0.68
Pembrolizumab	1,784 (36.7)	625 (37.5)	1,159 (36.2)
Atezolizumab	254 (5.2)	85 (5.1)	169 (5.3)
ICI/chemotherapy combination, n (%)
Yes	696 (14.3)	231 (13.9)	465 (14.5)	0.52
No	4,171 (85.7)	1,436 (86.1)	2,735 (85.5)

Percentages may not sum to 100 because of rounding.

^a Clinically impactful irAEs were considered, defined as irAE-coded claims accompanied by systemic immunosuppressant use within 14 days and ICI treatment disruption (cycle gap >50% of the expected dosing interval).

^b t-tests were used to assess significant differences for continuous variables, and Pearson chi-square tests were assessed for categorical variables. Characteristics are shown for descriptive purposes; covariate balance after weighting was assessed using standardized mean differences.

^c Race and ethnicity: White, Black, and Asian/Pacific Islander refer to non-Hispanic individuals. “Other” includes non-Hispanic American Indian or Alaska Native, other racial groups, or individuals of multiple races.

^d Stage: Cancer stage was classified using the SEER Summary Stage variable and obtained from SEER registry data at the time of initial diagnosis.

ICI = immune checkpoint inhibitor; irAE = immune-related adverse event; SEER = Surveillance, Epidemiology, and End Results.

In unweighted analyses, the proportion of patients with preexisting autoimmune disease was higher among those who developed irAEs compared with those who did not (20.9% vs 16.0%; P < 0.001), as was the proportion with prior immunosuppressant use (10.4% vs 7.7%; P = 0.001). In contrast, the proportion of patients with brain metastases (7.3% vs 9.5%; P = 0.03) and bone metastases (14.3% vs 18.1%; P = 0.002) was lower among patients who developed irAEs. After application of inverse probability weights, covariate balance improved substantially, with all absolute standardized mean differences below 0.1 (Supplementary Table 4 ^{(625.9KB, pdf)} ). The median follow-up time was 13 months (IQR = 4-19 months), with a maximum follow-up of 54 months.

HEALTH CARE USE

In adjusted analyses, patients with irAEs had higher PPPM all-cause health care use than those without irAEs (mean, 15.8 [SE = 0.12] vs 12.0 [SE = 0.12] visits; adjusted difference, 3.9 [95% CI = 2.9-4.8]) (Figure 2; Supplementary Table 5 ^{(625.9KB, pdf)} ). Patients with irAEs had higher odds of inpatient hospitalization (odds ratio, 2.21 [95% CI = 2.15-2.29]; P < 0.001) and outpatient visits (odds ratio, 1.20 [95% CI = 1.15-1.24]; P < 0.001).

FIGURE 2.

Adjusted per-Patient-per-Month Mean All-Cause Health Care Use Among Patients With and Without Clinically Impactful irAEs

Estimates are adjusted using marginal structural models with stabilized inverse probability of exposure and censoring weights, with irAE status modeled as a time-varying exposure.

irAE = immune-related adverse event.

Cumulative mean use increased over follow-up in both groups but remained higher among patients with irAEs. Use reflects aggregated health care services, including visits, pharmacy dispensing events, durable medical equipment, and home health services. At 24 months, cumulative use averaged 273.6 (SE = 9.9) among patients with irAEs compared with 213.1 (SE = 8.9) among those without (adjusted difference, 60.5 [95% CI = 60.5-71.4]), driven primarily by outpatient and physician encounters (Table 2). Among patients who experienced irAEs, adjusted irAE-related use averaged 2.9 (SE = 3.8) visits PPPM, with outpatient encounters accounting for the largest share, followed by inpatient and pharmacy services (Supplementary Figure 2 ^{(625.9KB, pdf)} ).

TABLE 2.

Adjusted Cumulative Mean All-Cause Health Care Use and Costs Among Patients With and Without irAEs at 6, 12, and 24 Months After ICI Initiation

Outcome	6 mo			12 mo			24 mo
	irAE	No irAE	Difference (95% CI)	irAE	irAE	Difference (95% CI)	irAE	No irAE	Difference (95% CI)
Health care use, mean (SE)
Total medical visits	72.77 (5.78)	56.68 (4.68)	16.09 (17.75-18.22)	167.90 (9.56)	130.76 (8.67)	37.13 (34.74-35.80)	273.57 (9.89)	213.05 (8.86)	60.51 (60.49-71.38)
Inpatient visits	2.89 (0.02)	1.48 (0.01)	1.41 (0.46-2.37)	3.62 (1.23)	1.87 (1.11)	1.75 (0.86-2.64)	4.65 (3.22)	2.41 (1.23)	2.24 (1.37-3.11)
Health care costs, mean (SE)
Total medical costs	$485,539 ($4,197)	$431,411 ($3,729)	$54,129 ($53,211-$55,046)	$673,182 ($7,932)	$598,124 ($7,048)	$75,057 ($73,323-$76,792)	$846,013 ($12,786)	$751,606 ($11,358)	$94,407 ($91,606-$97,207)
Inpatient costs	$143,533 ($1,268)	$49,255 ($439)	$94,277 ($92,650-$95,904)	$186,698 ($2,094)	$63,485 ($708)	$123,213 ($120,496-$125,931)	$218,715 ($2,917)	$73,722 ($967)	$144,993 ($141,169-$148,816)

Estimates are adjusted using marginal structural models with stabilized inverse probability of exposure and censoring weights. irAE status was modeled as a time-varying exposure. Clinically impactful irAEs were defined as irAE-coded claims accompanied by systemic immunosuppressant use within 14 days and ICI treatment disruption (cycle gap >50% of the expected dosing interval). Cumulative outcomes represent adjusted mean totals per patient over each follow-up window. All costs were inflation adjusted to 2019 US dollars using the Medicare component of the Consumer Price Index.

ICI = immune checkpoint inhibitor; irAE = immune-related adverse event; mo = month.

HEALTH CARE COSTS

Adjusted all-cause PPPM costs were higher among patients with irAEs than among those without ($16,042 vs $12,720; adjusted difference, $3,322 [95% CI = $3,283-$3,360]) (Figure 3; Supplementary Table 5 ^{(625.9KB, pdf)} ). Inpatient care accounted for the largest component of excess monthly costs (adjusted difference, $2,922 [95% CI = $2,903-$2,941]) (Figure 4).

FIGURE 3.

Adjusted per-Patient-per-Month Mean All-Cause Health Care Costs Among Patients With and Without Clinically Impactful irAEs

Estimates are adjusted using marginal structural models with stabilized inverse probability of exposure and censoring weights, with irAE status modeled as a time-varying exposure. Costs were inflation adjusted to 2019 US dollars using the Medical Care component of the Consumer Price Index.

irAE = immune-related adverse event.

FIGURE 4.

Incremental PPPM Cost Differences by Health Care Setting Among Patients With and Without Clinically Impactful irAEs

Estimates are adjusted using marginal structural models with stabilized inverse probability of exposure and censoring weights, with irAE status modeled as a time-varying exposure. Costs were inflation adjusted to 2019 US dollars using the Medical Care component of the Consumer Price Index.

DME = durable medical equipment; irAE = immune-related adverse event; PPPM = per-patient-per-month.

Cumulative mean costs increased over time in both groups but remained consistently higher among patients with irAEs. At 24 months, mean cumulative costs were $846,013 (SE = $12,786) for patients with irAEs and $751,606 (SE = $11,358) for patients without irAEs, yielding an adjusted difference of $94,407 (95% CI = $91,606-$97,207) (Table 2). Across time horizons, inpatient services were the dominant contributor to cumulative cost differences. Adjusted irAE-related costs averaged $2,242 (SE = $527) PPPM, with inpatient services representing the largest component (mean, $1,096 [SE = $136]) (Supplementary Figure 2 ^{(625.9KB, pdf)} ).

SENSITIVITY ANALYSES

When ICI drug costs were included, total PPPM cost differences were attenuated (adjusted difference, −$587 [95% CI = −$590 to −$585]), likely reflecting reduced ICI exposure and associated drug spending among patients with irAEs because of treatment interruption or discontinuation following toxicity. However, inpatient costs remained higher among patients with irAEs (adjusted difference, $1,103 [95% CI = $1,097-$1,110]) (Supplementary Table 6 ^{(625.9KB, pdf)} ; Supplementary Figure 3 ^{(625.9KB, pdf)} ). The descriptive event-time analyses demonstrated a decline in outpatient and ICI drug spending following irAE onset but sustained increases in inpatient costs after an irAE (Supplementary Figure 4 ^{(625.9KB, pdf)} ).

Using Lin’s weighted regression for censored cost data, results were consistent with the primary marginal structural model analyses (Supplementary Table 7 ^{(625.9KB, pdf)} ). During treatment months, patients with clinically impactful irAEs had higher adjusted all-cause costs than those without irAEs (mean difference, $1,236 PPPM; 95% CI = $1,211-$1,261), driven primarily by higher inpatient spending (difference, $2,255 PPPM; 95% CI = $2,238-$2,272), despite lower outpatient costs. When accounting for gaps in ICI therapy, total adjusted all-cause costs were modestly lower among patients with irAEs (difference, −$247 PPPM; 95% CI = −$252 to −$243), reflecting reduced outpatient and carrier spending, whereas excess inpatient costs persisted (difference, $1,207 PPPM; 95% CI = $1,205-$1,209).

After excluding patients with preexisting autoimmune disease, findings remained consistent with the primary analysis. During treatment months, patients with irAEs had higher total all-cause costs (adjusted difference, $1,696 PPPM; 95% CI = $1,603-$1,791) driven by higher inpatient spending (adjusted difference, $1,004; 95% CI = $956-$1,063) despite lower outpatient costs (adjusted difference, −$1,455; 95% CI = −$1,478 to −$1,432). When accounting for ICI treatment gaps, total all-cause costs were lower among patients with irAEs (adjusted difference, −$1,900 PPPM; 95% CI = −$1,956 to −$1,845), whereas excess inpatient costs persisted (adjusted difference, $252; 95% CI = $213-$293) (Supplementary Table 8 ^{(625.9KB, pdf)} ).

Discussion

In this population-based study of fee-for-service Medicare beneficiaries with advanced NSCLC treated with ICIs, clinically impactful irAEs occurred in 1 in 3 patients and were associated with substantial increases in health care use and costs. The observed irAE frequency is consistent with Cook et al, who reported that 37% of 803 patients with mNSCLC receiving ICIs developed clinically meaningful irAEs defined by treatment delay or discontinuation and systemic corticosteroid use.⁷ Our economic findings also align with prior real-world evidence. George et al reported that patients with adverse events (AEs) suspected to be immune related had more than double the risk of hospitalization and $24,301 higher 6-month costs, with inpatient visits accounting for nearly 90% of AE-related costs.¹³ Similarly, Gunturu et al found higher rates of irAEs, all-cause hospitalizations, and ED visits among commercial and Medicare Advantage patients receiving ICIs, with combination therapy amplifying these differences.¹⁶ Notably, the economic burden of irAEs differs from that of conventional chemotherapy-related toxicities, which are typically hematologic and gastrointestinal in nature, more predictable in onset, and managed with established supportive care protocols.^9,37 In contrast, irAEs are immunologically driven, can affect virtually any organ system, and often require prolonged immunosuppressive therapy and multidisciplinary management, contributing to the outsized role of inpatient costs observed in our study and others.¹⁹ The relevance of these findings is underscored by the expanding reach of ICIs in oncology practice. By 2018, an estimated 44% of patients with cancer were eligible for ICI therapy, with NSCLC indications contributing most to this increase.³⁸ ICIs have since become a cornerstone of treatment for advanced NSCLC without actionable genomic alterations, used as monotherapy or in combination with chemotherapy across treatment lines.^39,40 As ICI use continues to expand, the irAE-related use and cost burden documented here has broad implications for health systems, payers, and value-based oncology models. In our study, patients who experienced irAEs had higher use of nearly all health care services, with inpatient care emerging as the dominant contributor to both monthly and cumulative costs. This pattern persisted over extended follow-up and remained evident even when ICI drug costs were included, indicating sustained downstream consequences of irAE-related care.

Building on prior observational evidence linking irAE to increased hospitalization and short-term medical spending,^12,13,41,42 this study extends the literature by providing a focused, longitudinal assessment of the use and cost implications of clinically impactful irAEs in advanced NSCLC. By restricting the analysis to a single, high-burden cancer type and applying a longitudinal causal framework that accounts for time-varying exposure and censoring, we more precisely characterize downstream health care use and costs associated with irAEs over the course of treatment and follow-up. Earlier work has documented increased hospitalization and short-term medical spending following irAEs, often reporting that 20% to 40% of patients treated with ICIs require inpatient care for complications such as pneumonitis, colitis, or endocrine dysfunction.^12,13,41 In this Medicare population of adults aged 66 to 85 years, clinically impactful irAEs were associated with substantially higher health care use, averaging almost $4,000 per patient on a monthly basis, with inpatient services accounting for the largest share (∼80%) of both monthly and cumulative cost differences. To contextualize this burden, real-world AE-related costs for first-line cytotoxic chemotherapy in mNSCLC have been reported at approximately $6,323 PPPM, as compared with $4,259 PPPM for immunotherapy, with patients who are immunotherapy treated being 40% less likely to incur high AE costs.⁴² In a SEER-Medicare analysis of second-line chemotherapy for advanced NSCLC, AE-related costs averaged $1,036 PPPM, representing approximately 10% of total treatment costs.⁴³ Among patients receiving epidermal growth factor receptor–targeted therapy, total monthly treatment costs of approximately $20,000 have been reported, with hospitalization alone accounting for 47% of spending, highlighting that inpatient care is a major cost driver across treatment modalities in advanced NSCLC.⁴⁴ However, these estimates largely reflect broadly defined AEs. Our study focused specifically on clinically impactful irAEs—those requiring immunosuppression, treatment interruption, or hospitalization—which may account for the higher per-event cost increment observed. The approximately $4,000 PPPM attributable to clinically impactful irAEs underscores that, although irAEs occur less frequently than chemotherapy-related toxicities, their severity and management complexity, driven predominantly by hospitalization, impose a disproportionately high economic burden, reinforcing the need for targeted strategies to mitigate irAE-driven inpatient spending. Our population-based design and extended follow-up further demonstrate that these economic consequences are not confined to the period immediately surrounding toxicity management but continue to accrue over time among older adults treated with ICIs. Our finding likely reflects both the acuity of severe irAEs and the limited physiologic reserve of patients who are older, who may require hospitalization for management of complications that could otherwise be addressed in outpatient settings among younger populations.^22,25,45 Although outpatient visits and pharmacy services also increased following irAE onset, their relative contribution to total costs was modest compared with inpatient care. The steady growth in pharmacy spending over time among patients with irAEs may reflect, in part, ongoing management following toxicity, including prolonged immunosuppressive or supportive care. However, this trend is likely multifactorial, being that pharmacy spending generally increases over time in older populations because of accumulating comorbidities and polypharmacy,⁴⁶ and our data do not allow us to distinguish irAE-specific pharmacy costs from broader age-related trends. These findings highlight the importance of guideline-based irAE management, which emphasizes early recognition and appropriate outpatient treatment as potential opportunities to optimize care and reduce downstream resource use.

Including ICI drug acquisition costs attenuated or reversed differences in total spending between patients with and without irAEs, likely reflecting reduced ICI exposure because of treatment interruption or discontinuation following irAEs and underscoring the dominant role of drug pricing in oncology expenditures.^47–50 Importantly, excess inpatient costs attributable to irAEs persisted regardless of drug cost inclusion, indicating that irAE-related medical care represents an independent source of economic burden. This distinction is relevant for health economic evaluations and value-based oncology models, in which high drug costs may obscure downstream medical consequences of treatment-related toxicities.

Despite the economic impact of irAEs documented here, ICIs remain a favorable treatment option from a toxicity-cost perspective. In a retrospective claims analysis of nearly 10,000 patients initiating first-line nontargeted therapy for mNSCLC, Engel-Nitz et al found that patients receiving immunotherapy experienced fewer AEs and significantly lower total AE-related costs compared with those receiving cytotoxic chemotherapy ($16,319 vs $23,009; P < 0.001) and were 40% less likely to incur high AE costs.⁴² However, these findings should be interpreted in the context of a relatively small immunotherapy cohort, potential channeling of patients to treatment based on unmeasured clinical factors such as performance status, claims-based AE identification without causality determination, and the use of a broad AE definition that did not distinguish clinically meaningful irAEs, such as those requiring immunosuppression, treatment interruption, or hospitalization, from milder, self-limiting events. Nonetheless, taken together with our findings, these data suggest that although irAEs impose a meaningful economic burden when they occur, the overall AE-related cost profile of ICI may compare favorably to that of conventional chemotherapy, reinforcing the value of ICIs as a treatment backbone while highlighting the need for targeted strategies to mitigate irAE-driven inpatient spending.

Our focus on a Medicare population is clinically and operationally relevant. Older adults account for the majority of patients with advanced NSCLC in routine practice yet remain underrepresented in ICI trials.^21,51,52 Higher baseline comorbidity and differences in toxicity recognition may influence both the likelihood and consequences of irAEs in this group; however, disentangling irAE-attributable costs from underlying comorbidity-related health care needs is inherently challenging in observational data. To mitigate this concern, our primary analysis used marginal structural models with inverse probability weights that accounted for both baseline and time-varying comorbidities, and a sensitivity analysis excluding patients with preexisting autoimmune disease yielded consistent results, suggesting that the observed cost differences are not primarily driven by baseline immune-mediated conditions. As ICI use expands, driven by their demonstrated survival benefits and central role in the treatment of advanced NSCLC,⁴ so does the potential for toxicity-related medical spending.³⁹ In this fee-for-service Medicare population, clinically impactful irAEs were associated with sustained increases in hospitalization and downstream medical costs, highlighting the importance of optimizing toxicity management to preserve the clinical benefits of ICIs while minimizing avoidable resource use. These findings provide real-world inputs for budget impact analyses, cost forecasting, and value assessments of ICIs in older adults. The disproportionate contribution of inpatient services to irAE-related costs warrants consideration alongside current management guidelines, which recommend outpatient management with corticosteroids and supportive care for most grade 1 to 2 irAEs, reserving hospitalization for grade 3 or higher events or clinical deterioration.⁹ This suggests opportunities to reduce irAE-driven spending through earlier recognition, structured outpatient monitoring, and timely initiation of immunosuppressive therapy, potentially averting hospitalizations for irAEs that could be managed in ambulatory settings. Managed care organizations may consider incorporating irAE risk into care coordination programs, targeted toxicity monitoring, early outpatient intervention pathways, and hospital readmission reduction initiatives to mitigate avoidable inpatient use while preserving access to effective cancer treatment, particularly in older adults who may be at higher risk for both irAE severity and hospitalization. Ongoing evaluation of use and cost trends will remain important as ICI use and toxicity management strategies continue to evolve.

LIMITATIONS

This study has several limitations that should be considered when interpreting the findings. First, claims-based identification of irAEs is inherently imperfect. Diagnosis codes do not directly encode important parameters such as clinical certainty, toxicity grade, laboratory values, or clinician attribution. Additionally, treatment decisions captured in claims may reflect clinician practice patterns; for example, in older populations, clinicians may be more likely to initiate immunosuppressive therapy or modify treatment earlier, which could influence the identification of clinically impactful irAEs. As a result, some events may be misclassified, and severity cannot be measured using Common Terminology Criteria for Adverse Events in claims. To improve clinical relevance, we used a claims-based proxy for clinically impactful irAEs, but this approach may preferentially capture more severe events and miss irAEs managed without systemic immunosuppression or measurable treatment gaps. Second, censoring from death and Medicare disenrollment are substantial in advanced NSCLC and may bias cumulative estimates downward at longer horizons. Although we addressed censoring analytically and conducted sensitivity analyses, informative censoring may persist if dropout is related to both irAE risk and downstream use or costs. Third, attribution of use and costs to irAEs is challenging in observational claims data because post-irAE care may reflect toxicity management, cancer progression, comorbidity decompensation, or treatment changes. Our causal framework strengthens attribution by aligning exposure and confounding over follow-up, but residual confounding remains possible, particularly from time-varying clinical factors not captured in SEER-Medicare (eg, performance status, smoking history, molecular characteristics, symptom burden). Fourth, exposure timing may be imprecise because onset was defined at the month level and relied on claims data, which may be affected by coding delays or bundled billing. Fifth, part D pharmacy capture may be incomplete over follow-up, and claims-based patient responsibility amounts do not reflect true out-of-pocket burden. Sixth, generalizability is limited to fee-for-service Medicare beneficiaries aged 66 to 85 years with continuous parts A/B/D coverage. Additionally, practice patterns and clinician experience with irAE recognition and management have evolved since the study period (2015-2019), the most recent year of available data, which may influence the generalizability of findings to contemporary care, although core management approaches and associated resource use have remained similar. Relatedly, consistent with final year of study data period, cost estimates reflect 2019 US dollars and may not fully capture changes in pricing or health care delivery in more recent years. Future studies using more recent data are needed to assess whether these changes have influenced use and costs. Finally, although stabilized weights and sensitivity analyses were used, model misspecification and unmeasured confounding cannot be fully excluded.

Conclusions

Among older adults with advanced NSCLC treated with ICIs, clinically impactful irAEs were common and were associated with sustained increases in health care use and costs. Inpatient services accounted for the largest share of excess spending and remained elevated over follow-up. These population-level findings position irAEs as a long-term management challenge in routine oncology practice and support care approaches focused on early toxicity recognition, coordinated outpatient management, and prevention of avoidable hospitalizations while maintaining the clinical benefits of ICIs in patients who are older. Managed care strategies that incorporate toxicity risk identification, enhanced outpatient monitoring, and case management support may help reduce avoidable inpatient use while preserving the clinical benefits of ICIs.

Disclosures

The authors declare no competing interests.