can optimize early intervention and preventive maintenance,
thereby reducing ER visits and inpatient admissions.
ICD- 9 is an imperfect means of finding patients who share
similar characteristics. It may require casting a wider net than
needed to ensure all possible patients are identified.
In the case of poorly controlled diabetes, ICD- 9 adds the
words “uncontrolled” to many diabetes codes, each specifying
a diabetic complication. ICD- 9 creates “combination codes,”
where each code title conveys two things about a patient. For
example, code 250.42 indicates “this patient has diabetic renal
manifestations” and “this patient’s diabetes is uncontrolled.”
Thus the need to cast a wide net. No harm in that…yet.
However, say that an organization wants to use ICD- 10 to find
all its patients with poorly controlled diabetes. The organization
reads only the large print and autoconverts its systems, taking
each ICD- 9 code on its list and translating it to its corresponding
ICD- 10 equivalent.
Because of structural differences between ICD- 9 and ICD- 10,
what was one code in ICD- 9 is two separate codes in ICD- 10.
For each ICD- 9 combination code on the organization’s list, the
GEMs autoconversion adds two ICD- 10 codes to the list—one
code indicating “this patient has diabetes with renal manifes-
tations” and one code indicating “this patient has poorly con-
The GEMs cannot infer the purpose of the conversion; they
can only translate the original ICD- 9 codes. So in its effort to
fully translate each individual ICD- 9 code, the autoconversion
has added a whole bundle of separate codes for diabetes to the
ICD- 10 “poorly controlled diabetes” list without any mention of
The reasons conversions
cannot be fully automated are
not technical. They come from
differences in language and
structure between the code sets.
Fast forward to the care management program in the world of
A patient with one ICD- 10 code on his health record that indicates “this patient has diabetes with renal manifestations” has
been placed in the poorly controlled diabetes care management
program. So have the patients with circulatory manifestations,
ophthalmic manifestations, and peripheral neuropathy—in
fact, almost all diabetes patients are being referred to the care
Clearly this was not a correct conversion result for this system.
A subject matter expert who understands the purpose of the
system would have tailored the results of the conversion to produce an ICD- 10 version consistent with the system’s purpose.
ICD- 10 Fine Print
Four Truths in the Conversion Fine Print
1. Conversions cannot be fully automated, and they require
the active involvement of a team that has a solid understanding of both the system being converted and the
2. A conversion is not a single event with a well-defined end,
but a multiphase process that will span several years.
3. It is not possible to create a valid “all-purpose” translation
of ICD- 9 data to ICD- 10 data for general use in impact
4. Stakeholders negotiating proposed ICD- 10–based changes to contractual arrangements will try to estimate financial impact with little or no actual ICD- 10 data, finding
themselves forced into a contest of expert opinions rather
than a discussion of reliable, empirical results.
Replicated versus Optimized Conversions
Recognizing that there is formidable short-term financial risk
and long-term legal risk in the ICD- 10 conversion, many organizations are carefully converting their systems and policies so
that they will behave the same using ICD- 10 as they behave using ICD- 9. This conservative strategy can be called replication.
For a correctly coded record in ICD- 10, a replicated ICD- 10 system should yield the same result as the original ICD- 9 system
applied to the same record coded in ICD- 9.
Since conversion errors may produce unintended payment
gains and losses, the draft ICD- 10 MS-DRGs published by the
Centers for Medicare and Medicaid Services were converted to
replicate the ICD- 9 MS-DRGs so that unintended payment gains
and losses are minimized. The MS-DRG conversion involved extensive review and evaluation by subject matter experts.
Some organizations may attempt to optimize a system for
ICD- 10 from the outset—that is, make use of the increased precision and specificity of ICD- 10. An optimized ICD- 10 system
will intentionally not yield the same result as the original ICD- 9
system. Because ICD- 10 coded data are not readily available, estimating the actual impact of a prospectively optimized ICD- 10
conversion is necessarily an exercise in guessing the future. For
payment systems, ICD- 10 optimization without sufficient actual
ICD- 10 data risks unintentional payment gains and losses.
For systems that do not relate to historical prices or norms, it
may in fact be desirable to optimize for ICD- 10. For example,
claims-editing systems that identify clinical and other inconsistencies may need to be optimized for ICD- 10 from the outset.
Although a valid optimization of systems like claims editing can
be done, the precise efficacy of ICD- 10 optimized edits cannot
be predicted without significant actual ICD- 10 coded data.
Because many systems will first be replicated and then optimized, ICD- 10 conversion will be a multiphase process spanning several years surrounding ICD- 10 implementation. Even
systems that can be directly optimized will need refinement as
ICD- 10 use and coding practices evolve.