Expenditures in Young Adults with Hodgkin Lymphoma: NCI-Designated Comprehensive Cancer Centers versus Other Sites

This article is featured in Highlights of This Issue, p. 1

Hodgkin lymphoma (HL) patients diagnosed between age 15 and 39 years (adolescents and young adults: AYA) have not seen the same improvement in survival when compared with children (<15 years; refs. 1 and 2). Previously, our team revealed that among young adults (YA: 22–39 years) with HL, treatment at an NCI-designated Comprehensive Cancer Center (CCC) mitigates these outcome disparities among older AYAs (3). In the setting of findings such as these across cancer types, the NCI issued a mandate to address disparities in AYA outcomes (2). HL represents a quintessential AYA malignancy with peak incidence and inferior survival in young adulthood (4, 5). Although there has been general criticism of specialized cancer centers (such as CCC) for higher costs of delivering care (6), the expenditures for YA with HL at CCC-designated facilities as compared with other facilities are unknown. We aimed to compare expenditures in the first year following diagnosis at CCC versus non-CCC sites for YA with newly diagnosed HL. We hypothesized that expenditures would be higher at CCCs and aimed to examine drivers of differences in expenditures. In order to ensure that patients were treated by comparable oncology services (i.e., all adult services rather than possibly pediatric or adult for the 15- to 21-year-old patients, we used deidentified claims data from the OptumLabs Data Warehouse (OLDW) to examine this question among 22- to 39-year-olds with newly diagnosed HL.

Deidentified administrative claims data from OLDW include medical and pharmacy claims, laboratory results, and enrollment records for >88 million privately insured individuals during the study period. The database contains enrollees with a diverse mixture of ages and ethnicities across all 50 states in the United States (7). These data have been used previously to evaluate healthcare expenditures and utilization in oncology populations (8–11). This study was conducted in accordance with the US Common Rule, and approved by the Institutional Review Board of the University of Alabama at Birmingham.

We included patients diagnosed with HL when they were ages 22 to 39 years, between 2001 and 2016. The period of study was restricted to patients diagnosed through 2016 to allow for one-year follow-up from diagnosis (claims available through 2017). Patients were included if they were continuously enrolled during the six months prior to diagnosis and for ≥30 days post-diagnosis. Administrative billing codes (ICD-9: 201.4–201.9; ICD-10:C81.9x) were used to identify patients. We captured monthly data after HL diagnosis, retaining data only for periods during which enrollment exceeded 15 continuous days of the month. In order to avoid unintentional inclusion of off-therapy patients new to the payor, analyses were restricted to patients with at least 10 claims for chemotherapy with/without radiation [using Healthcare Common Procedure Coding System (HCPCS) codes and J-codes] during the year following HL diagnosis. Patients were not included if they had claims related to blood/marrow transplantation, due to differences in healthcare utilization associated with transplant as well as use of transplant in recurrent disease (vs. newly diagnosed). We excluded 18 patients who were outliers, with expenditures beyond the 99th percentile (Supplementary Fig. S1).

The primary outcomes included: (i) average monthly health-plan paid (HPP) expenditures and (ii) average monthly amount owed by the patient (out-of-pocket: OOP; including deductibles, copayments, and OOP pharmacy expenditures). Total expenditures in each category (HPP and OOP) were totaled separately, and included claims related to care provided in the inpatient and outpatient (outpatient hospital and office visits) setting, in the emergency department (ED), and other settings (long-term care, home health, and laboratory).

We evaluated the contribution of each of the following to HPP and OOP: (i) adjusted monthly rates of visits (inpatient, outpatient), (ii) length of inpatient stay (LOS), (iii) high-intensity care (HIC) after HL diagnosis (intubation, intensive care unit, mechanical ventilation, dialysis, and identified using ICD-9/10 codes), and (iv) treatment (chemotherapy and radiation) expenditures (HCPCS and J-code). Among older (Medicare-aged) cancer populations, expenditures differ between outpatient hospitals (hospital-affiliated outpatient centers) and office visits (private physician offices; refs. 12, 13); therefore, sub-analyses examined these separately.

If patients received any cancer-related care at a CCC, they were classified as CCC. Patients who did not receive any cancer-related care at a CCC were classified as non-CCC patients. Although the proportion of AYA with HL receiving >50% of care at a CCC (11%) or >25% of care at a CCC (15%) was similar to the authors' previous work (3), it did not allow for the fully adjusted models to run robustly; thus, unadjusted sensitivity analyses were performed by repeating the original analyses, considering CCC patients to be those who received >50% of their care at a CCC. In order to present the most robust findings possible, the adjusted models are presented here.

Demographic characteristics were derived from enrollment (sex), household-level consumer data (race/ethnicity, annual household income), and census block group (education). Clinical characteristics included age at HL diagnosis. Year of diagnosis was included (2001–2005, 2006–2010, and 2010–2016). For comorbidities diagnosed prior to HL, we modified the Charlson comorbidity index (14) by excluding malignancies, and adding asthma and obesity to enhance its relevance to YAs. We examined characteristics of patients and admissions requiring HIC (ICU, mechanical ventilation, and/or dialysis) in the year following HL diagnosis, primarily because of the potential for downstream spending generated by these events (15, 16).

In order to account for advanced disease, we used the administrative billing code for HL in multiple sites (C81.9) to derive a variable representing advanced disease stage. Because stage II patients may be treated as advanced stage, administrative billing codes were used to differentiate stage I from stages II–IV HL based on the Lugano staging schema (17). ICD10 codes identified involved nodal regions including single sites that do not confer higher risk (C81.91–96), single sites that do confer excess risk (C81.97 and C81.99) and multiple sites (C81.98), indicating a patients is stage II or higher.

HPP and OOP expenditures were evaluated as estimated average monthly expenditures per patient from diagnosis until death, disenrollment from the health plan, one year after diagnosis, or end of study (March 31, 2017), whichever occurred first. To account for inflation, we adjusted all expenditures to 2016 consumer price index at the claim-line level. To create the predicted expenditures, we adjusted for age at diagnosis, year of diagnosis, pre-HL comorbidities, advanced disease, race/ethnicity, income, education, occurrence of post-HL HIC, and then divided the total expenditures by the patient-months to create monthly predicted averages. We used bootstrapping to calculate 95% confidence intervals (95% CI) for average monthly expenditures. In order to make statistical comparisons between groups, we estimated the generalized linear model (GLM) with a log-link and gamma distribution per-person month treating costs as continuous variables (18). Cost ratios were calculated to help interpret the magnitude of difference in the raw coefficients; to do so, we divided the expenditures incurred by CCC patients by those incurred by non-CCC patients. We calculated incidence rate ratios (IRR) to compare the monthly visit rates (inpatient, outpatient, and ED) of CCC and non-CCC patients, using negative binomial regression and adjusting for the variables included in GLM. We focused on findings with P < 0.01 to highlight meaningful differences.

Overall, 1,797 patients met the inclusion criteria (Supplementary Fig. S1); 36% received care at a CCC (n = 643). The mean age at diagnosis was 31 years (SD, 5.1; Table 1). Patients were continuously enrolled for a median of 12 months and mean of 10.7 months (SD, 2.7) after HL diagnosis. There was no difference between CCC and non-CCC patients with respect to age at HL diagnosis, sex, race/ethnicity, months enrolled, preexisting comorbidities, receipt of radiotherapy, or proportion of patients requiring HIC. A larger proportion of CCC patients had advanced-stage disease (21% vs. 14%, P = 0.04). A larger proportion of non-CCC patients had a high school diploma or less (21% vs. 19%, P < 0.001) and earned less than $50,000/year (12% vs. 9%, P = 0.04).

In the year following HL diagnosis, adjusted monthly HPP expenditures per person were higher among CCC patients when compared with non-CCC patients ($12,869 vs. $10,688, P = 0.001). The HPP cost ratio of 1.2 reflects that the adjusted excess HPP monthly expenditures were 20% higher at CCCs when compared with non-CCCs (Fig. 1; Table 2). The OOP expenditures were comparable ($444 vs. $464, P = 0.3; Supplementary Table S1).

Across sites, the largest proportion of HPP expenditures was accounted for by outpatient expenditures (CCC: 82%, non-CCC: 85%; Fig. 1; Table 2). Outpatient HPP expenditures were higher in CCC than in non-CCC patients ($10,496 vs. $9,102; P = 0.001). The cost ratio reflected that the adjusted excess HPP monthly outpatient expenditures were 20% higher at CCCs when compared with non-CCCs. There was no difference between OOP outpatient expenditures (CCC: $361 vs. non-CCC: $383, P = 0.9; Fig. 1; Supplementary Table S1).

We examined outpatient expenditures by location (outpatient hospital, office visit; Table 3). All outpatient hospital expenditures were higher among CCC patients; the cost ratios reflected higher expenditures both related to treatment (chemotherapy: 2.3 times higher; radiation: 1.3 times higher) and not related to treatment (1.5 times higher). On the other hand, office visit expenditures were lower for CCC patients both when related to and not related to chemotherapy (chemotherapy: cost ratio = 0.7, non-chemotherapy: cost ratio = 0.7).

Expenditures related to inpatient admissions accounted for the second largest proportion of HPP (CCC: 10%, non-CCC: 11%) and OOP (CCC: 14%, non-CCC: 10%) expenditures (Table 2; Supplementary Table S1). Inpatient HPP expenditures were higher at CCCs ($1,823 vs. $1,091, P = 0.001) with a cost ratio reflecting 1.7-times higher expenditures at CCCs (Fig. 1; Table 2). There were no differences in inpatient OOP expenditures between sites (CCC: $54 vs. non-CCC: $50, P = 0.1; Supplementary Table S1).

Outpatient hospital expenditures related to radiation were higher at CCCs ($936 vs. $733, P = 0.001) with a cost ratio reflecting 30% higher expenditures at CCCs (Fig. 1; Table 3). On the other hand, radiation-related office visit expenditures were statistically comparable between sites ($218 vs. $306, P = 0.2). There were no significant differences between CCC and non-CCC patients with respect to monthly HPP or OOP expenditures for ED or other expenditures (Fig. 1; Table 2; Supplementary Table S1).

When compared with non-CCC patients, CCC patients had 1.6 times higher predicted average monthly rates of non-chemotherapy outpatient hospital visits (P = 0.001; Table 4). CCC patients also had a higher rate of non-chemotherapy office visits, but with a smaller magnitude of effect (IRR = 1.1, P = 0.01). For non-chemotherapy visits, CCC patients had lower estimated per-visit expenditures for outpatient hospital (CCC: $2,399, non-CCC: $2,511) and office (CCC: $713, non-CCC: $1,077) visits. When compared with non-CCC patients, CCC patients had lower predicted average monthly rates of chemotherapy-related office visits (IRR = 0.7, P = 0.001) and relatively comparable monthly rates of outpatient hospital visits (IRR = 0.9, P = 0.03). There were no differences in visit rates related to radiation (P = 0.07); a comparable proportion of patients across CCC and non-CCC sites received radiation (P = 0.4; Table 1). Estimated per-visit expenditures for outpatient chemotherapy-related visits were higher among CCC patients for outpatient hospital (CCC: $633/visit, non-CCC: $259/visit) and office (CCC: $1,104/visit, non-CCC: $998/visit) visits.

The predicted average monthly rate of chemotherapy-related admissions was 3.1 times higher for CCC patients (P = 0.001; Table 4). CCC patients had a relatively comparable rate of non-chemotherapy admissions (IRR = 1.2, P = 0.02) based on our significance parameters.

A comparable proportion of CCC and non-CCC patients had claims related to HIC (CCC: 10.3%, non-CCC: 10.3%; P = 0.1; Table 1). Among CCC patients with at least one inpatient admission (n = 316), median LOS was 4.0 days (range, 1–26). Among non-CCC patients with at least one admission (n = 421), median LOS was 4.0 days (range, 1–38). Across sites, patients had longer LOS if they also received HIC (IRR = 1.7, 95% CI, 1.5–2.0, P < 0.001) or had advanced disease (IRR = 1.5, 95% CI, 1.3–1.7, P < 0.001). Patients at CCCs with HIC had slightly shorter LOS than patients with HIC at non-CCCs, as demonstrated a modestly significant (P = 0.03) interaction between CCC and HIC.

Sensitivity analyses (Supplementary Tables S2–S4) support the findings as presented in the manuscript. Both the CRs comparing CCC with non-CCC predicted monthly expenditures (inclusive of detailed analyses of outpatient expenditures) and the IRRs reflecting average monthly rates of visits remain in the same direction as the adjusted analyses presented.

Using a commercial claims database, we found estimated monthly HPP expenditures per person to be higher for YAs with newly diagnosed HL receiving any of their care at a CCC, than among those not receiving any care at a CCC. Although outpatient HPP expenditures accounted for the largest proportion of expenditures across sites, this was 20% higher for CCC patients, driven largely by outpatient hospital expenditures. We also found higher inpatient HPP expenditures for CCC patients, accounted for by higher chemotherapy-related admission rates. OOP expenditures did not differ between CCC and non-CCC.

We found higher estimates of monthly per-patient outpatient hospital expenditures for CCC patients, with cost ratios reflecting expenditures that were 1.3 to 2.3-times higher. However, this did not hold true for office visits, which is likely characteristic of the structural differences between facility types. This phenomenon of facility-level differences driving healthcare expenditures is further reflected in the higher chemotherapy-related outpatient hospital expenditures among CCC patients, which are supported by higher per-visit outpatient hospital expenditures for CCC patients. These findings among the CCCs highlight the well-documented shift from office-based practices to hospital-affiliated outpatient centers among US cancer patients; this shift has been associated with higher expenditures, especially among commercial insurers (12, 13). Despite this phenomenon having been explored among Medicare patients (13), it has not been addressed in YA to our knowledge (19). Centers for Medicare and Medicaid Services proposed a rule (2020) aiming to make reimbursement more comparable across outpatient settings, including the outpatient hospital scenario underscored here.

CCC patients experience 50% higher outpatient hospital HPP expenditures unrelated to cancer therapy; a 60% higher monthly rate of outpatient hospital visits unrelated to cancer therapy could have driven (in part) these excess expenditures. Although monthly rates of the non-therapy office visits were higher among CCC patients, these were associated with lower expenditures at CCCs. The NCI designation as a CCC signifies that these sites approach care comprehensively, including supportive care, multidisciplinary decision-making, approaches to therapy delivery, and access to clinical trials along with individuals involved in research (20). Thus, it is reasonable to consider that the higher rates of non-therapy outpatient visits could reflect the provision of comprehensive care by the CCCs, which would be a sign of their NCI-designated status.

The 70% higher inpatient HPP expenditures at CCCs were driven by a higher rate of inpatient admissions, in the setting of comparable LOS and post-diagnosis HIC. Our data suggest that CCC patients received inpatient chemotherapy more often than non-CCC patients. Although the CCC patients also had a higher rate of non-chemotherapy admissions (IRR = 1.2), the magnitude of effect was larger for the rate of chemotherapy-related admissions (IRR = 3.1); this may speak to differences in treatment regimen (for example, pediatric-style regimens are administered inpatient more than are adult-style regimens). Unfortunately, it is difficult to identify expenditures associated with chemotherapy administered inpatient due to the nature of claims data.

Alternative hypotheses are important to consider, specifically other potential differences between CCC and non-CCCs that may account for differences in expenditures. One may consider patient-level differences between CCC and non-CCC patients such as comorbidities that require more frequent or more intense care, or higher stage disease requiring more intense therapy and/or with a higher likelihood of requiring treatment for persistent disease; both are adjusted for in the multivariable cost modeling. Additionally, the way in which patients were identified would exclude patients who had an ICD-10 code for cancer with no immediate treatment code, thus excluding likely recurrent or progressive disease. It is possible that clinical teams at CCCs are more likely to follow formal guidelines, which may require additional surveillance; conceivably, this may be consistent with our findings that CCC patients had a higher rate of visits not related to treatment. Unfortunately, insufficient data were available regarding clinical trial enrollment for patients; patients enrolled on clinical trials may have more follow-up visits. Finally, system-level differences such as being structured as an outpatient hospital versus physician office will impact billing practices; this is the most likely hypothesis and is supported by our findings as outlined above. Claims-level data afford investigators a window into real-world healthcare delivery at the patient level across a breadth of sites; however, such data are limited by the administrative nature of data capture. We could not examine patient-level clinical details; we derived the staging variable from billing codes. Classification of patients as CCC/non-CCC using administrative data may be limiting; although patients were considered CCC if they received any care at a CCC, sensitivity analyses in which CCC patients were considered those receiving >50% of care at a CCC revealed similar findings (Supplementary Tables S2–S4).

Although administrative data provide large cohorts, they face limitations in granularity. OLDW data did not capture managed care plans, which may reflect a portion of patients with HL. For patients treated at multiple sites, we could not identify the type of site (CCC vs. non-CCC) for each visit; thus, there may have been differences between CCC and non-CCC patients who we were unable to capture. We examined expenditures in the one year following diagnosis, yet included patients who were continuously enrolled six months prior to diagnosis and for >30 days after diagnosis. Both CCC and non-CCC patients were enrolled for a median of 12 months (CCC: 12 months [IQR = 12–12]; non-CCC: 12 mos [IQR = 11.5–12], P = 0.01); 75% of all patients were enrolled for at least one year (CCC: 79%, non-CCC: 74%). Insurance churning (transitioning between insurance policies and coverage) is more common among YAs than other age groups (21, 22). Thus, it was more important to be inclusive in order to capture the breadth of patients; hence the decision to include patients who were enrolled for >30 days after diagnosis. In order to account for the variable length of time that patients were in the cohort, we divided expenditures by patient-months. The data include commercially insured YAs, limiting the generalizability of these findings outside of that population; thus, these questions should be asked in data that include publicly insured patients. Nevertheless, over 70% of 22 to 39 year-old HL patients in our previous population-level study were privately insured (3); thus, these findings can be considered relevant for the majority of the YA HL population. We adjusted for available and measurable confounders, which were previously found to be associated with survival (income, education, advanced disease, and comorbidities) or with treatment at a CCC versus non-CCC (race/ethnicity) (3). The number of patients missing income, education, and race/ethnicity in the current data limits the conclusions that can be drawn from these variables; in light of measures to ensure the data are statistically certified as deidentified, partners do not have access to both individual-level variables and geographic identifiers that would enable use of area-level measures as well. In light of inpatient billing practices, one cannot separate chemotherapy expenditures from total inpatient expenditures; instead, our evaluation of chemotherapy expenditures focused on rates of chemotherapy-related admissions. Due to lack of cancer stage in OLDW, the approach to identifying advanced disease was derived based on clinically accepted staging systems and billing practices; to our knowledge, there is no validated approach to identifying advanced disease using billing codes, however it would have been problematic to not adjust for the extent of disease.

In summary, care of YA with newly diagnosed HL at a CCC is associated with higher HPP expenditures when compared with care at a non-CCC. HPP expenditures were significantly higher at CCCs than non-CCCs for both outpatient and inpatient care. Excess outpatient expenditures at CCCs were driven by higher rates of outpatient hospital visits not related to delivering cancer therapy, which may plausibly represent provision of comprehensive care. These two outpatient findings are likely reflective of the impact of facility structure and billing models (outpatient hospital vs. office visit, etc.), which is supported by higher per-visit expenditures for chemotherapy-related outpatient hospital visits. The higher inpatient HPP expenditures at CCCs were driven by the higher rates of inpatient admissions for chemotherapy. Differences in expenditures did not meet statistical significance either among ED and other HPP expenditures or among any OOP expenditures. Previously, we examined 1,094 patients 1 to 39 years of age (88% of whom were AYAs), using Los Angeles County cancer registry data and revealed superior survival among YAs with HL at CCCs; in specific, older YAs at non-CCC sites had over twice the risk of mortality as compared with children (HR = 2.2; 95% CI, 1.4–3.4, P < 0.001) whereas comparable patients at CCCs did not (HR = 0.7; 95% CI, 0.1–7.2, P = 0.8; ref. 3). Similarly, adolescents 15–19 years with HL in Georgia had superior survival if treated at Children's Oncology Group sites compared with other sites (HR = 0.14; 95% CI, 0.03–0.77, P < 0.05; ref. 23). In the setting of these findings that show superior survival at CCC sites, the results presented in this manuscript may indicate that the cost differences at CCCs are worth the benefits achieved. Although quantifying added value of care at a CCC for YAs with HL is challenging, the association between higher expenditures at CCCs and increased rates of non-therapy outpatient visits may indicate supportive, comprehensive care. Thus, it is plausible that promoting the utilization of CCCs may be one approach to improving outcomes while further work is needed to achieve this in the setting of decreased spending. It is conceivable that incentivizing institutions to modify facility-level billing practices may be one method to consider. Significant work remains to move toward the myth of systemic change.

D. Bernstein reports being a full-time employee of Stand Up To Cancer, a nonprofit organization that helped support the work through a competitive, peer-reviewed grant. The grant was made to a larger, multi-institutional team for a multi-aim project. The work described in this paper represents a sub-aim of the larger grant. Neither SU2C nor Bernstein received funds to support Bernstein's effort on this project. G.H. Lyman reports research funding from Amgen (Inst); speaking or advisory role in G1 Therapeutics, Partners Healthcare, BeyondSpring, Sandoz, ER Squibb (Inst), Merck, Jazz Pharm, Kallyope, TEVA, Seattle Genetics, and Samsung. L.E. Winestone reports grants from Stand Up To Cancer during the conduct of the study. No disclosures were reported by the other authors.

J.A. Wolfson: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S. Bhatia: Conceptualization, supervision, methodology, writing–review and editing. J.P. Ginsberg: Funding acquisition, writing–review and editing. L. Becker: Data curation, writing–review and editing. D. Bernstein: Funding acquisition, writing–review and editing. H.J. Henk: Resources, data curation, supervision, validation, methodology, project administration, writing–review and editing. G.H. Lyman: Writing–review and editing. P.C. Nathan: Writing–review and editing. D. Puccetti: Writing–review and editing. J.J. Wilkes: Writing–review and editing. L.E. Winestone: Writing–review and editing. K.M. Kenzik: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing.

J.A. Wolfson, S. Bhatia, and K.M. Kenzik were supported by a Stand Up To Cancer Award, Grant Number SU2C-AACR-SUPCHOP01. Stand Up To Cancer is a division of the Entertainment Industry Foundation. Research Grants are administered by the American Association for Cancer Research, the scientific partner of SU2C.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.