Most medical experts in medical malpractice are influenced by bias. Yet, they produce a certificate of merit. There are many types of biases, but I concentrate on only those biases that represent a systematic distortion that directly undermines the reliability of certainty in data and are measured by Type I and Type II errors.

Under this circumstance, merit can be regarded as certainty minus those biases. Those biases are selection bias, avoidance bias, and confirmation bias. Every medical intervention is influenced by these biases and has a measurable merit whether the null hypothesis is true or false.

The three biases of merit

Selection bias is the probability of selecting incorrect data so that a null hypothesis is retained or rejected as the case may be. When a null hypothesis is retained, selection bias, or the probability of selecting incorrect data that would reject a null hypothesis that might be true, is the same as Type I error. However, when a null hypothesis is rejected, selection bias is 100 percent minus the Type I error.

Avoidance bias is the probability of avoiding correct data so that the null hypothesis is rejected or vice versa as the case may be. When the null hypothesis is retained, selection bias or the probability of avoiding correct data that would reject a null hypothesis that might be false, is the same as Type II error. However, when a null hypothesis is rejected, selection bias is 100 percent minus the Type II error.

Confirmation bias is the probability of confirming a wrong conclusion. Confirmation bias is the average of selection bias and avoidance bias.

Merit is the probability that data are reliable.

Merit = 100% – (selection bias + avoidance bias + confirmation bias).

Interestingly, when attorneys claim that all the cases they represent have merit, they are being disingenuous, but they are not wrong. All cases have merit; attorneys just do not measure it. When merit is 0 or has a negative value, there is no substance in data which support the context of their conclusion. Claim as they may, these are frivolous lawsuits.

The problem with permissible inference

In court, the plaintiff attorney always has the burden of proof and the defense attorney needs only to cast doubt on that proof. However, in medical malpractice litigation, there is “permissible inference.” If there is no conclusive proof, permissible inference permits “triers of fact,” or jurors, to conclude medical malpractice based on preconceived notions, which are biases.

Permissible inference is always at play in medical malpractice litigation; however, nowhere is it more apparent than with res ipsa loquitur. When using abductive or inductive reasoning, plaintiff attorneys infer from circumstantial evidence alone that an injury of the nature described would ordinarily not occur without a departure from the standard of care. Jurors are allowed to decide using “permissible inference.”

Nevertheless, negligence should always have proof. Yet, determining res ipsa loquitur using abductive or inductive reasoning only requires decision-making based on a preponderance of evidence, which, in hypothesis testing, corresponds to a Type I error of 50 percent. Ergo, res ipsa loquitur, which is always a medical error, is indistinguishable from a random error of nature, which is never a medical error.

Hypothesis testing vs. legal reasoning

When Type I error is 50 percent, Type II error, or 100 percent power, is usually 12.5 percent. Essentially, “permissible inference” ignores a Type I error of 50 percent, a Type II error of 12.5 percent, and their attendant selection, avoidance, and confirmation biases.

Unlike abductive or inductive reasoning, which infers res ipsa loquitur, hypothesis testing proves it. There are four features:

• First, at least one phase in the medical intervention has an incident risk no less than 100 percent.
• Second, the medical intervention in question and the medical intervention in the standard of care are the same by definition of “null hypothesis.” Therefore, both independent variables represent res ipsa loquitur.
• Third, according to this condition, to prove res ipsa loquitur, the null hypothesis must be retained. When both independent variables correspond to res ipsa loquitur, the probability of rejecting a true null hypothesis is 0 percent. Therefore, alpha must be 0. Since the p value is always greater than 0, the null hypothesis is always retained.
• Fourth, when alpha is 0, beta is also 0. When the null hypothesis is retained and alpha = 0, merit = 100% – (selection bias + avoidance bias + confirmation bias) which equals 100% – (0+0+0). Hence, merit is 100 percent, as it should be.

However, when the null hypothesis is retained and alpha = 0.5 and corresponds to preponderance of evidence, merit = 100% – (50% + 12.5% + 31.25%) = 6.25%.

A new standard for the certificate of merit

Although statistical literacy is an accepted standard in both the medical and the legal professions, jurors are not expected to be statistically literate. However, they do understand the phrase “whatever gets measured, gets managed.” When merit in the plaintiff’s argument is 6.25 percent and it should be 100 percent, how would a rational person decide?

There will always be medical malpractice lawsuits but, when merit is measured, imagine the impact a “certificate of merit” would have when it quantifies merit. Needless to say, permissible inference, selection bias, avoidance bias, and confirmation bias will no longer play a role in legal proceedings. Other biases remain, but they are societal and cultural in nature. They are beyond the parameters of hypothesis testing, and are best left to voir dire and other challenges.

Statistics do not change the game. Statistics level the playing field.

Howard Smith is an obstetrics-gynecology physician.