If you are arrested for drunk driving, you will usually be given either a breath or a blood test.
The breath test is far more commonly given in DUI cases in most jurisdictions; it is quick, cheap, easy and gives immediate results. In many cases, however, a blood sample will be withdrawn – typically at the police station if a phlebotomist is available, at a law enforcement crime laboratory, or at a private lab contracting with law enforcement. This may be because a breath machine is unavailable, because the suspect is being treated at a hospital for injuries from an auto accident, or – increasingly – blood is forcefully taken where the suspect refuses to breathe into the machine.
Generally speaking, the blood test is more accurate than the breath test. See, for example, How Breathlayzers Work (and Why They Don't). But blood testing in DUI cases has its own sources of inaccuracy. See, for example, Fermentation in Blood Samples Produce…Alcohol and Can Coagulation of the Blood Sample Raise the Alcohol Level?
One source of error in blood testing, however, is encountered in the situation mentioned where the DUI suspect has been involved in an automobile accident and has been taken to a medical facility for treatment. A blood sample may be withdrawn for diagonostic reasons — during which blood-alcohol levels may be determined and given to law enforcement. Or the arresting officer accompanying the drunk driving suspect may direct hospital personnel to extract a blood sample for alcohol analysis at the hospital. In either case, there exists a built-in source of error — which may lead to the conviction of an innocent citizen.
From my book Drunk Driving Defense (7th edition):
Blood samples obtained in drunk driving cases are generally — but not always — analyzed as whole blood (sometimes called "legal blood"). If the sample is analyzed for medical purposes, however, the test will probably be done with serum (often referred to as "medical blood"). Serum is the clear yellowish fluid obtained from separating whole blood into its solid and liquid components (usually by centrifuging the sample); the liquid portion of the blood is called plasma, which is similar to serum. A third method involves precipitating proteins from the blood sample and centrifuging it; the result is a clear liquid called supernatant which is then analyzed.
Will analysis of serum/plasma or supernatant result in the same blood-alcohol readings as analysis of the whole blood? In a study entitled "Distribution of Ethanol: Plasma to Whole Blood Ratios" (Hodgson and Shajani, 18 Forensic Science Journal 73, 1985), scientists attempted to determine the answer to this very question. The conclusion: Blood-alcohol concentrations in plasma were approximately 11 percent higher than that of whole blood, and those in supernatant were about 5 percent higher….
For a study that found that serum-alcohol concentration can be up to 20 percent higher than blood-alcohol concentration, see Frajola, "Blood Alcohol Testing in the Clinical laboratories: Problems and Suggested Remedies", 39(3) Clinical Chemistry 377 (1993).
Bottom line: Any method of analyzing the amount of alcohol in a DUI case is subject to wide-ranging sources of error.