Improving Diagnosis in Healthcare; What can Health IT do?

Written by Charles Sneiderman, MD PhD

A new Institute of Medicine report on medical error estimates that 5% of US adults who get outpatient care each year are incorrectly diagnosed. Thus the majority of adults will experience a medical diagnostic error during their lifetime. The IOM definition of diagnostic error is failure to establish an accurate and timely explanation of a patient’s health problem or to communicate that explanation to the patient.(1)

I believe that the method used to estimate the frequency of diagnostic error in this report is badly flawed. I say this because the major data source for the studies summarized was medical records. Outpatient medical records, even in this era of electronic record-keeping are notoriously incomplete. Clinicians get paid for providing care, not for documenting it. Medical records are highly distilled notes to remind the clinician of the patient’s findings and treatment in a language which is highly abbreviated and codified.

Diagnosis is an open-ended process; clinicians are required to report an interim conclusion summarized with a codified label after each encounter with a patient if they want to get paid. Patients usually seek medical care either because they have experienced a symptom (like pain), a sign (like rash or fever), or been told that they have some risk to their health that needs further evaluation. Clinicians are trained to look for patterns of symptoms, signs, and other findings like abnormal laboratory or imaging results and match them to patterns of known diseases. Usually patients will have a pattern of findings that match more than one disease and the clinician has to decide which is most likely and whether there is any additional testing that can improve the certainty of that conclusion. This intellectual process is rarely documented and only the codified conclusion is generally recorded, because that information is required For reimbursement. Ironically just this month, Medicare and other insurers are now requiring healthcare providers to code using a new version of the International Classification of Diseases (ICD10) which is an order of magnitude more detailed than the previous version.

Analysts who have not had extensive clinical training and experience naively assume that this codified product is “diagnosis” rather than the process that produced that conclusion. Thus it is hardly surprising that many diagnoses are incorrect; disease processes evolve over time. My job as a primary care physician in the community is to decide whether my patients are sufficiently safe to wait as the process evolves. I have to live with a high degree of uncertainty. Depending on the risk of death or or serious disability, we accept a social cost of sending some people with chest pain to the hospital to find out it was only indigestion or remove a normal appendix in people with lower abdominal pain. One of my favorite sayings is “Through the retrospectoscope, you can always see 20/20”. Sometimes a finding that was either subliminally present on a previous encounter or was presumed to fit in another pattern in retrospect fits better in the pattern that matches the disease that a patient now “clearly has”. In our minds we generate a “differential diagnosis” — a list of possible explanations for the findings to date. In ambulatory practice this is rarely documented in the medical record. Nevertheless, diagnostic errors do occur. My previous blog on medical error addressed how the systematic application of known rules of patient care can be incorporated by health IT and in particular EMR systems.

One of the first demonstrations that computerized reminders improve physician adherence to clinical practice guidelines was published in 1977 by Clem McDonald’s group in Indiana.(2)

Clem had postdoctoral training at the Massachusetts General Hospital in Octo Barnett’s lab; Octo had developed the Computer Stored Ambulatory Record (COSTAR) which was also adopted by the Family Medicine Department at the Medical University of South Carolina where I did my residency in those years. I became a disciple and joined the Lister Hill National Center for Biomedical Communication (LHNCBC) at the National Library of Medicine after residency and my “postdoc” became a three decade career! Ironically Clem left Indiana to become scientific director of the LHNCBC a year before I retired from NLM in 2010.

Computer assistance in medical diagnosis has a long history. When I first began my career at the National Library of Medicine over thirty five years ago, we funded a research project at the University of Pittsburgh in which the wisdom of Dr. Jack Myers, a highly respected diagnostician was programmed into a statistical matrix which, when fed the findings of the New England Journal of Medicine’s “case of the week” came up with the correct diagnostic conclusion more often than most recent graduates of medical residency training like myself at that time. There have been numerous such programs developed over the intervening years, but none have had widespread adoption because the process to input all the findings had to be done manually. The final frontier of computer assisted diagnosis is being explored today with the use of natural language processing to extract the findings from the text of clinicians notes in EMRs. Hopefully I will live to see the Holy Grail.

1. Balough EP, Miller BT, Ball JR, et al., eds., Improving Diagnosis in Health Care, Washington: National Academy of Sciences, 2015.
2. McDonald CJ, Murray R, Jeris D, Bhargava B, Seeger J, Blevins L. A computer-based record and clinical monitoring system for ambulatory care. Am J Public Health. 1977 Mar;67(3):240-5.