Determination of lung cancer incidence in the elderly using Medicare claims data

Validation of an algorithm for identifying incident cancer cases based on long-term illness and diagnosis related group program data from the French National Health Insurance Information System (SNDS)

PURPOSE

Three generic claims-based algorithms based on the Illness Classification of Diseases (10th revision- ICD-10) codes, French Long-Term Illness (LTI) data, and the Diagnosis Related Group program (DRG) were developed to identify retirees with cancer using data from the French national health insurance information system (Système national des données de santé or SNDS) which covers the entire French population. The present study aimed to calculate the algorithms' performances and to describe false positives and negatives in detail.

METHODS

Between 2011 and 2016, data from 7544 participants of the French retired self-employed craftsperson cohort (ESPrI) were first matched to the SNDS data, and then toFrench population-based cancer registries data, used as the gold standard. Performance indicators, such as sensitivity and positive predictive values, were estimated for the three algorithms in a subcohort of ESPrI.

RESULTS

The third algorithm, which combined the LTI and DRG program data, presented the best sensitivities (90.9%-100%) and positive predictive values (58.1%-95.2%) according to cancer sites. The majority of false positives were in fact nearby organ sites (e.g., stomach for esophagus) and carcinoma in situ. Most false negatives were probably due to under declaration of LTI.

CONCLUSION

Validated algorithms using data from the SNDS can be used for passive epidemiological follow-up for some cancer sites in the ESPrI cohort.

Characterizing Health Outcomes in Idiopathic Pulmonary Fibrosis using US Health Claims Data

BACKGROUND

Idiopathic pulmonary fibrosis (IPF) is a life-threatening interstitial lung disease (ILD). Characterizing health outcomes of IPF patients is challenging due to disease rarity.

OBJECTIVE

This study aimed to identify the burden of disease in patients newly diagnosed with IPF.

METHODS

Patients with ≥1 claim with an IPF diagnosis were identified from a United States healthcare insurer's database (2000-2013). Patients with other known causes of ILD or aged <40 years were excluded. Subgroups were compared based on the 2011 change in International Classification of Diseases, 9th Revision (ICD-9) definition of IPF and occurrence of IPF testing. The prevalence and incidence of preselected health conditions of clinical interest were estimated.

RESULTS

Median age of newly diagnosed patients (n = 7,298) was 62 years (54.0% male). Restricting to patients with IPF diagnostic testing did not substantially affect cohort characteristics, nor did ICD-9 IPF coding change. Mean follow-up was 1.7 years; 16.8% of patients died; and a substantial proportion of patients were censored due to end of health plan enrollment (50.7%) and other causes of ILD (19.6%). The incidence of pulmonary hypertension, lung cancer, and claims-based algorithm proxy for acute respiratory worsening of unknown cause was 22.5, 17.6, and 12.6 per 1,000 person-years, respectively.

CONCLUSIONS

Patients with IPF had a high disease burden with a variety of health outcomes observed, including a high rate of mortality. Database censoring due to changes in enrollment or other ILD diagnoses limited follow-up. Altering cohort entry definitions, including IPF testing or ICD-9 IPF coding change, had little impact on cohort baseline characteristics.

Characterizing idiopathic pulmonary fibrosis patients using US Medicare-advantage health plan claims data

BACKGROUND

Idiopathic pulmonary fibrosis (IPF) is a rare life-threating interstitial lung disease (ILD). This study characterizes demographics, health care utilization, and comorbidities among elderly IPF patients and estimates prevalence and incidence rates for selected outcomes.

METHODS

Cohort study using a large US health insurance database (Optum's Medicare Advantage plan).

INCLUSION CRITERIA

≥ 1 diagnosis code for IPF (2008 - 2014), age ≥65 years, no diagnosis of IPF or other ILD in prior 12 months. Demographics, health care utilization, comorbidities and incidence rates for various outcomes were estimated. Follow-up continued until the earliest of: health plan disenrollment, death, a claim for another known cause of ILD, or end of the study period.

RESULTS

4,716 patients were eligible; 53.4% had IPF diagnostic testing. Median age was 77.5 years, 50.3% were male, median follow-up time was 0.8 years. Incidence rates ranged from 1.0/1,000 person-years (lung transplantation) to 374.3/1,000 person-years (arterial hypertension). Baseline characteristics and incidence rates were similar for cohorts of patients with and without IPF diagnostic testing.

CONCLUSIONS

Elderly IPF patients experience a variety of comorbidities before and after IPF diagnosis. Therapies for IPF and for the associated comorbidities may reduce morbidity and associated health care utilization of these patients.

Evaluation of algorithms to identify incident cancer cases by using French health administrative databases

PURPOSE

Administrative databases are increasingly being used in cancer observational studies. Identifying incident cancer in these databases is crucial. This study aimed to develop algorithms to estimate cancer incidence by using health administrative databases and to examine the accuracy of the algorithms in terms of national cancer incidence rates estimated from registries.

METHODS

We identified a cohort of 463 033 participants on 1 January 2012 in the Echantillon Généraliste des Bénéficiaires (EGB; a representative sample of the French healthcare insurance system). The EGB contains data on long-term chronic disease (LTD) status, reimbursed outpatient treatments and procedures, and hospitalizations (including discharge diagnoses, and costly medical procedures and drugs). After excluding cases of prevalent cancer, we applied 15 algorithms to estimate the cancer incidence rates separately for men and women in 2012 and compared them to the national cancer incidence rates estimated from French registries by indirect age and sex standardization.

RESULTS

The most accurate algorithm for men combined information from LTD status, outpatient anticancer drugs, radiotherapy sessions and primary or related discharge diagnosis of cancer, although it underestimated the cancer incidence (standardized incidence ratio (SIR) 0.85 [0.80-0.90]). For women, the best algorithm used the same definition of the algorithm for men but restricted hospital discharge to only primary or related diagnosis with an additional inpatient procedure or drug reimbursement related to cancer and gave comparable estimates to those from registries (SIR 1.00 [0.94-1.06]).

CONCLUSION

The algorithms proposed could be used for cancer incidence monitoring and for future etiological cancer studies involving French healthcare databases. Copyright © 2017 John Wiley & Sons, Ltd.

Validating the use of Medicare Australia billing data to examine trends in skin cancer

Background:  Epidemiological data surrounding non-melanomatous skin cancer (NMSC) is highly variable, in part due to the lack of government cancer registries. Several studies employ the use of Medical Australia (MA) rebate data in assessing such trends, the validity of which has not been studied in the past. Conversely, melanoma skin cancer is a notifiable disease, and thus, MA and cancer registry data is readily available. The aim of the current study is to assess the use of MA for epidemiological measures for skin cancers, by using melanoma as a disease sample.

Methods:  Following ethics approval, data from MA and Victorian Cancer Registry (VCR) from 2004-2008 were extracted. Incidence of MA and VCR unique melanoma cases were compared and stratified by age and local government area (LGA). Regression and a paired-samples t-test were performed.

Results: During the study period; 15,150 and 13,886 unique melanoma patients were identified through VCR and MA data sources respectively. An outlier in the >80­ year age group was noted between MA and VCR data. When stratified by age, significant correlation between MA and VCR was observed for all patients (gradient 0.91, R²= 0.936) and following exclusion of >80 patients (gradient 0.96, R²= 0.995). When stratified by LGA, a high degree of observation was observed for all patients (gradient 0.94, R²= 0.977) and following exclusion of >80 patients (gradient 0.996, R²= 0.975).

Conclusion: Despite the inclusion of outlier data groups, acceptable correlation between MA and VCR melanoma data was observed, suggesting that MA may be suitable for assessing epidemiological trends. Such principals may be used to validate the use of MA data for similar calculations assessing NMSC trends.

Incidence and prevalence of non-melanoma skin cancer in Australia: A systematic review

Non-melanoma skin cancer (NMSC), including basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), is the most common cancer occurring in people with fair skin. Australia has been reported to have the highest incidence of NMSC in the world. Using a systematic search of the literature in EMBASE and Medline, we identified 21 studies that investigated the incidence or prevalence of NMSC in Australia. Studies published between 1948 and 2011 were identified and included in the analysis. There were six studies that were conducted on national level, two at state level and 13 at the regional level. Overall, the incidence of NMSC had steadily increased over calendar-years in Australia. The incidence of NMSC per 100,000 person-years was estimated to be 555 in 1985; 977 in 1990; 1109 in 1995; 1170 in 2002 and 2448 in 2011. The incidence was higher for men than women and higher for BCC than SCC. Incidence varied across the states of Australia, with the highest in Queensland. The prevalence of NMSC was estimated to be 2% in Australia in 2002. The incidence and prevalence of NMSC still need to be accurately established at both national and state levels to determine the costs and burden of the disease on the public health system in Australia.

Estimation of national colorectal-cancer incidence using claims databases

Background. The aim of the study was to assess the accuracy of the colorectal-cancer incidence estimated from administrative data. Methods. We selected potential incident colorectal-cancer cases in 2004-2005 French administrative data, using two alternative algorithms. The first was based only on diagnostic and procedure codes, whereas the second considered the past history of the patient. Results of both methods were assessed against two corresponding local cancer registries, acting as “gold standards.” We then constructed a multivariable regression model to estimate the corrected total number of incident colorectal-cancer cases from the whole national administrative database. Results. The first algorithm provided an estimated local incidence very close to that given by the regional registries (646 versus 645 incident cases) and had good sensitivity and positive predictive values (about 75% for both). The second algorithm overestimated the incidence by about 50% and had a poor positive predictive value of about 60%. The estimation of national incidence obtained by the first algorithm differed from that observed in 14 registries by only 2.34%. Conclusion. This study shows the usefulness of administrative databases for countries with no national cancer registry and suggests a method for correcting the estimates provided by these data.

Is it possible to estimate the incidence of breast cancer from medico-administrative databases?

One approach to estimate cancer incidence in the French Départements is to quantify the relationship between data in cancer registries and data obtained from the PMSI (Programme de Médicalisation des Systèmes d'Information Médicale). This relationship may then be used in Départements without registries to infer the incidence from local PMSI data. We present here some methodological solutions to apply this approach. Data on invasive breast cancer for 2002 were obtained from 12 Départemental registries. The number of hospital stays was obtained from the National PMSI using two different algorithms based on the main diagnosis only (Algorithm 1) or on that diagnosis associated to a mention of "resection" (Algorithm 2). Considering registry data as gold standard, a calibration approach was used to model the ratio of the number of hospital stays to the number of incident cases. In Départements with registries, validation of the predictions was done through cross-validation. In Départements without registries, validation was done through a study of homogeneity of the mean number of hospital stays per patient. Cross-validation showed that the estimates predicted by the model were true with data extracted by Algorithm 1 but not by Algorithm 2. However, with Algorithm 1, there was an important heterogeneity between French Départements as to the mean number of hospital stays per patient, which had an important impact on the estimations. In the near future, the method will allow using medico-administrative data (after calibration with registry data) to estimate Départemental incidence of selected cancers.

French claims data as a source of information to describe cancer incidence: predictive values of two identification methods of incident prostate cancers

Claims data from the "Programme de Médicalisation du Système d'Information" (PMSI) have been commonly used for several years to complement cancer registries and describe cancer incidence in France. It is less clear whether or not it is possible to use these data as an independent source of information to assess cancer incidence, in the absence of a regional cancer registry. Following a similar study on breast cancer, we present a study which aimed to evaluate two methods of identifying incident prostate cancer using claims data. These methods were developed using claims data from the Hospices Civils de Lyon (HCL) and their validity was tested against medical records. The first method (M1) identified incident patients as those who had at least one stay with a principal diagnosis of prostate cancer. The second method (M2) had a prostate cancer treatment code in addition to the criteria for the first method. Both methods of identification had similar results, indicating a low rate of false negatives (negative predictive values: M1 = 100 [CI95: 93.8-100], M2 = 98.6 [CI95: 90.1-99.6]) and a high rate of false positives (positive predictive values: M1 = 33.3 [CI95: 23.2-42.1], M2 = 33.7 [CI95: 24.2-43.2]). The sample size did not allow us to produce consistent estimates of sensitivity and specificity. Our results showed that an estimation of the number of incident cases of prostate cancer using both methods of identification would be biased because of the high rate of false positives. Statistical methods that correct identification errors should be used.

Comorbid illnesses and health care utilization among Medicare beneficiaries with lung cancer

We evaluated the association of chronic obstructive pulmonary disease (COPD) and congestive heart failure (CHF) with outcomes in a 5% Medicare sample of 4447 elderly beneficiaries with lung cancer. Twenty-nine percent of patients had COPD and 13% had CHF. Patients with COPD or CHF had significantly decreased survival (hazard ratios 1.14 (1.05-1.25) and 1.38 (1.18-1.62), respectively); most of this differential was within 2 months after diagnosis. Patients with COPD or CHF were significantly less likely to receive surgery or chemotherapy than patients with neither COPD nor CHF. The association with less chemotherapy was similar in patients with the highest probability of surviving more than 2 months after the cancer diagnosis. In Medicare beneficiaries with lung cancer, COPD and CHF were common and were associated with both short-term mortality and decreased use of cancer treatments. Accounting for these two comorbid illnesses is important in evaluating health care utilization in this population.

The added value of claims for cancer surveillance: results of varying case definitions

OBJECTIVE

As cancer diagnosis and treatment has moved to the outpatient healthcare setting, traditional cancer surveillance tools are less effective for complete and unbiased capture of incident cases. This study evaluates the potential for Medicare data to supplement cancer surveillance in a unique manner by using a standard that is independent of a central cancer registry.

DESIGN

State cancer registry records were matched with Medicare data. Case validation included inpatient record abstraction combined with a mail/telephone survey of treating physicians. The positive predictive value (PPV), sensitivity (capture rate), and potential additional cases were calculated for 6 Medicare claims-based case definitions.

RESULTS

The PPV varied according to cancer site and definition, ranging from 70%-97% (prostate) to 87%-98% (breast). Sensitivity varied inversely with PPV, ranging from 51%-94% (breast) to 10%-88% (lung). The most important factors that predicted being missed by the registry were having no admission to an ACOS-certified hospital and no surgical treatment.

CONCLUSION

Medicare data represent a valid resource for supplementing state cancer registries in surveillance efforts. This potential is especially applicable to cancers predominantly diagnosed and treated outside the hospital setting.

[Sensitivity and specificity of two methods used to identify incident breast cancer in specialized units using claims databases]

BACKGROUND

Hospital claims databases from acute care units are available nationwide and contain most patients at the beginning of their cancer. The goal is to define the ability of these databases to provide a number of incident breast cancer cases using identification methods. Two identification methods were assessed in three specialized sections of a teaching hospital.

METHODS

The first method identified women who had at least one stay with a principal diagnosis of breast cancer. The second, which is more restrictive, identified women who had at least one stay with a principal diagnosis of breast cancer and a breast cancer-specific surgical treatment code. Both methods were applied to 4588 women 20 Years of age or older hospitalized in three specialized sections of the Hospices Civils de Lyon in 2000. To categorize these women in two groups, incident breast cancer cases or non-incident breast cancer cases, 150 women were randomized in each of two groups, one for incident breast cancer cases and one for non-incident breast cancer cases. Their medical records were used as references.

RESULTS

Sensitivity, specificity and their credibility intervals were respectively 99.4% (84-99.9) and 91.7% (90.3-93.3) for the first method and 93.8% (76.2-98.7) and 97.3% (96.1-98) for the second. Among women wrongly identified with an incident breast cancer in 2000, 75.4% (43/57) had a breast cancer that was not incident that Year with the first method, compared to 96% (24/25) with the second. Among these women wrongly identified with an incident breast cancer, coding errors of the principal diagnosis were found for 24.6% (14/57) of patients with the first method and for 4% (1/25) with the second. Their correction led to 99.2% (86.5-99.9) sensitivity and 92.9% (91.4-94.6) specificity for the first method and to 94.2% (76.5-98.7) sensitivity and 97.3% (96.2-98.1) specificity for the second.

CONCLUSIONS

The second method using cancer-specific surgical codes appeared more specific with a slight loss in sensitivity. The use of identification methods to assess the number of incident cancer cases still have to be defined.

Using Medicare Data to Estimate the Number of Cases Missed by a Cancer Registry

BACKGROUND

Cancer surveillance is essential for assessing patterns of cancer occurrence. State cancer registries do not capture all available cases potentially biasing results. Secondary data may be useful in identifying new cases and estimating the number of cases missed.

OBJECTIVE

We sought to create 2 distinct data sources from Medicare claims to use in combination with registry data as 3 sources for a capture-recapture analysis to estimate the capture rate and bias in capture of a statewide cancer registry.

METHODS

Data from the Virginia cancer registry (Registry) were merged with Medicare inpatient (Part A) as well as Medicare outpatient and physician claims (Part B) to provide 3 sources to estimate missing cases. A 3-source loglinear model was used to estimate the number of missing cancer cases for breast, lung, colorectal, and prostate cancer. Models included main effects and interactions. Additional analysis looked at the effect of demographic and comorbidity variables.

RESULTS

Loglinear models demonstrated mostly positive dependence between the 3 sources, implying that 2-source models would underestimate missing cases and overestimate capture rates. Using capture-recapture estimates of total number of cancer cases as the denominator, capture rates for Registry ranged from 59% (colorectal) to 74% (lung). When the aggregate of cases found by either Medicare or Registry were used the capture rates ranged from 74% (prostate) to 89% (breast). Further analysis indicated that capture rates differed by demographic characteristics.

CONCLUSION

We conclude that Medicare claims are useful to supplement a Registry, estimate the number of missing cases, and assess bias in capture.

Critical care use during the course of serious illness

Despite its expense and importance, it is unknown how common critical care use is. We describe longitudinal patterns of critical care use among a nationally representative cohort of elderly patients monitored from the onset of common serious illnesses. A retrospective population-based cohort study of elderly patients in fee-for-service Medicare is used, with 1,108,060 Medicare beneficiaries at least 68 years of age and newly diagnosed with serious illnesses: 1 of 9 malignancies, stroke, congestive heart failure, hip fracture, or myocardial infarction. Medicare inpatient hospital claims from diagnosis until death (65.1%) or fixed-right censoring (more than 4 years) were reviewed. Distinct hospitalizations involving critical care use (intensive care unit or critical care unit) were counted and associated reimbursements were assessed; repeated use was defined as five or more such hospitalizations. Of the cohort, 54.9% used critical care at some time after diagnosis. Older patients were much less likely to ever use critical care (odds ratio, 0.31; comparing patients more than 90 years old with those 68-70 years old), even after adjustment. A total of 31,348 patients (2.8%) were repeated users of critical care; they accounted for 3.6 billion dollars in hospital charges and 1.4 billion dollars in Medicare reimbursement. We conclude that critical care use is common in serious chronic illness and is not associated solely with preterminal hospitalizations. Use is uneven, and a minority of patients who repeatedly use critical care account for disproportionate costs.

Marriage, widowhood, and health-care use

Despite suggestive evidence, there has been no adequately powered systematic study of the ways in which marital status influences health care consumption. Using a novel data set of 609016 newly diagnosed, seriously ill elderly individuals in the USA, and employing hierarchical linear models, we look at differences in the experience of hospitalization as a function of marital status. We find that the married consistently use higher quality hospitals and have shorter lengths of stay. On the other hand, the married and the widowed appear to receive similar quality care once they are in the hospital. Marital status thus has a substantial impact on the health care obtained by the elderly. We suggest that these patterns are most consistent with spouses exerting their benefits by functioning as higher-order decision-makers than as home health assistants.

Evaluation of an algorithm to identify incident breast cancer cases using DRGs data

Hospital databases have the potential to be inexpensive, timely and nationally representative sources of information about cancer. This study examines the utility of the French hospital database adapted from the Diagnosis Related Group (DRG) classification and named 'Programme de médicalisation des systèmes d'information (PMSI)', as an independent source to identify incident cancer cases. From the 19 679 women hospitalized and treated in 1998 in the public hospitals of the Somme area in France, we identified those diagnosed with breast cancer in the PMSI database. These women were matched with women in the cancer registry of the Somme area who had been diagnosed with breast cancer in 1998. An algorithm was used to identify cancer-related diagnoses and procedures reported to PMSI. The sensitivity, specificity and positive predictive value (PPV) of the PMSI database were calculated using the cancer registry as a gold standard. The PMSI database had 85% sensitivity, 99.9% specificity and 97% PPV for women hospitalized with breast cancer as a principal diagnosis. The sensitivity was higher by 9% for hospitalization with breast cancer as a secondary diagnosis but had a lower PPV (78%). In conclusion, the PMSI database seems to offer an interesting potential to assess breast cancer incidence, because of its high sensitivity, in particular when secondary diagnosis was considered, and its very high specificity and PPV. However, these preliminary results need to be confirmed by other studies in France before such databases are used, particularly in areas without cancer registries.

Ability of Medicaid claims data to identify incident cases of breast cancer in the Ohio Medicaid population

BACKGROUND

The use of Medicaid data to study cancer-related outcomes would be highly desirable. However, the accuracy of Medicaid claims data in the identification of incident cases of breast cancer is unknown.

OBJECTIVES

(1) To estimate the sensitivity of Medicaid claims data for case ascertainment of breast cancer, and (2) to determine the positive predictive value (PPV) of diagnostic and procedure codes retrieved from Medicaid claims, using the Ohio Cancer Incidence Surveillance System (OCISS) as the gold standard.

METHODS

The study used the linked OCISS and Medicaid enrollment files, 1997-1998 (n = 1,648). The claims search yielded 2,635 incident cases, of which 1,132 were also identified through the OCISS-Medicaid files. Sensitivity and PPV of Medicaid data were calculated in subgroups of the population.

RESULTS

The overall sensitivity was 68.7 percent, but varied greatly across the subgroups of the population. It was lower among women enrolled in Medicaid only for part of the study year than those enrolled in Medicaid for 12 months of the study year (56.7 percent and 78.0 percent respectively, p < 0.0001), and lower among those who are dual Medicare-Medicaid eligible compared to those not participating in the Medicare program (63.1 percent and 78.6 percent respectively, p < 0.0001). The overall PPV was 43.0 percent, increasing up to 86.6 percent in the presence of procedure codes indicating the presence of mastectomy and lumpectomy, in addition to that of breast cancer diagnosis.

CONCLUSIONS

The sensitivity of Medicaid claims for case ascertainment of breast cancer is somewhat low, but improves considerably when accounting for women enrolled in Medicaid for the entire duration of the study year. The PPV is poor due to a high rate of false positives. The higher PPV obtained in the presence of procedure codes, in addition to diagnosis codes, will help researchers to correctly identify incident cases of breast cancer using Medicaid claims data.

Care after the onset of serious illness: a novel claims-based dataset exploiting substantial cross-set linkages to study end-of-life care

To date, there has not been a study using a large, nationally representative group of patients with serious illness who are at risk for hospice use and who are followed forward in time to understand the determinants of hospice use. In this paper, we outline the development of a large new cohort of 1,221,153 Medicare beneficiaries newly diagnosed with 1 of 13 serious conditions in 1993, a cohort that can be used to study end-of-life care in the United States. In describing our methods, we illustrate the possible utility of Medicare claims for end-of-life research. The members of our cohort are followed forward for hospice and other health care use through December 1997, and for mortality through June 1999. Medicare claims data on their inpatient and outpatient hospital use is also collected. Based on the ZIP Codes and counties in which cohort members lived, we were also able to characterize the health care markets of cohort members, as well as obtain other socioeconomic information about them. Information about cohort member's health care providers is also available. Detailed health information about cohort members' spouses was also collected. We conclude by highlighting the types of analyses that can be conducted in this data set.

The lack of effect of market structure on hospice use

OBJECTIVE

To describe the relative importance of health care market structure and county-level demographics in determining rates of hospice use.

DATA SOURCES

Medicare claims data for a cohort of elderly patients newly diagnosed with lung cancer, colon cancer, stroke, or heart attack in 1993, followed for up to five years, and linked to Census and Area Resource File data.

STUDY DESIGN

Variation between markets in rates of hospice use by patients with serious illness was examined after taking into account differences in individual-level data using hierarchical linear models. The relative explanatory power of market-level structure and local demographic variables was compared.

DATA COLLECTION METHODS

The cohort was defined within the Medicare hospital claims data using validated algorithms to detect incident cases of disease with a three-year lookback. Use of hospice was determined by linkage at an individual level to the Standard Analytic Files for Hospice through 1997. Individual-level data was linked to the Area Resource File using county identifiers present in the Medicare claims.

PRINCIPAL FINDINGS

There is substantial variation in hospice use across markets. This variation is not explained by differences in the major components of health care infrastructure: the availability of hospital, nursing home, or skilled nursing facilities, nor by the availability of HMOs, doctors, or generalists.

CONCLUSIONS

Intercounty heterogeneity in hospice use is substantial, and may not be related to the set-up of the medical care system. The important local factors may be local preferences, differences in the particular mix of services provided by local hospices, or differences in community leadership on end of life-issues; many of these differences may be amenable to educational efforts.

Racial, ethnic, and affluence differences in elderly patients' use of teaching hospitals

OBJECTIVE

To understand the role of race, ethnicity, and affluence in elderly patients' use of teaching hospitals when they have that option.

METHODS

Using a novel data set of 787,587 Medicare patients newly diagnosed with serious illness in 1993, we look at how sociodemographic factors influence whether patients use a teaching hospital for their initial hospitalization for their disease. We use hierarchical linear models to take into account differences in the availability of teaching hospitals to different groups. These models look within groups of people who live in the same county and ask what demographic factors make an individual within that county more or less likely to use a teaching hospital.

RESULTS

We find that blacks are much more likely than whites to use teaching hospitals (odds ratio [OR], 1.75; 95% confidence interval [95% CI], 1.73 to 1.77). However, Hispanics and Asian-Americans are less likely to use teaching hospitals than are whites (Hispanic OR, 0.92; 95% CI, 0.88 to 0.97; Asian-American OR, 0.89; 95% CI, 0.82 to 0.97). Medicaid patients are less likely to use teaching hospitals (given their opportunities) than are non-Medicaid recipients (OR, 0.91; 95% CI, 0.90 to 0.92). And we find a curvilinear relationship with affluence, with those in the poorest and those in the wealthiest neighborhoods most likely to use a teaching hospital.

CONCLUSION

The use of teaching hospitals is more complex that heretofore appreciated. Understanding why some groups do not go to teaching hospitals could be important for the health of those groups and of teaching hospitals.

Use of Medicare claims data to measure county-level variation in breast carcinoma incidence and mammography rates

BACKGROUND

National-level population-based data about breast carcinoma incidence and its association with screening mammography are currently not available.

METHODS

Inpatient, hospital outpatient and physician/supplier Medicare claims were used to identify incident cases of breast carcinoma in women > or = 65 years from 1996 to 1997 and calculate county-level incidence rates. The 1994-1995 claims data were used to determine county-level rates of mammography, and determine the correlation with incidence.

RESULTS

The median 2-year incidence rate for women > or = 65 was 979/100,000, and substantial variation in incidence between counties was observed. (i.e. 25th percentile 789/100,000, 75th percentile 1186/100,000). Two-year county-level mammography rates also varied among counties (i.e. 25th percentile 30.5%, 75th percentile 40.9%) and were higher in white women than in black women (median 36.8 and 26.3%, respectively). Counties with higher rates of mammography also had higher age-adjusted incidence rates.

CONCLUSIONS

Medicare claims may provide an alternative source of population-based data, particularly for areas in which registry data are not readily available, or are of limited scope. The data highlight the geographic variation in incidence and screening rates that may be useful for targeted interventions, and also suggest that mammography remains in a growth phase.

Geographic variations in breast cancer survival among older women: implications for quality of breast cancer care

BACKGROUND

Breast cancer care, such as utilization of screening procedures and types of treatment received, varies substantially by geographic region of the United States. However, little is known about variations in survival with breast cancer.

METHODS

We examined breast cancer incidence, survival, and mortality in the 66 health service areas covered by the Surveillance, Epidemiology, and End Results (SEER) program for women aged 65 and older at diagnosis. Incidence and survival data were derived from SEER, while breast cancer mortality data were from Vital Statistics data.

RESULTS

There was considerable variation in breast cancer survival among the 66 health service areas (chi2 = 202.7, p <.0001). There was also significant variation in incidence and mortality from breast cancer. In a partial correlation weighted for the size of the health service area, both incidence (r =.812) and percent 5-year survival (r = -.587) correlate with mortality. In a Poisson regression analysis, the combination of variation in incidence and variation in survival explains 90.9% of the variation in mortality.

CONCLUSIONS

There is considerable geographic variation in survival from breast cancer among older women, and this contributes to variation in breast cancer mortality. Geographic variations in breast cancer mortality should diminish as the quality of breast cancer care becomes more standardized.

Administrative data for public health surveillance and planning

Electronically available administrative data are increasingly used by public health researchers and planners. The validity of the data source has been established, and its strengths and weaknesses relative to data abstracted from medical records and obtained via survey are documented. Administrative data are available from a variety of state, federal, and private sources and can, in many cases, be combined. As a tool for planning and surveillance, administrative data show great promise: They contain consistent elements, are available in a timely manner, and provide information about large numbers of individuals. Because they are available in an electronic format, they are relatively inexpensive to obtain and use. In the United States, however, there is no administrative data set covering the entire population. Although Medicare provides health care for an estimated 96% of the elderly, age 65 years and older, there is no comparable source for those under 65.

Determination of county-level prostate carcinoma incidence and detection rates with Medicare claims data

BACKGROUND

To the authors' knowledge, national-level population-based data regarding prostate carcinoma incidence and detection currently are not available. The availability of such data could identify those regions with a disproportionately high cancer incidence as well as the population-level association between prostate carcinoma detection and incidence.

METHODS

Inpatient, hospital outpatient, and physician/supplier Medicare claims from 1997 were used to identify incident cases of prostate carcinoma in men age > or = 65 years and to calculate state and county-level incidence rates. The 1991 and 1997 claims data were used to determine small area rates of prostate-specific antigen (PSA) testing and prostate biopsy and to determine their correlation with incidence.

RESULTS

The calculated incidence rates for 1997 were 890 per 100,000 and 1196 per 100,000, respectively, in white males and African-American males and varied substantially between counties (i.e., 25--75th percentile, 676--1124 per 100,000). Rates of PSA and prostate biopsy increased markedly from 1991 to 1997 in both white men (1580 per 100,000 to 24,286 per 100,000) and African-American men (1277 per 100,000 to 15,190 per 100,000), and considerable variation in detection between counties was observed. Counties that had higher rates of prostate biopsy also had higher age-adjusted incidence rates, and county-level PSA testing was found to be associated with incidence in African-American patients, but not in white patients.

CONCLUSIONS

Medicare claims may provide an alternative source of population-based data, particularly for areas in which registry data are not readily available or are of limited scope. In addition, claims provide otherwise unavailable national data concerning cancer detection.

An introduction to epidemiological research with medical databases

In most regards, database research is like any other epidemiological endeavor: excellent research can be conducted, but there are many potential difficulties. Training in appropriate epidemiological and statistical methodology, together with knowledge of the databases and their coding systems, is likely to magnify the advantages of databases and also minimize the potential problems. As in all epidemiological investigations, the quality of the data and the methodology employed need to be carefully considered in the context of the research questions at hand.

Cancer prevalence estimates based on tumour registry data in the Surveillance, Epidemiology, and End Results (SEER) Program

BACKGROUND

The Connecticut Tumor Registry (CTR) has collected cancer data for a sufficiently long period of time to capture essentially all prevalent cases of cancer, and to provide unbiased estimates of cancer prevalence. However, prevalence proportions estimated from Connecticut data may not be representative of the total US, particularly for racial/ethnic subgroups. The purpose of this study is to apply the modelling approach developed by Capocaccia and De Angelis to cancer data from the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute to obtain more representative US site-specific cancer prevalence proportion estimates for white and black patients.

METHODS

Incidence and relative survival were modelled and used to obtain estimated completeness indices of SEER prevalence proportions for all cancer sites combined, stomach, cervix uteri, skin melanomas, non-Hodgkin's lymphomas, lung and bronchus, colon/rectum, female breast, and prostate. For validation purposes, modelled completeness indices were computed for Connecticut and compared with empirical completeness indices (the ratio of Connecticut based prevalence proportion estimates using 1973-1993 data to 1940-1993 data). The SEER-based modelled completeness indices were used to adjust SEER prevalence proportion estimates for white and black patients.

RESULTS

Model validation showed that the adjusted SEER cancer prevalence proportions provided reasonably unbiased prevalence proportion estimates in general, although more complex modelling of the completeness indices is necessary for female cancers of the colon, melanoma, breast, cervix, and all cancers combined. The SEER-based cancer prevalence proportions are incomplete for most cancer sites, more so for women, whites, and at older ages. For all cancers combined, prevalence proportions tended to be higher for whites than blacks. For the site-specific cancers this was true for stomach, prostate, cervix uteri, and lung and bronchus (men only). For colon/rectal cancers the prevalence proportions were higher for blacks through ages 59 (men) and 64 (women), and then for the remaining ages they were higher for whites. Prevalence proportions were lowest for stomach cancer and highest for prostate and female breast cancers. Men experienced higher prevalence proportions than women for skin melanomas, non-Hodgkin's lymphomas, lung and bronchus, and colon/rectal cancers.

CONCLUSION

The modelling approach applied to SEER data generally provided reasonable estimates of cancer prevalence. These estimates are useful because they are more representative of cancer prevalence than previously obtained and reported in the US.

The utility of Medicare claims data for measuring cancer stage

BACKGROUND

The validity of using claims data for measuring tumor stage, one of the most important determinants of choice of therapy and long-term survival, is unknown.

OBJECTIVES

To determine the relative accuracy of both inpatient and hospital Outpatient Medicare claims for measuring the stage of disease of six commonly diagnosed cancers.

RESEARCH DESIGN

Analysis of a database linking Surveillance, Epidemiology, and End Results (SEER) registry data and Medicare claims in patients aged 65 years with cancer.

SUBJECTS

Three hundred twenty thousand, six hundred and thirty seven cases of invasive breast, colorectal, endometrial, lung, pancreatic, and prostate cancers diagnosed between 1984 and 1993.

MEASURES

Using SEER files as the "gold standard," concordance with Medicare claims, as well as sensitivity and positive predictive value of coding for each stage was measured.

RESULTS

Although Medicare data correctly categorized local, regional, and distant stage tumors in 97%, 33%, and 65%, respectively, the data substantially overestimated the proportion of localized tumors and underestimated the rate of regional stage disease. The highest concordance was observed for breast and colorectal cancer. However, the sensitivity and positive predictive values were never simultaneously 80% within one stage of a specific cancer. The accuracy of coding for stage in Outpatient files was inferior to inpatient data.

CONCLUSIONS

With few exceptions, Medicare claims have limited utility as a measure of cancer stage. If tumor registry data are not available, investigators should consider the trade offs in sensitivity and predictive value when considering a study that will use claims data.

Use of Medicare hospital and physician data to assess breast cancer incidence

OBJECTIVES

Health claims data have the potential of being an inexpensive, timely, and nationally representative source of information about cancer. This study examined the utility of Medicare hospital and physician data as an independent source to identify incident breast cancer cases.

METHODS

Data came from Medicare and the National Cancer Institute's SEER cancer registries. From 1992, for women residing in the SEER states (n = 659,260), Medicare hospital and physician claims were reviewed to identify women with a breast cancer diagnosis on a claim (n = 6,784). These women were matched with women in the SEER data who had been diagnosed with breast cancer in 1992 (n = 3,230). The sensitivity, specificity, and positive predictive value (PPV) of the Medicare data were calculated. Logistic regression models were used to identified cancer related procedures reported to Medicare that could distinguish true cases from false positive cases. Predicted values from these models were included to create plots of sensitivity versus false positive rates and sensitivity versus PPV.

RESULTS

Medicare hospital data had 62% sensitivity, 99.9% specificity, and 88% PPV. Physician claims increased sensitivity by 14%, with specificity of 99.4%, and a PPV of 10%. Inclusion of additional cancer related diagnoses and procedures improved the ability to distinguish true cases from false positives, although the number of false positive cases remained high.

CONCLUSIONS

The Medicare data overall offer limited potential to assess breast cancer incidence, largely because of low sensitivity and poor PPV. The Medicare data may have utility to identify women undergoing selected breast cancer treatments. In addition, the data may be used to help registries focus case-finding efforts, particularly for persons undergoing cancer related treatments.

The sensitivity of Medicare claims data for case ascertainment of six common cancers

BACKGROUND

Although Medicare claims data have been used to identify cases of cancer in older Americans, there are few data about their relative sensitivity.

OBJECTIVES

To investigate the sensitivity of diagnostic and procedural coding for case ascertainment of breast, colorectal, endometrial, lung, pancreatic, and prostate cancer.

SUBJECTS

Three hundred and eighty nine thousand and two hundred and thirty-six patients diagnosed with cancer between 1984 and 1993 resided in one of nine Surveillance Epidemiology and End Results (SEER) areas.

MEASURES

The sensitivity of inpatient and Part B diagnostic and cancer-specific procedural codes for case finding were compared with SEER.

RESULTS

The sensitivity of inpatient and inpatient plus Part B claims for the corresponding cancer diagnosis was 77.4% and 91.2%, respectively. The sensitivity of inpatient claims alone was highest for colorectal (86.1%) and endometrial (84.1%) cancer and lowest for prostate cancer (63.6%). However, when Part B claims were included, the sensitivity for diagnosis of breast cancer was greater than for other cancers (93.6%). Inpatient claim sensitivity was highest for earlier years of the study, and, because of more complete data and longer follow up, the highest sensitivity of combined inpatient and Part B claims was achieved in the late 1980s or early 1990s.

CONCLUSIONS

Medicare claims provide reasonably high sensitivity for the detection of cancer in the elderly, especially if inpatient and Part B claims are combined. Because the study did not measure other dimensions of accuracy, such as specificity and predictive value, the potential costs of including false positive cases need to be assessed.

Geographic variations in breast cancer mortality: do higher rates imply elevated incidence or poorer survival?

OBJECTIVES

Mortality rates from breast cancer are approximately 25% higher for women in the northeastern United States than for women in the South or West. This study examined the hypothesis that the elevation is due to decreased survival rather than increased incidence.

METHODS

Data on breast cancer incidence, treatment, and mortality were reviewed.

RESULTS

The elevated mortality in the Northeast is apparent only in older women. For women aged 65 years and older, breast cancer mortality is 26% higher in New England than in the South, while incidence is only 3% higher. Breast cancer mortality for older women by state correlates poorly with incidence (r = 0.28).

CONCLUSIONS

Those seeking to explain the excess breast cancer mortality in the Northeast should assess survival and should examine differences in cancer control practices that affect survival.

Validity of diagnostic and drug data in standardized nursing home resident assessments: potential for geriatric pharmacoepidemiology. SAGE Study Group. Systematic Assessment of Geriatric drug use via Epidemiology

OBJECTIVES

The Health Care Financing Administration requires that patients admitted to certified nursing homes be assessed with the Minimum Data Set, a data collection instrument containing more than 300 demographic, diagnostic, clinical, and treatment variables. Long-term care databases potentially may be used to assess the outcomes of specific treatments as well as drug effectiveness. The authors sought to ascertain reliability and validity of diagnostic and drug data in a database obtained by merging the Minimum Data Set with detailed information on drugs consumed by each resident.

METHODS

A population of 296,379 residents of 1,492 nursing homes in Kansas, Maine, Mississippi, New York, and South Dakota participated in the study between 1992 and 1994. Minimum Data Set clinical diagnoses were contrasted with selected resident characteristics and a variety of symptoms and treatments. Limited to individuals who had been hospitalized in the 6 months preceding the first assessment, Minimum Data Set diagnoses were compared with those on the hospital discharge claims maintained in the Medicare Provider Analysis and Review database. Finally, the probability that the use of selected drugs predicted the correspondent gender-specific, age-specific, or unique labeled indication was estimated.

RESULTS

The positive predictive value for Minimum Data Set diagnoses compared with gender or function measures exceeded 0.9, and it was 0.8 for specific symptoms and 0.6 for virtually all other comparisons. The positive predictive value for Minimum Data Set diagnoses compared with those from hospital claims was approximately 0.7 for all chronic medical conditions, except for depression and asthma/chronic obstructive pulmonary disease/emphysema. The positive predictive value for acute/subacute diagnoses (ie, pneumonia, urinary tract infection, anemia) that may resolve during hospital stay was less than 0.5. The positive predictive value for selected drugs, except estrogens, compared with age and gender was close to 1.0 in all cases. When compared to their labeled indication, the positive predictive value was more than 0.6 for all drugs considered, with 0.97, 0.91, and 0.87 for tacrine and Alzheimer's disease, antidiabetics and diabetes mellitus, and L-dopa and Parkinson's disease, respectively.

CONCLUSIONS

These findings point to the overall validity of the drug and clinical data in this Minimum Data Set-based data set. Additional validation efforts will determine whether this data set can be used for studies of geriatric pharmacoepidemiology and for analyses of the influence of different policies and practices on residents' outcomes.

Comorbidity measurement in elderly female breast cancer patients with administrative and medical records data

The inter-rater reliability, cross-source (Medicare claims versus medical record) agreement, and ability to predict all-cause mortality of three aggregate comorbidity indices were evaluated in a group of 404 elderly, incident breast cancer cases identified from the Virginia Cancer Registry and linked to Medicare administrative data files. Comorbidity was based on both medical records and Medicare claims data using indices from Charlson et al (1987), Satariano and Ragland (1994), and Kaplan and Feinstein (1974). Inter-rater agreement was good for all indices (kappas > or = 0.80). Agreement between comorbidity indices measured by claims and medical records was considerably poorer (kappas between 0.30 and 0.40). However, claims-based and medical records-based comorbidity indices were similarly associated with mortality. For the Charlson index, the index best predicting survival, the adjusted relative risk for an increase from a lower to higher comorbidity category was 1.48 (95% confidence interval 1.23, 1.78) based on medical records compared to 1.53 (95% confidence interval 1.23, 1.93) based on Medicare claims. The claims-based Charlson index score still appeared to be associated with survival (relative risk = 1.30; 95% confidence interval = 1.00, 1.70) after controlling for the medical records-based score. This suggests that both comorbidity data sources add valuable prognostic information and, conversely, that the use of either source alone will result in some misclassification of comorbidity.

Use of Medicare claims data to estimate national trends in stroke incidence, 1985-1991

BACKGROUND AND PURPOSE

Although stroke mortality has been declining in the United States for decades, recent trends in stroke incidence based on national data have not been described. We used Medicare hospitalization data to estimate national trends in the incidence of stroke among Americans aged 70 years or older, and we provide evidence of the validity of the estimate.

METHODS

We defined stroke as a principal diagnosis with International Classification of Diseases, 9th Revision, Clinical Modification codes 430 to 434 or 436 to 437. We excluded many recurrent cases from the analysis by eliminating persons hospitalized for stroke during the 5 years preceding the index stroke. We calculated annual adjusted incidence rates and examined trends graphically. We investigated the effect of different exclusion periods, trends in in-hospital mortality of stroke patients, and trends in out-of-hospital stroke mortality. We examined trends in relation to sex, race, and age.

RESULTS

The estimated age- and sex-adjusted stroke incidence declined 9.5% from 1985 to 1989, then increased 3.3% to 1991. The pattern did not vary with the length of the exclusion period or when all listed diagnosis rather than principal diagnoses were used to identify stroke cases. Incidence trends resembled the overall trend for both men and women, for 5-year age groups, and for whites; the trend did not change for blacks.

CONCLUSIONS

Stroke incidence declined steadily from 1985 to 1989 and then increased slightly to 1991. Several postulated potential sources of bias were investigated and found to be unlikely to account for the incidence decline, although some may have contributed to the subsequent incidence increase.

Projected numbers of cancers diagnosed in the US elderly population, 1990 through 2030

As based on Bureau of the Census population projections and age-specific cancer incidence rates for 1985 to 1989 from the National Cancer Institute's Surveillance, Epidemiology and End Results program, the number of incident cancers diagnosed annually in the United States among persons aged 65 and over is projected to reach 1.5 million by the year 2030, or 2.4 times the number estimated for 1990. These projections, which may be conservative because birth cohort patterns (based on Connecticut rates) suggest possible future increases in incidence rates for all cancer sites combined, have implications for planning expanded primary prevention efforts, such as smoking cessation (especially for women) and dietary modification programs, and for projecting health care needs and costs.

Measuring the incidence of cancer in elderly Americans using Medicare claims data

BACKGROUND

The Surveillance, Epidemiology, and End Results (SEER) program of the National Cancer Institute is the most frequently used and best estimate of the incidence of cancer in the United States. Although synthetic estimates based on the SEER information can be used to plan cancer prevention and intervention programs, the evaluation of these action programs and the monitoring of cancer incidence in states or other geographic areas requires information on the population for whom the program is directed.

METHODS

The age-adjusted incidence of six cancers among persons 65 years of age and older for 1986-1987 living in the five states participating in the SEER program was compared with the incidence derived from hospitalization records contained in the Health Care Financing Administration's (HCFA) administrative data files. Age-adjusted incidence rates for 1990 developed from HCFA data for persons living in the nine SEER program areas were contrasted with the incidence rates for persons living in the rest of the United States and were developed for each of the 50 states and the District of Columbia.

RESULTS

The comparison of the SEER and HCFA overall age-adjusted cancer incidence rates in the elderly for 1986-1987 showed that for four of the six cancers (breast, colon, lung, and corpus uteri) the rates differed by 5% or less. The HCFA derived rates were 6.37% and 7.65% greater than the SEER rates for prostate and esophagus cancer, respectively. The incidence of cancer between 1986 and 1990 was neither uniformly higher nor lower among elderly SEER program area residents compared with residents of the rest of the country. Incidence rates varied greatly among states for each of the cancers.

CONCLUSIONS

HCFA administrative data can be used by states or other geographic units to monitor the incidence of cancer in the elderly as well as to plan and evaluate cancer prevention and intervention programs.