The Impact of Intensity Modulated Radiation Therapy on Hospitalization Outcomes in the SEER-Medicare Population With Anal Squamous Cell Carcinoma

Abstract

Purpose:

We examined the impact of intensity modulated radiation therapy (IMRT) on hospitalization rates in the Surveillance, Epidemiology, and End Results (SEER) —Medicare population with anal squamous cell carcinoma (SCC).

Methods and Materials:

We performed a retrospective cohort study using the SEER-Medicare database. We identified patients with nonmetastatic anal SCC diagnosed between 2001 and 2011 and treated with chemoradiation therapy. We assessed the relation between IMRT and first hospitalization by use of a multivariate competing-risk model, as well as instrumental variable analysis, using provider IMRT affinity as our instrument.

Results:

Of the 1165 patients included in our study, 458 (39%) received IMRT. IMRT use increased over time and was associated more with regional and provider characteristics than with patient characteristics. The 3- and 6-month cumulative incidences of first hospitalization were 41.9% (95% confidence interval [CI], 37.3%−46.4%) and 47.6% (95% CI, 43.0%−52.2%), respectively, for the IMRT cohort and 46.7% (95% CI, 43.0%−50.4%) and 52.1% (95% CI, 48.4%−55.7%), respectively, for the non-IMRT cohort. IMRT was associated with a decreased hazard of first hospitalization compared with 3-dimensional radiation techniques (hazard ratio, 0.70; 95% CI, 0.58–0.84; PZ.0002). Instrumental variable analysis suggested an even greater reduction in hospitalizations with IMRT after controlling for unmeasured confounders. There was a trend toward improved overall survival with IMRT, with an adjusted hazard ratio of 0.77 (95% CI, 0.59–1.00; PZ.05).

Conclusions:

The use of IMRT is associated with reduced hospitalizations in elderly patients with anal SCC. Further work is warranted to understand the long-term health and cost impact of IMRT, particularly for patient subgroups most at risk of toxicity and hospitalization.

Summary

The relation between intensity modulated radiation therapy (IMRT) and first hospitalization in patients with squamous cell carcinoma was evaluated with the Surveillance, Epidemiology, and End Results-Medicare database. We found through a multivariate competing-risk model, as well as instrumental variable analysis, that the use of IMRT was associated with reduced hospitalizations compared with conventional radiation techniques. In addition, there was a trend toward improved overall survival with IMRT.

Introduction

Combined radiation therapy–chemotherapy with 5-fluorouracil (5-FU) and mitomycin C (MMC) has been the standard of care for squamous cell carcinoma (SCC) of the anal canal for nearly 4 decades (1–5). However, conventional radiation therapy approaches are associated with high treatment-related morbidity rates (3), which may compromise therapeutic efficacy because of prolonged treatment breaks and inability to deliver full doses of radiation therapy and/or chemotherapy (6–8). Intensity modulated radiation therapy (IMRT) is an advanced radiation technique that allows the conformal delivery of radiation to target tissue and minimizes dose to normal tissue including small bowel, bladder, external genitalia, femoral heads, and iliac crests (9). Radiation Therapy Oncology Group (RTOG) 0529—a phase 2 trial of IMRT with 5-FU and MMC—and several retrospective studies have reported a reduced frequency of treatment breaks, improved acute toxicity profile, positive impact on quality of life, and favorable early disease outcome with IMRT (10–14). However, IMRT requires expertise, careful target design, and intensive physics and quality-assurance support. Central review prior to treatment start on RTOG 0529 revealed that 81% of submitted plans required revision.

Given the complexities of IMRT, we sought to examine its impact on outcomes in patients with anal SCC as it is used in the real-world setting by use of the Surveillance, Epidemiology, and End Results (SEER)–Medicare database. We examined patterns of utilization of IMRT, as well as the impact of the use of IMRT on hospitalization rates and survival.

Methods and Materials

Study overview and cohort selection

We performed a retrospective cohort study using the 2014 linkage of the SEER-Medicare database to examine the utilization of IMRT in elderly patients with anal SCC, as well as the subsequent impact of this treatment on hospitalization rates. We identified patients with invasive, nonmetastatic, intact anal SCC diagnosed between 2001 and 2011 and treated with both chemotherapy (capecitabine, 5-FU, MMC, or cisplatin) and radiation therapy. Additional details about the SEER-Medicare database and our cohort selection process are available in Appendix E1 (available online at www.redjournal.org). The Institutional Review Board of Stanford University deemed this study exempt from review.

Primary exposure variable

We ascertained radiation therapy administration from the Medicare claims (Appendix E2; available online at www.redjournal.org). The cohort was categorized as having received IMRT if any IMRT delivery or planning codes were present. If patients did not have an IMRT planning or delivery code as part of their radiation claims, then they were assumed to have undergone conventional 3-dimensional (3D) radiation techniques.

Study covariates

We used the SEER database to obtain demographic, disease, and socioeconomic characteristics. We used state buyin status as an indicator of dual eligibility of patients for both Medicare and Medicaid. We created a composite measure for area socioeconomic status based on 3 different variables from the 2000 US Census data: median household income, percentage of persons aged ≥25 years with at least a high school education, and percentage of persons living below the poverty level (15). The composite measure was constructed by summing the z scores for each of the 3 variables and then classified into quartiles. We used the Area Resource File to determine radiation oncologist density in the Health Service Area (HSA) to which each patient belonged. An HSA is defined by the National Center for Health Statistics as a single county or cluster of contiguous counties that are relatively self-contained with respect to hospital care. The density of radiation oncologists per HSA was determined by dividing the number of radiation oncologists by the Medicare-eligible population for a given HSA and categorized into quartiles.

We calculated a modified Charlson Comorbidity Index by using inpatient and outpatient claims for an interval before cancer diagnosis of 1 to 12 months (16–20). We also included a validated measure of predicted poor disability status as a claims-based proxy for poor performance status (21, 22). We determined human immunodeficiency virus (HIV) status through International Classification of Diseases, Ninth Revision codes as outlined in the Centers for Disease Control and Prevention coding guidelines (23, 24). We identified chemotherapy administration with Medicare claims (Appendix E2; available online at www.redjournal.org) using previously described methods (25). To address potential stage migration, positron emission tomography (PET) use prior to treatment was determined. As with radiation therapy, we only considered chemotherapy claims within 1 month prior to and 6 months after the diagnosis date to avoid counting treatment courses for disease progression or recurrence.

We attributed radiation therapy to being delivered at a hospital-associated radiation treatment facility if radiation therapy claims were present in the outpatient claims (26). Patients whose radiation therapy claims were present in only the carrier claims were considered to have received their treatments at a freestanding radiation treatment facility. By use of the Hospital File based on data submitted to the Centers for Medicare & Medicaid Services in the Healthcare Cost Reports and the Provider of Service survey, hospital-associated facilities were additionally categorized as National Cancer Institute (NCI)—Designated Cancer Centers if they achieved a clinical or comprehensive cancer center designation; moreover, we determined whether they had a residency program during the study period.

Primary outcome

Our primary outcome was time to first hospitalization from start of radiation therapy. We identified hospitalizations as inpatient admissions using the Medicare Provider Analysis and Review hospital stay file. We excluded admissions to skilled nursing facilities. We defined unplanned hospitalizations as hospitalizations coming through the emergency department or with an admission type of urgent or emergent, excluding those for primarily chemotherapy, radiation therapy, or rehabilitation services (27). We considered the first noncancer diagnosis in the claims data to be the reason for admission (27) and used the Agency for Healthcare Research and Quality Clinical Classifications Software to assign International Classification of Diseases, Ninth Revision diagnoses into clinically meaningful categories. We also examined all-cause and cancer-specific mortality (28, 29).

Statistical analysis

Bivariate associations between covariates and receipt of IMRT were evaluated with Pearson χ² tests. We used multivariate logistic regression to evaluate associations between receipt of IMRT and patient characteristics. We used a reverse, stepwise selection process to construct a working model, retaining variables with P<.1.

We assessed the relation between IMRT and first hospitalization (unplanned and all cause) after radiation therapy start using a competing-risk model, with death as a competing risk (30). Patients were censored on December 13, 2013, or when they stopped receiving Part A or B coverage or began receiving health maintenance organization coverage (Appendix E1; available online at www.redjournal.org), whichever came first, if they had not died or experienced hospitalization by that time. We assessed the relation between IMRT and all-cause and cause-specific mortality using Cox proportional hazards models. Persons surviving past December 31, 2013, were censored. For cause-specific mortality, persons who died of noncancer causes were also censored. Predictors were checked for departures from the proportional hazards assumption visually and by use of Schoenfeld residuals (31). All models were adjusted for the following covariates that were selected a priori: age, gender, marital status, race, comorbidity, disability status, HIV status, tumor stage, year of diagnosis, chemotherapy regimen, PET scan at diagnosis, treatment at freestanding or NCI center, state buy-in, socioeconomic composite index, and SEER region.

To address bias due to unmeasured confounding factors, we performed an instrumental variable analysis (32) for our primary outcome. Other authors have used provider experience with IMRT at the site of interest to study the impact of IMRT on outcomes (33). Rather than using physician IMRT affinity for patients with anal cancer, we used physician IMRT affinity for treating non-anal gastrointestinal cancers as our instrument because we assumed that this instrument would be more likely to affect outcome through only IMRT treatment than other factors associated with experience with treating patients with anal cancer. Additional details of our instrumental variable analysis are provided in Appendix E3 (available online at www.redjournal.org). Statistical analyses were performed with SAS Enterprise Guide (version 7.12; SAS Institute, Cary, NC).

Results

We identified a total of 1165 patients with invasive, nonmetastatic anal SCC treated with chemoradiation therapy within 6 months of diagnosis between 2001 and 2011 (Appendix E1; available online at www.redjournal.org). Of these, 458 (39.3%) received IMRT. The median age was 70 years (interquartile range, 63–76 years). Table 1 summarizes the baseline characteristics of included patients. The median follow-up period for all patients in this study was 47.4 months.

Table 1.

Baseline characteristics associated with receipt of IMRT among patients with anal squamous cell cancer treated with chemoradiation therapy in SEER-Medicare database

Characteristic	n	% treated with IMRT	P value*
Entire cohort	1165	39.3
Age at diagnosis			.32
≤65 y	318	43.1
66–70 y	292	39.4
71–75 y	229	35.4
>75 y	326	38.3
Sex			.005
Male	414	44.7
Female	751	36.4
Race			.0005
White	1037	38.0
Nonwhite†	128	50.0
Marital status			.05
Married	401	35.9
Single‡	713	40.3
Unknown	51	52.9
Comorbidity index			.03
0	643	37.2
1	213	36.6
≥2	309	45.6
Disability status			.78
Good	1078	39.4
Poor	87	37.9
HIV			.06
Yes	95	48.4
No	1070	38.5
SEER historic stage			.50
Local	650	38.2
Regional	424	41.5
Unknown	91	37.4
AJCC T category			.52
T1	192	41.7
T2	401	39.9
T3	165	42.4
T4	71	32.4
Unknown or no primary identified	336	37.2
AJCC N category			.06
N0	826	39.0
N1	64	50.0
N2	107	43.0
N3	53	45.3
Unknown	115	29.6
AJCC composite stage			.41
I	156	41.0
II	409	38.6
III	273	42.9
Unknown	327	36.4
Year of diagnosis			<.0001
2001–2003§	208	5.3
2004 or 2005	214	12.1
2006 or 2007	213	32.4
2008 or 2009	256	54.7
2010 or 2011	274	77.4
SEER registry∥			<.0001
Northeast	149	22.1
Midwest	102	32.4
South	343	34.1
West	571	48.2
Chemotherapy			.05
Mitomycin based	878	41.0
Cisplatin based	128	29.7
Neither mitomycin nor cisplatin based	159	37.7
PET scan at diagnosis			<.0001
Yes	483	62.5
No	682	22.9
Dual eligible			.69
Yes	318	40.3
No	847	39.0
SES composite¶			.12
First quartile (lowest)†	354	37.0
Second quartile	282	44.3
Third quartile	273	35.5
Fourth quartile (highest)	256	41.0
Rural-urban classification			.1
Metropolitan	984	40.3
Urban or rural	181	33.7
HSA radiation oncologist density			.0009
First quartile (lowest)†	365	31.5
Second quartile	283	45.6
Third quartile	271	42.8
Fourth quartile (highest)	246	39.8
Radiation treatment center			.0004
Freestanding center	392	46.4
Hospital-based outpatient center	773	35.7
NCI-Designated Cancer Center			.0003
Yes	58	62.1
No	1107	38.1
Residency program			.5
Yes	482	38.2
No	683	40.1
Provider experience with IMRT			<.0001
Upper half (>3.4 patients/y)	601	50.4
Lower half	564	27.5

Abbreviations: AJCC = American Joint Committee on Cancer; HIV = human immunodeficiency virus; HSA = Health Service Area; IMRT = intensitymodulatedradiationtherapy;NCI = NationalCancer Institute; PET = positron emission tomography; SEER = Surveillance, Epidemiology, and End Results; SES = socioeconomic status.

^*P values are based on the Pearson χ² test.

^†Patients in whom this characteristic was deemed unknown are included in this category for privacy purposes because of the low number of patients.

^‡Single includes unmarried, divorced, separated, and widowed.

^§The 2001–2003 categories are combined in this table for privacy purposes because of the low number of IMRT patients.

^‖West comprises the San Francisco, Hawaii, New Mexico, Seattle, Utah, San Jose, Los Angeles, and Greater California Registries; Midwest comprises the Detroit and Iowa Registries; Northeast comprises the Connecticut and New Jersey Registries; and South comprises the Atlantic, Rural Georgia, Kentucky, Louisiana, and Greater Georgia Registries.

^¶A composite measure for area SES is presented based on the following variables from the 2000 US Census data: median household income, percentage of persons aged ≥25 years with at least a high school education, and percentage of persons living below the poverty level.

Factors associated with IMRT use

The use of IMRT increased over time, from 6.5% of patients in our cohort treated with IMRT in 2001 to 79.5% in 2011 (+8.4% per year, P<.0001). In addition to more recent calendar year, residence in the West, treatment at a freestanding or NCI-Designated Cancer Center, having a PET scan prior to treatment, and having treating physicians who had high IMRT affinity were all independently associated with IMRT use (Table 2).

Table 2.

Factors independently associated with receipt of IMRT in multivariate logistic model

Predictor	Adjusted OR for receipt of IMRT	95% CI	P value
Year of diagnosis
2001–2003*	Reference		-
2004 or 2005	2.25	0.98–5.19	.06
2006 or 2007	8.09	3.72–17.61	<.0001
2008 or 2009	20.02	9.11–44.03	<.0001
2010 or 2011	61.71	27.36–139.15	<.0001
PET scan
Yes	Reference		-
No	0.57	0.40–0.82	.002
SEER registry
West	Reference		-
Northeast	0.26	0.15–0.47	<.0001
Midwest	0.46	0.24–0.89	.02
South	0.48	0.32–0.73	.0006
HSA radiation oncologist density
First quartile (lowest)	Reference		-
Second quartile	2.05	1.27–3.32	.003
Third quartile	1.36	0.84–2.20	.21
Fourth quartile (highest)	1.51	0.84–2.19	.22
Radiation treatment center
Freestanding center	Reference		-
Hospital-based outpatient center	0.46	0.29–0.73	.001
NCI-Designated Cancer Center
Yes	Reference		-
No	0.23	0.10–0.51	.0003
Physician experience with IMRT
Upper half	Reference		-
Lower half	0.35	0.24–0.49	<.0001

Abbreviations: CI = confidence interval; HSA = Health Service Area; IMRT = intensity modulated radiation therapy; NCI = National Cancer Institute; OR = odds ratio; PET = positron emission tomography; SEER = Surveillance, Epidemiology, and End Results.

Variables with P<.1 were retained in the model.

^*The 2001 to 2003 categories are combined in this table for privacy purposes because of the low number of IMRT patients.

Hospitalization outcomes after IMRT

Figure 1 shows the unadjusted cumulative incidence curves of first hospitalization after start of radiation therapy among patients who received IMRT and those who received 3D radiation therapy. The unadjusted 3- and 6-month cumulative incidences of first hospitalization were 41.9% (95% confidence interval [CI], 37.3%−46.4%) and 47.6% (95% CI, 43.0%−52.2%), respectively, for the IMRT cohort and 46.7% (95% CI, 43.0%−50.4%) and 52.1% (95% CI, 48.4%−55.7%), respectively, for the non-IMRT cohort. IMRT was associated with decreased hazard of hospitalization compared with 3D radiation techniques (hazard ratio, 0.70; 95% CI, 0.58–0.84; P=.0002) (Table 3). The results of our instrumental variable analysis suggested an even greater risk reduction in hospitalization with the use of IMRT during the 3- and 6-month periods after initiation of radiation treatment (Appendix E3; available online at www.redjournal.org).

Fig. 1.

Unadjusted cumulative incidence curves of first hospitalization in intensity modulated radiation therapy (IMRT) cohort versus non-IMRT cohort, with death as a competing risk.

Table 3.

Unadjusted and adjusted HRs for first hospitalization, overall survival, and cause-specific survival with IMRT

	HR with IMRT	95% CI	P value
First hospitalization
Unadjusted	0.83	0.73–0.94	.005
Adjusted model	0.70	0.58–0.84	.0002
Overall survival
Unadjusted	0.87	0.72–1.06	.18
Adjusted model	0.77	0.59–1.00	.05
Cause-specific survival
Unadjusted	0.67	0.50–0.90	.008
Adjusted model	0.75	0.51–1.10	.14

Abbreviations: CI = confidence interval; HR = hazard ratio; IMRT = intensity modulated radiation therapy.

Other factors associated with increased hazard of hospitalization on multivariate regression were comorbidity, older age, HIV positivity, mitomycin- or cisplatin-based chemotherapy, and advanced tumor stage (Table 4). Table 5 lists the most common noncancer reasons for first hospitalization in IMRT versus non-IMRT patients within the first 3 months of starting radiation therapy.

Table 4.

Factors (in addition to radiation technique) independently associated with first hospitalization on multivariate regression

	Hospitalization
	Adjusted HR*	95% CI	P value
Age at diagnosis			<.0001
65 y	Reference	-
66–70 y	1.18	0.94–1.48
71–75 y	1.48	1.16–1.88
>75 y	1.79	1.41–2.27
Comorbidity index			.002
0	Reference	-
1	1.14	0.95–1.37
≔2	1.43	1.17–1.75
HIV			.01
Yes	Reference	-
No	0.69	0.51–0.92
SEER historic stage			.03
Local	Reference	-
Regional	1.17	1.01–1.36
Chemotherapy			.02
Mitomycin based	Reference	-
Cisplatin based	1.07	0.85–1.36
Neither mitomycin nor cisplatin based	0.75	0.60–0.93

Abbreviations: CI = confidence interval; HIV = human immunodeficiency virus; HR = hazard ratio; SEER = Surveillance, Epidemiology, and End Results.

^*We also adjusted for gender, marital status, race, disability status, tumor stage, year of diagnosis, positron emission tomography scan at diagnosis, treatment at freestanding or National Cancer Institute center, state buy-in, socioeconomic composite index, and SEER region; these were not independently associated with hospitalization and are not shown.

Table 5.

Most common noncancer reasons for first hospitalization within 3 months of start of radiation treatment by AHRQ Clinical Classifications Software

	IMRT		3D
	n	%	n	%
Total patients at risk	458	-	707	-
Total first hospitalizations	192	41.9	330	46.7
Digestive	41	9.0	66	9.3
Blood and/or bone marrow	45	9.8	60	8.5
Infection	31	6.8	51	7.2
Fluid and electrolyte disorders, renal, and/or GU	21	4.6	56	7.9
Cardiovascular and/or pulmonary	17	3.7	22	3.1
Injury and poisoning	20	4.4	23	3.3

Abbreviations: AHRQ = Agency for Healthcare Research and Quality; GU = genitourinary; IMRT = intensity modulated radiation therapy; 3D = 3-dimensional.

Unadjusted rates are presented. Categories with fewer than 11 patients are not shown or are combined because of SEER-Medicare privacy requirements.

IMRT was associated with decreased unplanned hospitalization compared with 3D radiation techniques, with an adjusted hazard ratio of 0.79 (95% CI, 0.65–0.96; P=.02). The 3- and 6-month unadjusted cumulative incidences of unplanned hospitalization were 37.1% (95% CI, 32.7%41.5%) and 42.4% (95% CI, 37.8%−46.9%), respectively, for the IMRT cohort and 38.5% (95% CI, 34.9%−42.1%) and 42.4% (95% CI, 38.8%−46.1%), respectively, for the non-IMRT cohort.

Patients treated with IMRT were in the hospital an average of 4.2 days (vs 5.5 days for non-IMRT patients, P=.04) during the first 3 months and an average of 5.6 days (vs 6.6 days for non-IMRT patients, P=.05) during the first 6 months of starting radiation therapy. Finally, patients treated with IMRT were more likely to receive ≥2 cycles of MMC- or cisplatin-based chemotherapy (56.1% of IMRT group received ≥2 cycles vs 45.5% of non-IMRT group, P=.0004).

Survival outcomes after IMRT

The 2-year overall and cause-specific survival rates among patients who received IMRT were 79.9% (95% CI, 75.9%−83.3%) and 89.5% (95% CI, 86.1%−92.0%), respectively. The 2-year overall and cause-specific survival rates among those who received 3D radiation therapy were 79.5% (95% CI, 76.3%−82.3%) and 85.7% (95% CI, 82.7%−88.0%), respectively. After adjustment for demographic, tumor, patient, treatment, and socioeconomic characteristics, there was a trend toward improved overall survival with IMRT, with an adjusted hazard ratio of 0.77 (95% CI, 0.59–1.00; P=.05) (Table 3).

Discussion

Although chemoradiation therapy is a curative, organ-preserving treatment approach for anal SCC, this treatment is highly toxic. Acute treatment-associated morbidity can result in hospital admissions, which contribute a substantial portion to Medicare costs in elderly patients with cancer (34). In addition, hospitalizations can interrupt treatment, increase risk of nosocomial infections, and lower quality of life for patients and their families. IMRT can improve tolerability of treatment by reducing dose to adjacent normal organs (9). We evaluated IMRT as it is used in the real-world setting and found that its use is associated with reduced hospitalizations in SEER-Medicare patients with anal SCC.

On the basis of the assumption that hospitalizations can serve as a surrogate for severe toxicity, our data reflect the significant acute toxicity seen with combined-modality treatment in the clinical trial setting. In our study the 3-month cumulative incidences of first hospitalization were 41.9% (95% CI, 37.3%−46.4%) and 46.7% (95% CI, 43.0%−50.4%) for the IMRT and non-IMRT cohorts, respectively, with the most common noncancer reasons for hospitalizations including digestive and hematologic issues, infection, and fluid and electrolyte disorders. RTOG 9811 reported rates of acute nonhematologic and hematologic grade 3 or 4 toxicities of 74% and 61%, respectively, in patients receiving MMC-based chemoradiation therapy with conventional radiation techniques (3). RTOG 0529 demonstrated lower rates of toxicity with IMRT: grade 3 or 4 nonhematologic toxicity rate of 44% and grade 3 or 4 hematologic toxicity rate of 58% during the period from treatment start to 2 months after treatment end. Other retrospective series of IMRT have also reported improved toxicity rates ranging from 7% to 10% for grade 3 or higher gastrointestinal toxicity, 24% to 59% for grade 3 or higher hematologic toxicity, and 21% to 37% for grade 3 or higher skin toxicity (11–14).

Other population-based studies of patients undergoing cancer therapy have found a similar burden of hospitalizations. In one study, among SEER-Medicare patients with colorectal cancer, 77.5% experienced at least 1 unplanned hospitalization during chemotherapy (35). Another study, using linked Texas Cancer Registry and Medicare database information, found that 55.9% of patients with anorectal cancer had at least 1 unplanned hospitalization within 1 year of diagnosis (27). Waddle et al (36) specifically looked at unplanned hospitalization rates within 90 days of starting radiation therapy and reported a 21% hospitalization rate for all gastrointestinal cancers. Our numbers are higher likely because our cohort consisted of mostly elderly patients who underwent chemotherapy in addition to radiation therapy, and Waddle et al found that concurrent chemotherapy predicted for a higher rate of hospitalization.

Additional factors that were independently associated with hospitalization included the presence of multiple comorbidities, higher tumor stage, HIV positivity, and older age; other investigators have reported many of these same factors as predictors of hospitalization in other cancer sites (27, 37, 38). Worse acute and late toxicities with chemoradiation therapy have been reported among HIV-positive patients (39, 40). We also found that patients who received MMC- or cisplatin-based chemotherapy were more likely to be hospitalized than patients who received 5FU or capecitabine alone, which is consistent with the increased toxicity seen with the addition of MMC compared with 5-FU alone (1). It is interesting that we found no difference in hospitalizations between patients who received MMC-based chemotherapy and those who received cisplatin-based chemotherapy. Ajani et al (3) found that grade 3 to 4 hematologic toxicities were worse with MMC compared with cisplatin but acute nonhematologic and late toxicities were similar.

As with any observational study, there is concern for confounding and selection bias. Although we adjusted for many observed confounders in our multivariate analyses, to address possible unmeasured differences between our IMRT and 3D cohorts, we performed an instrumental variable analysis using provider IMRT affinity in the treatment of gastrointestinal cancers other than anal cancer as our instrument. The advantage of instrumental variable analysis over multivariate regression or propensity score analyses is that all possible confounders do not need to be identified to provide unbiased estimates of treatment effect. Our instrumental variable analysis showed an even greater risk reduction of hospitalization with the use of IMRT, suggesting that if we account for unmeasured confounders, the true impact of IMRT on reducing hospitalizations is likely greater than what we found in our primary analysis.

Given the complexity of IMRT planning and delivery, as well as concern for tumor miss and suboptimal disease control with user variability, we also evaluated IMRT’s integration into general practice and its subsequent impact on survival. We found that IMRT was more likely to be delivered by providers who had more experience with using IMRT. Furthermore, IMRT was associated with a trend toward improved, and not inferior, survival. We speculate that by reducing toxicities and hospitalizations, IMRT can potentially translate into improved outcomes by minimizing treatment interruptions. This is supported by our finding that patients treated with IMRT received more cycles of chemotherapy. Finally, while other studies have suggested that lower socioeconomic status and black race were associated with a lower rate of IMRT use (41–44), we did not find that area socioeconomic status, dual-eligibility status, or nonwhite race was independently associated with IMRT use. We also did not find any racial or socioeconomic disparities in early access to IMRT when we looked at only the years prior to 2007, before RTOG 0529 opened. Although we adjusted for HIV positivity based on claims, it is possible that HIV status confounded these associations, as patients with HIV were more likely to receive IMRT. Instead, we found that IMRT use was more influenced by regional and provider characteristics than by patient characteristics. In addition to provider experience with IMRT, we found that IMRT was used more in the West, in metropolitan areas, and at freestanding centers, which other authors have also found in breast or head and neck cancers (26, 42, 44).

Our study has limitations, many of which are associated with the nature of administrative claims data (28, 45–47). Claims data are designed to support financial transactions rather than to convey clinical information, and they do not reflect many aspects of cancer management including patient preferences. However, the large sample size, long follow-up, and real-world settings make these types of databases appealing for studying patterns of care and effectiveness of IMRT use throughout the United States. Anal cancer is also an ideal cancer to study through these large databases given the relative uniformity and stability of the treatment regimen over the past 4 decades. Although we were not able to evaluate the impact of IMRT on toxicities not requiring hospitalizations, hospitalization serves as a proxy for serious toxicities and is relevant given its implications on health care costs. Furthermore, retrospective evaluation of toxicity is challenging, and hospitalization outcomes are less subject to misclassification in claims data. Finally, our study cohort consisted of Medicare-eligible patients in the SEER database and may not be generalizable to all patients with anal SCC.

In conclusion, although much progress has been made in the curative treatment of anal cancer, there continues to be significant treatment-associated morbidity. We found that IMRT reduces hospitalizations in patients undergoing chemoradiation therapy for anal cancer. Further work is needed to understand the long-term health and cost impact of this treatment, particularly for those patient subgroups most at risk of toxicity and hospitalization. Determining the total cost of care related to the use of IMRT in the real-world setting would help to characterize the value of this advanced, resource-intensive radiation treatment modality and to potentially develop episode-based bundled payment models for anal SCC. Additional advances in radiation therapy may lead to further reductions in toxicity as well as entirely new treatment paradigms (48).