Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 5
Chapter 2
STUDY METHODS
In the Institute of Medicine's study of the reliability of the six
items selected frog the Medicare record, a field term independently
abstracted selected patient records within participating study hos-
pitals. The results of the independent abstracting were compared
with Medicare data on the same hospitalizations obtained from the HCFA.
(The Medicare information had been compiled by HCFA frog claim forms
submitted by the study hospitals to the fiscal intermediary for payment.)
Discrepancies between the Institute of Medicine (IOM) abstract and the
Medicare record were noted, and the patient records were re-examined
in an attempt to understand the reasons for discrepancies. For most
records with discrepancies, an attempt was made to locate the hospital
copy of the appropriate claim form (Medicare form number 1453) and
transfer the claims information to the abstract form. These data,
supplemented by information on hospital procedures for processing
Medicare claims, constituted the basis for the analysis.
The sampling plan, research instruments, field work, data processing,
and analytic techniques developed for the previous re-abstracting
study are generally applicable here. Unique methodological aspects of
the current study are discussed below.
SAMPLING PLAN
As before, a three-stage sampling plan was used. The plan included an
initial national sample of hospitals suitable for use in the larger
PSRO evaluation; a smaller subsample of hospitals included in this study;
and within each study hospital, aa sample of Medicare records that were
used in the abstracting.
Initial National Sample of Hospitals
The Institute of Medicine had previously drawn a national probability
sample of short-te'-m general hospitals using a two-way controlled selection
5
OCR for page 6
6
process . t 1 ] The sample consisted of ten panels of hospitals with
less than 1,000 beds and an additional group of hospitals with 1,000 or
more beds, which were included in the sample with certainty. Each panel
constitutes a national sample in its own right and can be combined with
other panels, plus the certainty hospitals, to create a range of repre-
sentations of the national hospital universe, depending on the sample
size and level of precision desired.
Two of the ten panels were randomly selected and combined with the
certainty hospitals to serve as the sampling frame for the previous
re-abstracting steady. Therefore, eight panels were available for in-
clusion in this study. One panel was chosen at random and, together
with a subsample of certainty hospitals, provided 303 hospitals that
served as the sampling frame for this study.
Subsample of Hospitals for Abstracting
_
Within the sampling frame, a final sample of 125 hospitals was drawn,
using an approximation to controlled selection. Although the initial
goal of the study was to include seventy to seventy-five hospitals,
oversampling was necessary because of the limited time for follow-up
contacts and an expected high refusal rate. A letter to administrators
of sampled hospitals requesting their participation is in Appendix B.
Eighty-four of the 125 hospitals (65 percent) agreed to participate.
This number was reduced to seventy-two, using the stratification vari-
ables, in order to stay within the budget. The hospitals that finally
were asked to participate had similar characteristics and were distrib-
uted in proportion to those in the original sample of 125. One hospital
in the final sample was excluded because it was not possible to retrieve
the necessary medical records with the available identifying information.
Therefore, a total of seventy-one hospitals were included in the study.
The hospitals declining to participate did not appear to differ from
the participants in any systematic manner. The basic hospital weights
were adjusted to reflect the reduced sample size.
Sample of Medicare Records
The sample design for selecting records to be independently abstracted
was guided by several considerations. Because t'ne study was intended,
at least in part, to determine the usefulness of Medicare records for
evaluating the effects of PSROs, the diagnoses were selected to conform
R. Goodman and L. Kish, .'Controlled Selection - A Technique in Prob-
ability S~mpling'.8 Journal of the American Statistical Association,
45 (September 1950~: 350-72; see also Irene Hess, Donald C. Riedel,
and Thomas B. Fitzpatrick, Probability Sampling of Hospitals and
Patients, 2nd ed. (Ann Arbor: Health Administration Press, 1975J.
OCR for page 7
7
with those specified in the PSRO Evaluation Plan as appropriate for de-
termining PSRO impact on utilization of hospital services.~2] This had
the added advantage of facilitating comparisons between the results of
this study with the previous re-abstracting study. However, some diag-
noses from the previous study were not appropriate for a Medicare pop-
ulation and were excluded--for example, hypertrophy of tonsils and
adenoids. Other diagnoses especially important for a Medicare popula-
tion were included, even though they had not been used in the previous
study--for example, diseases of the prostate.
In addition to examining specific diagnoses, it is sometimes useful to
aggregate information to reflect broader, homogeneous groupings of
patients with generally similar conditions. In this regard, HGFA's
Office of Policy, Planning and Research has analyzed utilization data
using AUTOGRP categories as a research tool.~3] Therefore, an addi-
tional sampling consideration was the desirability of analyzing the
reliability of data by AUTOGRP categories, as well as by specific
diagnoses.
A sampling plan was developed that concentrated on the fifteen AUTOGRP
Diagnosis Related Groups (DRGs) for conditions most frequently occur-
ring in the Medicare population. Most DRGs were divided into specific
and residual sub-groups. Specific diagnostic sub-groups include diag-
noses identified for special study because of their importance for the
Medicare population and/or their inclusion in the previous study. They
were sampled at a greater frequency than the residual diagnostic sub-
groups, which include everything in the DRGs except the specific diag-
noses. As an example, cataract is a specific diagnosis from the DRG
o f "Di seases of the Eye'' and was sampled with a higher frequency than
the residual sub-group within that DRG. In some cases, the DRG con-
tained only the specific diagnosis - for example, diabetes or cerebro-
vascular diseases - and there was no residual sub-group. Where there
were residuals, they often included diagnoses referred to as "satel-
lites" in the previous study. "Satellite" diagnoses are frequently and
erroneously coded as principal in place of the specific diagnosis with
which they are associated. This is another feature of the sample design
that facilitated comparisons between this study and its predecessor.
U.S. Department of Health, Education, and Welfare, Office of the Assistant
Secretary for Health, Office of Professional Standards Review, Program
Evaluation Plan: Professional Standards Review Organizations, by Martin A.
Baum et al. (22 September 1975), pp. 119-21.
3
The AUTOGRP system, developed at Yale University, classifies patients
into homogeneous categories that are clinically and statistically mean-
ingful and reflect similar patterns of hospital resource consumption.
See Ronald Mill, Robert B. Fetter, Donald C. Riedel, and Richard Averill,
"AUTOGRP: An Interactive Computer System for the Analysis of Health
Care Data'" Medical Care 14 (JU1Y 19761: 603-615.
OCR for page 8
8
In order to give some chance of inclusion in the sample to all diag-
noses, a sixteenth DRG was created that encompassed all diagnoses not
included in any of the other fifteen DRGs. This sixteenth category,
plus the residual sub-groups in the other DRGs, permit the calculation
of net and gross difference rates, as was done in the previous study.
The total sample represents the universe of all diagnoses included in
the ICDA-8 classification system, as adapted by HCFA.~4] Figure 1
shows the diagnostic groups included in the sample and their ICDA-8
code numbers. A more comprehensive listing of code numbers is found
in Appendix C.
A computerized sampling procedure was used to select the abstracts for
inclusion in the study. The universe of abstracts eligible for selection
in each hospital was known and used to develop the rates with which each
diagnostic group within each hospital was sampled. All abstracts were
for Medicare beneficiaries age 65 and over, who were discharged from the
study hospitals during calendar year 1974 and included in the twenty per-
cert sample maintained by HCFA. The year 1974 was used because this is
the baseline year for assessing the effects of PSROs.
This procedure was expected to yield a total of 4,908 abstracts from
seventy-one hospitals. In some instances, however, the medical record
was not available in the medical record department, and no substitutions
were made. As a result, 4,745 medical records were actually indepen-
dently abstracted by the IOM field team.
Eighth Revision of International Classification of Diseases, Adapted
for Use by the Social Security Administration, U.S. Department of Health,
Education 9 and Welfare, Social Security Administration, Pubn. No. 23-72'
undated.
OCR for page 9
9
Figure 1. Components of Sample of Medicare Records
-
Diagnosis Sub- ICDA-8 code
related group (DRG) groups numbers*
1. Ischemic heart diseases except
acute myocardial infarction
(AMI)
Specific diagnosis Chronic ischemic 412.0;412.9
heart disease**
Residual diagnosis Subacute ischemic 411.0;411.9
heart disease
(satellite)
Angina pectoris
Asymptomatic
ischemic heart
disease
2. Cerebrovascular Diseases
413.0;413.9
414.0;414.9
Specific diagnosis Cerebrovascular 430.0-438.9
diseases**
Residual diagnosis
3. Fractures
Specific diagnosis
Residual diagnosis
none
Fracture, neck 820.0-820.9
of femur**
Fracture of other 821.0-821.9
and unspecified
parts of femur
(satellite)
Fracture of skull, 800.0-809.9
spine, and trunk
*Discrete and non-continuous codes are listed individually. Where a
diagnosis has a lengthy list of continuous codes, however, only the first
and last codes are given, separated by a dash. A complete listing of
codes for each diagnosis is found in Appendix C. The reader may wish to
refer to the Appendix before considering the influence of coding discre-
pancies in the analysis.
**Indicates specific or 'target" diagnosis used in prior re-abstracting
study.
OCR for page 10
10
Figure 1 ~ cont inued ~
~ ~ . .
Diagnosis Sub- ICDA-8 code
re lated group ~ DRG)
Fracture of upper
limb
Fractures of
lower limb; ex-
cludes fractures,
neck of femur
4. Diseases of the eye
810 . 0-819 . 9
822 .0-829 .9
Speci tic diagnosis Cataract*X 374. 0-374 . 9
Res idual diagnosis Inf laudatory 360 . 0-369 .9
di seases of the
eye
Other diseases 370 . 0-373 . 9;
and conditions 375 . 0-378 . g
0 f the eye;
excludes cataract
and blindnes
5. Acute myocardial infarction
Specific diagnosis Acute myocardial 410.0-410~9
infarction'*
Res idual diagnosis none
6. Hernia of abdominal cavity
Specific diagnosis
Residual diagnosis
Inguinal hernia 550
without ment ion
of ob struct ion^-
Inguinal hernia 552
with ob struct ion
~ satellite)
Other hernia of
abdominal cavi ty
without mention
of ob struct ion
Other hernia of
abdominal cavi ty
wi th ment ion o f
ob s true t ion
551 .0-551 .9
553 . 0-553 .9
OCR for page 11
11
Figure 1 (continued)
Diagnosis
related group (DRG)
Sub-
groups
ICDA-8 code
numbers
7. Diabetes mellitus
Specific diagnosis
Residual diagnosis
8. Diseases of the prostate
Specific diagnosis
Residual diagnosis
9. Pneumonia
Specific diagnosis
Residual diagnosis
10. Diseases of the gall bladder
and bile duct
Specific diagnosis
Residual diagnosis
11. Miscellaneous diseases of
the intestine and peritoneum
Specific diagnosis
Diabetes**
mellitus
none
Hyperplasia
of prostate
250.0-250.9
600
Prostatitis.and 601;602
other diseases
of the prostate
Bronchopneumonia 485 ;486
organism not
specified and
pneumonia organism
and type not
specified
Pneumonia,
type and organism
specified
480 ;481;
482 .0-482 .9;
483; 484
Cholelithiasis 574. 0-574 . 9;
and cholecystitis** 575
Other diseases
of the gall
bladder and
bile duct
Intestinal
obstruction
without mention
of hernia
576. 0-576 .9
560 .0-560 .9
OCR for page 12
12
Figure 1 (continued)
Diagnosis
related group (DRG?
ICDA-8 code
groups
numbers
Residual diagnosis
12. Heart failure
Specific diagnosis
Residual diagnosis
13. Enteritis, diverticula, and
functional disorders of
intestine
Specific diagnosis
Residual diagnosis
14.
Bronchitis
Specific diagnosis
Residual diagnosis
Peritonitis,
peritoneal
adhesions, and
other diseases
of intestine
and peritoneum
Congestive
heart failure
and left
ventricular
failure
Acute heart
failure, undefined
Diverticulosis
of intestine
Noninfectious 5 61
gastroenteritis
and colitis,
except ulcerative
Chronic enteritis
and ulcerative
colitis
Functional
disorders of
intestine
Bronchitis
none
567 .0-567 .9;
568; 569 . 0-
569.9
427.0 j427.1
782 .4
562.0 ; 562.1
563. 0-563 .9
564. 0-564 .9
466;490;
491
o
OCR for page 13
13
Figure 1 (continued)
.
. .
Diagnosis Sub- ICDA-8 code
related group (DRG) groups numbers
15. Malignant neoplasm of the
respiratory system
Specific diagnosis
Residual diagnosis
16. All else
Specific diagnosis
Residual diagnosis
Malignant neo- 162.1
plasm of
bronchus and
lung
Primary malignant 160.0-160.9;
neoplasm of respi- 161.0-161.9;
ratory system 162.0;
except of bran- 163.0-163.9
chus and lung
all else
none
-
*Discrete and non-continuous codes are listed individually. Where a
diagnosis has a lengthy list of continuous codes, however, only the first
and last codes are given, separated by a dash. A complete listing of
codes for each diagnosis is found in Appendix C. The reader may wish to
refer to the Appendix before considering the influence of coding discre-
pancies in the analysis.
**Indicates specific or "target" diagnosis used in prior re-abstracting
study.
Retrieval of Medicare Claim Forms
As noted earlier, the field team was instructed to consult the hospi-
tal's copy of the Medicare claim form when a discrepancy was found
between the IOM abstract and the Medicare record that was not attrib-
utable to the field work. (Mistakes by the field team were expected
to result in a determination that the Medicare record was correct, so
there would be no need to consult the claim form.)
Retrieving claim forms was difficult because of the lack of uniformity
in filing procedures. Some are filed according to the date on which
the account is closed; others are physically removed from the hospital.
Some hospitals had not retained the form. Since there were 2,878 IOM
abstracts with at least one discrepancy not attributable to the field
work, the same number of claim forms should have been available and ex-
amined. However, only 2,377, or 82.6 percent of those needed, could
be retrieved. Ten hospitals were unable to produce any Medicare claim
forms. They were located primarily in Standard Metropolitan Statistical
OCR for page 14
14
Areas, as was the general study population. However, they included
fewer voluntary hospitals, more governmental hospitals, and a dispropor-
tionately lower share of hospitals with 200 to 500 beds than the rest
of the study population. Because of the limited availability of claims
forms, these data were analyzed as a separate data set.
RESEARCH INSTRUMENTS
Two research instruments were developed: the abstracting form was in-
tended to record information obtained in the process of abstracting
selected medical records (see Appendix D); the informal checklist was
intended to describe the flow of claims information within hospitals
and to ascertain the extent to which hospital definitions conform to
those used by the field team and by HCFA (see Appendix E).
Abstracting Form
Information items to be abstracted included date of hospital admission,
date of discharge, sex, admitting diagnosis, principal and other diag-
noses, and principal and other procedures. Additional items appear on
the claim form, such as date of surgery and charges for laboratory and
other hospital services. Because of the complexities and costs of add-
ing more items to the study, however, the information abstracted was
restricted to that mentioned above. Age and/or date of birth was not
abstracted because it is validated by HCFA when a person enrolls for
benefits.
All abstracting was done in accord with item definitions developed
for the Uniform Hospital Discharge Data Set (included in Appendix D),
since PSROs use UHDDS definitions. Furthermore, consensus has been
reached within the Department of Health, Education, and Welfare to
require use of UHDDS in federal reporting systems. Diagnostic coding
was based on ICDA-8 as modified by HCFA.~5] Procedural coding was
based on four-digit Surgical Current Procedural Terminology (CPT),
first edition, as modified by HCFA.~6]
for Us ~ 1th
Education, and Welfare, Social Security Administration, Pubn. No. 23-72,
undated.
6
Numerical Surgical Current P , U.S.
Department of Health, Educa scurry Adminis-
tration, Office of Program Policy and Planning, ORS, Pubn. No. 013-75
(9/75~. Although the title specifically refers to surgical procedures,
the contents include procedures which are not surgical--for example,
transfusion--and which also appear on the Medicare claim form.
Throughout this report the term "procedure" is used, rather than
"surgical procedure," since it is clear that this broader concept is
appropriate.
OCR for page 15
15
Although all diagnoses and procedures for a given hospitalization were
recorded on the abstract form, only the admitting and principal diag-
noses and principal procedure were coded. Strict guidelines were pro-
vided for designating an admitting diagnosis, relying on only the limited
information available at the time of admission or soon thereafter. This
information was used to explore the possibility that the primary diag-
nosis, contained in the Medicare record, may be in reality an admitting
diagnosis, rather than a more definitive principal diagnosis, because
of the urgency to submit a claim form for reimbursement.
The presence of additional diagnoses was also noted--again, in accord
with specific guidelines. For example, a "history of" a particular
diagnosis would not be included unless it were clinically significant
for the hospitalization under review. A bone fracture with surgery or
other treatment would be included, whereas a slight degree of osteo-
arthrit~s noted only as an observation and for which a patient was not
treated, would not be included. The notation of significant additional
diagnoses was intended to determine whether they are appropriately rec-
orded on the Medicare record. (All diagnoses may be recorded on the
Medicare claim form. The Medicare record, however, includes a code for
the primary diagnosis only; the presence of additional diagnoses is
noted, but they are not coded.) It was hypothesized that the reli-
ability of data for patients with additional diagnoses might be lower
than for those with a single diagnosis only.
There were several steps involved in the abstracting process, each in-
tended to increase understanding of the reasons for discrepancies and
the relative degree of error that would be involved if such data were
used for evaluative purposes. After the field team had independently
abstracted all information items from the hospital medical record, the
results of that process were compared with corresponding information
on the Medicare record. Where the two disagreed, the medical record
was re-examined and an attempt was made to determine the correct data
source, as well as the reason for the discrepancy. In all cases, these
determinations were made before consulting the claim form. The options
available to denote reasons for discrepancy varied, depending on the
data item, as outlined below.
Reasons for discrepancy - date of hospital admission, discharge, and
sex:
.
a
Clerical - discrepancies attributable to mistakes of a
coding clerk, such as obvious transposition of numbers
in the year of admission.
Completeness - discrepancies resulting from an inadequate
review of the medical record. For example, an item may be
missing from the admitting sheet, but clearly stated in the
discharge summary.
Reasons for discrepancy - admitting vs. principal diagnoses (both
determined by field team):
OCR for page 16
16
.
Completeness - discrepancies between the admitting and princi-
pal diagnoses resulting from an incomplete review of those por-
tions of the medical record designated for use in determining
admitting diagnosis.
o Coding refinement - discrepancies resulting from the avail-
ability of additional details when the principal diagnosis
was determined. Although the general diagnostic category
does not change, the final determination is more refined.
For example, the admitting diagnosis may be pneumonia
(486.0), although the principal diagnosis is pneumococcal
pneumonia (481.0~.
Investigation - discrepancies occurring when an admitting
diagnosis (such as headache) is based on preliminary find-
ings or symptoms, although the principal diagnosis (hypo-
glycemia) is based on more complete medical investigation.
Other - all reasons not included in any of the above. For
such cases, the field team member was instructed briefly to
explain the reason for the difference.
Reasons for discrepancy - principal diagnosis and procedure:
These reasons for discrepancy were grouped into two broad categories:
ordering and coding. The possibility of an ordering discrepancy had
to be eliminated before considering a coding discrepancy. In general,
ordering discrepancies stem from uncertainty about whether a diagnosis
or procedure should be regarded as "principal" or '.other," in accord
with UHDDS definitions. The following specific types of ordering
discrepancies were considered:
Ordering: Medicare definition - discrepancies reflecting dif-
ferences between the UHDDS definition and that required for the
Medicare claim form, which should occur only if a hospital con-
sciously and consistently used the Medicare definition in com-
pleting the claim forms in 1974. For example, a patient is ad-
mitted for an open fracture reduction and while on the operating
table suffers an acute myocardial infarction which requires a
three-month hospitalization. Using the UHDDS definition, frac-
ture reduction would be chosen as the principal diagnosis, be-
cause fracture is the diagnosis explaining the cause of admis-
sion. If the definition in the "Medicare Hospital Manual'' for
principal diagnosis is used, however, the diagnosis might be
AMI. The manual states: "the primary diagnosis is the diag-
nosis or the illness or condition which was the primary reason
for the patient's hospitalization." It might be noted that
this Medicare definition is appropriate primarily for patients
included in the twenty percent sample. For other Medicare bene-
ficiaries, the "intermediary may authorize the hospital to use
the 'working' diagnosis rather than the final diagnosis if this
OCR for page 17
17
would reduce delays in the submission of billing forms.''t6] In
hospitals with this authorization, billing office personnel would
have to explicitly determine whether a particular patient was in
the twenty percent sample and report the diagnosis accordingly.
Therefore, this reason for discrepancy was not expected to be
frequently used.
o Ordering: Hospital list - discrepancies stemming from a routine
hospital practice (in 1974) of choosing the first listed diag-
nosis or procedure on the face sheet as principal.
.
Ordering: Completeness - discrepancies caused by selecting a
narrative for principal diagnosis based on an incomplete review
of the medical record. For example, the principal diagnoses re-
corded by TOM and Medicare refer to different diseases, each of
which the patient had during the hospital episode under question.
However, if the chart had been searched more thoroughly, it would
have been clear that one, rather than the other, was the proper
principal diagnosis in accord with UHDDS definitions.
Ordering: Judgment - discrepancies representing an honest dif-
ference of opinion in determining which of several diagnoses is
principal. One example of this might be a record for a patient
with diabetes and glaucoma and insufficient documentation to
decide whether either diagnosis would conform to the UHDDS def-
inition for principal diagnosis. Similarly, a record may in-
dicate carcinoma of several sites and may not be sufficiently
documented to permit a determination of a principal diagnosis
using the UHDDS definition.
Ordering: Other - all reasons not included in any of the above.
The field team was instructed to write a note explaining the
necessity to use this reason.
Ordering: Dependent - this option applied only to discrepancies
on principal procedure and was used if an earlier discrepancy in
the selection of principal diagnosis resulted in a corresponding
or dependent discrepancy in the selection of principal procedure.
After the ordering options were eliminated as possible reasons for dis-
crepancies between the Medicare record and the ION abstract, coding op-
tions could be considered. In some respects, the term "coding" may be
a slight misnomer for this type of discrepancy because the code should
be assigned by HCFA personnel, who have access only to information on
the claim form. A more accurate description of this type of discrepancy
might refer to the specificity of detail in the narrative on which the
code is later based, but this is rather cumbersome. In some hospitals
a code is assigned in place of or in addition to the narrative, in which
Medicare Hospital Manual, U.S. Department of Health, Education, and
Welfare, HIM Pubn. 10 (6-66), Reprint, August 1975, p. 81.1-82.2.
OCR for page 18
18
case the earlier use of '~coding" discrepancies is quite appropriate.
Occasionally fiscal intermediaries also assign codes. Furthermore,
the distinction between ordering and coding discrepancies facilitates
comparison with the prior study.
Therefore, for abstracts where there was a general agreement on what the
principal diagnosis or procedure should be, but not the resultant code,
the following reasons for discrepancy were available:
· Coding: Clerical - discrepancies caused by transposing diag-
nostic code numbers or using non-existent codes.
0 Coding: Completeness - discrepancies caused by including
incomplete narrative information for assigning a code, which
resulted from an incomplete review of the medical record. For
example, the presence of a nine as a fourth digit, indicating
the diagnosis is not otherwise specified, when a more careful
review of the chart would have resulted in a more specific
fourth digit code.
Coding: Procedural - discrepancies caused by routine and
systematic misuse or misunderstanding of the coding system.
For example, reliance on the diagnostic index without ref-
erence to tabular listings or failure to heed inclusion and/
or exclusion advice from the tabular listing.
a Coding: Importance - this option applies only to the coding
of procedures and was used to explain discrepancies caused
by differences of opinion over how significant a procedure
must be to warrant coding. This is particularly important
in hospitals which routinely code all procedures without
regard to the UHDDS definition.
· Coding: Judgment - discrepancies caused by absence of com-
plete word-for-word correspondence between the recording of
the diagnosis or procedure in the medical record and/or claim
and the wording in the coding manuals, which requires relying
on judgment. For example, a diagnosis may be recorded as
"recurrent.`' It is unclear from the record whether '`acute'.
or "chronic" is the more appropriate qualifier, and these
are the only two options available in the coding manual.
· Coding: Other - All reasons not included in any of the above.
This option was expected to be used to explain differences
when the field team coded a procedure, but Medicare coders did
not because the information on the claim forms was illegible
or inappropriate. A note of explanation was required if this
option was selected.
OCR for page 19
19
Reasons for discrepancy - presence of additional diagnoses:
Completeness - discrepancies resulting from an incomplete re-
view of the medical record in hospitals where additional diag-
noses are expected to be noted on Medicare claims forms.
Hospital definition - discrepancies resulting from a hospital
policy whereby only one diagnosis, presumably the principal
one, is routinely entered on the Medicare claim form. By
definition, no additional diagnoses would appear on the Medi-
care record.
Importance - discrepancies resulting from IOM guidelines for
determining additional diagnoses. For example, osteoarthritis
may be listed as a diagnosis in the medical record and the Medi-
care claim form might indicate the presence of an additional
diagnosis. However, if osteoarthritis were noted in the record
only as an observation, then the field team would not regard it
as an additional diagnosis. IOM guidelines define an additional
diagnosis as one whose presence affects treatment methods
or utilization of services. (See Specific Instructions in
Appendix D).
For each discrepancy, the field team was asked to code only one explan-
atory reason. Frequently a subjective assessment was required in order
to determine which reason might be influential. It is particularly dif-
ficult to know whether coding errors should be traced to the hospital
or to HCFA. Therefore, the reliability of these responses may be less
than for the remainder of the data. Nevertheless, the potential value
of the information was judged to outweigh its partially subjective nature.
After all records had been abstracted and the reasons for discrepancy
determined, information from the claim form was abstracted. The diag-
noses and procedures were entered on the abstracting form in the order
in which they were listed on the claim form. The first-listed diagnosis
and related procedure was then coded by the field team since this
procedure is usually followed by Medicare coders. To expedite the
field work, comparisons between the claims data and the Medicare record
were performed by computer at the National Academy of Sciences (NAS),
rather than by the field team.
When the narrative summary of diagnoses and procedures is transferred
frog the claim form into a numerical code by HCFA coders for inclusion
in the twenty percent data base, occasionally special coding guidelines
are used. The field team did not use these guidelines, because of the
potential lessening of reliability that might result if they altered
their usual coding practices. This raises the possibility that dis-
crepancies between the field team's work and the Medicare record may
stem from HCFA procedures, rather than from inaccurate data submitted
by the hospital. Therefore, claims data from all abstracts with dis-
crepancies were coded by senior Registered Record Administrators (RRAs)
OCR for page 20
20
employed by BCFA to determine whether the HCFA codes had been correctly
applied. In the analysis, special attention was given to those cases
in which the IOM abstract and hospital claim agreed, but the Medicare
record was different. This enabled an estimation of the extent to which
HCFA coding accounted for the discrepancy.
Information Checklist
The checklist was developed to describe the flow of information from
the medical record department to the hospital billing office and fiscal
intermediary (see Appendix E). It was not administered as a formal
questionnaire. Instead, it provided an informal guide for the field
team to help them understand hospital procedures and the reasons for
which discrepancies might occur. This information was needed for
several reasons.
During initial meetings with local hospitals and Medicare officials to
determine the feasibility of conducting this study, it became apparent
that the paths by which claims information eventually enters the HCFA
computer vary considerably. In some hospitals, when a patient is dis-
charged the medical record department sends the billing office a dis-
charge list, including information on diagnoses and procedures needed
to complete the claim form. Elsewhere, a copy of the face sheet of the
medical record or discharge summary may be sent to the billing office,
where the necessary information is retrieved. If the principal diag-
nosis is determined by a billing clerk, short and easily spelled diag-
noses may receive preference. HCFA requires that information on diag-
nosis and procedure be provided in a narrative (not coded) form. In
one hospital visited, diagnostic information was transformed to H-ICDA
codes by the medical record department and sent to the billing office,
which then translated that information back to a narrative for forward-
ing to the intermediary. In other instances, the hospital provided
coded information to the intermediary, which then converted the codes
to narrative for transmittal to HCFA. Occasionally, HCFA may receive
claim forms for the twenty percent sample that are already coded.
Nevertheless, Medicare coders are instructed to disregard the codes
and enter their own.
Although the fact of variation was known, the permutations along that
path and the influence on data reliability were not known. It there-
fore became desirable to try to document the variability and use that
information in analyzing the reliability of the Medicare records. Thus,
information was obtained to reflect the source from which the billing
office obtained information on diagnoses and procedures and the train-
ing levels of persons retrieving that information, the number of days
after discharge that diagnostic information was transmitted to the bil-
ling office for entry on the claim form, whether later, more definitive
diagnostic information was forwarded to the billing office and inter-
mediary, whether the billing office received narrative or coded infor-
mation and, if coded, whether it was translated back into a narrative.
OCR for page 21
21
Information on the definition of principal diagnosis and procedure used
by each hospital during 1974 was also obtained. This was needed in order
to determine whether the hospital consciously followed Medicare's defini-
tion, which may have led to discrepancies, or whether the hospital had
developed a unique definition that differed from both UHDDS and Medicare
definitions, which again may have resulted in discrepancies.
FIELD WORK
The field work was conducted by four Registered Record Administrators
(RRAs), recruited because of their extensive experience in diagnostic
coding, research, and administration. They were specially trained for
this study by a consultant under the general guidance of the TOM staff.
Two of the field team members and the technical consultant participated
in the previous re-abstracting study and were able to draw upon that
experience in refining the methods for this study. The consultant re-
viewed HCFA's adaptation of the ICDA-8 classification scheme, which con-
sisted primarily of an expansion of fourth digit codes, and instructed
the field team in its use.
Before the field work began, the supervisor of the medical record de-
partment in each study hospital was asked to locate and have available
the selected medical records. When the field team member arrived, she
first informally discussed the checklist items with the supervisor to
acquaint herself with any unusual hospital procedures that might in-
fluence the abstracts, and immediately began abstracting. ~
The completed abstracts were compared with the Medicare records (con-
tained in a sealed envelope to avoid conditioning the abstracting) and
any differences between the two were noted. If a discrepancy between
the Medicare record and TOM abstract was observed, the patient's medical
record was re-examined to determine what information appeared to be cor-
rect and what factors might account for the discrepancy. For those dis-
crepancies not attributable to the field team member, the claim forms
were reviewed. A roster of all records under study had previously been
sent to the director of the billing department, so that the claims were
available, if needed. Details on the re-abstracting process are found
in Appendix D.
The field team's ability fully to understand the reasons for discrepancy
was somewhat constrained by the break in continuity between the examina-
tion of the medical record and review of the claim form. Ideally, after
abstracting the medical record one would compare the TOM abstract, the
Medicare record, and the claim form by placing them side-by-side and
referring to the medical record in hopes of understanding discrepancies.
During the feasibility phase of the study it became apparent that this
would not be possible. Such a comparison would have required copying
or physically removing either the medical record or the claim form
frog its customary location, which created confidentality problems.
OCR for page 22
22
Furthermore, the difficulty in obtaining claim forms often led to
delay or return visits, which might have required accessing the
medical record twice and, in any case, would have created logistical
problems. Therefore, the method used was regarded as the best com-
promise .
To check the reliability of the field work, a subsample of abstracts
was independently assessed by the consultant who trained the field
team. Comparisons were made between these results and those initially
compiled by the field team. In conducting this work, the consultant
did not know which member of the field team had done the initial ab-
stracting, or whether discrepancies were initially detected. The as-
sessment process and its results are described in Appendix F.
CONFIDENTIALITY
To retrieve the medical records and claim forms within study hospitals,
it was necessary for HCFA to provide the IOM with the health insurance
claim number, name and birth date of the beneficiary, and date of hos-
pital admission and discharge for each hospital episode reviewed. To
review the accuracy of diagnostic and procedural coding, a partial re-
view of the medical record was required. Thus, the protection of in-
dividual privacy became very important, and special measures were re-
quired to assure confidentiality. All information provided by HCFA was
carefully secured. The computer tapes on which the data were compiled
for analytic purposes were stored in the NAS computer center, which is
accessible only to authorized persons. All information which might
identify individuals was removed and destroyed as tape files were
created. The report contains statistical summaries only, which do not
permit the identification of patient, physician, or hospital. These
provisions were carefully delineated in a notice in the Federal Register,
dated October 14, 1976.
DATA PROCESSING
After completion of the field work, the abstracts and Medicare records
were returned to TOM for data processing. Each abstract and informal
checklist was scanned visually, keypunched and verified, and subjected
to computer edits. After the accuracy of the raw data was assured,
weights were added for use in the analysis.
A single composite weight was assigned to each abstract which reflected
its probability of inclusion at each step in the sampling process and
was adjusted to account for the non-participation of hospitals and the
unavailability of medical records. The weights were applied through-
out the analysis to permit generalizing to the broader national universe
of all 1974 hospital discharges for Medicare beneficiaries age 65 and
over and eligible for inclusion in the Medicare utilization data file.
Although analyses are based on weighted data, the unweighted sample
sizes are reported in the tables presented in the following chapter.
Representative terms from entire chapter:
field team
