Say what?

Dr. Barton (orthopedist): "Give the patient 75 mg of morphine for pain."

[ILLUSTRATION OMITTED]

[ILLUSTRATION OMITTED]

Ms. Stevens (charge nurse): "Do you mean 75 mg of morphine or 75 mg of meperidine?" (She questions the order but reads it back as she got it.)

Barton: "Oh, thanks. I was thinking of something else. Yes, meperidine." (The read back caught the error.)

Stevens: "And her five grains of quinine each night?" (Apothecary units)

Barton: "Fine. And be sure to check her weight every day. She was 148 pounds when she came in, and the cardiologist told me that we need to keep her under 155." (English units)

Stevens (looking at the chart): "She was 68.3 kg this morning." (Metric units)

The slip is common. What might happen if the slip involved a decimal point in a child--0.5 mg of morphine vs. 5 mg? Or 0.025 mg of digoxin instead of 0.25 mg? And if the nurse was afraid to ask or is only used to working with adult dosages?

Different units

While pleasantly old fashioned and relate to the "art" of medicine, different units are a prescription for errors. Quickly, how many milligrams in a grain? If you give a patient 100 cc/hour of IV fluid, how many pounds will this be in a day?

Imagine you are a new night nurse on a medical floor. You see NTG 1/150 gr. ordered for a patient. Is that 1/150 of a grain, which is 0.4 mg, or is it 1/150 of a gram, which is about 7 mg? Most physicians think the answer is obvious. But non-physicians may feel differently--and they are giving the medication.

Remember the Mars Climate Orbiter that disappeared from communication shortly after it was due to land? The issue turned out to be a programming glitch between metric and English units.

A medical calculating test was administered to nurses, finding that nearly half scored less than 70 percent. Frequent errors occurred when there was more than one calculation and when milligrams were converted to grains. Calculators are of little help with orders of magnitude difference, such as micrograms to milligrams. (1)

The famous "Gimli glider" incident, where a B-767 ran out of fuel over Ontario was due to use of a correction factor of pounds, rather than kilograms, when fueling the aircraft. (2)

Three rechecks each missed the fact that kilograms, not pounds, match with liters. Multiple rechecks, if done exactly the same (wrong) way, may not catch errors.

More slips

What if nobody catches the slip? What if the hierarchy is so strong that nobody dares to correct a mistake?

Hierarchy issues were involved in an aircraft crash in Portland when the pilot was so pre-occupied with a possible landing gear problem that he didn't realize the plane was running out of fuel. The flight engineer knew the problem but was unable to convince the rest of the crew of the severity. (3)

When performing a familiar process automatically, we are prone to slips, especially if interrupted. Have you ever driven on a familiar road leading to a different destination? It is easy to end up missing the proper turn, especially if distracted. Such slips are called "habit capture." (4)

All of us have these annoying experiences daily. These normal slips in automatically performed functions may become issues in practice. We may think we gave an order but didn't, or vice versa.

Distortion

One theory of medical decision making is that physicians make diagnoses on the basis of the underlying probability of the disease, the presence of certain symptoms and the sensitivity and specificity of the specific diagnostic tests involved.

The theory is logical, elegant and--wrong.

Causality dominates much of our routine problem solving--one problem causes another. Here are two useful rules: (5)

1. When possible, use physiological principles and causal relations to solve diagnostic problems. In other words, the better you know anatomy, physiology, and pathology, the better diagnoses you will tend to make.

2. Probabilistic diagnoses are helpful when causal or statistical relations between clinical variable are uncertain or unproved.

Example: If I have a heart attack, I will have chest pain. But if I have chest pain, that doesn't mean I'm having a heart attack.

A particular cancer is present in 0.1 percent of the population; that is, one in a thousand. A test for the disease is 90 percent sensitive and 90 percent specific, so that for every 100 who have the disease 90 will test positive.

For every 100 who don't have the disease, 90 test negative. A patient tests positive. What is the probability the test is a false positive?

a. 1 percent

b. 30 percent

c. 75 percent

d. More than 90 percent

Of the 1008 patients with positive tests, 9, or about 1 percent, have the disease and 99 percent are false positive!

[ILLUSTRATION OMITTED]

Interpret all positive tests in light of the prevalence of the disease! Remember that the probability of a disease given a positive test is proportional to the baseline probability of the disease before you perform the test. We are not normally Bayesian thinkers, but there are situations where the approach is useful. (6)

Corollary 1: Using Bayesian techniques requires a complete and inclusive set of mutually exclusive possibilities. While not likely one may obtain a set, if it is possible to do so, many potential traps may be avoided.

Corollary 2: The sensitivity and specificity of a test depend upon a "gold standard" of diagnosis, which is dependent upon equipment (imaging, for example), operator skill or the stage of disease. (6,7)

Corollary 3: If the prevalence of a condition is low, under 1 percent, even with high sensitivities and specificities, the likelihood of a positive test being false positive is considerable. If the sensitivity and specificity are lower, which they often are, the likelihood is even greater.

Corollary 4: Since we don't pick patients at random from the population but rather classify them into high or low probabilities of having a disease, we need to use that prior knowledge along with the test's sensitivity and specificity to get an idea of whether the test is a false positive.

Example: You have reasonable prior knowledge of a patient's probability of having syphilis with a positive VDRL if the patient is an IV drug abuser with a history of pelvic inflammatory disease vs. whether the patient is a nun who is concerned about having syphilis because of impure thoughts. (8) Would you not treat a positive VDRL differently in these two women?

Corollary 5: The post-test probability of a disease after a negative test is a function of the sensitivity. High sensitivity--negative test--Essentially excludes diagnosis. If the sensitivity is low, a negative result slightly lowers the probability of having the disease. (5)

Example: A physician was diagnosed as having HIV/AIDS on the basis of two ELISA tests and one Western blot test on one blood sample. His only risk factor was being around infected patients. His life was put on hold for several days--indeed, he was told he had AIDS--until a more confirmatory test turned out to be negative.

Prevalence of 0.1% (10 in 10000), sensitivity and specificity each 90%:

Disease/Test  Positive           Negative           Totals

Positive      9                  1                      10
              (90% Sensitivity)
Negative      999                8,991               9,990
                                 (90% Specificity)
Totals        1,008              8,992              10,000

References:

1. Bogner, MS, Human Error in Medicine. Lawrence Erlbaum Associates, Hillsdale, NJ, 1994, p. 353.

2. www.wadenelson.com/best.html

3. www.airsafetyonline.com/accidents/ayers/173.shtml

4. Reason, J. Human Error. Cambridge University Press. Cambridge, 1990.

5. Kassirer, JP and Kopelman, RI. Learning Clinical Reasoning, Williams and Wilkins, Baltimore, 1991.

6. Harris, JM. "The Hazards of Bedside Bayes," JAMA, 246(22), 1981.

7. Chang, PJ. "Bayesian Analysis Revisited: A Radiologist's Survival Guide," Am. J. Radiol. 152: 721-727, 1989.

By Michael S. Smith, MD, MS

Michael S. Smith, MD, MS, a statistician, wants to help people in the medical community use statistics to make better, faster and easier decisions. He is self-employed and may be reached at 520-410-7917 or [e-mail]