Disease-related cost of care and survival among Medicare-enrolled patients with myelodysplastic syndromes

Abstract

Background

While newer treatments for myelodysplastic syndromes (MDS), particularly hypomethylating agents (HMAs), are expensive, it is unclear whether MDS-related cost of care is associated with overall survival. We evaluated the relationship between MDS-related cost and survival among Medicare beneficiaries with MDS.

Methods

Eligible patients were identified from the SEER-Medicare database using International Classification of Disease for Oncology, 3^rd edition, codes for MDS. The patients were diagnosed between 1/1/2005 and 12/31/2011, aged ≥66 years, and followed through death or end of study (12/31/2012). Medicare payments were used to estimate costs. Cumulative costs in a propensity-score-matched group of cancer-free Medicare beneficiaries were subtracted from costs in the MDS cohort in each registry to estimate MDS-related costs. Hazard ratios (HRs) and 95% confidence intervals (CI) were derived from multivariate Cox proportional hazards models adjusting for patient and disease characteristics.

Results

Of the 8,580 eligible patients, 15.7% received HMAs. The overall two-year survival was 48.7%, and the median two-year registry-specific MDS-related cost per patient was $63,291 (range: $40,793-$78,156 across 16 registries). The two-year MDS-related cost was not associated with survival in the overall study population (1^st tertile: reference; 2^nd tertile, HR=0.96, 95% CI: 0.89–1.04, p=0.29; 3^rd tertile, HR=0.98, 95% CI: 0.91–1.06, p=0.64), or subgroups of patients who did or did not receive HMAs.

Conclusion

Medicare expenditure for elderly patients with MDS varied across registries but was not associated with survival. Lack of association between cost and outcome warrants additional research as it may help identify potential areas for cost-saving interventions without compromising patient outcomes.

Introduction

With the advent of effective but expensive therapies and longer survival of patients, costs of care for cancer patients continue to rise and contribute significantly to the rapidly increasing costs of healthcare in the United States (US). Substantial geographic variations in healthcare spending have been documented in the US,¹ including regional variations in costs of care for several types of solid malignancies such as breast, colon, lung, prostate and bladder cancers^2–6. Importantly, geographic variation in spending was not associated with outcomes of patients with solid tumors, and in particular, overall survival (survival).^2–5 Little is known about regional variation in the cost of treating patients with hematologic malignancies, including myelodysplastic syndromes (MDS), and whether the cost of care for these patients correlates with survival.

Care costs for MDS patients are high because the drugs (e.g., azacitidine, lenalidomide, decitabine) and supportive care (e.g., transfusions, hematopoietic growth factors, iron chelators) are both costly.^7–9 One study reported a higher cost of care for MDS patients during the first five years after diagnosis than reported comparable costs for the 18 most prevalent types of cancer at the time.^7–9 It is therefore important to understand any possible regional variation in the cost of care for MDS patients and the correlation with survival of patients, if any.

Since 80% of MDS patients are ≥65 years at the time of diagnosis,^{10, 11} Medicare claims-based analyses can capture the majority of patients. The Surveillance, Epidemiology, and End Results (SEER)-Medicare linked database is frequently used to study population-level outcomes including cost and survival analyses in cancer patients.^{6, 12} By leveraging the SEER-Medicare database, we assembled a large, population-based cohort of Medicare-enrollment beneficiaries with MDS to assess regional variation in the cost of care and to study the relationship between MDS-related cost of care and survival. Given the existing literature on the lack of survival improvement in geographic areas with higher spending for cancer care,^2–5 we hypothesize that there is no association between the cost of care for MDS patients and their survival.

Methods

Data Source

To assemble the study population, we used the SEER-Medicare database, a longitudinal cohort containing information on cancer diagnosis and sociodemographic characteristics linked to Medicare claims including healthcare utilization and cost information for cancer patients.⁹ SEER program consists of population-based cancer registries in 17 geographic areas and covers approximately 28% of the US population. Among individuals included in SEER files who were ≥65 years, 93% were matched with Medicare enrollment file.¹³ In addition to Medicare claims for individuals with cancer diagnoses reported to SEER, the database also includes a file that identifies a 5% random sample of Medicare beneficiaries without cancer residing in SEER areas. These non-cancer beneficiaries were incorporated into our study to estimate MDS-related costs.

Identification of Patients

We assembled a cohort of MDS patients who did not have other cancer and fulfilled the following eligibility criteria: (1) International Classification of Disease for Oncology, 3^rd edition (ICD-O-3) codes: 9980 for refractory anemia (RA), 9982 for RA with ring sideroblasts (RARS), 9983 for RA with excess blasts (RAEB), 9985 for refractory cytopenia with multilineage dysplasia (RCMD), 9986 for MDS associated with chromosome 5q deletion (del5q-MDS), 9987 for therapy-related MDS (t-MDS), 9989 for not-otherwise-specified MDS (MDS-NOS), 9991 for refractory neutropenia, or 9992 for refractory thrombocytopenia^{14, 15} diagnosed between 1/1/2005 and 12/31/2011; (2) aged ≥66 years at diagnosis; and (3) not identified exclusively through death certificates or autopsy. To ensure completeness of claims data, we excluded patients not continuously enrolled in Medicare Parts A and B, as well as patients who were enrolled in a health maintenance organization any time from 12 months before MDS diagnosis through date of death or end of 2012, whichever was earlier. We limited patients to those diagnosed at ≥ 66 years to obtain Medicare claims for 12 months before MDS diagnosis. We did not limit to beneficiaries with Medicare Part D (prescription drug) coverage because this would significantly reduce the sample size. The final study cohort included 8,580 MDS patients (Figure 1).

Figure 1.

Process of patient selection

Control Selection with Propensity Score Matching

Control subjects were selected from the 5% random sample of Medicare beneficiaries without cancer. A “pseudo-date of diagnosis” was assigned to each control to ensure that the MDS patients and controls had a comparable time period under study. Controls fulfilled the same inclusion/exclusion criteria as MDS patients did but had no history of cancer based on SEER records. To increase the comparability of MDS patients and controls, we utilized propensity score matching, which minimizes the impact of selection bias in observational studies¹⁶. For each registry, a propensity score was computed using a logistic regression model, with the outcome being disease status (MDS or not), and covariates including year of diagnosis, sex, age group (66–74, 75–79, 80–84, or ≥85 years), comorbidities (0, 1, and ≥2), and quartile of pre-diagnosis cost of care (see “construction of variables” below). Each patient in the MDS cohort was matched with the closest control using a 1:1 “nearest neighbor matching” technique, with a caliper of 0.2×standard deviation of the estimated propensity score¹⁷. The effectiveness of propensity score matching was assessed by comparing the prematch and postmatch balance of identified covariates.¹⁸ A standardized difference between cases and controls (mean difference expressed as a percentage of the average standard deviation of the variable’s distribution across the MDS and control groups) of <10% was considered indicative of good balance, which was accomplished with the current matching.

Construction of Variables

Both survival and cost of care were examined for up to two years after diagnosis. We chose this period because the median survival in older MDS patients was about two years in this analysis, and two-year survival was an important endpoint assessed in several clinical trials involving newer MDS therapies.^19–22 Survival of MDS patients was defined as the duration between the date of diagnosis and the date of death due to any cause, end of study (12/31/2012), or two years after diagnosis, whichever was earlier.

To measure cost of care, we used inpatient, outpatient and carrier claims to tally Medicare payments, which are considered a good proxy for true economic costs²³. Costs were adjusted to 2012 US dollars using the Consumer Price Indices issued by the Bureau of Labor Statistics. The two-year cumulative cost was calculated following the method of Zhao et al.²⁴ Monthly costs were computed for each subject for each month post diagnosis and then integrated over patients’ cost histories to correct for censoring. Observations were weighted by the inverse probability of being censored. In order to estimate costs attributable to MDS, costs in matched controls were subtracted from the costs in the cancer cohort.²⁵ The adjusted MDS-related costs were summed at the registry level. As Rural Georgia had only 14 patients, we combined Rural Georgia with Greater Georgia, resulting in a total of 16 registries for analysis. Two-year MDS-related costs weighted by the number of patients in each registry were categorized into tertiles.

MDS subtypes were assessed using the ICD-O-3 codes. To derive pre-diagnosis costs, we aggregated all Medicare payments for claims made within the 12 months before the date of diagnosis, and then categorized this into quartiles according to the distribution among MDS patients. To measure comorbidities, all inpatient, outpatient and carrier claims within 12 months before the date of diagnosis were identified to calculate a modified Elixhauser comorbidity score.^{26, 27} We also used a method developed by Davidoff et al. to calculate each patient’s disability status as a proxy for performance status.^{28, 29} Median household income at the zip code level was categorized into tertiles.

Statistical Analyses

Frequencies and percentages were used to describe various characteristics of the study population. Pearson’s correlation coefficients were calculated between two-year registry-specific MDS-related cost and variables of interest. The Kaplan–Meier approach was utilized to describe the probability of survival at various time points, stratified by cost group. Log-rank test was employed to compare survival across groups. Because the cost of care is usually correlated with the duration a patient survived and we do not have a reliable measure of disease severity for all patients, it is questionable to include individual cost in a survival model that evaluates the impact of individual-level characteristics on the hazard of death. We therefore performed cost-survival analysis using cost at the registry level. Cox proportional hazards models were used to evaluate the association between two-year registry-specific MDS-related costs (in tertiles) and survival, adjusting for age at diagnosis, sex, race (white vs. non-white), comorbidities, disability status, pre-diagnosis cost, and median household income at the zip code level. Specifically, hazard ratios (HRs) and 95% confidence intervals (CIs) were derived from the Cox models. In addition to the overall study cohort (n=8,580), we conducted the cost-survival analysis in two subcohorts: patients who received hypomethylating agents (HMAs, azacitidine or decitabine) during follow-up (HMA users, n=1,267), and patients who did not receive HMAs (HMA non-users, n=7,313). All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC), and all tests were two-sided with an alpha of 0.05.

Results

Of the 8,580 patients, the median age was 80 (interquartile: 74–85) years, and the majority were white (86.7%) and male (53.0%) (Table 1). As for histological subtype at the time of diagnosis, 1,006 (11.7%) patients had RA, 653 (7.6%) had RARS, 1055 (12.3%) had RAEB, and 237 (2.8%) had del5q-MDS. Since only 16 patients had refractory neutropenia or refractory thrombocytopenia, two newly added MDS subtypes,¹⁵ we combined these with RA into a single group refractory cytopenias (RC). Consistent with prior Medicare analyses of MDS, a large percentage of patients (58.5%) were classified as MDS-NOS. MDS diagnoses were distributed almost evenly across the seven years of study (2005–2011). Only 17.8% of patients reported no comorbidity. Compared with non-users, HMA users were more likely to be younger, white, male, reported as RAEB, to have more comorbid conditions and higher cost before diagnosis, and to reside in neighborhoods with higher median household income and SEER registries with higher MDS-related cost (Table 1).

Table 1.

Characteristics of the Patients Diagnosed with Myelodysplastic Syndromes (2005–2011)

			Receipt of HMAs within 2-year after diagnosis
	Overall (n=8,580)		Ever (n=1,267)		Never (n=7,313)		p
	n	%	n	%	n	%
Age at diagnosis (in years)
66 – 74	2190	25.5	429	33.9	1761	24.1	<0.01
75 – 79	1890	22.0	362	28.6	1528	20.9
80 – 84	2149	25.1	311	24.6	1838	25.1
≥85	2351	27.4	165	13.0	2186	29.9
Race
White	7441	86.7	1124	88.7	6317	86.4	<0.01
Other	1139	13.3	143	11.3	996	13.6
Sex
Female	4031	47.0	473	37.3	3558	48.7	<0.01
Male	4549	53.0	794	62.7	3755	51.4
Histological subtype (ICD-O-3)
RC (9980, 9991, 9992)	1022	11.9	67	5.3	955	13.1	<0.01
RARS (9982)	653	7.6	62	4.9	591	8.1
RAEB (9983)	1055	12.3	342	27.0	713	9.8
RCMD (9985)	563	6.6	102	8.1	461	6.3
Del-5q (9986)	237	2.8	28	2.2	209	2.9
t-MDS (9987)	34	0.4	8	0.6	26	0.4
MDS, NOS (9989)	5016	58.5	658	51.9	4358	59.6
Year of diagnosis
2005	1134	13.2	138	10.9	996	13.6	0.04
2006	1292	15.1	197	15.6	1095	15.0
2007	1280	14.9	178	14.1	1102	15.1
2008	1182	13.8	194	15.3	988	13.5
2009	1234	14.4	183	14.4	1051	14.4
2010	1216	14.3	171	13.5	1045	14.3
2011	1242	14.5	206	16.3	1036	14.2
Elixhauser score
0	1528	17.8	282	22.3	1246	17.0	<0.01
1 – 2	3121	36.4	530	41.8	2591	35.4
3+	3931	45.8	455	35.9	3476	47.5
Disability status score
1st quartile (low)	2143	25.0	433	34.2	1710	23.4	<0.01
2nd quartile	2145	25.0	376	29.7	1769	24.2
3rd quartile	2146	25.0	304	24.0	1842	25.2
4th quartile (high)	2146	25.0	154	12.2	1992	27.2
Pre-diagnosis cost
1st quartile (low)	2145	25.0	359	28.3	1786	24.4	<0.01
2nd quartile	2145	25.0	350	27.6	1795	24.6
3rd quartile	2145	25.0	326	25.7	1819	24.9
4th quartile (high)	2145	25.0	232	18.3	1913	26.2
Zip code level median household income
1st tertile (low)	2686	31.3	364	28.7	2322	31.8	<0.01
2nd tertile	2686	31.3	399	31.5	2287	31.3
3rd tertile (high)	2686	31.3	418	33.0	2268	31.0
Unknown	522	6.1	86	6.8	436	6.0
Two-year MDS-related registry-specific cost
Low	2802	32.7	411	32.4	2391	32.7	<0.01
Medium	2851	33.2	345	27.2	2506	34.3
High	2927	34.1	511	40.3	2416	33.0

Abbreviations: HMAs, hypomethylating agents; ICD-O-3, International Classification of Diseases for Oncology, 3^rd edition; RC, refractory cytopenia; RARS, RA with ring sideroblasts; RAEB, RA with excess blasts; RCMD, refractory cytopenia with multilineage dysplasia; del5q-MDS, myelodysplastic syndrome associated with chromosome 5q deletion; t-MDS, therapy-related myelodysplastic syndrome; and MDS-NOS, not-otherwise-specified myelodysplastic syndrome.

By the end of two-year follow-up, 4,403 patients (51.3%) had died. Median survival of the entire cohort was 1.84 years. Two-year survival rates for HMA users and non-users were 37.4% and 49.4%, respectively. Across histologic subtypes, patients with RAEB had the lowest two-year survival rate (21.8%, median survival: 0.80 years).

There was a substantial variation in two-year MDS-related costs for Medicare-enrolled patients across the 16 registries (Figure 2). Two-year registry-specific MDS-related costs ranged from $40,793 in New Mexico to $78,156 in Detroit, Michigan. HMA users had higher cost than non-users (Figure 3). Except for New Mexico ($50,418), the two-year MDS-related costs for RABE patients were nearly $100,000 or higher, ranging from $103,621 (Kentucky) to $148,943 (Michigan). However, two-year MDS-related cost was not significantly correlated with percentage of HMAs users (r=0.42, p=0.11) or that of RAEB patients in the registry (r=0.14, p=0.60).

Figure 2.

Two-year MDS-related cost (in 2012 US dollars), percentages of patients who received hypomethylating agents, and percentage of RAEB patients by registry.

Figure 3.

Two-year MDS-related cost (in 2012 US $) by receipt of hypomethylating agents and registry

There was no correlation between the two-year registry-specific MDS-related cost and survival (r=−0.28, p=0.29) (Figure 4a). A log-rank test comparing Kaplan-Meier curves across the three levels of registry-specific cost did not suggest a difference, either (p=0.68; Figure 4b). Using a Cox proportional hazards model that adjusted for multiple patient characteristics and disease subtype, we also found that the 2-year MDS-related cost was not associated with survival (1^st tertile: reference; 2^nd tertile, HR=0.96, 95% CI: 0.89–1.04, p=0.29; 3^rd tertile, HR= 0.98, 95% CI: 0.91–1.06, p=0.64) (Table 2). Among the characteristics included in the multivariate Cox model, female gender, non-white race, younger age at diagnosis, RA and RARS subtypes, lower pre-diagnosis health cost, lower Elixhauser comorbidity score, and lower disability status score were associated with longer survival. Similarly, no association was observed between 2-year MDS-related cost and survival for either HMA users or non-users (Table 2).

Figure 4A.

A scatter plot of two-year MDS-related cost (in 2012 US dollars) and two-year survival by registry.

Figure 4B.

Kaplan-Meier survival curves by two-year registry-specific MDS-related cost among 8,580 elderly patients with myelodysplastic syndromes

Table 2.

Hazard Ratios and 95% Confidence Intervals for the Association between Different Factors and Survival among 8,580 Elderly Patients with Myelodysplastic Syndromes

	Overall (n = 8,580)		HMA users (n = 1,267)		HMA non-users (n = 7,313)
	HR (95% CI)	p	HR (95% CI)	p	HR (95% CI)	p
Two-year registry-specific MDS-related cost
1st tertile (low)	1.00		1.00		1.00
2nd tertile	0.96 (1.89–1.04)	0.29	1.10 (0.91–1.33)	0.33	0.93 (0.86–1.01)	0.10
3rd tertile (high)	0.98 (0.91–1.06)	0.64	1.06 (0.88–1.27)	0.55	0.97 (0.89–1.06)	0.53
Hypomethylating agents
Never	1.00
Ever	1.07 (0.99–1.17)	0.09
Age at diagnosis (in years)
66 – 74	1.00		1.00		1.00
75 – 79	1.10 (1.00–1.21)	0.04	1.10 (0.91–1.32)	0.34	1.11 (0.99–1.23)	0.07
80 – 84	1.38 (1.27–1.51)	<.01	1.23 (1.02–1.49)	0.03	1.44 (1.30–1.59)	<.01
≥85	1.69 (1.55–1.84)	<.01	1.58 (1.26–1.97)	<.01	1.72 (1.56–1.89)	<.01
Race
White	1.00		1.00		1.00
Other	0.88 (0.81–0.97)	<.01	1.06 (0.85–1.33)	0.61	0.87 (0.78–0.96)	<.01
Sex
Female	1.00		1.00		1.00
Male	1.31 (1.23–1.39)	<.01	1.04 (0.89–1.20)	0.65	1.35 (1.27–1.45)	<.01
Histological subtype (ICD-O-3)
RC (9980, 9991, 9992)	1.00		1.00		1.00
RARS (9982)	0.78 (0.65–0.93)	<.01	0.75 (0.43–1.32)	0.32	0.77 (0.64–0.93)	<.01
RAEB (9983)	3.49 (3.07–3.97)	<.01	2.23 (1.51–3.29)	<.01	3.92 (3.41–4.50)	<.01
RCMD (9985)	1.47 (1.26–1.73)	<.01	1.30 (0.83–2.04)	0.25	1.44 (1.21–1.71)	<.01
Del-5q (9986)	1.31 (1.05–1.63)	0.02	1.35 (0.73–2.50)	0.34	1.25 (0.98–1.59)	0.07
t-MDS (9987)	2.18 (1.42–3.36)	<.01	2.27 (0.93–5.54)	0.07	1.99 (1.20–3.28)	<.01
MDS, NOS (9989)	1.72 (1.54–1.92)	<.01	1.54 (1.05–2.25)	0.03	1.71 (1.52–1.92)	<.01
Elixhauser score
0	1.00		1.00		1.00
1 – 2	0.99 (0.90–1.09)	0.89	1.01 (0.83–1.24)	0.90	1.01 (0.91–1.13)	0.81
3+	1.26 (1.14–1.41)	<.01	1.43 (1.13–1.81)	<.01	1.26 (1.11–1.42)	<.01
Disability status score
1st quartile (low)	1.00		1.00		1.00
2nd quartile	1.14 (1.04–1.25)	<.01	0.99 (0.83–1.19)	0.95	1.21 (1.08–1.34)	<.01
3rd quartile	1.39 (1.27–1.52)	<.01	1.07 (0.88–1.30)	0.48	1.50 (1.35–1.66)	<.01
4th quartile (high)	1.71 (1.56–1.88)	<.01	1.13 (0.88–1.44)	0.33	1.83 (1.65–2.03)	<.01
Pre-diagnosis cost
1st quartile (low)	1.00		1.00		1.00
2nd quartile	0.91 (0.82–1.00)	0.04	0.84 (0.69–1.03)	0.10	0.92 (0.82–1.02)	0.12
3rd quartile	0.98 (0.88–1.08)	0.62	0.80 (0.64–1.00)	0.05	1.02 (0.91–1.14)	0.78
4th quartile (high)	1.22 (1.10–1.36)	<.01	0.76 (0.59–0.98)	0.03	1.31 (1.16–1.47)	<.01
Zip code level median household income
1st tertile (low)	1.00		1.00		1.00
2nd tertile	1.04 (0.97–1.12)	0.29	0.94 (0.78–1.14)	0.54	1.07 (0.98–1.16)	0.13
3rd tertile (high)	0.97 (0.90–1.06)	0.52	1.05 (0.86–1.28)	0.62	0.96 (0.88–1.05)	0.37
Unknown	0.94 (0.80–1.08)	0.37	0.72 (0.52–1.01)	0.06	0.97 (0.84–1.12)	0.67

Abbreviations: HMA, hypomethylating agent; HR, hazard ratio; 95% CI, 95% confidence interval; ICD-O-3, International Classification of Diseases for Oncology, 3rd edition; RC, refractory cytopenia; RARS, RA with ring sideroblasts; RAEB, RA with excess blasts; RCMD, refractory cytopenia with multilineage dysplasia; del5q-MDS, myelodysplastic syndrome associated with chromosome 5q deletion; t-MDS, therapy-related myelodysplastic syndrome; and MDS-NOS, not-otherwise-specified myelodysplastic syndrome.

Discussion

Here we report the first large study on regional variations in MDS-related cost and evaluate the association between cost of care and survival in MDS patients using the SEER-Medicare data. Our study shows that MDS-related cost as reflected by Medicare expenditure in older adults varied substantially across registries in the US during the two years post-diagnosis. Interestingly, higher expenditures on MDS treatment did not translate into better survival even after adjusting for patient and disease characteristics.

Direct costs of care for patients with MDS include but are not limited to the costs of blood product transfusions, HMAs, lenalidomide, erythropoiesis-stimulating agents, iron chelators, laboratory tests, hospitalizations, clinic and emergency room visits for management of infections, bleeding and other complications, and, for a minority of patients, the costs of chemotherapy and stem cell transplantation. This study confirms that the management of MDS is costly in the era of wider use of MDS-directed therapies. A study before the approval of HMAs found that transfusion dependence among MDS patients was associated with an incremental cost of $31,255 per patient annually compared to transfusion-independent MDS patients.⁸ Using Medicare reimbursement schedule for 2007 to evaluate the costs of National Cancer Comprehensive Cancer Network-recommended drugs for MDS, Greenberg et al. estimated the annual costs for erythropoiesis-stimulating agents at $26,076 to $87,300, azacitidine at $55,332 and decitabine at $74,160.⁹ In a Medicare-based study, patients who were diagnosed after the approval of HMAs and received HMAs had a significantly higher 2-year MDS-related costs compared to patients diagnosed prior to the approval of HMAs and did not receive HMAs (97,977 vs. 42.628 in 2009US$).⁷

The geographic variation in MDS-related Medicare expenditures may be affected by many factors. We observed higher MDS-related -cost in registries where more patients had higher pre-diagnosis cost and more comorbidities, which may be due to more physician visits and diagnostic testing. The fact that the higher-spending registries did not experience better survival may suggest that patients with more comorbidities as reflected in higher pre-diagnosis cost or comorbidity score were less likely to benefit from MDS treatment.

This study contributes to a growing literature suggesting that geographic variation in spending is not associated with outcomes, and in particular, cancer survival.^3–6 Skolarus et al. even observed superior bladder cancer-specific survival among patients in lower-spending hospital referral regions (HRR) than those in the higher-spending HRRs (HR = 0.83; 95% CI: 0 .71–0.97).⁵ Two other studies adopted average HRR-level Medicare inpatient spending among beneficiaries in the last 6 months of life as the proxy of area-level healthcare spending.^3,4 Landrum et al. did not observe any association between end-of-life inpatient expenditure and 3-year mortality at the HRR level among colon cancer patients.⁴ Other researchers found an inverse association between area-level spending and one-year mortality among lung cancer patients in the Medicare cohort, while no association was found among lung cancer patients in the Veterans Health Administration cohort or colon cancer patients from either population.³ One potential explanation is that higher-spending regions were more likely to use recommended care but at the same time, were also more likely to use discretionary and non-recommended care.^3,4

Among MDS patients, a prior Medicare-based study found a significantly better survival over a period of two years (HR = 0.80, 95% CI: 0.70–0.93) and higher MDS-related costs in patients diagnosed between July 2004 and December 2007 who received HMAs, compared to patients diagnosed prior to the approval of HMAs (between January 2001 and June 2004) who did not receive HMAs and had lower MDS-related costs.⁷

Our study, like any other observational study, has limitations that should be taken into consideration. Our study population was restricted to Medicare-eligible patients, so caution should be taken when generalizing to the broader MDS population. Medicare claims data are comprised of administrative records submitted by providers for the purpose of obtaining reimbursement for services provided and are limited to the services covered by Medicare. A large number of patients were reported as MDS-NOS in SEER, and important clinical variables such as karyotypic abnormalities, blast count, and blood counts are not available in the SEER-Medicare database. Although we adjusted for histologic subtypes of MDS in the Cox model, we were limited by the fact that 58.5% of the patients had the uninformative category of MDS-NOS. Therefore, we could not fully control for severity of disease. Additionally, other significant costs such as the costs of oral drugs prescribed through Part D (e.g., lenalidomide), costs of travel, lost income, and out-of-pocket costs were not accounted for in our current analysis. Therefore, we probably have underestimated the overall cost of MDS. Furthermore, we were unable to separate the explicit impact of the cost of end-of-life care, although we used a rigorous approach that is designed to account for censoring and various follow-up times.²³ Finally, the choices of therapy in these patients are also subject to the discretion of clinicians and/or patients that can’t be directly observed.

In conclusion, there was substantial geographic variation in Medicare expenditure on the treatment of older adults with MDS. Our observation that a higher expenditure was not associated with better observed or risk-adjusted survival at the registry level calls for future research on the different elements/modalities contributing to the overall cost of treatment for older adults with MDS, and the effectiveness of different modalities given specific patient characteristics. The underlying reasons behind a possible disconnection between cost and outcome warrant additional research.