Terminate Terminology Terror in EMS Research

When shall we three meet again? In thunder, lightning or in rain? A little awkward, I thought to myself as I sat at my desk in Mr. Bebbington's Grade 7 English class, reading Shakespeare for the first time.

When the hurlyburly's done, when the battle's lost and won. What the heck does hurlyburly mean? That will be ere the set of sun. Huh? Ere? Ok, time to turn to Coles Notes. Well, 10 years later, I found that if there was anything more frustrating to read and understand than a Shakespeare tragedy, it must surely be a medical journal.

Scientists, for better or worse, love their terminology. Scientific publications are scattered with terms foreign to most native speakers of the English language. This can make it difficult for clinicians who browse the literature to grasp the message of any research paper and determine if the results are applicable to the patients they encounter. For those of you terrorized by terminology, here's a guide to the "s" word: statistics. Feel free to print this out and stick it in your protocol book.

Normal Distribution

All data is distributed across a range. For example, if we took all the response times in your area, we would find a few really fast ones (the call next door to the station!) and a few really slow ones (driving across town in a snowstorm). But most of the calls will be centered in the middle. We call this a normal distribution. Sometimes data is not normally distributed. It is skewed away from the middle. Consider time-to-backboard. Usually, we can backboard people quickly, within a few minutes. But sometimes, the fire department will need an hour to extricate the patient, thus skewing the distribution in one direction. This creates outliers. Outliers are results that are far away from the expected.

Average

This term can be misleading. Here is an example: 2, 4, 6, 8 and 40. These numbers represent how many pairs of shoes five people report owning. We have an outlier who owns 40 pairs of shoes, which skews the distribution of the data. We can report this with a mean or median, and get drastically different "averages."

Mean

To determine the mean, add up all the responses and divide by the number of responses. In our example, the mean is 12. However, this is not really reflective of how many pairs of shoes people in our sample own. The outlier has skewed the data, and the mean is not representative of the average. Mean is used when the data is normally distributed. Standard deviation measures dispersion or how close or far responses are to the mean. One standard deviation represents where about 70% of the results fall. A low standard deviation indicates that data points tend to be very close to the mean, whereas a high standard deviation indicates that data are spread out over a large range of values.

Median

Arranging values from lowest to highest, take the middle response to determine median. In our example, the median is 6. This measure is appropriate when data is not normally distributed. Range measures the dispersion of data by reporting the highest and lowest figure. In our example, the range is 2 to 40. Interquartile range is the range of the 25th and 75th percentile. It eliminates high and low outliers that skew the data by showing us where the middle 50% of values lie.

Type 1 (Alpha) Error

This is when a conclusion renders a false positive, the belief that there is a difference between two findings when in fact there is no difference. For example, thinking drug A is better than drug B, when in fact they are equally beneficial (or equally harmful!).

Type 2 (Beta) Error

This is when your conclusion renders a false negative, believing there is no difference when in fact there is. For example, you might conclude defibrillator A doesn't save more lives compared to defibrillator B, when in fact it does.

P Value

The "probability value," also known as significance, quantifies the probability that an observation is due to chance and not an actual difference. In other words, it describes the probability of making a type 1 error. In medicine, a P value of 0.05 is the highest allowable P for results to be considered "statistically significant." A P of 0.05 means there is a 95% chance the results are actual and not caused by chance. Statistically significant results must be analyzed by clinicians for clinical significance. If fentanyl decreases pain by 30%, and morphine decreases pain by 32%, are we going throw out all the fentanyl? Probably not. Although these results may be statistically significant, with a P value of less than 0.05, they (in my mind) don't justify throwing out the fentanyl.

Confidence Interval

This describes the possible variation of a value within the margin of acceptable alpha error. For example, the odds of death for patients treated by circus clowns (compared to paramedics) may be 2.0* with a confidence interval of 1.8 to 2.2. This means that the odds of death are twice that for people treated by clowns, and any value between 1.8 and 2.2 has a P value of less than 0.05 and is considered statistically significant. If a confidence interval spans 1 (i.e., 0.8-1.4), the P value is greater than 0.05. (8The author has no evidence to support or refute the claim that clowns are harmful.)

Odds Ratio

This value compares the odds of experiencing an outcome between two groups; for example, the odds of death in smokers compared to nonsmokers, or the odds of survival in a control group compared to an experimental group. An odds ratio of 1 means the two groups experience the event of interest (death, survival, etc.) equally. An odds ratio greater than 1 means the first group experiences the event more than the second group. An odds ratio of less than 1 means the first group experiences an event less often than the second group.

Relative Risk

This calculation compares the probability (rather than the odds) of experiencing an outcome between two groups. A relative risk of 1 means there is no difference between the two groups. A relative risk of less than 1 means the event is less likely to occur in the experimental group than in the control group. A relative risk greater than 1 means the event is more likely to occur in the experimental group than in the control group.

Odds Ratio vs. Relative Risk: What's the Difference?

The odds ratio and relative risk compare the likelihood of an event between two groups. Let's use the Titanic survivors as an example. There were 462 female passengers: 308 survived and 154 died. There were 851 male passengers: 142 survived and 709 died.

The odds ratio calculates the odds of death for passengers on board the Titanic as follows. Females faced odds of 2 to 1 against dying (154/308=0.5). The odds of death for males was 5 to 1 (709/142=4.993). The odds ratio is 9.986 (4.993/0.5). There was a tenfold greater odds of death for males than for females.

Relative risk compares the probability of death instead of the odds of death. The probability of death for females was 33% (154/462=0.3333). The probability of death for males was 83% (709/851=0.8331). The relative risk of death was 2.5 (0.8331/0.3333), meaning males had a probability of death 2.5 times greater than females.

The choice to use an odds ratio or a relative risk is complicated and depends on the study design and question being asked. Think twice about any reported odds ratio or relative risk before interpreting the findings of a study.

Whether reading The Tragedy of MacBeth or the Annals of Emergency Medicine, it's important to consider the message of the text. When in doubt, do a quick dictionary search on the Internet to make sure you have interpreted the message correctly. After all, you'd hate to get lost in the plot. Fair is foul and foul is fair: Hover through the fog and filthy air. I'll leave the interpretation of that line up to you...

Blair Bigham, ACP, MSc, is an advanced care flight paramedic and prehospital science investigator in Toronto. When not roaming the streets and skies practicing clinical paramedicine, he studies resuscitation and patient safety related to EMS and transport medicine. Trained at Sunnybrook Health Sciences Centre and the University of Toronto, he now studies lifesaving interventions at Rescu, the world-renowned prehospital research program at St. Michael's Hospital and U.T. Contact Blair at bighamb@smh.ca.