Electronic Radiology Reports Research Articles

O-P08 Optimising design of national collaborative studies in surgery: national availability of electronic data in the NHS

Abstract Background Clinical trials are constrained in part by cost and limited funding opportunities. Trainee-led collaborative, multi-centre research models have become increasingly popular, enabling clinicians to undertake impactful national and international practice-changing studies with limited funding. However, ensuring equitable and efficient data collection from participating centres requires understanding data accessibility. Here, we reflect upon our strategies in designing and delivering collaborative research (The RICOCHET Study). Specifically, we consider the availability of data sources for UK hospitals, to enable effective design of study case report forms to ensure study success. Methods RICOCHET was a National trainee-led audit of UK hospitals treating patients with pancreatic cancer. All registered hospitals were asked to complete a site-survey at point of registration. The site survey was completed by the lead consultant (surgeon or gastroenterologist) or the lead trainee at each site. The site-survey detailed hospital demographics, including size, availability of services and availability of patient data electronically. The survey was completed online via REDCap. Sites lacking full reporting were excluded from the analyses. Results 98 UK sites registered for RICOCHET, and 94 completed the site-survey (26 specialist pancreatic sites; 68 non-specialist sites). 100% sites had electronic radiology reports. However, electronic ward-round notes were available at only 19% (5/26, 19.2% specialist vs. 13/68, 19.1% non-specialist; p = 0.99). Furthermore, sources such as external letters were low (29%; 11/26 42.3% specialist vs. 16/68, 23.5% non-specialist; p = 0.072). Although electronic interventional radiology reports were broadly available at 79% sites, this was significantly lower at non-specialist sites (49/68 72.1% vs. 25/26 96.2%; p = 0.011). Conclusions Availability of electronic data is not consistent across sites managing pancreatic cancer patients. Significant disparity exists between specialist and non-specialist centres, highlighted by lack of availability of interventional radiology reports. To ensure success in trainee-led cohort studies we recommend designing concise case report forms that can be easily completed from electronic data sources. Until this report, there was no data available for the availability of electronic data across the NHS. This represents the most comprehensive report of such data sources, and will guide future study design of national studies at both specialist and non-specialist pancreatic centres in the UK.

Abstract P243: Identifying Cerebral Venous Thrombosis Through Administrative Data: Icd-10 Case Ascertainment Depends on Clinical Context

Background: Recent reported population-based rates of cerebral venous thrombosis (CVT) are higher than in older studies, though the context of these diagnoses is not well-defined. To better understand these trends, we examined the accuracy of administrative codes ( ICD-10 ) for CVT in different clinical scenarios. Methods: Cases of CVT presenting to a tertiary center between 2008-2018 were identified in two ways: free text search through all hospital electronic radiology reports regardless of modality and body part and any ICD-10 discharge codes (see Table 1). Electronic medical records were reviewed to verify diagnoses of CVT and their clinical context (Figure 1) to calculated Positive Predictive Value (PPV) of ICD-10 codes. Additionally, sensitivities of ICD-10 codes were calculated against all CVTs identified using either searches that were verified on chart review as the gold standard. Results: There were 289 confirmed cases: 239 new diagnoses, 204 of which were acute events. Only 75 cases (37%) were new, symptomatic CVTs not provoked by trauma or structural processes. Sensitivity and PPV for ICD-10 codes depending on clinical context is reported in Table 1. Conclusion: The majority of CVT identified at our institution were incidentally diagnosed in context of intracranial processes such as trauma, surgery, infection, or masses; 37% were symptomatic, non-structural incident diagnoses. Our findings have implications in interpreting CVT rates identified through administrative data, as the management and prognosis of CVT may differ based on clinical context.

Patient Safety Analysis in Radiation Burden of Head Computed Tomography Imaging in 1185 Neurosurgical Inpatients

We performed a retrospective analysis in a cohort of 1185 patients at our institution who were identified as undergoing ≥1 head computed tomography (CT) examinations during their inpatient stay on the neurosurgery service, to quantify the number, type, and associated radiation burden of head CT procedures performed by the neurosurgery service. CT procedure records and radiology reports were obtained via database search and directly validated against records retrieved from manual chart review. Next, dosimetry data from the head CT procedures were extracted via automated text mining of electronic radiology reports. Among 4510 identified adult head CT procedures, 88% were standard head CT examinations. A total of 3.65 ± 3.60 head CT scans were performed during an average adult admission. The most common primary diagnoses were neoplasms, trauma, and other hemorrhage. The median cumulative effective dose per admission was 5.66 mSv (range, 1.06-84.5 mSv; mean, 8.56 ± 8.95 mSv). The median cumulative effective dose per patient was 6.4 mSv (range, 1.1-127 mSv; mean, 9.26 ± 10.0 mSv). The median cumulative radiation burden from head CT imaging in our cohort equates approximately to a single chest CT scan, well within accepted limits for safe CT imaging in adults. Refined methods are needed to characterize the safety profile of the few pediatric patients identified in our study.

Identification of Patients with Osteoporotic Vertebral Fractures via Simple Text Search of Routine Radiology Reports.

Secondary fracture prevention programs mostly identify patients with symptomatic non-vertebral fractures, whereas asymptomatic vertebral fractures are usually missed. We here describe the development and validation of a simple method to systematically identify patients with radiographic vertebral fractures using simple text-based searching of free-text radiology reports. The study consisted of two phases. In the development phase (DP), twelve search terms were used to identify vertebral fractures in all X-ray and CT reports issued over a period of 6months. Positive reports were manually reviewed to confirm whether or not a vertebral fracture had in fact been reported. The three search terms most effective in detecting vertebral fractures during the DP were then applied during the implementation phase (IP) over several weeks to test their ability to identify patients with vertebral fractures. The search terms 'Loss of Height' (LoH), 'Compression Fracture' (CoF) and 'Crush Fracture' (CrF) identified the highest number of imaging reports with a confirmed vertebral fracture. These three search terms identified 581 of 689 (84%) of all true vertebral fractures with a positive predictive value of 76%. Using these three terms in the IP, 126 reports were identified of which 100 (79%) had a vertebral fracture confirmed on manual review. Amongst a sample of 587 reports in week 1 of the IP, 7 (1.2%) were false negatives. Many patients with vertebral fractures can be identified via a simple text-based search of electronic radiology reports. This method may be utilised by secondary fracture prevention programs to narrow the 'care gap' in osteoporosis management.

Intraperitoneal metastases after transarterial embolization of hepatocellular carcinoma: An observational study.

Transarterial embolization is frequently used to treat local hepatocellular carcinoma (HCC). While various complications are known to occur following transarterial embolization, only one prior case of peritoneal spread of HCC occurring shortly after transarterial chemoembolization has been reported. We present five cases of peritoneal spread of HCC following transarterial embolization (including bland embolization, conventional transarterial chemoembolization (TACE), and doxorubicin-eluting beads TACE) and identify features common among those cases. Search of electronic radiology reports and images identified five patients with imaging before and after treatment of HCC with transarterial embolization and with newly developed peritoneal metastases after treatment. Various patient demographics and tumor characteristics were noted. The mean maximal diameter of the treated HCC tumors was 3.7cm (range 1.4-11.9cm). Three of the patients had ascites and treated tumors in the posterior right hepatic lobe, and all patients had subcapsular tumors treated with transarterial embolization before developing peritoneal metastases. The mean time from treatment with transarterial embolization to the development of peritoneal metastases was four months. Intraperitoneal metastatic disease should be considered a rare but potential complication of transarterial embolization of subcapsular HCC, particularly in patients with ascites and tumors that are in the posterior segments of the right lobe. This potential complication should perhaps be considered when planning transarterial HCC treatment, and radiologists interpreting imaging after transarterial embolization of HCC should assess for peritoneal metastases.

Plenary III-02: Accuracy of Natural Language Processing to Identify Pneumonia from Electronic Radiology Reports

Background/Aims Pneumonia is common and can be devastating in older adults. Health plan data hold promise for studying pneumonia, but ICD-9 codes have poor accuracy for this condition. Natural language processing (NLP) offers potential to accurately and efficiently identify pneumonia from electronic medical records (EMRs). Our aims were to train one NLP tool to identify pneumonia from electronic radiology reports and to assess its validity compared to manual review.

Percutaneous Vertebroplasty to Treat Painful Myelomatous Vertebral Deposits: Long Term Efficacy Outcomes.

Abstract 1860Poster Board I-885 Introduction:Initially developed for the treatment of benign conditions such as osteoporotic crush fractures and vertebral haemangiomas, percutaneous vertebroplasty (PV) has since been shown to provide symptomatic and functional relief in patients with lesions related to neoplastic conditions including multiple myeloma (MM).1,2 The aim of this study is to present the experience of PV for a cohort of patients with MM treated at our tertiary referral centre over a 5 year period between 2004 and 2009. Patients and Methods:The electronic radiology reports for all patients treated with PV in the Imperial Healthcare NHS Trust were studied and those treated for lesions related to MM identified. Their medical records were then reviewed to extract baseline patient and disease characteristics as well as pain score and analgesic regimens pre and post the procedure. Any procedural related complications were also recorded. Following this, where possible, patients were contacted directly by telephone and asked to assess their benefit from the intervention. They were asked to describe their pain score, analgesia use and mobility post procedure and assessed for performance status based on the European Cooperative Oncology Group (ECOG) scale. Where patients reported a benefit they were asked whether that benefit had been sustained. Results:Of 132 procedures performed between 2004 and 2009, 26 patients were treated for painful lesions secondary to MM. Between 1 and 3 vertebral levels were treated, according to assessment of the patient and spinal MRI images by the consultant interventional radiologist at the pre-procedure workup appointment. Sixty-five percent of patients rated their pain as severe and 35% as moderate prior to the procedure. Seventy-seven percent reported a significant improvement with a reduction in their pain score. Seven patients (26%) reported some benefit on waking from the anesthetic, which then improved further over the following days to weeks. The median time of onset of relief post procedure was 3 days with a mean of 8.8 days. Fifty-eight percent of patients were able to reduce their systemic analgesia post procedure: 9 patients stopped all forms of opiate analgesia and the others were able to reduce their doses requirement by 30-75%. Fifty-eight percent of patients reported a significant improvement in their mobility and 5 had an improvement in their formal ECOG score - primarily due to being able to recommence work. Follow up times varied between 20 days and 42 months (mean 19 months) and those that experienced benefit reported that it was sustained for a mean of 13.5 months (range 1.5 – 35 months). The majority of patients whose pain returned had a concurrent progression of their myeloma, however 9 patients who were under ongoing follow up an average of 16 months from their procedure had not experienced any disease progression and reported sustained back pain relief up to the time of the study. Discussion and Conclusions:The results of this study serve to confirm and extend those reported in recent literature. Most studies have involved limited patient numbers (<15) with the majority of patients being treated at a single vertebral level.3–5 Only one study to date has focused on several vertebral level treatment in myeloma patients.2 In our study 50% of patients had 3 vertebrae treated simultaneously. Ongoing advances in MM therapies mean that patients can be expected to survive longer, resulting in a need for increased access to effective therapies that improve quality of life. Although the recent publication of the first two randomized trials of PV in osteoporotic fractures have cast doubt of its benefit over placebo6,7 all published observational data points to a positive benefit in myeloma. Clearly the proposed rationale that natural healing of the fracture over time may account for the non superiority results in these trials is not applicable to myeloma in which the pathology is one of progressive bony destruction. Our data support the consideration of PV as a first line treatment for painful, myeloma related spinal column disease, with a good safety profile and sustained subjective and objective measures of analgesic and functional benefit. Future large scale, prospective, randomized studies will be required to further define the role of this treatment for a patient population that is potentially expanding in line with advances in existing treatment strategies and emerging innovative therapies. Disclosures:No relevant conflicts of interest to declare.

Extraction of Recommendation Features in Radiology with Natural Language Processing: Exploratory Study

The purposes of this study were to validate a natural language processing program for extraction of recommendation features, such as recommended time frames and imaging technique, from electronic radiology reports and to assess patterns of recommendation features in a large database of radiology reports. This study was performed on a radiology reports database covering the years 1995-2004. From this database, 120 reports with and without recommendations were selected and randomized. Two radiologists independently classified these reports according to presence of recommendations, time frame, and imaging technique suggested for follow-up or repeated examinations. The natural language processing program then was used to classify the reports according to the same criteria used by the radiologists. The accuracy of classification of recommendation features was determined. The program then was used to determine the patterns of recommendation features for different patients and imaging features in the entire database of 4,211,503 reports. The natural language processing program had an accuracy of 93.2% (82/88) for identifying the imaging technique recommended by the radiologists for further evaluation. Categorization of recommended time frames in the reports with the 88 recommendations obtained with the program resulted in 83 (94.3%) accurate classifications and five (5.7%) inaccurate classifications. Recommendations of CT were most common (27.9%, 105,076 of 376,918 reports) followed by those for MRI (17.8%). In most (85.4%, 322,074/376,918) of the reports with imaging recommendations, however, radiologists did not specify the time frame. Accurate determination of recommended imaging techniques and time frames in a large database of radiology reports is possible with a natural language processing program. Most imaging recommendations are for high-cost but more accurate radiologic studies.

Use of Radcube for Extraction of Finding Trends in a Large Radiology Practice

The purpose of our study was to demonstrate the use of Natural Language Processing (Leximer), along with Online Analytic Processing, (NLP-OLAP), for extraction of finding trends in a large radiology practice. Prior studies have validated the Natural Language Processing (NLP) program, Leximer for classifying unstructured radiology reports based on the presence of positive radiology findings (F (POS)) and negative radiology findings (F (NEG)). The F (POS) included new relevant radiology findings and any change in status from prior imaging. Electronic radiology reports from 1995-2002 and data from analysis of these reports with NLP-Leximer were saved in a data warehouse and exported to a multidimensional structure called the Radcube. Various relational queries on the data in the Radcube were performed using OLAP technique. Thus, NLP-OLAP was applied to determine trends of F (POS) in different radiology exams for different patient and examination attributes. Pivot tables were exported from NLP-OLAP interface to Microsoft Excel for statistical analysis. Radcube allowed rapid and comprehensive analysis of F (POS) and F (NEG) trends in a large radiology report database. Trends of F (POS) were extracted for different patient attributes such as age groups, gender, clinical indications, diseases with ICD codes, patient types (inpatient, ambulatory), imaging characteristics such as imaging modalities, referring physicians, radiology subspecialties, and body regions. Data analysis showed substantial differences between F (POS) rates for different imaging modalities ranging from 23.1% (mammography, 49,163/212,906) to 85.8% (nuclear medicine, 93,852/109,374; p < 0.0001). In conclusion, NLP-OLAP can help in analysis of yield of different radiology exams from a large radiology report database.