Share
|
Unfortunately health care planers and even public health base medical policy on this vast corrupted pool of diagnostic data.
Diagnostic errors in hospitals account for 17% of adverse events and 10.5% of adverse events resulted from misdiagnosis.[2] Given that 15% of the time the clinician lists the wrong initial diagnosis, clinical decision support (CDS) – elsewhere called computer decision support system (CDSS) -- makes a lot of sense. While CDS finds its way into most commercial vender based electronic health records (EHR) and internally programed institutional EHRs, very few include diagnostic decision support beyond facilitating the problem list. Wright, Sittig and Ash et al, in the Journal of Informatics (AMIA) list taxonomy of 53 CDS front-end applications taken from a survey of seven selected vendors and four institutions. Only three provide any diagnostic support. Most CDS applications focus on medications, orders, incompatibilities and treatment protocols based on “best evidence.” Obviously, the best treatment protocol applied to a wrong diagnosis does more harm than good. If the clinician lists all of the possibilities as in a differential diagnosis, the list will likely contain the right diagnosis. It then becomes a process of selection and elimination. The final answer may still be in dispute at autopsy but there will be fewer misstatements along the way and a far more credible database. CDS can provide a differential diagnosis along with statical probabilities based on the patient data available. If the CDS contains the criteria for diagnosis and the patient meats that criteria, then the clinician will likely be on the right track. The human mind under the best of circumstances struggles with both memory and statistics. In no way can a computer replace the judgment, the intuition, and the wisdom of a well-educated and experienced physician, but well written database applications can remember it all unerringly, and apply statistical interpretation. Furthermore, to an even higher level of accuracy, the relational database can internally improve its statistical interpretation based on later proven diagnosis and or outcome. Ongoing analysis of internal data amounts to a simple artificial intelligence, but not of the human variety. Therefore, CDS acts as a complement to clinical thinking, not a replacement and an ongoing source of continuing medical education. (CME)
Table 1Diagnostic Error Rates from the literature[4]
Aortic Aneurism 61%
Sub Arachnoid Hemorrhage 30%
Breast Cancer / Mammogram 21%
Bipolar Disorder 69%
Appendicitis 18%
Cancer Path 2-9% GYN; 5-12% non-GYN
Endometriosis 18%
Psoriatic Arthritis / Standard Patient 39%
Atrial Fibrillation / EKG Machine 35%
Infant Botulism 50%
Diabetes Mellitus 18%
Chest XR by / ER 18%
A review of the literature published in the Green Journal[1]
Table1 lists diagnostic errors by clinical condition from various studies.
Striking among them, ruptured abdominal aortic aneurysm was missed 61% and
dissecting thoracic aneurysm 35% of the time. The radiologist missed breast
cancer on the mammogram 21% of the time. The initial diagnosis of bipolar
disorder was wrong 69% of the time. Among patients with fatal pulmonary
embolism, the diagnosis was unsuspected 55% of the time. Psoriatic arthritis
was missed or wrong 39% of the time. Amazingly, the machine EKG missed atrial
fibrillation in 35% of the tracings and the reviewing clinician failed to pick
it up in 24% of those.
In my own observation, misdiagnosis is rampant among less
threatening clinical conditions especially those offered from outside of the
relevant specialty among specialists, as well as the more unusual conditions
encountered by primary care physicians. It is not uncommon to misdiagnose
various forms of sinus disorder, otitis and respiratory disease. Pneumonia too,
while recognized as such often lacks a credible etiology -- due in part to less
scrupulous sputum collection and identification by microbiology. With a
multitude of causes, pneumonia may be diagnosed by circumstance such as community-acquired
pneumonia and treated with a high percentage antibiotic for the so-called
community acquired condition.
Trial lawyers claim that the first physician misdiagnoses
cancer 45% of the time -- probably not an overstatement.
Medicine is both an art and a science. Some physicians lean
so to the side of art that they lay claim to the diagnosis itself. When they
“make a diagnosis,” that diagnosis is laid in stone. The treatment, whether
appropriate or not, is credited with the success whilst the body cures itself
with or in spite of the treatment. So much is the body’s responsiveness to
caring, faith and encouragement that the placebo effect is real. Thus, the patient’s
response reinforces the physician’s self-confidence and further obscures misdiagnosis.
Then too, there is the right treatment for the wrong
diagnosis, often developed over time based on experience, the outcome may be
quite effective – or just an accident. Was this how a drug for Parkinson’s came
to be effective for influenza, how an anti tuberculosis drug became effective
for depression, or how a depression medication came to be effective for pain.
Diagnosis too is like an amoeba; it changes with passing
specialties, fashion and even science. Current medical terminology changes as if
with the weather, placing a slightly or even a profoundly different slant on
the basic concept of pathophysiology and today a new understanding of the
basics of disease through advances in molecular biology.
I’ve made the point now so often that is a cliché to add the
distortion in diagnostic records produced by insurance clerks. Seeking to
submit only the diagnosis that claims processors will accept for reimbursement,
the insurance database bears little relationship to the actual doctor’s notes. Furthermore,
the ICDMA consistently leagues behind current terminology. Whereas ICDA makes
room for acute, sub acute and chronic, it makes little provision for
differential diagnosis, presumptive, working or established diagnosis. One can write
3 or 4 digit codes and state the diagnosis simply as chest pain without
specific cause or abdominal pain the same way, but in the wisdom of the
reimbursement system, a 4 digit code does not garner reimbursement or authorization
for further testing. Expediency thus codifies the wrong diagnosis in the
accumulated database of medicare, medicaid and other third party claims
processing agencies.
A further distortion results from the physician’s reluctance
to saddle the patient with a diagnosis, which might cause the patient to lose
insurance coverage. The same applies to a prior condition. Sometimes the
physician may downplay a frightening diagnosis pending further counseling with
the patient. In another case, the physician may avoid the true diagnosis to
protect the patient’s confidentiality when a family member or others view or
co-pay the insurance claim.Unfortunately health care planers and even public health base medical policy on this vast corrupted pool of diagnostic data.
Diagnostic errors in hospitals account for 17% of adverse events and 10.5% of adverse events resulted from misdiagnosis.[2] Given that 15% of the time the clinician lists the wrong initial diagnosis, clinical decision support (CDS) – elsewhere called computer decision support system (CDSS) -- makes a lot of sense. While CDS finds its way into most commercial vender based electronic health records (EHR) and internally programed institutional EHRs, very few include diagnostic decision support beyond facilitating the problem list. Wright, Sittig and Ash et al, in the Journal of Informatics (AMIA) list taxonomy of 53 CDS front-end applications taken from a survey of seven selected vendors and four institutions. Only three provide any diagnostic support. Most CDS applications focus on medications, orders, incompatibilities and treatment protocols based on “best evidence.” Obviously, the best treatment protocol applied to a wrong diagnosis does more harm than good. If the clinician lists all of the possibilities as in a differential diagnosis, the list will likely contain the right diagnosis. It then becomes a process of selection and elimination. The final answer may still be in dispute at autopsy but there will be fewer misstatements along the way and a far more credible database. CDS can provide a differential diagnosis along with statical probabilities based on the patient data available. If the CDS contains the criteria for diagnosis and the patient meats that criteria, then the clinician will likely be on the right track. The human mind under the best of circumstances struggles with both memory and statistics. In no way can a computer replace the judgment, the intuition, and the wisdom of a well-educated and experienced physician, but well written database applications can remember it all unerringly, and apply statistical interpretation. Furthermore, to an even higher level of accuracy, the relational database can internally improve its statistical interpretation based on later proven diagnosis and or outcome. Ongoing analysis of internal data amounts to a simple artificial intelligence, but not of the human variety. Therefore, CDS acts as a complement to clinical thinking, not a replacement and an ongoing source of continuing medical education. (CME)
Diagnosis is like an onion, it has many layers. Changing patterns
of cause and effect, a long and complex history and the patient’s adaptations
both emotionally and physically yield hidden layers. The problem list helps as
in problem oriented charting. Still not universally used, it at least provides
a list of the elements. Sir Wm. Osler, professor of medicine at the U of
Pennsylvania, 1884, Johns Hopkins, 1888, Oxford 1905[3]
and the acknowledged father of modern internal medicine, suggested that the physician,
in dealing with multiple problems not subscribe multiple diagnoses but rather
look for an underlying cause. He also suggested that the physician’s primary
role was convincing patients to take fewer medicines; hardly in keeping with
today’s `Medicine Wagon` cornucopia of expensive drugs.
Most planners and non-clinicians miss the degree to which
diagnosis, treatment and medical knowledge change over time, like an amoeba
moving this way and that sometimes rapidly and other times slowly. The changes
come with new knowledge, with current fashion and with genomic reinterpretation
of the pathophysiology itself. Furthermore,
patients vary greatly as individuals. Geographical regions, even small ones,
experience significantly different epidemiological and environmental problems.
Aggregating all of the known diagnoses and syndromes in the
World into one database with criteria presents a formidable task. It would have
not have been attempted just a few years ago due to limitations in storage
capacity. That limitation no longer exists. Accumulating such an open source
universal database for diagnostic CDS may go a long way in improving diagnostic
accuracy. However, health care institutions and vendors may avoid incorporating
a resource sensitive volume of material into their database system.
There are many reasons for the problem of diagnostic
inaccuracy. As mentioned one is the
changing criteria and understanding of the etiology of disease with advances in
genomics proteinomics and micro-molecular biology. Other problems result from
the insurance clerk’s attempt to list a diagnosis compatible with the
requirements for testing and or reimbursement. Another results from the snap
diagnosis associated with high throughput, efficiencies insisted upon by
profit-motivated management. There must be some significant reasons for the
dismal ranking of US medicine compared with the rest of the world. Not to beat
a dead horse, but a ranking of 36th in infant mortality implies
something more than mal distribution of providers and over-treatment. I submit
that diagnostic CDS will go a long way towards closing the gap.Table 1Diagnostic Error Rates from the literature[4]
Tuberculosis / Autopsy 50%
Pulmonary Embolism 55%Aortic Aneurism 61%
Sub Arachnoid Hemorrhage 30%
Breast Cancer / Mammogram 21%
Bipolar Disorder 69%
Appendicitis 18%
Cancer Path 2-9% GYN; 5-12% non-GYN
Endometriosis 18%
Psoriatic Arthritis / Standard Patient 39%
Atrial Fibrillation / EKG Machine 35%
Infant Botulism 50%
Diabetes Mellitus 18%
Chest XR by / ER 18%
No comments:
Post a Comment