Introduction

The first chimeric antigen receptor T-cell immunotherapy (CAR-T) received FDA approval in August 2017 for the treatment of B-cell precursor acute lymphoblastic leukemia (ALL).¹ Since then, several additional CAR-T therapies have been approved for treatment of various hematologic cancers, and next-generation CAR-T therapies are advancing beyond hematology with studies in solid tumor cancers, autoimmune disorders, chronic viral infections, and heart failure.²

The current CAR‑T treatment patient journey involves a series of coordinated steps, beginning with assessment of patient eligibility through lab panels and clinical history, followed by collection of the patient’s white blood cells via leukapheresis and administration of the genetically modified cells, and concluding with monitoring and treatment for adverse events.^3,4 These steps and the corresponding billing codes for each procedure are outlined in manufacturer billing guides.

These therapies present unique challenges to health insurers today due to:

• Low disease incidence: The hematologic cancers approved for treatment with CAR-T therapies are low incidence, making the occurrence of new cases unpredictable, especially among smaller insured populations.
• Single administration treatments: Approved CAR-T therapies do not have a multi-dose or ongoing treatment option. It is more difficult to forecast utilization over time since there is not a continuing patient base.
• High costs: In a study of 2022–2025 commercial claims data, CAR-T therapies had an average wholesale acquisition cost (WAC) near $443,600, but an estimated mean episode cost of $809,800 per treatment episode; episode costs were higher than $1.1 million for 15% of cases.⁵

High costs paired with the rarity of the cancers can strain premiums and reserves, increase financial risk, and reduce the predictability needed for sustainable plan pricing. This report evaluates how CAR-T therapies and related treatment steps are documented in claims data, including:

• The extent to which expected procedures associated with CAR-T treatment episodes are explicitly documented in claims
• Differences in claims interpretation based on whether treatment costs can be attributed to a specific claim line item or only to a treatment date
• Cost variation of CAR-T therapies and other associated costs for those patients where CAR-T treatment costs could be attributed to a specific claim line

By assessing actual claims documentation patterns, we intend to provide valuable context to payers engaged in similar exercises.

Methods

Claims for services rendered from January 1, 2022, to April 30, 2025, were analyzed from Milliman’s Consolidated Health Cost Guidelines Source Database (CHSD), a longitudinal administrative claims database with nearly 67 million commercially insured members. Patients treated with CAR-T therapy were identified by the observation of a CAR-T-specific Healthcare Common Procedure Coding System (HCPCS) code (typically outpatient/professional claims) or International Classification of Diseases, 10th Revision, Procedure Coding System (ICD-10-PCS) procedure code (inpatient claims), based on a review of manufacturer billing code guidelines for each product included in this study (Appendix Figure 4).

Some patients had procedure codes associated with multiple treatments or multiple dates. The allocation of costs across claim lines for a particular claim ID also varied from claim to claim, reflecting varying reimbursement dynamics (e.g., fee-for-service, case rates). Each patient was allocated to a single CAR-T treatment and date by the following algorithm:

1. Attributable to a claim line item and date: Among all claim lines on the same date as the CAR-T-specific procedure code, the CAR-T therapy product was attributable to a specific line item if all of the following were true:
1. The most expensive line item associated with the claim ID had an allowed cost in excess of $100,000.
2. The most expensive line item associated with the claim ID was distinct.
3. The most expensive line item associated with the claim ID had a specific pharmacy-related revenue code.^*
4. The patient had no claim line reversals for that date.

   The date of that line item was labeled as the CAR-T treatment date.

2. Not attributable to a claim line item, but attributable to a date: If the patient was not assigned a date via step 1, then the date with the greatest total allowed charges involving a relevant CAR-T therapy HCPCS or ICD-10-PCS code was labeled as the treatment date, but the treatment was not attributed to a specific claim line item.

This algorithm enabled us to assign each patient to a single treatment date. It also discerns the level of detail available in claims, enabling consideration of cost dynamics by the level of detail available.

After treatment dates were assigned, we examined incurred costs leading up to the treatment date stratified by observation of specific key steps (harvest/leukapheresis, treatment preparation, and administration) explicitly documented in claims. These key steps were compiled from the billing guidelines,^4,6,7,8,9,10 along with the relevant revenue and HCPCS codes (see Appendix Figure 4).

In cases where a specific claim line was identified for the CAR-T therapy cost with reasonable certainty (method 1 above), we were able to consider cost variation of the drug itself compared with the variation in cost of other healthcare utilization. We measured cost variation with standard deviation, and by comparing the average cost overall to the average cost of the top decile of patients by overall healthcare spend in the study period under the medical benefit.

Results

Within the commercial claims data, we identified 1,370 patients treated with a CAR-T therapy. For these claims, we evaluated the presence of associated procedure codes, assessed allocation of treatment costs on claim lines, and estimated the cost variation for the product costs and other costs.

Presence of CAR-T-associated procedure codes from billing guidelines

We evaluated whether claims captured the expected clinical steps associated with preparation and administration of CART-T therapy (as defined using codes suggested in manufacturer billing guidelines) and found that claims for each step were not consistently observed for each patient. In the 60 days leading up to treatment, 51% of the patients identified had no explicit documentation (i.e., billing codes suggested by the manufacturer billing guides) of leukapheresis, 44% had no explicit documentation of preparation, and 54% had no explicit documentation of the administration. All patients in the study had an observation of a HCPCS and/or ICD-10-PCS code associated with a specific CAR-T therapy, suggesting they received the treatment, but the billing guidelines’ recommended codes for associated procedures were not used.

The standard deviation of allowed cost per patient in Figure 1 is substantially larger than the difference between the smallest and largest average allowed cost per patient across documentation patterns. This suggests that explicit documentation of prerequisite procedures is not a strong predictor of total treatment cost.

Identified treatment dates and claim lines

Overall, we observed that 23% of patients had a specific CAR-T therapy HCPCS code only, 68% had an ICD-10-PCS code only, and 9% had both a HCPCS and ICD-10-PCS code.

We evaluated whether CAR-T therapy costs could be attributed to a specific claim line or only to a treatment date, since this distinction affects how confidently product costs can be isolated from broader episode costs. In total, 62% of CAR-T patients’ treatments were attributed to a single claim line and date, with the remainder attributed to a treatment date but not to a single claim line (Figure 2). Where we were unable to assign a specific treatment claim line, coding patterns varied, including cases in which nearly all claim costs were allocated to a general revenue code not tied to CAR-T therapy administration or costs were distributed across multiple claim lines.

Cost variation

For the 845 patients where the CAR-T therapy administration was attributable to a claim line and date (Figure 2), we examined variation in allowed cost for that specific claim line across patients, and the variation in per patient allowed costs on the treatment date plus or minus 60 days excluding the cost of the CAR-T therapy product. Overall, the average cost of CAR-T therapy product was $586,032, and the average cost among the top decile of patients by total allowed incurred during the study period was $997,283 (a difference of $411,251). Other costs incurred 60 days before through 60 days after the treatment were $216,543 and $377,400 for the average and top decile patients, respectively (a difference of $160,857). Although ancillary costs varied among patients, the standard deviation in the cost of the CAR-T therapy product itself was much larger, suggesting that variability in the treatment episode is driven more by the product cost than other related and supporting service costs. Figure 3 presents the variation in costs of the CAR-T therapy product itself, stratified by specific products.

Discussion

This analysis assessed the costs and coding practices in claims data associated with patients treated with CAR-T therapies and provides insight into how clearly the steps associated with CAR-T procedures were captured in claims. This study identified a large amount of coding variability associated with these treatments, including costs and documentation of associated procedures. In particular:

1. Most patients did not have clear documentation for each expected step associated with CAR-T treatment. Despite the availability of specific codes for each treatment stage, such as those described in manufacturer billing guidelines, these codes were often not utilized in practice. Only 19% of CAR-T episodes had specific codes for all associated steps of cell harvesting, preparation, and administration.
2. More than one third of patients who had a CAR-T therapy procedure code did not have costs clearly associated with a single claim line. For the 62% of patients with CAR-T procedure costs clearly assigned to a single claim line item, treatment-date costs were generally consistent with expectations (e.g., near or above wholesale acquisition costs). For the remaining 38% of patients, claim costs were distributed across claim lines equally or in unexpected ways, making it difficult to parse the product costs for other related procedures associated with the claim.
3. Variation in CAR-T episode costs was driven more by the cost of the CAR-T therapy product itself, rather than costs for surrounding services. Services surrounding treatment added substantially to the episode (nearly $200,000 on average), but the CAR-T therapy product costs contributed the largest portion of the episode spend and had high standard deviations. Cost variations for the surrounding services and the CAR-T product can reflect differences in facility charges/mark-ups, negotiated reimbursement rates, and treatment complications (e.g., readmissions).

Frequent deviations from the billing guidelines and large variation in allowed costs attributed to CAR-T-specific procedure codes suggest that norms have yet to be established for how treatment episodes are coded and reimbursed. Inconsistent coding patterns, as presented in this study, suggest that billing practices for CAR-T therapies remain heterogeneous across treatment settings, limiting the reliability of analyses that rely on a narrow set of procedure codes or revenue codes. Reliable interpretation requires both product code identification and episode-level validation of surrounding costs, timing, and claim-line attribution. In short, nuanced analyses are necessary when assessing costs and utilization of CAR-T therapies using claims data.

Conclusion

Among claims experience for patients treated with CAR-T therapy, the variation in service documentation, claim line detail, and product reimbursement rates present meaningful challenges for commercial health insurers, complicating both forecasting and financial risk management. This study found that, although CAR-T therapy procedure codes were clearly identifiable for many patients, a substantial proportion had inconsistent coding for treatment steps, variation in how costs were allocated across claims, and wide variation in the cost of the CAR-T therapy product itself. As the CAR-T therapy landscape continues to evolve and mature, nuanced analytical approaches remain essential for reliably interpreting CAR-T therapies in claims data. Additional analyses may be needed to distinguish the effects of negotiated reimbursement arrangements, such as fee-for-service versus bundled billing practices on CAR-T product costs.

Limitations

This white paper presents an overview of healthcare costs and services for patients treated with CAR-T therapies and how aggregate patterns differ from expected if manufacturer billing guidelines are consistently followed. In performing this analysis, we relied on publicly available information and Milliman’s CHSD commercial claims data. We have not audited or verified this data and other information. If the underlying data or information is inaccurate, incomplete, or outdated, the results of this analysis may also be inaccurate, incomplete, or outdated.

These findings reflect commercially insured members in the dataset and may not generalize to Medicare, Medicaid, or other insured or uninsured populations. CAR-T therapy billing and reimbursement can differ substantially by payer, meaning results presented herein may vary substantially even for other commercially insured populations.

The logic used to assign a single treatment/date and identify a product claim line is reasonable but algorithmic. Some patients may have been misclassified or missed, especially when costs were bundled, distributed across multiple lines/dates, reversed, or billed under nonspecific revenue codes.

Although deviation from manufacturer billing guidelines was observed, this study does not intend to imply that the services associated with harvesting, preparation, and administration did not occur. Services that were not clearly observable or attributable to a specific procedure code may have been captured in the product code if bundled or on other general codes that were not included in this study.

Jessica Naber is a consulting actuary with Milliman. She is a member of the American Academy of Actuaries and meets the Qualification Standards of the American Academy of Actuaries to render the actuarial opinions contained herein. To the best of her knowledge and belief, this report is complete and accurate and has been prepared in accordance with generally recognized and accepted actuarial principles and practices.