Linking the Ohio Cancer Incidence Surveillance System with Medicare, Medicaid, and Clinical Data from Home Health Care and Long Term Care Assessment Instruments: Paving the Way for New Research Endeavors in Geriatric Oncology

Abstract

This study describes the Ohio Cancer-Aging Linked Database, which mirrors in structure the linked Surveillance, Epidemiology, and End-Results (SEER)-Medicare files, but also incorporates data from Medicaid enrollment and claims files, the home health care (HHC) Outcome Assessment Information Set (OASIS), and the long term care (LTC) Minimum Data Set (MDS). This article also discusses the potential uses of this database, particularly in addressing new research questions emerging from the nascent and rapidly developing field of geriatric oncology.

Introduction

Elders carry a disproportionate burden of cancer. Data from the Surveillance, Epidemiology, and End Results (SEER) indicate that 56% of all incident cases of cancer and 71% of cancer deaths occurred in patients 65 years of age or older during the period 1998–2002.¹ The number of elders in the United States will double to nearly 70 million by 2030,¹ reflecting a growing burden of cancer to the society in the years ahead.

The linked SEER-Medicare files have served as the basis to a vast body of cancer-related health services research (HSR) literature in the past 15 years, documenting patterns of cancer care and outcomes and disparities thereof in subgroups of the Medicare population. However, an increasingly more encompassing approach to the management of cancer in elders warrants the development of new research methodologies and/or the refinement of existing ones, and an adapted version of the existing SEER-Medicare files to meet emerging needs in geriatric cancer-HSR.

Receipt of various treatment modalities and relevant outcomes in elders are often influenced and confounded by other diseases, disabilities, symptoms, and/or other vulnerabilities—such as low socioeconomic status (SES),^2,3 and inadequate social support^4,5—that are superimposed on the burden of cancer. The recent body of literature discussing the management of cancer in old age emphasizes the need to characterize the “functional age” of older patients,⁶ and focuses increasingly on geriatric assessment.^7,8,9,10,11 Hence, the development of the Comprehensive Geriatric Assessment (CGA), an instrument that incorporates principles of geriatrics in the practice of oncology. The CGA measures several domains, including, but not encompassing, comorbidities, functional limitations (FL), and geriatric syndromes (GS).

While the value of this more encompassing approach to the management of cancer in elders is gradually gaining recognition in clinical practice, population-based research in this area has remained relatively stagnant. This is due, in great part, to the limitations of the SEER-Medicare files—the most widely used database to study cancer-related outcomes in elders.

State-level cancer registry or surveillance data have been previously linked to various administrative databases, including Medicare data,^12,13 Medicaid data,^{14,15,16,17,18} hospital discharge data,¹⁹ and primary medical record data.²⁰ To date, however, cancer registry data linked with both Medicare and Medicaid data are available only from the state of Michigan²¹—a database developed with the distinct aim to study cancer-related disparities, especially by dual Medicare-Medicaid eligibility status. The Ohio Cancer-Aging Linked Database (CALD), described below in detail, further expands on the structure of the Michigan database by appending data from clinical assessment instruments in anticipation of the emerging needs in geriatric cancer-HSR.

The objectives of this study are twofold: (1) to describe the CALD in detail; and (2) to discuss the potential uses and limitations of the CALD in addressing various epidemiologic and health services research questions in the realm of cancer and aging.

Methods

Overview

The CALD was developed by linking records from the Ohio Cancer Incidence Surveillance System (OCISS), with that of Medicare and Medicaid enrollment and claims files, data from the US Census, and Ohio death certificate files. In addition, OCISS records were linked with assessments from the home health care Outcomes Assessment Information Set (OASIS), and the long term care Minimum Data Set (MDS) (see Figure 1). Data linkage was accomplished by using patient identifiers, as detailed below.

Figure 1.

The current version of the CALD includes data for Ohio residents 65 years of age or older, and diagnosed with incident breast, prostate, or colorectal cancer in the years 1997–2001. The CALD incorporates Medicare and Medicaid enrollment and claims files for the time period 1996–2002 to allow for a minimum of 12 months follow-up period before or after the initial date of cancer diagnosis for all study subjects. MDS and OASIS data were not made available to researchers until July 1998 and July 1999 respectively; therefore the CALD lacks relevant data for the period prior to these dates.

The CALD was developed after obtaining approval from all relevant agencies, including the Institutional Review Board (IRB), University Hospitals of Cleveland. Thus, use of the OCISS was approved by the IRB of the Ohio Department of Health, which maintains the cancer surveillance system. Use of Ohio Medicaid data for the purposes of CALD-based studies was approved through a Data Users Agreement (DUA) with the Ohio Department of Job and Family Services, which administers the Medicaid program. Medicare enrollment and claims files, the OASIS, and the MDS were obtained from the Centers for Medicare and Medicaid Services (CMS). The studies were approved through relevant DUAs, as well as by the CMS privacy board.

Data Sources

The Ohio Cancer Incidence Surveillance System (OCISS)

Established in 1991, the OCISS is administered by the Ohio Department of Health (ODH). It includes data reported by all hospitals, as well as clinics and private physician offices. With the exception of non-melanoma skin cancers, and carcinoma in situ of the cervix, all primary cancers diagnosed on or after January 1, 1992, are required to be reported to the OCISS.

The OCISS contains patient identifiers, including first name, last name, social security number (SSN), date of birth, gender, race, county, address of residence, and zip code; as well as cancer-specific data, including date of diagnosis, anatomical site, stage at diagnosis, and tumor grade.

The completeness of reporting in the OCISS for year 2000 at the state level for all cancer sites and types combined has been estimated at 92%.²²

Medicare files

Medicare claims files include the Medicare Provider Analysis and Review (MEDPAR) for inpatient admissions, the Outpatient Standard Analytic Files (SAF) for all outpatient hospital encounters and ambulatory surgical centers, the Physician Supplier files for services received in non-hospital outpatient settings and through non-institutional providers, as well as data from the Durable Medical Equipment (DME) SAF, the Home Health Agency SAF, and hospice SAF. The Medicare Denominator file provides pertinent information on the individuals’ Medicare enrollment history, and their participation in managed care programs.

The MEDPAR data include up to 10 diagnosis codes and 10 procedure codes according to International Classification of Diseases, 9^th Clinical Modification (ICD-9-CM), and billing data. The Medicare outpatient claim record includes up to 10 diagnosis codes according to ICD-9-CM, and one procedure code according to Current Procedural Terminology, 4^th edition (CPT-4) at the line item level, and billing data. Records from the Physician/Supplier file, and the Durable Medical Equipment (DME) file include up to 8 diagnosis codes according to ICD-9-CM, and one procedure code according to CPT-4 or Healthcare Common Procedure coding System (HCPCS) at the line item level, and billing data. Similarly, the home health agency SAF and the hospice SAF include up to 10 diagnosis codes, and one procedure code according to CPT-4 or at the line item level.

The Ohio Medicaid Policy Database (MPD)

The MPD includes all Medicaid enrollment and claims files. The enrollment files include patient identifiers, as well as data on individuals’ enrollment history on a monthly basis, including the eligibility category, enrollment in Medicare Part A or Part B, and participation in the spend-down program. A unique identifier enables users to retrieve individual-specific eligibility and claims data across time and across service categories. Medicaid claims files include claim records for services received by Medicaid beneficiaries, and paid for by the Medicaid program. All claims for services received through the traditional fee-for-service system (FFS) are present in the files, including inpatient care, outpatient hospital and non-institutional care, nursing home, and pharmacy. Encounter level data may be incomplete for individuals participating in the spend-down program, and for those who are dually eligible for the Medicaid and Medicare program—hence the importance of linking Medicare and Medicaid data for dually eligible individuals in order to obtain a more complete assessment of their care trajectory.

Institutional claims (eg, inpatient hospital, outpatient hospital) include up to 5 diagnosis codes, and up to 3 procedure codes, according to ICD-9-CM. The line items for these claims carry procedure codes (one per line item) according to CPT-4, and billing data. The non-institutional outpatient claims include up to 2 diagnosis codes according to ICD-9-CM, and the line items include one procedure code according to CPT-4, and billing data.

The Home Health Outcome Assessment Information System (OASIS)

The OASIS is a repository of mandatory comprehensive assessments gathered by clinical staff providing home health services that are reimbursed by Medicare and Medicaid. The assessments are completed upon admission to and discharge from HHC, on a 60-day interval if the patient is not discharged from HHC, as well as upon transfer to other care settings or death.

The OASIS record includes patient identifiers, including SSN, date of birth, and gender, as well as a wealth of clinical and functional assessment data. Of particular relevance are diagnosis codes retrieved from the patient’s medical charts; Activities of Daily Living (ADL) and Instrumental Activities of Daily Living (IADL); and a number of variables providing a detailed assessment of the patient’s behavioral and cognitive status. It also includes date of assessment; a variable indicating whether a given assessment was completed upon admission (or “Start of Care Assessment”), follow-up care, or discharge; as well as variables describing the patient’s functional status in the 14 days prior to the date of assessment, and variables describing the patient’s current functional status.

Because the OASIS assessment is required for home health care to be reimbursed by Medicare, the completeness of the OASIS databases is believed to be very high, and depending on the item examined, inter-rater reliability has been deemed to be highly adequate,²³ substantial, or excellent.²⁴

The Long Term Care Minimum Data Set (MDS)

The MDS is a standardized primary screening and assessment tool of health status used to collect data on all residents of long-term care facilities that are certified to participate in Medicare or Medicaid.²⁵ Assessments must be completed upon admission, quarterly, and/or when a significant change in health status occurs. Additionally, a full annual assessment must be completed within 12 months of the most recent full assessment. MDS records include data encompassing a variety of domains.²⁶ Of particular relevance to the proposed studies is the assessment of functional and cognitive status, in addition to patient identifiers, which have been used to link with the OCISS records. MDS files have been available since mid 1998, and used to address a variety of questions in health services research.

Like the HHC assessments, the MDS assessment is also required by Medicare and Medicaid for reimbursement, hence the high rate of completeness. Studies analyzing inter-rater reliability have reported favorable results^27,28 implying that data errors may be low. On the other hand, a recent comparison of diagnosis codes between the MDS and hospital discharge data from Canada showed varying sensitivity across diagnostic categories,²⁹ suggesting under-recording of diagnostic data in the MDS.

Ohio Death Certificate files

Death certificate files include records for nearly every deceased individual who was a resident of the state. In addition to patient identifiers, including SSN, date of birth, gender, and first and last names, the death certificate carries the date of death, as well as the cause of death.

US Census data

Residence address at the time of diagnosis, as recorded in the OCISS, was used to identify the patient’s census block group by geocoding. The census block group identifier was then used to retrieve data on income and educational attainment at that geographic level.

Data linkage

The presence of patient identifiers in the OCISS files made it possible to link records across the OCISS and the Medicare/Medicaid enrollment and claims files, as well as with the OASIS and MDS files, the Ohio death certificate files, and data from the US Census.

The OCISS served as the “source file,” meaning that OCISS records were linked individually with records from each of the relevant files. Through this process, the unique patient “dummy” identifier (PATID) from the OCISS was appended to records from each of the files, thus serving as the common link variable across the CALD. Because of the presence of multiple records for cases that were for the same individual, cancer site, and date of diagnosis, data linkage followed a process of unduplication of records in the OCISS file. Further, the final OCISS denominator included unique individuals, identified by unduplicating records for the same individual across different primary cancer sites (eg, a woman diagnosed with breast and colorectal cancer in the same year).

OCISS-Medicare linkage

Following approval from the relevant agencies, the OCISS staff provided a data file including records for all elders diagnosed during years 1997–2001 with incident breast, colorectal, or prostate cancer. This file, which included patient identifiers, was then shared with CMS, which performed a search based on the SSN and gender. CMS then provided a file with Medicare identifiers (HIC) obtained through this search process. A testing of this file, also referred to as the SSN-HIC conversion file, indicated that at least 97% of patients in the OCISS were also identified in the Medicare files.

OCISS-Medicaid linkage

Consistent with previous studies,^18,30 OCISS and Medicaid records were linked using the following algorithm (with the first and last names truncated to the first 6 digits only):

• Step 1: SSN, Last Name, First Name,

• Step 2: SSN, Last Name, Date of Birth (Month), Gender

• Step 3: SSN, First Name, Date of Birth (Month), Gender

• Step 4: First Name, Last Name, Date of Birth (Month and Year), Gender

This algorithm is a slight variant of that used in linking SEER and Medicare files.³¹ Nearly 85% of patients were identified through the first step, and 8%–10% through the last step. Step 4, which is based solely on name, date of birth and gender, is a strategy designed to account for Medicaid-only patients who may not qualify for Medicare. Such a situation may present, for example, when an individual does not have a SSN, or has little or no credits in the Social Security system—a scenario likely to apply to domestic workers who may have maintained their income through payments in cash.³²

OCISS-OASIS and OCISS-MDS linkage

For those identified successfully in Medicare files, OCISS records were linked with each of the OASIS and MDS records using the combination of SSN and gender.

OCISS-Death Certificate linkage

The method used to link OCISS and death certificate records was similar to that employed in linking records across OCISS and Medicaid files. To improve the completeness of data on patient vital status, we resorted to the Medicare Denominator file as well. This strategy, leading to the identification of 486 additional decedents (or 1.4% of patients identified as deceased through 2005), made it possible to identify deaths that occurred to patients who were Ohio residents at the time they were diagnosed with cancer, but may have moved to a different state later in the course of their disease. Unfortunately, however, the cause of death will not be identified for this subgroup of deceased patients.

Study Variables

Cancer site (breast, prostate, or colon/rectum), date of cancer diagnosis, cancer stage (SEER summary stage), and demographics were retrieved from the OCISS. Age at diagnosis was categorized in 5-year increments: 65–69, 70–74, 75–79, 80–84, and 85+. Given the demographic profile of Ohio, and the small number of patients who are identified as Hispanic or ones of other racial/ethnic groups (Asian, Pacific Islander, or Native Americans/American Indians), we categorized the race variable as African American and All Other. The date of cancer diagnosis was used to stratify patients by year of cancer diagnosis. The date of cancer diagnosis was also used in conjunction with the date of assessment on the OASIS and MDS records to group patients by their timing of admission to home health care or to a nursing home, relative to their cancer diagnosis.

Analysis

SAS version 9.1 (Cary, NC) was used in linking the datasets and in all other analyses. Chi-square tests were performed to compare demographics and cancer stage between Medicare and non-Medicare groups, and between Medicaid and non-Medicaid groups.

Results

In any given year and primary cancer site, CMS was able to identify 96% to 98% of all patients identified in the OCISS file (Table 1). This match rate was consistent across years, and across cancer sites.

Table 1.

Number of Incident Cases Reported in the OCISS before and after Unduplication, Percent Matched with Medicare and Medicaid Files, and Percent Patients Deceased by December 31, 2005

Anatomic Site/ Year of diagnosis	# of cases reported in the OCISS	# of cases reported in the OCISS unduplicated by date and site*	Unique Individuals	% matched with Medicare files **	% matched with Medicaid files	% deceased by December 31, 2005
Female Breast:
1997	4,535	4,473	4,422	96.13	15.11	50.3
1998	4,826	4,742	4,677	96.69	14.33	42.7
1999	4,898	4,808	4,745	96.38	12.90	37.9
2000	4,511	4,449	4,386	97.24	11.47	34.6
2001	4,542	4,454	4,399	97.32	10.21	28.8
Prostate:
1997	4,978	4,976	4,959	97.76	6.37	46.0
1998	4,755	4,751	4,730	97.72	6.43	41.8
1999	5,485	5,480	5,448	97.72	6.09	38.5
2000	5,390	5,379	5,339	97.71	5.71	33.4
2001	5,414	5,407	5,375	98.03	5.00	27.9
Colorectal:
1997	5,202	5,109	5,068	96.17	15.21	71.4
1998	5,402	5,271	5,189	97.17	13.99	66.1
1999	5,341	5,245	5,149	96.83	13.50	61.1
2000	5,003	4,910	4,823	96.85	12.54	56.7
2001	4,931	4,832	4,772	97.84	11.25	52.3

^*Individuals with multiple records with the same date of diagnosis and the same cancer site were counted once.

^**Denominator is the number of unique individuals

The match rate with Medicaid varied considerably by cancer site, with lower match rates observed in prostate cancer patients than in breast or colorectal cancer patients. Table 1 also presents the number of decedents by cancer site and year of diagnosis. The proportion of patients deceased by the end of 2005 was 39.2%, 37.3%, and 62.3% in each of breast, prostate, and colorectal cancer patients, respectively.

Table 2 compares demographics and cancer stage data between patients who were and were not successfully matched with Medicare and Medicaid enrollment files. The results indicate that patients who are older, women (in the case of colorectal cancer), of African-American descent, those diagnosed with distant cancer stage, and those who had unstaged cancer or cancer of unknown stage are disproportionately represented in those who were not matched with the Medicare files. These differences, also reported in the SEER-Medicare files,³¹ were consistent across all three cancer sites. Among colorectal cancer patients, we also observed a disproportionately higher representation of women in the group not matched with Medicare files. Additional analyses indicated that not all these patients fell in the Medicaid-only subcategory of patients (data not shown). Patterns in demographic and cancer stage data among Medicaid beneficiaries were similar to those observed among patients in the non-Medicare group.

Table 2.

Comparison of Cases by Matching Status with Medicare and Medicaid Files

Anatomic Site	Cases matched successfully with Medicare	Cases not matched with Medicare files	p-value	Cases matched successfully with Medicaid	Cases not matched with Medicaid Files	p-value
Female Breast:
Age:
65–69	5,186 (23.9%)	158 (21.5%)		524 (18.2%)	4,820 (24.6%)
70–74	5,697 (26.2%)	167 (22.7%)		599 (20.7%)	5,265 (26.9%)
75–79	4,899 (22.6%)	162 (22.1%)		564 (19.5%)	4,497 (23.0%)
80–84	3,331 (15.3%)	123 (16.8%)		521 (18.0%)	2,933 (15.0%)
85+	2,612 (12.0%)	124 (16.9%)	0.0006	681 (23.6%)	2,055 (10.5%)	<0.0001
Race:
Af-Am	1,559 (7.2%)	84 (11.4%)		530 (18.4%)	1,113 (5.7%)
All Other	20,166 (92.8%)	650 (88.6%)	<0.0001	2,359 (81.6%)	18,457 (94.3%)	<0.0001
Stage:
In situ	2,857 (13.2%)	80 (10.9%)		265 (9.2%)	2,672 (13.6%)
Local	11,825 (54.4%)	342 (46.6%)		1,293 (44.7%)	10,874 (55.6%)
Regional	4,091 (18.8%)	147 (20.0%)		616 (21.3%)	3,622 (18.5%)
Distant	910 (4.2%)	**		184 (6.4%)	775 (4.0%)
Unstaged	1,910 (8.8%)	109 (14.9%)		511 (17.7%)	1,508 (7.7%)
Unknown	132 (0.6%)	**	<0.0001	20 (0.7%)	119 (0.6%)	<0.0001
Total number of Cases	21,725	734		2,889	19,570
Prostate:
Age:
65–69	7,168 (28.4%)	173 (30.3%)		347 (22.7%)	6,994 (28.8%)
70–74	7,812 (30.9%)	132 (23.1%)		348 (22.8%)	7,596 (31.2%)
75–79	5,667 (22.4%)	143 (25.0%)		337 (22.1%)	5,473 (22.5%)
80–84	2,881 (11.4%)	70 (12.3%)		249 (16.3%)	2,702 (11.1%)
85+	1,749 (6.9%)	53 (9.3%)	0.0011	245 (16.1%)	1,557 (6.4%)	<0.0001
Race:
Af-Am	2,686 (10.6%)	90 (15.8%)		488 (32.0%)	2,288 (9.4%)
All Other	22,591 (89.4%)	481 (84.2%)	<0.0001	1,038 (68.0%)	22,034 (90.6%)	<0.0001
Stage:
In situ	29 (0.1%)	0 (0.0%)		**	**
Local	18,081 (71.5%)	368 (64.4%)		816 (53.5%)	17,633 (72.5%)
Regional	1,518 (6.0%)	33 (5.8%)		92 (6.0%)	1,459 (6.0%)
Distant	1,133 (4.5%)	37 (6.5%)		165 (10.8%)	1,005 (4.1%)
Unstaged	4,504 (17.8%)	133 (23.3%)		448 (29.4%)	4,189 (17.2%)
Unknown	12 (0.1%)	0 (0.0%)	0.0018	**	**	<0.0001
Total number of Cases	25,277	571		1,526	24,322
Colorectal:
Age:
65–69	4,616 (19.1%)	133 (17.7%)		476 (14.3%)	4,273 (19.8%)
70–74	5,635 (23.4%)	157 (20.8%)		602 (18.2%)	5,190 (24.1%)
75–79	5,634 (23.4%)	171 (22.7%)		685 (20.7%)	5,120 (23.8%)
80–84	4,191 (17.4%)	130 (17.2%)		604 (18.2%)	3,717 (17.2%)
85+	4,040 (16.7%)	163 (21.6%)	0.0099	947 (28.6%)	3,256 (15.1%)	<0.0001
Gender
Male	11,443 (47.5%)	273 (36.2%)		1,025 (30.9%)	10,691 (49.6%)
Female	12,673 (52.5%)	481 (63.8%)	<0.0001	2,289 (69.1%)	10,865 (50.4%)	<0.0001
Race:
Af-Am	1,846 (7.6%)	86 (11.4%)		595 (17.9%)	1,337 (6.2%)
All Other	22,270 (92.4%)	668 (88.6%)	0.0002	2,719 (82.1%)	20,219 (93.8%)	<0.0001
Stage:
In situ	1,568 (6.5%)	**		169 (5.1%)	1,435 (6.7%)
Local	7,075 (29.4%)	189 (25.1%)		834 (25.2%)	6,430 (29.8%)
Regional	8,806 (36.5%)	252 (33.4%)		1,166 (35.2%)	7,892 (36.6%)
Distant	3,140 (13.0%)	129 (17.1%)		369 (11.1%)	2,900 (13.5%)
Unstaged	3,302 (13.7%)	141 (18.7%)		742 (22.4%)	2,701 (12.5%)
Unknown	225 (0.9%)	**	<0.0001	34 (1.0%)	198 (0.9%)	<0.0001
Total number of Cases	24,116	754		3,314	21,556

^**In accordance with Centers for Medicare and Medicaid Services (CMS) rules to protect patient confidentiality, cells smaller than 11 are masked.

When the masked cell could be calculated by subtracting the other numbers from the total, another (randomly selected) cell was masked as well.

For years 2000 and 2001, for which full-year data for both the OCISS and the OASIS were available, the proportion of OCISS patients also identified through the OASIS was 9.3% and 9.2%, respectively. For years 1999, 2000, and 2001, for which full-year data were available for both the OCISS and the MDS, the proportion of OCISS patients also identified through the MDS was 7.9%, 5.7%, and 5.8%, respectively.

Figures 2 and 3 show the distribution of patients diagnosed in 2001 and receiving home health care and nursing home care, respectively, according to their timing of initiation of such care relative to their date of cancer diagnosis. In these analyses, patients are counted only once.

Figure 2.

Figure 3.

As indicated in Figure 2, over 50% of breast cancer patients and nearly 60% of colorectal cancer patients initiated HHC in the first 6 months following cancer diagnosis. Conversely, the proportion of prostate patients initiating HHC in the 6 months following cancer diagnosis approached only 35%. The proportion of patients having initiated HHC a year or longer before cancer diagnosis ranged between 15% and 20% across all cancer sites. On the other hand, the proportion of patients initiating HHC 6 months after cancer diagnosis or later was highest among prostate cancer patients.

Figure 3 shows the distribution of LTC patients according to their initiation of nursing home care, relative to cancer diagnosis. The proportion of patients being admitted to LTC in the first 6 months following cancer diagnosis was highest in colorectal cancer patients (nearly 45%), compared to 20%–25% in breast and prostate cancer patients. On the other hand, the proportion of patients having initiated LTC one year or more prior to cancer diagnosis was highest in breast cancer patients (approximately 40%), compared with 25%–30% in prostate and colorectal cancer patients.

With regard to geocoding, nearly 97% of all cases were geocoded at the street address level, as recorded in the OCISS dataset (data not shown). Less than 4% were geocoded at the zip code centroid level, and very few cases could not be geocoded.

Discussion

This study reported on the Ohio CALD, which, while mirroring the SEER-Medicare in structure, also incorporates data from the Medicaid enrollment and claims files, as well as clinical assessment data from the OASIS and the MDS. With respect to the linking of Medicare and Medicaid data with state-based cancer surveillance or registry data, this compares closely with a database developed in Michigan, although the latter does not incorporate data from the OASIS or the MDS.

Research on cancer outcomes in elders must reflect current efforts in the practice of geriatric oncology to integrate principles of geriatrics in the delivery of cancer care to this patient population. To that end, a number of additional data elements measuring components of the above-referenced Comprehensive Geriatric Assessment should be accounted for, especially as we strive to refine existing risk adjustment methodologies, and gain a better understanding on variations in patterns of care and outcomes in this patient population.

In addition to comorbid conditions, the additional data elements needed for improved risk adjustment techniques include measures of functional limitation and geriatric syndromes. Since such data elements are not available through the SEER-Medicare files, the CALD incorporated data from the OASIS and the MDS, both of which offer a rich array of clinical assessment data. As such, the CALD presents the opportunity to develop new endeavors of research, as illustrated in the studies that were completed recently, or ones that are currently underway, as follows:

a) Methodological studies

The availability of the Ohio CALD has provided unprecedented opportunities to conduct a number of validation studies to assess the value of deriving pertinent variables from one source or another, or from multiple sources combined. In one study, for example, and using the MDS data as the gold standard, we examined the ability of Medicare claims data to identify patients receiving nursing home care.³³ In another study, and in collaboration with other investigators, and using the above-referenced linked cancer registry-Medicare-Medicaid database from Michigan, we assessed the ability of the Medicare Denominator file to identify dually eligible Medicare-Medicaid beneficiaries with cancer in two contiguous states from the Midwest, while accounting for the linked Medicaid-cancer registry file as the gold standard. In other studies currently underway, we are evaluating the added value of accounting for comorbidity measures derived from Medicaid claims data, from the OASIS, and/or from the MDS, relative to capturing these measures from Medicare claims data alone. These studies will greatly contribute to the refinement of methodologies employed in cancer-related HSR.

b) Cancer-related outcomes studies

The Ohio CALD has also provided opportunities to conduct empirical studies on patterns of cancer care and outcomes. In one study,³⁴ we compared the proportion of unstaged cancer cases across patients receiving nursing home care, those receiving home health care, and those who are community dwelling and not receiving home health care. Findings from this study demonstrated that unstaged cases were disproportionately represented in nursing home patients. In another study focusing on cancer patients receiving HHC for the first time in the 30 days before or after being diagnosed with incident cancer, we used baseline comorbidity, functional and cognitive assessment data from the OASIS to better characterize their clinical presentation.³⁵ In a third study, also limited to the patient population initiating HHC around the time of cancer diagnosis, we evaluated patterns of care for colorectal cancer and survival outcomes in relation to comorbidity, functional limitations, and geriatric syndromes.

The above offers just a few examples of the various research opportunities that the CALD can offer. A number of additional research studies under consideration include a characterization of nursing home cancer patients’ clinical presentation and the study of patterns of care and outcomes in relation to comorbidities, as well as functional limitations and geriatric syndromes; and trajectory of care and outcomes in relation to clinical presentation at baseline, as well as treatment modalities. Unfortunately, obtaining detailed data on patients’ clinical presentation including data on functional and cognitive status in addition to comorbidities requires that we limit our patient population to those receiving HHC or LTC because such data are available only in the OASIS and the MDS databases.

The timing of initiation of HHC or LTC as shown in Figures 2 and 3 is important not only to study trajectory of care, but also to assess the number of patients for whom clinical assessment data may be obtained at baseline, or around the time of cancer diagnosis. It remains to be shown whether those initiating HHC or LTC upon cancer diagnosis differ significantly in health status from those having initiated care prior to cancer diagnosis. From a clinical perspective, however, it is likely that among patients receiving HHC or LTC, those initiating HHC or LTC around the time of cancer diagnosis are healthier and with less comorbidities than those having initiated care prior to cancer diagnosis. In many instances, patients may experience a sudden change in health status following receipt of cancer treatment, necessitating HHC. In fact, for example, results from preliminary analyses have indicated that 85% of colorectal cancer patients initiating HHC in the 30 days before or after cancer diagnosis did so within a month after undergoing colorectal surgery (data not shown).

While representing a relatively modest proportion of geriatric oncology patients, outcomes in the subgroup of this population receiving nursing home care or home health care are likely to greatly affect the outlook in the geriatric population at large. In particular, for patients admitted to HHC and/or LTC in the month immediately preceding or following cancer diagnosis, data obtained from the relevant data sources can provide important insight on their clinical presentation at diagnosis, or at baseline, coinciding with the time at which clinicians develop a care management plan for the patient. Of course, with the rapidly growing elderly population, collecting data on all Medicare beneficiaries’ functional status and other measures pertinent to geriatric syndromes—in addition to comorbidities—becomes imperative. However, this will require a fundamental change in the way administrative data are collected, as discussed by Iezzoni and Greenberg in the context of data on functional status.³⁶

The logistical issues involved with the development of the CALD are noteworthy. This endeavor was costly, requiring a budget of well over $100,000. In the case of this particular database, which was built primarily with support from the author’s career development grant from the National Cancer Institute, the budget did not provide the needed amount in full from the beginning of the grant period, making it inevitable to obtain the data for one or two years at a time. As well, the various activities associated with obtaining data use agreements from the relevant government agencies and Institutional Review Boards required considerable time and effort that could have been devoted to the investigation itself. Last but not least, the absence of a data linking strategy that is accepted and disseminated as “standard” or uniform in the research community is bound to cause idiosyncrasies, affecting linkage results across state-based databases that are built by different investigators. Because of these difficulties, and given the benefits of the CALD described above, it would be highly desirable that the SEER-Medicare database incorporate data from Medicaid, the OASIS, and the MDS.

Some of the observed findings deserve to be discussed. As indicated by the findings in Table 1, the lower match rate with Medicaid among prostate cancer patients was expected, given that the Medicaid program serves primarily women, and older women are more likely than older men to have the income threshold to qualify for Medicaid benefit. Of note, however, is the greater proportion of men with colorectal cancer than those with prostate cancer enrolling in Medicaid (8.8% vs. 5.9%, respectively, p < 0.001; data not shown), possibly explained by the greater intensity of care necessitated for the treatment of colorectal cancer, compared with that of prostate cancer. Second, the decreasing proportion of Medicaid cases towards the end of the study period likely results from the fact that patients enroll in Medicaid when or after they are diagnosed with cancer. For example, a patient diagnosed with cancer in 2001 may not have been enrolled in Medicaid in previous years. Conversely, patients diagnosed with cancer in 1997 may have enrolled in Medicaid in subsequent years, and stayed on Medicaid for a number of years.

The drop in the proportion of patients identified in both the OCISS and the MDS from 1999 is likely due to the fact that the MDS data for the latter two years used for the linking consisted of Ohio patients only, whereas the MDS data for all other years used in this study were nationally representative, drawing from a larger pool of patients. Of note is that the MDS data for 2000 and 2001 were re-used from a previous project, and the data for other years were obtained from CMS as part of this study. Hence the differences in the match rates across the years.

Several limitations should be noted. First, the CALD is specific to Ohio; as a result, the generalizability of the findings from studies using this database may be limited. However, the demographic profile of the population in Ohio is very similar to that of neighboring states in the Midwest. Also, little has been published on cancer-related outcomes from Ohio that is comparable in breadth and scope to analyses using SEER-Medicare data and, for the first time, the CALD makes it possible to compare outcomes between Ohio and other states/regions that are represented in the SEER-Medicare files.

Second, patients admitted to HHC and/or LTC for whom measures of functional status and geriatric syndromes are available do not constitute the majority. However, as noted above, these patients represent the most vulnerable subgroup of elders with cancer, with varying levels of dependence. These patients have been traditionally grouped with all others, despite the fact that their clinical presentation and the complexity of their health care needs differ so greatly from their non-HHC/non-LTC counterparts.

In closing, the growing number of elders with cancer and the integration of principles of geriatrics in the delivery of oncology services to this patient population warrant that the data sources and research methodologies adapt to new paradigms in clinical practice. In the shorter term, appending OASIS and MDS data to the SEER-Medicare files would greatly improve the utility of SEER-Medicare data in addressing the emerging research needs in geriatric oncology. Longer-term strategies will require fundamental changes in the content of administrative data.