404
Editor's column
3. Diaz J, Schain RJ, and Bailey BG: Phenobarbital-induced brain growth retardation in artificially reared rat pups, Biol Neonate 32:77, 1977. 4. Diaz J and Schain RJ: Phenobarbital: effects of long-term administration on behavior and brain of artificially reared rats, Science 199:90, 1978.
The Journal of Pediatrics September 1979
5. Cranbach LJ: in Murphy G, editor: Essentials of psychological testing, ed 2, New York, 1960, Harper and Brothers, pp 181-189. 6. Anastasi A: Psychological testing, ed 4, New York, 1976, The Macmillan Company, pp 242-245.
Commentary: The cut-off point for TSH measurement or recalls in a screening program for congenital hypothyroidism using primary 7"4screening I y v I E w of the increasing n u m b e r of screening programs for congenital hypothyroidism b e i n g implemented in North America and involving primary T 4 screening backed up by a TSH measurement. '-3 the choice of the cut-off point for recalling infants or measuring TSH in spots with low T4 values is becoming an important issue? The goal is to minimize the n u m b e r of false positive measurements while not missing any hypothyroid infants. STATISTICAL
INFERENCE
The problem is to discriminate between two populations: the hypothyroid and normal infants (Figure). If the cut-off point is C. the frequency of false negatives is equal to a and the frequency of the false positives is equal to fi (Table I). In terms of screening, 1-a is called the sensitivity of the diagnostic test: that is. the proportion of hypothyroid infants diagnosed correctly as abnormal. If 1-a is high (close to one). the test is recognized as sensitive for the disease. 1-fi is called the specificity of the test; thus. if l-fi is high the test is considered specific for that disease. We have first of all to determine how blood T~ concentrations are distributed in the two populations. In our program the distribution of the unaffected population is log-normal with a m e a n T4 level of 1.73 n g / b l o o d spot and a standard deviation of 1.35 ng!spot? The mean varies from day to day, due mainly to technical factors; thus results are standardized daily to normalize mean and
1Iol. 95, No. 3, pp. 404-406
X
Figure. The curve on the right is the distribution of the logarithm of T~ concentration from the normal population. The one on the left is the distribution of screened hypothyroid T, concentration. C represents the cut-off point, a and fi are, respectively, the proportion of false negative and false positive cases. v a r i a n c e values so ~,at they can be pooled from different assays. If we consider the hypothyroid infants as an homogeneous population, m a k i n g no distinction between thyroid agenesis, hypoplastic and ectopic glands, we have a Abbreviations used T~: thyroxine TSH! thyroid-stimulating hormone T~: tri-iodothyronine population which includes 72 patients; the m e a n T4 is 0.52 ng/spot with a standard deviation of 0.188 ng/spot, the distribution seems normal rather than log-normal, and a test of goodness of fit test does not reject this assumption. However, the n u m b e r of samples is small. Estimation of
0022-3476/79/090404+03500.30/0 9 1979 The C. V. Mosby Co.
Volume 95 Number 3
Editor's column
Table I. Hypothesis
405
Table II
Infant having hypothyroidism
Normal infant
sD
fillse positive*
false negative
Probability of missed ease (per million)
0.10 0.05 0.03 0.025 0.018 0.011
0.0003 0.003 0.013 0.016 0.034 0.072
0.06 0.6 2.5 3.2 6.7 14
1-a
T~ below the cutoff point
Proportion of hyProportion of normal infants diagpothyroid infants diagnosed nosed as sick as sick ...................................... test ......................................
T~ above the cutoff point
Proportion of hypothyroid inrants diagnosed as well
1-fi Proportion of norreal infants diagnosed as well
the hypothyroid population distribution would be more accurate it we could use data from all screening programs; unfortunately, these data are not available. Even if they were, it would be impossible to pool them since different centers do not use exactly the same methods. PROBABILITY
OF MISSING
A CASE
Table II summarizes the values of a, the frequency of false negatives and the value of fi, and the frequency of false positives for various ciat-off points. F o r example, if the cut-off point is at --2.1 SD from the geometric mean, fi is equal to 0.025 and a = 0.016. According to the conditional probability theory, the probability of missing a case is a multiplied by the frequency of the disease, which in our population is i/5,000. T h e n 1 -x 0.016 5000 gives 3.2 per million newborn infants screened. To be more accurate in our estimation of missed or false negative cases, we should consider the different types of hypothyroidism and their respective populations. Those hypothyroid infants more likely to be missed are those with ectopic or hypoplastic glands. Because of residual thyroid tissue and an elevated serum TSH level (thus a possible increase in T4 secretion), these infants might be missed in a primary T4 screening program using a low cut-off p o i n t : This argument has been put forward in Europe, where primary T S H screening is widely employed:. ~ In Quebec we have detected eight hypothyroid infants with ectopic glands, four infants with dyshormonogenetic goiters (organification defect), and four patients with transient hypothyroidism secondary to maternal treatment with antithyroid drugs (Table III). In our procedure all of these infants were detected, even those with an eventual normal serum T~ concentration at
1.3 1.7 2.0 2.1 2.3 2.6
1.17 1.03 0.95 0.92 0.86 0.79
*Equivalent to percentile cut-off point from the normal distribution.
which is deviant of about 0.007
three weeks of age. Only one infant (Patient 2 with - 2 . 2 SD) was close to our cut-off point of 2.1 SD. The mean filter paper spot T4 concentration for these infants was 0.62 ng/spot with a standard deviation of 0.23 ng/spot. This value is higher than the m e a n concemration in our population with thyroid agenesis (mean 0.47 ng). However, the probability of missing these infants is not necessarily higher, since the probability is also a function of the frequency. When first screened, these two populations are treated similarly with regard to the TSH measurement. Thus our estimation on the probability of missing a case of thyroid agenesis is overestimated at 3.2 for a million screened infants (--2.1 SD for the geometric mean). Only one infant was not detected, but was referred at three days of age because of a large goiter. At that time he had a normal T4 c0ncentraton, which progressively fell to an abnormal level (Patient 12). For our network the probability of missing an infant due to the set of the cut-off point for filter paper Spot T, values, is three infants over a ten-year period (100,000/ year). On the other hand. the frequency of false negative values by human error is: E M(1-a)I where E, M, and I, are, respectively, the n u m b e r of missed cases, the n u m b e r of screened infants, and the incidence. There are only two screening centers with a large sample volume who do not use a computerized filing system: Oregon and New England. These two centers have reported three cases missed by h u m a n error over 540,000 screened infants. Therefore the frequency of false negative results is 0.028, which is slightly less than the false negative rate due to the cut-off point at 2.3 SD. Thus the probability of missing a case by h u m a n error is about 6 x 10 '~. Furthermore. if we change the cut-olf point from 2.5% to 10%, the probability of missing a case decreases by about 3 • 10 '*; this amounts to about half
406
Editor's column
The Journal of Pediatrics September 1979
Table III. T h y r o i d f u n c t i o n o f d e t e c t e d n e w b o r n infants w i t h t h y r o i d dysgenesis, d y s h o r m o n o g e n e t i c goiter, a n d i a t r o g e n i c - i n d u c e d t r a n s i e n t h y p o t h y r o i d i s m
Serum
Filter T, paper spot Patient Ectopic thyroid
1 2 3 4 5 6
Dyshormonogenetic goiter
latrogenic
8 9 10 11 12" 13 14 15 16
ng/401~l [ SD'~ [ TSH (~u/40~O 0.60 0.90 0.60 0.78 0.56 0.71 0.53 0.80 0.29 0.49 0.66 1.2 0.60 0.57 0.22 0.46
3.5 2.2 3.5 2.6 3.7 3.0 3.9 2.5
Editor's column
3. Diaz J, Schain RJ, and Bailey BG: Phenobarbital-induced brain growth retardation in artificially reared rat pups, Biol Neonate 32:77, 1977. 4. Diaz J and Schain RJ: Phenobarbital: effects of long-term administration on behavior and brain of artificially reared rats, Science 199:90, 1978.
The Journal of Pediatrics September 1979
5. Cranbach LJ: in Murphy G, editor: Essentials of psychological testing, ed 2, New York, 1960, Harper and Brothers, pp 181-189. 6. Anastasi A: Psychological testing, ed 4, New York, 1976, The Macmillan Company, pp 242-245.
Commentary: The cut-off point for TSH measurement or recalls in a screening program for congenital hypothyroidism using primary 7"4screening I y v I E w of the increasing n u m b e r of screening programs for congenital hypothyroidism b e i n g implemented in North America and involving primary T 4 screening backed up by a TSH measurement. '-3 the choice of the cut-off point for recalling infants or measuring TSH in spots with low T4 values is becoming an important issue? The goal is to minimize the n u m b e r of false positive measurements while not missing any hypothyroid infants. STATISTICAL
INFERENCE
The problem is to discriminate between two populations: the hypothyroid and normal infants (Figure). If the cut-off point is C. the frequency of false negatives is equal to a and the frequency of the false positives is equal to fi (Table I). In terms of screening, 1-a is called the sensitivity of the diagnostic test: that is. the proportion of hypothyroid infants diagnosed correctly as abnormal. If 1-a is high (close to one). the test is recognized as sensitive for the disease. 1-fi is called the specificity of the test; thus. if l-fi is high the test is considered specific for that disease. We have first of all to determine how blood T~ concentrations are distributed in the two populations. In our program the distribution of the unaffected population is log-normal with a m e a n T4 level of 1.73 n g / b l o o d spot and a standard deviation of 1.35 ng!spot? The mean varies from day to day, due mainly to technical factors; thus results are standardized daily to normalize mean and
1Iol. 95, No. 3, pp. 404-406
X
Figure. The curve on the right is the distribution of the logarithm of T~ concentration from the normal population. The one on the left is the distribution of screened hypothyroid T, concentration. C represents the cut-off point, a and fi are, respectively, the proportion of false negative and false positive cases. v a r i a n c e values so ~,at they can be pooled from different assays. If we consider the hypothyroid infants as an homogeneous population, m a k i n g no distinction between thyroid agenesis, hypoplastic and ectopic glands, we have a Abbreviations used T~: thyroxine TSH! thyroid-stimulating hormone T~: tri-iodothyronine population which includes 72 patients; the m e a n T4 is 0.52 ng/spot with a standard deviation of 0.188 ng/spot, the distribution seems normal rather than log-normal, and a test of goodness of fit test does not reject this assumption. However, the n u m b e r of samples is small. Estimation of
0022-3476/79/090404+03500.30/0 9 1979 The C. V. Mosby Co.
Volume 95 Number 3
Editor's column
Table I. Hypothesis
405
Table II
Infant having hypothyroidism
Normal infant
sD
fillse positive*
false negative
Probability of missed ease (per million)
0.10 0.05 0.03 0.025 0.018 0.011
0.0003 0.003 0.013 0.016 0.034 0.072
0.06 0.6 2.5 3.2 6.7 14
1-a
T~ below the cutoff point
Proportion of hyProportion of normal infants diagpothyroid infants diagnosed nosed as sick as sick ...................................... test ......................................
T~ above the cutoff point
Proportion of hypothyroid inrants diagnosed as well
1-fi Proportion of norreal infants diagnosed as well
the hypothyroid population distribution would be more accurate it we could use data from all screening programs; unfortunately, these data are not available. Even if they were, it would be impossible to pool them since different centers do not use exactly the same methods. PROBABILITY
OF MISSING
A CASE
Table II summarizes the values of a, the frequency of false negatives and the value of fi, and the frequency of false positives for various ciat-off points. F o r example, if the cut-off point is at --2.1 SD from the geometric mean, fi is equal to 0.025 and a = 0.016. According to the conditional probability theory, the probability of missing a case is a multiplied by the frequency of the disease, which in our population is i/5,000. T h e n 1 -x 0.016 5000 gives 3.2 per million newborn infants screened. To be more accurate in our estimation of missed or false negative cases, we should consider the different types of hypothyroidism and their respective populations. Those hypothyroid infants more likely to be missed are those with ectopic or hypoplastic glands. Because of residual thyroid tissue and an elevated serum TSH level (thus a possible increase in T4 secretion), these infants might be missed in a primary T4 screening program using a low cut-off p o i n t : This argument has been put forward in Europe, where primary T S H screening is widely employed:. ~ In Quebec we have detected eight hypothyroid infants with ectopic glands, four infants with dyshormonogenetic goiters (organification defect), and four patients with transient hypothyroidism secondary to maternal treatment with antithyroid drugs (Table III). In our procedure all of these infants were detected, even those with an eventual normal serum T~ concentration at
1.3 1.7 2.0 2.1 2.3 2.6
1.17 1.03 0.95 0.92 0.86 0.79
*Equivalent to percentile cut-off point from the normal distribution.
which is deviant of about 0.007
three weeks of age. Only one infant (Patient 2 with - 2 . 2 SD) was close to our cut-off point of 2.1 SD. The mean filter paper spot T4 concentration for these infants was 0.62 ng/spot with a standard deviation of 0.23 ng/spot. This value is higher than the m e a n concemration in our population with thyroid agenesis (mean 0.47 ng). However, the probability of missing these infants is not necessarily higher, since the probability is also a function of the frequency. When first screened, these two populations are treated similarly with regard to the TSH measurement. Thus our estimation on the probability of missing a case of thyroid agenesis is overestimated at 3.2 for a million screened infants (--2.1 SD for the geometric mean). Only one infant was not detected, but was referred at three days of age because of a large goiter. At that time he had a normal T4 c0ncentraton, which progressively fell to an abnormal level (Patient 12). For our network the probability of missing an infant due to the set of the cut-off point for filter paper Spot T, values, is three infants over a ten-year period (100,000/ year). On the other hand. the frequency of false negative values by human error is: E M(1-a)I where E, M, and I, are, respectively, the n u m b e r of missed cases, the n u m b e r of screened infants, and the incidence. There are only two screening centers with a large sample volume who do not use a computerized filing system: Oregon and New England. These two centers have reported three cases missed by h u m a n error over 540,000 screened infants. Therefore the frequency of false negative results is 0.028, which is slightly less than the false negative rate due to the cut-off point at 2.3 SD. Thus the probability of missing a case by h u m a n error is about 6 x 10 '~. Furthermore. if we change the cut-olf point from 2.5% to 10%, the probability of missing a case decreases by about 3 • 10 '*; this amounts to about half
406
Editor's column
The Journal of Pediatrics September 1979
Table III. T h y r o i d f u n c t i o n o f d e t e c t e d n e w b o r n infants w i t h t h y r o i d dysgenesis, d y s h o r m o n o g e n e t i c goiter, a n d i a t r o g e n i c - i n d u c e d t r a n s i e n t h y p o t h y r o i d i s m
Serum
Filter T, paper spot Patient Ectopic thyroid
1 2 3 4 5 6
Dyshormonogenetic goiter
latrogenic
8 9 10 11 12" 13 14 15 16
ng/401~l [ SD'~ [ TSH (~u/40~O 0.60 0.90 0.60 0.78 0.56 0.71 0.53 0.80 0.29 0.49 0.66 1.2 0.60 0.57 0.22 0.46
3.5 2.2 3.5 2.6 3.7 3.0 3.9 2.5
