Automated right ventricle-to-left ventricle diameter ratio predicts ICU stay for acute pulmonary embolism on CTPA examinations in the emergency department

Introduction

Acute pulmonary embolism (PE) is a leading cause of cardiovascular death throughout the world (1). Acute right ventricular (RV) dysfunction, defined as a rapid progressive syndrome from acute obstruction of the pulmonary artery (PA) by thrombus, results in impaired RV filling and/or reduced RV flow output (2). It is a critical determinant of short-term mortality in patients with PE and is commonly used in clinical guidelines as a risk stratification tool and prognostic marker in assessing outcomes (3-5). When RV enlargement is identified via CT or echocardiography, patients are classified as intermediate or high risk, contingent upon additional factors such as hemodynamic status and biomarker levels (3). The simplified pulmonary embolism severity index (sPESI) scoring system has been adapted for severity assessment of acute PE and incorporation of RV overload marker to the sPESI increases the identification of intermediate and high-risk PE patients and predicts short-term complications (6). Treatment for acute PE is based on risk stratification. Systemic thrombolysis is the first-line recommendation for patients with high-risk PE, however with hemodynamic instability or with contraindications to thrombolysis, surgical or percutaneous mechanical thrombectomy may be needed (7).

Computed tomography pulmonary angiography (CTPA) is recommended as a first-line examination for the diagnosis of acute PE (8). Not only does CTPA identify the pulmonary emboli, but it also allows the identification of RV dysfunction (9,10). Right ventricle-to-left ventricle diameter ratio (RV/LV) >1, main PA to ascending aorta diameter ratio, morphology of the interventricular septum, degree of contrast reflux into the inferior vena cava (IVC) are markers that support the diagnosis of right heart dysfunction on CTPA (9-12). Patients with RV dysfunction have a higher risk of adverse outcomes and often require inpatient treatment and close monitoring (13). RV/LV can be measured on the standard axial view that shows four chambers of the heart in the CTPA. An accurately measured RV/LV >1 on CTPA examinations and routine chest CT examinations with contrast can suggest RV enlargement (12).

Manual RV/LV is not routinely performed on all CTPA examinations, which may be due to increased time to do so in the era of ever-increasing imaging volumes, and radiology staff shortage. Additionally, manual measurements are subjected to inter-reader variability. Therefore, a reproducible and reliable automated artificial intelligence (AI) measurement tool to measure RV/LV can potentially be useful and increase the radiologist’s confidence in reporting it. In this study, we investigate the relationship between an automated RV/LV in acute positive PE examinations detected by AI on CTPA studies in the emergency department (ED) and various clinical parameters in predicting mortality and outcomes. The outcomes of interest for this study include 30- and 90-day mortality, intensive care unit (ICU) admission, ICU length stay, interventional procedures, including thrombolysis and thrombectomy, administration of medication, and ventilation use in the ICU. We present this article in accordance with the STROBE reporting checklist (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-aw-526/rc).

Methods

Study design and dataset collection

Selection of the cohort of acute PE examinations for the RV/LV

This retrospective study is Health Insurance Portability and Accountability Act (HIPAA) compliant and was approved by the institutional review board of the University of Texas Southwestern Medical Center (No. STU-2023-0769, on December 20, 2023). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The requirement for written consent was waived. The investigation was conducted at a tertiary care university hospital. During the study period from April 2022 to October 2023, for purposes unrelated to this investigation, a Food and Drug Administration (FDA)-approved commercially available AI tool (BriefCase for PE, Aidoc) trained to detect acute PE on non-electrocardiogram (ECG) gated CTPA examinations was used to evaluate all contrast-enhanced chest CT examinations with PE protocol, the results of which were available to the radiologist. An automated measurement of RV/LV was also provided by a commercially available FDA-approved AI tool by the same vendor (14) when a positive acute PE was detected by the PE detection AI algorithm. The result of the automated measurement of the RV/LV was not available to the radiologist at the time of clinical interpretation. All results of the AI tool were stored in a departmental data warehouse for investigational purposes.

The department data warehouse was searched for positive acute PE examinations designated by an AI algorithm on consecutive conventional CT angiogram examinations (i.e., CTPA protocol). The patient status, demographics, AI results, report narrative, and report impression were extracted. Inclusion criteria for this study included adult patients older than 18 years, patients enrolled in the ED, and examinations classified as acute PE by the AI algorithm. Repeat examinations of the same patient were excluded from the cohort to maintain sample independence in assessing patient outcomes (Figure 1).

Figure 1 Flow chart shows patient selection. AI, artificial intelligence; CTPA, computed tomography pulmonary angiography; PE, pulmonary embolism; RV/LV, right ventricle-to-left ventricle diameter ratio.

For purposes of this investigation, a natural language processing (NLP) algorithm provided by the AI vendor (NLP Research Engine, Aidoc) (15) was used to evaluate the clinical reports for language indicating the presence of acute PE. If the NLP results of the clinical reports matched AI results, the findings were considered congruent. The reports for the discrepant examinations i.e., positive by AI algorithm and negative by NLP were manually reviewed by one of the two trainee students [medical student (XX) and undergraduate student (XY)] following a training session (that included review of 20 reports), by a senior radiologist (YY) for detection of acute PE in the radiology reports. The same senior radiologist established a final consensus when necessary. The final classifications were assigned after the corrections were added to the cohort (Figure S1). The RV/LV provided by the AI algorithm was extracted for the above-selected cohort of acute PE examinations. The radiology reports of the final cohort were processed to search for words indicating RV dysfunction by “regular expression” (described in detail in Appendix 1) in the “Cardiovascular” subsection of the structured report using keywords described in detail (Appendix 1). Any structured reports that failed to detect the specified keywords, and any unstructured reports without a “Cardiovascular” subsection, were manually reviewed by a medical student (XX) trained by a senior cardio-thoracic radiologist with 14 years of experience. The senior radiologist corrected any indeterminate classifications. Signs of RV strain include PA enlargement, septal flattening, RV enlargement or dilated IVC, and reflux of contrast in the hepatic veins.

Variables, data measurement

In addition to conventional demographics, comorbidities like the history of chronic cardiopulmonary disease, cancer, and chronic PE were included in the patient’s data. Results of investigations including the highest heart rate, the lowest and highest temperature, the lowest systolic blood pressure, and the lowest arterial blood gas were recorded ±12 hours of the CTPA examinations. A sPESI score was calculated using the above variables (3).

Patient management and outcome

Records of medications, including vasopressors, thrombolytics, anticoagulants (Figure S2) and interventional procedures performed for catheter-directed thrombolysis/thrombectomy within ±12 hours of the CTPA examinations were extracted. The outcomes studied for the above cohort included ICU admission within ±12 hours of the CTPA examination, ICU duration stay and time of death from the electronic medical record. A 30- and 90-day all-cause mortality was calculated based on the CTPA examination date and death date.

Statistical methods

Median, inter-quartile range, mean, standard deviation, counts and percentages were reported when appropriate. All examinations were grouped based on the RV/LV (measured by AI algorithm) threshold of one. The Wilcoxon rank sum test and Chi-squared tests were used to assess the difference between the RV/LV group for continuous and categorical variables, respectively.

Logistic regression analyses were used to estimate the association between RV/LV and outcome (ICU admission, ventilation utilization in ICU, number of hours in ICU, 30-day mortality, and 90-day mortality) after adjusting for sPESI score. Sensitivity Analysis was also performed using a threshold of 1.18 for the RV/LV to assess mortality and ICU admission (Table S1).

Results

Initial search of the radiology data warehouse identified 4,520 CTPA examinations (i.e., CTPA protocol) performed in the study period, out of which 795 examinations performed in 708 unique patients were deemed positive for acute PE by the AI algorithm. A total of 197 inpatient and 84 outpatient examinations were excluded. Fifty-six additional repeat examinations were excluded to maintain sample independence in assessing outcomes. There were no new acute PEs in the excluded examinations during the study period. Out of the remaining cohort, 50 additional examinations were excluded following the adjudication process and manual review of the discrepant reports, as illustrated in Figure S1. The latter also included exclusion of non-diagnostic examinations and examinations with an impression of chronic PE in the radiology report. Therefore, the final cohort comprised 408 examinations from 408 patients (210 women, 197 men; mean age 62±17 years; median age, 65 years; age range, 49–74 years) to be included in the analysis (Figure 1).

Table 1 summarizes the characteristics of patients and examinations. Five patients were not included in the sPESI score category and four patients in the sPESI high-risk category as information on one of the variables for the score was not available in the electronic medical record (Table 2).

In patients with acute PE, 142 examinations (35%) had an RV/LV ≤1 and 266 examinations with acute PE (65%) had an RV/LV >1. The mean age in the RV/LV >1 group was significantly higher than the other group (66 vs. 62 years). The non-Hispanic whites and females constituted the largest group, 204 out of 408 examinations (50%). There were no significant differences in race, ethnicity, gender or sPESI score between the two groups.

There were nineteen unstructured reports. A total of 346 structured radiology reports were searched using a regular expression tool, and 62 reports—either unstructured or those in which the regular expression tool did not detect the specified keywords—were reviewed manually. Seventy-seven of 408 radiology reports (19%) for acute PE mentioned terms suggestive of RV dysfunction. Among the 266 examinations with an RV-LV >1, only 25% (66/266) of radiology reports contained keywords indicating RV dysfunction (Table 1), and the term “RV enlargement” was mentioned in 27 reports specifically, out of which 23 had an AI-detected RV/LV >1. Examinations with an AI-measured RV/LV >1 had a higher likelihood of reporting signs of RV dysfunction (25%) than those with an AI-measured RV/LV ≤1 (7.7%, P<0.001).

Outcomes

Out of 266 patients with RV/LV >1 on CTPA, 66 patients had ICU admission as compared to 21 out of 142 with an RV/LV ≤1. Patients with RV/LV >1 were more likely to be admitted into ICU ±12 hours of CTPA (25% vs. 15%, P=0.02) and more likely to undergo interventional radiology (IR) procedures (6.8% vs. 0.7%, P=0.006) compared to those with RV/LV ≤1 (Table 3). The 30-day mortality was significantly higher in the high-risk sPESI score (P<0.003). No significant difference was detected in medication, ventilation utilization, and number of hours in ICU if they were admitted to ICU. No significant difference was found in 30- and 90-day mortality in the two groups (P=0.40 and 0.70, respectively).

Multivariable logistic regression analyses

Patients with RV/LV >1 had higher odds of ICU admission both before and after adjusting for sPESI [unadjusted odds ratio (OR): 1.90 (1.12, 3.33), P=0.02; adjusted OR: 2.00 (1.14, 3.64), P=0.02]. No significant association between the RV/LV and the 30- or 90-day mortality rate was observed with or without adjusting for sPESI (Table 4).

Discussion

Our study investigated the prognostic value of an AI-based, automated RV/LV measurement in patients seen in the ED with acute PE on CTPA exams. We found that an automated RV/LV ratio of >1 was significantly associated with ICU admission and IR procedures.

Acute PE patients with accompanying RV enlargement have a higher risk for decompensation and death (16). Radiology reports use a variety of subjective and qualitative terms to describe an abnormal appearing RV in the setting of PE (17). The RV/LV ratio measured manually has been used as a quantitative measure of RV enlargement and has been shown to be of value (18). Increased RV/LV >0.9 has been shown as a negative prognostic factor even in patients without acute PE (19). However, manual measurements are known to be subject to inter-reader variability based on slice selection, measurement location, and reader experience (20,21). Therefore, a reproducible, automated measurement of RV/LV ratio could be useful in the routine, standardized assessment of RV enlargement which may be overlooked by the radiologist and increase the confidence of radiologists in reporting RV enlargement.

Prior studies using manual measurement of RV/LV ratio have demonstrated its association with patient outcomes. A recent study by Cho et al. (22) found a significantly higher ICU admission rate (28.05% vs. 11.61%) in RV/LV >1 group on CTPA, measured manually by the ED specialists for early risk stratification of acute PE patients. Our study also found that the automated RV/LV ratio was similarly associated with ICU admission.

In our study, an RV/LV threshold of >1 was not associated with a change in 30-day or 90-day mortality, similar to results reported by Cho et al. (22). RV dilatation in acute PE reflects early hemodynamic strain that often results in prompt management and close monitoring in critical care. ICU admission is therefore a more sensitive marker of acute physiological instability, whereas mortality is determined by a myriad of factors, including age, comorbidities, treatment decisions, and disease burden. Additionally, the low mortality rate in our cohort may have limited statistical power to detect a relationship with mortality. Other studies using manual measurement have found the RV/LV ratio >1 to be associated with an increased risk for all-cause mortality (10). However, our results are similar to those of Foley et al. (23) who did not find a significant change in mortality with an automated RV/LV threshold of one. Interestingly, they did find that an automated RV/LV of 1.18 was predictive of 30-day mortality, with a sensitivity of 100% and specificity of 54%. We did not find this in our study. The authors also investigated the prognostic value in patients with acute PE. However, their study included a relatively small sample size and was conducted in the United Kingdom with limited information regarding population heterogeneity. In contrast, our study includes a larger cohort drawn from an ED setting, variable population demographics and evaluates clinical outcomes in addition to mortality.

Prediction of mortality is complex due to heterogeneity in terms of comorbidities, symptoms, and initial presentation (24,25). These differences may make the use of a single threshold difficult to predict mortality.

In our study, 30-day mortality was significantly associated with a high-risk sPESI score, consistent with established literature (6). Although sPESI demonstrates high sensitivity (80–90%), its relatively low specificity (40–60%) limits its ability to reliably discriminate mortality risk. Recent studies have shown that deep learning-based prediction models incorporating clinical, demographic, and imaging features have shown superior performance compared with sPESI alone (26). Additionally, the pan-immuno-inflammation value (PIV) measured at admission has been reported as a predictor of in-hospital mortality and appears noninferior to the PESI score, highlighting the potential value of alternative and complementary risk stratification tools (27).

Our study has limitations. First, it is a single academic, tertiary center study. The study population was predominately white, which may not be generalizable to a diverse patient population. The study included a vendor-designed automated RV/LV measurement conditioned on PE positive by AI, and any false negative examinations by AI were not included in this study. Although the AI algorithm has demonstrated high sensitivity and specificity in a meta-analysis, a small number of acute pulmonary emboli may have been missed (28-30). The primary aim of this study was to assess outcomes associated with the RV/LV ratio among AI-positive examinations. Given the higher sensitivity of AI algorithms for central relative to peripheral emboli, any missed emboli were likely peripheral (31). The prevalence of RV/LV >1 in our cohort was 65%, comparable to the prevalence reported by Schoepf et al. (32) indicating that these missed peripheral emboli were unlikely to have significantly contributed to the RV strain. This study did not differentiate chronic (preexisting) AI or imaging-detected RV enlargement from acute enlargement which is dependent on availability of prior imaging studies. Since the study period included the initial implementation of the AI algorithm, earlier CT scans did not have automated RV/LV assessments, resulting in heterogeneous time points and variable reporting of RV enlargement. Importantly, the aim of our study was to evaluate the association between the automated RV/LV ratio and clinical outcomes, rather than to establish a causal relationship between acute PE and RV enlargement. This question could be a potential research question in future studies.

Our study was focused on the use of the tool in patients presenting to an ED and excluded the inpatient and outpatient populations; therefore, the results may not be generalizable to these groups.

Conclusions

Acute PE examinations in the ED with an automated measured RV/LV >1 have an association with ICU admission and are more likely to undergo interventional procedures. This study showed that potential applications of the AI tool can be used to assess the prognosis of acute PE patients with RV enlargement.

Acknowledgments

We would like to acknowledge the contributions of the faculty, residents, and staff of the Department of Radiology at University of Texas Southwestern Medical Center to the performance of this project. In addition, we would like to thank the members of the Health Information Resources Department at University of Texas Southwestern Medical Center for their support on this project.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-aw-526/rc

Data Sharing Statement: Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-aw-526/dss

Peer Review File: Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-aw-526/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-aw-526/coif). T.S. received research grants as principal investigator from Janssen R & D, Liquidia Technologies, United Therapeutics and Keros Therapeutics. T.S. has participated in advisory board for Liquidia Technologies and United Therapeutics Corporation. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the institutional review board of University of Texas Southwestern Medical Center (No. STU-2023-0769) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Goldhaber SZ, Bounameaux H. Pulmonary embolism and deep vein thrombosis. Lancet 2012;379:1835-46. [Crossref] [PubMed]
2. Gafoor K, Patel S, Girvin F, et al. Cavitary Lung Diseases: A Clinical-Radiologic Algorithmic Approach. Chest 2018;153:1443-65. [Crossref] [PubMed]
3. Konstantinides SV, Meyer G, Becattini C, et al. 2019 ESC Guidelines for the diagnosis and management of acute pulmonary embolism developed in collaboration with the European Respiratory Society (ERS). Eur Heart J 2020;41:543-603. [Crossref] [PubMed]
4. Howard LSGE, Barden S, Condliffe R, et al. British Thoracic Society Guideline for the initial outpatient management of pulmonary embolism (PE). Thorax 2018;73:ii1-ii29. [Crossref] [PubMed]
5. Wang D, Fan G, Zhang X, et al. Prevalence of long-term right ventricular dysfunction after acute pulmonary embolism: a systematic review and meta-analysis. EClinicalMedicine 2023;62:102153. [Crossref] [PubMed]
6. Barrios D, Morillo R, Lobo JL, et al. Assessment of right ventricular function in acute pulmonary embolism. Am Heart J 2017;185:123-9. [Crossref] [PubMed]
7. Carlon TA, Goldman DT, Marinelli BS, et al. Contemporary Management of Acute Pulmonary Embolism: Evolution of Catheter-based Therapy. Radiographics 2022;42:1861-80. [Crossref] [PubMed]
8. Torbicki A, Perrier A, Konstantinides S, et al. Guidelines on the diagnosis and management of acute pulmonary embolism: the Task Force for the Diagnosis and Management of Acute Pulmonary Embolism of the European Society of Cardiology (ESC). Eur Heart J 2008;29:2276-315. [Crossref] [PubMed]
9. Chornenki NLJ, Poorzargar K, Shanjer M, et al. Detection of right ventricular dysfunction in acute pulmonary embolism by computed tomography or echocardiography: A systematic review and meta-analysis. J Thromb Haemost 2021;19:2504-13. [Crossref] [PubMed]
10. Bach AG, Nansalmaa B, Kranz J, et al. CT pulmonary angiography findings that predict 30-day mortality in patients with acute pulmonary embolism. Eur J Radiol 2015;84:332-7. [Crossref] [PubMed]
11. Aviram G, Cohen D, Steinvil A, et al. Significance of reflux of contrast medium into the inferior vena cava on computerized tomographic pulmonary angiogram. Am J Cardiol 2012;109:432-7. [Crossref] [PubMed]
12. Meinel FG, Nance JW Jr, Schoepf UJ, et al. Predictive Value of Computed Tomography in Acute Pulmonary Embolism: Systematic Review and Meta-analysis. Am J Med 2015;128:747-59.e2. [Crossref] [PubMed]
13. Martinez Licha CR, McCurdy CM, Maldonado SM, et al. Current Management of Acute Pulmonary Embolism. Ann Thorac Cardiovasc Surg 2020;26:65-71. [Crossref] [PubMed]
14. Imbio CT RV/LV User Manual. 2022 [v3.1.0]. Available online: https://imbio-web-public.s3.amazonaws.com/RVLV+Analysis/Versions/3.1.0/UM-97-Imbio_CT_RVLV_SW_Manual.pdf
15. Wismüller A. Early-stage COVID-19 pandemic observations on pulmonary embolism using nationwide multi-institutional data harvesting. NPJ Digit Med 2022;5:120. [Crossref] [PubMed]
16. Gupta R, Ammari Z, Dasa O, et al. Long-term mortality after massive, submassive, and low-risk pulmonary embolism. Vasc Med 2020;25:141-9. [Crossref] [PubMed]
17. Kang DK, Thilo C, Schoepf UJ, et al. CT signs of right ventricular dysfunction: prognostic role in acute pulmonary embolism. JACC Cardiovasc Imaging 2011;4:841-9. [Crossref] [PubMed]
18. Becattini C, Agnelli G, Germini F, et al. Computed tomography to assess risk of death in acute pulmonary embolism: a meta-analysis. Eur Respir J 2014;43:1678-90. [Crossref] [PubMed]
19. Tang A, Hunsaker A, Hammer M. Significance of Right-to-Left Ventricular Ratio as a Quantitative Computed Tomography Biomarker in Patients With Negative Computed Tomography Pulmonary Angiograms. J Thorac Imaging 2022;37:181-6. [Crossref] [PubMed]
20. Araoz PA, Gotway MB, Harrington JR, et al. Pulmonary embolism: prognostic CT findings. Radiology 2007;242:889-97. [Crossref] [PubMed]
21. Lanham S, Maiter A, Swift AJ, et al. The reproducibility of manual RV/LV ratio measurement on CT pulmonary angiography. BJR Open 2022;4:20220041. [Crossref] [PubMed]
22. Cho SU, Cho YD, Choi SH, et al. Assessing the severity of pulmonary embolism among patients in the emergency department: Utility of RV/LV diameter ratio. PLoS One 2020;15:e0242340. [Crossref] [PubMed]
23. Foley RW, Glenn-Cox S, Rossdale J, et al. Automated calculation of the right ventricle to left ventricle ratio on CT for the risk stratification of patients with acute pulmonary embolism. Eur Radiol 2021;31:6013-20. [Crossref] [PubMed]
24. Lankeit M. Always think of the right ventricle, even in “low-risk” pulmonary embolism. Eur Respir J 2017;50:1702386. [Crossref] [PubMed]
25. Côté B, Jiménez D, Planquette B, et al. Prognostic value of right ventricular dilatation in patients with low-risk pulmonary embolism. Eur Respir J 2017;50:1701611. [Crossref] [PubMed]
26. Cicek V, Orhan AL, Saylik F, et al. Predicting Short-Term Mortality in Patients With Acute Pulmonary Embolism With Deep Learning. Circ J 2025;89:602-11. [Crossref] [PubMed]
27. Çiçek V, Yavuz S, Şaylık F, et al. Evaluation of pan-Immuno-Inflammation value for In-hospital mortality in acute pulmonary embolism patients. Rev Invest Clin 2024;76:065-79.
28. Weikert T, Winkel DJ, Bremerich J, et al. Automated detection of pulmonary embolism in CT pulmonary angiograms using an AI-powered algorithm. Eur Radiol 2020;30:6545-53. [Crossref] [PubMed]
29. Batra K, Xi Y, Bhagwat S, et al. Radiologist Worklist Reprioritization Using Artificial Intelligence: Impact on Report Turnaround Times for CTPA Examinations Positive for Acute Pulmonary Embolism. AJR Am J Roentgenol 2023;221:324-33. [Crossref] [PubMed]
30. DePry EX, Parmar V, Rajendran S, et al. Performance of FDA-Approved AI Algorithms in Detecting Acute Pulmonary Embolism on Computed Tomographic Pulmonary Angiography (CTPA): A Meta-Analysis of Real-World Retrospective Studies. Cureus 2025;17:e94391. [Crossref] [PubMed]
31. Cheikh AB, Gorincour G, Nivet H, et al. How artificial intelligence improves radiological interpretation in suspected pulmonary embolism. Eur Radiol 2022;32:5831-42. [Crossref] [PubMed]
32. Schoepf UJ, Kucher N, Kipfmueller F, et al. Right ventricular enlargement on chest computed tomography: a predictor of early death in acute pulmonary embolism. Circulation 2004;110:3276-80. [Crossref] [PubMed]