ACADEMIC EMERGENCY MEDICINE 2012; 19:24–30 © 2012 by the Society for Academic Emergency Medicine
Objectives: Physical examination, ultrasonography, and laboratory tests fail to reliably establish the preoperative diagnosis of tubal rupture in patients with ectopic pregnancy (EP), leading to a high rate of diagnostic laparoscopy. The aim of this study was to construct and to evaluate a clinical prediction rule for tubal rupture screening based on a self-assessment questionnaire, among patients with EP.
Methods: A standardized questionnaire was constructed via semistructured interviews of patients with acute pelvic pain. Features associated with tubal rupture were then identified in 141 prospectively included patients with tubal pregnancy, including 30 with tubal rupture, in five hospitals. Multiple logistic regression was used to select the best combination of independent features for predicting tubal rupture. Cross-validation was with the jackknife method. The main outcome measure was diagnostic accuracy of the questionnaire for ruling out tubal rupture.
Results: Eighty-nine items characterizing acute pelvic pain were identified. Among them, four contributed independently to the diagnosis of tubal rupture: vomiting during pain, diffuse abdominal pain, acute pain for longer than 30 minutes, and flashing pain. The presence of one or more of these features had 93% sensitivity (95% confidence interval [CI] = 84% to 100%) and 44% specificity (95% CI = 35% to 53%) for tubal rupture, with a negative likelihood ratio for ruling out tubal rupture of 0.16.
Conclusions: These results suggest that a standardized questionnaire may contribute to ruling out tubal rupture in patients with EP.
Ectopic pregnancy (EP) occurs in 0.6% to 2.1% of all pregnancies in the United States1,2 and accounts for up to 5% of all maternal deaths in developed countries.3 The most common location for EP is the Fallopian tube, and growth of the pregnancy leads to tubal rupture in 18% to 35% of cases.4,5 Ruptured tubal pregnancy requires immediate surgery to avoid life-threatening bleeding with massive hemoperitoneum.
The combination of transvaginal ultrasound (US) and serial serum human chorionic gonadotropin (hCG) determination has proved to be very reliable for the early diagnosis of EP,6,7 but not for diagnosing tubal rupture.5,8 Although US can diagnose hemoperitoneum, tubal rupture may cause abundant bleeding after a normal initial US scan.
The diagnosis of tubal rupture is obvious in a woman with hemodynamic instability. In most cases, however, the clinical symptoms of tubal rupture are more subtle.8–10 The only available diagnostic method for definitively ruling out tubal rupture is laparoscopy.8 However, routine diagnostic laparoscopy is not recommended, as pharmacological treatment for uncomplicated and nonruptured EP is available.11,12
Careful history-taking has been suggested as contributing to the diagnosis of ruptured tubal pregnancy.8–10 A systematic analysis of the characteristics of acute pelvic pain may help to establish the diagnosis when other diagnostic tools are inadequate, as demonstrated for acute renal colic.13 At present, no questionnaire is available for identifying specific pain features related to life-threatening events in gynecologic emergencies. We designed a prospective study to examine whether standardized history taking using a self-assessment questionnaire based on the characteristics of acute pelvic pain can help to rule out tubal rupture in patients with known EP.
This study had two parts: 1) a qualitative study to design a self-assessment questionnaire for gynecologic emergencies (SAQ-GE) and 2) a clinical evaluation of the SAQ-GE in a prospective multicenter observational study, in which a clinical prediction rule for ruptured EP was developed using the SAQ-GE among women consulting for acute pelvic pain and diagnosed with EP. The study was approved by the French Department of Higher Education and Research (n°06.336) and by the French National Committee of Information Technology and Individual Liberties (n°906253). The Assistance Publique-Hopitaux de Paris (AP-HP) funded the study but had no role in data collection, management, analysis, or interpretation or in manuscript preparation, review, or approval.
Survey Content and Administration
Construction of the SAQ-GE. Selected women admitted for acute pelvic pain in 2003 at the Poissy-Saint Germain-en-Laye, France, hospital were interviewed. We interviewed patients diagnosed with gynecologic emergencies carrying a risk of life-threatening complications (e.g., hemodynamic instability and/or need for blood transfusion or progressive disease) if not treated appropriately and promptly. These diagnoses consisted of EP with tubal rupture or active bleeding, pelvic inflammatory disease, adnexal torsion, and appendicitis. We also included patients who were not at risk for life-threatening complications within the next few hours; here, the diagnoses were complicated benign functional cyst, uncomplicated EP, and miscarriage. In each diagnostic group, at least five patients were considered necessary to interview after surgical treatment.
In the semistructured interview, we obtained the patient’s own description of the pain and associated symptoms. Supplementary questions were asked, if necessary, to obtain additional information on pain location, intensity, course, and associated symptoms. All interviews were recorded and then transcribed. An adaptation of Colaizzi’s method14 was used for a qualitative assessment of the full-length recordings and transcripts. Important phrases and sentences were collected. Identified meanings were grouped into themes, then into theme clusters with common traits, and finally into more general categories. Topics and items reflecting common subjective experiences were extracted from the interviews by a research psychologist under supervision by one of the authors (AF).
The topics and items collected by semistructured interviews were then submitted to seven French experts identified by searching Medline for publications on pregnancy-related emergencies. Each expert was asked to rate each item on a 0–10 scale according to his or her perception of its usefulness for each diagnosis, to restrict the number of items of the questionnaire.
The SAQ-GE was finally built with items identified from the interviews that were often associated with potentially life-threatening gynecologic emergencies and/or specific for potentially life-threatening gynecologic emergencies and/or given high ratings by the experts. To improve comprehension by the patients, we worded the items using the phrases and sentences collected from the semistructured interviews and modified during a pilot study (data not shown).
Derivation of a Clinical Prediction Rule for Ruptured EP Using the SAQ-GE
Study Setting and Population. From September 2006 to March 2008, we asked all patients aged 18 years or older presenting at the gynecology emergency departments (EDs) of the study hospitals for acute pelvic pain and/or vaginal bleeding to complete the SAQ-GE on a voluntary basis. In France, the diagnosis of acute pelvic pain can be made in general EDs, in which case the patient is referred to a gynecologic ED, or directly in gynecologic EDs, to which all patients have free access. Thus, all patients with gynecologic emergencies are seen in gynecologic EDs.
The patients were enrolled at five gynecologic EDs in the Paris greater metropolitan region. Four departments were in teaching hospitals (Poissy-Saint Germain en Laye, Créteil, Port-Royal, and Louis Mourier) and one was in a general hospital (Versailles).
Study Protocol. The questionnaire was self-completed at arrival in the gynecologic ED or within the first few hours after admission to the gynecologic ED, before surgery if surgery was needed, and after appropriate pain management. The ED staff gave the questionnaire to the patients and encouraged them to complete it. Patients who underwent laparoscopy completed the SAQ-GE before the procedure. Exclusion criteria were a history of chronic pelvic pain, neurologic or psychiatric disease, hemodynamic instability, no knowledge of French, and no health insurance. Only patients whose final diagnosis was tubal pregnancy were included in this part of the study.
Ectopic pregnancy was diagnosed using either an algorithm15 or laparoscopy. All centers used similar algorithms based on transvaginal US and quantitative hCG measurement. Urine hCG testing was performed routinely before serum hCG measurement and US. Clinical and US examinations were performed by the gynecology resident on duty at the gynecologic ED, under the supervision of a senior gynecologist. Transvaginal US and the hCG assay were repeated every 48 hours until the definitive diagnosis was established. Laparoscopy was performed in patients with suspected complications of EP (abundant hemoperitoneum with active bleeding) or contraindications to medical treatment.
Patients were selected for medical treatment of EP as recommended by the French College of Gynecologists and Obstetricians based on absent or moderate spontaneous pelvic pain, minimal symptoms by physical examination, hCG < 5000 mIU/mL, hematosalpinx < 4 cm, hemoglobin > 10 g/dL, and absence of contraindication to methotrexate.16 All medically treated patients were followed until the hCG assay became negative and were therefore classified as having nonruptured EP.
Key Outcome Measures. The reference standard for the diagnosis of tubal rupture was laparoscopic visualization of a tubal wall rupture. The operative reports were standardized and the amount of peritoneal blood was recorded. Patients in whom tubal rupture occurred during laparoscopy or who had an intact tubal serosa were classified as having nonruptured EP.
Data Analysis. The required sample size was estimated as follows. We expected to be able to derive, based on the appropriate selection of several questionnaire items, a clinical prediction rule for preoperative tubal rupture screening in patients with known EP. To rule out ruptured EP, our analysis had to focus on sensitivity (SE),17 to ensure the lowest possible false-negative rate (1 – SE). To be of clinical interest, the rule had to have at least 95% SE.18 The questionnaire would be deemed ineffective if we were unable to derive a model having at least 80% SE. Further investigations would be considered necessary if the SE was between 80 and 95%. On a pragmatic basis, the risk of type III errors19 has been used to determine the required sample size. We chose this approach because our study was done from a pragmatic perspective.19 The two type III errors are defined here as the probability that the observed SE is above 95% while the true SE is equal to 80%, and the probability that the observed SE is under 80% while the true SE is equal to 95%. Based on the binomial distribution, and assuming a mean rate of EP rupture of 25%, including 120 patients with EP, guarantees that the two type III errors are below 0.025 (one-sided).
We compared the questionnaire items selected by patients with and without tubal rupture. Pearson’s chi-square test or Fisher’s exact test was used for the bivariate analyses. For the variables with p-values smaller than 0.05 in the bivariate analysis, we computed SE, specificity (SP), the positive likelihood ratio (LR+), the negative likelihood ratio (LR–), and the crude diagnostic odds ratio (OR). To avoid item redundancy, we performed a principal component analysis (PCA) to assess correlations between the selected items.
We then used multiple logistic regression analysis to select the best combination of variables for predicting tubal rupture among variables with p-values smaller than 0.05 in the bivariate analyses. We performed a forward stepwise procedure to select variables based on the crude diagnostic OR. We sought to identify combinations of variables independently associated with tubal rupture (p < 0.05). Adjusted diagnostic ORs were computed with their 95% confidence intervals (95% CIs), and robust estimations of variances were performed using the sandwich estimator.20 The predictive performance of the logistic regression model was then assessed by determining the probability of tubal rupture in the overall study population. Cross-validation of the model was with the jackknife procedure.21
To construct a clinical prediction rule, selected variables were combined. The performance of each combination for diagnosing tubal rupture was assessed by calculating SE, SP, LR+, and LR– based on the patients in our sample, thus defining several risk groups. All analyses were carried out using Stata version 11.0 (Stata Corp., College Station, TX).
Thirty-nine patients were interviewed and evaluated after surgery, including 13 with EP and five with miscarriages. Of the 13 patients with EP, seven had Fallopian tube rupture. Based on these semistructured interviews, we built an 89-item SAQ-GE. The 89 items were divided into six topics: qualitative pain description, pain intensity, pain location, pain course, whether vaginal bleeding was present, and whether other signs were present.
Multicenter Observational Study
At the five study centers, 507 patients completed the SAQ-GE (Figure 1). Among them, 142 were diagnosed with EP. Their mean (±SD) age was 31.7 ± 5.3 years. Laparoscopy was performed in 120 patients and tubal rupture was diagnosed in 30 patients (21% of those with EP). Medical treatment was used in 22 patients and failed in one patient, who underwent secondary laparoscopy; the other 21 patients each had an uneventful course during follow-up until the hCG assay reverted to negative. One patient had a nontubal epiploic pregnancy and was excluded, leaving 141 patients for the analysis.
The bivariate analyses identified 15 items that predicted tubal rupture (Table 1). Among them, only pain during movement was more than 95% sensitive for tubal rupture. Three items were more than 90% specific for tubal rupture, namely, shoulder pain, vomiting during pain, and syncope (Table 1). PCA showed no correlations >0.5 between any pairs of significant items. Multiple logistic regression identified six variables independently associated with tubal rupture (Pearson chi-square = 28.80; p = 0.99): vomiting during pain, diffuse abdominal pain, pain lasting longer than 30 minutes, flashing pain, no brown vaginal bleeding, and pain during movement (Table 2). Robust estimation of variance retrieved the six variables independently associated with tubal rupture.
|Questionnaire Item||Group Selecting||Total, n/N* (%)||Tubal rupture, n/N (%)||Nonruptured EP, n/N (%)||Sensitivity (%)||Specificity (%)||Crude DOR† [95%CI]|
|Sudden onset of pain||Experts||77/127 (60.6)||21/27 (77.8)||56/100 (56)||78||44||2.75 [1.00–7.55]|
|Pain duration >30 minutes||Patients||47/109 (43.1)||16/24 (66.7)||31/85 (36.5)||67||64||3.48 [1.29–9.41]|
|Ovarian pain||Patients||75/128 (58.6)||21/27 (77.8)||54/101 (53.5)||78||46||3.05 [1.11–8.39]|
|Shoulder pain||Both experts and patients||12/131 (9.2)||6/27 (22.2)||6/104 (5.8)||22||94||4.67 [1.32–16.5]|
|Diffuse abdominal pain||Both experts and patients||42/129 (32.6)||18/27 (66.7)||24/102 (23.5)||67||76||6.50 [2.40–17.6]|
|Pelvic pain radiating to stomach||Patients||27/128 (21.1)||11/27 (40.7)||16/101 (15.8)||41||84||3.65 [1.39–9.63]|
|Pain during movement||Patients||94/128 (73.4)||26/27 (96.3)||68/101 (67.3)||96||33||12.6 [1.51–106]|
|No vaginal brown bleeding||Patients||66/131 (50.4)||20/28 (71.4)||46/103 (44.6)||71||55||3.10 [1.22–7.88]|
|Nausea||Patients||69/133 (51.9)||21/28 (75)||48/105 (45.7)||75||54||3.56 [1.35–9.39]|
|Vomiting during pain||Patients||20/131 (15.3)||12/27 (44.4)||8/104 (7.7)||44||92||9.60 [3.03–30.5]|
|Dysuria||Patients||23/133 (17.3)||9/28 (32.1)||14/105 (13.3)||32||87||3.08 [1.14–8.34]|
|Diaphoresis, hot flashes||Experts||63/132 (47.7)||23/28 (82.1)||40/104 (38.5)||82||62||7.36 [2.39–22.6]|
|Fatigue||Both experts and patients||76/133 (57.1)||22/28 (78.6)||54/105 (51.4)||79||49||3.46 [1.26–9.51]|
|Flashing pain||Patients||30/122 (24.6)||12/26 (46.1)||18/96 (18.8)||46||81||3.71 [1.42–9.73]|
|Syncope||Experts||10/115 (8.7)||7/26 (26.9)||3/89 (3.4)||27||97||10.6 [2.25–49.5]|
|Model||Adjusted Diagnostic OR||95% CI||Sensitivity (%)||Specificity (%)||Unbiased Sensitivity (%)||Unbiased Specificity (%)|
|Complete model, AUC = 0.92||0.88–0.97||70.0||93.7||47.0||92.7|
|Pain during movement||29.8||1.37–651|
|Vomiting during pain||26.0||5.03–134|
|Diffuse abdominal pain||9.17||2.35–35.9|
|Pain duration > 30 minutes||9.02||2.06–39.5|
|No vaginal brown bleeding||5.08||1.33–19.3|
|Simplified model, AUC = 0.89||0.83–0.95||50.0||95.5||43.0||94.6|
|Vomiting during pain||21.4||5.11–89.5|
|Diffuse abdominal pain||7.23||2.29–22.9|
|Pain duration > 30 minutes||6.12||1.69–22.2|
The jackknife procedure showed that SE decreased slightly, whereas SP remained unchanged (Table 2). We then constructed a simplified model by removing the two items not present in all jackknife models, namely, absence of brown vaginal bleeding (absent twice) and presence of pain during movement (absent three times). The final model without these two variables was significant in all ten iterations and was more than 95% specific when the logistic regression model equation was used: Pr = 1/(1 + e(4.36 0–1.64 × flashing pain −1.98 × diffuse abdominal pain −1.81 × pain lasting longer than 30 minutes −3.06 × vomiting during pain) (Table 2). All four items were selected from the patient interviews during questionnaire development. Only diffuse abdominal pain was selected both by the patients and by the experts.
Two clinical prediction rules were built using the four selected criteria (Table 3). The first clinical prediction rule defined patients at low risk for rupture. In this group of patients having none of the criteria, the probability of tubal rupture was 3.9%, SE was 44%, and SP was 93%. Inversely, presence of one or more of the four criteria had 93% SE (95% CI = 84% to 100%) for tubal rupture with an LR– of 0.16 for ruling out tubal rupture. Two of the patients with none of the four criteria had tubal rupture (false-negatives). However, laparoscopy was indicated in both, based on US signs of hemoperitoneum. The second clinical prediction rule defined patients at high risk for rupture, defined as patients having three or four of the four criteria. The probability of tubal rupture in this high-risk group was 73%. Presence of three or more of the criteria had 97.3% SP (95% CI = 94% to 100%) for tubal rupture with an LR+ of 9.9 (Table 3).
|Risk||Patients With the Characteristic||Probability of Tubal Rupture||Sensitivity (%)||Specificity (%)||LR+||LR–|
|None of the criteria†||51||3.9%||44*||93*||6.3*||0.60*|
|One or more criteria||90||31.1%||93||44||1.66||0.16|
|Three or more criteria||11||72.7%||26.7||97.3||9.9||0.75|
We developed a clinical prediction rule for tubal rupture in patients with EP, based on four simple items from a new self-assessment questionnaire: vomiting during pain, diffuse abdominal pain, acute pain for longer than 30 minutes, and flashing pain. Although the sensitivity of our clinical prediction rule was lower than expected, the diagnostic performance of our clinical prediction rule led to an LR– of 0.16, which is encouraging18 and suggests usefulness for triaging patients in gynecologic EDs.
Two previous studies suggested that asking patients about spontaneous acute pelvic pain might help to diagnose tubal rupture in patients with EP.8,10 However, neither study provided a detailed analysis of the pain features. The only prospective study specifically designed to evaluate history-taking for tubal rupture diagnosis in patients with EP concluded that clinical symptoms were poor predictors of tubal rupture.9 However, only a few symptoms were studied, and only the presence of abdominal pain, dizziness, and shoulder pain were associated with tubal rupture in bivariate analyses. The discrepancy with our study may be ascribable to the difference in the way the symptoms were assessed.
A major strength of our study is that the descriptors of pelvic pain and other symptoms were based on the subjective experience of the patients14 and not on physicians’ perceptions of symptoms, which were used in earlier studies.22,23 Colaizzi’s method was chosen because it rests on a phenomenologic approach, i.e., on the progressive emergence of the patient’s subjectivity through an open interview.14 As opposed to most methods used in sociologic or anthropologic studies, this approach is free of theoretical preconceptions and is therefore particularly well suited to medical situations. Of the four criteria independently associated with tubal rupture, only diffuse abdominal pain was among the criteria selected by the panel of experts in the preliminary qualitative phase of the study. Furthermore, shoulder pain and syncope, two signs widely held by physicians to indicate tubal rupture and selected by the experts,24 were not significantly associated with tubal rupture in the final model. Shoulder pain and syncope were associated with tubal rupture in the bivariate analyses, and the strength of the associations was similar to that in the study by Hirata et al.9 Last, items of the SAQ-GE that significantly predicted tubal rupture were not correlated in the PCA.
The relations between verbal descriptors of pain and the pathologic abnormalities identified during laparoscopy suggest that the characteristics of pelvic symptoms as described verbally by patients with gynecologic emergencies may reflect the underlying sensory experience specifically associated with each pathologic abnormality.25,26 Pathophysiologic considerations may help to explain the associations found in our study. Vomiting during pain may indicate autonomic nervous system damage during tubal rupture or a vagal reflex triggered by severe pain.25,27“Flashing pain” is a term used in the McGill Pain Questionnaire26 and translated in the validated French and Canadian version as “douleur en éclair,”28 which was the term used in our questionnaire. Lightning pain is another possible translation. In an experimental study of electrical ileosigmoidostomy stimulation in nine patients, flashing pain was one of the most common terms of the McGill Pain Questionnaire used to describe the effect of high-intensity stimuli.29 Tubal rupture may result in severe intraluminal visceral nociception. Thus, the words used by women to describe pain from tubal rupture may closely reflect pathophysiologic mechanisms.25,26
First, the pain features were identified in ED patients with acute pelvic pain who had not yet undergone diagnostic or therapeutic interventions. This recruitment strategy may have biased the population toward patients with less severe illness. Our clinical prediction rule identifies cases of ruptured EP of sufficient clinical severity to result in laparoscopy. We cannot exclude that some patients not identified by the rule had minor tubal rupture with spontaneous discontinuation of bleeding. Although excluding hemodynamically unstable patients was reasonable, patients with severe pain may have been less likely to complete the SAQ-GE, leading to selection bias. To limit this source of bias, a strict pain management protocol was implemented in the EDs of all study centers. Furthermore, pelvic pain intensity in the patients included in the study was high (75% of patients had a score of 7 or more on an 11-point verbal rating scale). On the other hand, the small proportion of patients with EP who were managed medically in our study may reflect the fact that patients with uncomplicated EP may have no pain and would therefore not have been eligible for the study. Although diagnostic bias is a possibility, it is highly unlikely that any of the 21 patients managed medically had ruptured EP, as none experienced any adverse events during follow-up until reversion of the hCG assay to negative. Finally, tubal rupture was diagnosed in 21% of our patients with EP, which was lower than the expected proportion of 25%, suggesting spectrum bias.30
Last, there was a high risk of type I error when identifying items associated with tubal rupture.21,31 However, the risk of type I errors was decreased by deriving all questionnaire items from an independent qualitative analysis specifically designed to investigate verbal descriptors of acute pelvic pain used by patients with gynecologic emergencies. According to the recommended minimum “rule of thumb” of 10 events per predictor,32 overfitting of our initial model may have occurred before cross-validation. However, Vittinghoff and McCulloch33 have argued in favor of relaxing this rule of 10 events per variable in logistic regression models. In addition, we used a stringent variable selection procedure with a robust variance estimator for our logistic regression analysis, and we performed a jackknife procedure to correct for overfitting.31
This study highlights the potential usefulness of standardized history-taking based on a self-assessment questionnaire for the diagnosis of tubal rupture among patients with ectopic pregnancy. We suggest that the four pain descriptors identified in our study be used for triaging women with suspected ectopic pregnancy (early pregnancy with bleeding and/or pain) before making decisions about medical or surgical management. Our prediction rule needs to be prospectively tested in a large population of women presenting with ectopic pregnancy to determine its exact value.