1 Department of Clinical Epidemiology and Biostatistics and 2 Centre for Reproductive Medicine, Academic Medical Centre, University of Amsterdam, Amsterdam, the Netherlands
![]() |
Abstract |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Key words: algorithms/ectopic pregnancy/human chorionic gonadotrophin/probabilistic decision rules
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Since the variety of non-invasive clinical tools offers numerous possibilities for developing diagnostic strategies, there was a need for clinical guidelines in the diagnosis of ectopic pregnancy. To support the clinician in daily practice, several authors have developed algorithms for the management of suspected ectopic pregnancy, incorporating transvaginal sonography and serum HCG measurement, sometimes preceded by serum progesterone measurement (Ankum et al., 1993; Carson and Buster, 1993
; Stovall and Ling, 1993
). These algorithms have shown excellent sensitivity and specificity in heterogeneous populations of women with suspected ectopic pregnancy, thereby reducing the number of unnecessary laparoscopies.
As a consequence of the general availability of non-invasive diagnostic tools, and since these tools are a minor burden for the patients, the clinical context of the patient with suspected ectopic pregnancy has changed dramatically. Nowadays, the classical clinical case, a woman with severe abdominal pain and a possible pregnancy, is rarely seen, since ectopic pregnancy is often already suspected in early pregnancy, sometimes even before the onset of clinical symptoms (Cacciatore et al., 1994; Mol et al., 1997
). In these patients, the prevalence of ectopic pregnancy is lower than in those with the classical clinical picture, thus increasing the hazard of false positive diagnoses (Sackett et al., 1985
).
These differences in the strength of suspicion of ectopic pregnancy should have consequences for diagnostic management. In women with a clinical profile corresponding to a low degree of suspicion, test results corresponding with higher likelihood ratios are required for a diagnosis ectopic pregnancy. In women with a clinical profile corresponding to a high degree of suspicion of ectopic pregnancy, the probability of an ectopic pregnancy need only be increased a little further in order to establish the diagnosis and to warrant therapeutic action.
This heterogeneity in the degree of suspicion of ectopic pregnancy has not been incorporated in the diagnostic algorithms available in the medical literature. Since these algorithms all use rigid cut-off levels for clinical decisions, they do not take into account differences in pre-test probability. For example, the likelihood ratio of a serum HCG level of 3000 IU/l is 15 where sonography shows neither an intrauterine pregnancy nor adnexal abnormalities (Mol et al., 1998). A pre-test probability for ectopic pregnancy of 5%, well suited for a symptom-free patient with a previous ectopic pregnancy, results in this patient having a post-test probability for ectopic pregnancy of 44% (Mol et al., 1997
). The same serum HCG value in a patient with abdominal pain, with a pre-test probability of 40%, results in a post-test probability of 92%. Such differences could have clinical implications, since confirmatory laparoscopy or treatment with systemic methotrexate is only justified when the probability of ectopic pregnancy is high. The use of rigid cut-off levels for the diagnosis of ectopic pregnancy could lead to unwarranted therapeutic action and harm to the patient.
In this study, we present a more flexible individualized diagnostic algorithm that uses probabilistic decision rules for the evaluation of suspected ectopic pregnancy. We have compared its diagnostic performance with that of a general algorithm based on rigid cut-off levels.
![]() |
Materials and methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
In the inflexible algorithm, an intrauterine pregnancy is diagnosed when an intrauterine gestational sac is visualized at transvaginal sonography. If an intrauterine gestational sac cannot be visualized, both adnexal regions are scanned; ectopic pregnancy is diagnosed if a yolk sac, a fetal pole, fetal cardiac activity or an ectopic mass and fluid in the pouch of Douglas is seen. If transvaginal sonography does not lead to a diagnosis, serum HCG measurement is performed. Taking into account the `discriminatory zone principle', as defined previously (Kadar et al., 1981), ectopic pregnancy is diagnosed for serum HCG concentration >2000 IU/l, whereas in patients in whom sonography showed an adnexal mass or fluid in the pouch of Douglas, a serum HCG concentration of 1500 IU/l or higher suffices (Kadar et al., 1981
; Mol et al., 1998
). When the serum HCG concentration is <1500 IU/l, transvaginal sonography and serum HCG are repeated at 2 day intervals, until a diagnosis is made. The cut-off level for diagnosis of ectopic pregnancy at repeat serum HCG measurement is 1000 IU/l. If the second repeat HCG measurement does not reach this threshold, ectopic pregnancy is diagnosed when serum HCG is increasing, whereas decreasing serum HCG concentrations are supposed to rule out ectopic pregnancy.
The flexible, individualized algorithm, which uses probabilistic decision rules, is shown in Figure 1. Based on observations in our own hospital, we let the pre-test probability of an ectopic pregnancy in a patient depend on three findings of the medical history: abdominal pain, vaginal bleeding and previous exposure to a risk indicator. In the absence of risk indicators, the pre-test probability for ectopic pregnancy is 34% in a patient with abdominal pain, 18% in a patient with vaginal bleeding, and 39% when a patient presents with both symptoms. If one or more risk indicators are present, these probabilities are 42, 34 and 54% respectively. The pre-test probability of ectopic pregnancy in the absence of abdominal pain and vaginal bleeding in a woman without a risk indicator is <1%. In symptom-free women with at least one risk indicator this pre-test probability is 6%. Risk indicators are a history of ectopic pregnancy, tubal surgery or pelvic inflammatory disease; tubal disease detected by hysterosalpingography or laparoscopy; in-utero exposure to diethyl stilboestrol (DES), sterilization, and an intrauterine contraceptive device (IUCD) in situ at the moment of conception (Mol et al., 1995
; Ankum et al., 1996
).
|
The first test is scanning of the intrauterine cavity. The presence of an intrauterine gestational sac is associated with a likelihood ratio of 0.07; absence of an intrauterine gestational sac has a likelihood ratio of 2.2 (Mol et al., 1998). If the calculated post-test probability drops below a threshold that we presume to be low enough to rule out ectopic pregnancy, say 1%, the diagnosis of ectopic pregnancy is rejected and further diagnostic tests can be delayed. Alternatively, if the post-test probability exceeds a threshold that we presume to be high enough to justify confirmatory laparoscopy, or the start of medical treatment, say 95%, the diagnosis ectopic pregnancy can be made, and further diagnostic tests can also be delayed. For example, a pregnant patient with abdominal pain and a previous ectopic pregnancy has a pre-test probability of 42%. When transvaginal sonography in this patient does not show an intrauterine pregnancy, one can derive from the Appendix that the probability of ectopic pregnancy increases to 60%. This post-test probability does not reach the threshold of 95% that was required for the diagnosis ectopic pregnancy. Consequently, we need a second test, i.e. scanning of the adnexal region.
Scanning of the adnexal region can show ectopic cardiac activity, an ectopic gestational sac, an ectopic mass and/or fluid in the pouch of Douglas. Ectopic cardiac activity proves the presence of an ectopic pregnancy (likelihood ratio of infinity), whereas the other features are associated with likelihood ratios of 23, 3.6 and 4.4 respectively (Mol et al., 1998). Presence of an ectopic mass with fluid in the pouch of Douglas has a likelihood ratio of 9.9. A completely normal adnexal region has a likelihood ratio of 0.55.
Starting from the probabilities for ectopic pregnancy that were obtained after scanning the intrauterine cavity, and using the likelihood ratios for the findings at scanning of the adnexal region, we can derive new post-test probabilities. Again, ectopic pregnancy can be rejected without further testing if the lower threshold is reached (probability <1%), or, alternatively, ectopic pregnancy can be diagnosed without further non-invasive testing (probability >95%). In all other cases, the next diagnostic test must be performed, i.e. serum HCG measurement.
After serum HCG measurement, the probabilities for ectopic pregnancy are recalculated, taking into account findings in the adnexal region at sonography (Mol et al., 1998). If the probability for ectopic pregnancy remains between 1 and 95%, transvaginal sonography and, if necessary, serum HCG measurement can be repeated at 2 day intervals. Since the findings at repeated sonography are related to previous findings at sonography, risk adjustment should only be done if the sonographic picture changes as compared with earlier sonography. This is not shown in the flow chart. In general, ectopic pregnancy can be ruled out in the presence of an intrauterine pregnancy, whereas the diagnosis can be made where ectopic cardiac activity or an ectopic gestational sac are visualized.
Although for clinical practice we would recommend repeat serum HCG measurements when three repeat measurements do not reveal a diagnosis, in the absence of an intrauterine pregnancy, the algorithms in the present study reject a diagnosis of ectopic pregnancy if the serum HCG level at the second repeat measurement declines and accept the diagnosis if it increases.
Statistical analysis
The diagnostic characteristics of the inflexible algorithm, using rigid cut-off levels and the performance of a flexible algorithm, based on a probabilistic decision rule, were compared using data observed in a cohort of 800 women, of whom 200 had an ectopic pregnancy (Mol et al., 1998). Based on the test results and findings in these patients, fictitious cohorts of 10 000 women with suspected ectopic pregnancy were put together, varying in the prevalence of ectopic pregnancy. Test results of patients with and without an ectopic pregnancy in the fictitious cohorts were drawn from the distributions of results in the respective subgroups in the original cohort of 800 patients.
The random number generator in the SAS® statistical software package was used to draw random samples from the test results of patients in the original cohort. The distributions of the initial serum HCG concentrations were assumed to be log normal. For the repeat serum HCG measurements, we used the empirical distribution of the serum HCG course after 2 and 4 days in each individual patient, expressed in terms of the percentage change relative to the previous value. This serum HCG course was then combined with the initial serum HCG concentration to obtain a repeat serum HCG concentration at 2 and 4 days.
Subsequently, the two algorithms under study were applied to the cohorts with 10 000 fictitious patients. For each of the algorithms, we calculated sensitivity and specificity as well as predictive values for prevalences varying between 5 and 50%. Sensitivity was defined as the fraction of women with an ectopic pregnancy in whom the algorithm makes this diagnosis correctly, whereas specificity was defined as the fraction of women without an ectopic pregnancy in whom the algorithm rejected ectopic pregnancy correctly. The positive predictive value was defined as the fraction of women in whom the algorithm diagnosed ectopic pregnancy who turn out to have an ectopic pregnancy, whereas the negative predictive value was defined as the fraction of women in whom the algorithm rejected ectopic pregnancy, but who turned out to have an ectopic pregnancy. Furthermore, the fraction of patients in whom the diagnosis was established with 2 days delay and with 4 days delay was calculated.
In the initial analysis, the flexible, individualized algorithm used a threshold probability of 1% to exclude the presence of an ectopic pregnancy, whereas a threshold probability of 95% was used to establish the diagnosis of ectopic pregnancy. To evaluate the impact of the use of different thresholds in the flexible individualized algorithm, we constructed two receiver-operating-characteristic (ROC) curves. In a ROC plot, for each threshold probability, the sensitivity of the flexible, individualized algorithm is plotted against `1specificity'. In doing so, it is possible to assess the impact of the use of different threshold probabilities on the performance of the flexible, individualized algorithm, thereby facilitating the choice of the optimal threshold probabilities for ruling in or ruling out ectopic pregnancy. The prevalence of ectopic pregnancy in these ROC analyses was set at 10%. In the first ROC analysis, the threshold probability required to diagnose ectopic pregnancy was varied between 55 and 97%, whereas in the second ROC analysis the threshold probability required to rule out ectopic pregnancy varied between 1 and 15%.
To compare the clinical value of the algorithms, we weighted the consequences of the diagnostic outcomes of the algorithms on a disutility scale (Pauker and Kassirer, 1975). On that scale, a true positive diagnosis and a true negative diagnosis are considered to generate no disutility. A false positive diagnosis is valued as 1, and a false negative diagnosis is then valued 10 times, 4 times, equally, 0.25 times and 0.10 times worse than a false-positive diagnosis. For each of these five possible valuations, the disutility reduction that was expected after using the flexible algorithm as compared with the inflexible algorithm was calculated. A positive value would indicate that the flexible algorithm would be superior, i.e. generates less disutility, whereas a negative value would indicate that the inflexible algorithm would be superior.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
At a prevalence of 5%, the flexible algorithm had a positive predictive value of 78%, 20% higher than the inflexible algorithm, with a negative predictive value that was only slightly higher (0.5%). If the prevalence of ectopic pregnancy increased, the positive and negative predictive values also increased, to become 96 and 5.4% for a prevalence of 50%. In the inflexible algorithm, the diagnosis of ectopic pregnancy was delayed for 2 days in 8% of the patients and for 4 days in 27% of the patients at a prevalence of 5%. In the flexible algorithm, these rates were 15 and 39% at a prevalence of 5%, and 15 and 28% at a prevalence of 50%.
Figure 2 shows two ROC curves, the first evaluating the impact of different thresholds for the probability required to diagnose ectopic pregnancy, the second evaluating the impact of different thresholds for the probability required to rule out ectopic pregnancy. Both ROC curves were constructed assuming a pre-test prevalence of 10%. At a threshold probability to diagnose ectopic pregnancy of 95% the algorithm had a sensitivity of 92% and a specificity of 98%. Above this threshold, the sensitivity dropped slightly without an increase in specificity. If the threshold decreased below 95%, the specificity decreased strongly without an increase in sensitivity. At a threshold probability of 1% to rule out ectopic pregnancy sensitivity and specificity were 92 and 98% respectively. An increase of this threshold to 5% resulted in a strong decrease in sensitivity for a very small increase in specificity.
|
|
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
We used computer-simulated fictitious cohort patients to compare the diagnostic performance of the algorithms. In fact, this fictitious cohort has the same characteristics as our own cohort of 800 patients. Since the diagnostic characteristics that were used in the flexible algorithm are derived from this same population, the performance of that algorithm might be overestimated. However, the cut-off values of serum HCG values used in the inflexible algorithm were also based on ROC analyses in that particular cohort, thereby facilitating a comparison of both algorithms (Mol et al., 1998).
Surprisingly, sensitivity and specificity of the inflexible algorithm were slightly less than those of the algorithm reported in a previous study, in which they were reported to be 96 and 97% respectively (Ankum et al., 1993). This difference might be due to minor artefacts in our simulation. Another explanation could be that in the original study sonography was performed with knowledge of the results of the serum HCG measurements, thereby improving the performance of the algorithm as compared with a strict evaluation by protocol, as was done in the present study.
In the literature the performance of diagnostic tests is commonly reported in terms of sensitivity, specificity and likelihood ratios. When such parameters are used, an important assumption is that these indices remain constant for patients with different clinical characteristics. Unfortunately, empirical evidence for this crucial assumption is rare. We previously reported that the diagnostic performance of initial serum HCG measurement depended significantly on sonographic abnormalities in the adnexal region, whereas abdominal pain and vaginal bleeding had no impact on the diagnostic performance of serum HCG measurement (Mol et al., 1998). In our cohort, the diagnostic performance of scanning the intrauterine cavity and the adnexal region was not significantly dependent on abdominal pain and vaginal bleeding (unpublished data; vaginal bleeding and intrauterine cavity P = 0.60, abdominal pain and intrauterine cavity P = 0.39, vaginal bleeding and adnexal pathology P = 0.06, abdominal pain and adnexal pathology P = 0.91). However, it is likely that the distribution of diagnostic test results in patients with and without ectopic pregnancy is not uniform. This issue will remain unresolved until more data are available on the performance of sonography and serum HCG in patients with different characteristics.
Apart from differences in diagnostic performance of serum HCG between patients with and without abnormalities in the adnexal region, it is clear that the sensitivity and specificity vary with the disease prevalence in the flexible algorithm (Table I). This phenomenon can be explained if one takes into account the fact that the cut-off levels for test positivity differ depending on the pre-test probability of ectopic pregnancy. For example, a patient without abnormal findings at transvaginal sonography and a pre-test probability for ectopic pregnancy of 60% requires a serum HCG value >2000 IU/l to increase the post-test probability for ectopic pregnancy beyond 95%. The post-test probability in a patient with identical findings at sonography and a pre-test probability of only 10% would be 63%, and the serum HCG value has to be >6000 IU/l in order to increase the pre-test probability beyond 95%. The sensitivity of the flexible algorithm is lower for lower pre-test probabilities, whereas specificity increases.
The same mechanism explains why predictive values are virtually constant in the flexible algorithm, whereas they vary considerably in the inflexible algorithm. Although, in general, predictive values depend on the pre-test probability and should therefore not be used as characteristics of a test's worth, they are of utmost relevance to the clinician interpreting the test result, since they express the probability of presence or absence of disease conditional on the test result. By using different cut-off values, the fluctuation of these predictive values is eliminated in the flexible approach.
The likelihood ratio of 0.07 for an intrauterine pregnancy on scanning the uterine cavity indicates that for high pre-test probabilities the post-test probability is still too high to reject the diagnosis of ectopic pregnancy. It should be noted that this likelihood ratio does not take into account the quality of the intrauterine pregnancy. In the present algorithm, for example, similar value is given to a small gestational sac and a viable intrauterine pregnancy with a yolk sac and cardiac activity at sonography. Of course, in clinical practice one could take the quality of the intrauterine pregnancy into account in decision making. Furthermore, subsequent serum HCG measurement cannot lead to a diagnosis of ectopic pregnancy in these patients, since serum HCG measurement has only diagnostic significance in patients without an intrauterine sac at sonography (Kadar et al., 1981). If a patient with a high pre-test suspicion of ectopic pregnancy has a small gestational sac at sonography, and continues to have severe abdominal pain, close clinical and sonographical monitoring is recommended. Heterotopic pregnancy should not be excluded in these patients.
In 1981, Kadar et al. proposed the use of a discriminatory zone for serum HCG, defined as the minimal HCG concentration above which the sac of an intrauterine pregnancy can always be identified by sonography (Kadar et al., 1981). Since then, the resolution of sonography has strongly improved, thereby decreasing the discriminatory zone. Studies reporting on the use of serum HCG measurement combined with sonography in the diagnosis have recommended cut-off levels varying between 1000 and 6500 IU/l (Kadar et al., 1981
; Romero et al., 1985
; Nyberg et al., 1988
; Cacciatore et al., 1990
; Ankum et al., 1993
). The present study shows that there exists no optimal cut-off level for serum HCG values in the diagnosis of ectopic pregnancy. Rigid cut-off levels should be abandoned, and the available diagnostic information should be used to calculate the probability of an ectopic pregnancy.
ROC analysis indicated that the optimal cut-off rates for establishing and ruling out ectopic pregnancy were 95 and 1% respectively. It should be emphasized that the threshold for establishing a diagnosis can vary. For example, if a woman with suspected ectopic pregnancy wants to be sure about the diagnosis within a short time, confirmatory laparoscopy can be performed, even when the probability for ectopic pregnancy is below 95%. On the other side of the spectrum, one should urge caution with increasing the threshold for ruling out ectopic pregnancy. In our opinion, the consequence of missing an ectopic pregnancy is so serious that it justifies increased follow-up of patients until ectopic pregnancy is ruled out with certainty. Since in the evaluation of the flexible algorithm in this study a diagnosis was always made after the second repeat serum HCG measurement, we speculate that the performance of the algorithm can improve even if the serum HCG course is followed over a longer period.
The improved diagnostic performance in the flexible algorithm can be explained, since in this algorithm the final diagnosis is delayed in patients with probabilities between 1 and 95%, thereby allowing a time component in the diagnostic process. The improvement of diagnostic performance should therefore be weighed against the disadvantages; the possibility of tubal rupture and the extra costs generated by further testing. With respect to tubal rupture, a prevalence of 8% has been observed in patients in whom the diagnosis was delayed (Mol et al., 1999). With respect to costs, it can be assumed that the decrease of false positive and false negative diagnoses reduces costs in such a way, that an increase in the use of transvaginal sonography and serum HCG measurement is justified.
The merits of probabilistic diagnosis have been known for decades, and they are applied in various fields of obstetrics and gynaecology (Habbema et al., 1978). In prenatal screening for Down's syndrome, for instance, the results of the triple test are always interpreted in relation to maternal age, which is closely linked to the baseline risk of giving birth to a Down's syndrome child (Fletcher et al., 1995
). The present analysis has shown that such an approach is useful in the diagnostic work-up of women with suspected ectopic pregnancy.
Our algorithm was based on data from the medical history, transvaginal sonography and serum HCG measurement, but did not incorporate serum progesterone measurement, due to the fact that serum progesterone was not measured in our cohort. Of course, serum progesterone measurement could be added to our algorithm, especially in symptom-free women with a low suspicion of ectopic pregnancy (McCord et al., 1996). However, one should be cautious with the combined use of serum HCG measurement and serum progesterone measurement, since these two markers are likely to be mutually dependent.
In conclusion, we state that the use of probabilistic decision rules in the algorithms for the work-up of suspected ectopic pregnancy increases the diagnostic performance of such algorithms as compared to inflexible algorithms using rigid cut-off values. Clinicians should therefore incorporate probabilistic decision rules in algorithms used for the diagnosis of ectopic pregnancy.
![]() |
Appendix |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
|
|
![]() |
Notes |
---|
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Ankum, W.M., Mol, B.W.J., van der Veen, F. and Bossuyt, P.M.M. (1996) Risk-factors for ectopic pregnancy: a meta-analysis. Fertil. Steril., 65, 10931099.[ISI][Medline]
Cacciatore, B., Stenman, U.H. and Ylostalo, P. (1990) Diagnosis of ectopic pregnancy by vaginal ultrasonography with a discriminative serum HCG level of 1000 IU/l. Br. J. Obstet. Gynaecol., 97, 904908.[ISI][Medline]
Cacciatore, B., Stenman, U.H. and Ylöstala. P. (1994) Early screening for ectopic pregnancy in high-risk symptom-free women. Lancet, 343, 517518.[ISI][Medline]
Carson, S.A. and Buster, J.E. (1993) Ectopic pregnancy. N. Engl. J. Med., 329, 11741181.
Fletcher, J., Hicks, N.R., Kay, J.D.S. and Boyd, P.A. (1995) Using decision analysis to compare polices for antenatal screening for Down's syndrome. Br. Med. J., 311, 351356.
Habbema, J.D.F., Hilden, J. and Bjerregaard, B. (1978) The measurement of performance of probabilistic diagnosis I. The problem, descriptive tools, and measures based on classification matrices. Meth. Inf. Med., 17, 217226.[ISI][Medline]
Kadar, N., DeVore, G. and Romero, R. (1981) Discriminatory HCG zone: its use in the sonographic evaluation for ectopic pregnancy. Obstet. Gynecol., 58, 156161.[Abstract]
McCord, M.L., Muram, D., Buster, J.E. et al. (1996) Single serum progesterone as a screen for ectopic pregnancy: exchanging specificity and sensitivity to obtain optimal test performance. Fertil. Steril., 66, 513516.[ISI][Medline]
Mishalani, S.H., Seliktar, J. and Braunstein, G.D. (1994) Four rapid serum-urine combination assays of choriogonadotrophin (HCG) compared and assessed for their utility in qualitative determinants of HCG. Clin. Chem., 40, 19441949.
Mol, B.W.J., Ankum, W.M., Van der Veen, F. and Bossuyt, P.M.M. (1995) Contraception, and the risk for ectopic pregnancy: a meta-analysis. Contraception, 52, 337341.[ISI][Medline]
Mol, B.W.J., Hajenius, P.J., Engelsbel, S. et al. (1996) Diagnostiek en behandeling van de extra-uteriene graviditeit in Nederland. Ned. Tijdsch. Geneesk., 140, 13581361.
Mol, B.W.J., Hajenius, P.J., Ankum, W.M. et al. (1997) Screening for ectopic pregnancy in symptom-free women at increased risk. Obstet. Gynecol., 89, 704707.
Mol, B.W.J., Hajenius, P.J., Engelsbel, S. et al. (1998) Serum human chorionic gonadotrophin measurement in the diagnosis of ectopic pregnancy when transvaginal sonography is inconclusive. Fertil. Steril., 70, 972981.[ISI][Medline]
Mol, B.W.J., Hajenius, P.J., Engelsbel, S. et al. (1999) Can non-invasive diagnostic tools predict tubal rupture or active intra-abdominal bleeding in patients with ectopic pregnancy? Fertil. Steril., 71, 167173.[ISI][Medline]
Nyberg, D.A., Mack, L.A., Laing, F.C. and Jeffreu, R.B. (1988) Early pregnancy complications: endovaginal sonograpic findings correlated with human chorionic gonadotrophin levels. Radiology, 167, 619622.[Abstract]
Pauker, S.G. and Kassirer, J.P. (1975) Therapeutic decision making: a cost-benefit analysis. N. Engl. J. Med., 293, 229234.[Abstract]
Romero, R., Kadar, N., Jeanty, P. et al. (1985) Diagnosis of ectopic pregnancy: value of the discriminatory human chorionic gonadotrophin zone. Obstet. Gynecol., 66, 357360.[Abstract]
Sackett, D.L., Haynes, R.B., Guyatt, G.H. and Tugwell, P. (1985) Clinical Epidemiology: a Basic Science for Clinical Medicine. Little, Brown and Company, Boston/Toronto/London.
Stovall, T.G., Ling, F.W. (1993) Ectopic pregnancy, diagnostic and therapeutic algorithms minimizing surgical intervention. J. Reprod. Med., 38, 807812.[ISI][Medline]
Submitted on November 17, 1998; accepted on August 25, 1999.