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 120
16
The Uniform Clinical Data Set
Henry Krakauer
To properly assess the Uniform Clinical Data Set, it must be clearly
understood that it was designed for a very specific purpose, namely, to meet
the operational needs of the Health Care Financing Administration (HCFA)
in assuring, through the Peer Review Organizations (PROs), the quality of
the care that Medicare beneficiaries receive. It will, therefore, satisfy a
limited array of needs for clinical data, but the extent to which it does will
have to be determined empirically, that is, through experience with its use.
PROs exert considerable influence on the practice of medicine through
the financial and disciplinary actions at their disposal. They are authorized to:
· deny reimbursement for inappropriate admissions,
· deny reimbursement for substandard care,
· initiate sanctions by the Inspector General, and
correct aberrant patterns of medical care.
The difficulties inherent in these activities may be illustrated by the
process of denial of reimbursement for care judged to have been substandard.
Table 1 presents a theoretical but reasonable algorithm leading to such a
denial. First, it is necessary to demonstrate that the patient suffered harm,
that is, an adverse outcome. Beyond that, it is necessary to demonstrate
that the adverse outcome was avoidable; that is, it should not have been
predictable with a reasonably high level of probability from the condition of
the patient at admission. Finally, it is necessary to establish that a breach of
protocol occurred, in other words that there was negligence or incompetence,
in order to establish culpability.
120
OCR for page 121
COLLECTION OF PRIMARY DATA
TABLE 1 Problems in Denying Reimbursement for Substandard Care
121
SPECIFIC
Denials for substandard care
Harm death, disability, physiological impairment, Increased intensity or
duration of care
SCREENING
Avoidable physiological condition at admission
Objective standards
CULPABILITY
Breach of protocol negligence, incompetence
THE PRO CASE REVIEW PROCESS
The current process of case review by the PROs begins with a screening
of the medical record to identify instances in which the care is suspicious
enough to merit further expert attention. A case identified in the screening
is then referred to a physician advisor for review. If the system of peer
review is not to be perceived as arbitrary and capricious, it is necessary to
(a) develop and uniformly apply well-defined screening criteria that identify
where there is a reasonable probability of a deficiency and (b) develop and
uniformly apply objective standards that would permit the physician reviewer
to ascertain with a high level of probability that a deficiency in care did
occur.
The best guide in these matters is actual experience. Experience permits
one to judge that an act of omission or commission results in harm reason-
ably often and that it was therefore reasonably likely to have done so in the
case in question. Given the realities of medical practice, that experience
should be passed through the filter of consensus to make it acceptable.
Once this has been accomplished, the devising of consistent screening criteria
and objective standards and their uniform application become straightforward.
A strategy for the efficient accumulation and evaluation of experience in
the Medicare environment is displayed in Table 2. It consists of a sequen-
tial process that begins with assessment of the health of the Medicare ben-
eficiaries and of the time trends and geographic variations therein two
problem-finding tools and proceeds with the assessment of the effectiveness
of interventions, be they medical or administrative, as the problem-solving
step. The final step is feedback of the results of the evaluations, coupled
with disciplinary activities and financial incentives to ensure their proper
and timely use. This approach makes extensive use of observational techniques
to evaluate the natural history of conditions as they are currently being
OCR for page 122
22
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
TABLE 2 Strategy for Improvement of the Effectiveness of Medical
Interventions for Medicare Beneficiaries
Health Care Financing Administration
1. Monitoring time trends
a. population-based
b. medical interventions (feasible with available billing and census data)
2. Analyzing geographic variations
a. population-based
b. medical interventions (feasible with available billing arid census data)
3. Assessing effectiveness of interventions (longitudinal, to develop objective
star~dards)
a. monitoring, as in step 1 above (retrospective, based on billing data)
b. use of data from medical records (retrospective, natural history)
Uniform Clinical Data Set (also for case finding for PRO review
[medicolegal] ~
National Center for Health Services Research, HCFA
c. clinical demonstrations (prospective, natural history, especially for
emerging technologies)
National Institutes of Health, NCHSR, HCFA
d. randomized, controlled clinical trials
Health Resources arid Services Administration, NIH, HCFA
4. Feedback (educational, disciplinary, financial)
treated and to begin identifying which of the available and competing courses
of treatment appear most beneficial. Because the approach is sequential
and begins with an assessment of the universe of patients at risk for a
condition, every successive step will, while decreasing the number of pa-
tients and increasing the detail of the pertinent data (down to the randomized
clinical trial), allow the researcher to generalize the findings to the patient
universe. In addition, each earlier, broader step informs planning for the
subsequent, more specific step in the analytic sequence.
THE UNIFORM CLINICAL DATA SET AND PRO NEEDS
The Uniform Clinical Data Set occupies a specific niche in this process.
It is a tool for extracting data from medical records to permit effective risk
adjustment in assessing the treatment of a given patient.
The composition of the Uniform Clinical Data Set is dictated, as was
indicated above, by two requirements. It must enable the PROs to screen
cases efficiently and uniformly in order to identify those in which the effec-
tiveness of the care delivered was problematic, and it must enable reviewing
physicians to develop objective tools by which to judge the cases. Thus, it
OCR for page 123
COLLECTION OF PRIMARY DATA
123
must satisfy an operational need, screening, and simultaneously must sup-
port an epidemiological activity, the development of objective standards
through evaluation of the effectiveness of medical interventions. The contents
of the data set were defined by a task force of clinicians and research
personnel who had these two objectives clearly in mind.
USING THE UNIFORM CLINICAL DATA SET
The content of the Uniform Clinical Data Set is illustrated by Tables 3
through 7. These tables are excerpted computer screens, or menus, that
prompt the entry of data from the medical record. Table 3 displays the
main menu, which lists the major classes of information that are to be
extracted from the hospitalization record. The order of the screens follows
roughly the order in which the data are likely to be encountered in the
medical record. The hardware and software are, however, so quick that the
order of abstraction is dictated by the order of appearance of data in the
medical record rather than the order presented on the list.
The richness of the data collected is best illustrated with specific examples,
TABLE 3 Excerpt from Uniform Clinical
Data Set Computer Screen: Peer Review
Screens (Main Menu)
A. Sociodemographic data
B. Admission status
Admission medication history
History permanent anatomic changes
History and physical
Laboratory: chemistry/blood gases
Laboratory: hematology/urinalysis
Laboratory: microbiology
J. Laboratory: cytologylhistology
K. Admission diagnostic tests
Endoscopy
M. Operative episodes
N. Treatment interventions
O. Recovery phase
P. Discharge status
Q. Discharge planning
R. HDI master
Type letter corresponding to a menu item:
F3=Set HDI_ID F10=Leave
OCR for page 124
124
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
such as the screens which address the patient's history and physical exami-
nation. These are accessed by striking the letter that labels the "history and
physical" menu entry. The subscreens (Table 4, the first of three "history
and physical" screens) provide a menu from which more specific subscreens
(which refer to organ systems) may be selected and indicate whether data
pertaining to the organ system identified have been entered. When data
have been entered, the flag "F" (false) next to that item changes to "T"
(true), as shown for several entries in Table 4.
At the next (organ) level of data, using the cardiovascular examination as
an example (Table S), the data recorded include specific abnormalities,
findings within normal limits ("normal"), and "other findings". Because
more data may be recorded than can fit on one screen, an additional screen
may be entered by striking "+". The abstractor toggles the F to a T for any
finding by striking the appropriate letter, resulting in the recording of that
datum (for example, for jugular venous distention in Table 5~.
The results of diagnostic tests are also recorded for specific periods dur-
ing the hospitalization. For the laboratory tests (chemistry, hematology,
enzymes, urinalysis, microbiology, and cytology), the worst result obtained
within the first 24 hours of hospitalization is recorded as the "initial" value;
in the case of some enzymes, such as the cardiac enzymes, a window of 48
hours is specified. If no admission data are on the record, preadmission
results obtained within a week before admission are accepted.
TABLE 4 Excerpt from the Uniform Clinical Data Set Computer Screen:
Peer Reviews Screens History and Physical A
Chronic necrologic disease
History of necrologic surgery
Current neuroloic exam findings
Chronic cardiac disease
Chronic vascular disease
History of cardiovascular surgery
Current cardiovascular exam findings
I. Chronic pulmonary disease
History of pulmonary surgery
K. Current pulmonary exam findings
Chronic psychiatric disease
M. Current psychiatric exam findings
N. History of cancer
F
F
F
T
F
T
F
F
F
F
+/- GO TO OTHER HISTORY AND PHYSICAL INFORMATION
(+ = SCREEN B AND - = SCREEN C)
Blank=Criteria F10=Leave
Type letter corresponding to item _
OCR for page 125
COLLECTION OF PRIMARY DATA
125
TABLE 5 Excerpt from the Uniform Clinical Data Set Computer Screen:
Cardiovascular Examination Findings
Enter item letter on 1st line to change or enter T or F beside item
ENTER ITEM LETTER: CHANGE ITEM
ENTER = ITEM BY ITEM, F10 = LEAVE
PRESS "+" FOR OTHER CV EXAM FINDINGS
Item Description Value
A
B
C
D
E
F
G
H
I
K
L
M
N
o
p
Normal
Shock
Pulmonary edema
Peripheral edema
Jugular venous distension T
Tachycardia
Bradycardia
Murmur
Arrhythmia
Cardiomegaly
Gallop rhythm
Peripheral pallor
Bruit
Thrill
Friction rub
Pulse Deficit-peripheral
F
F
F
F
F
F
F
F
F
F
F
Current CV exam Findings
Enter item letter on 1st line to change or enter T or F beside item
-
ENTER ITEM LETTER: CHANGE ITEM
ENTER = ITEM BY ITEM, F10 = LEAVE
PRESS "+" FOR OTHER CV EXAM FINDINGS
Item Description
Value
A Ischemic ulcers F
B Stasis ulcers F
C Venous/varicose ulcer F
D Gangrene F
E Dependent rubor F
F Delayed capillary fill F
G Chest pain (steady) F
H Other findings F
OCR for page 126
26
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
TABLE 6 Excerpt from the Uniform Clinical Data Set Computer Screen:
Peer Reviews Screens-Treatment Interventions
Nonsurgical procedures
A. Blood products
F
T
T
B. Inhalation therapy
C. Professional services
Medication therapy in hospital
D. Prescribed medications
E. Adverse reaction to medications F
F.
T
Delivery systems for medications T
Blank=Criteria F10=Leave Type letter for item desired
Prescribed medication
Use Arrows, Home, End, Pgup, Pgdn and Enter to Choose.
FlOtoCarlcel
Drug name Route Start End
Heparin 2 05/09/88 05/12/88
Theophylline, anhydrous 1 05/09/88 05/15/88
Bactr~m 1 05/12/88 05/15/88
Pepcid 1 05/13/88 05/15/88
Carafate 1 05/14/88 05/16/88
Codeine 1 07/08/88 07/10/88
In addition to the laboratory findings that apply to the period immedi-
ately surrounding the admission, the results of the last test prior to discharge
are recorded (although this may have occurred considerably before discharge)
and, for selected tests, the worst value between the initial and final value
(the "interim" value) are recorded. The date a test was drawn is also
recorded, as well as whether, given the calibration of the equipment on
which the test was performed, the result fell outside the hospital's normal
limits. Although the worst interim result is not usable for epidemiological
analyses, it is used in screening cases by means of the HCFA Generic
Quality Screens and so must be collected.
Table 6 illustrates the kinds of information being collected about treat-
ment, such as nonsurgical procedures and drugs administered during the
hospitalization. The medications data sought include route of administra-
tion (self-administrable, thus not requiring specific skills, or invasive, requiring
specific skills for administration), the start date (the date when the drug was
first given), and the end date (the last date of administration). The entry of
drugs is guided by a dictionary that contains about 6,000 trade and generic
names. This ensures that the drugs recorded are recognizable and associates
OCR for page 127
COLLECTION OF PRIMARY DATA
127
the names with therapeutic categories used by the expert system that as-
sesses the appropriateness of admission and the quality of care.
APPLICATION OF FINDINGS
This expert system represents one of the two applications of the ab-
stracted data. Immediately following abstraction, a sequence of about 3,000
logical rules is applied to the data. These rules embody the criteria currently
used by the PROs to (a) verify that the patient's illness was sufficiently
severe to justify the admission and that services that require the patient to
be hospitalized were in fact rendered (the "admission necessity" algorithms)
and (b) ascertain whether breaches of protocol pertaining to inpatient man-
agement and the discharge of the patient may have occurred (the "generic
quality" and the "discharge" algorithms). The product, placed on the screen
of the computer in about two and a half minutes, is a case summary that
contains the results of the evaluation (Table 7) and an ordered listing of all
the findings abstracted (not shown).
TABLE 7 Excerpt from the Uniform Clinical Data Set: Flag Settings and
Reasonsa
Algorithm
Flags
AD00 Admission necessary PA
CASE FAILS ADMISSION NECESSITY SCREENS. REFER TO PA.
ES00 Elective admission PR
SP Elective admission flag.
ES04 Cardiac revascularization PR
2B INDICATIONS FOR INPATIENT ELECTIVE SURGERY NOT
PRESENT. PA FLAG
ES04 PA
DP01 Ischemic heart disease / chest pain PR
8D APPROPRIATE HOSPITALIZATION AND SERVICES. OK
FLAG
DPO1 OK
OP09 Central nervous system MO
A CASE REQUIRES MONITORING FOR SEVERITY OF ILLNESS
DS01 Discharge status/disposition MO
N17 Surgical patient with final hemoglobin missing or result less than
admission result with difference >= 3 and < 4 grams per
deciliter, discharge pulse > 110
DS01 Discharge status/disposition
N29 No creatinines
MO
aExample for actual hospitalization
OCR for page 128
28
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
The case shown represents an actual hospitalization in which a coronary
angioplasty was performed, ostensibly for a malfunction of a prior coronary
artery bypass graft. The "flag" report states that the admission was elec-
tive, for cardiac revascularization, but that sufficient indications for the
procedure were not present. In fact, the results of cardiac catheterization,
specified in the case summary, included a left ventricular ejection fraction
of 56 percent and a 50 percent stenosis of the right coronary. To justify
revascularization, the algorithms require at least 70 percent stenosis. Con-
sequently, the case is to be referred to a physician advisor (PA) for further
review of the necessity for the admission.
The result of the surgical procedure algorithm (those labeled ES) results,
in turn, in the summary recommendation (AD00) on admission necessity
because elective surgery was performed. Had the admission not been elec-
tive, the results of the "disease-specific" (DP) and the more generic "organ-
specific" algorithms would also have been considered, and an "OK" in any
of the admission necessity algorithms would have resulted in no referral for
further review. In this instance, there were enough findings and services to
justify the hospitalization (but not the angioplasty) for ischemic heart dis
ease.
There were some signs of disease of the central nervous system (OP09),
but not enough to either justify or deny the admission, resulting in a recom-
mendation that the case be accumulated in a data base for monitoring (MO)
for patterns, but only if the monitoring flag appeared in the absence of an
"OK" flag. In this case, it is disregarded.
No generic quality screens (DS) were failed, but two problems were
identified by the discharge screen. Neither of these was severe enough to
merit referral to a physician advisor, but they were serious enough to merit
tracking (MO) to ascertain whether they are recurring problems at the hos-
pital that cared for the patient.
The results of the case-finding algorithms suggest the level of clinical
judgment they incorporate: rather rudimentary, but sufficient to give rise to
controversy. This is a problem that, in the current environment of medical
uncertainty, will dog the application of any expert system to the evaluation
of medical practices.
The other application of the data acquisition system, the development of
an epidemiological data base, has as its ultimate purpose the reduction of
that uncertainty so that case-finding rules and judgments rendered by PRO
physician advisors might be more more objective and substantial. In a more
general sense, when patterns of care and patterns of outcomes are compared
among providers of medical care, adjustment for risks contributed by patients
and therefore not attributable to care, must be made.
The first example of the epidemiological application of abstracted clinical
data is, in fact, an adjustment of mortality rates of hospitalized patients,
grouped by hospital (the hospital is the provider). Table 8 and Figure 1
OCR for page 129
COLLECTION OF PRIMARY DATA
129
present results obtained with data abstracted from medical records using
MedisGroups, a commercial system.
Table 8 compares measures of goodness-of-fit of models that employ
data available from the HCFA claims files (core model), the core model
plus the MedisGroups admission severity grade (ASG) (a measure that makes
use of the abstracted data but selects findings and weights them according
to clinical judgment)' and a model that consists of the core variables plus
specific clinical findings as abstracted. The outcome is the probability of
death of individual patients. The fit improves progressively as the compre-
hensiveness of the model increases. Variables that identify hospitals and
whose regression coefficients estimate the contribution of the hospital to
the probability of patient death (patient risk factors being equal) contribute
little at any level, but progressively less as patient risk factors are included
in greater detail.
The two measures of goodness-of-fit the proportion of concordant pairs
or area under the ROC (receiver operating characteristics curve, and the
rank order correlation coefficient of observed and predicted probabilities of
death-are directly related. The area under the ROC curve ranges from 0.5,
if the model is totally ineffective, to 1.0, if it is perfectly predictive. The
value of 0.9 achieved by inclusion of the specific clinical findings is sub-
stantial. It indicates that in about 90 percent of pairs of patients, one of
whom died and one of whom did not, the patient who died had the higher
predicted probability of death. I am not expert in these matters, but I am
TABLE X Evaluation of Goodness-of-Fit of Regression Models of the
Probability of Death of Individual Patients
Variables in Model
Demographic only
Demographic and hospital
Core
Core and hospital
Core and MedisGroups ASGb
Core, MedisGroups ASG, and hospital
Core and clinical findings
Core, clinical findings, and hospital
Proportion of
Concordant Pairsa
Rank Correlation of
Observed and
Predicted Deaths
0.640
0.689
0.838
0.852
0.883
0.890
0.896
0.902
0.279
0.378
0.675
0.704
0.767
0.781
0.792
0.804
aConcordant pairs: in pairs consisting of one patient who did and one who did
not die, those pairs in which the patient who died had the higher predicted
probability of dying.
bASG admission severity grade.
OCR for page 130
130
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
rather impressed by the power of the clinical findings to improve the good-
ness-of-fit of the model.
A more precise indication of what improvement is achieved in assessing
the hospital's contribution to the probability of patient death when clinical
data are added to the model is suggested by Figure 1. The figure plots
estimates of that contribution obtained with claims data alone (core model)
and with claims and clinical data (full model). Further discussion of this
matter is best left for another occasion, but the potential uses of detailed
clinical data in risk adjustment should be clear.
A more compelling application of detailed clinical data is to the estima-
tion of the influence of patient risk factors and of treatments on outcomes,
illustrated in Table 9. It presents a very useful example because of its
2
~1
=)
-
z
O
-
o
In -1
I
-2
· id
·
· ^'
_ _ ~e-
.
.
/
/; ~
.
· )~~
~ ~ ~ ·.,p, ~
. - ,
·~
.
~
.
-2 -1 0 1 2
HOSPITAL COEFFICIENTS, CORE MODEL
FIGURE 1 Comparison of Hospital Regression Coefficients from Core and Full
Models
OCR for page 131
COLLECTION OF PRIMARY DATA
TABLE 9 Risk Factors for Death Up to Two Years After Admission for
Acute Myocardial Infarction
131
Risk Factor
Relative Risk
of Dyinga
Age, 80 vs. 65 years
Leukocytosis, 20,000 vs. 7,000
Hypokalemia, 3.2 vs. 4.3
Alkalosis, pH 7.49 vs. 7.41
Prior admission within 30 days, yes vs. no
Myocardial ischemia (EKG), yes vs. no
Myocardial infarction, age undetermined, yes vs. no
Blood glucose, 300 vs. 90
Atrioventricular dissociation, yes vs. no
Congestive heart failure (X-ray), yes vs. no
Blood urea nitrogen, 60 vs. 15
Arterial oxygen pressure, 60 vs. 90 mm Hg
Disoriented, x2 or x3, yes vs. no
Coma/stupor, yes vs. no
Heart murmur, yes vs. no
Systolic blood pressure, 60 vs. 120
Tachypnea, 32 vs. 12 per minute
History of diabetes, yes vs. no
History of stroke or transient ischemic attack, yes vs. no
History of congestive heart failure, yes vs. no
History of myocardial infarction
Comorbidities (by ICD-9-CM codes)
cancer, yes vs. no
Chronic renal disease, yes vs. no
Streptokinase (IV or IC), yes vs. no
Coronary angioplasty, yes vs. no
Coronary bypass surgery, yes vs. no
Parenteral drugs (first 48 hours)
Beta blocker, yes vs. no
Calcium channel blocker, yes vs. no
Digitalis, yes vs. no
Intravenous nitroglycerine, yes vs. no
Loop diuretic, yes vs. no
Pressor agent, yes vs. no
Short-acting nitroglycerine, yes vs. no
1.25
1.15
0.87
1.14
1.34
0.83
0.84
1.19
2.64
1.54
1.24
1.22
1.59
2.46
1.48
1.71
1.27
1.29
1.37
1.22
1.16
1.88
1.49
Treatment Covariates
0.87(P > 0.5)
0.47(P < 0.001)
0.48(P < 0.001)
0.66(P > 0.1)
1.12(P > 0.5)
0.92(P > 0.4)
0.76(P < 0.003)
0.72(P ~ 0.1)
1.48(P < 0.001)
1.68(P ~ 0.2)
aBased on the Cox proportional hazards model and stepwise regression. Follow-
up is 12 to 24 months. All patient-specific risk factors are statistically significant at
P < 0.05.
OCR for page 132
132
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
complexity and the difficulty of its interpretation. The results shown were
obtained by the application of the Cox proportional hazards model.
The upper portion is straightforward, consisting of estimates of changes
in the risk (relative risks) of death due to acute myocardial infarction (AMI)
associated with the specified risk factors, all other risk factors being held
constant. Only highly statistically significant (P < 0.05) predictors of death
are listed in this portion. The risk factors include demographic characteris-
tics, results of the admission physical examination, laboratory tests and
other diagnostic tests carried out in the first 48 hours of hospitalization (or
prior to surgery), and historical data.
The lower portion, which addresses treatments, is intriguing. The data
suggest that coronary angioplasty and bypass are (or were in 1985, the year
the treatments were administered) highly effective tools for the treatment of
patients with AMI, controlling for the patient risks specified in the upper
portion of the table. In fact, they reduced the probability of death by about
half. This effect persists if patients who died on the day of admission are
excluded, because they may not have lived long enough to become candidates
TABLE 10 Risk Factors for Rehospitalization Following Acute
Myocardial Infarction
Risk Factor
Relative Risk of
Rehospitalizationa
Leukocytosis, 20,000 vs. 7,000
Hypocalcemia, 7 vs. 9.5
Hypokalemia, 3.2 vs. 4.3
Prior admission within 30 days, yes vs. no
Blood urea nitrogen, 60 vs. 15
Arterial oxygen pressure below 75 mm Hg
Edema, >2+
Systolic blood pressure, 60 vs. 120
Tachypnea, 32 vs. 12 per minute
Left ventricular ejection fraction, 35 vs. 62%
History of diabetes, yes vs. no
History of congestive heart failure, yes vs. no
History of chronic obstructive pulmonary disease, yes vs. no
History of immunosuppressive therapy, yes vs. no
Currently on anticoagulants' yes vs. no
Comorbidities (by ICD-9-CM code)
cancer, yes vs. no
1.18
0.66
0.85
1.54
1.23
1.16
1.34
0.78
1.19
1.33
1.15
1.23
1.30
1.26
1.41
1.61
aBased on the Cox proportional hazards model and stepwise regression. Follow-
up is 12 to 24 months. All patient-specific risk factors are statistically significant at
P < 0.05.
OCR for page 133
COLLECTION OF PRIMARY DATA
133
for revascularization, and if the "center effect" is controlled for, because the
superior outcome associated with revascularization may reflect the fact that
patients admitted to hospitals that perform coronary revascularization may
have received better care overall. The adverse effect of the use of presser
agents, controlling for hypotension, is also intriguing.
The analyses presented in Table 9 are observational and must, therefore,
be approached with some caution. Nevertheless, the power of detailed
clinical data in describing the natural history of conditions as they are
currently being treated is well illustrated.
Of course, mortality is not the only outcome that may be addressed by
the combination of detailed clinical and claims data. Table 10 illustrates, in
a fashion analogous to Table 9, analyses addressing rehospitalization rates.
In all, to characterize adequately the effectiveness of medical interventions,
and therefore their impacts on the health of patients, at least from a public
health perspective, it is necessary to measure mortality, morbidity, disabil-
ity, and expenditures for health services, in order to try to track the effec-
tiveness of interventions.
CONCLUSIONS
HCFA's objectives in assessing the effectiveness of interventions were
as follows:
· to assess the overall merits of competing procedures,
· to provide information to assist clinicians in the management of patients,
· to provide information to assist in peer review of care,
· to guide in the formulation of policy on the allocation of resources.
The initial intent was to provide PROs with more effective tools for the
review and evaluation of patient care. Clearly, the information generated in
this process has broader applications, the most important being to assist
clinicians in the treatment of patients by providing assessments of the rela-
tive merits of treatment strategies overall and for patients with specific risk
factors. A further useful by-product is guidance for the allocation of resources,
at whatever level such decisions are made, by providing measures of the
impacts of those decisions on the health of patients.
OCR for page 134
OCR for page 135
Development and Use of
Outcome Measures: Introduction
G. Richard Smith, Session Moderator
The primary thrust of the Effectiveness Initiative is to determine what
works in the practice of medicine. One can determine if something works
only by knowing what happens to the patient. Therefore, a major emphasis
of the committee's work has been on the outcomes of patient care.
Previously, we could determine outcomes only in terms of whether a
patient was alive or dead or by using some kind of medical test. We had
very limited knowledge about what effect various medical interventions had
on patients. Over the past 15 years, a new technology has been developed
to enable us to understand some of the effects of medical care. This technology
is called health status assessment, functional status assessment, or even at
times quality of life. As a result of this new technology, we are now able to
quantify a number of aspects of the state of a patient's health. For example,
we can determine the effect on a patient's physical health of developing
asthma or the effect on a patient's mental health of being told that she has
breast cancer.
When these tools are applied in a systematic and thoughtful fashion, we
can tell much about the effect of our medical care system not only on our
people as a whole, but upon our individual patients.
Donald L. Patrick is currently professor of health services and director of
the Social and Behavioral Sciences Program at the University of Washington
School of Public Health. Dr. Patrick discusses selection of outcomes; use
of generic and disease-specific measures; progress toward short, reliable,
valid, and responsive measures; and interpreting observed changes in measures
and what these changes mean.
Paul D. Cleary is an associate professor in the Department of Health
Care Policy at the Harvard Medical School. His chapter describes current
135
OCR for page 136
136
EFFECTIVENESS AND OUTCOMES IN HEALTH CARE
and recent research efforts in the development and use of outcome measures.
The presentation highlights the use of patient self-reports.
Audrey Burnam is a senior behavioral scientist at The RAND Corporation.
Her chapter concentrates on the depression part of the Medical Outcomes
Study and summarizes the group's approach to studying depression outcomes.
Initial findings from the study are presented.
Representative terms from entire chapter:
uniform clinical
