Introduction

Autoimmune Atrophic Gastritis (AAG) is a chronic immune-mediated condition that mainly affects the body and fundus of the stomach. AAG is often under-diagnosed due to the slow progression of the disease and a variable spectrum of symptoms. Recent data suggests a potential role of altered gastric microbiota in the natural history of AAG, due to the potential overgrowth of bacteria linked to the development of neoplastic complications (neuroendocrine tumors- NETs and gastric cancer) (1,2). Moreover, the same data hypothesizes the use of fecal microbiota as a possible non-invasive marker to diagnose and monitor AAG. Thus, we aimed to prospectively characterize the salivary and fecal microbiota of patients with AAG to provide further insights in this field.

Aims & Methods

We collected salivary and fecal samples from consecutive patients with an established diagnosis of AAG, enrolled at the Gastroenterology Unit of Azienda Ospedale Università Padova from March to June 2023 at the time of endoscopic follow-up examination. All patients needed to be off Proton-Pump Inhibitors, antibiotics, and probiotics for at least one month before enrolment. Stool samples were collected by the patients on the day of the endoscopy, transferred in a buffer solution, and stored at -20°C. Oral samples were collected through salivary commercial kits and freezed at -20°C. Both salivary and faecal samples were analyzed by BMR Genomic (Padua, Italy) that amplified the V3-V4 hypervariable ribosomal RNA region. Sequencing of the 16S rRNA gene V3- V4 region was performed using the Illumina MiSeq. Preliminary analyses were conducted on individual fecal samples compared to an internal population of healthy subjects (provided by BMR genomics) using Qiime2 tools. Shannon–Wiener index was used to calculate α-diversity and then the microbial biodiversity index (MBI) and Dysbiosis Index (DI). Mann–Whitney tests were performed to verify statistical differences in the α-diversity across AAG patients and healthy subjects.

Results

We enrolled 15 AAG patients with a mean age of ± 63 ± 11 years (9/15 females). OLGA 2 stage was reported in 10/15 patients. Sixty percent of the patients showed a MBI between 7.11 and 12.45, indicating that each sample contained phylogenetically similar species. The remaining patients have an index greater than 12.45 compared to the healthy population. More than half of the patients had dysbiosis (>0.082). The highest indices of dysbiosis were found in patients with a history of NET (n=3), gastric cancer (n=1), and previous HP- infection and OLGA 3 stage (n=1). Firmicutes, Bacteroidetes, and Proteobacteria were more frequent phyla found in the fecal samples, whereas Enterobacteriaceae and Streptococcaceae were the most common bacterial families (13/15, 87%). In particular, Klebsiella and Campylobacter pathogens genera were found in some samples. Streptococcus and Prevotella were the most abundant genera found in the salivary samples.

Conclusion

Patients with AAG were found to have a great proportion of Firmicutes and Bacteroidetes, which aligns with the literature (3,4). Moreover, a dominant of Proteobacteria, especially Enterobacteriaceae commonly associated with proinflammatory states, was found. (5) Our preliminary data showed a possible positive association between higher degrees of dysbiosis and a previous history of NET or gastric cancer.

References

1. Parsons BN, Ijaz UZ, D'Amore R, Burkitt MD, Eccles R, Lenzi L, Duckworth CA, Moore AR, Tiszlavicz L, Varro A, Hall N, Pritchard DM. Comparison of the human gastric microbiota in hypochlorhydric states arising as a result of Helicobacter pylori-induced atrophic gastritis, autoimmune atrophic gastritis and proton pump inhibitor use. PLoS Pathog. 2017 Nov 2;13(11):e1006653.

2. Conti L, Annibale B, Lahner E. Autoimmune Gastritis and Gastric Microbiota. Microorganisms. 2020 Nov 19;8(11):1827.

3. Castaño-Rodríguez N, Goh KL, Fock KM, et al. Dysbiosis of the microbiome in gastric carcinogenesis. Sci Rep. 2017;7(1):15957.

4. Conti L, Borro M, Milani C, Simmaco M, Esposito G, Canali G, Pilozzi E, Ventura M, Annibale B, Lahner E. Gastric microbiota composition in patients with corpus atrophic gastritis. Dig Liver Dis. 2021 Dec;53(12):1580-1587.

5. Rizzatti G, Lopetuso LR, Gibiino G, Binda C, Gasbarrini A. Proteobacteria: A Common Factor in Human Diseases. Biomed Res Int. 2017;2017:9351507.

Disclosure

Nothing to disclose

Introduction

Fatigue, defined as “an overwhelming sense of tiredness, lack of energy and feeling of exhaustion” [1] is a frequent complaint in celiac disease. A strict, lifelong gluten free diet is the only established treatment for celiac disease, however its long-term effect on fatigue is unknown.

Aims & Methods

Our aims were to investigate the course of fatigue severity and prevalence in celiac patients from diagnosis to five to seven years on a gluten free diet, and to investigate whether diet adherence, disease activity and selected biochemical markers are associated with fatigue.
We conducted a single-center prospective study in 45 patients with newly diagnosed celiac disease. The primary endpoints were fatigue severity, as measured by the Fatigue Severity Scale (FSS) [2], the fatigue Visual Analog Scale (fVAS) [3], and the Vitality subscale of the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36vs) [4]. SF-36vs scores were inverted (100 – SF-36vs = inverted SF-36vs) to ease comparison between the other scales, so that higher scores indicate more fatigue. Clinically relevant fatigue was defined by an FSS score of ≥4, a fVAS score of ≥50, and an inverted SF-36vs score of ≥65. Study visits were performed at the time of diagnosis (Visit 1), after adhering to a gluten free diet for one year (Visit 2) and for five to seven years (Visit 3). Diet adherence was evaluated through urinary gluten immunogenic peptides (GIP) and the Celiac Diet Adherence test (CDAT) [5]. Disease activity was assessed through Celiac Symptom Index (CSI) [6], celiac autoantibodies (anti tissue transglutaminase IgA and deamidated gliadin peptide IgG) and histologic evaluation (Marsh score). Nutritional status was assessed through BMI, hemoglobin, 25-hydroxy vitamin D, cobalamin, ferritin and folic acid levels.

Results

Fatigue severity was significantly reduced from time of diagnosis, for all three instruments (Table 1). Fatigue prevalence rates were significantly decreased from Visit 1 (40-53%) to Visit 3 (27-31%). Fatigue was significantly lower in patients with good adherence compared to patients with poor adherence according to CDAT scores (For FSS p=0.022, fVAS p=0.017, and for inverted SF-36vs p=0.033). There was however no difference in fatigue severity between adherent and nonadherent patients when evaluating adherence through urinary GIP. Fatigue was lower in patients in clinical remission according to CSI scores (for FSS p=0.002, fVAS p=0.004, and for inverted SF-36vs p=0.002), but not associated with autoantibodies or nutritional status. In multiple linear regression analysis, depression was the only independent variable significantly associated with all three fatigue measures at Visit 3, whereas pain was positively associated with inverted SF-36vs scores.

Table 1.

Conclusion

Fatigue remains a frequent complaint in celiac disease even after five to seven years on a gluten free diet. However, fatigue severity is reduced from time of diagnosis. Fatigue was associated with dietary adherence as assessed by CDAT. Disease activity according to CSI was associated with fatigue, whereas nutritional status, autoantibodies and histologic changes were not.

References

1. Krupp, L.B. and D.A. Pollina, Mechanisms and management of fatigue in progressive neurological disorders. Curr Opin Neurol, 1996. 9(6): p. 456-60.
2. Krupp, L.B., et al., The fatigue severity scale. Application to patients with multiple sclerosis and systemic lupus erythematosus. Arch Neurol, 1989. 46(10): p. 1121-3.
3. Wolfe, F., Michaud K, Pincus T, Preliminary evaluation of a visual analog function scale for use in rheumatoid arthritis. J Rheumatol, 2005. 32(7): p. 1261-6.
4. Ware, J.E., Jr. and C.D. Sherbourne, The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care, 1992. 30(6): p. 473-83.
5. Leffler, D.A., et al., A simple validated gluten-free diet adherence survey for adults with celiac disease. Clin Gastroenterol Hepatol, 2009. 7(5): p. 530-6, 536.e1-2.
6. Leffler, D.A., et al., A validated disease-specific symptom index for adults with celiac disease. Clin Gastroenterol Hepatol, 2009. 7(12): p. 1328-34, 1334.e1-3.

Disclosure

Berit Mære Skjellerudsveen: Received lecture honoraria from Takeda for lectures on IBD and pregnancy. This study is partly funded by a grant from the patient organization “The Norwegian Coeliac Society”.

Introduction

The randomized controlled OPTIC trial reported that optimised mercaptopurine treatment was superior to placebo in achieving clinical, endoscopic and histological outcomes at one year in ulcerative colitis (UC) patients failing 5-aminosalicylates after receiving induction treatment with corticosteroids.

Aims & Methods

We aimed to assess cost-effectiveness and cost-utility of therapeutic drug monitoring (TDM)-guided mercaptopurine treatment in UC. Data were collected during the OPTIC trial to perform a health economic evaluation for mercaptopurine treatment versus placebo in UC patients. The economic evaluation was performed from a health care and societal perspective using a cost-effectiveness and cost-utility analysis with a time horizon of one year. Volumes and costs of in-hospital care, out-of-hospital care, out-of-pocket expenses and productivity loss were assessed. Main outcomes were the costs per patient who achieved combined corticosteroid-free clinical remission and endoscopic improvement at one year and the costs per quality-adjusted life-year (QALY).

Results

In total, 59 patients were randomized to the intervention (n = 29) and control (n = 30) group. Patients on mercaptopurine treatment showed a non-significant gain in QALYs of 0.047 (-0.024 to 0.117) (P=0.184) corrected for baseline health utility. Non-significant differences in costs were €63 (-€1267 to €1434; P=0.927) in favour of placebo from a health care perspective and -€742 (-€3683 to €2016; P=0.644) in favour of mercaptopurine from a societal perspective, including patients’ out-of-pocket expenses and employers’ productivity loss. The probability of optimised mercaptopurine treatment being cost-effective was 0.80 at a willingness to pay per additional QALY of €20,000.

Conclusion

Optimised mercaptopurine treatment in UC patients failing 5-aminosalicylates is a cost-effective strategy from a societal perspective.

References

1. de Boer NKH, Peyrin-Biroulet L, Jharap B, et al. Thiopurines in inflammatory bowel disease: new findings and perspectives. J Crohns Colitis 2018;12:610–20.
2. Löwenberg, M., Volkers, A., van Gennep, S., Mookhoek, A., Montazeri, N., Clasquin, E., Duijvestein, M., van Bodegraven, A., Rietdijk, S., Jansen, J., van Asseldonk, D., van der Zanden, E., Dijkgraaf, M., West, R., de Boer, N., & D'Haens, G. (2023). Mercaptopurine for the Treatment of Ulcerative Colitis: A Randomized Placebo-Controlled Trial. Journal of Crohn's & Colitis, 17(7), 1055–1065.

Disclosure

NA

Introduction

Intestinal ultrasound (IUS), in particular bowel wall thickness (BWT), is accurate in determining endoscopic disease activity in ulcerative colitis (UC). Reduction of BWT six weeks into advanced anti-inflammatory treatment predicted endoscopic response and clinical remission. Filgotinib is a preferential JAK1-inhibitor effective in UC. Currently, there is no data on IUS-kinetics or early prediction of treatment response with filgotinib.

Aims & Methods

In this study, we aimed to identify early IUS predictors of treatment response to filgotinib in patients with moderate to severe UC. This prospective observational study included UC patients with active endoscopic disease (i.e. endoscopic MAYO score (EMS) ≥2) starting with filgotinib. Clinical scores (UCEIS, Lichtiger), biochemical parameters and IUS parameters of the sigmoid were assessed at baseline, after 4 weeks (W4) and at the time of the second endoscopy (between W8 and W23). Endoscopic remission (i.e. EMS=0), endoscopic response (i.e. decrease in EMS ≥1 compared to baseline) and clinical remission (SCCAI ≤2 or Lichtiger score ≤3) were assessed. IUS response (i.e. defined as a reduction in BWT of >25% or >2.0 mm, or >1.0 mm and one colour Doppler signal) was assessed at week 4. All normally distributed data were reported in mean ± standard deviation [SD] and non-normally distributed in median and interquartile range [IQR]. Mann–Whitney U tests were used to compare continuous non-parametric variables, chi square tests for dichotomous variables.

Results

A total of 20 patients were included. 13 patients underwent second endoscopy, 2 refused a follow-up endoscopy and 5 patients were still in follow-up. At baseline, 9 patients (9/20, 45%) had received ≥2 biologicals, 16 patients (16/20, 80%) had previous anti-TNF exposure, and no patients were exposed to other JAK-inhibitors. 8 patients (8/20, 40%) had severe endoscopic disease at baseline (i.e. EMS=3). 5 patients (5/13, 38%) were endoscopic responders at W8-23, and no patient reached endoscopic remission. BWT was significantly lower in the sigmoid colon from W4 onwards for patients reaching endoscopic response (Table 1). Decrease in BWT compared to baseline was statistically greater in endoscopic responders from W4 onwards (Table 1). IUS response was seen at W4 in 6 patients, of which 5 were endoscopic responders (83%, p=0.002). Decrease in BWT from W4 onwards was statistically greater in patients reaching clinical remission at W8-23 (-45% (-55% to -36.6%) vs -9% (-25% to +8%), p=0.014).

Conclusion

On IUS, UC patients treated with filgotinib have significantly lower BWT from week 4 onwards compared to baseline, which could predict endoscopic response.

Disclosure

MP received speakers fees from Janssen and Tillotts
CT received speakers fees from Tillotts
FV received speaker or honoraria fees from AbbVie, Janssen and Tillotts
ML received consultancy/lecturefees from Abbvie, Bristol Myers Squibb , Eli Lilly, Galapagos, Janssen-Cilag, Johnson & Johnson, Medtronic, Pfizer, Takeda and Tillotts.
RG Received speakers fee from Janssen and Tillotts
GD has served as a consultant and/or received speaker’s fees from AbbVie, Alimentiv, AstraZeneca, Boehringer Ingelheim, Bristol Myers Squibb, Celltrion Healthcare, Cytoki Pharma, Eli Lilly, Ferring, Galapagos, Glaxo SmithKline, Gossamer Bio, Immunic, InDex Pharmaceuticals, Janssen, Johnson & Johnson, Kaleido, Pfizer, Progenity, Prometheus Biosciences, Prometheus Laboratories, Protagonist Therapeutics, Takeda, Tillotts, Ventyx Biosciences.
KG has received grants from Pfizer Inc, Celltrion and Galapagos; consultancy fees from AbbVie, Arena Pharmaceuticals, Galapagos, Gilead, Immunic Therapeutics, Janssen Pharmaceuticals, Novartis, Pfizer Inc., Samsung Bioepis and Takeda and speaker’s honoraria from Abbvie, Celltrion, Ferring, Janssen Pharmaceuticals, Novartis, Pfizer Inc, Samsung Bioepis, Takeda and Tillotts.

Introduction

MASLD patients is increasing worldwide and the clinical outcomes of different phenotypes had shown varied especially the occurrence of of liver related events (LRE) or major cardiovascular events (MACE)(Eslam M, et al. Hepato Inter 2020, 14:889–919) In addition, the prognostic factors for LRE. or MACE. are not well studied.

Aims & Methods

Aims: We aimed to study the rate of LRE. or MACE and the predictors for LRE. Or MACE. in MASLD patients with liver biopsy proven.
Methods: In a single-tertiary-center prospective cohort, we collected data of MASLD patients with biopsy-proven who were followed up of at least 1 years at King Chulalongkorn Memorial Hospital during 2005-2022. The primary outcome was the presence of LRE. or MACE. which included cirrhotic complications: acute decompensations; EV/GV bleeding, ascites, hepatic encephalopathy or HCC or acute complications requiring liver transplantation or death. MACE was defined by the occurrence of stroke or acute coronary disease or myocardial infarction. Related baseline characteristics, LSM, FIB-4, and liver biopsy were collected.

Results

Among 286 MASLD patients, the MASLD patients with mean age of 50.1 +/- 14.5 years old (range 19-80 yr.) Male and female were 44% and 56 % respectively. MASLD patients were categorized into 3 groups which were type 2 diabetes mellitus (T2DM) in 119 (41.6%), overweight or obese group in 149 (52.1%), and lean MASLD (BMI<23) in 18 (6.3%). Their baseline characteristics were shown in Table 1.
The average BMI was 41.3 kg/m² (range 19.1-88.7). Other metabolic abnormalities were hypertension in 165 (57.7%), and dyslipidemia in 157 (54.9%). There were 6 cirrhotic cases (2.1%) at diagnosis. Presence of advanced liver fibrosis (TE>12KPA) was found in 15 cases (5.2%). After a mean follow-up of 9.5 years (range 1-19); there were 11 cases (3.8%) developed coronary artery disease and 3 cases (1%) developed stroke. LRE. were present in 7 cases (2.4%) which included acute decompensation (3 cases; 1%), HCC (3 cases; 1%), and death (1 cases; 0.3%). The overall complications of LRE. or MACE. were present in 16 cases (5.6%). By using multivariate analysis, the predictors of LRE. or MACE. were in presence of hypertriglyceridemia (aOR 1.01 (1.00-1.01); p =0.002) and older age (aOR 1.06 (1.01-1.11); p =0.023).

Conclusion

The overall complication of LRE. or MACE. was not uncommon (5.6% of this cohort with a mean follow-up of 9.5 years). Older age and hypertriglyceridemia were predictors of LRE. or MACE. occurrence.

References

Eslam M, et al. Hepato Inter 2020, 14:889–919.

Introduction

Biliary stenting by endoscopic retrograde cholangiopancreatography (ERCP) is a common palliative endoscopic procedure for unresectable malignant biliary obstruction or for patients unfit for surgery.

Aims & Methods

We conducted a multicentre retrospective cohort study in 3 centres in Portugal, involving patients submitted to first palliative biliary drainage for malignant biliary obstruction by ERCP, between 2015 and 2022. This study aimed to identify predictive factors of 90-day mortality after endobiliary drainage and propose a predictive model. We have analysed data from 217 patients. Statistical analysis was performed using SPSS v.29.

Results

In our cohort, 51,2% (N=111) of patients were female and the mean age was 74 years. 52,5% (N=114) had stage IV disease and in 88% (N=191) the biliary obstruction resulted from primary tumour involvement. The most frequent tumour was primary pancreatic cancer (53,5%, N=116), followed by cholangiocarcinoma (26,7%, N=58). Metallic stents were used in 53,9% (N=117) of the patients, mainly with fully uncovered self-expandable metallic stents (66,7%, N=78). In the multivariate regression model, the independent predictors for 90-day mortality after first ERCP-guided stent placement were primary tumour size <3 cm (p = 0,014; OR 0,394; 95% CI, 0,187-0,830), stage IV malignant disease (p = 0,005; OR 3,776; 95% CI, 1,507-9,465), intrahepatic biliary obstruction (p = 0,004; OR 3,727; 95% CI, 1,518-9,150) and palliative chemotherapy after ERCP biliary drainage (p <0,0001; OR 0,44; 95% CI, 0,015-0,132). Using these variables, we developed a prediction model for the risk of mortality within 90 days after biliary drainage (Hosmer-Lemeshow X² = 2,907, p = 0,893) with R² of 44,1% in our cohort. The AUROC for our prediction model was 0,833 (p <0,0001; 95% CI, 0,774-0,892). By the Youden’s J statistic, the optimal cut-off value for was 0,4, with a sensitivity of 73,9% and specificity of 76,3%. Patients with a score < 0,2 had a low likelihood of death, whereas a score ≥ 0,9 had a high likelihood of death.

Conclusion

The present study was able to identify statistically significant predictive factors of 90-day mortality after ERCP-guided biliary stenting in unresectable malignant biliary obstruction, which may help to stratify the risk of death within 90 days post-stent placement. Since all these predictive factors can be determined before ERCP, this risk stratification could help in the definition of the proper palliative strategy for each patient.

References

Termsinsuk, P., Charatcharoenwitthaya, P., & Pausawasdi, N. (2022). Development and validation of a 90-day mortality prediction model following endobiliary drainage in patients with unresectable malignant biliary obstruction. Frontiers in oncology, 12, 922386. https://doi.org/10.3389/fonc.2022.922386;
Sripongpun, P., Attasaranya, S., Chamroonkul, N., Sookpaisal, T., Khow-Ean, U., Siripun, A., Kongkamol, C., Piratvisuth, T., & Ovartlarnporn, B. (2018). Simple Clinical Score to Predict 24-Week Survival Times in Patients with Inoperable Malignant Distal Biliary Obstruction as a Tool for Selecting Palliative Metallic or Plastic Stents. Journal of gastrointestinal cancer, 49(2), 138–143. https://doi.org/10.1007/s12029-017-9918-9;
Afshar, M., Khanom, K., Ma, Y. T., & Punia, P. (2014). Biliary stenting in advanced malignancy: an analysis of predictive factors for survival. Cancer management and research, 6, 475–479. https://doi.org/10.2147/CMAR.S71111.

Introduction

The use of artificial intelligence (AI) systems for decision support has recently gained great popularity in the medicine.¹ However, recent studies have shown possible limitations emerging from the interaction between humans and AI. Among these, the term “automation bias (AB)” refers to human errors deriving from over-relying on AI’s false alarms such that the human changes a correct initial interpretation.² AB is poorly investigated in the medical field and was never tested in the gastroenterological setting.

Aims & Methods

The aim of the study was to investigate the impact of AI on human diagnostic performance in gastrointestinal endoscopy, including AB.
Ninety high-quality endoscopic videoclips (16 seconds) extracted from small bowel (SB) capsule endoscopy (N=60) and proctosigmoidoscopy (N=30) were edited with the addition of a mock AI system designed with diagnostic accuracy, sensitivity and specificity of 80%, 86% and 75%. The ground truth was settled on the agreement of 9 established experts about the presence of SB lesions with significant hemorrhagic potential according to the Saurin classification (30 videoclips), inflammatory lesions relevant for the diagnosis of SB Crohn's disease according to the Lewis score (30 videoclips) and assessing the Ulcerative Colitis (UC) Endoscopic Index of Severity (30 videos). The videoclips were randomly divided into three sessions and submitted around Europe to fully trained endoscopist, trainees and endoscopic technicians. Participants were instructed to the accuracy measures of AI annotations and scored each videoclip first without, then with AI support. Participants’ performances with and without AI were calculated. Subgroup analyses were made after stratification by participants’ experience level in each endoscopic setting and test performances.

Results

A total of 12.660 videoclips (both with and without AI) were visualized in 422 sessions (182, 118 and 122 participants completed the first, second and third session, respectively). A strong and significant difference between the respondents' performance (N=281) when they were unaided by AI (average accuracy (M)=0.78, [0.77, 0.79] and when they were supported by the AI (M=0.81, [0.80, 0.82], paired t-test, t(276)=11.2, p<0.001) was found. Size effect and the evidence in favour of a significant positive effect of AI were large, 0.67 (Cohen's D) and very large (Bayes Factor>100), respectively. Interestingly, no significant difference between trainees and specialists both in regard to their performance, either unaided, M=0.77 [0.74, 0.79] vs M=0.78, [0.77, 0.79] or aided M=0.80 [0.78, 0.82] vs M=0.81 [0.80, 0.82] and their accuracy improvement (M=0.03, [0.02, 0.05] vs M=0.03, [0.02, 0.04] (unaided t-value=-0.89, p=0.37; aided t-value=-0.69, p=0.49, difference t-value=0.54, p=0.59) were found. A strong leveller effect emerged, with low-performers improving their accuracy much more than top-performers (average improvement 5%, [0.04, 0.06] vs 1%, [0.00, 0.02], paired T test, t=6.53, p<0.001). Nine percent of the final decisions involved disregarding correct AI advice in favour of incorrect human judgment, while AB occurred in 2% of cases.

Conclusion

In this study, the application of AI to assess SB bleeding, SB inflammatory lesions, and UC activity brings significant improvement in diagnostic accuracy, with the benefit appearing to be greater in the low-performers. However, in rare cases, AI may induce diagnostic errors due to AB. Future studies directly evaluating the impact of AI in real-life clinical settings are needed to confirm these data.

References

1. Cabitza F, Campagner A, Angius R, Natali C, Reverberi C. 2023. AI Shall Have No Dominion: on How to Measure Technology Dominance in AI-supported Human decision-making. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI '23). Association for Computing Machinery, New York, NY, USA, Article 354, 1–20. https://doi.org/10.1145/3544548.3581095
2. Reverberi, C., Rigon, T., Solari, A. et al. Experimental evidence of effective human–AI collaboration in medical decision-making. Sci Rep 12, 14952 (2022). https://doi.org/10.1038/s41598-022-18751-2