Prevalence of and Eligibility for Surveillance Without Anticoagulation Among Adults With Lower-Risk Acute Subsegmental Pulmonary Embolism

Key Points Question How prevalent is structured surveillance without anticoagulation for subsegmental pulmonary embolism in community practice, and what proportion of patients are surveillance eligible using modified American College of Chest Physicians (CHEST) criteria? Findings In this cohort study of 666 outpatients examined over 5 years after publication of the 2016 CHEST guideline, only 1 patient (<1%) with subsegmental pulmonary embolism underwent surveillance without anticoagulation in a community setting with excellent follow-up access. Using modified CHEST criteria, 35 patients (5%) with subsegmental pulmonary embolism were surveillance eligible. Meaning These findings suggest that only a small proportion of patients with subsegmental pulmonary embolism may be surveillance eligible, and structured surveillance is rarely used despite the CHEST guideline.


eMethods 1. Development of a Natural Language Processing Algorithm to Identity Radiology Reports of Computed Tomography Pulmonary Angiography Likely Positive for Subsegmental Pulmonary Embolism
We used natural language processing (NLP) algorithms developed for a separate study 50 to identify computed tomography (CT) pulmonary angiography radiology reports that were likely positive for pulmonary embolism. In that study, colleagues from our CREST Network manually reviewed a training dataset of 700 CT reports to identify tomograms performed to evaluate for pulmonary embolism and, if ordered to diagnose pulmonary embolism, then to identify positive or negative pulmonary embolism findings. This dataset was used to develop and iteratively refine 2 NLP algorithms: the first to identify CTs undertaken to diagnose pulmonary embolism (NLP1) and the second to identify CTs positive for pulmonary embolism (NLP2). Once both NLP algorithms had been finalized, they applied them to the remaining CT reports and reviewed an additional 1,500 reports (500 for NLP1 and 1000 for NLP2) to validate NLP performance. To establish inter-rater reliability, a random sample of 10% of the NLP derivation dataset was reviewed by 3 investigators. Against manual report review as the criterion standard, the NLP algorithms used in tandem demonstrated 93% sensitivity and 98% specificity in identifying pulmonary embolism, with positive predictive value of 85% and negative predictive value of 99%. The area under the receiver operator characteristic curve was 0.96 for correctly identifying negative CT pulmonary angiography studies (Wald 95% confidence interval [CI] 0.93 to 0.98). Interrater reliability kappa scores were 0.93 (95% CI 0.79-1.0) for NLP1 and 1.0 (perfect agreement) for NLP2. We went on to use these NLP algorithms in other pulmonary embolism studies. 35,41 For this current study, we used the above NLP algorithms to screen 151,275 CT reports and identified 14,196 (9.4%) that were likely positive for pulmonary embolism. We then developed a text string algorithm using an iterative process to identify the subcohort of reports with probable subsegmental pulmonary embolism. We searched for cases with 'filling defects?|emboli|embolism|embolus|thrombi|thrombus|clot) (up to 8 words) subsegmental' or 'subsegmental (up to 6 words) (embolus|embolism|emboli|thromboemboli|nonocculusive clot|filling defects?|foci of PE|PE)' to identify. We excluded cases with 'segmental (up to 2 words) subsegmental' or 'extending (up to 3 words) subsegmental' or 'saddle embolus/embolism/emboli' or 'not excluded at the subsegmental artery level'. We identified 917 (6.5%) of 14,196 reports that were likely positive for subsegmental pulmonary embolism. Study physicians then manually reviewed these 917 CT reports for criteria of study eligibility ( Figure 1 We included only outpatients, whose PE diagnosis was made in the outpatient clinic or ED.
"…have active cancer (particularly if metastatic or being treated with chemotherapy);"

Active cancer.
Any cancer treatment (chemotherapy, radiation, surgery) within the prior 6 months, Stage IV disease, or receiving comfort care.
"…or have no reversible risk factor for VTE such as recent surgery" No documented reversible VTE risk factor. These include recent (<3m): • Chronic pulmonary disease, including asthma • Heart failure, systolic or diastolic "Marked symptoms that cannot be attributed to another condition"

Marked symptoms:
• Those with syncope or presyncope 35 • Those whose general appearance was documented as "in distress" Clinical conditions not included in 2016 CHEST guideline and expert panel report Clinical instability c Defined by one or more of the following documented during the clinic or ED evaluation: Patient 2 did not meet modified CHEST or enhanced criteria for structured surveillance.
ED physician discussed case with hospitalist who recommended pulmonology consultation. Per pulmonologist, if no DVT on bilateral lower-extremity compression ultrasound, then defer anticoagulation. Patient transferred to observation unit. US negative. Patient remained stable and was discharged home without anticoagulation. Follow-up one day after ED discharge with PCP. PCP discussed case with a different pulmonologist who recommended anticoagulation. As the patient was still symptomatic, PCP initiated rivaroxaban and scheduled follow-up with pulmonology who recommended lifelong anticoagulation. No bleeding complications or VTE recurrence within 90 days.