Health Affairs, 24, no. 4 (2005): 928-937 doi: 10.1377/hlthaff.24.4.928 © 2005 by Project HOPE	New Online   Senate Health Reform Bill   Rewarding Providers   Public Option Policy Brief   Health Reform & Abortion   Delivery System Reform

This Article Abstract Reprint (PDF) Submit a response to this article Alert me when this article is cited Alert me when Comments are posted Alert me if a correction is posted Services E-mail this article to a friend Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Archive Download to Citation Manager Reprints & Permissions Citing Articles Citing Articles via Web of Science (5) Citing Articles via Google Scholar Google Scholar Articles by Malenka, D. J. Articles by Wennberg, J. E. Search for Related Content PubMed PubMed Citation Articles by Malenka, D. J. Articles by Wennberg, J. E. Related Collections Health Reform Legal/Regulatory Issues Medicare Business Of Health Pharmaceuticals Research And Technology Consumer Issues

FDA Under Scrutiny

Postmarketing Surveillance Of Medical Devices Using Medicare Claims

Abstract

 
The current system of postmarketing surveillance of high-risk medical devices could be improved by taking advantage of the administrative billing data collected by the Centers for Medicare and Medicaid Services (CMS) to systematically monitor for adverse events that may signal device-related problems. In this paper we use the current concern about the excess risk associated with drug-eluting coronary stents to highlight the strengths and weaknesses of claims data for postmarketing surveillance and propose a pilot collaboration between government, industry, and academe to systematically explore the use of Medicare claims data for this purpose.

Because of the complexity of understanding the full range of device-related outcomes, Congress has given the FDA the responsibility for overseeing a program of postmarketing surveillance. The FDA has several mechanisms for conducting this surveillance.² Voluntary reporting of device-related concerns by health care institutions and professionals is facilitated through the MedWatch program. The Medical Device Reporting (MDR) system mandates that manufacturers and importers report device-related deaths, serious injuries, and malfunctions; user facilities must report device-related deaths to the FDA and deaths plus serious injuries to the manufacturer. Congressional concerns with the system of mandatory reporting led to the statutory requirement for the Medical Product Surveillance Network (MedSun), a nationwide sample of facilities specially trained to collect complete and accurate data on device-related problems. As a condition of pre-market approval, the FDA can request that manufacturers report information on a broader range of adverse events than mandated through the MDR. Under Section 522 authority, or as another condition of premarket approval, the FDA can also mandate that manufacturers conduct postmarket studies based on concerns about specific safety issues or longer-term effectiveness. Finally, the FDA collaborates with similar agencies throughout the world in an effort to pool important device-related information.

Despite this varied approach to postmarketing surveillance, the current system of surveillance has some weaknesses. Adverse events are underreported, for a variety of reasons.³ There is no systematic approach to detecting device-related adverse events. The reporting system is passive and relies on manufacturers’ and providers’ good will and watchfulness; it assumes that they will be able to recognize adverse events and associate them with the use of a medical device. Health care providers are not necessarily well informed about their obligation to report or how to use the system of reporting. Manufacturers and providers are concerned about issues of liability and adverse publicity. Submitted data may be incomplete, inaccurate, and, typically, unverified. There is no routine mechanism for determining the true rate of an adverse event because no denominator data on how often a device is used are routinely collected.

In this paper we describe the potential of the Medicare claims system for postmarketing surveillance, describe its advantages and limitations in assessing adverse events following the introduction of drug-eluting stents, and recommend specific steps for improving the usefulness of claims data for this purpose. We then propose a pilot project to demonstrate the value of Medicare claims to supplement current efforts at postmarketing surveillance.

The Medicare claims database. In 2003 more than 34.8 million Americans age sixty-five and older were enrolled in traditional, fee-for-service Medicare—Parts A and B. For these enrollees, the CMS maintains a database recording the use of all services that are eligible for payment. The Part A database contains a record for each Medicare hospitalization. These records include unique identifiers for the hospital and patient, the dates of admission and discharge, admitting diagnosis, procedures performed, and additional diagnoses representing comorbid conditions. Procedures and diagnoses are coded using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) convention.⁴ Part B covers physician services such as physician visits and surgical and diagnostic procedures. These claims contain unique patient and physician identifiers with the billable service coded using Current Procedural Terminology (CPT) codes, as well as the diagnoses for which the service was performed (coded in ICD-9-CM).⁵

Medicare also maintains an enrollment file, updated monthly, which includes important demographic information such as age, sex, race, place of residence, and date of death (if applicable). Since individual records can be linked, claims data in essence constitute a person-specific registry of medical histories.

The implications for postmarketing surveillance are that all patients enrolled in traditional Medicare who receive a specific device could be identified through the Part A or Part B records, and all subsequent billable events could be temporally related to the date on which the device was implanted. Therefore, patients who receive a specific device can be assembled into cohorts and, to the extent the database allows, the incidence of adverse events estimated and compared with event rates for the existing standard of care. As long as the patient remains in the traditional Medicare program, records for hospitalizations and physician services accumulate, and there is no loss to follow-up. If additional clinical detail about a patient or an event is needed, unique patient and provider identifiers make locating the medical record for chart review easy and reliable.

Example: subacute thrombosis in drug-eluting stents. Since the 1980s percutaneous coronary interventions (PCIs) have grown to become the preferred strategy for treating patients with blockage(s) of one or more coronary arteries because of atherosclerotic plaque. The original technology used a balloon-tipped catheter to fracture the plaque and stretch the blood vessel. The Achilles’ heel of this approach was that as much as half the time, the blockage would recur within six to twelve months. This problem stimulated the development of coronary stents: slotted tubes that could be placed across an area of blockage to buttress open the vessel and prevent restenosis. These devices reduced the risk of restenosis but did not eliminate it. In response, the device industry developed drug-eluting stents, which in contrast to the existing "bare-metal stents" were coated with a drug designed to prevent the overexuberant healing response associated with restenosis. Based on a premarket experience with 673 patients, the FDA approved the first drug-eluting stent in the United States, the Cordis/Johnson and Johnson CYPHER, for general use 23 April 2003.

During the first five months of general distribution, more than 260,000 stents were shipped. However, during this time the FDA, via Johnson and Johnson, began to receive reports of subacute thrombosis following placement of the stents.⁶ By October 2003 the FDA recognized a significant increase in the number of reported cases of subacute thrombosis compared with what it had been receiving before the drug-eluting stents were introduced. On 29 October 2003 the agency posted a public health notification to physicians describing the receipt, through the voluntary medical device reporting system, of more than 290 reports of subacute thrombosis and 60 deaths associated with use of the CYPHER stent.⁷

The notification became a major news item and prompted a flurry of calls from apprehensive patients to physicians asking what they should do. The physician community was left trying to put the FDA’s concern in context, and patients were left to deal with their anxiety about having a coronary event. It was unclear at the time, and remains unclear, whether this flurry of reported cases represented a true increase in the rate of subacute thrombosis over that seen with bare-metal stents, or a lower threshold for reporting this complication, driven by the high profile of the new device. What was missing was a quantitative estimate of the risk for subacute thrombosis, an estimate that required identifying the exposed population—in this case, the number of patients in whom the device was inserted—and an accurate ascertainment of all patients who subsequently experienced the adverse event.

Could Medicare claims data have helped the FDA? Claims covering the time frame when drug-eluting stents were introduced are not yet available. However, the claims data available in historical files do cover the period when bare-metal stents were in common use. We used these data to estimate the risk of subacute thrombosis following placement of bare-metal stents and compared our results with those in the literature.

Study Data And Methods

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 
The cohort. For this analysis we used a 20 percent national sample of all hospital (Part A) and physician (Part B) claims submitted to the CMS in 2001, a time period when only bare-metal stents were available. Patients undergoing a PCI with stent placement were identified by the presence of a Part B physician claim for stenting with an aligned (based on dates) Part A hospital claim. Using the unique patient identifier, we linked all claims in 2001 for each of these patients to create a patient history.

To reasonably compare our results with those in the literature, we used clinical exclusion criteria similar to those in the Stent Anticoagulation Restenosis Trial Study (STARS).⁸ Patients were excluded if (1) they were admitted with a primary diagnosis of an ST-segment elevation myocardial infarction (MI) on their Part A claim; (2) they were admitted within seven days of discharge from a prior hospitalization; (3) they were within thirty days of coronary artery bypass graft (CABG) surgery or had evidence of bypass graft disease on their index hospitalization (eliminating patients who might have a PCI in a bypass graft); and (4) they were within thirty days of a PCI.

Outcomes. As in STARS, we assumed that any death, MI, or coronary revascularization procedure within thirty days of stenting was secondary to a subacute thrombosis. Data on survival were obtained from the Medicare Denominator File. PCI and CABG were defined based on the presence of a Part B physician claim with an appropriate CPT code. A CABG surgery occurring on the same day as the index PCI was attributed to a technical complication of the procedure and not to subacute thrombosis. PCI as an outcome required a procedure date after that of the index procedure. On the index admission we could not distinguish if the coding of MI represented the indication for the procedure or an outcome of the procedure; the former is much more common than the latter. Therefore, MI as an adverse event was counted only if there was a second admission during the thirty days of observation with an ICD-9-CM code for MI. The combined endpoint of major adverse cardiac events was the union of death, revascularization, and MI.

Study Results

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 
We identified 23,094 patients meeting study criteria between 1 February and 31 November 2001.⁹ Thirty-day outcomes are shown in Exhibit 1, along with comparisons with results reported in recent literature. In a relatively contemporaneous group of randomized controlled trials (RCTs) of bare-metal stents (Burzotta), the pooled combined endpoint at thirty days was 3.54 percent (compared with 7.30 in our sample). In a meta-analysis of three RCTs and seven single-center registries (Bhatt), the rate for the combined endpoint ranged from 2.10 percent to 4.00 percent. In a series of consecutive patients from one high-volume European center (Lemos), the combined endpoint at thirty days was 4.2 percent.

Limitations Of Medicare Claims Data For Postmarketing Surveillance

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 
Identifying the device. We were able to identify patients as recipients of only a bare-metal stent because in 2001 a drug-eluting stent had yet to be approved by the FDA. Using 2003 data, we will be able to identify patients receiving only a drug-eluting stent because a hospitalization for drug-eluting stent has a diagnosis-related group (DRG) that is different from that for a bare-metal stent, not because drug-eluting stents were assigned a unique CPT code. In neither case can we identify stents made by a particular manufacturer. Claims-based postmarketing surveillance would benefit greatly from the timely availability of more specific codes.

Identifying the patients. We used an algorithm based on medical codes, procedure codes, and dates (as described in our methods section) to attempt to identify patients who met study criteria. How well this algorithm performed is unknown. In addition, our intention was to compare the experience bare-metal stents with that of drug-eluting stents. This immediately raises the question of whether the patient populations will be comparable. Also, the claims databases do not contain some clinically relevant data. In our example, we had no information on the details of the coronary anatomy, which is known to be linked to procedural success or failure. On the other hand, there is evidence that major comorbidities (such as congestive heart failure, diabetes mellitus, and peripheral vascular disease) are coded nearly as well in administrative data as in clinical data sets.¹⁰ This same study also showed that when one looks across multiple hospitalizations, the prevalence of other comorbidities more closely approximates the clinical data. To what extent uncertainty about case-mix will compromise an analysis remains to be determined.

Describing the procedure. Claims data contain no specific procedural information. In our example, we do not know where stents were placed, how many stents were placed, stent size, or what adjunctive therapy was given. These procedural practices are linked to outcome as well and could affect our comparison of subacute thrombosis following drug-eluting versus bare-metal stents. Although not the case in our example, in some circumstances (such as hip replacement) it will be important to know on what side of the body a device was placed.

Describing outcomes and defining causality. Cardiac catheterization—not death, MI, or repeat revascularization—is the gold standard for identifying subacute thrombosis, and catheterization data are not part of the claims databases. However, even clinical studies do not necessarily require angiography when assessing for subacute thrombosis. Just as we did, STARS was willing to label any death within thirty days of stenting and any MI without an associated catheterization as suggestive enough of subacute thrombosis. That being said, most patients in STARS suspected of having such a thrombosis did undergo catheterization. Access to clinical data also allowed the STARS investigators to distinguish a second PCI on a different lesion from a subacute thrombosis—something we could not do with claims data. The absence of catheterization data is associated with an overestimation of the true rate of subacute thrombosis. As mentioned in the methods section, we could not identify patients who had such a thrombosis on the index admission, were not revascularized for some reason, and had an MI, because we could not distinguish MI as the indication for the procedure (an exclusion criterion) from MI as a procedural outcome. Lastly, a claims-based study cannot establish causality and thus cannot determine what patient or provider characteristics, procedural practices, or process variables were responsible for the observed rate of subacute thrombosis. As we discuss below, many of these concerns become nonissues if the claims data are selectively supplemented with clinical information.

Other populations. An obvious limitation of any analysis using Medicare data for postmarketing surveillance is that surveillance is limited to patients age sixty-five and older. Although the point estimate for subacute thrombosis might increase with age, we think it unlikely that the relative comparison of event rates for bare-metal versus drug-eluting stents will change much with age, which makes the results generalizable. For postmarketing surveillance of devices used largely in younger populations (such as breast implants for augmentation), other administrative data sets maintained by third-party payers would be needed.

Despite these limitations, our estimate of the thirty-day risk of subacute thrombosis with the use of bare-metal stents is an outcome measure that should be reproducible year to year, as it depends on hard endpoints that are well recorded in the Medicare claims data. If these data had been available in "real time," the FDA might have been able to determine whether the flurry of reports of subacute thrombosis following the release of drug-eluting stents represented a reporting bias or a real increase in the incidence of this serious adverse outcome.

Improving The Usefulness Of Medicare Claims Data For Postmarketing Surveillance

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 
In light of the preceding discussion, we have several recommendations for improving the usefulness of Medicare claims data for this purpose.

Coordinate release of codes with procedures’ first use. Federal agencies and the American Medical Association (AMA) should make it a priority to assign specific CPT and ICD-9 codes before a device is approved for general use. This will permit tracking of the device from its first approved use. The need for device-specific codes for new technologies was recently recognized by the AMA’s CPT Editorial Panel with the creation of Category III codes—temporary sets of tracking codes for new and emerging technologies.¹¹

Allow chart review. Most of the limitations we have mentioned could be addressed by medical record reviews. In many instances, chart review could be selective. For example, in our study of subacute thrombosis, the charts of only those patients identified as having an adverse event—including those patients with a secondary diagnosis of MI on the index admission—could be reviewed to validate the outcome as consistent with subacute thrombosis. Any concerns about patient selection, case-mix, or severity of illness could be addressed by reviewing an appropriate sample of charts. In some instances, it could be necessary to supplement the claims data with clinical information on the entire study cohort. This would be the case if we wished to compare subacute thrombosis for bare-metal stents made by different manufacturers, because there are no codes to distinguish among them. It would also be the case if one needed to distinguish on what side of the body a device was placed, as for a total hip replacement. Certainly, for these examples and possibly for much of postmarketing surveillance, only a limited amount of information would need to be abstracted from each chart.

Speed up release of data. A major limitation of using Medicare claims data for postmarketing surveillance is the gap between the date when the device is inserted and the date when Medicare claims become available for analysis. Typically, the CMS makes the claims files available only after a delay of eighteen to twenty-four months. The delay is to assure that all bills have been submitted and adjudicated, so that the file, when released, is a complete record of CMS use and spending for a given period of time. However, for hospitalizations and for expensive procedures, the billing cycle itself is thought to be much quicker—perhaps no more than four to six weeks after the date of service. For postmarketing surveillance to work, the CMS will need to make the data available in as close to real time as is practical.

A Proposal For A Pilot Test

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 
We believe that the value of Medicare claims for postmarketing surveillance is promising enough to warrant an evaluation of its usefulness by the FDA and CMS. The first task would be for the FDA and interested clinicians to develop a list of medical devices that would comprise the "study population." In collaboration with industry representatives and clinical experts, we suggest that the FDA identify the outcomes of interest for each device. In as near as possible to real time, the CMS would provide all Parts A and B claims to the FDA for people whose claims history indicates that they have received one or more of the listed devices. Working with analysts experienced in the use of claims data for cohort studies, the FDA could identify diagnoses and procedure codes that have a high probability of defining an adverse event and then estimate the incidence of these events. These estimates would then be validated by chart review. The CMS’s Quality Improvement Organizations (QIOs) provide a ready-made vehicle for this task. QIOs are experienced in conducting sophisticated chart reviews and have privileged access to medical records of individual Medicare enrollees. Also, this investigation of possible adverse events would seem to be in keeping with QIOs’ quality improvement mandate. We therefore suggest that the FDA contract with the CMS’s QIO program to undertake chart review to validate the coding of adverse events and to obtain additional information that might be required to more fully understand the association between device implantation and outcomes. The accuracy and efficiency of this method of identifying adverse events and obtaining estimates of their incidence could then be compared with that gathered from other sources.

Postmarketing surveillance is essential to better understanding the risks associated with new medical technology. However, it is difficult to consistently obtain accurate event rates, and there is no one method that accomplishes this task best—it requires a variety of approaches. We believe that Medicare claims data have untapped potential for contributing important information on device-related risks and should be put to the test through a unique collaboration of the FDA, the CMS, industry, and academe. We have proposed a process for moving forward with this novel approach to postmarketing surveillance and would hope that in this era of heightened public and congressional concerns about the safety of new drugs and devices, both regulatory agencies and industry would be willing to become active participants.

Editor's Notes

 
David Malenka (David.Malenka{at}Hitchcock.org) is an associate professor of medicine and of community and family medicine in the Section of Cardiology, Dartmouth-Hitchcock Medical Center, in Lebanon, New Hampshire. Aaron Kaplan is an associate professor of medicine there. Sandra Sharp is a research associate in Community and Family Medicine, Dartmouth Medical School, in Hanover, New Hampshire. John Wennberg is the Peggy Y. Thomson Professor for Evaluative Clinical Sciences and a professor of medicine and of community and family medicine at the Center for the Evaluative Clinical Sciences, Dartmouth Medical School.

NOTES

Top Study Data And Methods Study Results Limitations Of Medicare Claims... Improving The Usefulness Of... A Proposal For A... NOTES

 

1. L.H. Monsein, "Primer on Medical Device Regulation, Part II: Regulation of Medical Devices by the U.S. Food and Drug Administration," Radiology 205, no. 1 (1997): 10–18[Free Full Text]; and D.W. Feigal, S.N. Gardner, and M. McClellan, "Ensuring Safe and Effective Medical Devices," New England of Journal of Medicine 348, no. 3 (2003): 191–192.
2. As early as 1973, the U.S. Food and Drug Administration (FDA) requested voluntary reporting by health care professionals of device-related adverse events, a program that continues today as MedWatch. The Medical Device Regulation of 1984 (MDR, Federal Register 49, no. 36325, 14 September 1984) and the Safe Medical Devices Act of 1990 (P.L. 101-629) required the FDA to mandate universal reporting by facilities and distributors of serious adverse events associated with high-risk devices. The Medical Device Amendments of 1992 (P.L. 102-300) and the FDA Modernization Act of 1997 (P.L. 105-115) greatly modified these statutory requirements for postmarketing surveillance, repealing the mandate for such surveillance by distributors and requiring the FDA to develop a system for universal reporting among a representative subset of facilities. This has led to the formation of the Medical Product Surveillance Network (MedSun), which expects to be operating by 2005. As part of premarketing approval or Section 522 of the FDA Modernization Act of 1997, or both, the FDA still mandates postmarketing surveillance of high-risk devices when deemed appropriate. For a detailed description of the FDA’s role in insuring the safety of medical devices, see T.P. Gross and L.G. Kessler, "Surveillance for Medical Devices," in Pharmacovigilance, ed. R.D. Mann and E.B. Andrews (Indianapolis: John Wiley and Sons, 2002), 411–422.
3. Task Force on Risk Management, Managing Risks from Medical Product Use: Creating a Risk Management Framework, Report to the FDA Commissioner (Rockville, Md.: U.S. Department of Health and Human Services, Food and Drug Administration, May 1999).
4. ICD-9-CM: International Classification of Diseases, Ninth Revision, Clinical Modification (Los Angeles: Practice Management Information Corporation, 1999).
5. American Medical Association, Current Procedural Terminology: CPT 2001 (Chicago: AMA Press, 2001).
6. Subacute thrombosis refers to the formation of a blood clot within the stent that inhibits the flow of blood through the stent and, if not removed, usually results in a myocardial infarction (MI). It is termed "subacute" because it occurs after the stent has been successfully deployed and the patient has left the catheterization laboratory. The risk of subacute thrombosis is increased if for technical reasons the stent is incompletely opposed to the vessel wall. It is also increased for a period of days to weeks following stent deployment while a layer of cells, which normally lines the inside of blood vessels and inhibits clot formation, slowly covers the newly placed stent. The presence of a subacute thrombosis can be documented only with the use of coronary angiography, which visualizes the total occlusion of the blood vessel at the site of the stent. However, it can be inferred when a patient who recently received a stent presents with typical symptoms, electrocardiographic changes, and laboratory findings of MI, most often prompting a cardiac catheterization confirming the thrombosis and a repeat PCI to reopen the vessel. A sudden death within thirty days of stent placement is usually attributed to MI from subacute thrombosis associated with a life-threatening arrhythmia. Based on studies in which all patients underwent cardiac catheterization thirty days after receiving a stent, it is known that there is a rare patient with subacute thrombosis who has minimal symptoms and does not have MI.
7. FDA, "FDA Public Health Web Notification: Information for Physicians on Sub-Acute Thromboses (SAT) and Hypersensitivity Reactions with Use of the Cordis CYPHER Coronary Stent," 29 October 2003, www.fda.gov/cdrh/safety/cypher.html (18 April 2005).
8. We used exclusion criteria based on the Stent Anticoagulation Restenosis Trial Study (STARS). See M.B. Leon et al., "A Clinical Trial Comparing Three Antithrombotic-Drug Regimens after Coronary-Artery Stenting: Stent Anticoagulation Restenosis Study Investigators," New England Journal of Medicine 339, no. 23 (1998): 1665–1617.[Abstract/Free Full Text]
9. The study starts on 1 February because a thirty-day period of observation is required to eliminate patients with a recent revascularization. It ends 30 November to ensure a full thirty days of observation for the last enrolled patient.
10. P.S. Romano et al., "A Comparison of Administrative versus Clinical Data: Coronary Artery Bypass Surgery as an Example, Ischemic Heart Disease Patient Outcomes Research Team," Journal of Clinical Epidemiology 47, no. 3 (1994): 249–260.[CrossRef][Web of Science][Medline]
11. M. Beebe, "CPT Category III Codes Cover New, Emerging Technologies," Journal of the American Health Information Management Association 74, no. 9 (2003): 82–84, 86, 88.[Medline]

                      What's this?