1

Thank you very much for the reference, Mr. Carter!

2

Anyone have a link to the military's sheet of instructions on how to make a baptismal font? Just curious, would like to see.

Powerful story.

3

On the electoral college count, there's also this site: http://2.004k.com/trend/

4

Regarding the home pregnancy tests, there are, I think, three problems here. One is that the tests are, truly, pretty bad when used in the earliest days after a missed period. From the original article:

Although [pregnant women's] levels of hCG varied widely, most levels were too low for most tests to detect [one day after a missed period]. Only First Response Early Results, which sells for about $12, would detect more than 95 percent of pregnancies.

Only three of the 18 tests produced a positive result when testing urine that contained the amount of hCG typically present during the second day after a woman's first missed period. By the third day, eight tests appeared effective.

"If one waits a week after the missed period," said Cole, "any home pregnancy test will do." . . .

The second is that the FDA regulations make no realistic demands for either accuracy or truth in labeling for the tests:

So how can manufacturers claim the tests are 99 percent effective? A small asterisk on the test packaging refers consumers to a notation saying the claim is based on "typical" hormone levels. Cole says the claims are misleading.

An FDA spokesman said manufacturers are required only to prove their tests kits are as effective as a previously sold test kit in order to make that claim -- but there is nothing preventing them from using a 1970s test as a comparison.

"I didn't write the law. I only follow it," said Steven Gutman, a doctor and director of the FDA office of in-vitro diagnostics. He said Cole's research has sparked "some internal discussion" at the FDA about ways to craft "more appropriate language."

The tests are "99%" effective in detecting any pregnancy that would have been detected by a test sold 30 years ago - not 99% effective in detecting any actual pregnancy. Truth in advertising, FDA-style.

Finally, another problem that I strongly suspect is entering in here - though it is not discussed in the article - is the distinction between accuracy of a positive result and accuracy of a negative result. As medical statisticians know, but many doctors forget and almost none of the public has even heard, "accuracy" is not a reasonable evaluative standard for a medical test. Careful and honest evaluations are based on the "predictive value" of the test, which is not the same thing.

For any simple yes/no test, like a dipstick pregnancy test, you can have four possible outcomes - a true positive result, a false positive result, a true negative result, or a false negative result. Someone who talks about the "accuracy" of a test is almost always referring to the "sensitivity" - the percentage of positive results that are actually detected. This is the ratio of true positive results to the total of true positives plus false negatives (e.g., of all the pregnant women who take the test, how many will it actually say are pregnant?). But people taking a test don't want to know "If I'm really pregnant, what are the odds that this test will say so?" - they want to know something completely different: for any given test result positive or negative, what are the odds that result is true? ("If this test says I'm pregnant, what are the odds that I really am?" "If this test says I'm not pregnant, what are the odds I'm really not?") For those you need the ratio of true positives [or negatives] to all positives [or negatives] true or false, known as the "positive predictive value" or "negative predictive value" of the test.

It is possible for a test to have both high sensitivity and high "specificity" (percentage of all negative results that are reported accurately) but not have a high predictive value for one or the other. The reason is that most test populations have a large percentage of either positive or negative individuals; this means that, while the percentage of false negatives or positives may be low, the actual numbers of false results in one direction or the other will be high, because there is a large group from which those results can come.

Now, assuming that the pregnancy test is 99% "sensitive" (it correctly reports pregnancy for 99% of the pregnant women who take the test) and 99% "specific" (it correctly reports negative results for 99% of the non-pregnant women who take the test), and also that a large percentage of women who use the test are in fact pregnant (they wouldn't take the test if they didn't already have reason to believe they were), what will happen? It will report as pregnant 99% of the pregnant women and 1% of the relatively low percentage of non-pregnant women - meaning that almost all of the women it reports as pregnant really are pregnant, and the test is "accurate" in their case. But it will report as non-pregnant 99% of the truly non-pregnant women and also 1% of the relatively large group of women who really are pregnant - meanting that, not most, but quite a few of the women who take the test and get a negative result really are pregnant, even though the test has 99% accuracy in detecting non-pregnancy.

[Take an example: 10,000 women take the test; 90% are truly pregnant and 10% are truly not; the test has 99% sensitivity and 99% specificity. So: 9,000 women are pregnant, and the test reports 99% = 8,910 as pregnant and 1% = 90 as non-pregnant; 1,000 are not pregnant and the test reports 99% = 1,990 as non-pregnant and 1% = 10 as pregnant. That gives (8,910 + 10) reported as pregnant, with 8,910 of those being actually pregnant: there is a probability of (8,910 / 8,920) = 99.9% that a postive test result is accurate (the "positive predictive value" of the test). At the same time, there are (1,990 + 90) reported as non-pregnant, of which 1,990 are actually not pregnant. This gives a probability of (1,990 / 2,080) = 95.7% that any given negative result is correct (the "negative predictive value" of the test). Thus, about 1 in 23 women who take the test and get a negative result will actually be pregnant, while only 1 in 1,000 women who get a positive result will actually be non-pregnant, even though the test is 99% accurate for both positive and negative results. This is because the population pools of women who could be reported falsely either positive or negative are so imbalanced.

This is true for all tests where the true-positive and true-negative pools are very different percentages of the population. Testing of groups for which there is some independent reason to believe they are likely to be positive - such as self-selected samples of women who have missed a period, or already-diagnosed populations in a treatment clinic - always produces large numbers of false negatives. Contrarily, testing large populations for which there is no reason to think that any given individual is positive - such as mass-screening programs for rare diseases - invariably turns up large numbers of false positives. This is why mass-screening programs for the general population, which are often very popular, are discouraged by health authorities.]

So, for these tests, the manufacturers may be able to honestly report that the test is "99% accurate" in detecting results (although the article suggests it is not even accurate in that sense at the time that many women are likely to use it), without mentioning that this does not mean that any given result is 99% likely to be true. That, of course, is what women really want to know, but that is a technical issue that confuses even many health professionals, let alone the general public, and about which they don't inform you. Given that women using these tests have a much higher likelihood of being pregnant than women in the general population, the higher concentration of true positive women increases the likelihood of false-negative results, which is from most women's point of view the worst possible result. Somehow the manufacturers can't find room on the package to mention that fact.