One of the greatest challenges in treating patients with cancer is avoiding emergency department (ED) visits and hospitalizations. The ED is widely considered a suboptimal site of care due to long wait times, lack of continuity with primary care, and excessive testing; these conditions may be especially challenging for cancer patients experiencing exacerbations of symptoms.1 A recent study of patients with gastrointestinal cancers and 2 studies of patients with various cancers, all based on clinician interpretation of medical records or assessment of their own medical encounters, estimated that 19%–23% of ED visits and hospitalizations for cancer patients are potentially avoidable.2–4 Unplanned hospitalizations may cause delays in a patient’s scheduled cancer treatments and disrupt the continuum of care.5 A recent study of Medicare patients with gastrointestinal cancer in Texas estimated that 58% of hospitalizations were unplanned, and many of the presenting complaints could have been treated in primary care.5 Another study using Medicare claims linked to the Surveillance, Epidemiology and End Results (SEER) database estimated the cost of unplanned hospitalizations at $20,412 per admission.6 In addition, 48% of spending and 67% of variation in spending in cancer treatment are attributable to acute hospital care.7
Two models have emerged with the potential to improve symptom management for cancer patients at home or in outpatient settings to reduce reliance on acute care. The oncology medical home uses principles of the patient-centered medical home such as enhanced access to and coordination of care to manage treatment for cancer patients.8,9 Although patient navigation programs vary, the general purpose of these models is to identify and address barriers to accessing cancer care among underserved populations to improve the timeliness of diagnosis and treatment and to ensure consistency in follow-up care. Patient navigators assist patients with the logistics of getting to appointments as well as the interpretation of diagnoses and implementation of treatment plans and also connect patients with supportive services.10 Through the Centers for Medicare and Medicaid Services (CMS) Health Care Innovation Awards, 2 programs were funded to test the oncology medical home and patient navigation models.
Innovative Oncology Business Solutions implemented the Community Oncology Medical Home (COME HOME) model at 7 outpatient oncology practices in New Mexico, Texas, Georgia, Ohio, Florida, and Maine. The main goal of the model was to support patient symptom management and reduce utilization and spending for oncology patients, achieved through implementation of 3 main components: triage pathways to help first responders and nurses identify and manage patient symptoms; extended access to outpatient care during evenings and weekends for the practices’ symptomatic patients after chemotherapy, radiation therapy, or surgery; and diagnosis and treatment pathways based on nationally recognized, evidence-based standards to guide clinical decision making and support patient self-management.11
The University of Alabama at Birmingham (UAB) implemented the Patient Care Connect Program (PCCP), a patient navigation program at 12 oncology practices throughout the south in Alabama, Florida, Georgia, Mississippi, and Tennessee. The PCCP employed and trained nonclinical navigators who educated and empowered cancer patients and survivors, connected patients and caregivers with resources, and improved adherence to care plans.12 In addition, navigators acted as patient advocates and liaisons between patients and providers, often clarifying treatment plans and voicing concerns. Because of the broad scope of navigation activities, navigators provided services in multiple settings, including in patients’ homes, in the hospital, in the ED, and through telephone.
CMS contracted with the NORC at the University of Chicago (NORC) to conduct an independent evaluation of COME HOME and PCCP. This article presents the results of this evaluation. The purpose of our analysis was to estimate the impacts of the 2 programs and assess their relative effectiveness to inform future models to reduce utilization and spending in cancer patients.
For this study, we used 100% Medicare claim files from 2010 to 2015 on the CMS Virtual Research Data Center data enclave environment linked to program files provided by the COME HOME and PCCP models. The program files identified targeted patients and their enrollment dates, as well as practices participating in the program; these files were then linked to Medicare claims. Comparison groups for the 2 programs were constructed from Medicare claims from the same time period.
Patients included in our analysis were Medicare fee-for-service patients, with both Part A and Part B coverage, served by the COME HOME (N=3322) and PCCP (N=1989) programs from July 1, 2013 through June 30, 2015. To ensure comparability of our results across programs, we only included patients with International Classification of Disease, ninth edition (ICD-9) codes for active cancers that were frequently treated in both the COME HOME and PCCP practices (breast, lung, lymphoma, and colorectal cancers). We defined patients with active cancers as those having 1 inpatient or 2 outpatient claims with diagnoses codes for cancer within 3 months before or after enrollment in the program. Using codes available on claims, we also identified patients undergoing chemotherapy to understand the role symptom management played in the effectiveness of each program.13 More information on the program population and how we identified patients undergoing chemotherapy can be found in the Appendix (Supplemental Digital Content 1, http://links.lww.com/MLR/B459).
To identify comparison groups, we first found practices that were similar to the organizations implementing each program. For COME HOME, we used propensity scores to identify comparison sites, using logistic regression models that estimated the probability of a site being in the awardee’s program. Variables in the propensity score model included geographic region and practice-level characteristics such as patient demographics, Medicare oncology patient volume, and volume of oncology services such as chemotherapy and radiation therapy. For PCCP, we selected 2 similar National Cancer Institute Comprehensive Cancer Centers (CCCs) and their 18 affiliated practices that were closest in geographic distance to UAB and mirrored the arrangement between UAB’s CCC and its affiliated hospital sites.
For both COME HOME and PCCP, comparison group participants were associated with specific comparison practices based on their practice identifiers on Medicare claims (tax identification number or CMS certification number). We defined “enrollment” for comparison patients based on their time of seeking treatment for their active cancers, as indicated by diagnoses and procedure codes (for chemotherapy, surgery, or radiation therapy) on claims. We then used propensity score models14 to match each program participant to a similar patient seen at a comparison practice, based on cancer type (breast, colorectal, lung, lymphoma, other), demographic characteristics (age, sex, black race), comorbidities [based on Hierarchical Condition Category (HCC) scores], and hospitalizations and ED visits in the year before program enrollment.14,15 We categorized variables as displayed in Table 1. We performed the matching within each type of cancer to ensure that program participants were matched to a comparator with the same type of cancer. In each of the matched datasets, standardized differences between the treatment and comparison group for demographic covariates, comorbidities, and prior year hospitalizations and ED visits were reduced; this indicates that propensity score matching greatly improved the comparability of the treatment and comparison groups.
Given the lack of information on cancer stage in claims, we use 4 variables as proxies for cancer severity in our propensity score model: diagnosis and procedure code(s) in the 3 months before or after program enrollment for metastatic cancer, surgery for cancer, chemotherapy for cancer, and radiation therapy for cancer.
Study Design and Measures
We conducted a retrospective cohort analysis following each participant and propensity-matched comparator from 2 years before enrollment through up to 3 years after enrollment (or to the end of the study period or death), using an intent-to-treat approach. For participants who died during the study period, we excluded their matched comparators from subsequent analysis, and vice versa, to maintain equivalence between the treatment and comparison groups. We measured quarterly total Medicare spending and utilization (hospitalizations, 30-day readmissions, and ED visits) outcomes for participants and comparators from Medicare claims during each quarter. Total Medicare spending included both Part A and Part B spending. We used logit models with log link and robust SEs to estimate program effect for the quarterly utilization outcomes. For the quarterly spending outcome, we used log-linked generalized linear regression models with a gamma distribution to obtain the program effect estimates. Utilization outcomes are specified as binary (eg, did the patient have at least 1 hospitalization during each quarter?) and presented per 1000 participants. Spending outcomes are continuous and presented as dollars per participants.
We assessed average treatment effects for the 2 programs using difference-in-differences (DID) models.16 These models compare the average outcomes between patients in the treatment group and patients in the comparison group across the entire preenrollment and postenrollment periods while limiting the influence of selection bias and secular trends. For each program, the primary parameter of interest in the DID is the difference in average outcome between the treatment group and a comparison group after exposure to program minus the difference in average outcome between the treatment group and a comparison group before exposure to the program.
The specifications of the DID model is:
Here, post is an indicator variable for time after versus before the program went into effect. Thus, the interaction term β3 allows for estimating the impact of the program for the entire post period. Both unadjusted and adjusted models are presented; adjusted models include covariates for demographic characteristics (age, race/ethnicity, sex), HCC score, dual eligibility, original Medicare entitlement, indicators for cancer severity (metastatic cancer, surgery for cancer, chemotherapy for cancer, and radiation therapy for cancer), and CCC or affiliate status (PCCP only). Statistical analyses were conducted using Stata 13.1 and are presented with 95% confidence intervals (CIs).
As shown in Table 1, a majority of the patients included in the analysis in both models were female and the highest percentage of patients were between the ages of 65 and 69 years. However, there were approximately twice as many black patients in PCCP as in COME HOME, and PCCP participants had more severe comorbidities (as measured by the HCC score), and higher total Medicare spending in the year before enrollment. COME HOME had a higher percentage of patients with breast cancer (46.8% compared with 36.8% for PCCP), and PCCP enrolled far fewer patients under the age of 65 (0.2% compared with 9.6% for COME HOME).
Overall, approximately one-third of COME HOME and PCCP participants died during the study period; for both models, this was significantly lower than the percentage of comparison participants who died during the same period. Mortality during the study period was significantly lower for COME HOME participants with breast cancer (9.5%) and lung cancer (68.5%), relative to the comparison group (13.6% and 74.0%, respectively). Among PCCP participants, we observed significantly lower rates of mortality within all cancer types, relative to the comparison group.
Unadjusted estimates for the 2 programs and comparators in the preimplementation period and the postimplementation period are presented online (Figs. S1–S4, Supplemental Digital Content 1, http://links.lww.com/MLR/B459). For COME HOME participants, we observe 9 fewer hospitalizations (95% CI, −15 to −3) and 13 more ED visits (95% CI, 5–21) in the preenrollment period, as well as 26 fewer 30-day readmissions in the postenrollment period, per 1000 patients relative to the comparison group. In the preenrollment period, PCCP participants have 8 fewer hospitalizations (95% CI, −15 to −1) and 33 fewer 30-day readmissions (95% CI, −56 to −10), per 1000 patients. In the postenrollment period, PCCP participants have 31 fewer hospitalizations (95% CI, −43 to −20), 28 fewer ED visits (95% CI, −41 to −14), and 27 fewer 30-day readmissions (95% CI, −50 to −3), per 1000 patients. In addition, PCCP participants also have significantly lower costs in the preenrollment period (−$403 per beneficiary; 95% CI, −$652 to −$156) and the postenrollment period (−$772 per beneficiary; 95% CI, −$1268 to −$277), relative to the comparison group.
Adjusted DID estimates for both programs are presented in Table 2. COME HOME was associated with significantly fewer ED visits (15 per 1000 patients/quarter; 95% CI, −24 to −6) and a reduction in Medicare spending of $675 per patient per quarter (95% CI, −$1118 to −$232), relative to a comparison group. The COME HOME program was also associated with a nonsignificant reduction in 30-day readmissions. PCCP was significantly associated with 11 fewer hospitalizations (95% CI, −20 to −2) and 22 fewer ED visits (95% CI, −33 to −11) per 1000 patients per quarter relative to the comparison group, but did not achieve statistically significant reductions in spending or 30-day readmissions. For patients undergoing chemotherapy, both COME HOME and PCCP were associated with decreases in ED visits of 18 (95% CI, −31 to −5) and 28 (95% CI, −43 to −13) per 1000 patients per quarter, respectively. PCCP also showed a reduction in hospitalizations of 13 per 1000 patients per quarter (95% CI, −24 to −2), relative to a comparison group. There is no evidence that either program significantly affected the total spending on care for the subgroup of chemotherapy patients.
Through the CMS Health Care Innovation Awards, COME HOME and PCCP were given the opportunity to test alternate models of oncology care. Both were successful in reducing ED visits. This result is consistent with patient and caregiver-reported feedback on the models. Patients from both programs expressed how the enhanced access to the clinic or navigator helped them avoid unnecessary ED visits.17 Although COME HOME and PCCP offered 2 different care delivery models, both programs encouraged patients to call their physician practice or navigator first rather than go directly to the ED if they had concerns. For COME HOME patients, the dedicated triage line gave patients direct access to nurses, and patients could also take advantage of same-day appointments and extended and weekend hours. PCCP patients were encouraged to contact their navigators with any issues or concerns.
PCCP had stronger effects on utilization, particularly for patients undergoing chemotherapy, whereas COME HOME was associated with larger reductions in spending. Variation in program features and the composition of patients between the 2 programs may explain the differences in the results for COME HOME and PCCP. For instance, COME HOME had more participants with breast cancer, who tended to be women. Interviews with program staff suggested that women utilized the triage line and extended office hours more frequently than men. PCCP served a population with a greater percentage of African American patients with more barriers to obtaining cancer care and poorer outcomes for cancer, and higher prior-year medical expenditures. Although PCCP may have been well suited to many underserved patients as evidenced by reduced utilization, they may not have been able to effectively reach the most vulnerable patients and thus did not reduce overall spending.
This preliminary evidence suggests that the effectiveness of innovative models of oncology care varies for key subpopulations. In designing future models for cancer care, policy makers and advocates should consider the specific needs of the population to be served. Different clinical profiles and sociodemographic factors may influence how responsive patients are to particular interventions.
Our study makes several important contributions to the literature. This is the first study using a comparison group to provide evidence of the effectiveness of oncology medical homes. The one published evaluation of an oncology medical home relied on pre-post data and found improvements in utilization.18 This is also the first study to demonstrate a link between patient navigation programs for oncology patients and reduced utilization. In previous research, patient navigation has been associated with more timely cancer screening, diagnostic resolution, and treatment initiation but no link has been made to decreased ED visits or hospitalizations.19,20
Our findings are particularly relevant considering the recently launched Oncology Care Model (OCM),21 which will allow elements of both the oncology medical home and patient navigation models to be tested on a larger scale. OCM provides episode-based payments to physician practices to manage cost and quality for patients undergoing chemotherapy.21 Among the practice redesign activities that OCM sites are required to implement are patient navigation and 24/7 access to clinicians.20,22 In evaluating the effectiveness of OCM, it will be important to note differential effects by intervention type and population subgroups served.
This analysis has several limitations. First, we lacked reliable measures of cancer severity and disease stage due to our reliance on claims data. We approximated this information using indicators for metastatic cancer (for which prior evidence suggests somewhat limited validity)23 and previous cancer treatment (surgery, radiation, and/or chemotherapy), but without access to medical records we cannot completely adjust for severity and stage.
Second, we only analyzed Medicare fee-for-service enrollees and with cancers common to both programs. Although these restrictions were necessary for obtaining claims data and conducting comparisons across the 2 programs, we may not have captured the overall impact of each program on their entire participant population.
Third, the chemotherapy subgroup analysis included only patients receiving infused chemotherapy covered under Medicare Part B, and not patients receiving only oral chemotherapy covered under Medicare Part D. Moreover, our estimates of total Medicare spending did not include Part D spending. It is estimated that ∼20% of patients receive oral chemotherapy, whether alone or in conjunction with intravenous chemotherapy24,25 (see Appendix for additional details, Supplemental Digital Content 1, http://links.lww.com/MLR/B459).
In addition, it is worth noting that our analysis did not directly assess the necessity of each ED visit or hospitalization and we could not extrapolate the potential reason for ED visits or hospitalizations that were prevented. Therefore it is conceivable that some of the averted utilization could have been clinically appropriate rather than avoidable and unnecessary.
Finally, we lacked data on the quality of care provided by the awardees. Our evaluation focused on core utilization and cost measures, and we did not quantitatively assess patient experience with care, adherence to clinical standards, or other measures of quality of care.
Both the COME HOME oncology medical home and the PCCP patient navigation programs showed promise in reducing utilization among cancer patients, but the oncology medical home model showed greater impacts in spending and the patient navigation program showed larger decreases in hospitalizations and ED visits. The differential effects may be explained by the particular populations served by the respective programs. Evidence from our evaluation can help inform new models of coordinating care for patients with cancer and should be considered in evaluating the OCM.
The authors would like to acknowledge the assistance of Dan Gilden of JEN Associates in constructing analytic files.
1. Committee on the Future of Emergency Care in the United States Health System. Hospital-based Emergency Care: At the Breaking Point. Washington, DC: National Academies Press; 2007.
2. Brooks GA, Abrams TA, Meyerhardt JA, et al. Identification of potentially avoidable hospitalizations in patients with GI cancer. J Clin Oncol. 2014;32:496–503.
3. Delgado-Guay MO, Kim YJ, Shin SH, et al. Avoidable and unavoidable visits to the emergency department among patients with advanced cancer receiving outpatient palliative care. J Pain Symptom Manage. 2015;49:497–504.
4. Brooks GA, Jacobson JO, Schrag D. Clinician perspectives on potentially avoidable hospitalizations in patients with cancer. JAMA Oncol. 2015;1:109–110.
5. Manzano JG, Luo R, Elting LS, et al. Patterns and predictors of unplanned hospitalization in a population-based cohort of elderly patients with GI cancer. J Clin Oncol. 2014;32:3527–3533.
6. Fessele KL, Hayat MJ, Mayer DK, et al. Factors associated with unplanned hospitalizations among patients with nonmetastatic colorectal cancers intended for treatment in the ambulatory setting. Nurs Res. 2016;65:24–34.
7. Brooks GA, Li L, Uno H, et al. Acute hospital care is the chief driver of regional spending variation in Medicare patients with advanced cancer. Health Aff. 2014;33:1793–1800.
8. Eagle D, Sprandio J. A care model for the future: the oncology medical home. Oncology. 2011;25:571. 575–576.
9. Page RD, Newcomer LN, Sprandio JD, et al. The patient-centered medical home in oncology: from concept to reality. Am Soc Clin Oncol Educ Book. 2015;35:e82–e89.
10. Wells KJ, Battaglia TA, Dudley DJ, et al. Patient navigation: state of the art or is it science? Cancer. 2008;113:1999–2010.
11. Waters TM, Webster JA, Stevens LA, et al. Community oncology medical homes: physician-driven change to improve patient care and reduce costs. J Oncol Pract. 2015;11:462–467.
12. Rocque GB, Partridge EE, Pisu M, et al. The Patient Care Connect Program: transforming health care through lay navigation. J Oncol Pract. 2016;12:e633–e642.
13. Sikorskii A, Given CW, Given B, et al. Symptom management for cancer patients: a trial comparing two multimodal interventions. J Pain Symptom Manage. 2007;34:235–264.
14. Seeger JD, Williams PL, Walker AM. An application of propensity score matching using claims data. Pharmacoepidemiology Drug Saf. 2005;14:465–476.
15. Pope GC, Kautter J, Ellis RP, et al. Risk adjustment of Medicare capitation payments using the CMS-HCC model. Health Care Financ Rev. 2004;25:119–141.
16. Freedland SJ, Humphreys EB, Mangold LA, et al. Mortality following prostate cancer recurrence after radical prostectomy-reply. JAMA. 2005;294:2969–2970.
18. Kuntz G, Tozer JM, Snegosky J, et al. Michigan Oncology Medical Home Demonstration Project: first-year results. J Oncol Pract. 2014;10:294–297.
19. Freund KM, Battaglia TA, Calhoun E, et al. Impact of patient navigation on timely cancer care: the Patient Navigation Research Program. J Natl Cancer Inst. 2014;106:dju115.
20. Paskett ED, Harrop JP, Wells KJ. Patient navigation: an update on the state of the science. CA Cancer J Clin. 2011;61:237–249.
23. Chawla N, Yabroff KR, Mariotto A, et al. Limited validity of diagnosis codes in Medicare claims for identifying cancer metastases and inferring stage. Ann Epidemiol. 2014;24:666–72.e2.
24. Weingart SNBE, Bach PB, Eng K, et al. NCCN task force report: oral chemotherapy. J Natl Compr Canc Netw. 2008;6:S8–S14.
25. Zerillo JASS, Fraile B, Dodek AD, et al. Understanding oral chemotherapy prescribing patterns at the end of life at a comprehensive cancer center: analysis of a Massachusetts Payer Claims Database. J Oncol Pract. 2015;11:372–377.