Statement of William Hogan and Vergil Slee, The Rods Laboratory (at the University of Pittsburgh), Pittsburgh, Pennsylvania

SUMMARY

It would be a mistake to switch from ICD-9-CM to ICD-10-CM in the Medical Record Health Information System (MRHIS)^[1] at the present time.

The resource expenditure required would present an unnecessary, perhaps insurmountable, obstacle to the efforts of the federal government to modernize the healthcare industry with information technology (IT), most especially the adoption by physicians of the electronic medical record (EMR).

The claim that implementing ICD-10-CM is critical to biosurveillance for such threats as SARS and avian influenza is inaccurate – the public health reporting system does not use ICD-9-CM codes for this purpose.

The basic problem that the healthcare system and the federal government need to address is that there is no standard set of codes for diagnosis INPUT.  Our system is obsolete.  It uses diagnosis OUTPUT codes^[2] for diagnosis INPUT  – both ICD-9-CM and ICD-10-CM are OUTPUT codes. 

Instead of switching to ICD-10-CM, we should develop and implement a modern system for diagnosis INPUT^[3].  No system for this purpose exists today.

OUR CREDENTIALS

William Hogan, MD, MS is an Assistant Professor of Medicine in the School of Medicine of the University of Pittsburgh and a senior analyst in its Realtime Outbreak and Disease Surveillance (RODS) Laboratory.  RODS carries out research under funding from DARPA, CDC, NLM, DHS, AHRQ, and NSF.   Hogan has extensive experience in building biosurveillance systems, as well as conducting basic biosurveillance research.  Prior to joining the RODS Laboratory in 2002, he worked at Health Language, Inc, where he gained expertise with vocabulary standards including ICD‑9‑CM and SNOMED.  He is the expert on vocabulary standards at the RODS Laboratory, and led the effort at the RODS Laboratory to map proprietary codes used by eight hospital laboratories to LOINC^[4] and SNOMED CT for electronic laboratory reporting.  He has written on vocabulary standards, biosurveillance, and the intersection of the two.

Vergil Slee, MD, MPH, FACP, FACHE (Hon) was responsible for the first deployment of ICD in hospitals as a tool for diagnosis indexing, a task for which ICD was admirably suited at that time (1955).  In 1975 he represented the U. S. at the WHO conference which designed ICD-9.  In 1976 he became President of the Council on Clinical Classifications which, in collaboration with the U. S. National Center for Health Statistics, developed ICD-9-CM (1978).  He has analyzed ICD-10-CM (Reference 6) and has written extensively on the expanding demands on medical record information (Reference 7), demands which have destroyed ICD’s suitability for diagnosis input. 

STATEMENT OF THE ISSUE: 

The Federal government is being urged to replace ICD-9-CM diagnosis codes with ICD-10-CM diagnosis codes in the Medical Record Health Information System (MRHIS)^.[5]This system uses medical record data for three major purposes:

PRIMARY PURPOSE:   Patient care.  The medical record’s primary, nonnegotiable purpose is to be the memory and communication tool for the physician.  It has no substitute for the care of the individual patient.

SECONDARY PURPOSES: (1) Billing and (2) Statistics on health and health care.

BASIC FACTS: 

Two essential facts about the MRHIS must be included in discussing the question of switching from ICD-9-CM to ICD-10-CM:

For patient care, the physician must have diagnoses in their greatest detail.

For billing and statistics, diagnoses must be grouped.

“ICD coding,” the coding used in the MRHIS, is category coding^[6], which captures only the labels of the groups (categories) in a clinical modification of ICD.  The ICD series^[7] was designed for the OUTPUT of data for statistics (and, in the U. S., the clinical modification (CM) is essential for billing).  The ICD series was never intended for INPUT, the purpose for which we (mis)use it.  The precise diagnoses are simply discarded, except in the rare instance where a category has only one diagnosis (see the illustration on page 8).  The result is that ICD coding – category coding – is not useful for the physician in the care of the patient.

Category coding has an especially pernicious effect for statistics.  The coded data are already aggregated, and aggregated data can never be disaggregated – they can only be combined into larger groups (such as DRGs).  This means that our health care system is in a  “one-size-fits-all” situation and must use the same statistics for such disparate purposes as public policy, quality review, facility management, and evidence-based medicine.  Common sense dictates that each of these uses has unique information demands and should have its own grouping of diagnoses.  We propose a solution below.      

SWITCHING FROM ICD-9-CM TO ICD-10-CM

Proponents of switching argue that switching is a cost-effective and necessary step to modernize our healthcare information system, that we must keep in step with other nations for international comparisons of morbidity, and that ICD-10-CM is more up-to-date, has more room for “things,” can more easily accommodate new diseases such as SARS, and is better suited to biosurveillance for terrorist and emerging disease threats. 

UP TO DATE:  ICD-9, the parent of ICD-9-CM, was written in 1975 and ICD-9-CM was put into use in 1978.  ICD-10 was written in 1989 and published in 1992.  ICD-10-CM is still in draft form.  The U.S. agreement with the World Health Organization (WHO) (the author of the ICD series of classifications) states that any clinical modifications (CM versions) must be “collapsible” back into the categories of ICD itself, which greatly reduces our freedom to keep it current with medical progress.

LACK OF SPACE:  A second argument is that ICD-9-CM has few remaining codes to assign to new diagnoses.  Actually it uses only about 13,000 codes out of the over 100,000 permitted by its structure.  The problem is that it uses an antiquated code structure, where the code indicates the location of a category in a hierarchy of categories.^[8]  Modern computer methods do not require that codes be in such a numerical hierarchy.  ICD-10-CM (2003 draft) has 67,000 codes, and thus one might expect that it has more clinically relevant detail, but this number is misleading.  For example, one category, Code S82 Fracture of lower leg, including ankle, with its mandatory extensions, accounts for 3,248 of the codes in this count.

NEW DIAGNOSES: We have no system for promptly coding new conditions such as avian influenza, SARS, and Gulf War Syndrome (real or suspected) with ICD-9-CM, nor would we with ICD-10-CM.  For example, there were no ICD-9-CM codes for SARS until October 1, 2003, nearly 3 months after the WHO lifted all its travel advisories and considered the outbreak under control. Nor is the system ready to uniquely identify avian influenza.  The inability of a category coding system to classify new entities is due to the necessity of deciding what they are before deciding where to put them in the classification.  There are also the requirements that a committee make the decisions, and that changes, which must be implemented first by human coders dispersed throughout the nation, are only made once a year. 

INTERNATIONAL MORBIDITY DATA: The U. S. is under no obligation to use ICD for anything other than mortality data.  In personal communication with Dr. Slee, a WHO statistician stated that for virtually all international morbidity studies, special data collection is required.

BIOSURVEILLANCE:  Proponents of ICD-10-CM argue that it is essential to make the switch to have better disease surveillance for terrorist and emerging infectious disease threats.   However, the current biosurveillance system makes little use of even ICD-9 codes for detecting disease outbreaks.  When physicians, hospitals, and laboratories report notifiable diseases to public health, they do not use (nor are they required to use) ICD-9-CM codes.  Influenza surveillance relies in part on mortality statistics (which are retrospective, of course), but mortality statistics are already compiled using ICD-10 (not ICD-10-CM) codes.  We have already noted the inadequacy of ICD-9-CM for accommodating emerging disease threats such as SARS.

In view of these facts, the arguments as to advantages of ICD-10-CM over ICD-9-CM lose a great deal of their weight. 

FINANCIAL IMPACT:

MONETARY COSTS:  Two estimates have been made of the cost of switching to ICD-10-CM from ICD-9-CM.  The RAND Corporation figure, for the Centers for Disease Control, was from $425 million to $1.125 billion over 10 years. The Robert E. Nolan Company figure, for the Blue Cross Blue Shield Association, was from $6 billion to $14 billion over 2-3 years.  The higher figures are likely to be more accurate, because Nolan based its estimates on similar information technology conversions in the past, namely actual costs to the healthcare system of the year 2000 (Y2K) remediation and Health Insurance Portability and Accountability (HIPAA) compliance.  Nolan also gave some details of actual costs incurred by Canada, Australia, and the United Kingdom during their switch to ICD-10.^[9]

Regardless, the switch would require the commitment of human and financial resources which would thus not be available for solving the underlying problem with health care information: we use diagnosis output codes for diagnosis input.  Detailed diagnosis input is a building block on which the entire MRHIS must be (re)built.  It is critical that the healthcare system and the federal government devote resources to the solution of this problem, as we discuss below.

SAVINGS:  The most optimistic estimate of the benefit of switching to ICD-10-CM is $7.7 billion over 10 years (Rand).  By contrast, the projected savings to the healthcare system of a national health information network (NHIN), as envisioned by the Office of the National Coordinator for Healthcare Information Technology (ONCHIT), is $337 billion over the first 10 years, and $77.8 billion per year thereafter (Reference 1).  This estimate assumes the use of the EMR, the adoption of which requires diagnosis input codes in place of ICD-9-CM codes.^[10]  Importantly, none of the benefit of NHIN results from switching to ICD-10-CM.

BUILDING NHIN: Achieving this goal would be seriously delayed.  In the event of a switch to ICD-10-CM physicians would have to give the new code system priority over the EMR so that they could maintain revenue.  Health plans considering financial incentives to physicians for adopting EMRs would have to divert resources to the ICD-10-CM switch.   Hospitals, health plans, physicians, and state governments would all have fewer resources to devote to developing health care data exchange.

INFORMATION COSTS:  The costs of the switch in its effects on health and healthcare information cannot be predicted, but they may well be, in the long run, more important than money.  One particular effect:

Longitudinal studies (studies which cross the date on which classifications change, e.g., from ICD-9 to ICD-10) usually are seriously disrupted and often have to be abandoned.  Disturbance of such studies costs money as well as information.  To illustrate with one such study: 

ICD-10 has been used for U.S. mortality tabulations effective with 1999 death certificates.  In Florida, the AIDS death rate, which had been declining by about 6% per year until 1999 took a sudden rise of about 6% in that year.  Investigation showed that this 12-13% jump was entirely due to the coding change; the true trend had been badly distorted (See Reference 2).

INFORMATION QUALITY: Information quality always sags, often for several years, simply as the result of changing coding (the effects are prolonged by the fact that implementation cannot be achieved on a single date; all elements of the system, from coders, through computers, must be up to speed before the system becomes reliable).

HOSPITAL AND PHYSICIAN REIMBURSEMENT: Our reimbursement system is in financial equilibrium.  Any change in coding would require recalibration, involving collection of both medical and financial information on huge numbers of patients (enough to give statistical validity to each DRG, for example), and parallel operation of both the old and the new systems until reliability could be guaranteed.  This is truly a non-trivial aspect of switching.

MODERNIZATION OF THE MRHIS: The switch would delay attention to actually modernizing the system, as outlined below.

MODERNIZATION OF THE MRHIS

We contend that we should not try to put a “band-aid” on our obsolete MRHIS.  Instead, we should develop the missing component, a system for INPUT of diagnoses, and bring the system up to 21^st century standards.  Modernization should start with these steps:

1.    DELAY THE SWITCH to ICD-10-CM: We should continue to use  ICD-9-CM, modifying it as necessary for the reimbursement system, in order to free the resources, human and financial, needed to modernize the system.

2.    DEVELOP A CODED DIAGNOSIS INPUT SYSTEM:  Add diagnosis entity coding^[11] to the medical record.

It is a basic principle in science that original observations – in this instance, original medical records – should be preserved permanently, without alteration, in order to permit review and further analysis as needed.  Preserving diagnoses only after they have been placed in groups and have lost their individual identity, as we do today with category coding input, is a gross violation of this basic principle.  The practice is an embarrassment and should be stopped immediately.  Entity coding would permanently preserve the original diagnoses.

The needed input system would be simple and user-friendly for the physician – with no look-up or coding required.  This would demand three critical attributes.  The system would

a.    Accept natural language input for any term the physician wishes to use (See Reference 8 which discusses “free vocabulary” and References 3 and 4 which discuss the need for an “interface” vocabulary for users of EMRs).

b.    Map, by computer, this free language term to a standard vocabulary such as SNOMED CT.  (SNOMED CT is a reference terminology for well-studied conditions, not an input terminology, although its preferred terms are, of course, often those the physician normally uses.)

c.    Provide instant, “realtime” codes for new diagnoses such as SARS – coding for new terms is delayed in the present system for months or years, as was the case with AIDS.

3.    DEPLOY THE INPUT SYSTEM:   Make the entity diagnosis coding system freely and readily available to all elements of health care.

ADVANTAGES OF THE MODERNIZED SYSTEM

The modernized system would

1.            Provide physicians the detailed diagnoses they need for patient care.

2.            Remove a major barrier to acceptance of the EMR.

^[1]  The MRHIS is the information system which originates with data in the medical records of hospitals and physicians’ offices.  It must first meet the needs of the physician as a memory and communication tool in care of the individual patient.  Some medical record content is then used to support billing and to create statistics on health and healthcare.

^[2]  Output codes are codes which represent the labels of the categories of a classification which has been constructed for statistical purposes.  Each diagnosis output category (code) contains a collection of the individual diagnoses which are its input.  In both ICD-9-CM and ICD-10-CM, several hundred thousand specific diagnoses are collected into the few thousand categories of the classification.  Most individual diagnoses, therefore, have no unique codes.

^[3]  Input diagnosis codes are codes which exactly represent diagnosis entities as expressed by the physician.  The physician may or may not record a diagnosis using its “preferred term;” he may or may not even use a standard synonym.  Following input, the computer system will add the “standard code” for the preferred term in order to facilitate finding it in all medical records.

^[4]  LOINC – Logical Observation Identifiers, Names, and Codes.

^[5]  NOTE RE PROCEDURE CODING:  Proponents of the switch to ICD-10-CM imply that it would also  require the switch, for hospital inpatient records, from Volume 3 of ICD-9-CM to ICD-10-PCS (which was written by 3M under contract from HHS).   Procedure coding is a separate issue from diagnosis coding.  No simultaneous switch is necessary; the system could stay with the present procedure coding or consider the procedure coding available with SNOMED CT (Standardized Nomenclature of Human Medicine – Clinical Terminology).  SNOMED CT is a reference terminology developed and maintained by the College of American Pathologists (CAP).  SNOMED CT has been made available to the healthcare system by HHS under a contract with CAP.

^[6]  Category coding of diagnoses is coding in which each code (number) represents (the title of) a category of diagnoses, e.g., “Other diseases of the liver.”  In category coding, the coding of the diagnosis itself and its classifying are combined into one step.

^[7]  ICD, the International Classification of Diseases, is a serial publication of WHO.  The U.S. created ICD-9-CM (a clinical modification, CM) in 1978 for use in hospitals and doctors’ offices.  ICD-10-CM based on ICD-10, first edition (1992), is now in draft in a version dated June 2003.  WHO issued a second edition of ICD-10 in 2004.  Presumably the U. S. ICD-10-CM would have to be made to correspond with each new edition.

^[8]ICD-10-and ICD-10-CM maintain this antiquated hierarchical code structure, but have a higher number of possible codes.  Depending on how our knowledge of disease increases, ICD-10 could also eventually “run out of codes” in some parts of its category hierarchy.

^[9]  Canada created and implemented its own modification of ICD-10 called ICD-10-CA.  Similarly, Australia created and implemented ICD-10-AM.  The U.K. uses ICD-10 for morbidity with no modifications.  Note that each has used a different version of ICD-10, thus making international comparisons far more difficult than if the same code version (clinical modification) were used.

^[10]  The ICD-9-CM coding would continue, however, since it is at the heart of the billing process, and financial stability must be maintained.

^[11]  Diagnosis entity coding, necessary to modernize MRHIS, is coding in which each code represents an individual diagnosis entity.  The diagnoses can, with this coding, be placed in ANY classification, e.g., ICD-9-CM, ICD-10-CM, or policy, planning, or research groupings.