Introduction

ChatGPT is a large language model with applications in numerous fields, including medicine^1,2 . Post-colonoscopy surveillance is crucial in colorectal cancer prevention, but adherence to guidelines remains low^{3, 4} . This study aims to evaluate ChatGPT's ability to streamline post-colonoscopy patient management by processing endoscopy and pathology results and providing guideline-based recommendations and patient summaries.

Aims & Methods

A proof of concept pilot study was conducted with 20 clinical scenarios prepared by a gastroenterologist, covering various recommendations from society guidelines. Scenarios were prompted to ChatGPT as structured endoscopy reports and free text clinical notes. Two senior gastroenterologists evaluated ChatGPT's responses, marking them as correct or incorrect. Fleiss' Kappa coefficient was calculated to evaluate inter-rater agreement, and accuracy was assessed in terms of concordance with guidelines.

Results

Of the 20 scenarios, 90% (18/20) of ChatGPT's responses adhered to guidelines. In 17/20 (85%) cases, both endoscopists marked the responses as correct and adherent to guidelines. Inter-rater agreement was very good, with a Fleiss' kappa coefficient of 0.84 (p<0.01). ChatGPT interpreted free text clinical notes similarly to structured endoscopic responses in 19/20 scenarios (95%). The model composed concise, easily understandable patient letters with comprehensive advice on maintaining a healthy lifestyle.

Conclusion

ChatGPT shows potential in providing accurate post-colonoscopy surveillance recommendations, which could aid healthcare providers in decision-making and improve guideline adherence. However, limitations include inherent randomness in responses, and the possibility of incorrect recommendations. Future prospective research is essential for exploring the integration of ChatGPT or similar language models into electronic health record systems and evaluating their effectiveness as decision support tools and means of enhancing patient communication.

References

1. OpenAI. GPT-4 Technical Report. 2023 Mar 15 [cited 2023 Mar 29]; Available from: https://arxiv.org/abs/2303.08774v3
2. Kung TH, Cheatham M, Medenilla A, Sillos C, De Leon L, Elepaño C, et al. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLOS Digit Heal [Internet]. 2023 Feb 9 [cited 2023 Mar 29];2(2):e0000198. Available from: https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000198
3. Djinbachian R, Dubé AJ, Durand M, Camara LR, Panzini B, Bouchard S, et al. Adherence to post-polypectomy surveillance guidelines: A systematic review and meta-analysis. Endoscopy [Internet]. 2019 [cited 2023 Mar 24];51(7):673–83. Available from: http://www.thieme-connect.com/products/ejournals/html/10.1055/a-0865-2082
4. Zagari RM, Frazzoni L, Fuccio L, Bertani H, Crinò SF, Magarotto A, et al. Adherence to European Society of Gastrointestinal Endoscopy Quality Performance Measures for Upper and Lower Gastrointestinal Endoscopy: A Nationwide Survey From the Italian Society of Digestive Endoscopy. Front Med. 2022 Apr 6;9:847.

Introduction

Colonoscopy is the gold standard for diagnosing colorectal cancer (CRC)¹, with post-colonoscopy colorectal cancer (PCCRC) defined by the World Endoscopy Organization (WEO) as CRC diagnosed between 6 and 48 months after a negative colonoscopy². PCCRC rates vary from 2.5% to 7%², and the prognosis of PCCRC is generally poorer than that for CRC detected during colonoscopy³.

Aims & Methods

To evaluate the risk factors associated with PCCRC in a major tertiary medical center.
We analyzed all consecutive colonoscopic procedures conducted between 2001 and 2019. Data was retrieved from electronic medical records (EMR) and cross-referenced with the Israeli Cancer Registry. We defined PCCRC based on the WEO criteria, and two physicians independently confirmed potential PCCRC cases to ensure data accuracy. Root-cause analysis of PCCRC was performed according to the WEO algorithm as follows:
A) possible missed lesion, index examination adequate;
B) possible missed lesion, index examination inadequate;
C) detected lesion, not resected; or
D) likely incomplete resection of previously identified lesion.
We then identified colonoscopy-detected CRCs (CDCRC, controls) and compared them to the PCCRC cases at a ratio of 1:3. We assessed patient median age at the procedure, gender, religion, family history of CRC, diabetes, smoking status, body mass index (BMI), obesity (defined in the EMR or BMI > 30) and endoscopic parameters: colonoscopy indication [surveillance (follow-up colonoscopy according to recommendations), abnormal imaging, patient symptoms], adequate Boston Bowel Preparation Scale (BBPS>6 and >2 in all segments), cecal intubation rate, diverticulosis, rectal retroflection, and tumor location.
Multivariable logistic regression assessed the association of multiple factors (endoscopy indications and obesity) with PCCRC, maintaining a 95% confidence interval (CI).

Results

Out of 54,154 colonoscopies, 2,339 patients were diagnosed with CRC, of which 49 (2.1%) were confirmed PCCRCs, while 2,157 (92.2%) were detected during colonoscopy. Of patients with PCCRC, 65.4% (n=32) had adequate BBPS, and 91.8% (n=45) cecal intubation was documented. The two most common reasons for PCCRC were category A (n=29, 59.1%), and category B (n=17, 34.6%), although in a lesser extent. We then compared PCCRC cases to 150 randomly selected CDCRC cases. Significant differences in patient characteristics included BMI>30 (p = 0.041) and obesity (p=0.015). Statistically significant endoscopic factors included colonoscopy indication (p < 0.001), cecal intubation rate (p < 0.001), and tumor location (p < 0.001). As expected, PCCRC was more commonly detected in the right colon (n=24, 55.8%) or rectum (n=12, 27.9%). Multivariable logistic regression revealed that patients with abnormal imaging as a colonoscopy indication had a lower risk for PCCRC (adjusted odds ratio (aOR) = 0.148, 95% CI: 0.033-0.455), while patients that performed colonoscopy for surveillance had a higher risk (aOR = 7.334, 95% CI: 2.355-25.963). Obesity was also a significant risk factor (aOR = 3.035, 95% CI: 1.326-7.035).

Conclusion

The most frequent cause of PCCRC across various patient populations and colonoscopy indications was a possible missed lesion. Factors such as patient obesity and undergoing colonoscopy for surveillance may increase the likelihood of PCCRC. These findings highlight the importance of physician awareness regarding PCCRC risks during colonoscopy.

References

1. Bretthauer M, Løberg M, Wieszczy P, et al. Effect of Colonoscopy Screening on Risks of Colorectal Cancer and Related Death. N Engl J Med. 2022;387(17):1547-1556.
2. Rutter MD, Beintaris I, Valori R, et al. World Endoscopy Organization Consensus Statements on Post-Colonoscopy and Post-Imaging Colorectal Cancer. Gastroenterology. 2018;155(3):909-925.e3.
3. Govindarajan A, Rabeneck L, Yun L, Tinmouth J, Paszat LF, Baxter NN. Population-based assessment of the outcomes in patients with postcolonoscopy colorectal cancers. Gut. 2016;65(6):971-976.

Introduction

There is a steadily increasing interest among patients in obtaining medical information from electronic sources. ChatGPT, a conversational agent utilizing artificial intelligence, has been demonstrated as a reliable source of information for patients before a colonoscopy (1) as well as in various medical purposes.

Aims & Methods

The aim of this study is to determine the ability of ChatGPT 3.5 to issue recommendations after a screening colonoscopy for colorectal cancer. This is a single-center observational cross-sectional study conducted in 2023. Only colonoscopies for screening purposes were included. Colonoscopies performed for gastrointestinal bleeding, follow-up of chronic inflammatory bowel diseases, post-operative follow-up of colorectal cancer, or in patients under 40 years old were excluded. Colonoscopy reports and corresponding pathology reports were entered into a conversation with ChatGPT 3.5. The colonoscopy report includes the quality of colonic preparation, withdrawal time, results, and specimens sent for histology. ChatGPT 3.5 was asked to determine the time interval for the follow-up colonoscopy using the Consensus of the American multisociety taskforce of 2020 (2). The answers from ChatGPT were compared to the recommendations, and were validated by two gastroenterologists. The answers were categorized as “correct”, “underestimated” if the interval recommended by ChatGPT is shorter than the one provided by the consensus, and “overestimated” if the interval recommended by ChatGPT is longer than the one provided by the consensus.

Results

A total of 1259 colonoscopies were considered for analysis. The average age at the time of colonoscopy was 61.6 ± 11.0 years old. Among the participants, 688 (54.6%) were women and 517 (45.4%) were men.
ChatGPT® issued a correct recommendation in 892 cases (70.8%). The surveillance interval was underestimated in 260 (20.7%) cases and overestimated in 107 (8.5%) cases.
The accuracy of the recommendations for each identified category is represented in the table below.

Table 1. Accuracy of the recommendations issued by ChatGPT 3.5 according to the ​colonoscopy result.

Conclusion

ChatGPT appears to be a reliable tool for issuing recommendations after a screening colonoscopy. It is limited by the recognition of poor preparation requiring closer surveillance. Artificial intelligence could be exploited and integrated into colorectal cancer screening programs. It could also be used by a patient to remind them of their next colonoscopy, except in cases of poor colonic preparation.

References

1. Lee TC, Staller K, Botoman V, Pathipati MP, Varma S, Kuo B. ChatGPT Answers Common Patient Questions About Colonoscopy. Gastroenterology. 2023;165(2):509-511.e7. doi:10.1053/j.gastro.2023.04.033
2. Gupta S, Lieberman D, Anderson JC, et al. Recommendations for Follow-Up After Colonoscopy and Polypectomy: A Consensus Update by the US Multi-Society Task Force on Colorectal Cancer. Gastrointest Endosc. 2020;91(3):463-485.e5. doi:10.1016/j.gie.2020.01.014

Introduction

The adherence of doctors and patients to published colorectal post-polypectomy surveillance guidelines varies greatly, and patient follow-up is critical but time-consuming.

Aims & Methods

Aims: We aimed to develop an automatic surveillance (AS) system to accurately identify post-polypectomy patients, assign surveillance intervals for different risks of patients and proactively follow up with patients on time.
Methods: The endoscopic and pathological reports from 47,544 patients from three hospitals were used to develop AS system based on natural language processing. The performance of AS system was fully evaluated in internal and external tests according to five different guidelines worldwide and compared with that of doctors. Furthermore, a multi-reader-multi-case (MRMC) trial was conducted to evaluate the effect of AS system on improving doctors’ guideline adherence and efficiency, and a prospective trial was performed to evaluate the success rate for automatically informing patients and its effect on reducing the burden of nurses.

Results

In internal, external 1, and external 2 tests, the system reached an overall accuracy of 99.73%, 99.54%, and 99.77% for identifying different types of patients, respectively; and achieved an overall accuracy of ≥99.45%, ≥98.89%, and ≥98.56% for stratifying patients' risk levels and assigning surveillance intervals according to five guidelines. The system greatly increased the accuracy of doctors (98.67% vs. 78.10%, p=0.043) in MRMC trial. In a prospective trial, AS system successfully informed 93·18% of patients, and significantly reduced the follow-up time cost (127.4s vs. 14.02s per patient).

Conclusion

This system has great potential to improve doctors' adherence to guidelines while relieving the workload of doctors and nurses.

Disclosure

The authors declare no conflict of interest.

Introduction

Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer-related deaths globally, with increasing incidence and a trend toward younger patients. CRC develops through the adenoma-carcinoma sequence, typically over 10 to 15 years, offering a critical “golden window” for early diagnosis and intervention. Early prevention, diagnosis, and treatment are key to reducing CRC mortality, with high-quality colonoscopy playing a vital role. The adenoma detection rate (ADR), a key measure of colonoscopy quality, is influenced by factors such as withdrawal time, bowel preparation, physician experience, cecal intubation rate, and patient characteristics. This study aims to identify risk factors for colorectal adenoma and develop a risk prediction model to improve colonoscopy screening, enhancing adenoma detection rates and reducing CRC risk.

Aims & Methods

This study aims to investigate the risk factors for colorectal adenoma based on patient characteristics and factors related to the colonoscopy procedure, and to develop a risk prediction model for colorectal adenoma.
A retrospective study design was adopted, enrolling 830 patients who underwent painless colonoscopy at the outpatient clinic of the Affiliated Wuxi People's Hospital of Nanjing Medical University between January and June 2023. Data collected included patient characteristics such as gender, age, BMI, medical history, clinical symptoms, and personal history, as well as colonoscopy-related information such as the number, anatomical location, and histological type of adenomas, insertion time, withdrawal time, bowel preparation, and the physician's experience level. Univariate and multivariate logistic regression analyses were performed to identify independent risk factors for colorectal adenoma. A nomogram risk prediction model for colorectal adenoma was constructed using R language, and its predictive performance was evaluated using ROC curves, calibration curves, and decision curves. Additionally, random forest (RF), XGBoost (eXtreme Gradient Boosting), and LightGBM (Light Gradient Boosting Machine) algorithms were applied to further validate the factors associated with the adenoma detection rate (ADR).

Results

Multivariate logistic regression analysis revealed the following independent risk factors for colorectal adenoma: age (OR = 1.05, 95% CI: 1.03–1.08, p < 0.001), male gender (OR = 1.79, 95% CI: 1.32–2.41, p = 0.005), physician experience level (OR = 1.79, 95% CI: 1.20–2.68, p = 0.005), and withdrawal time (OR = 1.00, 95% CI: 1.00–1.00, p < 0.001). The ROC curve for the risk prediction model of colorectal adenoma in the modeling cohort had an area under the curve (AUC) of 0.720, indicating good discriminative ability. The calibration curve showed good agreement between the predicted and observed values, and the decision curve analysis suggested that the model has high clinical predictive value. RF, XGBoost, and LightGBM algorithms also confirmed that age, gender, physician experience, and withdrawal time are important factors influencing ADR.

Conclusion

Male, older age, higher physician experience, and longer withdrawal time are independent risk factors for the occurrence of colorectal adenoma. This study established a risk prediction nomogram model for colorectal adenoma based on easily accessible clinical indicators, which can effectively guide colonoscopy screening, improve the detection rate of colorectal adenomas, and help reduce the incidence of colorectal cancer.

Introduction

Post colonoscopy colorectal cancer (PCCRC) is colorectal cancer diagnosed after a normal colonoscopy with no cancer found. In England, the rate of PCCRCs has been decreasing but remains around 6.5% for colonoscopies performed from 2011-2013¹. English study suggested that 80% of PCCRCs are avoidable². We aim to analyse PCCRC cases using the local data and determine the demographics, nature of cancer and the cause of PCCRC.

Aims & Methods

The PCCRCS are identified following the linkage of Hospital Episode Statistics, Bowel cancer screening programme (BCSP) and cancer registry datasets. As new colorectal cancers are registered with the National Cancer Registration and Analysis Service (NCRAS), PCCRC cases will be identified. Each colorectal cancer (ICD-10 code C18-20) provisionally registered within the Hospital Episode Statistics dataset will be identified and determined whether a colonoscopy was performed within 6- 48 months before the diagnosis which failed to identify the cancer.
This audit includes patients diagnosed with PCCRC 6-48 months after a colonoscopy was performed in a single medical centre in England.
Once the cases had been confirmed, we reviewed the clinical, pathology and radiological findings of each case. We also looked at interval cancer which the World Endoscopy Organisation (WEO) defines as one diagnosed before the next recommended interval. Important to note that all interval cancers are PCCRCs but not all PCCRCs are interval cancer.

Results

39 cases were included and 29 were confirmed to be true PCCRC which fulfilled the above criteria. 7 did not meet the criteria. (2 index colonoscopies performed at other trusts, 3 missing data, 2 other types of cancers such as Neuroendocrine tumour and appendiceal tumour, 1 has <6 months to cancer diagnosis from index colonoscopy, 2 flexible sigmoidoscopy was performed as index colonoscopy to investigate per rectal bleeding but subsequent cancer was found in ascending colons).
52% (n=15) of patients were male with a median age of cancer diagnosis at 74 years old (range 22-84). In terms of PCCRC, 24% (n=7) presented via the National Bowel Screening Programme (BCSP), 14%(n=4) presented with per rectal bleeding, 7% (n=2) presented with abnormal imaging and the same proportion of patients presented due to anaemia, non-specific change in bowel habit, weight loss, diarrhoea, abdominal pain and surveillance for family history of gene mutation. 3% (n=1) were diagnosed after presenting for polypectomy and 3% (n=1) presented with intestinal obstruction.
In terms of investigations that led to PCCRC, 45% (n=13) were via 2-weeks-wait colonoscopy, 24% (n=7) were via National Bowel Screening Programme, 17% (n=5) were diagnosed by Computed Tomography (CT) imaging, 7% (n=2) via flexible sigmoidoscopy and 7% (n=2) was via surgically resected sample. Upon diagnosis of PCCRC, 7% (n=2) tested positive for Lynch syndrome, 24% (n=7) was tested but no marker was found and 69% (n=20) was not tested. In terms of size of PCCRC, 66% (n=19) had tumour size >2cm. 66% (n=17) of tumour size was determined by histopathological sample, 21 (n=6) via endoscopy, 3% (n=1) via CT imaging and 17% (n=6) was not known.
In terms of PCCRC staging, 48% (n=14) was Dukes’ A staging, 21% (n=6) was Dukes’ B, 7% (n=2) was Dukes’ C and 24% (n=7) was Dukes’ D. 83% (n=24) was planned for curative treatment and 17% (n=5) was planned for palliative treatment. 52% (n=16) had surgical resection only, 17% (n=29) had surgical resection and chemotherapy, 3% (n=11) had surgical resection and radiotherapy, 3% (n=1) had surgical resection, chemotherapy and radiotherapy, 7% (n=2) had chemotherapy and radiotherapy, 3% (n=1) had chemotherapy only, 3% (n=1) had endoscopic resection only and 7% (n=2) had no treatment.
In terms of the type of interval cancer, 14% (n=4) was non-interval A, 7% (n=2) was non-interval B, 66% (n=19) was non-interval C and 21% (n=4) was interval cancer. Important to note was 66% (n=19) had no further follow-up or test arranged due to adequate examination on the index endoscopy. 90% of index endoscopies report no difficulties. 75% (n=22) was due to clinician factors, 10% (n=3) was due to administrative errors such as booking delay and 10% (n=3) was due to patient factors such as patient’s decision and intercurrent illness.

Conclusion

The majority of the PCCRC had treatment with curative intent. Most PCCRC cases are due to clinician factors and measures needed to improve this. The high percentage of non-interval type C PCCRC would be explained by either a fast-growing cancer or a missed lesion since up to 90% of the index procedures reported no difficulty.

References

1. Burr NE, Derbyshire E, Taylor J, et al. Variation in post-colonoscopy colorectal cancer across colonoscopy providers in English National Health Service: population-based cohort study. BMJ 2019;367:l6090. doi:10.1136/bmj.l6090
2. Anderson R, Burr NE, Valori R. Causes of Post-colonoscopy Colorectal Cancers Based on World Endoscopy Organization System of Analysis. Gastroenterology Published Online First: January 2020. doi:10.1053/j.gastro.2019.12.03