Systematic evaluation of common natural language processing techniques to codify clinical notes.

Proper codification of medical diagnoses and procedures is essential for optimized health care management, quality improvement, research, and reimbursement tasks within large healthcare systems. Assignment of diagnostic or procedure codes is a tedious manual process, often prone to human error. Natural Language Processing (NLP) has been suggested to facilitate this manual codification process. Yet, little is known on best practices to utilize NLP for such applications. With Large Language Models (LLMs) becoming more ubiquitous in daily life, it is critical to remember, not every task requires that level of resource and effort. Here we comprehensively assessed the performance of common NLP techniques to predict current procedural terminology (CPT) from operative notes. CPT codes are commonly used to track surgical procedures and interventions and are the primary means for reimbursement. Our analysis of 100 most common musculoskeletal CPT codes suggest that traditional approaches can outperform more resource intensive approaches like BERT significantly (P-value = 4.4e-17) with average AUROC of 0.96 and accuracy of 0.97, in addition to providing interpretability which can be very helpful and even crucial in the clinical domain. We also proposed a complexity measure to quantify the complexity of a classification task and how this measure could influence the effect of dataset size on model's performance. Finally, we provide preliminary evidence that NLP can help minimize the codification error, including mislabeling due to human error.

Can Publicly Available Artificial Intelligence Successfully Identify Current Procedural Terminology Codes for Common Procedures in Neurosurgery?

OBJECTIVE: Coding for neurosurgical procedures is a complex process that is dynamically changing year to year, through the annual introduction and removal of codes and modifiers. The authors hoped to elucidate if publicly available artificial intelligence (AI) could offer solutions for neurosurgeons with regard to coding. METHODS: Multiple publicly available AI platforms were asked to provide Current Procedural Terminology (CPT) codes and Revenue Value Units (RVU) values for common neurosurgical procedures of the brain and spine with a given indication for the procedure. The responses of platforms were recorded and compared to the currently valid CPT codes used for the procedure and the amount of RVUs that would be gained. RESULTS: Six platforms and Google were asked for the appropriate CPT codes for 10 endovascular, spinal, and cranial procedures each. The highest performing platforms were as follows: Perplexity.AI identified 70% of endovascular, BingAI identified 55% of spinal, and ChatGPT 4.0 with Bing identified 75% of cranial CPT codes. With regard to RVUs, the top performer gained 78% of endovascular, 42% of spinal, and 70% of cranial possible RVUs. With regard to accuracy, AI platforms on average outperformed Google (45% vs. 25%, P = 0.04236). CONCLUSIONS: The ability of publicly available AIs to successfully code for neurosurgical procedures holds great promise in the future. Future development of AI should focus on improving accuracy with regard to CPT codes and providing supporting documentation for its decisions. Improvement on the existing capabilities of AI platforms can allow for increased operational efficiency and cost savings for practices.

Development and multicentre validation of the FLEX score: personalised preoperative surgical risk prediction using attention-based ICD-10 and Current Procedural Terminology set embeddings.

BACKGROUND: Preoperative knowledge of surgical risks can improve perioperative care and patient outcomes. However, assessments requiring clinician examination of patients or manual chart review can be too burdensome for routine use. METHODS: We conducted a multicentre retrospective study of 243 479 adult noncardiac surgical patients at four hospitals within the Mass General Brigham (MGB) system in the USA. We developed a machine learning method using routinely collected coding and patient characteristics data from the electronic health record which predicts 30-day mortality, 30-day readmission, discharge to long-term care, and hospital length of stay. RESULTS: Our method, the Flexible Surgical Set Embedding (FLEX) score, achieved state-of-the-art performance to identify comorbidities that significantly contribute to the risk of each adverse outcome. The contributions of comorbidities are weighted based on patient-specific context, yielding personalised risk predictions. Understanding the significant drivers of risk of adverse outcomes for each patient can inform clinicians of potential targets for intervention. CONCLUSIONS: FLEX utilises information from a wider range of medical diagnostic and procedural codes than previously possible and can adapt to different coding practices to accurately predict adverse postoperative outcomes.

Forecasting The Impact of the 2023 Current Procedural Terminology Coding Changes On Ventral Hernia Work Relative Value Units. A Cross-Sectional Study.

BACKGROUND: In January 2023, significant changes were implemented to ventral hernia repair Current Procedural Terminology codes, with new codes replacing previous codes. The new codes were assigned a 0-day global period. The impact of these changes on clinical productivity remains unclear. Our objective was to forecast the impact of Current Procedural Terminology changes on ventral hernia-related work relative value units using historical data. METHODS: Ventral hernia repairs performed between March 2021 and December 2022 on adults by a single surgeon with available 90-day follow-up were retrospectively retrieved from the Abdominal Core Health Quality Collaborative. Demographic, hernia, and operative and postoperative data were collected. The ventral hernia repairs were coded twice using the previous and new Current Procedural Terminology codes, and work relative value units were calculated using both systems. The median work relative value units per case were compared using the Wilcoxon signed-rank test. RESULTS: A total of 143 ventral hernia repairs were included. The median age was 59 years, and 50% of patients were male. Median hernia width and length were 3.5 and 5.0 cm, respectively. The most common ventral hernia types were incisional 57% and umbilical 33%. Twenty percent of hernias were recurrent, and 99% were elective repairs. 49% of the procedures were open, 30% robotic, and 21% laparoscopic. Component separation was performed in 16%. The median length of stay was 0.0, and the median number of 90-day outpatient postoperative visits was 1.0. The new Current Procedural Terminology coding system was associated with a higher median 90-day work relative value units per case (14.1) than the previous system (13.8) (P = .002). Subset analysis identified statistically higher median 90-day work relative value units per case using the new versus previous Current Procedural Terminology codes for hernias with the largest defect dimension >10 cm (23.3 vs 18.8), umbilical/epigastric/Spigelian hernias (9.2 vs 7.1), recurrent hernias (20.1 vs 17.3) and open ventral hernia repairs (9.8 vs 7.1), all P < .05. Median 90-day work relative value units per case were statistically lower using the new versus previous codes for non-recurrent (11.6 vs 13.8) and incarcerated/strangulated (14.8 vs 14.9) hernias, all P < .05. In the new coding system, postoperative care within 90-days contributed to a median of 1.3 work relative value units per case (9% of total 90-day work relative value units). CONCLUSION: We forecast that in our practice, the 2023 ventral hernia repair Current Procedural Terminology changes will result in a modest impact on clinical productivity. The impact of these changes on a particular practice depends on surgical practice patterns and ventral hernia case mix.

GRU-D-Weibull: A novel real-time individualized endpoint prediction.

BACKGROUND: In the era of healthcare digital transformation, using electronic health record (EHR) data to generate various endpoint estimates for active monitoring is highly desirable in chronic disease management. However, traditional predictive modeling strategies leveraging well-curated data sets can have limited real-world implementation potential due to various data quality issues in EHR data. METHODS: We propose a novel predictive modeling approach, GRU-D-Weibull, which models Weibull distribution leveraging gated recurrent units with decay (GRU-D), for real-time individualized endpoint prediction and population level risk management using EHR data. EXPERIMENTS: We systematically evaluated the performance and showcased the real-world implementability of the proposed approach through individual level endpoint prediction using a cohort of patients with chronic kidney disease stage 4 (CKD4). A total of 536 features including ICD/CPT codes, medications, lab tests, vital measurements, and demographics were retrieved for 6879 CKD4 patients. The performance metrics including C-index, L1-loss, Parkes' error, and predicted survival probability at time of event were compared between GRU-D-Weibull and other alternative approaches including accelerated failure time model (AFT), XGBoost based AFT (XGB(AFT)), random survival forest (RSF), and Nnet-survival. Both in-process and post-process calibrations were experimented on GRU-D-Weibull generated survival probabilities. RESULTS: GRU-D-Weibull demonstrated C-index of ~0.7 at index date, which increased to ~0.77 at 4.3 years of follow-up, comparable to that of RSF. GRU-D-Weibull achieved absolute L1-loss of ~1.1 years (sd≈0.95) at CKD4 index date, and a minimum of ~0.45 year (sd≈0.3) at 4 years of follow-up, comparing to second-ranked RSF of ~1.4 years (sd≈1.1) at index date and ~0.64 years (sd≈0.26) at 4 years. Both significantly outperform competing approaches. GRU-D-Weibull constrained predicted survival probability at time of event to smaller and more fixed range than competing models throughout follow-up. Significant correlations were observed between prediction error and missing proportions of all major categories of input features at index date (Corr ~0.1 to ~0.3), which faded away within 1 year after index date as more data became available. Through post training recalibration, we achieved a close alignment between the predicted and observed survival probabilities across multiple prediction horizons at different time points during follow-up. CONCLUSION: GRU-D-Weibull shows advantages over competing methods in handling missingness commonly encountered in EHR data and providing both probability and point estimates for diverse prediction horizons during follow-up. The experiment highlights the potential of GRU-D-Weibull as a suitable candidate for individualized endpoint risk management, utilizing real-time clinical data to generate various endpoint estimates for monitoring. Additional research is warranted to evaluate the influence of different data quality aspects on prediction performance. Furthermore, collaboration with clinicians is essential to explore the integration of this approach into clinical workflows and evaluate its effects on decision-making processes and patient outcomes.

Relative billing complexity of in-person versus telehealth outpatient encounters.

RATIONALE: Video visits became more widely available during the coronavirus disease (COVID-19) pandemic. However, the ongoing role and value of video visits in care delivery and how these may have changed over time are not well understood. AIMS AND OBJECTIVES: Compare the relative complexity of in-person versus video visits during the COVID-19 pandemic and describe the complexity of video visits over time. METHODS: We used billing data for in-person and video revisits from non-behavioural health specialities with the most video visit utilisation (≥50th percentile) at a large, urban, public healthcare system from 1 January 2021 to 31 March 2022. We used current procedural terminology (CPT) codes as a proxy for information gathering and decision-making complexity and time spent on an encounter. We compared the distribution of CPT codes 99211-99215 between in-person and video visits using Fisher's exact tests. We used Spearman correlation to test for trends between proportions of CPT codes over time for video visits. RESULTS: Ten specialities (adult primary care, paediatrics, adult dermatology, bariatric surgery, paediatric endocrinology, obstetrics and gynaecologist, adult haematology/oncology, paediatric allergy/immunology, paediatric gastroenterology, and paediatric pulmonology) met inclusion criteria. For each speciality, proportions of each CPT code for in-person visits and for video visits varied significantly, and patterns of variation differed by speciality. For example, in adult primary care, video visits had smaller proportions of moderate/high complexity visits (99214 and 99215) and greater proportions of lower complexity visits (99211-99213) compared with in-person visits (p < 0.001), but in paediatric endocrinology, the opposite was seen (p < 0.001). Trends in CPT codes over time for video visits in each speciality were also mixed. CONCLUSION: In-person and video visits had differing proportions of complexity codes (typically skewing towards lower complexity for video visits). The complexity of video visits changed over time in many specialities. Observed patterns for both phenomena varied by speciality.

Controversies in CPT® coding in the neonatal intensive care unit: - critical vs. intensive care.

Professional reimbursement to neonatal providers is based on the level of Current Procedural Terminology (CPT®) coding in the NICU, newborn nursery and other areas where neonatal care is provided. Four levels of evaluation and management (E&M) care-critical, intensive, routine-hospital care or normal newborn care can be provided to neonates. The work relative value units (wRVUs) associated with these four levels of care vary widely. This manuscript provides a brief review of basic features associated with each of these four levels with a specific perspective on differences between critical and intensive care codes. Coding and billing are constantly evolving fields with significant variation in interpretation and readers are encouraged to review the current publications on CPT® coding and make an informed decision on the best codes to be used for their patients.

Trends in the Management of Pediatric Trigger Thumb in the United States.

BACKGROUND: Recent literature suggests that initial observation of pediatric trigger thumb without early surgical interventions can lead to spontaneous resolution. We sought to analyze current trends in the management of pediatric trigger thumb and compare real-world data with what the literature supports. METHODS: We conducted a retrospective study of data collected using the PearlDiver database between 2015 and 2018. Patients who were aged younger than 10 years with a diagnosis of trigger thumb were identified using International Classification of Diseases codes. Current Procedural Terminology codes were used to identify patients who had an operation for trigger thumb. Patient demographics, comorbidities, utilization of hand therapy, and treatment cost were also collected. RESULT: Of the 997 patients included in the study, 69% were diagnosed with trigger thumb between the age of 2 and 5 years. In all, 492 patients (49%) had surgery for trigger thumb: 65% of patients had surgery within 1 year of diagnosis, and 76% patients had surgery before the age of 5 years. This treatment pattern was similar across multiple regions of the United States, and there were no significant predictors for surgery. The average cost of treating patients without surgery was $593/patient, whereas that for patients with surgery was $1363/patient. CONCLUSIONS: Nationwide data show that pediatric trigger thumb may be managed surgically at higher frequencies and in patients at younger ages than supported by the existing literature. Possible overtreatment is not only detrimental to patients but also burdens the health care system with unnecessary cost.

Comparison of Estimated No Surprises Act Qualifying Payment Amounts and Payments to In-Network and Out-of-Network Emergency Medicine Professionals.

Importance: The No Surprises Act (NSA), which took effect on January 1, 2022, applies a qualifying payment amount (QPA) as an out-of-network payment reference point. An understanding of how QPA measures compare with the in-network and out-of-network payments physicians received before the NSA implementation may be useful to policy makers and stakeholders. Objective: To estimate the QPA for geographic and funding markets and compare QPA estimates with in-network and out-of-network payments for 2019 emergency medicine claims. Design, Setting, and Participants: This cross-sectional study of US commercial insurance claims assessed the Health Care Cost Institute's 2019 commercial professional emergency medicine claims (Current Procedural Terminology [CPT] codes 99281-99285 and 99291) and included enrollees in commercial health maintenance organizations, exclusive provider organizations, point of service, and preferred provider organizations self-funded and fully insured through Aetna, Humana, and some Blue Health Intelligence plans. Claims with missing or inconsistent data fields were excluded. Data were analyzed November 1, 2021, to April 7, 2022. Main Outcomes and Measures: The QPA was calculated as the median allowed amount of all observed claims within strata defined by geographic region, CPT code, and funding market. For each stratum, the ratio of mean in-network allowed amounts to QPAs and mean out-of-network allowed amounts to QPAs were calculated. Then the volume-weighted mean of these ratios was computed across CPT codes within each geographic and funding market stratum. Results: The analytic sample included 7 556 541 professional emergency claims with a mean (SD) allowed amount of $313 ($306) and mean (SD) QPA of $252 ($133). Among the 650 geographic and market strata in the sample, the mean in-network allowed amounts were 14% (ratio, 0.96) higher than the estimated QPA. For the subset of strata with a sufficient sample of out-of-network claims (n = 227), the mean out-of-network payments were 112% (ratio, 2.12) higher than the QPA. More generous out-of-network payments were from self-funded plans (120% [ratio, 2.20] higher than the QPA estimate) vs fully insured plans (43% [ratio, 1.43] higher than the QPA estimate). Mean in-network allowed amounts for nonphysician clinicians were 4% (ratio, 1.04) lower than the QPA, whereas mean in-network allowed amounts for physicians were 15% (ratio, 1.15) higher than the QPA estimates. These differences remained after adjusting for geographic region. Conclusions and Relevance: The findings of this cross-sectional study of US commercial insurance claims suggest that the NSA may have heterogeneous implications for out-of-network payments and negotiating leverage experienced by emergency medicine physicians in different geographic markets, with the potential for greater implications in the self-funded market.

Recent Trends in Medicare Utilization and Reimbursement for Spinal Cord Stimulators: 2000-2019.

OBJECTIVE: Spinal cord stimulators (SCS) allow spine surgeons to provide relief for patients who suffer from chronic pain due to several disorders, such as failed back surgery syndrome, complex regional pain syndrome, and neuropathy. Despite this, there remains a paucity of data regarding the utilization and reimbursement of SCS. Therefore, the purpose of this study is to evaluate the monetary and procedural trends of spinal cord stimulators among the Medicare database from 2000 to 2019. METHODS: Medicare Part B National Summary Data files, which are publicly available, were used. These files contain data from the years 2000-2019 on all services billed to Medicare within that time frame. Each service is given a Current Procedural Terminology (CPT) code and the number of times that service was performed, as well as the total physician Medicare charges and reimbursements for each service annually are included in that data set. The CPT codes for percutaneous and open placement of spinal cord stimulators were identified: 63650 and 63655, respectively. The total allowed services allowed charges and actual payments were isolated from the data set for each year for each CPT code. The total allowed charges and actual payments for the year were then divided by the total allowed services to find and trend the allowed charges and actual payment for each individual service performed for both percutaneous and open placement of spinal cord stimulators. RESULTS: There were 992,372 Medicare-approved total percutaneous spinal cord stimulator operations and 99,736 Medicare-approved total open spinal cord stimulator operations from 2000 to 2019. Medicare paid $1.02 billion (2019 U.S. dollars) in reimbursement to physicians for percutaneous spinal cord stimulator operations and nearly $145 million (2019 U.S. dollars) in reimbursement to physicians for open spinal cord stimulator operations. From the years 2000 to 2019, there was an average 21.9% increase annually in Medicare-approved percutaneous spinal stimulator placement operations and a 18.4% increase annually in Medicare-approved open spinal stimulator placement operations. During this time, there was also an average 8.7% increase annually in Medicare reimbursement per each percutaneous spinal stimulator placement operation and a 9.1% increase annually in Medicare reimbursement per each open spinal stimulator placement operation. CONCLUSIONS: The results of this study show that the number of percutaneous and open procedures have steadily increased from 2000 to 2019. Reimbursement per procedure has also increased steadily over this time. Identifying these trends is important to promote research into costs of these surgeries and ensure adequate resource allocation.

New Evaluation and Management Code Level Selection Trends in Hip and Knee Osteoarthritis Patients.

BACKGROUND: On January 1, 2021, the American Medical Association implemented changes regarding the outpatient Evaluation and Management (E/M) criteria dictating Current Procedural Terminology code level selection to help diminish administrative burden and emphasize medical decision-making as the primary determinant in E/M level of service (EML). The goal of this study was to describe EML coding trends in outpatient visits for hip and knee osteoarthritis after the 2021 Centers for Medicare and Medicaid Services changes to the E/M system. METHODS: All outpatient visits for primary hip and knee osteoarthritis within the divisions of Joint Replacement, Operative Sports Medicine, and Nonoperative Sports Medicine at a single orthopaedic practice were retrospectively analyzed during 2 separate 10-month timeframes in 2019 and 2021. The primary endpoint was the visit EML (1 through 5) based on Current Procedural Terminology E/M codes. RESULTS: In 2019, 7.8% of all visits were billed as level 2, 85.8% of all visits were billed as level 3, and 6.3% of all visits were billed as level 4. In 2021, 2.8% of visits were billed as level 2, 54% of visits were billed as level 3, and 41.3% of visits were billed as level 4. Level 1 and Level 5 visits did not exceed 2% in either year. Across all 3 divisions, level 2 and 3 visits decreased significantly (P < .05), while level 4 visits increased significantly (P < .05). CONCLUSION: Since the E/M coding criteria overhaul in 2021, there has been a significant trend towards higher level of service code selection across multiple divisions in our orthopaedic practice.

Hospital and Professional Charges and Reimbursement Patterns in Pediatric Extracorporeal Membrane Oxygenation.

BACKGROUND: Extracorporeal membrane oxygenation (ECMO) is an expensive therapeutic modality. We sought to identify the main charge contributors to patient bills and analyze their patterns of reimbursement. We additionally sought to evaluate the impact of 2015 Current Procedural Terminology (CPT) code changes in professional billing for pediatric surgeons. METHODS: A retrospective review of ECMO cases at a standalone quaternary children's hospital between 2008-2017 was performed. Itemized hospital and professional bills were analyzed. RESULTS: Top charges included room rates, nitric oxide, medications, invasive support and monitoring, and laboratory testing. Average reimbursement was ∼60% for hospital and ∼36% for professional bills. CPT code changes in 2015 represented a 65% reduction in RVUs and 46% reduction in professional charges. Medicaid reimbursement for professional billing remained stable at 9%, and commercial reimbursement fell from 70% to 59% during the study period. CONCLUSIONS: The main drivers of ECMO charges are unrelated to ECMO supplies or surgery. Evidence-based guidelines for ECMO management could make a difference in healthcare expenditure. Modern CPT codes depreciate RVUs and professional charges, compromising revenue. As the infrastructure required to provide this service is costly, diminishing returns may limit access to this therapy.

Identification of pregnancies and infants within a US commercial healthcare administrative claims database.

PURPOSE: Health care insurance claims databases are becoming a more common data source for studies of medication safety during pregnancy. While pregnancies have historically been identified in such databases by pregnancy outcomes, International Classification of Diseases, 10th revision Clinical Modification (ICD-10-CM) Z3A codes denoting weeks of gestation provide more granular information on pregnancies and pregnancy periods (i.e., start and end dates). The purpose of this study was to develop a process that uses Z3A codes to identify pregnancies, pregnancy periods, and links infants within a commercial health insurance claims database. METHODS: We identified pregnancies, gestation periods, pregnancy outcomes, and linked infants within the US-based Optum Research Database between 2015 and 2020 via a series of algorithms utilizing diagnosis and procedure codes on claims. The diagnosis and procedure codes included ICD-10-CM codes, Current Procedural Terminology (CPT) codes, and Healthcare Common Procedure Coding System (HCPCS) codes. RESULTS: We identified 1 030 874 pregnancies among 841 196 women of reproductive age. Of pregnancies with livebirth outcomes, 84% were successfully linked to infants. The prevalence of pregnancy outcomes (livebirth, stillbirth, ectopic, molar, and abortion) was similar to national estimates. CONCLUSIONS: This process provides an opportunity to study drug safety and care patterns during pregnancy and may be replicated in other claims databases containing ICD-10-CM, CPT, and HCPCS codes. Work is underway to validate and refine the various algorithms.

Are work relative value units correlated with operative duration of common surgical procedures?

OBJECTIVES: Work relative value units (wRVUs) quantify physician workload. In theory, higher wRVU assignments for procedures recognize an increase in complexity and time required to complete the procedure. The fairness of wRVU assignment is debated across specialties, with some surgeons arguing that reimbursement may be unfairly low for longer, more complex cases. For this reason, we sought to assess the correlation of wRVUs with operative time in commonly performed surgeries. STUDY DESIGN: We analyzed the National Surgical Quality Improvement Program database, selecting the 15 most performed surgical procedures across specialties in a 90-day global period, using Current Procedural Terminology codes. METHODS: Calculation and comparison of mean operative time and mean wRVUs were performed for each of the 15 procedures. Cases with missing values for wRVUs or operative time and cases with an operative time of less than 15 minutes were excluded. The Spearman correlation coefficient was calculated to evaluate the strength of correlation between operative duration and wRVUs. RESULTS: A total of 1,994,394 patients met criteria for analysis. The lowest mean wRVU was 7.78 (95% CI, 7.77-7.78) for inguinal hernia repair; the highest was 43.50 (95% CI, 43.37-43.60) for pancreatectomy. The shortest mean operative time was 51.0 (95% CI, 50.8-51.1) minutes for appendectomy; the longest was for pancreatectomy at 324.6 (95% CI, 323.2-326.0) minutes. The Spearman correlation coefficient was 0.81. CONCLUSIONS: In our analysis, we found a strong correlation between operative duration and wRVU assignment. Thus, the reimbursement of physicians depending on wRVUs is fair for the most commonly performed surgical procedures across specialties.

Growth in Thoracentesis and Paracentesis Performed by Radiology and Advanced Practice Providers: Medicare Volume and Reimbursement Trends From 2012 to 2018.

PURPOSE: To study trends in volume and reimbursement for paracentesis and thoracentesis by physicians and advanced practice providers (APPs) after the introduction of discreet Current Procedural Terminology codes for image guidance. METHODS: Medicare claims for 2012 to 2018 (paracentesis) and 2013 to 2018 (thoracentesis) were extracted using Current Procedural Terminology codes for blind and image-guided paracentesis and thoracentesis. Total volumes were analyzed by provider specialty. Nonfacility reimbursement and relative value units were compared. RESULTS: For blind paracentesis, volume decreased from 17,393 to 12,226 procedures from 2012 to 2018. Conversely, volume of image-guided paracentesis increased from 171,631 to 253,834 procedures. Radiology performed the majority of image-guide paracentesis (83.9% in 2012 and 77.1% in 2018). Volume and relative share for APPs dramatically increased (from 10.2% to 15.8%). For blind thoracentesis, volume decreased from 26,716 to 15,075 procedures from 2013 to 2018. Conversely, volume of image-guided thoracentesis increased from 187,168 to 222,673 procedures. Radiology performed the majority of image-guided thoracentesis (73.6% in 2013 and 66.2% in 2018). Volume and relative share for APPs dramatically increased (from 7.7% to 12.9%). Although reimbursement for both image-guided paracentesis and thoracentesis decreased, their reimbursement remained higher than that of blind paracentesis and thoracentesis throughout the study period. CONCLUSION: A higher percentage of these procedures are being performed using image guidance; radiologists performed a growing number but declining percentage of image-guided paracentesis and thoracentesis. APPs are playing an increasing role, particularly using image guidance. Given decreasing reimbursement for these procedures, APPs can provide a large cost advantage in procedural radiology practices.

Consensus Current Procedural Terminology Code Definition of Source Control for Sepsis.

INTRODUCTION: Unlike antibiotic and perfusion support, guidelines for sepsis source control lack high-quality evidence and are ungraded. Internally valid administrative data methods are needed to identify cases representing source control procedures to evaluate outcomes. METHODS: Over five modified Delphi rounds, two independent reviewers identified Current Procedural Terminology (CPT) codes pertinent to source control. In each round, codes with perfect agreement were retained or excluded, whereas disagreements were reviewed by the panelists. Manual review of 400 patient records meeting Sepsis-3 criteria (2010-2017) clinically adjudicated which encounters included source control procedures (gold standard). The performance of consensus codes was compared with the gold standard to assess sensitivity, specificity, predictive values, and likelihood ratios. RESULTS: Of 5752 CPT codes, 609 consensus codes represented source control procedures. Of 400 hospitalizations for sepsis, 39 (9.8%; 95% confidence interval [CI] 7.0%-13.1%) underwent gold standard source control procedures and 29 (7.3%; 95% CI 4.9-10.3%) consensus code-defined source control procedures. Thirty consensus codes were identified (20.0% gastrointestinal/intraabdominal, 10.0% genitourinary, 13.3% hepatopancreatobiliary, 23.3% orthopedic/cranial, 23.3% soft tissue, and 10.0% intrathoracic), which had 61.5% (95% CI 44.6%-76.6%) sensitivity, 98.6% (95% CI 96.8%-99.6%) specificity, 83.2% (95% CI 66.6%-92.4%) positive, and 95.9% (95% CI 93.9%-97.2%) negative predictive values. With pretest probability at sample prevalence, an identified consensus code had a posttest probability of 83.0% (95% CI 66.0%-92.0%), whereas consensus code absence had a probability of 4.0% (95% CI 3.0-6.0) for undergoing a source control procedure. CONCLUSIONS: Using modified Delphi methodology, we created and validated CPT codes identifying source control procedures, providing a framework for evaluation of the surgical care of patients with sepsis.