PSA results vary considerably due to several factors. The first is random lab error which is always present because nothing can be measured with 100% accuracy.

The second factor is called systematic error. These are errors that result because one lab may use a different analytical technique (assay) for reading the PSA, or a particular labs may calibrate differently.

Lastly, other events may, and do, influence the PSA level in the blood. Ejaculation, bike riding, or any manipulation of the prostate, such as a DRE, all have the tendency to raise the PSA in the blood. Furthermore, there are theories that the time of day a sample is drawn and the day within a 28 day cycle may affect the PSA level. Also, inflammation of the prostate (prostatitis) is also a frequent source of raised PSA.

To understand the significance of any particular PSA reading all of the above factors must be considered. We are usually interested in "relative" readings more then absolute readings. E.g., did my PSA significantly rise or fall from the last reading.

We can, and should, do certain things to eliminate many of these errors. Not having sex (of any kind that results in ejactulation), not bike riding, and not having a DRE within 48 hours of having blood drawn for a PSA test is a simple step to eliminate that source of error.

Another step within our control is to use the same lab and the same assay at that lab for all of our PSA test. That eliminates the systematic error induced by a particular labs technique, in most cases.

Although not proven yet as a source of error, just to be sure, we can insist that the PSA blood sample be drawn at the same time of day, and even at the same point in a 28 day cycle.

If we are concerned about a substantial raise in PSA over the last reading, we can eliminate prostatitis as a cause through the use of by antibiotics (CIPRO) and possibly anti-inflammatory (NSAID)drugs such as Aleve, Ibuprofen, or perscription drugs like Celebrex. (See NSAIDs, Statins and Prostate Cancer).

All of the above are things that are within our ability to control. We may have to insist that our doctors or labs allow us to do it, e.g., schedule appointment when we want, not when the lab wants, or insist that our doctors write Rx for anti-inflammatory and antibiotics, and that they do what is necessary to check out prostatitis. But, with a little effort, we can control this factors. NOTE: If your doctor will not cooperate, change doctors.

The one factor we cannot control is random lab error. Periodically a survey is taken of 1000's of labs to detect their random error in measuring PSA. Identical blood samples are sent to all labs for a PSA reading. In the study available to us at this time, six samples were sent to over 2500 labs, each sample contained a blood sample of a different PSA level from about 0.2 to 19.4. The results reported be each lab were analyzed to obtain the mean reading, the standard deviation from the mean for each lab and for each PSA level (of the six different samples sent to each lab). This allowed for the determination of a 95% confidence range -- a range around the mean value reported that there is a 95% chance the real PSA value falls within (hence, 1 in 20 reported readings will be out of this range).

The actual data, combined for all labs, even though they used different assays, is given in the table below. These random error range is higher then it would be if each lab were considered separately. Various labs use different brand instruments with different precision (random errors and detection limits). Various labs use different calibration standards and from different batches. This leads to a systematic relative error for each lab. These systematic errors, when pooled from 2600 labs, behave like random errors in the study. This is why all errors (systematic and random) can be combined. Thus, so-called standard deviations (SD) can be calculated, SD's only applying to random errors, not systematic errors.

For a given lab, any systematic error (due to inaccurate calibration) does not show up when doing repeated analysis on one blood sample. We would only see the random errors. Constant relative systematic errors are not a problem when we are interested in the PSA trend and this is the case when we always use the same lab. (I am ignoring here that within one lab, the relative systematic error may change when a switch is made to a different batch of calibration standards.)

	95% Confidence # of Labs Low Med High Mean SD %rekSD Range 2672 10.8 19.4 34.25 19.67 2.14 10.9 15.39- 23.95 2653 7.2 9.8 18.0 9.92 1.11 11.1 7.70-12.14 2689 5.3 7.3 12.8 7.36 0.79 10.7 5.78-8.94 2509 2.1 3.0 4.7 3.03 0.33 10.8 2.37-3.69 2504 0.6 0.7 1.5 0.73 0.11 14.5 0.51-0.95 2591 0.1 0.2 0.8 0.24 0.10 40.2 0.04-0.44

Explanation of the columns:

# of Labs - The number of participating laboratories

Low - the lowest reported PSA value

High - the highest reported PSA value

Med - the middle number of all the results

Mean - the average of all (about 2600) reported PSA values

SD - standard deviation from the Mean, a measure for the degree of spread of results

%relSD = 100 x SD / Mean, the relative SD in %.

The last two columns give the so-called 95% confidence intervals. For example, the first sample has a Mean PSA of 19.67 ng/ml and a SD of 2.14 ng/ml. The true PSA remains unknown, but there is a 95% chance (probability) that is lies between 15.39 and 23.95. The latter "from-to" values are calculated from Mean - 2 SD and Mean + 2 SD respectively.

At higher PSA values, we see that relative errors (%relSD or CV) are fairly constant, about 11%, whereas at low values, the absolute errors (SD in ng/ml) are fairly constant, about 0.10. (This behavior is quite normal in many analytical techniques). For a Mean of 0 ng/ml, the SD is more like 0.08. This implies that the detection limit is 0.16 (2 times SD at 0 PSA).

The meaning of this is best given by examples:

Suppose that your blood sample is measured by one of the 2600 labs and you do NOT know which lab that is.

EXAMPLE FOR HIGH PSA

Your PSA is reported as 15.0 ng/ml

Because this is a relatively high value, you must consider RELATIVE errors (in %). Taking 2 relative standards deviations (%relSD) both ways, you get a 95% chance that the TRUE PSA is between 15.0 - 22% and 15.0 + 22%, that is between 11.7 and 18.3 ng/ml.

EXAMPLE FOR LOW PSA

Your PSA is reported as 0.30 ng/ml

Because this is a relatively low value, you must consider ABSOLUTE errors (in ng/ml). Taking 2 standard deviations both ways, you get a 95% chance that the TRUE PSA is between 0.30 - 0.16 and 0.30 + 0.16, that is between 0.14 and 0.46 ng/ml.