Using AI to Identify Unremarkable Chest Radiographs for Automatic Reporting.

Background Radiology practices have a high volume of unremarkable chest radiographs and artificial intelligence (AI) could possibly improve workflow by providing an automatic report. Purpose To estimate the proportion of unremarkable chest radiographs, where AI can correctly exclude pathology (ie, specificity) without increasing diagnostic errors. Materials and Methods In this retrospective study, consecutive chest radiographs in unique adult patients (≥18 years of age) were obtained January 1-12, 2020, at four Danish hospitals. Exclusion criteria included insufficient radiology reports or AI output error. Two thoracic radiologists, who were blinded to AI output, labeled chest radiographs as "remarkable" or "unremarkable" based on predefined unremarkable findings (reference standard). Radiology reports were classified similarly. A commercial AI tool was adapted to output a chest radiograph "remarkableness" probability, which was used to calculate specificity at different AI sensitivities. Chest radiographs with missed findings by AI and/or the radiology report were graded by one thoracic radiologist as critical, clinically significant, or clinically insignificant. Paired proportions were compared using the McNemar test. Results A total of 1961 patients were included (median age, 72 years [IQR, 58-81 years]; 993 female), with one chest radiograph per patient. The reference standard labeled 1231 of 1961 chest radiographs (62.8%) as remarkable and 730 of 1961 (37.2%) as unremarkable. At 99.9%, 99.0%, and 98.0% sensitivity, the AI had a specificity of 24.5% (179 of 730 radiographs [95% CI: 21, 28]), 47.1% (344 of 730 radiographs [95% CI: 43, 51]), and 52.7% (385 of 730 radiographs [95% CI: 49, 56]), respectively. With the AI fixed to have a similar sensitivity as radiology reports (87.2%), the missed findings of AI and reports had 2.2% (27 of 1231 radiographs) and 1.1% (14 of 1231 radiographs) classified as critical (P = .01), 4.1% (51 of 1231 radiographs) and 3.6% (44 of 1231 radiographs) classified as clinically significant (P = .46), and 6.5% (80 of 1231) and 8.1% (100 of 1231) classified as clinically insignificant (P = .11), respectively. At sensitivities greater than or equal to 95.4%, the AI tool exhibited less than or equal to 1.1% critical misses. Conclusion A commercial AI tool used off-label could correctly exclude pathology in 24.5%-52.7% of all unremarkable chest radiographs at greater than or equal to 98% sensitivity. The AI had equal or lower rates of critical misses than radiology reports at sensitivities greater than or equal to 95.4%. These results should be confirmed in a prospective study. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Yoon and Hwang in this issue.

Hemodialysis and biomarkers of myocardial infarction - a cohort study.

OBJECTIVES: End-stage renal disease is associated with a high risk of cardiovascular disease. We compared the concentration and prognostic ability of high sensitivity cardiac troponin T (hs-cTnT) and I (hs-cTnI) and cardiac myosin-binding protein C (cMyC) among stable hemodialysis patients. METHODS: Patients were sampled before and after hemodialysis. We measured hs-cTnI, hs-cTnT and cMyC and used Cox regressions to assess the association between quartiles of concentrations and all-cause mortality and a combination of cardiovascular events and all-cause mortality during follow-up. RESULTS: A total of 307 patients were included, 204 males, mean age 66 years (SD 14). Before dialysis, 299 (99 %) had a hs-cTnT concentration above the 99th percentile, compared to 188 (66 %) for cMyC and 35 (11 %) for hs-cTnI. Hs-cTnT (23 %, p<0.001) and hs-cTnI (15 %, p=0.049) but not cMyC (4 %, p=0.256) decreased during dialysis. Follow-up was a median of 924 days (492-957 days); patients in the 3rd and 4th quartiles of hs-cTnT (3rd:HR 3.0, 95 % CI 1.5-5.8, 4th:5.2, 2.7-9.8) and the 4th quartile of hs-cTnI (HR 3.8, 2.2-6.8) had an increased risk of mortality. Both were associated with an increased risk of the combined endpoint for patients in the 3rd and 4th quartiles. cMyC concentrations were not associated with risk of mortality or cardiovascular event. CONCLUSIONS: Hs-cTnT was above the 99th percentile in almost all patients. This was less frequent for hs-cTnI and cMyC. High cTn levels were associated with a 3-5-fold higher mortality. This association was not present for cMyC. These findings are important for management of hemodialysis patients.

Hemodialysis leads to plasma depletion of lectin complement pathway initiator molecule ficolin-2.

INTRODUCTION: This study aimed to investigate changes in complement system-related molecules in patients undergoing hemodialysis. METHODS: Patients >18 years of age on maintenance hemodialysis were included. Using enzyme-linked immunosorbent assays (ELISA) methods complement related molecules ficolin-1, ficolin-2, ficolin-3 mannose-binding lectin, long pentraxin 3, complement activation products C3c, and complement activation potentials were measured before and after a single hemodialysis treatment. All patients were dialyzed with synthetic high flux filters >1.6 m2 , respectively, Polyamix and Polysulfone, and the Kt/V was maintained >1.3. FINDINGS: Three hundred and four patients were included. There was a modest decrease in plasma level of ficolin-1 (p < 0.001). Ficolin-2 was virtually depleted with median 3.9 (interquartile range [IQR]: 2.6-6.1, range 0.3-13.5) µg/ml before dialysis to median 0.0 (IQR: 0.0-0.5, range 0.0-5.5) µg/ml after dialysis (p < 0.001). No significant difference before and after hemodialysis was seen for mannose-binding lectin and long pentraxin 3 (p > 0.05). In a random subgroup of 160 patients ficolin-2-binding, ficolin-3-mediated lectin pathway capacity and classical pathway capacity were significantly decreased due to hemodialysis. The complement capacity of the alternative pathway was increased after hemodialysis (p = 0.0101), while mannose-binding lectin-mediated lectin pathway capacity was unaltered (p = 0.79). There was an increase in the complement activation product C3c (p < 0.0001), while the concentration of total C4 and C3 did not change (p > 0.158). Multivariate Cox proportional hazard analyses showed an increased risk for all-cause mortality with increasing ficolin-2 (p = 0.002) after hemodialysis. DISCUSSION: Plasma ficolin-2 was virtually depleted from the circulation after hemodialysis. However, elevated plasma ficolin-2 levels after hemodialysis was independently associated with increased mortality.

Mid-regional pro-atrial natriuretic peptide levels before and after hemodialysis predict long-term prognosis.

BACKGROUND: Mid-regional pro-atrial natriuretic peptide (MR-proANP) is a strong prognostic biomarker in cardiovascular disease but there is limited data for its use among patients undergoing dialysis. METHODS: This was a cohort study of patients receiving maintenance hemodialysis from two Danish centers. Blood sampling and echocardiography were performed before and after a dialysis session. We calculated the area under the curve (AUC) for the receiver operating characteristics for diagnosing heart failure and Cox regressions for cardiovascular events and all-cause mortality. RESULTS: Of the 306 patients, 284 (93%) had MR-proANP measurements both before and after dialysis. Median concentration was 642 pmol/L (IQR 419-858) before and 351 pmol/L (IQR 197-537) after dialysis, a mean decrease of 330 pmol/L (43%, CI 296-364, P < 0.001). MR-proANP concentration both before and after dialysis was negatively correlated to left ventricular ejection fraction with no difference in predictive ability for heart failure, AUC before and after dialysis were 0.60 (CI 0.50-0.70) and 0.61 (CI 0.51-0.71) (P = 0.40). Median follow-up was 32 months (IQR 31-33), during which 99 patients (32%) had a cardiovascular event and 110 (36%) died. A doubling of MR-proANP concentration was associated with a hazard ratio (HR) of 1.6 (CI 1.3-1.9) before and 1.7 (CI 1.4-2.0) after dialysis for mortality and a HR of 1.5 (CI 1.2-1.9) before and 1.4 (CI 1.2-1.7) after dialysis for cardiovascular events (all P < 0.001). CONCLUSION: The MR-proANP concentration is elevated among patients undergoing hemodialysis and decreases during dialysis. MR-proANP concentration both before, after and intra-dialysis change strongly predicted cardiovascular events and all-cause mortality.

[Diagnostic imaging findings in COVID-19].

This is a review of diagnostic imaging findings in coronavirus disease-2019 (COVID-19). CT findings are frequently bilateral, multilobar and peripheral ground-glass opacities with vascular enlargements. Consolidations often appear during progression as well as crazy paving and reticulation. Lymphadenopathy, pleural effusions and pneumothorax are rare and should raise concern for other disease. Initially the CT can be normal, and the diagnostic performance for COVID-19 is not optimal. However, findings of characteristic ground-glass opacities during the ongoing epidemic should lead to considerations, that the patient may have COVID-19.

Changes in left ventricular filling parameters before and after dialysis in patients with end stage renal disease.

The aim of this study was to investigate the grading of diastolic dysfunction (DD) in relation to hemodialysis in patients with end stage renal disease (ESRD) on hemodialysis (HD) Cardiovascular disease is prevalent in patients with ESRD and accounts for significant morbidity and mortality. Left ventricular hypertrophy (LVH) is common in ESRD but little is known about the impact of HD on currently recommended grading schemes for DD. Comprehensive echocardiographic data was obtained in consecutive patients with ESRD before (n = 247) and immediately after (n = 239) standard HD regimen. Grading of DD was performed according to current recommendations both pre- and post HD. Prior to HD, DD was classified as present in 83 patients (34%), indeterminate in 51 patients (21%) and absent in 113 patients (45%). Patients with DD at baseline compared to those without were older [67.3 years (13.1) vs. 63.2 (14.3), p = 0.037], were more likely to have diabetic- or hypertensive ESRD (43.4% vs. 35.4%, p = ns) and LVMi was significantly higher [119 g/cm2 (27.5) vs. 103 g/cm2 (24.3), p < 0.001]. After HD [mean HD time = 221 min (27.6), mean ultrafiltration volume = 2 L (1.1)], 39 patients (16%) exhibited sustained DD. These patients were older [69.4 years (14.5) vs. 65.0 years (13.9), p = 0.071], were more likely to have diabetic- or hypertensive ESRD (59% vs. 36%, p = 0.010). Myocardial adverse remodeling was more advanced with higher LVMi [127.4 g/m2 (27.5) vs. 106.5 g/m2 (25.3), p < 0.001], lower LVEF [44.7% (11.0) vs. 54.5% (8.7), p < 0.001] and more impaired GLS [- 13.4% (4.3) vs. - 15.8% (4.0), p = 0.006]. Echocardiographic evaluation of diastolic function in patients with ESRD on HD is critically dependent on timing relative to dialysis. The presence of sustained DD after volume unloading by HD identifies a population of patients with an adverse phenotype of blunted vascular response and severe cardiac remodeling.

FGF23 in hemodialysis patients is associated with left ventricular hypertrophy and reduced ejection fraction.

BACKGROUND: Fibroblast growth factor 23 (FGF23) is known to cause left ventricular hypertrophy (LVH), but controversy exists concerning its effect in dialysis. This study evaluated associations between FGF23 levels, echocardiography and prognosis in patients on hemodialysis (HD). METHODS: Patients >18 years on chronic HD were included in this cross-sectional study. Plasma C-terminal FGF23 concentration was measured with ELISA and transthoracic echocardiography was performed, both before and after HD treatment. RESULTS: 239 haemodialysis (HD) patients were included in the study. The FGF23 was median 3560RU/ml (IQR 1447-9952). The mean left ventricular mass index (LVMI) was 110.2±26.7g/m2 and the left ventricular ejection fraction (LVEF) was 52.7±9.9%. Defined by LVMI, LVH was found in 110 patients (46%), of which 92 (84%) had hypertension (p<0.01). Patients with LVH had FGF23 levels of 5319 RU/ml (IQR 1858-12,859) and those without 2496 RU/ml (IQR 1141-7028) (p<0.01). FGF23 was significant positive correlated with LVMI (p<0.01), and negatively to LVEF (p<0.01). In a multivariate analysis, FGF23 was correlated with LVEF (p<0.01), but only marginally to LVMI (p<0.01). Cardiovascular events in the follow up period was not correlated with FGF23. Furthermore, FGF23 was independently correlated with overall mortality (p<0.001). CONCLUSION: FGF23 was positively correlated with LVH and negatively to LVEF. FGF23 was an independent predictor for overall mortality.

The Copenhagen Triage Algorithm is non-inferior to a traditional triage algorithm: A cluster-randomized study.

INTRODUCTION: Triage systems with limited room for clinical judgment are used by emergency departments (EDs) worldwide. The Copenhagen Triage Algorithm (CTA) is a simplified triage system with a clinical assessment. METHODS: The trial was a non-inferiority, two-center cluster-randomized crossover study where CTA was compared to a local adaptation of Adaptive Process Triage (ADAPT). CTA involves initial categorization based on vital signs with a final modification based on clinical assessment by an ED nurse. We used 30-day mortality with a non-inferiority margin at 0.5%. Predictive performance was compared using Receiver Operator Characteristics. RESULTS: We included 45,347 patient visits, 23,158 (51%) and 22,189 (49%) were triaged with CTA and ADAPT respectively with a 30-day mortality of 3.42% and 3.43% (P = 0.996) a difference of 0.01% (95% CI: -0.34 to 0.33), which met the non-inferiority criteria. Mortality at 48 hours was 0.62% vs. 0.71%, (P = 0.26) and 6.38% vs. 6.61%, (P = 0.32) at 90 days for CTA and ADAPT. CTA triaged at significantly lower urgency level (P<0.001) and was superior in predicting 30-day mortality, Area under the curve: 0.67 (95% CI 0.65-0.69) compared to 0.64 for ADAPT (95% CI 0.62-0.66) (P = 0.03). There were no significant differences in rate of admission to the intensive care unit, length of stay, waiting time nor rate of readmission within 30 or 90 days. CONCLUSION: A novel triage system based on vital signs and a clinical assessment by an ED nurse was non-inferior to a traditional triage algorithm by short term mortality, and superior in predicting 30-day mortality. TRIAL REGISTRATION: Clinicaltrials.gov NCT02698319.

YKL-40 in patients with end-stage renal disease receiving haemodialysis.

PURPOSE: This study aimed to determine serum YKL-40 in patients with end-stage renal disease (ESRD) on haemodialysis (HD) and to evaluate the prognostic value of serum YKL-40. METHODS: Patients >18 years on maintenance HD were included. Serum YKL-40 was measured using ELISA before and after a single HD treatment. RESULTS: A total of 306 patients were included. Median serum YKL-40 concentration was 238 µgL-1 (IQR: 193-291 µgL-1) before HD treatment and 198 µgL-1 (IQR: 147-258 µgL-1) after HD treatment, which corresponded to age-corrected 93th percentile in healthy subjects. All-cause mortality after 2.8 years was 35.9%. Patients with serum YKL-40 in the highest quartile compared with the lowest quartile had a univariate HR of 4.0 (95% CI: 2.2-7.3, p < 0.001) for all-cause mortality which decreased to 2.4 (95% CI: 1.1-4.5, p = 0.01) in multivariate analysis. Time-dependent receiver operating characteristic curves showed that serum YKL-40 after HD treatment had significant higher area under the curves from 90 d (p = 0.004) and throughout the rest of the follow-up period when compared to serum YKL-40 before HD treatment. CONCLUSION: YKL-40 was highly elevated in patients with ESRD on HD, and dialysis reduced serum YKL-40 concentrations approximately one-sixth. YKL-40 measured after dialysis was independently associated with mortality in HD patients.

The prognostic significance of lung function in stable heart failure outpatients.

BACKGROUND: This study investigated the impact on all-cause mortality of airflow limitation indicative of chronic obstructive pulmonary disease or restrictive spirometry pattern (RSP) in a stable systolic heart failure population. HYPOTHESIS: Decreased lung function indicates poor survival in heart failure. METHODS: Inclusion criteria: NYHA class II-IV and left ventricular ejection fraction (LVEF) < 45%. Prognosis was assessed with multivariate Cox proportional hazards models. Two criteria of obstructive airflow limitation were applied: FEV1 /FVC < 0.7 (GOLD), and FEV1 /FVC < lower limit of normality (LLN). RSP was defined as FEV1 /FVC > 0.7 and FVC<80% or FEV1 /FVC > LLN and FVC

Prevalence of airflow obstruction in patients with stable systolic heart failure.

BACKGROUND: Chronic obstructive pulmonary disease (COPD) is an important differential diagnosis in heart failure (HF). However, routine use of spirometry in outpatient HF clinics is not implemented. The aim of the present study was to determine the prevalence of both airflow obstruction and non obstructive lung function impairment in patients with HF and to examine the effect of optimal medical treatment for HF on lung function parameters. METHODS: Consecutive patients with HF (ejection fraction (EF) < 45%) and New York Heart Association (NYHA) functional class II-IV at 10 different outpatient heart failure clinics were examined with spirometry at their first visit and after optimal medical treatment for HF was achieved. airflow obstruction was classified and graded according to the GOLD 2011 revision. RESULTS: Baseline spirometry was performed in 593 included patients and 71 (12%) had a clinical diagnosis of COPD. Mean age was 69 ± 11 years and mean EF was 30 ± 9%. Thirty-two % of the patients were active smokers and 53% were previous smokers. Mean FEV1 and FVC was 77.9 ± 1.7% and 85.4 ± 1.5% of predicted respectively. Obstructive pattern was observed in 233 (39%) of the patients. Of these, 53 patients (9%) had mild disease (GOLD I) and 180 (30%) patients had moderate to very severe disease (GOLD II-IV). No difference in spirometric variables was observed following up titration of medication. CONCLUSION: In stable patients with HF airflow obstruction is frequent and severely underdiagnosed. Spirometry should be considered in all patients with HF in order to improve diagnosis and treatment for concomitant pulmonary disease.

The Copenhagen Triage Algorithm: a randomized controlled trial.

BACKGROUND: Crowding in the emergency department (ED) is a well-known problem resulting in an increased risk of adverse outcomes. Effective triage might counteract this problem by identifying the sickest patients and ensuring early treatment. In the last two decades, systematic triage has become the standard in ED's worldwide. However, triage models are also time consuming, supported by limited evidence and could potentially be of more harm than benefit. The aim of this study is to develop a quicker triage model using data from a large cohort of unselected ED patients and evaluate if this new model is non-inferior to an existing triage model in a prospective randomized trial. METHODS: The Copenhagen Triage Algorithm (CTA) study is a prospective two-center, cluster-randomized, cross-over, non-inferiority trial comparing CTA to the Danish Emergency Process Triage (DEPT). We include patients ≥16 years (n = 50.000) admitted to the ED in two large acute hospitals. Centers are randomly assigned to perform either CTA or DEPT triage first and then use the other triage model in the last time period. The CTA stratifies patients into 5 acuity levels in two steps. First, a scoring chart based on vital values is used to classify patients in an immediate category. Second, a clinical assessment by the ED nurse can alter the result suggested by the score up to two categories up or one down. The primary end-point is 30-day mortality and secondary end-points are length of stay, time to treatment, admission to intensive care unit, and readmission within 30 days. DISCUSSION: If proven non-inferior to standard DEPT triage, CTA will be a faster and simpler triage model that is still able to detect the critically ill. Simplifying triage will lessen the burden for the ED staff and possibly allow faster treatment. TRIAL REGISTRATION: Clinicaltrials.gov: NCT02698319 , registered 24. of February 2016, retrospectively registered.

The formation and design of the TRIAGE study--baseline data on 6005 consecutive patients admitted to hospital from the emergency department.

BACKGROUND: Patient crowding in emergency departments (ED) is a common challenge and associated with worsened outcome for the patients. Previous studies on biomarkers in the ED setting has focused on identification of high risk patients, and and the ability to use biomarkers to identify low-risk patients has only been sparsely examined. The broader aims of the TRIAGE study are to develop methods to identify low-risk patients appropriate for early ED discharge by combining information from a wide range of new inflammatory biomarkers and vital signs, the present baseline article aims to describe the formation of the TRIAGE database and characteristize the included patients. METHODS: We included consecutive patients ≥ 17 years admitted to hospital after triage staging in the ED. Blood samples for a biobank were collected and plasma stored in a freezer (-80 °C). Triage was done by a trained nurse using the Danish Emergency Proces Triage (DEPT) which categorizes patients as green (not urgent), yellow (urgent), orange (emergent) or red (rescusitation). Presenting complaints, admission diagnoses, comorbidities, length of stay, and 'events' during admission (any of 20 predefined definitive treatments that necessitates in-hospital care), vital signs and routine laboratory tests taken in the ED were aslo included in the database. RESULTS: Between September 5(th) 2013 and December 6(th) 2013, 6005 patients were included in the database and the biobank (94.1 % of all admissions). Of these, 1978 (32.9 %) were categorized as green, 2386 (39.7 %) yellow, 1616 (26.9 %) orange and 25 (0.4 %) red. Median age was 62 years (IQR 46-76), 49.8 % were male and median length of stay was 1 day (IQR 0-4). No events were found in 2658 (44.2 %) and 158 (2.6 %) were admitted to intensive or intermediate-intensive care unit and 219 (3.6 %) died within 30 days. A higher triage acuity level was associated with numerous events, including acute surgery, endovascular intervention, i.v. treatment, cardiac arrest, stroke, admission to intensive care, hospital transfer, and mortality within 30 days (p < 0.001). CONCLUSION: The TRIAGE database has been completed and includes data and blood samples from 6005 unselected consecutive hospitalized patients. More than 40 % experienced no events and were therefore potentially unnecessary hospital admissions.