New evidence in diagnosis of pulmonary embolism during pregnancy

Abstract

Diagnosis of pulmonary embolism (PE) in pregnancy is notoriously difficult and lacking high quality evidence. Three studies (DiPEP, ARTEMIS and CT-PE-Pregnancy) evaluating a systematic approach to PE diagnosis have recently been published.

DiPEP is a retrospective case-control study that found a poor utility of clinical decision rules or D-dimer testing for PE diagnosis in pregnancy. ARTEMIS and CT-PE-Pregnancy are well conducted prospective management studies that proposed two algorithms with different clinical decision rules and D-dimer criteria for the diagnosis of PE in pregnancy. They included few events in high risk patients, which makes difficult the assessment of both algorithm's safety in women with a high probability of PE.

Considering this new evidence, D-dimer testing might be useful to avoid radiation imaging in pregnant women considered at low risk for PE. In contrast, a negative D-dimer cannot be considered sufficiently safe to rule out PE when clinicians estimate that PE is the most likely diagnosis.

Background

Pregnancy increases the risk of venous thromboembolism (VTE) fourfold to fivefold compared to the general population.^1,2 Pulmonary embolism (PE) is an important cause of morbidity and one of the leading causes of mortality during pregnancy in developed countries.^1,3–5 Diagnosis in pregnancy is challenging as symptoms and signs of PE (e.g. dyspnea and mild tachycardia) are similar to features of a normal pregnancy and clinical decision rules lack sensitivity and specificity during pregnancy.⁶ Use of D-dimer in pregnancy is controversial as it rises gradually during a normal pregnancy and can be increased by maternal conditions such as preeclampsia.^7,8 It often exceeds non-pregnant validated cut-off points, being likely to produce more false positive results.⁹ Studies evaluating D-dimer thresholds for VTE diagnosis in pregnancy only included a small number of patients.^10–12 Lower extremity ultrasonography yield is minimal in pregnant women with suspected PE without deep venous thrombosis (DVT) symptoms.¹³ Therefore, guidelines advise against the use of D-dimer in pregnancy and most recommend to proceed directly to thoracic diagnostic imaging such as CT pulmonary angiography (CTPA) or ventilation-perfusion scan in women with suspected PE and no symptoms of DVT.^14–16 However, because of a low-suspicion threshold, a final diagnosis of PE is only made in a minority (less than 7%) of pregnant women in whom it was initially suspected, as compared to 12–22% in the general population, leading to a high proportion of negative diagnostic imaging.^13,17–21

A validated diagnostic algorithm to evaluate suspected pregnancy-related PE could improve care by preventing unnecessary radiation to both mother and fetus.²² The Diagnosis of PE in Pregnancy (DiPEP), ARTEMIS and CT-PE-Pregnancy studies attempted to evaluate the use of clinical decision rules and D-dimer for the diagnosis of PE in pregnancy.

DiPEP

DiPEP is a multicenter prospective cohort study augmented with additional retrospective cases aiming to identify whether clinical features, clinical decision rules or D-dimer could be used to rule out PE in pregnancy. Cases with suspected PE in antepartum or postpartum women were identified from emergency departments or maternity units of 11 sites in the United Kingdom (UK). Retrospective cases with confirmed PE retrieved from the UK Obstetric Surveillance System (UKOSS) were also added. Women who required life support on admission or who were asymptomatic were excluded.

From March 2015 to August 2016, 324 eligible women with suspected PE were prospectively enrolled and 198 cases from the UKOSS database were identified. In the prospective patients, 18 PE were confirmed on imaging (5.56%), whereas 163 PE were diagnosed in UKOSS cases with post-mortem imaging or autopsy. The final population thus included 181 women, with a majority in third trimester (35%) or in postpartum (30%). Almost 50% of women with diagnosed PE had received thromboprophylaxis. Increasing age, personal history of VTE, increasing temperature, decreasing systolic blood pressure, decreasing oxygen saturation, and PE-related chest radiograph abnormalities were associated with increased odds of PE diagnosis in multivariate analysis. However, familial history of VTE reduced the likelihood of PE diagnosis. Six clinical decision rules were tested, including three rules developed following an expert consensus. Only the Well’s rule with strict interpretation (i.e. alternative diagnosis criterion positive only if PE is clearly considered the most likely diagnosis) had moderate efficacy (C-statistic of 0.73). All other rules (Pulmonary Embolism Rule-out Criteria, simplified revised Geneva score and the three consensus decision rules) had poor sensitivity or specificity. D-dimer measurement was done in 44 of the 198 women with diagnosed PE (22%), and in 156 of 324 women with suspected PE (48%). D-dimer was evaluated using the hospital laboratory threshold and trimester-specific thresholds (1.5x laboratory threshold in the second trimester and 2x in the third trimester). Poor sensitivity and specificity were found for both usual hospital threshold (88.4% and 8.8% respectively) and trimester-specific thresholds (69.8% and 32.8%). Other biomarkers were tested in a substudy of DiPEP and none was found to be useful for PE diagnosis.²³ The authors concluded that existing clinical decision rules and D-dimer have little discriminant value in the assessment of suspected PE in pregnant and postpartum women and that D-dimer in pregnant women produced imprecise estimates of accuracy.

Strengths of the DiPEP study included the high number of cases enrolled and inclusion of women with a high suspicion for PE (87% ended up having diagnostic imaging). One important limitation is that 90% of the diagnosed PE were obtained from the retrospective registry. Retrospective assessment of the most likely diagnosis required for clinical decision rules can be fraught with imprecisions. Selection bias confounds assessment of causality, for example considering long-haul travel and a family history of VTE that were found protective for PE, possibly due to a higher suspicion of PE in women with these known risk factors. Similarly, almost half of women diagnosed with PE had received thromboprophylaxis likely related to an increased baseline risk of VTE.

Chest radiograph abnormalities were associated with PE, but recruited women already had a high suspicion for PE so these findings would rarely change management. PE-related chest radiograph findings are notoriously insensitive and their presence in 7% of women without PE makes their findings clinically useless. Details are also missing for the exact radiographic findings and what imaging these women had to confirm PE. In women with initial chest radiograph abnormalities, the use of ventilation-perfusion scanning could lead to indeterminate or false positive results.²⁴

Considering D-dimer, results interpretation and generalization are difficult. The low specificity of D-dimer using hospital threshold (8.8%) is not unexpected due to its natural tendency to increase in normal pregnancies.⁷ However, its 88% sensitivity is surprisingly low potentially leading to dangerous false negative results. This is in contrast with its high sensitivity in the non-pregnant population and with another study which obtained a 100% sensitivity in pregnant women with suspected DVT.¹⁰ It must be emphasized that D-dimer levels were obtained in less than 25% of women with diagnosed PE and less than 50% of women with suspected PE. With such missing data selection bias is a real concern. The test characteristics might have been affected by the features of women that ended up being selectively tested for D-dimer (unclear symptoms or diagnosis, confounding features potentially leading to false positive imaging). There is no evidence proving that PE pathophysiology is different in pregnancy leading to a lower increase of D-dimer compared to outside pregnancy women.

Finally, recommendations for evaluation in non-pregnant patients with suspected PE include initially a pre-test probability assessment using a clinical decision rules then followed by D-dimer testing only in low-risk patients.^25,26 A significant proportion of women in DiPEP were potentially high risk due to features previously mentioned and measuring D-dimer in these women could lead to false negatives. The authors did not study D-dimer according to pre-test probability, which might have produced more relevant results for clinicians.

ARTEMIS

The ARTEMIS study is a multicenter prospective management study realized in 18 centers of Europe.²⁷ Pregnant women with suspected PE presenting to the emergency department or obstetrics ward were managed with a modified version of the previously published YEARS protocol.²¹ This algorithm evaluated the probability of suspected PE using three YEARS criteria and two D-dimer thresholds according to the number of positive criteria. The three YEARS criteria, derived from a prospective cohort study in a non-pregnant population, are signs of DVT, hemoptysis and PE considered as the most likely diagnosis.²⁸ A unilateral lower extremity ultrasound was obtained for women with signs of DVT and the presence of a proximal DVT was considered diagnostic for PE. If no YEARS criteria were present and D-dimer was below 1000 ng/ml, PE was considered ruled out. In women with one or more criteria, no evidence of DVT and a D-dimer below 500 ng/ml, PE was also considered ruled out. All other patients required imaging with CTPA.

From October 2013 through May 2018, 498 women were recruited of which 51% had no YEARS criteria and 49% met one to three criteria. The most frequent criterion was PE as the most likely diagnosis (89%), followed by DVT symptoms (19%) and hemoptysis (7.7%). Lower extremity ultrasound was performed in 43 of the 47 women with symptoms of DVT and was positive in three cases (7%). Ultrasound was also performed in 79 of the 498 patients without symptoms of DVT and was positive in one case (1%). A total of 20 women (4%) had confirmed PE: four women with DVT on ultrasound and 16 women with confirmatory CTPA. At three months follow-up, only one woman not initially diagnosed with PE (without YEARS criteria and with a D-dimer of 480 ng/ml) developed a proximal DVT (0.21%; 95% confidence interval, 0.04 to 1.2%). No PE was diagnosed during follow-up. The median D-dimer level was 505 ng/ml in the first trimester, 730 ng/ml in the second and 1120 ng/ml in the third. CTPA was avoided in 39% of patients in decreasing proportion from the first (65%) to third trimester (32%).

The authors concluded their algorithm provided solid evidence for the safe management of suspected PE in pregnant women with selective use of CTPA. The simple and easy to remember algorithm, the high inclusion and follow-up rate as well as the low rate of inconclusive CTPA (<10%) are notable strengths of this study. Women with prior VTE (6.0%) and known thrombophilia (2.8%) were included, but no details of thromboprophylaxis or rates of VTE in these higher risk women were provided. An important issue with ARTEMIS is its sample size calculated for a power of 80% to detect a maximum incidence of VTE recurrence of 2.7% at three months. This maximal VTE recurrence safety threshold is no longer recommended due to a decreasing prevalence of VTE in the general population in recent studies.²⁹ This is especially relevant in pregnant women with suspected PE as their VTE prevalence is even lower. A much higher number of recruited women would be required to achieve enough power to satisfy the safety cut-off now recommended. This is relevant as the trial might have been underpowered to detect a potentially clinically meaningful rate of VTE at three months.

A unique feature of the YEARS protocol is that if D-dimer were below 500 ng/ml, none of the YEARS criteria had importance as PE was automatically excluded. Out of 242 pregnant women with at least one YEARS criteria, only 31 had a negative D-dimer with no VTE during follow-up. This remains limited data considering that most algorithms in non-pregnant women suggest thoracic imaging in patients considered at high risk for VTE.^13,20 This also contrasts with cases of pregnant women with PE and D-dimer below the hospital threshold, which have been described in DiPEP.³⁰

Research in pregnancy is notoriously difficult and deviations to the protocol occurred in all groups.¹³ These deviations could affect study results, but it is reassuring that no PE was detected in women who underwent CTPA when it was not indicated.

When developing a clinical decision rule, variables used in the model are often derived from a specific population and then validated in a similar population, which can lead to optimistic estimates of performance.²⁹ It is therefore important to have an external validation to confirm the performance of this model especially because it was derived from a non-pregnant population. A total of 20 events were diagnosed, of which 16 required the algorithm (the 4 others had a positive lower extremity ultrasound). Four predictors were considered (three clinical criteria and the D-dimer level) for a number of events per predictor (EPP) of 4. Such a low number of events per predictor could unmask biases in estimates of predictor effects also causing an overfitted model.^29,31

D-dimer thresholds were derived directly from the YEARS protocol and not adjusted for the trimester of pregnancy. This lead to a much higher efficiency in the first trimester where 65% of women had PE ruled out without CTPA. This is noteworthy as the potential effect of fetal irradiation is thought to be most harmful for organogenesis and nervous system development between 4 and 15 weeks of gestation.^32,33 In contrast, most women in the second and third trimester had D-dimer levels above thresholds and the diagnostic efficiency of the algorithm declined (from 6.8% positive CTPA in the first trimester to 3.0% in the third trimester).

Finally, possible knowledge of D-dimer results by clinicians prior to the application of the YEARS criteria might have affected their assessment of PE as the most likely diagnosis. It is an important concern since this criterion was the most important and this bias could have increased the efficiency of the model.

CT-PE-Pregnancy

The CT-PE-Pregnancy study is a prospective management cohort study that enrolled outpatient pregnant women with suspected PE in 11 centers of France and Switzerland.¹³ Pre-test probability of PE was evaluated using the revised Geneva score (Table 1).³⁴ D-dimer was tested in all women and a cut-off of 500 ng/ml was used throughout the pregnancy. PE was excluded in women with low or intermediate probability and a negative D-dimer. Women with high pre-test probability or a positive D-dimer underwent bilateral whole leg compression ultrasound, followed by CTPA if the ultrasound was negative. Women with indeterminate CTPA results underwent ventilation-perfusion scanning.

Table 1.

Revised Geneva score.

Criteria	Points
Age above 65 years old	1
Previous DVT or PE	3
Surgery of fracture within one month	2
Active cancer	2
Unilateral lower limb pain	3
Hemoptysis	2
Heart rate
• 75–94 beats per minute	3
• 95 and above beats per minute	5
Pain on lower limb deep venous palpation and unilateral edema	4
Clinical probability
• Low	0–3
• Intermediate	4–10
• High	11 and above

DVT: deep venous thrombosis; PE: pulmonary embolism.

Between August 2008 and July 2016, 395 women were enrolled. The majority of women included had low (48.6%) or intermediate (50.6%) pre-test probability (n = 392). Of these 392 women, 46 (11.6%) had a negative D-dimer result, 341 (87%) had a positive D-dimer, and 5 (1.3%) had no D-dimer testing completed. As in ARTEMIS, the proportion of negative D-dimer declined with increasing gestational age from 25.2% in the first trimester to 4.2% in the third trimester. Only three women had a high pre-test probability; two were presumed with PE based on the diagnosis of proximal DVT at ultrasound, and one had a proven PE at CTPA. In total, PE was diagnosed in seven women with a proximal DVT on ultrasound, 19 with a positive CTPA and two with a high probability nuclear scan for a total of 28 women (7.1%). During the three months follow-up, no VTE were diagnosed.

Strengths of the trial include few exclusion criteria, no loss to follow-up and adjudication of all VTE events at follow-up. CT-PE-Pregnancy suffers from similar limitations than ARTEMIS. Its sample size was determined with a 3.0% upper limit of VTE incidence at three months, which might lead to an underpowered study with such a low disease prevalence in this population.³⁵ The low number of events again led to a very low events per clinical score predictor (2.6) questioning the validity of the revised Geneva clinical score in this population.^30,31

The revised Geneva clinical score has never been validated in the pregnant population and includes risk factors rarely (cancer, recent surgery) or never encountered (more than 65 years old) in this population. The score also allows points for a heart rate of 75 to 94 beats per minute, which is considered normal in late pregnancy.³⁶ This leads to a poor score calibration in pregnant women with suspected PE manifested by a low sensitivity and specificity.³⁷ As a consequence only three women out of the 28 diagnosed with PE were considered high risk proceeding directly to diagnostic imaging. Even in the non-pregnant population, the revised Geneva score might have lower accuracy than the Wells score or clinician gestalt.^38,39

The use of the standard D-dimer threshold for all women led to more false positives than in ARTEMIS, and thus avoided lung imaging in fewer women (11.6% of women compared to 39% in ARTEMIS).

Protocol deviations occurred in 49 women (12.4%), including five women who were not tested for D-dimer. The authors did not report whether these five women were ultimately diagnosed with PE. Four PE were diagnosed in women who did not undergo testing according to the protocol, which correspond to 14% of all PE diagnosed in the study. Also, of the 46 women with a negative D-dimer, 11 underwent thoracic imaging, which revealed to be negative in all cases.

Finally, the algorithm used leg compression ultrasound for most women with a minimal yield (proximal DVT was found in 2.0% of 328 examinations) and substantial resource utilization. This confirms that ultrasound in the pregnant population should be limited to women with symptoms of DVT.

Conclusion

Despite improvements in PE diagnosis for the general population, diagnosis of PE in pregnancy remains problematic. The three studies reviewed represent important steps toward better care of pregnant women with suspected PE and prove that high-quality prospective trial can be conducted in this population. However, results are conflicting and should be analyzed with caution.

Both ARTEMIS and CT-PE-Pregnancy aimed to rule-out PE with stratification according to pre-test probability and a negative D-Dimer result. However, they used different clinical decision rules and different D-dimer cut-offs. Overall, ARTEMIS results are more interesting and promising.

DiPEP showed that no unique clinical feature or clinical decision rule can be used to safely rule out PE during pregnancy. The presence of a personal history of VTE, hypotension or desaturation should increase clinical suspicion for PE and the presence of fever should not be used to exclude the diagnosis if PE is suspected. No D-dimer cut-off was found useful in DiPEP, but a high proportion of data was missing.

Implementation of these study protocols is impeded by issues such as potentially imprecise and overfitted models, low number of events and underpowered studies to detect a potentially clinically relevant VTE rate at follow-up in high-risk women. Since incidence of PE in pregnant women has decreased, it is likely that achieving a trial with a high rate of events at follow-up will be difficult, if possible at all. Despite these statistical issues, the fact that both algorithms have been prospectively validated in a single study and use divergent clinical decision rules and D-dimer cut-offs are issues difficult to overpass. Attempts to validate retrospectively these protocols in other cohorts are ongoing, but may not resolve these conflicting issues.^40,41

Considering this new evidence, D-dimer testing might be useful to avoid imaging in pregnant women considered at low risk for PE, especially in the first trimester. In contrast, a negative D-dimer cannot be considered sufficiently safe to rule out PE when clinicians estimate that PE is the most likely diagnosis. This is crucial as probability assessment of PE in pregnant women remains solely based on clinician’s experience and judgment.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Guarantor

Julien Viau-Lapointe.

Contributorship

All authors are acknowledged as contributing authors and approved manuscript submission.