The accuracy of hysterosalpingography in the diagnosis of tubal pathology: a meta-analysis

FERTILITY AND STERILITY Vol. 64, No.3, September 1995 Copyright ~ 1995 American Society for Reproductive Medicine Printed on acid-free paper in U. S. A The accuracy of hysterosalpingography in the diagnosis of tubal pathology: a meta-analysis Patricia Swart, M.D. * Ben W. J. Mol, M.D.t Fulco van der Veen, M.D.*:j: Marc van Beurden, M.D. * William K. Redekop, Ph.D.t Patrick M. M. Bossuyt, Ph.D.t Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands Objective: To assess the value of hysterosalpingography (HSG) in diagnosing tubal patency and peritubal adhesions using laparoscopy with chromopertubation as the gold standard. Design: Meta-analysis of 20 studies comparing HSG and laparoscopy for tubal patency and peritubal adhesions. Patients: Four thousand one hundred seventy-nine patients with infertility in 20 studies. Intervention: Hysterosalpingography and diagnostic laparoscopy as part of infertility workup. Main Outcome Measure: Tubal patency and peritubal adhesions. Results: For tubal patency the reported sensitivity and specificity differed between studies. In a subset of studies that evaluated HSG and laparoscopy independently, a point estimate of 0.65 for sensitivity and 0.83 for specificity was calculated. For peritubal adhesions a summary receiver operating characteristic curve could be estimated. Conclusions: Although HSG is of limited use for detecting tubal patency because of its low sensitivity, its high specificity makes it a useful test for ruling in tubal obstruction. For the evaluation of peritubal adhesions HSG is not reliable. Fertil Steril 1995; 64:486-91 Key Words: Hysterosalpingography, diagnostic laparoscopy, tubal patency, peritubal adhesions, meta-analysis, infertility Hysterosalpingography (HSG) and diagnostic laparoscopy with chromopertubation are two widely used methods to determine tubal function in infertile women. The diagnostic value of the HSG, however, is still a matter for debate. Some authors consider the HSG to be an indispensable test that should be performed before laparoscopy (1-3), whereas others favor omission of the HSG altogether (4, 5). A frequently noted point in favor of the HSG is its presumed ability to induce pregnancies (2, 6, 7). The aim of this study was to determine the value of the HSG in the diagnosis of tubal pathology. A systematic review of the literature was carried out. A meta-analysis was set up for studies comparing Received October 6, 1994; revised and accepted March 28, 1995. * Department of Obstetrics and Gynaecology. t Department of Clinical Epidemiology and Biostatistics. :j: Reprint requests: Fulco van der Veen, M.D., Academic Medical Center, Department of Obstetrics and Gynaecology, University of Amsterdam, P.O. Box 22700, 1100 DE Amsterdam, The Netherlands (FAX: 31-20-6963489). 486 Swart et al. HSG for tubal pathology: a meta-analysis HSG with diagnostic laparoscopy as the gold standard. Search Strategy MATERIALS AND METHODS A computerized Medline search to identify all registered articles in the English, French, German, and Dutch languages published between January 1968 and July 1994 was performed. Key words used were "hysterosalpingography" and "laparoscopy." In addition, all volumes of Fertility and Sterility, Human Reproduction, The British Journal of Obstetrics and Gynaecology, The American Journal of Obstetrics and Gynaecology, The Lancet and The British Medical Journal published between January 1974 a:nd June 1994 were reviewed manually. Cross-references in all selected articles were checked. Articles comparing HSG and laparoscopy for tubal pathology were included. Tubal pathology was defined in two ways: first, as the absence of tubal patency, i.e., the absence of filling or the absence of Fertility and Sterility

yes Summary Point Estimate Estimate Summary ROC curve Figure 1 The algorithm used for performing the meta-analysis on the diagnostic value of HSG (modified according to Midgette et al. [8]). overflow, and, second, as the presence of peritubal adhesions. If studies reported on proximal and tubal patency, these data were analyzed separately. Articles were excluded when data were not sufficient to construct a 2 X 2 table of the test (HSG) and the gold standard (laparoscopy) for tubal patency and/or peritubal adhesions. Articles that did not distinguish between tubal patency and peri tubal adhesions but only reported on "tubal pathology" also were excluded. In case of peritubal adhesions, only those patients in whom patency of at least one tube was demonstrated were included in the 2 X 2 table. Analysis The meta-analysis was conducted according to the methodology described by Midgette et al. (8) (Fig. 1). For each study, sensitivity, specificity, and likelihood ratios were calculated from the published data. Thereafter, tests for homogeneity by means of the X 2 test statistic (9) were performed for sensitivity and specificity separately. If homogeneity was not rejected for both sensitivity and specificity, a more precise summary point estimate of sensitivity and specificity, with its confidence intervals, was calculated. Sample sizes were used as the weight of each study (10). In case of heterogeneity in the reported sensitivity and specificity, the differences could be caused by a shift in cutoff levels for test positivity. In that case, higher sensitivity should be accompanied by lower specificity values and vice versa. This hypothesis can be explored by calculating a Spearman correlation coefficient (9). If there appears to be negative corre- Vol. 64, No.3, September 1995 lation (a correlation coefficient of -0.6 or lower was considered to be negative), the pairs of sensitivity and specificity can be thought of as originating from a single receiver operating characteristic (ROC) curve. A summary ROC curve then can be estimated (9). In the absence of a negative correlation between sensitivity and specificity, alternative explanations for the heterogeneity were examined. Subgroups of studies were distinguished by means of predefined gynecological and methodological criteria. Subgroup analysis examines the possibility that the variations in study results found can be attributed to differences in study population, study methodology, and/ or used techniques. Subgroups were selected on the following criteria: [1] prevalence of disease, i.e., absence of tubal patency or presence of peritubal adhesions (prevalence <35% or prevalence ~35%); [2] contrast medium used for HSG (oil- or water-soluble); [3J use of spasmolyticum while performing HSG; [4] setting (academic or nonacademic); [5] judgement of the laparoscopy with or without knowledge of the HSG (dependent or independent); [6] use of criteria for judgement of HSG; [7J time between HSG and laparoscopy; [8J number of patients included in the study; [9] HSG judged by a gynecologist or by a radiologist; and [10] type of cannula used for performing a HSG. Within each subgroup, the procedure reported earlier was repeated. Search Strategy RESULTS In total, 45 articles were found, 38 by Medline Search, 6 by cross-references, and 1 by hand search. Thirty-five articles fulfilled the inclusion criteria. Of these 35 articles, 15 articles were excluded according to the predefined criteria. Therefore 20 articles were analyzed (1-4, 6, 7, 12-25). Study characteristics and results of these articles are listed in Tables 1 and 2. Nineteen articles analyzed tubal patency (1-4,6, 7,12-24). Twelve ofthese articles distinguished between patients with two open tubes and patients with one or two closed tubes, whereas the other seven articles distinguished patients with two open tubes, patients with one open and one closed tube, and patients with two closed tubes. Defining disease as the presence of at least one closed tube, sensitivity, specificity, and likelihood ratios were calculated, comparing patients with one or two closed tubes with patients with two open tubes. Thirteen articles analyzed peritubal adhesions (6, 7, 12-21,25). Swart et a1. HSG for tubal pathology: a meta analysis 487

Table 1 Hysterosalpingography Versus Laparoscopy for Tubal Patency Likelihood ratio Author Year No. of patients* Prevalence Sensitivity Specificity Postitive test Negative test Articles that compare two groups of patients Swolin and Rosenkrantz (12) 1972 143 18 0.54 0.83 3.2 0.6 Maathuis et al. (13) 1972 207 38 0.72 0.83 4.2 0.3 Duignan et al. (14) 1972 273 25 0.85 0.89 8.0 0.2 Hutchins (1) 1976 409 20 0.76 0.90 7.3 0.3 Taylor (4) 1977 34 12 1.00 0.70 3.3 0.0 Phllipsen and Hansen (16) 1981 152 36 0.84 0.86 5.8 0.2 Donnez et al. (24) 1982 974* 34 1.00 0.99 71.8 0.0 Duff et al. (7) 1983 52* 40 0.67 0.84 4.1 0.4 Snowden et al. (22) 1984 77 48 0.73 0.88 5.8 0.3 Rice et al. (2) 1986 143 49 0.83 0.85 5.5 0.2 La Sala et al. (3) 1987 100 31 0.65 0.62 1.7 0.6 Loy et al. (20) 1989 77 26 0.80 0.72 2.9 0.3 Articles that compare three groups of patients Keirse and Vandervellen (15) 1972 50 54 0.79 0.83 4.5 0.3 EI Minawi et al. (6) 1978 352 26 0.89 0.90 8.9 0.1 Montanari et al. (17) 1982 85 35 0.83 0.89 7.6 0.2 Nordenskjold and Ahlgren (21) 1983 326 17 0.78 0.75 3.1 0.3 La Sala et al. (19) 1986 174 29 0.34 0.89 3.0 0.7 World Health Organization (23) 1986 121 60 0.82 0.55 1.8 0.3 Ismajovich et al. (18) 1986 215 78 0.74 0.68 2.3 0.4 * Number of tubes. % Analysis ency, which shows that 95% confidence intervals (CIs) do not overlap (Fig. 2). Only 2 of 20 articles The X 2 test statistic showed our data to be hetero- distinguished distal and proximal tubal occlusion, so geneous for tubal patency (P < 0.01; Table 3). For this distinction was not analyzed separately (1, 18). this reason the calculation of point estimates for sen- Because there was no negative correlation (Spearsitivity and specificity was not allowed methodologi- man correlation coefficient +0.22; Table 3), it was cally. Heterogeneity was confirmed by the plot of not possible to estimate a summary ROC curve. Visensitivity against specificity of HSG for tubal pat- sual inspection of the sensitivity and the specificity Table 2 Hysterosalpingography Versus Diagnostic Laparoscopy for Peritubal Adhesions No. of Likelihood ratio excluded Author Year patients No. of patients Prevalence Sensitivity Specificity Postitive test Negative test Swolin and Rosencrantz (12) 1972 29 114 41 0.79 0.61 2.0 0.3 Maathuis et al. (13) 1972 61 146 43 0.51 0.73 1.9 0.7 Duignan et al. (14) 1972 0 273 7 0.03 0.98 0.0 1.0 Keirse and Vandervellen (15) 1972 25 25 43 0.83 0.85 5.4 0.2 Hutchins (1) 1976 93 316 12 0.39 0.99 36.6 0.6 EI Minawi et al. (6) 1978 108 244 46 0.54 0.89 4.8 0.5 Phllipsen and Hansen (16) 1918 60 92 30 0.46 0.61 1.2 0.9 Montanari et al. (17) 1982 31 54 22 0.17 0.95 3.5 0.9 Duff et al. (7) 1983 19 33* 33 0.64 0.77 2.8 0.5 La Sala et al. (19) 1986 31 143 30 0.30 0.93 4.3 0.8 Loyet al. (20) 1989 77 77 53 0.29 0.92 3.5 0.8 Ismajovich et al. (18) 1986 53 162 58 0.77 0.50 1.5 0.6 Reshef et al. (25) 1989 0 215 35 0.13 0.87 1.0 1.0 * Number of tubes. 488 Swart et al. HSG for tubal pathology: a meta-analysis Fertility and Sterility %

~-...o. Table 3 Results of Tests for Homogeneity (P values of the X 2 test statistic) and Correlation (Spearman correlation coefficient). Sensitivity Specificity Spearman Tubal patency (1-4, 6-8, 12-24) Subgroups tubal patency Prevalence <35% (1-4, 12, 14, 19-21) Prevalence 2:35% (2, 6, 7, 13, 15-18,22,23) Contrast oil (1, 3, 6, 13, 19, 21) Contrast water (2,4,7, 12, 15-18,20,22-24) Spasmolyticum (15, 19, 22, 24) No spasmolyticum (1-4,6,7, 12, 13, 16-18,20,21,23) Academic setting (2, 4, 12, 13, 17, 21, 22, 24) Nonacademic setting (1, 3, 7, 14-16, 19,20) Independent judgment (7, 12, 15) Dependentjudgment(1-4,6,8, 13, 14, 16-24) Criteria (1, 7, 12, 13, 21, 23) No criteria (2-4, 6, 14-20, 22, 24) Peritubal adhesions (1, 6, 7, 12-20, 25) <0.01 <0.01 0.22 <0.05 <0.01 0.05 <0.01 <0.01 0.30 <0.01 <0.01-0.20 <0.01 <0.01 0.45 <0.01 <0.01 0.80 <0.01 <0.01 0.14 <0.01 <0.01 0.39 <0.01 <0.01 0.31 0.18 0.99 <0.01 <0.01 0.35 0.16 <0.01-0.53 <0.01 <0.01 0.25 <0.01 <0.01-0.69 0.75 0.25 - o i i - t-- -... _.--- - ---- -~ --- -~-----...- --.--< ~ also made it unlikely that the pairs of sensitivity and specificity originate from the same ROC curve but differ because of a shift in the criteria for test positivity (Fig. 2). The next step, analysis of the subgroups, only showed homogeneity for the three studies that judged HSG and laparoscopy independently (7, 12, 15) (P = 0.18 and P = 0.99; Table 3). The calculated point estimate for these studies is 0.65 for sensitivity (95% CI 0.50 to 0.78) and 0.83 for specificity (95% CI 0.77 to 0.88) (Fig. 2). For HSG judgment by a gynecologist or a radiologist and for the type of can- 11-1 c~ -~ "'-- -- '\J -- t- Point estimate and 95 % CI of studies with independent judgement of HSG and laparoscopy 0.75 I 0.5 specificity 0.25 o Figure 2 Sensitivity, specificity, and 95% CI in an ROC sheet for studies on tubal patency. Vol. 64, No.3, September 1995 nula used for performing a HSG, it was not possible to distinguish subgroups. In only one study could we identify that the judgment was performed by a radiologist (7). Ten of 20 articles did not mention which type of cannula was used, whereas the other 10 used seven different cannulas. All other subgroups were heterogenous and not negatively correlated. The X 2 test statistic showed our data to be heterogeneous for peritubal adhesions, too. However, there was a negative correlation (Spearman correlation coefficient ~0.69; Table 3). A summary ROC curve could be estimated for the ranges of reported sensitivity (0.0 to 0.83) and specificity (0.50 to 0.99). Figure 3 shows sensitivity, specificity, and 95% CIs for peritubal adhesions with the estimated ROC curve. DISCUSSION In this meta-analysis, we evaluated the value of hysterosalpingography in the diagnosis of tubal pathology. For tubal patency, an overall point estimate of sensitivity and specificity could not be calculated and a summary ROC curve could not be estimated. However, for those studies where HSG and laparoscopy were judged independently, a point estimate of 0.65 for sensitivity (95% CI 0.50 to 0.78) and of 0.83 for specificity (95% CI 0.77 to 0.88) could be calculated for tubal patency. This point estimate is the best assessment of the truth, first, because it represents a pooled estimate of several studies and is therefore more accurate compared with the result of a single study, and, second, because independent judgement prevents observer bias. For peritubal adhesions, data allowed the construction of a summary ROC curve (Fig. 3). This Swart et al. HSG for tubal pathology: a meta-analysis 489

0.75 ~. s; :;:; 'iii 0.5- e CD II) 0.25-- ---.~ t 0.75 0.5 specificity 0.25 o Figure 3 Sensitivity, specificity, and 95% CI for studies on peritubal adhesions with the estimated summary ROC curve. summary ROC curve represents the best assessment of the truth for peritubal adhesions. However, two methodological issues are of importance in the interpretation ofthe data. The first issue to be considered is that all studies were performed retrospectively and included only patients who underwent both HSG and laparoscopy. As a consequence, women who conceived after HSG and women who refused further investigations after HSG were not included. Because it is particularly those women who became pregnant after HSG that were likely to have had normal HSGs, abnormal HSGs will be over-represented in the populations studied. Common policy in the workup of infertility is to perform laparoscopy shortly after HSG when findings are abnormal and to wait 3 to 6 months when HSG is normal. The rationale for postponing laparoscopy is the supposed pregnancy-inducing effect of the HSG (2, 6,7). Recently, a meta-analysis was published of 10 studies in which the therapeutic role of oil- versus water-soluble contrast media are compared (26). Its conclusion was that oil-soluble contrast media have a pregnancy-inducing effect compared with watersoluble contrast media. One could overcome selection of patients by performing HSG and laparoscopy in one session. Only Swolin and Rosencrantz (12) performed laparoscopy on the day after HSG. This study has a sensitivity of 0.54 and a specificity of 0.83. The second issue is the fact that laparoscopy is not a perfect gold standard. Sometimes tubal obstruction at laparoscopy is due to artifacts, because 490 Swart et al. HSG for tubal pathology: a meta-analysis of technical failure and differences in resistance between the two tubes. However, laparoscopy is the best standard available. The clinical significance of a point estimate of 0.65 for sensitivity and of 0.83 for specificity is that a negative test result, i.e., tubal patency, will not rule out disease and that a positive test result, i.e., tubal obstruction, will rule in disease (27). The estimated ROC curve for the diagnosis of peritubal adhesions shows that HSG is an unreliable test. An alternative test for HSG in the infertility workup is Chlamydia serology. Recently, 0.74 for sensitivity (95% CI 0.56 to 0.87) and 0.92 for specificity (95% CI 0.78 to 0.98) were reported (28). Although only a single study was performed on the subject, Chlamydia serology may prove to be superior to the HSG. Obviously, a disadvantage of Chlamydia serology is that it provides no information on tubal integrity, whereas other possible causative agents are not identified. In conclusion, our data show that tubal obstruction on a HSG is a reliable test result and would, in this IVF era, not necessarily need confirmation by laparoscopy. Tubal patency on HSG, however, is not reliable and requires laparoscopy. Acknowledgments. The authors acknowledge Willem M. Ankum, M.D., Johan V.Th.H. Hamerlynck, M.D., and Frits B. Lammes, M.D., from the department of Obstetrics and Gynaecology, Academic Medical Center, University of Amsterdam, for critically reading the manuscript. REFERENCES 1. Hutchins CJ. Laparoscopy and hysterosalpingography in the assessment of tubal patency. Obstet Gynecol 1976;49: 325-7. 2. Rice JP, London SN, Olive DL. Revaluation ofhysterosalpingography in infertility investigation. Obstet Gynecol 1986; 67:718-21. 3. La Sala GB, Sachetti F, Degl'incerti-Tocci F, Dessanti L, Torelli MG. Complementary use of hysterosalpingography, hysteroscopy and laparoscopy in 100 infertile patients: results and comparison of their diagnostic accuracy. Acta Eur Fertil 1987; 18:369-74. 4. Taylor PJ. Correlations in infertility: symptomatology, hyste rosalpingography, laparoscopy and hysteroscopy. J Reprod Med 1977; 18:339-42. 5. Okonofua FE, Essen UI, Nimalaraj T. Hysterosalpingography versus laparoscopy in tubal infertility: comparison based on findings at laparotomy. Int J Gynaecol Obstet 1989; 28:143-7. 6. El Minawi MF, Abdel Hadi M, Ibrahim AA, Wahby O. Comparative evaluation of laparoscopy and hysterosalpingography in infertile patients. Obstet Gynecol 1978;51:29-32. 7. Duff DE, Fried AM, Wilson EA, Haack DG. Hysterosalpingography and laparoscopy: a comparative study. Am J Radiol 1983; 141:761-3. 8. Midgette AS, Stukel TA, Littenberg B. A meta-analytic method for summarizing diagnostic test performances: re- Fertility and Sterility

ceiver operating characteristic-summary point estimates. Med Decis Making 1993;13:253-7. 9. SAS Institute Inc. SAS/STAT user's guide, version 6. 4th ed. Vol. 1. Cary (NC): SAS Institute Inc., 1989. 10. DerSimonian R, Laird N. Meta-analysis in clinical trials, Controlled Clin Trials 1986; 7:177-88. 11. Littenberg B, Moses E. Estimating diagnostic accuracy from multiple conflicting reports: a new meta-analytic method. Med Decis Making 1993; 13:313-21. 12. Swolin K, Rosencrantz M. Laparoscopy vs. hysterosalpingography in sterility investigations, a comparative study. Fertil Steril 1972;23:270-3. 13. Maathuis JB, Horbach JGM, van Hall EV. A comparison of the results of hysterosalpingography and laparoscopy in the diagnosis of fallopian tube dysfunction. Fertil Steril 1972;23:428-31. 14. Duignan NM, Jordan JA, Coughlan BM, Logan-Edwards R. One thousand consecutive cases of diagnostic laparoscopy. J Obstet Gynaecol Br Commonw 1972;79:1016-24. 15. Keirse MJNC, Vandervellen R. A comparison of hysterosalpingography and laparoscopy in the investigation of infertility. Obstet Gynecol 1972;41:685-8. 16. Philipsen T, Hansen BB. Comparative study ofhysterosalpingography and laparoscopy in infertile patients. Acta Obstet Gynecol Scand 1981;60:149-51. 17. Montanari L, Bulgarelli C, Marra A. A comparison ofhysterosalpingography and laparoscopy in the investigation of infertility. Clin Exp Obstet Gynecol 1982;9:66-8. 18. Ismajovich B, Wexler S, Golan A, Langer L, Menachem PD. The accuracy of hysterosalpingography versus laparoscopy in evaluation of infertile women. Int J Gynaecol Obstet 1986; 24:9-12. 19. La Sala GB, Sachetti F, Cigarini C, Sartori F, Salvatore V. Hysterosalpingography versus chromosalpingolaparoscopy: comparison of results in 174 patients with sterility problems. Acta Eur Fertil 1986; 17:355-9. 20. Loy RA, Weinstein FG, Seibel MM. Hysterosalpingography in perspective: the predictive value of oil-soluble versus watersolubie contrast media. Fertil Steril 1989;51:170-2. 21. Nordenskjold F, Ahlgren M. Laparoscopy in female infertility. Diagnosis and prognosis for subsequent pregnancy. Acta Obstet Gynecol Scand 1983;62:609-15. 22. Snowden ED, Jarrett JC II, Dawood MY. Comparison of diagnostic accuracy oflaparoscopy, hysteroscopy, and hysterosalpingography in evaluation of female infertility. Fertil Steril 1984;41:709-13. 23. World Health Organization. Comparative trial oftubal insufflation, hysterosalpingography, and laparoscopy with dye hydrotubation for assessment of tubal patency. Fertil Steril 1986;46:1101-7. 24. Donnez J, Langerock S, Lecart CL, Thomas K. Incidence of pathological factors not revealed by hysterosalpingography but disclosed by laparoscopy in 500 infertile women. Eur J Obstet Gynecol Reprod BioI 1982; 13:369-75. 25. ReshefE, Daniel WW, Foster JC, Bradley EL, Blackwell RE, Younger JB. Comparison between I-hour and 24-hour followup in hysterosalpingography using oil based contrast media. Fertil Steril 1989;52:753-5. 26. Watson A, Vandekerckhove P, Lilford R, Vail A, Brosens I, Hughes E. A meta-analysis of the therapeutic role of oil soluble contrast media at hysterosalpingography: a surprising result? Fertil Steril 1994;61:470-7. 27. Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical epidemiology. A basic science for clinical medicine. 2nd ed. Boston: Little, Brown and Company, 1991. 28. Dabekausen YAJM, Evers JLH, Land JA, Stals FS. Chlamydia trachoma tis antibody testing is more accurate than hysterosalpingography in predicting tubal factor infertility. Fertil SteriI1994;61:833-7. Note. Additional references are available upon request from the author. Vol. 64, No.3, September 1995 Swart et al. HSG for tubal pathology: a meta-analysis 491