TL;DR
This study reveals significant disparities in access to CAR-T therapy for multiple myeloma patients within a large academic health system, with treatment location and patient race being key factors. Notably, a substantial portion of eligible patients did not have documented discussions about CAR-T, suggesting systemic barriers beyond clinical appropriateness.
What Problem Does This Paper Address?
Multiple myeloma (MM) is a serious blood cancer, disproportionately affecting Black patients. While CAR-T therapy offers a promising new treatment avenue, its limited availability and specialized nature raise concerns about equitable access. This paper investigates whether factors like where a patient receives care, their demographic background, and disease characteristics influence their likelihood of receiving CAR-T therapy, aiming to identify potential barriers to this advanced treatment.
Methods
The researchers conducted a retrospective cohort study using electronic health records from the University of California Health Data Warehouse (UCHDW). They employed logistic regression to analyze associations between patient demographics, disease characteristics (like ISS stage and CRAB features), treatment location (UC-1, UC-2, UC-3), and CAR-T therapy receipt. A novel aspect of this study was the use of a zero-shot GPT-4 model to analyze clinical notes. This AI approach was used to identify patients who were eligible for CAR-T therapy but had no documented discussion about it, and to categorize the reasons for eligibility determinations.
Datasets
The primary dataset used was the University of California Health Data Warehouse (UCHDW), containing electronic health records for over nine million individuals. This dataset spanned from 2012 to February 2025 and included information on patient demographics, diagnoses, treatments, and clinical notes. The study population specifically comprised 12,360 adult patients diagnosed with MM and treated at a UC facility offering CAR-T administration between January 2021 and January 2025. For the GPT-4 analysis, a subset of 270 UCSF patients with available clinical notes was used.
Key Insights and Ideas
- Location Matters: CAR-T receipt varied significantly across different UC medical centers, suggesting that the type of facility (e.g., specialty referral vs. comprehensive care) plays a role in access.
- Race as a Barrier: Even after accounting for disease severity and insurance, Black patients were significantly less likely to receive CAR-T therapy compared to White patients. This points to systemic issues rather than biological differences.
- The "Undiscussed" Eligible Patient: A striking finding is that a notable percentage of patients deemed eligible for CAR-T therapy had no recorded discussion about it with their providers. This highlights potential gaps in communication, referral pathways, or patient engagement.
- Disease Severity is a Factor: Patients with more severe disease (indicated by a higher number of CRAB features) were more likely to receive CAR-T, suggesting that while access is an issue, the urgency of the disease can drive treatment decisions.
- AI for Deeper Understanding: The use of GPT-4 to analyze clinical notes provided a unique window into undocumented aspects of care, such as the absence of CAR-T discussions for eligible patients.
Key Results
- Out of 12,360 MM patients, only 320 (2.6%) received CAR-T therapy.
- Black or African American patients had significantly lower odds of receiving CAR-T therapy compared to White patients (OR = 0.33).
- Patients treated at UC-3 had lower odds of receiving CAR-T compared to UC-1 (OR = 0.42).
- Among 270 UCSF patients assessed for eligibility, those identifying as Other Pacific Islander had the highest proportion (50%) of eligibility without a documented CAR-T discussion, followed by Black or African American patients (4.2%) and Asian patients (3.2%).
- Patients with a higher number of CRAB features had increased odds of receiving CAR-T therapy (OR = 1.43).
Limitations and Future Work
The authors acknowledge several limitations:
- Temporal Bias: Changes in CAR-T indications and clinical practice between 2021 and 2025 could introduce temporal bias.
- Data Completeness: ISS staging was unavailable for some patients due to missing lab data, limiting the use of the revised ISS. CRAB features reflect clinical manifestations, not comprehensive biological risk.
- Socioeconomic Measurement: The Area Deprivation Index (ADI) is a neighborhood-level measure and may not fully capture individual socioeconomic barriers.
- Clinical Notes Analysis: The GPT-4 analysis was limited to UCSF clinical notes and could not account for undocumented decisions, patient preferences, or unmeasured barriers.
- Direct Access Measures: The study evaluated documented receipt, not direct measures of access.
Future work should incorporate individual-level social determinants of health, detailed referral patterns, and a broader clinical context to fully understand the drivers of CAR-T utilization and promote equitable delivery of advanced therapies. Further investigation into why Hispanic or Latino patients had higher odds of receiving CAR-T compared to White patients is also warranted, as this contrasts with some prior reports.
Summary updated