Heterogeneous distribution of serum prolactin values in apparently healthy young women, and the effects of oral contraceptive medication*†

FERTILITY AND STERILITY Copyright c 1987 The American Fertility Society

Vol. 47, No.5, May 1987 Printed in U.SA.

Heterogeneous distribution of serum prolactin values in apparently healthy young women, and the effects of oral contraceptive medication *t

John B. Josimovich, M.D.t§ Marvin A. Lavenhar, Ph.D.11 Mona M. Devanesan, M.D.t Hee-Ja Sesta, M.D.t Sidney A. Wilchins, M.D.t Arlene C. Smith, M.s.t University of Medicine and Dentistry of New Jersey (UMDNJ), New Jersey Medical School, Newark, New Jersey

Controversy over effects of oral contraceptives (OCs) on serum prolactin (PRL) levels from retrospective studies suggested performing a prospective study. Statistical analyses of PRL levels in 552 reproductive-age, non medicated women indicated a provisionally lognormal distribution of values < 15 ng/ml, contaminated by a small number of abnormally high values ~ 90 ng/ml. Truncated samples were used to estimate a "normal range" of PRL levels for three subsets of the study sample, classified according to number of weeks after pregnancy. Fifty-microgram estrogen-containing OCs doubled basal PRL levels at 5 to 8 weeks in those whose initial control values fell below 15 ng/ml, but the PRL elevation was no longer evident at 6 months of drug use. These OCs induced a small but significant lowering of PRL at 5 to 8 weeks in those with control levels of 15 ng/ml or higher. Thirty-five-microgram estrogen-containing OCs failed to alter PRL levels at 5 to 8 weeks in those with control values < 15 ng/ml. Fertil SteriI47:785, 1987

Received August 5, 1986; revised and accepted January 9, 1987. *Presented in part at the Fortieth Annual Meeting of The American Fertility Society, April 2 to 7, 1984, New Orleans, Louisiana. tSupported in part by a grant to J.B.J. from the Foundation of the UMDNJ (25-79), the Hunterdon Health Fund, and the National Institutes of Health grant 5 SOl RR0593 (Biomedical Research Support grant); and by funds from an anonymous donor to S.A.W. :j:Department of Obstetrics and Gynecology. §Reprint requests: John B. Josimovich, M.D., Department of Obstetrics and Gynecology, UMDNJ-New Jersey Medical School, 185 South Orange Avenue, Newark, New Jersey 07103. IIDepartment of Preventive Medicine and Community Health. Vol. 47, No.5, May 1987

Controversy exists over whether or not oral contraceptive (OC) use might increase the incidence of clinically symptomatic prolactin (PRL)secreting chromophobe adenomas of the pituitary gland. One paperl reviewed 13 studies, most composed of small numbers of patients, in which OC-induced hyperprolactinemia was found in 4 but not in the other 9 studies. The review also presented data on a larger group of 123 patients taking OCs for 6 months or more, in whom 30% were found to have hyperprolactinemia. Interpretation of the data presented in the 13 studies is complicated by the fact that a "normal range" of serum PRL levels is not easily defined. Because clinical studies of the effects of bromergocriptine,

Josimovich et aI. Distribution of PRL values and DC effect

785

r a PRL suppressor, suggest that the drug may benefit certain infertile patients whose serum PRL levels are still usually within the "normal range,,,2 any large group of apparently healthy women may contain a minority whose PRL secretion control might vary from that found in the majority. The women in this minority may benefit from closer follow-up, as suggested by Jeffcoate,3 who found that PRL levels are not normally distributed, at least before logarithmic transformation of data. The current study was undertaken (1) to determine the extent to which log PRL levels in women of childbearing age follow a normal distribution, (2) to estimate a "normal range" ofPRL levels, and (3) to measure the effect of age, pregnancy history, and the type, amount, and duration of OC use on PRL levels. MATERIALS AND METHODS HORMONE ASSAYS

Blood samples were separated and the serum frozen at - 20°C for up to 2 years before assay. Radioimmunoassay tests were performed at least twice on thawed sera by a double-antibody system provided in Prolactin Tests Kits (Serono Laboratories, Inc., Randolph, MA) according to instructions provided with the kits. Intraassay variation was 6.0% and interassay variation 7.1%.

The majority of patients (468) were used only in the statistical analyses of PRL level distribution, while the other 84 patients were used in both the distribution studies and the study on the effects of OCs. Of the 84 patients given OCs, 59 were presented OCs containing 50 J.Lg estrogen. Fifty-two of these received formulations containing 1 mg norethindrone (44 [75% of the 59] combined with 50 J.Lg ethinyl estradiol [EE 2], 8 combined with 50 J.Lg mestranol), while 4 patients received 1 mg ethynodiol diacetate with 50 J.Lg EE 2, and 3 took 0.5 mg norgestrel with 50 J.Lg EE 2. Twenty-five patients who received lower-estrogen-dose OCs were divided as follows: 19 received 0.4 mg norethindrone plus 35 J.Lg EE 2; 2 were given 1.0 mg norethindrone plus 35 J.Lg EE 2; 2 were given 0.5 mg norethindrone plus 35 J.Lg EE 2; and 2 received 0.3 mg norgestrel plus 30 J.Lg EE 2. All 84 OC patients had a second sample of blood drawn for PRL determination 5 to 8 weeks after initiating therapy. Twelve patients who received 50-J.Lg containing OCs had a third sample drawn at 6 to 8 months. All patients found to have elevated PRL levels were requested to have a repeat blood sampling and if higher PRL levels persisted, OCs were discontinued and the patients referred to a gynecologic endocrinologist. In addition, 55 patients and personnel volunteers who did not receive OCs underwent a second blood sampling 1 week to 2 months after initial control blood sampling.

SUBJECTS INCLUDED FOR STUDY AND MEDICATIONS USED

A total of 703 samples was drawn from 552 subjects between 10:00 A.M. and noon. Of these, 12 samples were drawn from nine subjects at the Hunterdon County, NJ, branch of Planned Parenthood-Northwest Jersey, while the remainder were obtained from clients of the Family Planning Clinic and personnel volunteers at the UMDNJ-University Hospital, Newark, NJ. Subjects were requested to sign an informed consent and were excluded from participating if the following were found: history of breastfeeding during the past week or galactorrhea; use of medications other than aspirin or acetaminophen; use of steroidal contraceptives within 3 months; pregnancy less than 2 weeks previously; surgery or physical trauma within the past month; drug or alcohol abuse within the past year; or hypertension. 786

STATISTICAL METHODS

MUltiple regression analyses were applied to assess the association between six independent variables and baseline PRL levels (or log-transformed PRL levels) in 269 women 6 weeks or more postpregnancy or never pregnant. The following independent variables were employed: age (years); parity (0, 1, or 2 or more pregnancies); history of abortions (0, 1, or 2 or more); history of menstrual irregularity (yes, no); history of OC use (none, 35-J.Lg, 50-J.Lg estrogen dose); duration of prior (but not within 3 months) OC use (months). The effect of the number of weeks elapsed postpregnancy on baseline PRL levels was evaluated by Kruskal-Wallis test and by Dunn's approximate procedure for multiple comparisons based on Kruskal-Wallis rank sums. 4

Josimovich et aI. Distribution of PRL values and OC effect

Fertility and Sterility

Chi-square tests of goodness of fit were employed to determine whether the observed distributions of PRL levels could be characterized by normal or lognormal distribution. These tests and graphic analyses suggested that the observed PRL distributions could be considered a truncated sample from a lognormal population. Therefore, estimates of the mean ± standard deviation (SD) of serum PRL levels were calculated according to the procedures described by Bliss5 for analyzing provisionally normal data. These estimates were used to establish a "normal range" of PRL levels. McNemar's chi-square test for matched pairs was applied to assess whether the proportion of elevated serum PRL levels changed significantly 1 week to 2 months after the initial blood sampling, among 55 patients who did not receive OCs. The effect of OC use on serum PRL levels in women 6 weeks or more postpregnancy or never pregnant, whose baseline levels were within 2 SD of the recontructed mean estimates, was evaluated by means of Student's paired t-test or Wilcoxon's signed rank test (small samples). Similar tests and covariance analysis were made to compare values of those women given 50-fJ.g estrogen OCs with those untreated whose initial values lay beyond 2 SD of the reconstructed curve mean. Further studies were carried out to determine whether there had been a significant change in baseline PRL values after time in those with initially elevated levels when given no medication or given 50-fJ.g estrogen OCs or by applying Wilcoxon's signed rank test and covariance analysis.

RESULTS Multiple regression analyses could detect no significant associations between PRL levels (or log PRL levels) with the age, parity, number of abortions, history of irregular menses, or the type, amount, and duration of past (no more recent than 3 months) OC use among 269 women in the study sample who were 6 weeks or more postpregnancy or never pregnant. The distribution of initial serum PRL levels in 272 subjects 6 weeks or more postpregnancy or never pregnant is shown in Figure 1A indicating a marked skewing above the arithmetic mean of 10.8 ng/ml. The log-transformed values shown in Figure 1B were less skewed above a geometric mean of6.5 ng/ml. However, chi-square goodnessof-fit tests revealed significant departures from a Vol. 47, No.5, May 1987

40

A

30 cr UJ CD

:. ~ 20

50 85

95

30

a:

UJ CD

:. ::> z

PRL lng/mil

Figure 1 Distribution of PRL levels in 272 women of reproductive age. (A), Arithmetic distribution. (B), loglo distribution.

normal or lognormal distribution. These preliminary analyses suggested that serum PRL levels might be provisionally lognormal, with "normal" measurements contaminated by a small number of abnormally high measurements. Examination of the distribution of initial logserum PRL levels by number of weeks postpregnancy showed a progressive decrease in median levels ranging from 12 ng/ml at 2 to 3 weeks postpartum to 7.8 ng/ml at 6 or more weeks postpregnancy. Multiple comparisons based on Kruskal-Wallis rank sums suggested the division of initial serum PRL values into three groups: 175 patients who were 2 to 3 weeks postpartum, 99 who were 4 to 5 weeks postpregnancy, and 272 who were 6 weeks or more postpregnancy or never pregnant. A graphic analysis of the distributions of log-PRL levels in the first two groups showed that they were similar to the log-PRL levels in the third group, suggesting that all three samples were provisionally normal with contamination by some unknown number of abnormally high values.

Josimovich et al. Distribution of PRL values and OC effect

787

r

I

Table 1. Reconstruction of Normal Curves of Control Serum PRL Values at Different Times After Pregnancy

Group

Original no. of patients

Truncation point serum PRL level

Theoretical geometric mean of reconstructed normal curve, Jl

Theoretical 1 SD of Jl = ~

8.8 8.9 9.1 10.5 7.5 7.7 8.8 9.1 6.3 6.5 7.2 7.5

1.6 1.6 1.6 1.7 1.5 1.5 1.6 1.7 1.5 1.5 1.6 1.7

Jl +

2~

% values be-

yond Jl + 2~

Test of normality reconstructed curve,P

28.6 27.4 23.4 15.4 32.3 31.3 11.1 11.1 24.3 22.4 12.9 11.0

0.45 0.66 0.78 0.14 0.38 0.59 0.62 0.73 0.50 0.60 0.03 0.02

ng/ml

2-3 wk post-

175

pregnancy

4-5 wk post-

99

pregnancy

6 + wk postpregnancy or never pregnant

272

10.0 12.6 15.8 20.0 10.0 12.6 15.8 20.0 10.0 12.6 15.8 20.0

Assuming that the parent distributions were lognormal, the provisionally normal sample groups were truncated at several arbitrary points beyond the geometric mean, and the mean ± 2 SD of each reconstructed lognormal curve were estimated. With these estimates, the observed and expected frequencies of the truncated samples were compared to determine whether the observed distribution could be considered a truncated sample from a lognormal population. The results are shown in Table 1. The optimal point of truncation for each of the three study groups was considered to be the highest truncation point at which the estimates of the mean ± 2 SD remained stable and the goodness-of-fit of the lognormal distribution was not significant. On the basis of these criteria, the best estimates of the parameters of the reconstructed lognormal distributions expressed in original units (nanograms per milliliter) for the three study groups are summarized in Table 2. In Figure 2, the lognormal curve, reconstructed by truncating the original observations at 12.6 ng/ml in the 6 + weeks postpartum group, is superimposed on the original sample distribution of 272 patients, shown in Figure lB. Table 3 demonstrates that among the 55 patients who did not receive OCs and were retested 1 week to 2 months after the initial blood sampling, the proportion of elevated serum PRL levels differed significantly by chi-square tests (P < 0.025), depending on whether their initial levels were normal or elevated. Only 4 of the 36 patients with normal initial levels had elevated follow-up levels, whereas 15 of the 19 patients who had elevated initial levels had elevated follow-up levels. 788

20.9 21.5 22.6 30.5 16.9 17.5 23.2 25.6 13.7 14.4 18.5 20.9

Table 4 shows the statistically significant increase from a geometric mean PRL level of 6.3 to 11.6 ngiml after 5 to 8 weeks 50-j-Lg estrogencontaining OCs in 33 women. Nevertheless, values dropped almost to original control levels (7.48 ng/ml) in the 12 of these women followed for 6 months of OC use. Use of 30 to 35-j-Lg estrogen OCs failed to show a significant change in PRL levels in 21 women. When the effects of 50-j-Lg estrogen-containing OCs over 5 to 8 weeks in 26 patients with PRL levels elevated according to the criteria defined by Table 2 (range, 15 to 90 ng/ml) were compared with the changes over 5 to 8 weeks in 19 women with initially elevated levels (range, 15 to 92 ng/ ml) receiving no medication, mean changes of -12.6 ng/ml in the medicated group versus -' 5.8 ng/ml in the nonmedicated group were found. By covariance analysis (dependent variable log of the follow-up PRL level; independent variables log control PRL level, and presence or absence of OC medication) patients receiving medication had a significantly lower (P = 0.04) PRL level after 5 to 8 weeks than those given no medication. By Wilcoxon's Signed Rank test, PRL levels in nonmedicated group did not significantly change with time (P = 0.06, two-tailed), while a highly significant reduction in PRL was seen after 5 to 8 weeks of OCs (P = 0.0002, two-tailed). Table 2. Geometric Mean (fl) SD (~) and Upper Limits of Normal (+2~) of Reconstructed Normal Values in this Study Group Group

2-3 wk postpartum 4-5 wk postpartum 6 + wk postpartum

Josimovich et al. Distribution of PRL values and DC effect

Jl +

Jl 9.1 7.7 6.5

1.6 1.5 1.5

2~

22.6 17.5 14.4

or never pregnant

Fertility and Sterility

I

/l=6.5 ng/ml

t

30

0:

w

20

ED

:. :::>

z

to

2.0

4.0

6.0

to

20

40

60

tOO

PRL lng/mil

Figure 2 Superimposition of reconstructed normal distribution curve by use of truncation point from Table 2 on the log distribution of PRL samples seen in Figure lB.

DISCUSSION

Past confusion about whether significantly elevated PRL levels are seen in women taking OCs l might have depended on the method of assay, the time in relation to pregnancy in which control samples were taken, the distribution of values around a mean, or even the exclusion of far outlying values. The current study confirms3 that there is a more normal distribution of serum PRL values in women when they are logarithmically transformed (Fig. 1) and that there is a significantly lower geometric mean at 6 + weeks or more postpregnancy than at 2 to 5 weeks postpregnancy (Table 1). Most important, there is a bimodal or polymodal distribution of serum PRL levels in blood samples drawn randomly in late morning (Figs. 1 and 2) so that even though 95% of all values might fall below 20 to 25 ng/ml in a population of reproductive age women, such a population consists of a majority whose logtransformed values follow a statistically normal distribution below a level of 15 ng/ml. A large minority of women in this age group do not fit this distribution, however. That 11% of 36 women with initial values < 15 ng/ml (Table 3) fall at higher levels on follow-up (15 to 18 ng/mD may be due to the later sample having occurred during an episodic PRL surge. 6 Of great interest was the finding that 79% of 19 women with initially elevated levels continued to have values of 15 ng/ml or more, 1 or more weeks later. In 4 of 11 of these in whom more complete histories and breast examinations were carried Vol. 47, No.5, May 1987

out at a later date, there was also the development of oligomenorrhea and/or galactorrhea. Table 4 shows a temporary doubling in serum PRL levels (mean level rising from 6.3 to 11.6 ng/mD in women with initial control PRL values within the true normal range « 15 ng/mD after 5 to 8 weeks' administration of OCs, all containing 50 f..Lg EE2 and almost all norethindrone, but not at a later time (6 months or more of use). No significant effect at 5 to 8 weeks was seen in such women given 30- to 35-f..Lg estrogen OCs. Not only did OCs not have a marked or sustained effect in normoprolactinemic women, but the 50-f..Lg estrogen dose significantly decreased PRL levels (mean change, - 12.6 ng/mD in 26 women whose PRL levels were initially elevated. This group showed a small but highly significant decline in PRL when compared with 19 women with initially elevated levels of PRL who had received no medication. This suggestion of an ameliorating effect of OCs in mild cases of hyperprolactinemia remains to be confirmed in a larger study. The disappearance of clear-cut statistically significant effects of 50-f..Lg estrogen OCs by 6 months, after a clear-cut initial effect at 5 to 8 weeks, suggests adaptation to the estrogen effect, a phenomenon often found in studies of the actions of this class of hormones. 7 , 8 Alternatively, the progestogen in the OC (75% received 1 mg norethindrone) may have eventually vitiated the initial estrogenic effect. The lack of a significant, continued effect makes less likely a long-term adverse effect of OCs in accelerating the development of occult PRL-secreting pituitary gland adenomas, despite the earlier concerns of Sherman and co-workers. 9 The clear-cut hyperprolactinemic effects of estrogen in rats found by Chen and Meites lO and by Yen and colleagues in human beings l l were noted at much higher doses of Table 3. Persistence of Elevated Serum PRL Levels on 1 to 8 Weeks Follow-Up of Women with Initially Normal or Elevated Values Who Received No Medication Group

Total no.

No. with elevated follow-up

Initial PRL levels normal a Initial PRL values elevateda

36 19

elevated

P(McNemar's i')

4

11

< 0.025

15

79

%

a< 23 ng/ml for 2 to 3 weeks postpregnancy group; < 18 ng/ml for 4 to 5 weeks postpregnancy group; and < 15 ng/ml for 6 + weeks postpregnancy or never-pregnant group;. as derived from Table 2.

Josimovich et al. Distribution of PRL values and OC effect

789

l"

I

Table 4. Effects ofOCs on Serum PRL Levels in Women 6 Weeks or More Postpregnancy or Never Pregnant Whose Initial Values were Normal « 15 ng/mlr Group

No. of pairs

Comparison before and after drug administration Geometric mean initial Geometric mean final value value ng/ml

5--8 wk of 50-lLg estrogen OCs 6 mo or more of 50-lLg estrogen OCs 5--8 wk of 35-lLg estrogen OCs

P

ng/ml

11.6

< 0.001

33

6.33

12

6.74

7.48

> 0.5

21

6.62

6.61

> 0.5

aThree quarters contained norethindrone.

estrogen than used in this study. A lack of a stimulatory effect of OCs at the doses given to mildly hyperprolactinemic patients in the current study over 5 to 8 weeks does not, of course, guarantee that certain women might not have stimulation of growth in incipient microadenomas. Statistical analyses demonstrate that a PRL level of 15 ng/ml or higher in the study population may well represent a degree of pathologic hyperprolactinemia. Luciano and co-workers 12 found that 16% of single samples drawn before 10:00 A.M. were> 25 ng/ml (their overall mean control level + 2 SD), while in this study, only 11% were beyond a theoretical mean + 2 SD of 20.9 ng/ml (last line, Table 1), perhaps because of sampling later in the morning than in the Luciano study. With the truncation point of 12.6 ng/ml, 22% of values fell above p., + 2<;. Because only 79% of initially elevated values were found elevated on subsequent sampling (Table 3), 0.79 x 22% = 17% of those studied would appear to be normally elevated. This latter estimate correlates well with a 10% or higher incidence of micro adenomas found at autopsies,13 as well as the value of administration of bromergocriptine, in certain cases offemale infertility described by Ben-David and Schenker. 2 Although many clinicians state that most women with a mild degree of hyperprolactinemia will never require surgery for neurologically or endocrinologically threatening adenomas, patients should be counseled that sustained mild PRL elevations of 15 ng/ml or more may have need of therapy for infertility in the future or should be advised to have a series of PRL levels drawn at annual intervals. Patients should be educated to be aware of symptoms suggesting the development of clinically significant pituitary adenomas that would require more detailed diagnostic studies. Finally, such patients desiring to use com790

monly used formulations of OCs may be reassured that use of this medication is unlikely to worsen their hyperprolactinemia. For safety, however, a repeat PRL level should be performed at 2 months or more of OC use to indicate those individuals who may be exceptional in reacting with increased hyperprolactinemia. Multiple samples should be drawn and the PRL level averaged if a single elevated value is obtained initially. From these studies the following conclusions may be drawn: (1) the poly modal distribution of control serum, PRL values in women, as defined by methods such as the truncation technique reported here, must be taken into consideration before studying the effects of medications; (2) 50-f1g estrogen dose OCs exert a temporary increase in basal PRL levels, an effect that is no longer clearly evident at 6 months or more of therapy; (3) administration of lower-estrogen-dose (30 to 35 f1g) OCs is not likely to increase serum PRL levels; and (4) there may be a slight amelioration of elevated PRL levels by the cyclic use of 50-f1g estrogen OCs.

REFERENCES 1. Reyniak JV, Wenof M, Aubert JM, Stangel JJ: Incidence of hyperprolactinemia during oral contraceptive therapy. Obstet Gynecol 55:8, 1980 2. Ben-David M, Schenker JG: Transient hyperprolactinemia: a correctible cause of idiopathic female infertility. J Clin Endocrinol Metab 57:442, 1983 3. Jeffcoate SL: Prolactin secretion in normal and abnormal menstruation. Program of the IX World Congress on Gynecology and Obstetrics, Tokyo, Simul International, October 25 to 31, 1979, p 32 4. Hollander M, Wolfe DA: Non-parameteric Statistical Methods. New York, John Wiley & Sons, 1973, pp 115-116, 125-126 5. Bliss CT: Statistics in Biology, Vol 1. New York, McGrawHill, 1967, p 162

Josimovich et al. Distribution of PRL values and OC effect

Fertility and Sterility

6. Ehara Y, Siler TM, VandenBerg G, Sinha VU, Yen SSC: Circulatory prolactin levels during the menstrual cycle: episodic release and circhoral variation. Obstet Gynecol 117:962, 1973 7. Hisaw FL, Hisaw FL Jr: Action of estrogen and progesterone on the reproductive tract of lower primates. In Sex and Internal Secretion, Vol 1, Third edition, Edited by WC Young. Baltimore, Williams & Wilkins, 1961, p 556 8. Steinetz BG: Secretion and function of ovarian estrogens. In Handbook of Physiology, Section 7, Vol 2, Part I, Edited by RO Greep, EB Astwood, SR Geiger. Baltimore, Williams & Wilkins, 1973, p 439 9. Sherman BM, Schlechte J, Halmi NS, Chapler FK, Harris CE, Duello TM, VanGilde J, Granner DK: Pathogenesis of prolactin-secreting pituitary adenomas. Lancet 2:lO19, 1978

Vol. 47, No.5, May 1987

10. Chen CL, Meites J: Effects of estrogen and progesterone on serum and pituitary prolactin levels in ovariectomized rats. Endocrinology 99:1482, 1976 11. Yen SSC, Ehara Y, Siler 1M: Augmentation of prolactin secretion by estrogens in hypogonadal women. J Clin Invest 53:652, 1974 12. Luciano AA, Sherman BA, Chapler FK, Hauser KS, Wallace RB: Hyperprolactinemia and contraception: a prospective study. Obstet Gynecol 65:506, 1985 13. McCormick WF, Halmi NS: Absence of chromophobe adenomas from a large series of pituitary tumors. Arch Pathol 92:231, 1971

Josimovich et al. Distribution of PRL values and OC effect

791