Assessing Strength of Evidence of Appropriate Use Criteria for Diagnostic Imaging Examinations

Machine Learning Model Drift: Predicting Diagnostic Imaging Follow-Up as a Case Example

OBJECTIVE

Address model drift in a machine learning (ML) model for predicting diagnostic imaging follow-up using data augmentation with more recent data versus retraining new predictive models.

METHODS

This institutional review board-approved retrospective study was conducted January 1, 2016, to December 31, 2020, at a large academic institution. A previously trained ML model was trained on 1,000 radiology reports from 2016 (old data). An additional 1,385 randomly selected reports from 2019 to 2020 (new data) were annotated for follow-up recommendations and randomly divided into two sets: training (n = 900) and testing (n = 485). Support vector machine and random forest (RF) algorithms were constructed and trained using 900 new data reports plus old data (augmented data, new models) and using only new data (new data, new models). The 2016 baseline model was used as comparator as is and trained with augmented data. Recall was compared with baseline using McNemar's test.

RESULTS

Follow-up recommendations were contained in 11.3% of reports (157 or 1,385). The baseline model retrained with new data had precision = 0.83 and recall = 0.54; none significantly different from baseline. A new RF model trained with augmented data had significantly better recall versus the baseline model (0.80 versus 0.66, P = .04) and comparable precision (0.90 versus 0.86).

DISCUSSION

ML methods for monitoring follow-up recommendations in radiology reports suffer model drift over time. A newly developed RF model achieved better recall with comparable precision versus simply retraining a previously trained original model with augmented data. Thus, regularly assessing and updating these models is necessary using more recent historical data.

Representing narrative evidence as clinical evidence logic statements

Objective

Clinical evidence logic statements (CELS) are shareable knowledge artifacts in a semistructured “If-Then” format that can be used for clinical decision support systems. This project aimed to assess factors facilitating CELS representation.

Materials and Methods

We described CELS representation of clinical evidence. We assessed factors that facilitate representation, including authoring instruction, evidence structure, and educational level of CELS authors. Five researchers were tasked with representing CELS from published evidence. Represented CELS were compared with the formal representation. After an authoring instruction intervention, the same researchers were asked to represent the same CELS and accuracy was compared with that preintervention using McNemar’s test. Moreover, CELS representation accuracy was compared between evidence that is structured versus semistructured, and between CELS authored by specialty-trained versus nonspecialty-trained researchers, using χ2 analysis.

Results

261 CELS were represented from 10 different pieces of published evidence by the researchers pre- and postintervention. CELS representation accuracy significantly increased post-intervention, from 20/261 (8%) to 63/261 (24%, P value < .00001). More CELS were assigned for representation with 379 total CELS subsequently included in the analysis (278 structured and 101 semistructured) postintervention. Representing CELS from structured evidence was associated with significantly higher CELS representation accuracy (P = .002), as well as CELS representation by specialty-trained authors (P = .0004).

Discussion

CELS represented from structured evidence had a higher representation accuracy compared with semistructured evidence. Similarly, specialty-trained authors had higher accuracy when representing structured evidence.

Conclusion

Authoring instructions significantly improved CELS representation with a 3-fold increase in accuracy. However, CELS representation remains a challenging task.

Can emergency department provider notes help to achieve more dynamic clinical decision support?

OBJECTIVE

Assess whether clinical data were present in emergency department (ED) provider notes at time of order entry for cervical spine (c-spine) imaging that could be used to augment or pre-populate clinical decision support (CDS) attributes.

METHODS

This Institutional Review Board-approved retrospective study, performed in a quaternary hospital, included all encounters for adult ED patients seen April 1, 2013-September 30, 2014 for a chief complaint of trauma who received c-spine computed tomography (CT) or x-ray. We assessed proportion of ED encounters with at least 1 c-spine-specific CDS rule attribute in clinical notes available at the time of imaging order and agreement between attributes in clinical notes and data entered into CDS.

RESULTS

A portion of the clinical note was submitted before imaging order in 42% (184/438) of encounters reviewed; 59.2% (109/184) of encounters with note portions submitted before imaging order had at least 1 positive CDS attribute identified supporting imaging study appropriateness; 34.8% (64/184) identified exclusion criteria where CDS appropriateness recommendations would not be applicable. 65.8% (121/184) of encounters had either a positive CDS attribute or an exclusion criterion. Concordance of c-spine CDS attributes when present in both notes and CDS was 68.4% (κ = 0.35 95% CI: 0.15-0.56; McNemar P = 0.23).

CONCLUSIONS

Clinical notes are an underutilized source of clinical attributes needed for CDS, available in a substantial percentage of encounters at the time of imaging order. Automated pre-population of imaging order requisitions with relevant clinical information extracted from electronic health record provider notes may: (1) improve ordering efficiency by reducing redundant data entry, (2) help improve clinical relevance of CDS alerts, and (3) potentially reduce provider burnout from extraneous alerts.

Effectiveness of the clinical decision support tool ESR eGUIDE for teaching medical students the appropriate selection of imaging tests: randomized cross-over evaluation

OBJECTIVES

To evaluate ESR eGUIDE-the European Society of Radiology (ESR) e-Learning tool for appropriate use of diagnostic imaging modalities-for learning purposes in different clinical scenarios.

METHODS

This anonymized evaluation was performed after approval of ESR Education on Demand leadership. Forty clinical scenarios were developed in which at least one imaging modality was clinically most appropriate, and the scenarios were divided into sets 1 and 2. These sets were provided to medical students randomly assigned to group A or B to select the most appropriate imaging test for each scenario. Statistical comparisons were made within and across groups.

RESULTS

Overall, 40 medical students participated, and 31 medical students (78%) answered both sets. The number of correctly chosen imaging methods per set in these 31 paired samples was significantly higher when answered with versus without use of ESR eGUIDE (13.7 ± 2.6 questions vs. 12.1 ± 3.2, p = 0.012). Among the students in group A, who first answered set 1 without ESR eGUIDE (11.1 ± 3.2), there was significant improvement when set 2 was answered with ESR eGUIDE (14.3 ± 2.5, p = 0.013). The number of correct answers in group B did not drop when set 2 was answered without ESR eGUIDE (12.4 ± 2.6) after having answered set 1 first with ESR eGUIDE (13.0 ± 2.7, p = 0.66).

CONCLUSION

The clinical decision support tool ESR eGUIDE is suitable for training medical students in choosing the best radiological imaging modality in typical scenarios, and its use in teaching radiology can thus be recommended.

KEY POINTS

• ESR eGUIDE improved the number of appropriately selected imaging modalities among medical students. • This improvement was also seen in the group of students which first selected imaging tests without ESR eGUIDE. • In the student group which used ESR eGUIDE first, appropriate selection remained stable even without the teaching tool.

Characteristics of knowledge content in a curated online evidence library

Objective

To describe types of recommendations represented in a curated online evidence library, report on the quality of evidence-based recommendations pertaining to diagnostic imaging exams, and assess underlying knowledge representation.

Materials and Methods

The evidence library is populated with clinical decision rules, professional society guidelines, and locally developed best practice guidelines. Individual recommendations were graded based on a standard methodology and compared using chi-square test. Strength of evidence ranged from grade 1 (systematic review) through grade 5 (recommendations based on expert opinion). Finally, variations in the underlying representation of these recommendations were identified.

Results

The library contains 546 individual imaging-related recommendations. Only 15% (16/106) of recommendations from clinical decision rules were grade 5 vs 83% (526/636) from professional society practice guidelines and local best practice guidelines that cited grade 5 studies (P < .0001). Minor head trauma, pulmonary embolism, and appendicitis were topic areas supported by the highest quality of evidence. Three main variations in underlying representations of recommendations were "single-decision," "branching," and "score-based."

Discussion

Most recommendations were grade 5, largely because studies to test and validate many recommendations were absent. Recommendation types vary in amount and complexity and, accordingly, the structure and syntax of statements they generate. However, they can be represented in single-decision, branching, and score-based representations.

Conclusion

In a curated evidence library with graded imaging-based recommendations, evidence quality varied widely, with decision rules providing the highest-quality recommendations. The library may be helpful in highlighting evidence gaps, comparing recommendations from varied sources on similar clinical topics, and prioritizing imaging recommendations to inform clinical decision support implementation.

Discordance in Clinical Recommendations Regarding the Use of Imaging

As physicians strive to provide evidence-based care, challenges arise if different entities disseminate divergent Appropriate Use Criteria (AUC) or clinical guidelines on the same topic. To characterize these challenges in one field, this study reviews the literature on comparisons of clinical recommendations regarding medical imaging. The PubMed database was searched for the years 2013-2018 for studies describing discordance among clinical recommendations regarding the performance of imaging. Of the 406 articles identified, 15 met the selection criteria: 8 qualitative and 7 quantitative. Reasons for discordance varied, with lack of evidence often cited. Quantitative studies often found that different decisions would be reached depending on the clinical recommendation followed. Nonetheless, quantitative studies also tended not to consider one set of recommendations superior to another. The findings of this review might help clinicians seek guidance more thoughtfully and could inform use of guidelines and AUC for quality improvement and clinical decision support.

Fast Healthcare Interoperability Resources, Clinical Quality Language, and Systematized Nomenclature of Medicine-Clinical Terms in Representing Clinical Evidence Logic Statements for the Use of Imaging Procedures: Descriptive Study

Background

Evidence-based guidelines and recommendations can be transformed into “If-Then” Clinical Evidence Logic Statements (CELS). Imaging-related CELS were represented in standardized formats in the Harvard Medical School Library of Evidence (HLE).

Objective

We aimed to (1) describe the representation of CELS using established Systematized Nomenclature of Medicine—Clinical Terms (SNOMED CT), Clinical Quality Language (CQL), and Fast Healthcare Interoperability Resources (FHIR) standards and (2) assess the limitations of using these standards to represent imaging-related CELS.

Methods

This study was exempt from review by the Institutional Review Board as it involved no human subjects. Imaging-related clinical recommendations were extracted from evidence sources and translated into CELS. The clinical terminologies of CELS were represented using SNOMED CT and the condition-action logic was represented in CQL and FHIR. Numbers of fully and partially represented CELS were tallied.

Results

A total of 765 CELS were represented in the HLE as of December 2018. We were able to fully represent 137 of 765 (17.9%) CELS using SNOMED CT, CQL, and FHIR. We were able to represent terms using SNOMED CT in the temporal component for action (“Then”) statements in CQL and FHIR in 755 of 765 (98.7%) CELS.

Conclusions

CELS were represented as shareable clinical decision support (CDS) knowledge artifacts using existing standards—SNOMED CT, FHIR, and CQL—to promote and accelerate adoption of evidence-based practice. Limitations to standardization persist, which could be minimized with an add-on set of standard terms and value sets and by adding time frames to the CQL framework.

Medicare Imaging Demonstration: Assessing Attributes of Appropriate Use Criteria and Their Influence on Ordering Behavior

OBJECTIVE

Persistent concern exists about the variable and possibly inappropriate utilization of high-cost imaging tests. The purpose of this study is to assess the influence of appropriate use criteria attributes on altering ambulatory imaging orders deemed inappropriate.

MATERIALS AND METHODS

This secondary analysis included Medicare Imaging Demonstration data collected from three health care systems in 2011-2013 via the use of clinical decision support (CDS) during ambulatory imaging order entry. The CDS system captured whether orders were inappropriate per the appropriate use criteria of professional societies and provided advice during the intervention period. For orders deemed inappropriate, we assessed the impact of the availability of alternative test recommendations, conflicts with local best practices, and the strength of evidence for appropriate use criteria on the primary outcome of cancellation or modification of inappropriate orders. Expert review determined conflicts with local best practices for 250 recommendations for abdominal and thoracic CT orders. Strength of evidence was assessed for the 15 most commonly triggered recommendations that were deemed inappropriate. A chi-square test was used for univariate analysis.

RESULTS

A total of 1691 of 63,222 imaging test orders (2.7%) were deemed inappropriate during the intervention period; this amount decreased from 364 of 11,675 test orders (3.1%) in the baseline period (p < 0.00001). Of 270 inappropriate recommendations with alternative test recommendations, 28 (10.4%) were modified, compared with four of 1024 inappropriate recommendations without alternatives (0.4%) (p < 0.0001). Seventy-eight of 250 recommendations (31%) conflicted with local best practices, but only six of 69 inappropriate recommendations (9%) conflicted (p < 0.001). No inappropriate recommendations that conflicted with local best practices were modified. All 15 commonly triggered recommendations had an Oxford Centre for Evidence-Based Medicine level of evidence of 5 (i.e., expert opinion).

CONCLUSION

Orders for imaging tests that were deemed inappropriate were modified infrequently, more often with alternative recommendations present and only for appropriate use criteria consistent with local best practices.

Evaluating Terminologies to Enable Imaging-Related Decision Rule Sharing

Purpose: Clinical decision support tools provide recommendations based on decision rules. A fundamental challenge regarding decision rule-sharing involves inadequate expression using standard terminology. We aimed to evaluate the coverage of three standard terminologies for mapping imaging-related decision rules. Methods: 50 decision rules, randomly selected from an existing library, were mapped to Systemized Nomenclature of Medicine (SNOMED CT), Radiology Lexicon (RadLex) and International Classification of Disease (ICD-10-CM). Decision rule attributes and values were mapped to unique concepts, obtaining the best possible coverage with the fewest concepts. Manual and automated mapping using Clinical Text Analysis and Knowledge Extraction System (cTAKES) were performed. Results: Using manual mapping, SNOMED CT provided the greatest concept coverage (83%), compared to RadLex (36%) and ICD-10-CM (8%) (p<0.0001). Combined mapping had 86% concept coverage. Automated mapping achieved 85% mapping coverage vs. 94% with manual mapping (p<0.001). Conclusion: Although some gaps remain, standard terminologies provide ample coverage for mapping imaging- related evidence.