Existing Data Resources for Assessing
Quality of Care

other type of health care facility, to home with home health care services, discharged against medical advice, or died in the hospital). Data problems include coding errors and missing data.15

A study of Medicare discharges in California found that 23 percent of patient deaths within 20 days of discharge (as recorded on state vital statistics records) had been incorrectly coded on hospital discharge abstracts. Of these, a large proportion were reported as discharged alive when they had in fact died in the hospital. These problems have been confirmed by others. 16 HCFA's Bureau of Program Operations has initiated a study of the accuracy of mortality data, but results are not yet available.

Errors in the coding of inpatient deaths may have important implications both for the accuracy of some DRG coding and for the validity of PRO hospital and provider profiling efforts. Erroneous coding of discharge disposition can lead to incorrect DRG assignment. In cases of myocardial infarction, for example, patients discharged alive are assigned to DRG 121 or 122, which are reimbursed at a higher rate than the DRG for myocardial infarctions resulting in inpatient death (DRG 123). More important, erroneous coding may undermine the validity of the profiling activities that PROS use to help target their review activities. If the misreporting of inpatient mortality is random, it would not seriously affect the validity of analyses designed to identify outliers. But if particular hospitals seriously and routinely underreport inpatient mortality, they might not be identified as outliers when they should be so identified. Conversely, hospitals that report mortality data accurately could appear to have relatively high rates.

To some extent, the inclusion of date of death from Social Security Administration files on inpatient billing records could mitigate this problem. However, research in progress has suggested that this information is also problematic.

15GAO, Information Requirements for Evaluating the Impact of the Medicare Prospective Payment System on Long-term Care Services, GAO/PEMD-85-8 (Washington, D.C.: February 21, 1985), p. 6; HHS, Report to Congress: The Impact Of the Medicare Hospital Prospective Payment System, 1984 Annual Report (Washington, D.C.: November 1985), pp. 8-15.

16California Medical Review, Inc. Final Report, Premature Discharge Study, December 10, 1986. HCFA contract no. HCFA-500-87-0535-0032; M.S. Blumberg, "Comments on HCFA Hospital Death Rate Statistical Outliers," Health Services Research 21(6), pp. 716-8.

Existing Data Resources for Assessing
Quality of Care

Organization of the Data

Many analyses of quality of care require linking Medicare claims data
over time and across covered services. For example, some quality
review methods link diagnostic or procedure data from physician visits
with inpatient information to detect possible surgical complications. The
absence of linked data precludes this possibility. PROS' review of the
appropriateness of hospital readmissions could be strengthened if they
had access to information on the use of posthospital and ambulatory
services. Currently, beneficiary and provider identification numbers can
be used to link data from different Medicare files, but the size and struc-
ture of the files makes this very difficult. Manipulating both part A and
part B files to obtain a full picture of patients' use of various services
over time and comparing patterns of service use to care outcomes has
been particularly problematic. The creation of the MADRS file represents
one effort to provide this type of information for research and policy
analysis.17

HCFA is currently developing two major and closely related projects that will restructure the entire Medicare administrative data system. The first, the combined A/B working file project, is already well underway. It is designed to create an online beneficiary data file that contains all entitlement and utilization information in one file. This will allow contractors to have immediate access to information they need to process bills and will expand the possibilities for establishing prepayment screens. The system is also intended to provide a mechanism for future system enhancements, including the integration of PROS into HCFA's evolving on-line data system.

The second, even more ambitious undertaking is HCFA's Project to Redesign Information Systems Management (PRISM). This will include the design and installation of an entirely redesigned, integrated online Medicare and Medicaid administrative data system accessible to carriers, intermediaries, and PROS, as well as to HCFA managers and analysts.

A $9.28 million appropriation was requested for this project for fiscal year 1988. The analysis and system design phase of the project is now underway; the project is currently scheduled for completion in 1990. The statement of work calls for an interactive system including data required for the processing of queries regarding enrollment, eligibility, and so forth, plus a limited volume of part A bills, and a limited volume

17HCFA plans to make the MADRS file available to PROs in the next contract cycle. The Pennsylvania PRO is currently using MADRS to identify skilled nursing facility, home health, and hospital outpatient care for reviews of selected readmissions cases. (See appendix VII.)

Existing Data Resources for Assessing
Quality of Care

of part B bills. The statistical and decision support systems are to be designed so that users have access to both online and historical files.

Conclusions and
Recommendations

The data files maintained at HCFA contain information that could be used
to measure quality of care. It is possible to link provider-level and
patient-level files to create data bases to examine patterns of service
across providers, regions, types of service, and so forth. There are basic
problems, however, with the completeness of the information reported
to HCFA, with systems for verifying the accuracy of these data, and with
the way the data are reported and maintained. Basic information on
diagnoses is not included in HCFA part B claims data. Information that is
needed to analyze patient care outcomes, including inpatient mortality
and discharge destination, is rarely verified. The information obtained
from quality monitoring activities is usually not reported to HCFA in a
form that could be used to develop national estimates of the incidence or
distribution of quality of care problems. If these problems were cor-
rected, Medicare's capacity to measure and monitor quality of care
would be improved.

We believe that adding diagnostic information to part B physician claims
data would increase the usefulness of these files for monitoring care,
both in ambulatory settings and over episodes of illness.18 Diagnostic
information could be used to identify post-surgical complications and
aberrant patterns of posthospital care associated with particular ill-
nesses, as well as possible instances of inappropriate services before
hospitalization. Adding diagnostic information would not require chang-
ing current bill forms, nor would computer layouts for reporting part B
data need to be revised. This information is already required for part B
claims submitted electronically. The diagnostic coding system currently
in use for inpatient services, ICD-9-CM, is already being used by carriers in
their medical review of physician claims. The primary expense would be
costs of training additional people to use the coding system and keying
in the data. As we note in appendix IV, implementing this recommenda-
tion would involve no costs at all for some carriers. For others, the train-
ing and computer costs would be considerable. Savings derived from the
improved efficiency of prepayment and postpayment screens might off-
set the costs of implementing this recommendation, especially over the
longer term.

18This position has also been taken by the National Committee on Vital and Health Statistics in a report on Statistical Aspects of Physician Payment Systems, issued in July 1986.

Existing Data Resources for Assessing
Quality of Care

Recommendation

The introduction of systems for routinely coding and reporting part B
diagnostic information could be coordinated with planned changes in
contractor data reporting associated with the new PRISM system. Train-
ing costs might be reduced by implementing both simultaneously. Physi-
cians would have to supply diagnostic information on ambulatory
claims, whether they accept assignment or not. Further, accuracy would
have to be monitored through profiling and routine checks performed in
the course of claims review (see below).

Requiring that physicians submit diagnostic information on claims or take responsibility for providing this information to patients would require changing current regulations governing the submission of part B claims, and could require new legislation.

We recommend that the Secretary of HHS direct the Administrator of HCFA to require that physicians include written descriptions and ICD-9-CM diagnosis codes on part B claims, that this information be included in the carrier claims processing system, and that it be included in the B-MAD files and MADRS files.

A moderate level of error in Medicare data elements, particularly if randomly distributed, might not greatly impede analyses of quality of care. Of greater concern is the potential for systematic error in data elements such as diagnostic and procedure coding or mortality data (for example, systematic underreporting of in-hospital deaths). Errors in coding discharge destination and type or source of admission, information that could be useful in characterizing patients' condition at admission or discharge, also limit the usefulness and integrity of quality assessment efforts based on administrative data.

The current review systems involve extensive examination of both billing and medical record information; the resources required to document data errors uncovered in the course of claims processing and medical review would involve only marginal cost increases. Knowing the type and extent of such errors would be an important first step toward developing data quality standards that could be required of practitioners, providers, or suppliers submitting Medicare claims. HCFA could then monitor problems with these data (in the same way it currently monitors errors in data used to determine payment) and require some minimum level of compliance with reporting standards.

Existing Data Resources for Assessing
Quality of Care

Recommendation

We recommend that the Secretary of HHS direct the Administrator of HCFA to require PROS, intermediaries, and carriers to routinely document and report incidents in which key data elements required for monitoring the quality of care are inaccurate. In particular, errors in mortality (date of death) and discharge destination, as well as all diagnostic and procedure data (not limited to DRG assignment) should be monitored. Tracking errors in the source and type of admission data fields should also be considered. 19

Neither the existing central HCFA data systems nor the information generated by the medical peer review systems currently provide nationally representative or generalizable information on quality of care problems in the United States.

As discussed in chapter 4, the SuperPRO evaluation of PROS has been designed to validate each PRO's utilization and quality of care determinations. The SuperPRO re-reviews a representative sample of each PRO'S cases, applying the HCFA generic quality screens as well as the criteria developed by each PRO. But each PRO Selects cases based on its own quality objectives, a random sample of cases, and other cases where HCFA requires review. Because the reasons for including a case in the review have not been recorded by the SuperPRO and the data for each PRO are not comparable, the cases cannot be aggregated into a national sample. Identifying the randomly selected cases in the SuperPRO sample would allow analysts to create, using appropriate sampling weights, a nationally representative sample of PRO cases. This would, in turn, allow for the generation of nationally representative information on the types and distribution of quality problems identified by the generic screens. It would also allow for the creation of comparable national data for each SuperPRO review cycle, and thus allow estimates of changes in quality problems over time.

Information on the reason that a case was selected for review is readily available and could be added to SuperPRO files easily and cheaply. PROS use this information when reporting their review findings to HCFA, and many PROS currently include it in the abstracts attached to each case sent to the SuperPRO for review.

19

9As part of a related evaluation, we are currently examining the need for a nationally representative study of medical records to determine the nature and extent of missing and inaccurate data in the Medicare statistical system. Such a study would provide a systematic, definitive assessment of current levels of data accuracy. What is contemplated here is simply tracking blatant data errors and holding providers accountable for improvement.