|
|
Performance measurement is a powerful
policy tool for promoting efficient high quality
care. However, it is often underestimated
just how harmful performance
measures can be when they are poorly
constructed.1,2,3 Perhaps the hardest lesson is
that developing performance measures that
promote optimal care usually requires clinically
detailed data and complex measures,
and that simplistic or naive "good care"
measures can have strong perverse incentives.
This push for simple measures is strongest
in the community where the resources
and will to invest in electronic medical
records and chart-based review have been
lacking.3 However, since there is considerable
political pressure for the VA to adopt
those measures used in the community
(so as to allow benchmarking), the recent
push for all-or-none "good care" measures
is of considerable concern (e.g., A1c < 7
percent, blood pressure < 130/80, LDL <
100mg/dl, etc). Such measures are likely
to be very imprecise measures of efficient
high-quality care, and are also prone to
perverse incentives, such as promoting
treatment irrespective of how small the
potential benefit and how great the patient
burden or risks.1,2,3
Take annual diabetes eye exams for example.
Although well-timed photo coagulation
for early diabetic retinopathy is one of the
most beneficial treatments in all of medicine,
research conducted by VA HSR&D
suggests that almost all visual impairment
that is preventable by early detection will
be captured by: 1) screening those with no
known eye disease every two to three years,
and 2) close individualized surveillance
(every 4 to 12 months) after early retinopathy
has been detected.4 Therefore, annual
exams are not of high importance for the
vast majority of low-risk patients, and are
too infrequent for the highest risk patients
who will account for most preventable
complications. As a result, the conventional
performance measure (annual examinations
for diabetic patients), which provides a
strong incentive to focus resources on getting
everyone into clinic every 13 months,
provides no incentive to develop an effective
system to optimize care.
In fact, a health system that schedules
exams every 10 to 11 months and devotes
its scarce administrative resources to trying
to get anyone who misses this appointment
into the clinic as soon as possible,
is likely to improve its performance rating
while doing almost nothing to improve
outcomes. In contrast, a health system that
uses its administrative resources to aggressively
reschedule those needing close follow
up (because of known retinopathy) and
to have low-risk patients seen at least every
two years (those whose last retinal exam
was normal on their last visit), may make
its performance rating much worse while
substantially improving patient outcomes.
It should not be surprising therefore that
when we tried to develop an effective system
to improve eye screening and follow up
for diabetic patients, that the prevailing
annual eye exam performance measure was
one of the biggest barriers to implementing
the more effective system. Although many
may criticize clinic leaders for not doing
the "right" thing, these clinics and providers
have huge demands on their time and
attention. As has happened with so many
important quality improvement initiatives
that we've consulted on in the past
10 years, the clinicians and administrators
eventually said, "If the problem you want
us to address is really that important, then
get the performance measure changed. We
are struggling to meet dozens of demands
and we just do not have the time and personnel
to electively take on more things."
It is rare that "good care" can be measured
simply. Although the new NCQA
diabetes measures of A1c < 7 percent and
BP < 130/80 may seem straightforward
enough, they actually provide strong incentives
for speculative, costly, and potentially
dangerous polypharmacy. In addition, these
are unadjusted outcome measures that are
likely to be inaccurate. For example, the
A1c measure more strongly rewards adding
or increasing the dose of a glitazone,
which has high costs and limited data on
long-term safety, in someone with an A1c
of 7.5 percent than it does for doing so in
someone with an A1c of 8.5 percent, even
though a risk of an A1c of 7.5 percent is
trivial compared to the risk of an A1c of
8.5 percent.
Similarly, although adding up to three to
four medications at moderate doses in
pursuit of good blood pressure control in
a high CV-risk patient has been shown to
be highly beneficial, the benefits of pursuing
the 130/80 targets using more than
three to four medications is pure speculation.
Furthermore, there is ample reason to
be concerned about harmful effects from
polypharmacy or excessively lowering diastolic
blood pressure.
Trying to measure a complex clinical
scenario using simplistic performance
measures and wishful thinking is increasingly
promoted by disease advocates and
industry-sponsored experts in the community.
NCQA's new "good control" diabetes
measures and its resistance to revising its
eye care measure despite the strong advice
of evidence-based medicine experts is just
one example of this trend. This is not to
suggest that we should limit ourselves to
only measuring bad care, such as A1c > 9
percent or retinal screening exams > two
years. However, if we wish to measure
"good care," we will need more nuanced
performance measures that consider the
benefit of reaching these "optimal" treatment
goals, as well as the risks, costs, and
patient burden associated with treatments
needed to reach these optimal goals.
- Hayward RA. Performance Measurement in Search
of a Path. New England Journal of Medicine 2007;
356(9):951-3.
- McMahon LF, Hofer TP, Hayward RA. Physician-level
P4P–DOA? Can Quality-based Payment Be
Resuscitated? American Journal of Managed Care 2007;
13(5):233-6.
- Hayward RA. All-or-nothing Treatment Targets
Make Bad Performance Measures. American Journal
of Managed Care 2007; 13(3):126-8.
- Hayward RA, et al. Causes of Preventable Visual
Loss in Type 2 Diabetes Mellitus: An Evaluation of
Suboptimally-timed Retinal Photocoagulation. Journal
of General Internal Medicine 2005; 20:467-9.
top
|
