Saturday, November 17, 2012

A Fork in the Road


Share |

Medicine, the profession struggles with its identity. Will we be physicians in the likes of the great scientist philosophers and humanists since Hippocrates? Or, will we succumb to the trend of commodifying the profession? It is clear that health care planners intend to move us in the direction of standardized care. The extent to which the profession restores its leadership, humanity and science will largely depend upon how it deals with the coming universality of computerized patient records.
Meaningful Use requirements for incentive pay under the Affordable Care Act and various agencies will mandate certain quality metrics in the Electronic Health Record. (EHR) Evidence of compliance may be required by the uniform claims submission process for reimbursement.[1] Programmers will likely build these metrics into the software developed by various vendors of the EHR. Doctors are familiar with Best Evidence and some of those treatment standards are already appearing in EHRs. A new feature will be Clinical Decision Support (CDS). Best Evidence largely deals with treatment protocols. CDC will deal mostly with diagnosis. What follows is a discussion about how the later, CDS, deals with differential diagnosis and the role of the physician in an electronic data environment.

The review of systems (ROS) and the differential diagnosis lie at the heart of the data processing that attempts to deal with these two intellectual processes. Each item in the ROS, signs and symptoms related by the patient, becomes critical data in the process to follow. The negative response is as significant as the positive -- maybe more. Each positive response, when confirmed by further questioning, evokes a list. The list is critical as well and might as well be indelibly imprinted behind each data point or on the mind of the physician. When completing the ROS, one then has a number of responses, each with its list of etiological possibilities. One or more diagnostic possibilities appearing on separate lists tend to point to the underlying problem and contribute to the establishment of a tentative differential.
Here is where statistics comes into play The coupling of signs and symptoms with a medical information database, can assure the clinician that he or she did not overlook the less obvious problem.  The more often a diagnosis appears on multiple lists and the more completely the signs and symptoms on the ROS fulfill the attributes of a diagnosis, the more likely you are on the right track. However, all overlapping signs and symptoms are not equal and many illnesses have similar symptoms. Furthermore, individual patients seem to have a limited range of symptoms to account for a wide range of possible illnesses. Patients also react to illness in a highly individualized way.

Ongoing and real-time statistical analysis of the differential diagnosis compared with outcomes can validate the initial list of possibilities. Thus the coupling together of patient data with the vast store of medical information, can assist the clinician with his or her interpretation of the data. Furthermore, the computer generated differential can focus statistically on the more likely candidate diagnoses. Database mining software can provide further relationships that the mind cannot grasp or that have not previously been observed.

Playing the percentage game, however, can be risky. An old adage states, “If you hear hoof beats in the hall, don’t think of zebras.” The Internist might ask, “Is there a circus in town?” The problem with rare diseases is that there are so many of them. Encountering a rare condition unexpectedly is a common occurrence. Adding further to the dilemma, the easy diagnosis often masks an underlying cause. Examples abound: Valvular Heart Disease, Shingles, Atrial Fibrillation, and Duodenal Ulcer to mention only a few of the more obvious ones all have underlying causes. Diagnosis is thus challenged on multiple levels. Physicians must peal away the layers of the diagnostic onion. Programing information technology to dig down to those multiple levels of scrutiny may be tricky indeed.
Some would advocate a process in which the computer and pre-defined procedures defined by best evidence as written by experts, take the prime role and relegate the physician to insuring the validity of the data points. The rationale to this approach lies in the frequency of diagnostic error and the never ending expansion of diagnostic and treatment procedures that do more harm than good costing more and more.

Others would say, we need to go back to the physician as the humanist and scientist of old with vast experience and ongoing medical education in a professionally structured society, dedicated to excellence. Today we are at a cross roads. The road to subordination, however, may have already been taken. Whichever path we follow, patient safety, access and consistent in depth handling of the data are primary. The differential diagnosis is critical to achieving anything like the health metrics of the rest of the industrial world. --- We are currently something like 37th in perinatal mortality and longevity.
Man thinks, the computer remembers. The physician’s mind does far better at complex judgment and interpretation than any computer to date. The mindful and highly experienced physician sees the above analysis and assembles a differential diagnosis with a scanning logic. Most physicians navigate a differential diagnosis quite well. The differential diagnosis, however, demands a high degree of commitment to the art, medical education and extensive exposure to most every form of disease. Our mind thinks and makes individualized judgments that the computer cannot. The computer on the other hand remembers and does statistics. We do not remember so well and we do not manage statistics well on a large database.[2]

For the statistical linkages between patient data and the proven diagnosis to be valid, they must be relevant to the patient. Do you depend on statistics developed on a national scale, an international scale or limited to a local population that might be more relative to the patient at hand? I say might be because each patient is a one of a kind individual. The only statistics that matter might very well be the patient's own genomics. The individual genome, despite the nail biting, is falling in price and will soon be ubiquitous. We need to be sure the tests are not sold like snake oil by bathers, naturopaths, charlatans and opportunists.

The psychological or psychiatric diagnosis proves elusive for the computer generated history, whereas, the physician may grasp it intuitively. A ROS with phases of the MMPI[3] covering all classes of behavioral and psychological issues could be helpful. The biggest mistake is to tell a patient, “It’s all in your head;” that’s not a diagnosis. Again the physician must dig down to an underlying cause. Be suspicious. Clinical interaction must continuously measure outcome and look for errors. --- It may be difficult to teach a computer to be suspicious.
The present day inconsistencies in both diagnosis and treatment have lead to a movement among health care planers to standardize care. Indeed there are many missed and wrong diagnoses.[4] Misses are almost inevitably due to omitting the ROS or ignoring one or two of the responses when they do not fit the assumption. Larry Weed[5], the inventor of the problem oriented record (POMR), insists that any positive, not accounted for by the diagnosis, belongs on the active problem list. Weed further advocates that the physician’s training is backwards and that he or she should be trained to exacting and standardized skill in history taking, examination and technical surgical skills while leaving the basic science and pathology to the experts delivered by computer but with the provider guiding the patient in making the decisions in a compassionate way. This supposes that information technology will provide the best diagnosis, the best evidence and the best treatment leaving the decisions largely to the patient. This sentiment is further advanced in John Wennberg’s book, Tracking Medicine.[6] In which he documents over treatment and inconsistent treatment demonstrating superior outcomes in regions of less intense and less specialized care.

Arguably, we should do the thinking and follow a strategy in which the computer remembers the data and the physician exercises judgement based on the individual patient and that patient’s choices. The physician concentrates on the science, the judgment and the art of medicine. Bring the patient to the fore of decision making. The computer tracks lists and analyses statistically on a real-time basis. Embrace the benefits of HIT and CDS. Retain control of the data, however, at all costs. Let the computer couple your precise clinical data with the vast store of medical terminology, nosology, and the salient features of each. The value here is in not missing something and not getting stuck in a wrong assumption. The analysis should present to the physician a credible preliminary differential. With or without assisted memory and coupling to the vast store of medical knowledge, the ROS is key to patient safety and care. Without extensive branching logic, though, the computer administrated ROS may create variable data. We have all experienced the patient responding differently on different occasions, to different physicians or to the nurse. I think the responsibility for securing a valid database remains with the physician. Less experienced clerical questioners may fail to elicit valid data.
Physicians and Medical Schools struggle with a conflict between physicians in the classic role of science, art, diagnosis and a newly proposed role in which the physician practices in a role subordinate to the computer, algorithms, protocols, check lists and the authors of Best Evidence.

At the heart of the conflict lie the complex process of diagnosis and the choice of individual treatment. To meaningfully apply the dictate of Best Evidence, one needs the right diagnosis. We suggest that the best application of computers, statistics will be achieved when the computer applies its faultless memory and capacity for data with dynamic real-time statistical analysis, not on mass populations, but on the internal data at hand. This strategy provides data more relevant to the individual patient and keeps the physician in control of the data. The strategy also leaves the physician and the patient with the choices, the judgment and the mindfulness of that data. The physician thinks, the patient participates, the computer remembers.

Yogi Berra once famously said, "When you come to a fork in the road, take it!" As a profession, we should have it both ways, the physicians science and art plus the computers memory.


[1] Agencies include: Health and Human Services (HHS), National Institution of Health (NIH), Centers for Medicaid and Medicare Services (CMS), Insurance companies, and regional Health Information Exchanges (HIE),
[2] We do need to take into account the computer’s future capabilities, however. So far the computer does not think, but IBM's Watkins comes close.
 [3] Minnesota Multiphasic Inventory
[4] The American Journal of Medicine, Vol 121, (5A) May, 2008
“Overconfidence as a cause of Diagnostic Error in Medicine,”  ES BERNER & Mark L. Graber, M.D. American Journal of Medicine, 2008 Vol 121 S2-S23
[5] Medicine in Denial, Lawrence L. Weed and Lincoln Weed April 2011, Amazon
[6] Tracking Medicine, John Wennberg, 2011, Amazon

No comments:

Post a Comment