Journal of the National Cancer Institute Advance Access originally published online on March 11, 2008
JNCI Journal of the National Cancer Institute 2008 100(6):378-379; doi:10.1093/jnci/djn060
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© The Author 2008. Published by Oxford University Press.
EDITORIALS |
Enhancing Cancer Registry Data to Promote Rational Health System Design
Affiliation of author: Dana Farber Cancer Institute, Boston, MA
Correspondence to: Deborah Schrag, MD, MPH, Dana Farber Cancer Institute, 44 Binney St, Boston, MA 02115 (e-mail: deborah_schrag{at}dfci.harvard.edu).
Those of us in large chaotic health-care systems, like that in the United States, look with particular interest at the experience in smaller more contained health systems, like that in The Netherlands, to glean how we might apply rational principles to organizing care delivery. In classical epidemiology, we have learned a tremendous amount by examining the experiences of those in The Netherlands and other countries, such as Scandinavia, who couple meticulous records on the vital status and health of their populations with high-quality cancer registries. Similarly, in clinical epidemiology and health services research, we gain enormously from evaluating the experiences of other health-care systems with different models of care delivery.
It is therefore with great interest that we examine the study from The Netherlands by Vernooij et al. (1) that compares delivery of ovarian cancer care among generalized, semispecialized, and specialized hospitals. The authors sought to obtain information as to whether ovarian cancer patients who receive their care in specialty cancer hospitals have superior survival outcomes compared with those who receive their care in semispecialized or general hospitals. Using tumor registry data, they compared survival outcomes for the 40% of Dutch women who received ovarian cancer care at general hospitals with that of the 40% who received care at hospitals with some cancer expertise and the 20% who received care at specialty cancer centers. They found a 5-year relative survival of 38% at general hospitals, 39.4% at semispecialized hospitals, and 40.3% at specialty hospitals. They relied on 5-year relative survival—that is, the ratio of the observed survival to the survival that was expected based on the age- and period-specific mortality of the general population—as a proxy for cancer-specific survival, which is not reliably tracked by the cancer registry that they used.
The limitations of the analysis by Vernooij et al. (1) and by others (2–4) are those that are inherent in their data source—population-based tumor registries. Although tumor registries in The Netherlands and elsewhere collect meticulous high-quality data on cancer site, histology, and stage, several critical pieces of information that influence both choice of treatment and cancer outcomes are missing. In particular, there is no information about comorbidity or performance status. Consequently, it is exceedingly difficult, if not impossible, to determine whether the survival differences reported are due to differences in the quality of care or to underlying systematic differences in the characteristics of patients who seek care at general vs specialty hospitals. This potential for selection bias is the bugaboo of scores of articles on regionalization of specialty cancer care as well as the related voluminous literature on volume–outcome relationships in cancer surgery.
In the analysis of Vernooij et al. (1), we note that patients treated at general hospitals were an average of 5 years older and were substantially less likely to undergo primary surgical reduction or to receive systemic chemotherapy than those treated at more specialized centers. It is of course not possible to know whether the patients who did not undergo debulking surgery and chemotherapy would have received these interventions had they gone to a specialty center or whether these interventions would have improved their outcomes. However, both the age of the patients and the treatment differences strongly indicate that Dutch ovarian cancer patients who seek care at specialized hospitals have different characteristics from those who seek care at general hospitals. This potential for confounding makes it impossible to know whether the superior outcomes at specialty centers are a result of the patients’ underlying attributes and disease severity or of the care they received. Therefore, this analysis in and of itself does not justify regionalization of ovarian cancer surgery in The Netherlands to specialty centers. Although there are many reasons that explain why the care may be superior in specialty centers, the survival differences are quite modest and there is considerable evidence of selection when it comes to hospital choice. The data presented hint that generalized hospitals in The Netherlands may treat the very old and infirm, who do not undergo surgery at all, as well as patients with early-stage uncomplicated disease, who generalist gynecologists may be more comfortable managing. Such a heterogeneous mixture of patients would attenuate differences in survival outcomes between generalized and specialized centers.
The optimal strategy to mitigate selection bias is randomization. If women could be randomly assigned to receive their care in particular hospitals and systematic differences in their outcomes were evident, then we would have convincing evidence about how to organize care. However, the practical obstacles to randomly assigning patients to receive their care in particular centers are formidable. In the United States, the ability to conduct such a study is hindered by insurance constraints that specify which hospitals a patient may choose and by geographic barriers, such as specialized hospitals that are located far from the patient's home. However, even in The Netherlands, with universal health-care coverage and a compact, highly networked transportation system, it is not clear that such a study would be feasible.
The alternative to randomization is to use statistical techniques to minimize the impact of selection bias. Careful collection of data on potential parameters that may influence the association between choice of a specialized hospital and survival outcomes permits the application of statistical techniques such as multivariable logistic regression and propensity score analysis to adjust for the impact of any imbalances in the types of patients who seek care at different types of hospitals. However, these statistical techniques cannot account for selection that is based on characteristics that go unmeasured. Therefore, the key questions to consider in interpreting the results of Vernooij et al. (1) are, first, whether there is evidence of systematic differences in the types of patients who seek care at specialty vs general hospitals and, second, whether there is reason to suspect that unmeasured patient attributes might influence hospital choice.
Important information that is lacking in the study by Vernooij et al. (1) includes the comorbidities and performance status. A higher comorbidity burden among women receiving care at general hospitals may explain their inferior survival outcomes as well as their lower tendency to receive definitive surgery or chemotherapy. Comorbidity of older patients in the United States can be estimated because cancer registry data have been linked to Medicare claims data (5); however, comorbidity algorithms rely on patients’ concurrent medical conditions and diagnoses but fail to capture performance status, which measures a patient's ability to function. Performance status is a powerful predictor of outcomes, especially for ovarian cancer (6). It captures much of what seasoned clinicians ascertain in an instant as they watch a patient enter a room, rise from a chair, or clamber onto an exam table and distinguishes the patient who is in bed most of the time from the patient who remains active and fully functional.
Determining whether outcomes are superior when care is delivered in specialized centers or by specialty-trained physicians, such as gynecologic oncologists, is an important question. Therefore, rather than throwing up our hands in despair over the missing data elements and selection bias inherent in most registry studies, what strategies might improve the caliber of these analyses and thereby inform rationale health system design?
In many health-care systems, including those of The Netherlands and the United States, cancer registries are well established and indeed the crown jewels of national health surveillance systems. The existence of high-quality tumor registry data not only permits classical epidemiologic studies and monitoring of the overall cancer burden but also increasingly provides a platform for analysis of the performance of health-care delivery systems. Registries are typically staffed by well-trained, dedicated personnel with particular expertise in medical record abstraction who have the ability, training, and legal authorization to handle confidential patient information. Unfortunately, these registries are also woefully underfunded and, as a result, there is great reluctance to increase the number of data elements collected. However, as the number of cancer therapies increases and as the need to systematically evaluate their real-world clinical effectiveness grows, there is a need to optimize the data that can be gleaned from observational data sources. A compelling case can be made for expanding the scope of data that tumor registries collect. There are, of course, considerable hurdles to any such expansion. For example, although most clinicians document comorbidity and performance status, there is considerable variation in the systems used. Consensus on preferred systems would greatly facilitate abstraction of these data elements by registries. Finally, in contrast to static data elements such as stage and histology, comorbidity and performance status are dynamic and fluctuate over time. Development of timing rules for measuring and recording these parameters would facilitate their reporting to cancer registries. In addition, linking information about physician characteristics to registry data would allow analysis of associations between outcome and various physician characteristics.
Given the inability to evaluate every clinical intervention and strategy for organizing health-care delivery in the context of a randomized trial, it is inevitable that we will continue to rely heavily on observational population-based data such as the registry data used by Vernooij et al. (1). As medical record abstraction has been expedited by information and computer technology, it is important to consider how we maintain and indeed improve the quality of cancer registry data so that the important questions, such as those posed by Vernooij et al. (1), can be answered with greater ease. By working across international boundaries to develop consensus on data collection strategies, metrics, timing rules, and even systems for engaging patients in data collection more directly, we can optimize the value of cancer registry data to provide information not only about cancer incidence and mortality but also about the optimal design of cancer care delivery.
REFERENCES
1. Vernooij F, Heintz APM, Witteveen PO, van der Heiden-van der Loo M, Coebergh J-W, van der Graaf Y. Specialized care and survival of ovarian cancer patients in The Netherlands: nationwide cohort study. J Natl Cancer Inst. (2008) 100(6):399–406.
2. Vernooij F, Heintz P, Witteveen E, van der Graaf Y. The outcomes of ovarian cancer treatment are better when provided by gynecologic oncologists and in specialized hospitals: a systematic review. Gynecol Oncol (2007) 105(3):801–812.[CrossRef][Web of Science][Medline]
3. Silber JH, Rosenbaum PR, Polsky D, et al. Does ovarian cancer treatment and survival differ by the specialty providing chemotherapy? J Clin Oncol (2007) 25(10):1169–1175.
4. Earle CC, Schrag D, Neville BA, et al. Effect of surgeon specialty on processes of care and outcomes for ovarian cancer patients. J Natl Cancer Inst (2006) 98(3):172–180.
5. SEER-Medicare Linked Database. http://healthservices.cancer.gov/seermedicare/. Accessed February 1, 2008.
6. Carey MS, Bacon M, Tu D, Butler L, Bezjak A, Stuart GC. The prognostic effects of performance status and quality of life scores on progression-free survival and overall survival in advanced ovarian cancer. Gynecol Oncol (2008) 108(1):100–105.[CrossRef][Web of Science][Medline]
Correspondence about this Article
Related Articles in JNCI
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2008 100: 1414-1415.
J Natl Cancer Inst 2008 100: 399-406.
J Natl Cancer Inst 2008 100: 375.
This article has been cited by other articles:
|
|
 |
|
  M. N. Levine and J. A. Julian Registries That Show Efficacy: Good, but Not Good Enough J. Clin. Oncol., November 20, 2008; 26(33): 5316 - 5319. [Full Text] [PDF]
|
|
|
|
 |
|
 M. L. G. Janssen-Heijnen, H. A. A. M. Maas, and J. W. W. Coebergh Re: Enhancing Cancer Registry Data to Promote Rational Health System Design J Natl Cancer Inst, October 1, 2008; 100(19): 1414 - 1415. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||