Cost drivers for breast, lung, and colorectal cancer care in a commercially insured population over a 6-month episode: an economic analysis from a health plan perspective

Abstract

Aims: In the absence of clinical data, accurate identification of cost drivers is needed for economic comparison in an alternate payment model. From a health plan perspective using claims data in a commercial population, the objective was to identify and quantify the effects of cost drivers in economic models of breast, lung, and colorectal cancer costs over a 6-month episode following initial chemotherapy.

Research design and methods: This study analyzed claims data from 9,748 Cigna beneficiaries with diagnosis of breast, lung, and colorectal cancer following initial chemotherapy from January 1, 2014 to December 31, 2015. We used multivariable regression models to quantify the impact of key factors on cost during the initial 6-month cancer care episode.

Results: Metastasis, facility provider affiliation, episode risk group (ERG) risk score, and radiation were cost drivers for all three types of cancer (breast, lung, and colorectal). In addition, younger age (p < .0001) and human epidermal growth factor receptor-2 oncogene overexpression (HER2+)-directed therapy (p < .0001) were associated with higher costs in breast cancer. Younger age (p < .0001) and female gender (p < .0001) were also associated with higher costs in colorectal cancer. Metastasis was also associated with 50% more hospital admissions and increased hospital length of stay (p < .001) in all three cancers over the 6-month episode duration. Chemotherapy and supportive drug therapies accounted for the highest proportion (48%) of total medical costs among beneficiaries observed.

Conclusions: Value-based reimbursement models in oncology should appropriately account for key cost drivers. Although claims-based methodologies may be further augmented with clinical data, this study recommends adjusting for the factors identified in these models to predict costs in breast, lung, and colorectal cancers.

Keywords:

• Breast neoplasms
• lung neoplasms
• colorectal neoplasms
• economics
• risk adjustment

Introduction

Significant reductions in morbidity and mortality have arisen from early detection and advances in cancer treatments, but these advances have led to rising costs which are unsustainable^1,2. Cancer care costs in the US have risen from $72 billion in 2004 to $125 billion in 2010, and are expected to reach $173 billion by 2020¹. Despite the increasing cost of cancer care, an Institute of Medicine report from 2013 concluded that Page 2 of IOM report “cancer care is often not as patient-centered, accessible, coordinated, or evidence-based as it could be”¹, and that the various secular and economic trends presented a looming crisis. Preventive measures, early detection³, coordinated, patient-centered, evidence-based care⁴, and alternate payment models (APMs)^5–8 have been proposed as ways to stem this economic crisis.

Financial incentives that reward providers for improved quality with appropriate utilization have been shown to reduce the total cost of care in oncology, without compromising quality³. Cancer care can have multiple cost drivers that depend on different cancer sub-types, as well as factors such as stage, treatment options, goals of therapy, and patient preferences. The costs of novel cancer drugs continue to rise; as a health plan we conducted a focused analysis on patients who receive chemotherapy, due to its impact on cancer care costs³. Payers and providers face several challenges in measuring performance, defining value, measuring quality, and changing reimbursement to support effective cancer care. One such challenge for health plans is lack of detailed clinical data suitably robust across categories for risk stratification in order to accurately compare cancer care costs. Our analytic objective was to use multivariable regression models to identify and quantify the effects of cost drivers in breast, lung, and colorectal cancer costs over a 6-month episode following initial chemotherapy.

Patients and methods

We conducted a retrospective cohort study utilizing claims data from Cigna beneficiaries. Cigna is a large commercial health services company and insurance provider in the US. Beneficiaries aged 18–65 years with breast, lung, or colon cancer who began new chemotherapy from January 1, 2014 to December 31, 2015 were included. In order to classify new cases for inclusion, we included beneficiaries with continuous coverage and at least one chemotherapy claim (including cytotoxic, targeted, and immunotherapies, but excluding all oral drugs) paid from their medical benefit plan within the stated timeframe. For each beneficiary, we tracked back 6 months prior to verify that there were no previous chemotherapy claims; this was done to capture costs over a 6-month episode of care following new chemotherapy⁵. For each episode, the trigger date was identified as the first chemotherapy date, and the episode window was defined as the time from the trigger date to 6 months afterward. This time period was chosen because our analysis showed that the majority of spend occurred in this timeframe following initial chemotherapy; this timeframe for a care episode can be seen also in the Centers for Medicare and Medicaid Services Oncology Care Model⁶.

We included the following factors in the analysis: age, human epidermal growth factor receptor 2 oncogene overexpression (HER2+)-directed therapy (for breast cancer), radiation as determined by current procedural terminology (CPT) codes (see Appendix), provider affiliation based on tax ID and fee schedule (classified dichotomously as [a] community based, or independent providers, vs [b] facility based, or providers affiliated with a hospital), and metastasis.

For all cancers, the presence of metastasis was determined by international classification of diseases (ICD)-9 codes beginning with 197, or equal to 198, 198.0, 198.1, 198.2, 198.3, 198.4, 198.5, 198.6, 198.7, 198.81, 198.82, or 198.89. We used chemotherapy claims and episode risk group (ERG) scores⁷ to further stratify the population. We then used claims with Healthcare Common Procedure Coding System (HCPCS) codes for trastuzumab: J9355, C9131, or J9354 (ado-trastuzumab and pertuzumab) as proxies for HER2+ breast cancers. Claims for VEGF (vascular endothelial growth factor)-directed therapies bevacizumab (J9035) and EGFR (epidermal growth factor receptor)-directed therapies (J9055, J9303) including cetuximab (J9055) and panitumumab (J9303) were used as proxies for metastatic disease in colon cancer. Likewise VEGF-directed therapy bevacizumab (J9035) was used as a proxy for metastatic disease in lung cancer; this method is accepted for classifying metastasis in claims data^9,10. In order to get complete data for the defined episode of care following new infusion chemotherapy, we excluded beneficiaries who had incomplete claims, were not fully insured, did not have 6 months of enrollment, or were not aged 18–65 years (see Figure 1).

Figure 1. Flow diagram of study inclusion and exclusion criteria.

These cancer-related factors were used in addition to Episode Risk Group (ERG) risk scores in order to examine the additional impact on cost of specific cancer-related characteristics. Patient-level risk scores used for adjustment were calculated using Optum Insight’s (formerly Ingenix) methodology, which relies on underlying medical conditions for the general population. ERG risk score is an estimate of expected cost burden per patient based on both general diagnostic information and retrospective review of costs over the prior 12 months. ERG risk score is an established method to control for variations due to patient comorbidities, ignoring impacts of unit cost, steerage opportunities, and/or utilization¹⁰.

During the episode, we also included costs that were not cancer-related if at least $500 were incurred for a given comorbidity, including: (a) hypertension, (b) IHD, (c) chronic obstructive pulmonary disease, (d) congestive heart failure, and (e) cardiomyopathy.

Winsorization is a two-sided truncation adjustment that limits the impact of outliers on the average expenditures. We used this method to truncate episode expenditure estimates at the 5th and 95th percentiles of per-episode expenditures by cancer type. Specifically, episode expenditures below the 5th percentile by cancer type were set to the 5th percentile, and episode expenditures above the 95th percentile were set to the 95th percentile within cancer type. This approach aligns with the methodology that the Centers for Medicare and Medicaid Services (CMS) has used for the oncology care model⁶.

We calculated episode costs as total medical (cancer and non-cancer) costs per member per episode for 6 months, from the time of the first chemotherapy treatment. We excluded pharmacy costs because approximately half the study population did not have pharmacy benefits through Cigna. However, costs for treatments such as injectable chemotherapy or supportive drugs incurred on the medical benefit were included.

To understand the impact of factors on the 6-month episode cost, we used generalized linear regression models with a log link and gamma distribution, including as predictors patient demographic factors, underlying ERG risk score, stage, treatments, and facility type. The data analysis for this paper was generated using SAS software, Version 9 of the SAS System (© 2013 SAS Institute Inc.). Likelihood ratio statistics were computed between successive models. In addition, we performed cross-validation by temporarily withholding 30% of the sample as the validation dataset; the remaining 70% was the development dataset, using stratified randomization in order to maintain sample characteristics. The datasets were recombined for the final models. Model fit was assessed using Scaled Pearson Chi-Square.

Results

Sample demographic and clinical characteristics are shown in Table 1. Mean age at time of chemotherapy was 50.5 years (SD = 8.7) in breast cancer as compared to lung (Mean = 57.0 years, SD = 6.3) and colorectal (Mean = 52.7 years, SD = 8.1) cancers. Lung cancer patients had highest ERG risk score (Mean = 19.81, SD = 9.3) compared with the other two cancer types. Diabetes was the most common comorbidity across cancer types (see Table 1). As shown in Table 2, metastatic disease and facility-based provider affiliation were associated in a univariate fashion with significantly higher costs in all three cancer types (breast, lung, and colorectal). Results of the multivariable regression model in breast cancer (n = 5,692) indicated that metastatic disease, higher ERG risk score, radiation, younger age, HER2+-directed therapy, and facility provider type contributed significantly to higher cost (see Table 3). As shown in Table 4, in lung cancer (n = 1,618), significant cost drivers identified were metastatic disease, higher ERG risk score, radiation therapy, and facility provider type. As shown in Table 5, in colorectal cancer (n = 2,423), significant cost drivers identified were metastatic disease, ERG risk score, radiation therapy, younger age, female gender, and facility provider type. Model fit, as assessed by Scaled Pearson χ², was significant at α = 0.001 for the lung and colorectal multivariable regression models, and p = 0.21 for breast (test statistics shown in table notes).

Table 1. Sample demographic and clinical characteristics.

	Breast	Lung	Colon
	(n = 5,700)	(n = 1,621)	(n = 2,427)
Female, n (%)	5,700 (100%)	795 (49%)	1,094 (45%)
Age at time of chemotherapy start in years, mean (±SD)	50.5 (±8.7)	57.0 (±6.3)	52.7 (±8.1)
Median (min, max)	51.0 (21,65)	58.0 (20,65)	54.0 (20,65)
ERG risk score, mean (±SD)	14.47 (±5.9)*	19.81 (±9.3)**	18.04 (±7.0)***
Comorbidity, n (%)
Diabetes	461 (8%)	156 (10%)	283 (17%)
Hypertension	1,208 (21%)	440 (27%)	636 (39%)
Chronic obstructive pulmonary disease	60 (1%)	393 (9%)	49 (3%)
Ischemic heart disease	117 (2%)	153 (9%)	125 (8%)
Congestive heart failure	55 (1%)	35 (2%)	25 (2%)
Cardiomyopathy	55 (1%)	16 (1%)	22 (1%)

* n = 5,692 non-missing;

** n = 1,618 non-missing;

*** n = 2,423 non-missing.

Table 2. Costs by cancer type, facility type, year, and setting.

Item	n	Community cost			n	Facility cost
		Median	Mean	SD		Median	Mean	SD
Breast cancer	4,059	$71,617	$85,691	$59,390	1,641	$91,463	$116,965	$87,332
Metastasis, HER2+	376	$135,567	$152,790	$77,144	181	$178,403	$192,895	$104,633
No metastasis, HER2+	838	$113,293	$122,221	$67,510	278	$147,122	$178,271	$118,646
Metastasis, HER2–	944	$65,264	$73,901	$43,029	520	$87,772	$102,810	$61,959
No metastasis, HER2–	1,901	$56,346	$62,172	$36,470	662	$71,105	$81,580	$49,572
Lung cancer	1,187	$80,849	$97,632	$66,471	434	$120,397	$145,360	$101,859
Metastasis	684	$95,372	$111,396	$66,444	302	$120,481	$165,672	$77,848
Non-metastasis	503	$58,166	$71,227	$46,518	132	$88,994	$98,688	$62,650
Colorectal cancer	1,783	$69,991	$83,824	$59,743	644	$95,360	$119,903	$89,161
Metastasis	754	$88,739	$111,188	$63,819	384	$117,744	$146,713	$72,542
Non-metastasis	1,029	$53,454	$63,773	$44,343	260	$65,682	$80,306	$54,195

Costs reported in 2015 US dollars. All comparisons between provider type were significant via two-groups t-test at p < .001.

HER2+, human epidermal growth factor receptor positive status.

Community total n = 7,029. Facility total n = 2,719.

Table 3. Multivariable regression model maximum likelihood parameter estimates, cost drivers in breast cancer (n = 5,692).

Parameter	Estimate	Standard error	Wald chi-square	p-value
Intercept	11.004	0.051	46,071.8	<.0001*
Metastasis	0.140	0.017	71.8	<.0001*
ERG risk score	0.017	0.001	151.85	<.0001*
Radiation	0.217	0.016	186.11	<.0001*
Age	−0.006	0.001	45.67	<.0001*
HER2+	0.646	0.017	1,423.01	<.0001*
Facility provider	0.283	0.017	273.25	<.0001*

Scaled Pearson Chi-Square = 5,771.29, p = .21.

Eight members had missing Episode Risk Group (ERG) risk score.

* p < .01.

HER2+, human epidermal growth factor receptor positive status.

Table 4. Multivariable regression model maximum likelihood parameter estimates, cost drivers in lung cancer (n = 1,618).

Parameter	Estimate	Standard error	Wald Chi-square	p-value
Intercept	11.083	0.146	5,816	<.0001*
Metastasis	0.389	0.031	157.85	<.0001*
ERG risk score	0.012	0.002	57.54	<.0001*
Radiation	0.148	0.030	24.97	<.0001*
Age	−0.003	0.002	1.93	.164
Female	0.015	0.030	0.26	.607
Facility provider	0.206	0.033	37.55	<.0001*

Scaled Pearson Chi-Square = 1,813.09, p < .05.

Three members had missing Episode Risk Group (ERG) risk score.

* p < .01.

Table 5. Multivariable regression model maximum likelihood parameter estimates, cost drivers in colorectal cancer (n = 2,423).

Parameter	Estimate	Standard error	Wald Chi-square	p-value
Intercept	10.901	0.093	13,825	<.0001*
Metastasis	0.497	0.027	341.77	<.0001*
ERG risk score	0.020	0.002	105.79	<.0001*
Radiation	0.261	0.032	67.22	<.001*
Age	−0.006	0.002	15.8	<.0001*
Female	0.010	0.026	14.67	<.001*
Facility provider	0.215	0.029	53.42	<.0001*

Scaled Pearson Chi-Square = 2,620.61, p < .05.

Four members had missing Episode Risk Group (ERG) risk score.

* p < .01.

Metastasis was also associated with greater hospital admissions (mean for no metastasis in breast = 0.23 [±0.55 SD] admissions per member compared with mean for metastasis = 0.35 [±0.69 SD], p < 0.001; p < 0.001 for lung and colon), and hospital length of stay (p < 0.001) in all three cancers over the 6-month episode duration. Chemotherapy and supportive drug therapies accounted for the highest proportion (48%) of total medical costs across all cancer types among beneficiaries observed (Table 6).

Table 6. Distribution of medical costs.

	Breast	Lung	Colorectal	All
Inpatient	6%	13%	15%	10%
Chemotherapy/supportive drugs	53%	40%	42%	48%
Radiation	10%	16%	8%	11%
Imaging^a	2%	5%	3%	3%
Outpatient surgery	7%	2%	4%	5%
Other^b	22%	24%	29%	24%

^a Imaging includes computerized tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET).

^b Other includes emergency room, lab, diagnostic tests, pathology, office visits, and other ancillary costs.

Discussion

Our objective was to evaluate our method of using claims data from a commercially insured population to quantify factors that drive higher costs between baseline and performance cohorts, over the course of a 6-month episode of care among breast, lung, and colon cancer patients. Our findings should be used to improve the ability of health plans to quantify cost drivers in oncology with accurate risk adjustment, using claims data in the absence of clinical data.

We conducted a targeted analysis of cost drivers among patients who receive chemotherapy because they are expected, based on past research, to have higher costs³. This time frame is consistent with the CMS oncology cancer model, and should be considered a snapshot in time of longer-term costs. Consistent with recent studies, we also found that approximately half the cost in this duration was from chemotherapy and supportive care drug costs, and that there was significant variation in the treatment regimens^5,12. Future studies should include other prevalent cancer types, given recent advances and the expense of newer drugs¹³.

The higher facility costs we observed were consistent with existing research^12,14,15, and likely due to differences in fee schedules, as well as hospitals being tertiary referral centers and managing patients who have progressed through multiple lines of therapy. It is well known that staging is a major predictor of patient outcomes and cost; our data confirmed that metastatic disease was a significant driver across all three cancer types^{3,9,10,16–17}.

Our findings were consistent with existing research showing substantive contribution of the ERG risk score in accounting for comorbidities⁷. In breast cancer, we found that metastatic disease, HER2+ directed therapies, and younger age contributed to higher additional cost. Younger women are more likely to receive aggressive chemotherapy regimens for breast cancer than older women, which likely accounted for the finding on age¹⁸. Our findings were consistent also with existing studies on cost drivers in colorectal¹⁹ and lung cancers¹². Radiation has also been found in recent research to be a major source of expenditures, and has been proposed as a focus for future reimbursement reform^20,21.

One limitation of the present study was that we included patients based on chemotherapy treatment; this could affect cost estimates in breast cancer, because some patients can avoid chemotherapy during the course of their treatment. Future research can drive affordability by examining multiple triggers including surgery, biopsy, and over-utilization of various diagnostic and therapeutic strategies. Our analysis was limited by the availability of pharmacy costs in the data. Including pharmacy costs would give a better picture of the total costs of care in future analyses. Chemotherapy and supportive drug therapies accounted for the highest proportions of total medical costs in our study, but future analyses should include hormonal therapies. Another limitation was that we observed a model fit test result that was not statistically significant in our breast cancer model; as stated by de Jong and Heller, in this case conclusions based on likelihood ratio tests must be regarded as more reliable, as their statistical properties are superior²². We therefore, used the cost driver coefficients from the model to inform our conclusions.

Conclusions

Accurate risk adjustment is key to enable health plans to measure and compare costs and reward providers under APMs. Several stakeholders including health plans have been engaged in building APMs in oncology that optimize costs, utilization, and quality. This has been one of the strategies to shift from volume- to value-based care and to stem the tide of unsustainable rise in overall cost of cancer care. We recommend that payers take into account the influential cost drivers in oncology necessary to appropriately risk-adjust for disease severity when comparing baseline and performance cohorts in APMs²³.

Cigna’s Accountable Care Organization initiative has established a shared savings program that offers practices nationwide an incentive to improve quality and affordability. Our Specialty Care Collaborative in Oncology is built on a similar foundation, incentivizing providers to engage in APMs that reward improved quality and affordability. Our methodology builds on past research, providing a valid tool to estimate costs when only claims data are available. Inclusion of these cost drivers is critical to engage providers in APMs toward improving outcomes and reducing costs, benefitting patients, providers, and payers.

Transparency

Declaration of funding

Work was performed at, and funded by, Cigna. The Cigna name is a registered service mark used herein to refer to operating subsidiaries of CIGNA Corporation, including Cigna Health and Life Insurance Company and Cigna Behavioral Health, Inc.

Declaration of financial/other interests

All authors are employed by Cigna. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Ethical Board Review Statement

This research was conducted as a quality improvement initiative by Cigna in accordance with the Declaration of Helsinki.

Acknowledgments

The authors gratefully acknowledge the technical and editorial assistance provided by Dr Stuart Lustig, and the contributions to conception and design provided by Mary Deary-Weiss.

Appendix: Current Procedural Terminology (CPT) codes for radiation therapy

Table

0073T	0182T	0197T	330	333	339	344	973	32553	32701
330	333	339	344	49411	55875	55876	55920	61796	61797
61798	61799	61800	76950	76965	77011	77014	77261	77262	77263
77280	77285	77290	77295	77299	77300	77301	77305	77306	77307
77310	77315	77316	77317	77318	77321	77326	77327	77328	77331
77332	77333	77334	77336	77338	77370	77371	77372	77373	77385
77386	77387	77399	77401	77402	77403	77404	77406	77407	77408
77409	77411	77412	77413	77414	77416	77417	77418	77421	77422
77423	77427	77431	77432	77435	77470	77499	77520	77522	77523
77525	77600	77605	77610	77615	77620	77750	77761	77762	77763
77776	77777	77778	77781	77782	77783	77784	77785	77786	77787
77789	77790	77799	973	C1715	C1716	C1717	C1719	C1728	C1879
C2616	C2634	C2635	C2636	C2637	C2638	C2639	C2640	C2641	C2642
C2643	C2698	C2699	C9725	C9726	C9728	G0173	G0251	G0338	G0339
G0340	G6001	G6002	G6004	G6005	G6006	G6008	G6009	G6010	G6012
G6013	G6014	G6015	Q3001	S8030