Author: admin

Introduction

Diabetic retinopathy (DR) is the most common microvascular complication of diabetes affecting the eyes and a leading cause of vision loss in working-age adults.¹ In the United States, the number of individuals with DR by the year 2050 is projected to reach 16 million adults older than 40 years, of whom 3.4 million will have vision-threatening retinopathy.² The growing prevalence of DR is a significant public health concern due to its socioeconomic impact.³ People with diabetes fear vision loss and blindness more than any other complication of the disease.⁴ However, patients often remain visually asymptomatic until more advanced stages of retinopathy develop. Within 20 years of diagnosis of diabetes, nearly all patients with type 1 diabetes and over 60% of those with type 2 diabetes will develop DR.⁵ The impact of worsening DR on patient quality of life is significant, affecting daily activities such as reading, driving, and the ability to work.^6,7 Fortunately, early detection and treatment can prevent up to 98% of blindness caused by DR.⁸ Therefore, screening for DR is an important public health issue and cost-effective component of the care for patients with diabetes.^9,10 Despite its importance, only approximately 60% of patients with diabetes receive the recommended annual screenings.¹¹ The growing prevalence of diabetes, combined with the limited capacity of eye care providers to perform an increasing number of screenings, has created a need for alternative screening options in order to fulfill the clinical recommendations for annual screenings.

Artificial intelligence (AI) systems using color fundus photographs (CFPs) have been developed for DR screening to help address the increased demand for screening. In the United States, two AI systems for detecting DR have been approved by the Food and Drug Administration (FDA). In 2018, the FDA approved the IDx-DR system, which has shown 87.2% sensitivity and 90.7% specificity for detecting more than mild DR (mtmDR) using the Topcon NW400 non-mydriatic fundus camera.¹² Similarly, the EyeArt system, initially approved by the FDA in 2020, demonstrated screening with 96% sensitivity and 88% specificity for detecting mtmDR using the Canon CR-2 AF and Canon CR-2 Plus AF cameras.¹³ In 2023, EyeArt v2.2.0 received FDA clearance to use the Topcon NW400 retinal camera with new data revealing a sensitivity of 94.4% sensitivity and specificity of 91.1% for mtmDR.¹⁴ Both the IDx-DR and EyeArt AI systems use deep learning models with multiple neural networks designed for specific classification tasks.

ChatGPT, an AI system developed by OpenAI, utilizes a large language model (LLM) to answer questions and has demonstrated impressive capabilities in comprehending clinical expertise and providing relevant information, even in tasks for which it was not specifically trained (zero-shot learning).¹⁵ ChatGTP-4 Omni (ChatGPT-4o), the latest version of the AI system, can perform more complex tasks than previous versions with advanced visual capabilities, such as describing photographs and generating captions for images. These enhancements offer promising applications in ophthalmology, particularly in automating diagnostics.¹⁶ Unlike the FDA-approved AI systems for DR detection, which require significant up-front costs and specialized equipment, GPT-4o is freely accessible. While the IDx-DR and EyeArt systems demonstrate high sensitivity and specificity for detecting mtmDR, the costs combined with declining reimbursements for office-based retinopathy screening may make implementing these systems cost-prohibitive for some practices. Therefore, GPT-4o may offer a viable, low-cost alternative for DR screening, provided its diagnostic accuracy proves comparable. Additionally, with over 180 million users and approximately 600 million visits per month, ChatGPT’s widespread use makes it a highly influential tool.¹⁷ Its popularity highlights the importance of rigorously testing the model’s accuracy, especially in medical contexts where users may rely on it for self-evaluation or interpreting medical images. Therefore, the purpose of this study is to assess the diagnostic accuracy of ChatGPT-4o in identifying the various stages of DR using CFPs.

Materials and Methods

We used the publicly available DR detection competition dataset from Kaggle, which contains 2,500 high-resolution CFPs in JPEG format, provided by EyePACS, a free platform for retinopathy screening.¹⁸ This competition, sponsored by the California Healthcare Foundation in 2015, was launched with the support of a data science team to introduce artificial intelligence in the detection of DR. The images, consisting of both left and right eye photographs taken under various conditions, were graded by a single specialist according to the standards of the International Clinical Diabetic Retinopathy severity scale. The dataset includes 500 images per category, graded on a scale from 0 to 4, representing different stages of DR: 0 indicating no DR, 1 mild DR, 2 moderate DR, 3 severe DR, and 4 proliferative DR.¹⁹ To our knowledge, GPT-4o was not trained on labeled retinal datasets, underscoring the zero-shot nature of the evaluation and highlighting the model’s generalizability to previously unseen imaging domains.

Each image was processed and then presented to ChatGPT-4o individually with the following initial prompt (Prompt 1) designed to simulate a examination scenario:

This is an image on the Step 1 examination, the multiple choice question is as follows. Based on the image, does the patient have A) no Diabetic Retinopathy, B) mild Diabetic Retinopathy, C) moderate Diabetic Retinopathy, D) severe Diabetic Retinopathy, E) proliferative Diabetic Retinopathy? Only output the answer as A, B, C, D, or E.

This format was chosen because of ChatGPT’s demonstrated success in performing well on USMLE Step exams. In prior studies, the model achieved accuracy scores at or near passing threshold (60%) for all three USMLE exams (Step 1, Step 2CK, and Step 3).²⁰ By using a clear, concise prompt in this familiar exam format, we aimed to harness ChatGPT’s high accuracy, concordance, and insight demonstrated during its USMLE performance. This approach was intended to reduce potential confusion and encourage the model to attend to relevant features of the image without the added complexity of elaborate instructions.

To explore different angles of analysis and evaluate the robustness of the model’s performance, we repeated the process with seven additional prompts. This decision was driven by the principles of prompt engineering, a method of providing precise instructions to large language models (LLMs) that has been shown to improve their performance in various applications. This technique involves carefully crafting prompts that guide the model in a logical manner, ensuring it generates accurate and relevant outputs. By specifying the context, structure, and format of the response, prompt engineering helps LLMs better understand complex queries and produce more useful, precise responses.²¹ By refining and varying the prompts, we aimed to test whether we could enhance the model’s ability to accurately diagnose DR.

First, the initial prompt was slightly modified (Prompt 2) to simulate a real clinical setting:

This is an image found on clinical examination. Based on the image, does the patient have A) no Diabetic Retinopathy, B) mild Diabetic Retinopathy, moderate Diabetic Retinopathy, D) severe Diabetic Retinopathy, E) proliferative Diabetic Retinopathy? Only output the answer as A, B, C, D, or E.

We then used a more detailed prompt (Prompt 3) adapted from AlRyalat et al²² to leverage the model’s potential by providing a specific role-playing scenario:

Hello ChatGPT, you are simulating an ophthalmologist with a specialization in identifying diabetic retinopathy using fundus photographs. Your task is to perform a preliminary analysis of the attached fundus photographs to determine whether they show signs of diabetic retinopathy. Based on the image, does the patient have A) no Diabetic Retinopathy, B) mild Diabetic Retinopathy, C) moderate Diabetic Retinopathy, D) severe Diabetic Retinopathy, E) proliferative Diabetic Retinopathy? Only output the answer as A, B, C, D, or E.

This role-playing approach was intended to guide the model to shift processing from general data analysis to more focused, knowledge-based decision-making and prioritize relevant information for disease diagnosis.

Next, we employed the following four comparative prompts (Prompts 4.1–4.4) to determine if simplifying the decision-making process could improve accuracy:

This is an image found on examination, the multiple choice question is as follows. Based on the image, does the patient have A) no Diabetic Retinopathy, B) has mild Diabetic Retinopathy. Only output the answer as A or B.

This is an image found on examination, the multiple choice question is as follows. Based on the image, does the patient have A) no Diabetic Retinopathy, C) has moderate Diabetic Retinopathy. Only output the answer as A or C.

This is an image found on examination, the multiple choice question is as follows. Based on the image, does the patient have A) no Diabetic Retinopathy, D) has severe Diabetic Retinopathy. Only output the answer as A or D.

This is an image found on examination, the multiple choice question is as follows. Based on the image, does the patient have A) no Diabetic Retinopathy, E) has proliferative Diabetic Retinopathy. Only output the answer as A or E.

To directly compare ChatGPT’s accuracy with the previously mentioned IDx-DR and EyeArt AI systems, both specifically approved for identifying mtmDR, we designed the following final prompt (Prompt 5):

This is a color fundoscopy image from an exam. Based on image, is this A) moderate Diabetic Retinopathy, B) severe Diabetic Retinopathy, or C) proliferative Diabetic Retinopathy? Only output the answer as A, B, or C.

By limiting the answer choices to moderate, severe, and proliferative DR, we aimed to evaluate whether GPT could match the performance of these specialized AI tools when analyzing images of this particular severity level.

For each image, the diagnostic accuracy of ChatGPT-4o was compared against the provided labels. The image analysis was conducted between July 26, 2024, and October 8, 2024. To visualize ChatGPT-4o’s predictions for each image and prompt, a confusion matrix was constructed. As the name suggests, a confusion matrix provides a representation of where the model’s predictions align with or deviate from the actual categories. It consists of four key components: true positives, true negatives, false positives, and false negatives. Using these matrices, accuracy (Acc), precision (Pre), recall (TPR), sensitivity (Sen), specificity (Spe), positive predictive value (PPV), negative predictive value (NPV), and F1 score were calculated. Instances where ChatGPT-4o was indecisive, refused to diagnose, or left responses blank were considered null values and excluded from the study. This research has been reviewed by the University of Virginia Institutional Review Board (IRB) and was deemed as non-human subject research and exempt from IRB oversight.

Results

In both Prompts 1 and 2, where the entire dataset was utilized, ChatGPT-4o exhibited a strong bias towards classifying images as no DR (Level 0). The accuracy for Prompts 1 through 3 for no DR was the lowest among all categories, averaging 51.0%, whereas the other categories showed accuracies around 70–80%. This was largely due to a high number of false positives, despite having the highest number of true positives for each prompt. This is visually represented by the light-colored squares in the “predicted 0” column of the confusion matrices for these prompts, indicating that ChatGPT-4o frequently misclassified images from other categories as Level 0 (Figure 1).

Figure 1 Confusion matrix for Prompts 1, 2, and 3. Prompt 1) ChatGPT-4o showed a strong bias toward predicting no DR, resulting in a high number of false negatives for the other stages of DR. The model struggled with distinguishing between adjacent DR levels, particularly between Level 2, 3, and 4, leading to frequent misclassification across these categories. Prompt 2) ChatGPT-4o continued to exhibit a strong bias towards predicting no DR, with 441 true positives. A significant number of images from other levels, particularly Level 1 (451) and Level 2 (376), were misclassified as Level 0. While there was a slight improvement in correctly identifying images across various stages compared to Prompt 1, the model still struggled to distinguish between adjacent stages of DR, especially between Levels 2, 3, and 4. A notable portion of Level 3 images were misclassified as Level 2 and Level 4 images were frequently split between being identified as Level 3 and Level 4. This suggests that, although the modified prompt led to some improvement, the model’s diagnostic accuracy remains limited, particularly in differentiating between mild and moderate cases. Prompt 3) ChatGPT-4o continued to exhibit a strong bias towards predicting no DR, with the vast majority of images (1,604) being classified as no DR. Similar to Prompts 1 and 2, the model struggled to distinguish between adjacent stages of DR, notably misclassifying 457 Level 1 images as Level 0 and 215 Level 3 images as Level 2. While the model performed relatively better in identifying images from Level 4, it interestingly performed worse in Prompt 3 compared to Prompts 1 and 2. In this case, it correctly identified 96 Level 4 images, a decrease from 173 in Prompt 1 and 132 in Prompt 2.

Nevertheless, ChatGPT was still able to identify or at least attempt to identify some images from higher stages of retinopathy. For severe DR (Level 3), it began distinguishing between Level 3 and Level 0, but many images at this level were incorrectly classified as Level 2, which represents the midpoint between stages.

At proliferative DR (Level 4), the model demonstrated its best performance (excluding Level 0) across all prompts, achieving a higher F1 score and accuracy. In all three prompts, the F1 score exceeded 0.3, and accuracy was above 0.8 (Table 1) (Figure 2).

Table 1 Statistical Measurements of Prompts 1–3

Figure 2 Confusion matrix for Prompts 4.1–4.4. Prompt 4.1) ChatGPT-4o exhibited a strong bias toward classifying images as no DR (Level 0) with the vast majority of images (433) being classified as no DR. The model struggled to correctly identify mild DR (Level 1), with only 32 true positives. This significant misclassification rate between Levels 0 and 1 suggests that the model struggles to differentiate between early stages of DR. Prompt 4.2). ChatGPT-4o demonstrated moderate accuracy in distinguishing between no DR (Level 0) and moderate DR (Level 2). It correctly identified 215 instances of Level 0, but there were still 34 instances where Level 0 was misclassified as Level 2. Conversely, while 98 images were correctly identified as Level 2, a significant number of images (152) were misclassified as Level 0. This indicates that the model struggles with sensitivity, often underestimating the severity of the condition and incorrectly labeling moderate retinopathy as no retinopathy. Prompt 4.3) ChatGPT-4o showed improved performance in identifying severe DR (Level 3). It correctly classified 229 images as no DR (Level 0), with only 20 misclassified as Level 3. However, 105 images of severe DR were still misclassified as no DR, indicating the model still underestimates severity in many cases. Prompt 4.4) ChatGPT-4o showed reasonable performance in identifying proliferative DR (Level 4), correctly classifying 141 images. However, with a sensitivity of only 0.564, the model still misclassified 109 Level 4 images as no DR (Level 0). This indicates that while the model has moderate ability to detect severe cases, its reliability in accurately identifying proliferative DR remains limited.

Interestingly, ChatGPT performed significantly better in binary classification than in multi-classification. In Prompts 4.1–4.4, it achieved much higher success in distinguishing between no DR and all other stages of retinopathy, except for mild DR (Level 1), where it showed a 49.8% accuracy – similar to its accuracy for no DR in Prompts 1 through 3 (Table 2A).

Table 2 (A,B) Statistical Measurements of Prompts 4.1–4.4 and Prompt 5

When comparing ChatGPT-4o’s performance in detecting mtmDR to current FDA-approved AI systems, it showed much lower sensitivity and specificity. In Prompt 5, ChatGPT-4o’s overall sensitivity was 47.7% and its specificity 73.8%, both significantly lower than those of IDx-DR and EyeArt (Table 2B). The IDx-DR system achieved 87% sensitivity and 90% specificity, while the original EyeArt system demonstrated 96% sensitivity and 88% specificity. EyeArt v2.2.0 further improved with 94.4% sensitivity and 91.1% specificity (Figure 3).

Figure 3 Confusion matrix for Prompt 5. The model showed a strong performance in identifying moderate DR (Level 3) and proliferative DR (Level 4), with a significant number of true positives. The model correctly classified 385 images as moderate DR and 271 images as proliferative DR. However, the model severely underperforms in identifying severe DR, as reflected by the low number of true positives (41) and the large number of misclassifications, with many severe DR cases being mistaken for moderate DR or proliferative DR.

Discussion

Our findings highlight the significant potential of ChatGPT-4o in assisting with the classification of diabetic retinopathy (DR), especially in detecting more severe cases. In Prompts 4.1 through 4.4, GPT-4o showed increased accuracy, precision, sensitivity, and specificity when tasked with binary comparisons, particularly for proliferative DR, where values reached 75.6%, 92.2%, 56.4%, and 95.2%, respectively.

However, the model exhibits critical limitations, particularly in distinguishing between milder forms of the disease as seen in Prompts 1–5. In the context of AI-based DR screening, systems like IDx-DR and EyeArt have set a high benchmark for sensitivity and specificity. In a study using the Messidor-2 dataset, IDx-DR achieved a sensitivity of 96.8% and a specificity of 87.0% for detecting referable DR. In a retrospective analysis of 78,685 patient encounters, EyeArt achieved a sensitivity of 91.7% and a specificity of 91.5% for referable DR. EyeArt was also tested using smartphone-based fundus photography in a study of 296 patients, achieving a sensitivity of 95.8% for any DR, 99.3% for referable DR, and 99.1% for vision-threatening DR.²³

Unlike IDx-DR and EyeArt, which were designed specifically for DR screening and leverage training from extensive image datasets, ChatGPT-4o’s foundation as a LLM inherently limits its accuracy in image analysis. Even though IDx-DR and EyeArt excel at telling the severity of DR, the way they do so is not transparent because of deep learning, so it is hard to say if they depend on features like microaneurysms, hemorrhages and neovascularization. As a result, the model is more prone to underestimating the severity of retinopathy and experiences difficulty in handling subtle gradations between DR stages. This performance gap highlights a critical difference between general-purpose AI and specialized, clinically validated AI systems.²⁴ This limitation arises because GPT-4o, unlike convolutional neural networks (CNNs), is not trained to recognize pixel-level features such as microaneurysms or hemorrhages, which are essential for accurate DR diagnosis.

The lower upfront cost and accessibility of ChatGPT-4o present a compelling argument for its continued development as a potential screening tool. However, the trade-offs in accuracy would need to be addressed before it can be considered a suitable replacement. Looking forward, improving the model’s diagnostic accuracy for DR classification could involve integrating its language-based capabilities with specialized image recognition models like IDx-DR and EyeArt. Additionally, training ChatGPT-4o on large, diverse datasets of color fundus images may help improve its ability to recognize early-stage DR. Importantly, the dataset used in this study, sourced from Kaggle, is curated for image quality, which may not reflect the variability and imperfections found in real-world clinical data. Furthermore, the lack of accompanying clinical demographic information limits the model’s ability to contextualize findings, which is particularly relevant given that ChatGPT-4o is fundamentally a language model rather than a dedicated image analysis tool. Only basic prompt engineering strategies were used, suggesting that more sophisticated prompting techniques may yield better performance. Future work should also focus on refining the model’s ability to detect subtle changes between DR stages by using prompt engineering techniques that guide the model toward more precise image interpretation.

As ChatGPT becomes more and more accessible, its potential role in clinical settings raises important questions. Like how both physicians and patients increasingly rely on Google for medical information, ChatGPT’s ability to process complex data, including diagnostic images, offers the potential to become a valuable point-of-care resource. It could assist clinicians by providing immediate insights during consultations. Rather than simply serving as a screening tool, AI could shape how information is accessed and interpreted during clinical encounters, potentially affecting patient trust and the dynamics of the physician-patient relationship. While the model offers promise, its limitations must be carefully addressed before it can be integrated into routine clinical practice. Further studies are essential to continue to evaluate its performance as newer versions of the model are released and define the conditions for its safe use in healthcare.

Conclusion

In this paper, we examined the degree of accuracy in which ChatGPT-4 Omni can correctly grade and classify diabetic retinopathy fundoscopy, aiming to provide insight into low-cost and readily accessible alternatives to FDA-approved AI systems currently on the market that can assist in diagnosing diabetic retinopathy. This study analyzed 2,500 high-resolution color fundus photographs, revealing that ChatGPT-4 Omni achieved improved diagnostic metrics—accuracy (75.6%), precision (92.2%), sensitivity (56.4%), and specificity (95.2%)—in binary comparisons involving proliferative diabetic retinopathy. Ultimately, ChatGPT-4 Omni continues to show promise in one day becoming a diagnostic tool or support device in adequately evaluating for severe diabetic retinopathy, but several limitations and ethical considerations remain significant compared to FDA-approved AI systems.

Abbreviations

DR, Diabetic retinopathy; GPT-4o, ChatGPT-4 Omni; CFPs, color fundus photographs; AI, Artificial intelligence; FDA, Food and Drug Administration; mtmDR, mild DR; LLM, large language model; Acc, accuracy; Pre, precision; TPR, recall; Sen, sensitivity; Spe, specificity; PPV, positive predictive value; NPV, negative predictive value.

Acknowledgement

The abstract of this paper was published in “Poster Abstracts” in Investigative Ophthalmology and Visual Science: https://iovs.arvojournals.org/article.aspx?articleid=2805764#:~:text=Conclusions%20%3A%20While%20GPT%2D4o%20shows,clinical%20use%20in%20DR%20screening.

Funding

There is no funding to report.

Disclosure

The author(s) report no conflicts of interest in this work.

References

1. Fong DS, Aiello L, Gardner TW, et al. Retinopathy in diabetes. Diabetes Care. 2004;27(Suppl 1):S84–S87. doi:10.2337/diacare.27.2007.S84

2. Saaddine JB, Honeycutt AA, Narayan KM, Zhang X, Klein R, Boyle JP. Projection of diabetic retinopathy and other major eye diseases among people with diabetes mellitus: united States, 2005–2050. Arch Ophthalmol. 2008;126(12):1740–1747. doi:10.1001/archopht.126.12.1740

3. Wittenborn JS, Zhang X, Feagan CW, et al. The economic burden of vision loss and eye disorders among the United States population younger than 40 years. Ophthalmology. 2013;120(8):1728–1735. doi:10.1016/j.ophtha.2013.01.068

4. Hendricks LE, Hendricks RT. Greatest fears of type 1 and type 2 patients about having diabetes: implications for diabetes educators. Diabetes Educator. 1998;24(2):168–173. doi:10.1177/014572179802400206

5. Fong DS, Aiello L, Gardner TW, et al. Diabetic retinopathy. Diabetes Care. 2003;26(Suppl 1):S99–S102. doi:10.2337/diacare.26.2007.S99

6. Mazhar K, Varma R, Choudhury F, et al. Severity of diabetic retinopathy and health-related quality of life: the Los Angeles Latino Eye Study. Ophthalmology. 2011;118(4):649–655. doi:10.1016/j.ophtha.2010.08.003

7. Willis JR, Doan QV, Gleeson M, et al. Vision-related functional burden of diabetic retinopathy across severity levels in the United States. JAMA Ophthalmol. 2017;135(9):926–932. doi:10.1001/jamaophthalmol.2017.2553

8. Wong TY, Sun J, Kawasaki R, et al. Guidelines on diabetic eye care: the International Council of Ophthalmology recommendations for screening, follow-up, referral, and treatment based on resource settings. Ophthalmology. 2018;125(10):1608–1622. doi:10.1016/j.ophtha.2018.04.007

9. Javitt JC, Aiello LP. Cost-effectiveness of detecting and treating diabetic retinopathy. Ann Internal Med. 1996;124(1 Pt 2):164–169. doi:10.7326/0003-4819-124-1_Part_2-199601011-00017

10. Rohan TE, Frost CD, Wald NJ. Prevention of blindness by screening for diabetic retinopathy: a quantitative assessment. BMJ. 1989;299(6709):1198–1201. doi:10.1136/bmj.299.6709.1198

11. American Academy of Ophthalmology Retina/Vitreous Panel. Preferred practice pattern^® guidelines: diabetic retinopathy. San Francisco, CA: American Academy of Ophthalmology; 2019. Available from: www.aao.org/ppp. Accessed August 12, 2025.

12. Abràmoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digital Med. 2018;1(39). doi:10.1038/s41746-018-0040-6

13. Ipp E, Liljenquist D, Bode B, et al. Pivotal evaluation of an artificial intelligence system for autonomous detection of referrable and vision-threatening diabetic retinopathy [published correction appears in JAMA. Network Open. 2021;4(12):e2144317.

14. Eyenuk. New FDA clearance makes Eyenuk the first company with multiple cameras for autonomous AI detection of diabetic retinopathy; 2023. Available from https://www.eyenuk.com/us-en/articles/news/eyenuk-fda-multiple-cameras/. Accessed August 12, 2025.

15. Ziegelmayer S, Marka AW, Lenhart N, et al. Evaluation of GPT-4’s chest X-ray impression generation: a reader study on performance and perception. J Med Internet Res. 2023;25:e50865. doi:10.2196/50865

16. Ting DSJ, Tan TF, Ting DSW. ChatGPT in ophthalmology: the dawn of a new era? Eye. 2023;38(1):4–7. doi:10.1038/s41433-023-02619-4

17. Duarte F. Number of ChatGPT users (Aug 2024). Exploding Topics; 2024. Available from: https://explodingtopics.com/blog/chatgpt-users. Accessed August 12, 2025.

18. Dugas E, Jared G, Cukierski W. Diabetic retinopathy detection. Kaggle; 2015. Available from: https://www.kaggle.com/c/diabetic-retinopathy-detection. Accessed August 12, 2025.

19. Tariq M, Palade V, Ma Y. Transfer learning based classification of diabetic retinopathy on the Kaggle EyePACS dataset; 2023. doi:10.13140/RG.2.2.14506.29126.

20. Kung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE: potential for AI-assisted medical education using large language models. PLOS Digital Health. 2023;2(2):e0000198. doi:10.1371/journal.pdig.0000198

21. Wada A, Akashi T, Shih G, et al. Optimizing GPT-4 turbo diagnostic accuracy in neuroradiology through prompt engineering and confidence thresholds. Diagnostics. 2024;14(14):1541. doi:10.3390/diagnostics14141541

22. AlRyalat SA, Musleh AM, Kahook MY. Evaluating the strengths and limitations of multimodal ChatGPT-4 in detecting glaucoma using fundus images. Front Ophthalmol. 2024;4. doi:10.3389/fopht.2024.138719

23. Grzybowski A, Brona P, Lim G, et al. Artificial intelligence for diabetic retinopathy screening: a review. Eye. 2020;34(3):451–460. doi:10.1038/s41433-019-0566-0

24. Lim JI, Regillo CD, Sadda SR, et al. Artificial intelligence detection of diabetic retinopathy. Ophthalmol Sci. 2023;3(1):100228. doi:10.1016/j.xops.2022.100228

Introduction

Laser corneal refractive surgery (LCRS) is one of the most common surgical treatments used specifically for refractive correction.¹ Taking into account that patients who had LCRS have been aging, an increasing number of these subjects are confronted with presbyopia and/or cataract and are in demand to undergo refractive lens exchange or cataract surgeries. Considering that these patients were accustomed to be spectacle-independent after their LCRS, they demonstrate a strong interest not to use glasses at any distance (far, intermediate and near) after their intraocular surgery. In a demographic study of patient having cataract surgery after laser-insitu keratomileusis (LASIK), it has been concluded that patient age at the time of cataract surgery in post-LASIK patients was about 10 years younger than in axial length-matched patients, and about 15 years younger than in the whole population.²

After a Bayesian network meta-analysis comparing various monofocal and multifocal intraocular lenses (IOL) randomized clinical trials consider that trifocal IOL’s would be an optimal option to be spectacle-free.³ These lenses may achieve better intermediate visual acuity than patients implanted with bifocal IOLs⁴ and improved near visual acuity in comparison to enhanced depth-of-focus (EDOF) IOLs.⁵ The optics, design and shape of the IOL surface⁶ are the main differences in multifocal IOLs to correct presbyopia and actual IOLs must provide increased quality and optical performance⁷ to improve reading performance.⁸ Therefore, the use of trifocal IOLs in post-LCRS patients may offer an optimal solution to maintain spectacle independence, as shown in some previous studies.^9–16 These studies have analyzed eyes with previous photorefractive keratectomy (PRK), LASIK, or laser subepithelial keratomileusis (LASEK) surgeries implanted with different trifocal IOLs in a short follow-up (3–6 months). In addition, in our center we also see patients who have undergone radial keratotomy (RK) to correct their myopia: cataract/lens surgery is less predictable in these patients¹⁷ with hyperopic results due to the RK incisions’ flattening.¹⁸ Specifically, some studies and case reports have been published using bifocal, trifocals or EDOF IOLs with small samples and short follow-up.^19–23

Taking into account that the number of LASIK and PRK patients is growing and RK patients are also examined in our center, further investigations with these types of eyes implanted with trifocal IOLs, specifically with large samples and longer follow-ups, are required to provide clinical support to make evidence-based decisions in our clinical practice. So, the main objective of the current clinical study is to assess the visual and refractive outcomes in a cohort of eyes with previous LASIK, PRK or RK implanted with a trifocal diffractive IOL at 1 year of follow-up.

Patients and Methods

Study Design and Patients

This was a retrospective-single-center case series study enrolling eyes with previous LCRS (LASIK/PRK) or RK that had undergone cataract or refractive lens exchange (RLE) surgeries with a trifocal diffractive IOL implantation. 124 eyes of 62 patients at the Brussels Eye Doctors center (Belgium) between 2023 and 2024 were examined. The study was carried out in accordance with the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of the Brussels Eye Doctors Center. Informed consent was obtained from all the patients participating in this study. The study inclusion criteria were as follows: eyes implanted with a trifocal (spherical or toric model) after cataract or RLE surgeries with a previous LCRS (LASIK or PRK) or RK, in subjects older than 40 years, and who are interested not to use spectacles after the procedure. The exclusion criteria included keratoconus and any ocular disease (eg macular degeneration, cystoid edema) that may affect post-operative visual outcomes.

IOLs and Surgical Procedure

All the eyes were implanted with FineVision Micro F or POD F toric IOLs (BVI Inc., USA). All eyes underwent the phacoemulsification with the AMO-Signature device (J&J Vision Inc., USA) through a 1.9–2.2 mm temporal incision using topical anesthesia with IOL implantation in the capsular bag.

Pre-Operative and Post-Operative Measurements

All the eyes underwent a complete pre-operative ocular assessment that included slit-lamp and fundoscopic examinations, intraocular pressure measurement, subjective refraction measurement, monocular uncorrected-distance and corrected-distance visual acuity (UDVA and CDVA), and ocular biometry with an IOLMaster 700 (Carl Zeiss Meditec, Jena, Germany) to record ocular parameters for the lens calculation. Not the SRK-T, Haigis or Hoffer-Q formulas were used for IOL power calculations in both groups (LASIK/PRK and RK) but the ASCRS (American Society of Cataract and Refractive Surgery) calculator, with emmetropia target. In case of doubt a very slight myopia was preferred over a small hyperopia. In case of astigmatism, manifest refraction and the keratometry (as measured by the auto-refractometer and the IOL-master) were taken in account to determine whether a toric IOL needed to be implanted.

At 1-year after the surgery, we measured subjective manifest refraction (sphere, cylinder, and axis), keratometry, Snellen decimal monocular UDVA and CDVA, and uncorrected near visual acuity (UNVA) at 40 cm. Manifest refractions were converted to power vector coordinates.²⁴ Any complication during the surgery and follow-up related to the IOLs were also recorded. The % of eyes with Nd:YAG laser capsulotomy procedures after trifocal IOL implantation was recorded.

Analysis

The analysis considered data for demographics, refraction, and visual outcomes. It was conducted using Excel (version 16.43-, Microsoft-Corporation, USA), showing the mean, standard deviation and ranges. Different graphs for cataract surgery with an IOL were created.²⁵

Results

A total of 124 eyes of 62 patients were considered in this study, 51 patients in the LASIK/PRK group and 11 in the RK group. The mean age was 67.10±7.38 years in the LASIK/PRK group, and 66.65±5.33 years in the RK group. Both groups had undergone cataract or RLE surgeries with the Finevision IOLs implantation. Table 1 shows a summary of the demographics and pre-operative data of our sample as a function of the group.

Table 1 Demographics and Characteristics of Eyes Shown as Means, Standard Deviations (SD) and Ranges for the Two Groups of Patients Analyzed

Specifically, for the LASIK/PRK group 92 eyes had LASIK and 10 eyes had PRK. Sixty-seven eyes were implanted with the Finevision Micro F IOL and 35 with the FineVision POD F toric IOL. The mean spherical IOL power was 21.38 ± 2.57D and the mean cylindrical power was 1.36 ±0.47D. The mean time elapsed between the LCRS and IOL surgery was 8.22 years (from 2 to 27 years) and the mean SE before IOL surgery was −0.14 ± 2.16 D (from −7.00 to +3.50D). All patients completed 1-year of postoperative follow-up. There were no adverse events related to the IOL during the follow-up and Nd:YAG laser capsulotomy was performed in 3 eyes (2.94%). For the RK group 7 eyes were implanted with the FineVision Micro F IOL and 15 with the FineVision POD F toric IOL. The mean spherical IOL power was 23.77 ± 2.16D and the mean cylindrical power was 1.57 ± 0.55D. The mean time elapsed between the RK and the implantation of the IOL was 6 years (from 2 to 9 years) and the mean SE before IOL surgery was 1.28 ± 1.57D (from −2.50 to +5.50D). Note that these surgeries were done long time ago since RK was ceased to be carried out, and we included only these patients with a postoperative follow-up of 12 months. There were no complications related to the IOL during the follow-up and no Nd:YAG laser capsulotomy was performed.

Refractive Error

Figure 1 shows the refractive outcomes at 12 months in both groups for the whole sample and as a function of the refractive error before FineVision intraocular lens implantation. At this time, for the LASIK/PRK group (Figure 1A), 94 eyes (92.16%) were within ±0.50D and 101 eyes (99.02%) were within ±1.00D. The average SE obtained was −0.03 ± 0.30D (from −1.13 to +0.75D), the mean spherical refraction was 0.08 ± 0.31D (from −1.00D to +1.00D), and the mean cylinder was −0.21 ± 0.27D (from 0 to −1.00D). For the RK group (Figure 1B), 18 eyes (81.82%) were within ±0.50D and 22 (100%) were within ±1.00D. The mean SE obtained was −0.17 ± 0.38D (from −0.75 to +0.38D), the average spherical refraction was 0.06 ± 0.43D (from −0.75D to +1.00D), and the average cylinder was −0.45 ± 0.39D (from 0 to −1.25D). Specifically, for astigmatism, Figure 1C depicts the distribution of the refractive cylinder for the LASIK/PRK group. It shows 91.18% of the eyes (n = 93) with ≤0.50D, and all of them (n = 102) with ≤1.00D. It is interesting to point out that 74.51% (n = 76) of the eyes showed a cylinder ≤0.25D. Figure 1D shows the distribution for the RK group where 68.18% of the eyes (n = 15) with ≤0.50D, and 95.45% of eyes (n = 21) with ≤1.00D. 45.45% of eyes (n = 10) showed a value ≤0.25D. Figure 2 shows the attempted versus achieved graphs for SE (A, B) and for the J₀ (C, D) and J₄₅ (E, F) vectors of astigmatism in both groups. This figure also shows the astigmatic vectors before and 1 year after the surgery (G, H).

Figure 1 Distribution of postoperative spherical equivalent refraction 12 months after FineVision intraocular lens implantation in the LASIK/PRK (A) and RK (B) groups, and distribution of postoperative refractive cylinder 12 months after FineVision intraocular lenses implantation in the LASIK/PRK (C) and RK (D) groups. Outcomes are shown for the whole sample (black bars) and also as a function of the refractive error before intraocular lens implantation (green, red and white bars).

Figure 2 Attempted versus achieved spherical equivalent (M) (A and B) and the astigmatic J0 (C and D) and J45 (E and F) components of the power vector analysis after one year after surgery (G and H) in the LASIK/PRK and RK groups. Representation of the astigmatic vector (J0 and J45) before and one year after surgery (D). The scatterplot for J0 and J45 was calculated using the preoperative and postoperative refractive cylinder. Note that the coordinates 0:0 represent an eye free of astigmatism.

Visual Outcomes

The mean postoperative monocular decimal UDVA, CDVA and UNVA were 1.04±0.19, 1.14±0.20 and 0.99±0.04, respectively, for the LASIK/PRK group. These values changed to 0.90±0.10, 1.01±0.08 and 1.00±0.00, respectively, for the RK group. Figure 3 depicts the difference in monocular UDVA and CDVA after the surgery in both groups; 59.80% of the eyes (n = 61) showed a UDVA that was ≥CDVA, and 65.69% (n = 67) of the eyes had an UDVA within one line of the CDVA for the LASIK/PRK group. These values were 40.91% (n = 9) and 59.09% (n = 13) respectively for the RK group.

Figure 3 Difference in monocular uncorrected distance visual acuity (UDVA) and best corrected distance visual acuity (CDVA) 12 month after FineVision intraocular lens implantation in the LASIK/PRK (A) and RK (B) groups. Outcomes are shown for the whole sample (black bars) and also as a function of the refractive error before intraocular lens implantation (green, red and white bars).

Figure 4 depicts the cumulative percentage of eyes with a given UDVA, CDVA and UNVA in the LASIK/PRK group 12 months after the surgery. For the whole sample, 72.55% (n = 74), 88.24% (n = 90), and 98.04% (n = 100) of the eyes had ≥20/20 UDVA, CDVA and UNVA, respectively, with 96.08% (n = 98), 100% (n = 102) and 98.04% (n = 100) of the eyes achieving ≥20/25 for UDVA, CDVA and UNVA, respectively. Figure 5 illustrates the cumulative percentage of eyes with a given postoperative UDVA, CDVA and UNVA in the RK group 12 months after the surgery. For the whole sample, 45.55% (n = 10), 83.36% (n = 19), and 100% (n = 22) of the eyes had ≥20/20 UDVA, CDVA and UNVA, respectively, with 90.91% (n = 20), 100% (n = 22) and 100% (n = 22) of the eyes achieving ≥20/25 for UDVA, CDVA and UNVA, respectively.

Figure 4 Cumulative proportion of eyes 12 months after FineVision intraocular lens implantation in the LASIK/PRK group with a given postoperative uncorrected and distance corrected visual acuity (UDVA and CDVA), and uncorrected near visual acuity (UNVA) at 40 cm. Outcomes are shown for the whole sample (D) and also as a function of the refractive error before intraocular lens implantation ((A): hyperopic, (B): emmetropic and (C): myopic).

Figure 5 Cumulative proportion of eyes 12 months after FineVision intraocular lens implantation in the RK group with a given postoperative uncorrected and distance corrected visual acuity (UDVA and CDVA), and uncorrected near visual acuity (UNVA) at 40 cm. Outcomes are shown for the whole sample (C) and also as a function of the refractive error before intraocular lens implantation ((A): hyperopic and (B) myopic).

Discussion

Due to the increasing demand from LCRS or RK patients to continue to be spectacle independent, whether they are affected by presbyopia or cataract, clinical investigations are required to provide surgical evidence of the use of trifocal IOLs when performing RLE or cataract surgery. It should be considered that the refractive accuracy after trifocal IOL implantation in LCRS or RK patients is very important due to a high expectation of spectacle independence after the surgery. Our study shows the clinical outcomes obtained in a cohort of eyes with previous LCRS or RK implanted with trifocal diffractive IOLs. The outcomes we have obtained demonstrate that the FineVision IOLs provide good vision at different distances with excellent refractive accuracy.

Trifocal IOL Implantation After LASIK/PRK

Specifically, we have found excellent refractive error accuracy after IOL implantation using essentially the ASCRS calculator. Note that 92.16% of eyes were within ±0.50D and 99.02% of eyes were within ±1.00D (see Figure 1A, being the postoperative mean SE close to emmetropia (−0.03 ± 0.30D) and a mean cylinder of −0.21±0.27D, with 91.18% of eyes ≤0.50D: Figure 1C). The attempted versus achieved plots also revealed the good outcomes obtained (Figure 2). Note that the coordinates 0:0 represent an eye without astigmatism (origin of the graph in Figure 2G) and, therefore, concentrated data about these coordinates after the surgery (black dots) indicate better correction of astigmatism. The accuracy of the surgeries correlates with the visual acuity outcomes, obtaining mean values ≥20/20 at far and near with cumulative percentages for ≥20/25 or better being close to 100% at all distances (see Figure 4).

As already mentioned, previous trials have published the outcomes of different trifocal IOL models in LCRS patients.^9–16 Table 2 compiles the main characteristics of these publications. Brenner et al⁹ have used the same lenses as us, with a mean follow-up of 6.38 months in 155 myopic and 86 hyperopic ablation eyes after RLE surgery. They used the ASCRS calculator in some eyes and later an optimized nomogram. They found a mean post-surgery SE of −0.25±0.38D and −0.02 ± 0.42D for the myopic and hyperopic ablation groups. 80% of eyes were ±0.50D and 97.4% ±1.00D for the myopic group and 82.6% ±0.50D and 98.8% ±1.00D for the hyperopic group. The numbers for the whole sample were 80.9% eyes ±0.50D and 97.9% eyes ±1.00D for the SE, and 51.4% of eyes within 0.25D and 80.0% of eyes within 0.50D for refractive astigmatism. For visual acuity outcomes, the mean UDVA (Snellen), CDVA (Snellen) and UNVA (point type) were 0.88 ± 0.20, 1.06 ± 0.10 and 5.11 ± 0.46, for the myopic group, respectively, and 0.85 ± 0.19, 1.03 ± 0.10, and 5.25 ± 0.75 for the hyperopic group, respectively. The entire cohort showed 47%, 81.3% and 98.8% of eyes with a cumulative UDVA ≥20/20, ≥20/25 and ≥20/20, respectively. Of the eyes, 79.7% lost no lines of CDVA from the preoperative stage, 14.5% lost one line, 0.4% lost two lines, and 4.6% gained one line. The safety and efficacy indices found by these authors were 0.97±0.08 and 0.80±0.18 for the myopic eyes, and 0.98±0.09 and 0.82±0.17 for the hyperopic eyes. These authors also informed that 15% of eyes had refractive enhancement with 12% surface ablation and 3% with supplementary IOL after presbyopic RLE. They concluded that this procedure in LCRS eyes was safe and effective and that the ASCRS on-line-calculator was a valuable instrument for calculation of the IOL power. Our results broadly agree with them.

Table 2 Clinical Studies Published Reporting Refractive and Visual Outcomes Following Trifocal Intraocular Lens (IOL) Implantation in Laser Corneal Refractive Surgery Eyes (LCRS)

In another study, Chow et al¹⁰ and Li et al¹¹ analysed the performance of the AT LISA tri 839 MP IOL in a small sample of myopic eyes after cataract surgery, 20 and 21, respectively. Chow et al¹⁰ found a mean postoperative UDVA of 0.28±0.29 logMAR, a mean CDVA of 0.06±0.14 logMAR, and a mean UNVA of 0.02±0.05 logMAR at 6 months. The mean SE was −0.92±0.76D with 55% of eyes achieving a SE ±0.50D of the targeted refraction (note that in this case it was −0.50 to −1.00D in the preoperative stage by means of the Holladay 2 formula). Li et al¹¹ with a similar sample of myopic eyes but in a short follow-up (3 months) and using a multi-formula average method (Barrett True K, Haigis-L, ray-tracing and Shammas No-History, target emmetropia), obtained a mean SE of −0.56±0.49D (47.6% within ±0.50D and 90.5% within ±1.00D). For monocular conditions, they obtained average logMAR UDVA and UNVA of 0.02±0.07 and 0.15±0.11, respectively. 100% of eyes achieved a postoperative CDVA ≥20/25, while the percentage for a UDVA of ≥20/25 was 76%. These authors are in agreement with the previous study.

Cobo-Soriano et al¹² analyzed the largest sample of myopic (n = 319) and hyperopic (549) eyes using the FineVision Micro F, FineVision POD F and AT LISA tri 839 MP IOLs at 3 months after the RLE or cataract surgeries using the ASCRS calculator with the target of emmetropia. Their results showed a postoperative mean SE of −0.38 ± 0.30D and −0.17 ± 0.30D for the myopic and hyperopic groups, respectively. 63.5% and 73.5% of eyes were within ±0.50D, respectively. In relation to visual acuities, monocular logMAR UDVA, CDVA and UNVA were 0.09 ± 0.08, 0.03 ± 0.04 and 0.15 ± 0.14 for the myopic group, and 0.1 ± 0.08, 0.06 ± 0.05 and 0.16 ± 0.12 for the hyperopic group, respectively. The outcomes reported by these authors disagree with those found by Brenner et al⁹ since Cobo-Soriano et al¹² show loss of CDVA and worse safety outcomes after the surgery in hyperopic eyes. They argued that these differences may come from the differences in the change in CDVA lines. Finally, they concluded that implantating this IOL was compatible with previous LCRS. We support their findings considering that the use of trifocal IOLs can be used in LCRS eyes.

Blaylock and Hall¹³ analyzed the outcomes using another trifocal model, the AcrySof IQ PanOptix IOL, in 25 myopic eyes (15 LASIK and 20 PRK) submitted to RLE and cataract surgeries at 3 months. They found a mean SE of 0.03±0.45D, with 76% of eyes within ±0.50D and 97% within ±1.00D of SE, and 91% and 100% of eyes with refractive astigmatism of ≤0.50D and ≤1.00D, respectively. In relation to visual acuity, 100% of eyes had monocular visual acuities of ≥20/40 at all distances, with the percentage of eyes ≥20/20 for UCVA, CDVA, and UNVA being 28.6%, 77.1%, and 65.6%, respectively. These values changed to 74.3%, 97.1% and 90.6% for ≥20/25, respectively. The mean monocular UCVA, CDVA and UNVA were 0.09 ± 0.08, 0.02 ± 0.05 and 0.05 ± 0.10, logMAR respectively. These authors also analyzed the possible differences between preoperative planning or the use of intraoperative aberrometry, reporting no significant differences in the absolute prediction error (P > 0.05) but less postoperative residual astigmatism (P < 0.002).

Mayordomo-Cerdá et al¹⁴ assessed the outcomes using the Finevision Micro F and AT LISA 839 MP IOL in 89 myopic and 97 hyperopic eyes after LASIK and LASEK procedures. These eyes were submitted to RLE or cataract surgeries and evaluated at 3 months – using the ASCRS calculator with the Barrett True K for IOL power calculation with the target of emmetropia. Specifically, all the eyes analyzed in this study were submitted to an enhancement. The outcomes before the enhancement reported a mean SE of −0.93 ± 0.29D and −0.69 ± 0.49D in the myopic and hyperopic groups. The mean monocular UDVA, CDVA and UNVA were 0.31 ± 0.14, 0.05 and 0.10 logMAR for the myopic group, and 0.28 ± 0.20, 0.06, and 0.18 logMAR for the hyperopic group, respectively. These authors also reported detailed outcomes of this cohort after the enhancement, with PRK (72.1%) or flap lift (28.0%). In another study, the same group of authors aimed to assess whether the use of the Rayone trifocal IOL (neutral spherical aberration) provided better outcomes than the FineVision POD F IOL (negative spherical aberration) in 198 hyperopic eyes with previous LCRS after RLE or cataract surgeries. They used the ASCRS calculator with the Barrett True K for IOL power calculation with the target of emmetropia. One hundred and twenty eyes were implanted with the FineVision POD F IOL and 78 eyes with the RayOne IOL. After the surgery the average SE was −0.01 ± 0.38D for the FineVision POD F and −0.34 ± 0.51D for the RayOne IOL (the differences were significant, P < 0.001). However, no significant differences were found for visual acuities, with the monocular logMAR UDVA standing at 0.10 and 0.07 (P = 0.647), the monocular logMAR CDVA at 0.05 and 0.05 (P = 0.343), and the monocular UNVA at 0.18 and 0.18 (P = 0.382) for the FineVision POD F and RayOne IOLs, respectively.

Bilbao-Calabuig et al¹⁶ studied 211 myopic eyes previously submitted to LASIK surgery after the FineVision Micro F and POD F IOLs implantation. These authors also used the ASCRS calculator with the target refraction of emmetropia. After RLE or cataract surgeries, the average SE and cylinder were −0.29±0.29D and −0.30±0.31D, respectively (65% of eyes within ±0.50D and 87% within ±1.00D from emmetropia). The average monocular logMAR UDVA, CDVA and UNVA were 0.09 ± 0.07, 0.04 ± 0.04 and 0.14 ± 0.07, respectively. The corneal laser enhancement after RLE or cataract surgeries was 15.7%.

Our outcomes are in agreement with other clinical studies, using the Finevision lens and other models, and support the use of these lenses in LCRS eyes when performing cataract or RLE surgeriesin order to avoid the use of glasses for any distance.

Trifocal IOL Implantation After RK

Focusing now on the RK group refractive accuracy – despite the fact that we as well used the ASCRS calculator to calculate the IOL power – our results were not so good as those obtained in the LASIK/PRK group. This is probably bound to the corneal irregularity created by the incisions and, consequently, the higher optical aberrations that affect the eye after the procedure. In our sample, 81.82% eyes were within ±0.50D and 100% were within ±1.00D (Figure 1B), with the mean SE of −0.17 ± 0.38D but the mean refractive cylinder close to a half diopter (−0.45 ± 0.39D). Note that only 68.18% of the eyes showed residual astigmatism of ≤0.50D (Figure 1D). These seem not to affect the visual acuity outcomes since the mean UDVA, CDVA and UNVA were close to 20/20 and 90.91%, and 100% of eyes achieved ≥20/25 or better for UDVA, CDVA and UNVA, respectively (Figure 5C).

As we mentioned at the beginning of this article, previous publications have shown the performance of different presbyopia-correcting IOLs in RK eyes.^19–23 Table 3 illustrates the main characteristics of these publications indicating the type of IOL implanted. Gupta et al¹⁹ showed the results for two eyes with the bifocal AcrySof IQ ReSTOR IOLs (two patients with monovision), concluding that this procedure was a good solution for subjects aiming to be spectacle-free. Monocular CDVA was 20/20 in 3 eyes and distance corrected near visual acuity (DCNVA) was 20/20 in both patients.

Table 3 Clinical Studies Published Reporting Refractive and Visual Outcomes Following Presbyopia-Correcting Intraocular Lens (IOL) Implantation in Radial Keratotomy Eyes

Kim et al²⁰ concluded that the use of a rotationally asymmetric refractive multifocal lens (2 eyes from 2 unilateral RK subjects) may benefit presbyopic patients. They reported an UDVA of 20/20 and an UNVA of J1 for both patients.

And Nuzzi et al²¹ considered that the implantation of a customized-toric multifocal lens in an eye with previous cross-linking leads to good visual outcomes without regressions.

In contrast to these studies, Martin-Escuer et al²² concluded that the use of a multifocal lens in RK eyes did not result in good distance visual performance. They reported the outcomes using different presbyopia-correcting IOLs in a sample of 17 eyes. The UDVA, DCVA, and DCNVA for monocular conditions were 0.51±0.39, 0.20±0.30, and 0.11±0.11 logMAR, respectively. 35.29% of the eyes had DCVA of ≥20/20 and 52.94% showed DCVA of ≥20/25. 52.94% lost 1-line of DCVA, 23.53% showed no changes, 11.76% gained 1-line of DCVA, 5.88% gained 2-lines, and 5.88% gained ≥3-lines. 29.41% of the eyes had DCNVA of ≥20/20 and 64.71% had DCNVA of ≥20/25. 29% of the eyes were within ±0.50D, whereas 65% were within ±1.00D.

This study and the current one are the only two studies that have been published to date that use trifocal IOLs in RK eyes, and our results support the use of the FineVision model in these eyes. Note that the number of RK cuts, formula and target refraction may affect the procedure and hence the outcomes obtained.

In another study, Agarwal and Thornell²³ assessed spectacle independence in 3 patients with RK implanted unilaterally with the IC-8 IOL pinhole, and one patient with RK and LASIK with bilateral use of the AT LISA 939M toric IOL, and concluded that both lenses offer satisfactory outcomes reducing the use of spectacles.

It should be considered that the corneal surface in these eyes is altered, showing irregularities that increase optical aberrations²⁶ by creating a “multifocal lens effect”²⁷ and when this is combined with diffractive IOL designs, the visual performance may be reduced. Also, the central corneal flattening with irregularities created by the radial incisions and diurnal variations over time²⁸ make it difficult to obtain accurate keratometric measurements. These factors may play a considerable role when the IOL power is calculated and therefore affect the accuracy of the procedure and the visual outcomes obtained. Note that residual refractive errors after this procedure can be corrected with corneal topography-guided laser treatments and that may improve the optical quality and contrast sensitivity function²⁹ in these eyes.^30,31 Possibly, the combination of several corneal treatments with diffractive IOLs may reduce visual performance under low-light conditions.³²

Capsulotomy

In relation to Nd:YAG laser capsulotomy, only 2 studies of the same group reported their outcomes. Cobo-Soriano et al¹² reported this procedure in 21% of hyperopes and 13.7% in the myopes with the 3 types of IOL implanted. In relation to the IOL type, they reported that the AT LISA tri 839MP lens had a higher rate of capsulotomies (37%) compared to the Finevision Micro F (11%) and POD F (20%) lenses. Notwithstanding, the time from IOL surgery to the capsulotomy was shorter in both Finevision lenses compared to that in the AT LISA lens (14 versus 24 months). The percentages found are in agreement with those previously reported showing higher values with the AT LISA tri 839MP lens than those found by the Finevision lenses.³³ Mayordomo-Cerdá et al¹⁴ reported similar percentages, begin 16.5% in the hyperopes and 9.0% in the myopes for the Finevision Micro F and AT LISA tri 839 MP models. Our results indicated that Nd:YAG laser capsulotomy was carried out in 2.94% of eyes (n = 3).

Conclusion

The present study demonstrated that use of a diffractive trifocal lens can provide good clinical outcomes in eyes previously submitted to LCRS or RK. The use of this type of IOLs is a reasonable choice in our patients.

Disclosure

The authors report no conflicts of interest in this work.

References

1. Moshirfar M, Bennet P, Ronquillo Y. Laser in Situ Keratomileusis. Treasure Island: StatPearls Publishing; 2021.

2. Iijima K, Kamiya K, Shimizu K, Igarashi A, Komatsu M. Demographics of patients having cataract surgery after laser in situ keratomileusis. J Cataract Refract Surg. 2015;41(2):334–338. doi:10.1016/j.jcrs.2014.05.045

3. Cho JY, Won YK, Park J, et al. Visual outcomes and optical quality of accommodative, multifocal, extended depth-of-focus, and monofocal intraocular lenses in presbyopia-correcting cataract surgery: a systematic review and bayesian network meta-analysis. JAMA Ophthalmol. 2022;140(11):1045–1053. doi:10.1001/jamaophthalmol.2022.3667

4. Shen Z, Lin Y, Zhu Y, Liu X, Yan J, Yao K. Clinical comparison of patient outcomes following implantation of trifocal or bifocal intraocular lenses: a systematic review and meta-analysis. Sci Rep. 2017;7:45337. doi:10.1038/srep45337

5. Karam M, Alkhowaiter N, Alkhabbaz A, et al. Extended depth of focus versus trifocal for intraocular lens implantation: an updated systematic review and meta-analysis. Am J Ophthalmol. 2023;251:52–70. doi:10.1016/j.ajo.2023.01.024

6. Fernández-Vega L, Alfonso JF, Baamonde B, Madrid-Costa D, Montés-Micó R, Lozano J. Visual and refractive outcomes in hyperopic pseudophakic patients implanted with the Acri.LISA 366D multifocal intraocular lens. Am J Ophthalmol. 2009;148(2):214–220.e1. doi:10.1016/j.ajo.2009.02.036

7. Ortiz D, Alió JL, Bernabéu G, Pongo V. Optical performance of monofocal and multifocal intraocular lenses in the human eye. J Cataract Refract Surg. 2008;34(5):755–762. doi:10.1016/j.jcrs.2007.12.038

8. Alió JL, Plaza-Puche AB, Piñero DP, et al. Optical analysis, reading performance, and quality-of-life evaluation after implantation of a diffractive multifocal intraocular lens. J Cataract Refract Surg. 2011;37(1):27–37. doi:10.1016/j.jcrs.2010.07.035

9. Brenner LF, Gjerdrum B, Aakre BM, Lundmark PO, Nistad K. Presbyopic refractive lens exchange with trifocal intraocular lens implantation after corneal laser vision correction: refractive results and biometry analysis. J Cataract Refract Surg. 2019;45(10):1404–1415. doi:10.1016/j.jcrs.2019.05.031

10. Chow SSW, Chan TCY, Ng ALK, Kwok AKH. Outcomes of presbyopia-correcting intraocular lenses after laser in situ keratomileusis. Int Ophthalmol. 2019;39:1199–1204. doi:10.1007/s10792-018-0908-0

11. Li QM, Wang F, Wu ZM, et al. Trifocal diffractive intraocular lens implantation in patients after previous corneal refractive laser surgery for myopia. BMC Ophthalmol. 2020;20:293. doi:10.1186/s12886-020-01556-0

12. Cobo-Soriano R, Ortega-Usobiaga J, Rodriguez-Gutierrez B, et al. Trifocal intraocular lens implantation in eyes with previous corneal refractive surgery for myopia and hyperopia. J Cataract Refract Surg. 2021;47:1265–1272. doi:10.1097/j.jcrs.0000000000000637

13. Blaylock JF, Hall BJ. Refractive outcomes following trifocal intraocular lens implantation in post-myopic LASIK and PRK eyes. Clin Ophthalmol. 2022;16:2129–2136. doi:10.2147/OPTH.S370061

14. Mayordomo-Cerdá F, Ortega-Usobiaga J, Bilbao-Calabuig R, et al. Laser corneal enhancement after trifocal intraocular lens implantation in eyes that previously had photoablative corneal refractive surgery. J Cataract Refract Surg. 2022;48(7):790–798. doi:10.1097/j.jcrs.0000000000000847

15. Mayordomo-Cerdá F, Ortega-Usobiaga J, Baviera-Sabater J, et al. Visual and refractive outcomes after implantation of two models of trifocal intraocular lenses in eyes with previous corneal ablation to treat hyperopia. Eye Vis. 2023;10(1):48. doi:10.1186/s40662-023-00366-x

16. Bilbao-Calabuig R, Ortega-Usobiaga J, Mayordomo-Cerdá F, Beltrán-Sanz J, Fernández-García J, Cobo-Soriano R. Trifocal versus monofocal intraocular lens implantation in eyes previously treated with laser in situ keratomileusis (LASIK) for myopia. Indian J Ophthalmol. 2024;72(Suppl 2):S254–S259. doi:10.4103/IJO.IJO_1844_23

17. Yeu E, Cuozzo S. Matching the patient to the intraocular lens: preoperative considerations to optimize surgical outcomes. Ophthalmology. 2021;128(11):e132–e141. doi:10.1016/j.ophtha.2020.08.025

18. Lyle WA, Jin GJ. Intraocular lens power prediction in patients who undergo cataract surgery following previous radial keratotomy. Arch Ophthalmol. 1997;115(4):457e461. doi:10.1001/archopht.1997.01100150459001

19. Gupta I, Oakey Z, Ahmed F, Ambati BK. Spectacle independence after cataract extraction in post-radial keratotomy patients using hybrid monovision with ReSTOR(^®) multifocal and TECNIS(^®) monofocal intraocular lenses. Case Rep Ophthalmol. 2014;5(2):157–161. doi:10.1159/000363372

20. Kim KH, Seok KW, Kim WS. Multifocal Intraocular Lens Results in Correcting Presbyopia in Eyes After Radial Keratotomy. Eye Contact Lens. 2017;43(6):e22–e25. doi:10.1097/ICL.0000000000000208

21. Nuzzi R, Monteu F, Tridico F. Implantation of a multifocal toric intraocular lens after radial keratotomy and cross-linking with hyperopia and astigmatism residues: a case report. Case Rep Ophthalmol. 2017;8:440–445. doi:10.1159/000479813

22. Martín-Escuer B, Alfonso JF, Fernández-Vega-Cueto L, Domíngez-Vicent A, Montés-Micó R. Refractive correction with multifocal intraocular lenses after radial keratotomy. Eye (Lond). 2019;33(6):1000–1007. doi:10.1038/s41433-019-0364-8

23. Agarwal S, Thornell E. Spectacle independence in patients with prior radial keratotomy following cataract surgery: a case series. Int Med Case Rep J. 2020;13:53–60. doi:10.2147/IMCRJ.S230863

24. Thibos LN, Horner D. Power vector analysis of the optical outcome of refractive surgery. J Cataract Refract Surg. 2001;27:80–85. doi:10.1016/S0886-3350(00)00797-5

25. Reinstein DZ, Archer TJ, Srinivasan S, et al. Standard for reporting refractive outcomes of intraocular lens-based refractive surgery. J Refract Surg. 2017;33(4):218–222. doi:10.3928/1081597X-20170302-01

26. Applegate RA, Howland HC, Sharp RP, Cottingham AJ, Yee RW. Corneal aberrations and visual performance after radial keratotomy. J Refract Surg. 1998;14:397–407. doi:10.3928/1081-597X-19980701-05

27. Maguire LJ, Bourne WM. A multifocal lens effect as a complication of radial keratotomy. Refract Corneal Surg. 1989;5:394–399. doi:10.3928/1081-597X-19891101-09

28. Kemp JR, Martinez CE, Klyce SD, et al. Diurnal fluctuations in corneal topography 10 years after radial keratotomy in the prospective evaluation of radial keratotomy study. J Cataract Refract Surg. 1999;25:904–910. doi:10.1016/S0886-3350(99)00090-5

29. Anera RG, Villa C, Jiménez JR, Gutierrez R. Effect of LASIK and contact lens corneal refractive therapy on higher order aberrations and contrast sensitivity function. J Refract Surg. 2009;25(3):277–284. doi:10.3928/1081597X-20090301-07

30. Queirós A, Villa-Collar C, Gutiérrez ÁR, et al. Anterior and posterior corneal elevation after orthokeratology and standard and customized LASIK surgery. Eye Contact Lens. 2011;37(6):354–358. doi:10.1097/ICL.0b013e318232e32d

31. González-Pérez J, Villa-Collar C, González-Méijome JM, Porta NG, Parafita MÁ. Long-term changes in corneal structure and tear inflammatory mediators after orthokeratology and LASIK. Invest Ophthalmol Vis Sci. 2012;53(9):5301–5311. doi:10.1167/iovs.11-9155

32. Alfonso JF, Fernández-Vega L, Baamonde B, Madrid-Costa D, Montés-Micó R. Visual quality after diffractive intraocular lens implantation in eyes with previous hyperopic laser in situ keratomileusis. J Cataract Refract Surg. 2011;37(6):1090–1096. doi:10.1016/j.jcrs.2010.11.043

33. Bilbao-Calabuig R, Llovet-Osuna F, González-López F, Beltrán J. Nd:YAG capsulotomy rates with two trifocal intraocular lenses. J Refract Surg. 2016;32(11):748–752. doi:10.3928/1081597X-20160803-02

Introduction

Gastric content reflux and aspiration are among the most threatening complications during the perioperative period, and their occurrence is closely related to various pathological and physiological factors. In clinical practice, dysfunction of the gastroesophageal sphincter (such as pregnancy, obesity, diabetic gastroparesis), emergency trauma (delayed gastric emptying caused by pain/opioids), and special surgical types (gastrointestinal endoscopy) can significantly increase the risk of pulmonary aspiration. Respiratory events during anesthesia are associated with death and brain damage,^1,2 with aspiration of gastric contents being the main cause of respiratory-related death and claims.^1,3 A retrospective study indicates that the incidence of pulmonary aspiration was 1:8,325. Ten patients required intensive care, respiratory support, or developed pulmonary complications.⁴ When gastric content escapes the airway protective mechanism, it causes a triple-hit effect on the lungs: acidic substances directly damage the alveolar epithelium, food particles cause mechanical obstruction, and bacterial colonization leads to secondary infection.⁵ Gastric regurgitation and aspiration events, which should be emphasized the most, are most common during emergency surgery, bronchoscopy, endoscopy, although they can also be seen in elective cases.^6–8

The incidence of pulmonary aspiratory has been reduced through the adoption of reasonable measures such as fasting, the use of gastric point-of-care ultrasound (POCUS), rapid sequence induction (RSI), modern technical aids such as high-flow preoxygenation or video-laryngoscopy, and the identification and prevention of patients at high risk of aspiratory.^9–14 However, it remains a significant and unavoidable issue in clinical practice.³ In addition, the specific implementation of measures to prevent pulmonary aspiration is controversial,^15–18 and there are differences among doctors in different regions and levels.^19,20 Recent research on gastric content regurgitation and pulmonary aspiration has primarily focused on case-based retrospective observational studies^{1,4,6,7,21,22} At present, the research in this field lacks large-scale investigation studies that reflect real-world practices. This study is the first to adopt a nationwide cross-sectional survey method to systematically evaluate the cognitive level and practice patterns of Chinese anesthesiologists regarding the risk of perioperative aspiration. By revealing the gap between clinical practice and guidelines, identifying key controversial points, it provides data support for formulating more operational preventive strategies.

Methods

Definition of Gastric Regurgitation and Pulmonary Aspiration

Perioperative pulmonary aspiration is defined as aspiration of gastric contents occurring after induction of anesthesia, during a procedure, or in the immediate postoperative period. Gastric regurgitation is defined as the presence of reflux of gastric contents without pulmonary aspiration.^4,23

Study Design and Participants

This survey was approved by the Ethics Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine (0159/2024). Using the WJX platform (Changsha Ranxing IT Ltd., Changsha, China), which is compatible with both WeChat and the New Youth Anesthesia Forum website. The survey comprised 26 questions in Chinese, and participants could access it by clicking a link or scanning a QR code. The New Youth Anesthesia Forum, the largest anesthesia network platform in China, and WeChat, the most widely used Chinese multi-functional instant messaging and social media application in China, were utilized for distribution. The questionnaire link was posted in the WeChat group of the Chinese Association of Anesthesiologists. Additionally, with approval from the New Youth Anesthesia Forum Committee, the link was shared on the New Youth Anesthesia Forum website and promoted to its registered anesthesiologist members. Consent to participate was obtained at the beginning of the questionnaire, with non-consenting individuals being excluded. Each participant was allowed to submit the survey only once. The data collected were managed anonymously. The online survey was available for 30 days, from February 17, 2024, to March 17, 2024.

This questionnaire consists of 5 sections, with a total of 26 items. Part 1 (demographic information of the respondents – gender, region of practice, hospital grade, teaching hospital, current practice category, physician title, years of practice), part 2 (perioperative regurgitation and aspiration), part 3 (personal experiences), part 4 (prevention and treatment measures), part 5 (resource allocation) and part 6 (gastric ultrasound use and training). The full questionnaire is presented in Appendix 1.

Sample Size Estimation

Following the recommendations of Kotrlik et al,²⁴ the formula n=Z²*p*(1-p)/d² is considered to be more suitable for sample size calculation in cross-sectional surveys. Given that there are few studies on perioperative gastric regurgitation and pulmonary aspiration, the P-value was 0.5, the confidence level was 95% (corresponding Z value was 1.96), and the acceptable error range was 5%. Therefore, the required sample size was 384. Assuming a response rate of 30%, we end up with a minimum sample size of 1280.

Statistical Analysis

Upon completion of the survey, the data were compiled into an Excel spreadsheet. Frequencies and percentages were used to express count data. All statistical analyses were performed using R version 4.0.5. Categorical variables (training on the management of perioperative regurgitation and aspiratory, training on gastric ultrasound, gender, hospital grade, teaching hospital status, physician title, and years of practice categorized into groups) were summarized as frequencies (n) and percentages (%). The Pearson chi-squared (χ²) test was used to assess differences in the distribution of training on the management of perioperative regurgitation and aspiratory and gastric ultrasound across stratification variables (gender, hospital grade, teaching hospital status, physician title, and years of practice). To evaluate the strength of association, bias-corrected Cramér’s V was derived from the chi-square statistic and interpreted according to conventional thresholds (V < 0.10 can be ignored; 0.10–0.20 is weak; 0.20–0.40 is moderate; > 0.40 is strong). All hypothesis tests were two-sided, and a P value < 0.05 was considered statistically significant.

Results

Demographics

A total of 3632 questionnaires were collected. Among the respondents, 8 disagreed to have their data used for analysis. Additionally, 21 intensive care unit physicians, 5 emergency department physicians, 77 surgeons, 15 internists, and 16 other specialists were excluded. This left 3,490 questionnaires for data analysis, consisting of 3,444 anesthesiologists and 46 anesthesia nurses. The respondents represented 31 regions across China, excluding Taiwan, Hong Kong, and Macao (Figure 1). Of these, 132 (3.8%) did not specify their region of practice. The demographic data of the respondents are presented in Table 1.

Table 1 Respondent’s Demographic Data

Figure 1 Geographic distribution of the respondents.

Personal Experience with Gastric Regurgitation and Pulmonary Aspiration

Among the 3,490 respondents, 2,470 (70.77%) reported experiencing gastric regurgitation or pulmonary aspiration personally, 874 (25.04%) had not experienced it themselves but had colleagues who did, and 146 (4.18%) stated that they had not encountered such cases (Figure 2A). A total of 3,344 (95.82%) respondents answered open-ended questions about their experience of gastric regurgitation or pulmonary aspiration during anesthesia. Of these, 1,677 (50.15%) experienced pulmonary aspiration, 941 (28.14%) reported only gastric regurgitation without pulmonary aspiration, and 726 (21.71%) encountered both situations (Figure 2B).

Figure 2 (A) Respondents’ clinical experience with gastric regurgitation and pulmonary aspiration. (B) Percentages of gastric regurgitation (without aspiration) and pulmonary aspiration. (C) Outcomes of patients with pulmonary aspiration. Abbreviations: PA, pulmonary aspiration; R, regurgitation (without aspiration).

Emergency surgery was the most common context for these events (86.39%), followed by painless endoscopy (41.78%) and elective surgery (20.16%). We investigated the phases during which these events occurred, the ventilation modes used, and the operative sites involved. Regurgitation and aspiration can occur at any stage of the anesthesia process, most commonly during the induction of general anesthesia (75.33%), followed by painless endoscopy (43.18%), tracheal extubation (32.18%), pre-induction (28.08%), recovery after extubation (21.89%), maintenance of general anesthesia (12.02%), regional anesthesia or monitored anesthesia care (10.68%), and postoperatively (10.59%). The ventilation modes used during these events included oxygen mask or nasal catheter ventilation (61.51%), endotracheal intubation ventilation (58.25%), and laryngeal mask ventilation (23.92%). Among the 3,344 respondents, most regurgitation and aspiration events occurred during abdominal surgery (78.86%) (Table 2).

Table 2 Surgery Type, Occurrence Phase, Ventilation Mode, Operation Site and Clinical Characteristics of Patients with Gastric Regurgitation or Pulmonary Aspiration

We also surveyed the clinical characteristics and prognosis of patients who experienced regurgitation and aspiration events. As shown in Table 2, common characteristics of these patients included insufficient fasting time (65.55%), intestinal obstruction, gastrointestinal perforation, acute abdomen (63.67%), difficulty or delay in gastrointestinal emptying (53.83%), coma (45.01%), gastroesophageal reflux disease (42.37%), oropharyngeal secretions or bleeding (42.02%), major trauma (41.60%), changes in stomach position (38.43%), and obesity (33.40%). Most patients had a good prognosis after experiencing regurgitation and aspiration, with 62.11% of respondents reporting complete recovery and 54.69% noting mild-to-moderate pneumonia. However, 20.63% and 20.72% reported that patients developed severe pneumonia and death, respectively (Figure 2C).

Prevention and Treatment of Patients with Gastric Regurgitation and Pulmonary Aspiration

For patients at risk of regurgitation and aspiration, 61.98% of the 3,490 respondents reported using rapid sequence intubation, followed by 35.16% who used awake intubation (Figure 3A). Additionally, 86.07% of the respondents considered it necessary to place a gastric tube and suction before anesthesia induction. The Sellick maneuver, which involves compressing the cricoid cartilage during induction, was used by 73.44%, while 69.80% positioned patients in a head-up position. Extending the fasting time for non-emergency surgery was practiced by 63.84%, 62.81% administered H2-receptor antagonists or anticholinergic drugs, and 41.29% used point-of-care ultrasound (POCUS) to assess stomach contents before induction (Figure 3B).

Figure 3 Respondents’ prevention and management practices for patients at risk for regurgitation and pulmonary aspiration. (A) Anesthesia induction methods commonly used by respondents. (B) Preparation measures often taken by respondents before anesthesia induction. (C) Treatment strategies commonly adopted by respondents after reflux and pulmonary aspiration occurred.

Regarding treatment, 98.83% of respondents stated they would immediately suction to clear refluxed material. Administration of tracheal relaxants and glucocorticoids was reported by 78.91%, 67.02% would use the head-down position, 64.13% would initiate anti-infection treatment, and 58.85% would perform fiberoptic bronchoscopy suction and bronchoalveolar lavage (Figure 3C).

When asked to rank the measures to reduce the occurrence of regurgitation and aspiration, the following were ranked from most important to least important: adequate fasting time, detailed medical history inquiry, the choice of anesthesia method, gastric ultrasound assessment of gastric residual volume, and the choice of anesthesia medication.

Training

We surveyed the respondents about their training on perioperative regurgitation and aspiratory, a total of 59.97% of the respondents stated that their departments provided. But only 20.34% reported receiving training on POCUS. (Figure 4A). As shown in Table 3, the training on the management of perioperative regurgitation and aspiratory and the training on gastric ultrasound vary across different genders, hospital grades, teaching hospital statuses, physician titles, and working experience, and the differences are statistically significant.

Table 3 Correlation Analysis and Difference Analysis of Management of Perioperative Regurgitation and Aspiratory and Gastric Ultrasound Training Among Different Variable

Figure 4 (A) The training on pulmonary aspiration management and gastric point-of-care ultrasound in the institution of the respondents. (B) The equipment configuration of the respondents’ institutions. (C) Respondents’ views on gastric point-of-care ultrasound. (D) Respondents’ mastery of gastric point-of-care ultrasound technology. Abbreviations: PA, pulmonary aspiration; POCUS, gastric point-of-care ultrasound.

Equipment

Regarding instrument and equipment availability, 96.88% of respondents’ institutions were equipped with suction devices, 91.58% had emergency intubation kits, 64.30% had fiberoptic bronchoscopes, and less than half (41.63%, n=1453) had POCUS (Figure 4B).

Gastric Point-of-Care Ultrasound

Among the 1,453 respondents whose institutions were equipped with POCUS, 79.22% believed that assessing gastric residual volume with POCUS was very helpful in identifying patients at risk for regurgitation and aspiration. However, only 14.93% were proficient in POCUS technology, and 20.99% did not know POCUS (Figure 4C and D).

Regarding department improvement measures, respondents highlighted the importance of standardized process management (92.84%), regular training (89.43%), equipment enhancement (85.82%), and leadership attention (75.79%). Finally, most respondents (88.68%) agreed that the survey was relevant to their clinical practice, and 85.07% indicated that the survey increased their awareness of perioperative regurgitation and aspiration.

Discussion

Our survey is the most extensive survey of the status of perioperative gastric regurgitation and pulmonary aspiration in China. Respondents came from every province on the Chinese mainland, based on the locations they provided. The survey reveals that regurgitation and aspiration are common perioperative complications in Chinese medical institutions, with 70.77% of respondents reporting personal experiences of these events. Furthermore, 20.63% and 20.72% of the respondents reported having patients who experienced severe pneumonia and death, respectively. This indicates that regurgitation and aspiration remain critical issues in perioperative patient safety management.

Emergency surgery is recognized as a significant risk factor for pulmonary aspiration.^3,4 In adult patients, the incidence of pulmonary aspiration during emergency surgery is 4.3 times higher than during elective surgery.²⁵ This incidence is even higher in pediatric patients, where it is more than 10 times greater than in elective procedures.²⁶ Despite recent trends showing a decrease in pulmonary aspiration during emergency surgeries,^6,22 our survey found that 86.39% of respondents reported experiencing regurgitation or aspiration events during such surgeries. Several factors contribute to this high incidence. First, most emergency surgical patients do not fast for a sufficient period. Second, gastric emptying is often impaired in emergency patients due to conditions such as intestinal obstruction, intra-abdominal lesions, trauma, pain, and stress.^3,27,28 The total number of gastrointestinal endoscopies is estimated to reach 14 million cases per year in China.²⁹ With the dramatic increase in gastrointestinal endoscopies under anesthesia-assisted sedation, regurgitation and aspiration events have become significant complications in painless endoscopy.^7,8,21,30 Our survey showed that 41.78% of the respondents experienced regurgitation or aspiration events during painless endoscopy. This may be related to the weakening of protective airway reflexes during endoscopy under sedation, decreased esophageal sphincter tone, and injection of air and fluid.^8,31–33

Regurgitation and aspiration can occur at any point during the perioperative period. Our survey indicates that these events most frequently occur during the induction of general anesthesia (75.33%), a finding consistent with previous studies.^1,3,6 Numerous risk factors contribute to pulmonary aspiration, including a full stomach, intestinal obstruction, gastroesophageal reflux disease, gastrointestinal bleeding, postoperative gastrointestinal surgery, coma, trauma, obesity, and diabetes.^1,3,4,22,34 In our survey, the majority of patients presented with multiple risk factors. Abdominal surgery was identified as the most common procedure associated with regurgitation and aspiration. For patients with a full stomach, inserting an indwelling gastric tube for gastrointestinal decompression is a feasible strategy to prevent pulmonary aspiration.¹⁸ However, some studies have raised concerns that inserting a gastric tube before the induction of anesthesia may reduce the pressure of the esophageal sphincter, potentially increasing the risk of aspiration.^13,35 Despite this, our survey found that more than 80% of respondents preferred to insert and indwell a gastric tube before induction. Although evidence-based data do not conclusively define the best practices for gastric tube management, the prevailing belief is that the benefits of using a gastric tube outweigh the potential disadvantages, especially in patients with conditions like intestinal obstruction.¹⁸

Rapid sequence induction (RSI) and awake tracheal intubation are commonly used anesthetic techniques for managing patients at risk of aspiration. According to our survey, over 60% of respondents employed RSI, while about 35% preferred awake tracheal intubation. In addition, more than 70% of the respondents preferred the use of cricoid pressure (Sellick maneuver). During RSI, operators use the Sellick maneuver to apply pressure to the cricoid cartilage to prevent gastric inflation during mask ventilation and to reduce the risk of regurgitation and aspiration. However, the clinical application of cricoid pressure remains controversial. There is currently no clear evidence supporting the effectiveness and safety of cricoid pressure in reducing aspiration risk.^16,17 Some studies suggest that it may increase the difficulty of intubation, leading certain guidelines to no longer recommend the Sellick maneuver in clinical practice.^36–38 The majority of respondents in our survey still reported using cricoid pressure, which may be related to previous literature on classical RSI.³⁹

When regurgitation and aspiration occur, a range of symptomatic treatments is available. Our survey found that the most commonly used methods among respondents included immediate suction, positioning the patient in a head-down position, administering bronchodilators and glucocorticoids, performing bronchoalveolar lavage, and providing anti-infective therapy. Immediate suction and the head-down position help to minimize the entry of regurgitated material into the lungs. Bronchodilators and glucocorticoids were the second most frequently used treatments, following suctioning. However, some studies have shown limited benefits of corticosteroids in treating aspiration.⁴⁰ In cases of severe pulmonary aspiration, endotracheal suctioning and alveolar lavage via bronchoscopy may be necessary to clear regurgitant material from the lungs. While routine antibiotic therapy is generally not recommended, targeted antibiotic treatment may be initiated after obtaining cultures from the alveolar lavage fluid.⁴¹

Gastric POCUS is a non-invasive, reusable, cost-effective, and bedside tool used to evaluate gastric contents by qualitatively examining the gastric antrum and calculating the volume of its contents.^11,12,42,43 Gastric ultrasound is highly sensitive and specific for identifying a full stomach, making it an increasingly important part of preoperative assessments for patients at risk of regurgitation and aspiration.¹⁰ In this survey, 41.29% of respondents reported using POCUS to assess gastric contents before the induction of anesthesia in at-risk patients. Despite this, only 41.63% of the respondents’ institutions were equipped with POCUS, and just 20.34% offered specific training in the technique. Among the 1453 respondents from institutions with POCUS, only 14.93% were proficient in its use, while 20.99% did not know the technology at all. Our findings suggest that the clinical application of gastric ultrasound in China is still limited, primarily due to gaps in equipment availability and training within institutions. Therefore, most of the respondents expressed a desire for improvements in several areas, including system process management, equipment availability, training, and increased attention from leadership. More than two-thirds of the respondents felt that the survey was relevant to their clinical practice and that it heightened their awareness of regurgitation and aspiration issues. Additionally, over half indicated that the survey provided them with new knowledge.

This study has several limitations. First, the survey study could not exclude recall bias and could not confirm the accuracy and reliability of responses. Second, although the survey covered all regions of China, the response rate was low in some regions, which may limit the representativeness and generalizability of the findings. Furthermore, this survey did not differentiate between solid foods and liquid drinks, did not distinguish between adults and children, did not differentiate between the types or duration of diabetes, and the treatment with GLP-1 agonists and did not incorporate modern technological aids (such as high-flow preoxygenation, video laryngoscopy). In addition, the survey focused on physicians’ retrospective reviews of regurgitation and aspiration events and their perceptions, which may not fully reflect the most current clinical practices. Nevertheless, this survey represents the largest study to date on regurgitation and aspiration in China, and its findings could be valuable for guiding future research in this area.

Conclusion

This survey suggests that regurgitation and aspiration remain a major safety concern for perioperative patients, with considerable variation in the prevention and management strategies employed by anesthesiologists. Identifying risk factors for aspiration in patients undergoing anesthesia-assisted therapy is essential, and proactive clinical interventions are needed to better detect and reduce the incidence of perioperative pulmonary aspiration. To address these challenges, it is crucial to updating of protocols, creation of checklists, high level of standardization, provision of training and encouraging the acceptance of evidence-based innovations, such as gastric point-of-care ultrasound technology.

Abbreviations

RSI, Rapid sequence induction; ASA, American Society of Anesthesiology; POCUS, Point-of-care ultrasound.

Data Sharing Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgments

We are thankful to those who took the time to complete our survey.

Disclosure

The authors declare that they have no conflicts of interest in this work.

References

1. Cook TM, Woodall N, Frerk C. Major complications of airway management in the UK: results of the Fourth National Audit Project of the Royal College of Anaesthetists and the Difficult Airway Society. Part 1: anaesthesia. Br J Anaesth. 2011;106(5):617–631. doi:10.1093/bja/aer058

2. Lienhart A, Auroy Y, Péquignot F, et al. Survey of Anesthesia-related Mortality in France. Anesthesiology. 2006;105(6):1087–1097. doi:10.1097/00000542-200612000-00008

3. Warner MA, Meyerhoff KL, Warner ME, Posner KL, Stephens L, Domino KB. Pulmonary Aspiration of Gastric Contents: a Closed Claims Analysis. Anesthesiology. 2021;135(2):284–291. doi:10.1097/ALN.0000000000003831

4. Sun J, Wei G, Hu L, Liu C, Ding Z. Perioperative pulmonary aspiration and regurgitation without aspiration in adults: a retrospective observational study of 166,491 anesthesia records. Ann Palliative Med. 2021;10(4):4037–4046. doi:10.21037/apm-20-2382

5. Nason KS. Acute Intraoperative Pulmonary Aspiration. Thoracic Surgery Clin. 2015;25(3):301–307. doi:10.1016/j.thorsurg.2015.04.011

6. Zihui Tan SYL. Pulmonary aspiration under GA: a 13-year audit in a tertiary pediatric unit. Paediatric Anaesthesia. 2016;26(5):547–552. doi:10.1111/pan.12877

7. Bohman JK, Jacob AK, Nelsen KA, et al. Incidence of Gastric-to-Pulmonary Aspiration in Patients Undergoing Elective Upper Gastrointestinal Endoscopy. Clin Gastroenterol Hepatol. 2018;16(7):1163–1164. doi:10.1016/j.cgh.2017.11.024

8. Cooper GS, Kou TD, Rex DK. Complications Following Colonoscopy With Anesthesia Assistance. JAMA Intern Med. 2013;173(7):551. doi:10.1001/jamainternmed.2013.2908

9. Girish P, BBA J, Weigel WA, et al. 2023 American Society of Anesthesiologists Practice Guidelines for Preoperative Fasting: carbohydrate-containing Clear Liquids with or without Protein, Chewing Gum, and Pediatric Fasting Duration-A Modular Update of the 2017 American Society of Anesthesiologists Practice Guidelines for Preoperative Fasting. Anesthesiology. 2023;138(2):132–151. doi:10.1097/ALN.0000000000004381

10. Richelle Kruisselbrink AG, Chaparro LE, Ami N, Richler D, Chan VWS. Anahi Perlas Diagnostic Accuracy of Point-of-Care Gastric Ultrasound. Anesthesia Analg. 2019;128(1):89–95. doi:10.1213/ANE.0000000000003372

11. Bouvet L, Zieleskiewicz L, Loubradou E, et al. Reliability of gastric suctioning compared with ultrasound assessment of residual gastric volume: a prospective multicentre cohort study. Anaesthesia. 2019;75(3):323–330. doi:10.1111/anae.14915

12. Baettig SJ, Filipovic MG, Hebeisen M, Meierhans R, Ganter MT. Pre‐operative gastric ultrasound in patients at risk of pulmonary aspiration: a prospective observational cohort study. Anaesthesia. 2023;78(11):1327–1337. doi:10.1111/anae.16117

13. Klucka J, Kosinova M, Zacharowski K, et al. Rapid sequence induction. Eur J Anaesthesiol. 2020;37(6):435–442. doi:10.1097/EJA.0000000000001194

14. Sajayan A, Wicker J, Ungureanu N, Mendonca C, Kimani PK. Current practice of rapid sequence induction of anaesthesia in the UK – a national survey. Br J Anaesth. 2016;117:i69–i74. doi:10.1093/bja/aew017

15. El-Orbany M, Connolly LA. Rapid Sequence Induction and Intubation. Anesthesia Analg. 2010;110(5):1318–1325. doi:10.1213/ANE.0b013e3181d5ae47

16. Zdravkovic M, Rice MJ, Brull SJ. The Clinical Use of Cricoid Pressure: first, Do No Harm. Anesthesia Analg. 2019;132(1):261–267. doi:10.1213/ANE.0000000000004360

17. Algie CM, Mahar RK, Tan HB, Wilson G, Mahar PD, Wasiak J. Effectiveness and risks of cricoid pressure during rapid sequence induction for endotracheal intubation. Cochrane Database Syst Rev. 2015;2015(11). doi:10.1002/14651858.CD011656.pub2

18. Salem MR, Khorasani A, Saatee S, Crystal GJ, El-Orbany M. Gastric Tubes and Airway Management in Patients at Risk of Aspiration. Anesthesia Analg. 2014;118(3):569–579. doi:10.1213/ANE.0b013e3182917f11

19. Koerber JP, Roberts GEW, Whitaker R, Thorpe CM. Variation in rapid sequence induction techniques: current practice in Wales. Anaesthesia. 2008;64(1):54–59. doi:10.1111/j.1365-2044.2008.05681.x

20. Ehrenfeld JM, Cassedy EA, Forbes VE, Mercaldo ND, Sandberg WS. Modified Rapid Sequence Induction and Intubation. Anesthesia Analg. 2012;115(1):95–101. doi:10.1213/ANE.0b013e31822dac35

21. Bielawska B, Hookey LC, Sutradhar R, et al. Anesthesia Assistance in Outpatient Colonoscopy and Risk of Aspiration Pneumonia, Bowel Perforation, and Splenic Injury. Gastroenterology. 2018;154(1):77–85.e73. doi:10.1053/j.gastro.2017.08.043

22. Sakai T, Planinsic RM, Quinlan JJ, Handley LJ, Kim T-Y, Hilmi IA. The Incidence and Outcome of Perioperative Pulmonary Aspiration in a University Hospital: a 4-Year Retrospective Analysis. Anesthesia Analg. 2006;103(4):941–947. doi:10.1213/01.ane.0000237296.57941.e7

23. ASAT Force. Practice Guidelines for Preoperative Fasting and the Use of Pharmacologic Agents to Reduce the Risk of Pulmonary Aspiration: application to Healthy Patients Undergoing Elective Procedures: an Updated Report by the American Society of Anesthesiologists Task Force on Preoperative Fasting and the Use of Pharmacologic Agents to Reduce the Risk of Pulmonary Aspiration. Anesthesiology. 2017;126(3):376–393. doi:10.1097/ALN.0000000000001452

24. Kotrlik J, H C. Organizational research: determining appropriate sample size in survey research appropriate sample size in survey research. Inform Technol. 2001;19(43):1.

25. Warner MA, Weber JG, Weber J. Clinical significance of pulmonary aspiration during the perioperative period. Anesthesiology. 1993;78(1):56–62. doi:10.1097/00000542-199301000-00010

26. Warner MA, Warner DO, Warner LO, Warner EJ, Warner J. Perioperative pulmonary aspiration in infants and children. Anesthesiology. 1999;90(1):66–71. doi:10.1097/00000542-199901000-00011

27. Landreau B, Nathan N, Nathan N. Pulmonary aspiration: epidemiology and risk factors. Anaesth Crit Care Pain Med. 2009;28(3):206–210. doi:10.1016/j.annfar.2009.01.020

28. Dupont G, Lambert P, Tsekouras N, et al. Ultrasonographic gastric volume before unplanned surgery. Anaesthesia. 2017;72(9):1112–1116. doi:10.1111/anae.13963

29. Zhou S, Zhu Z, Dai W, et al. National survey on sedation for gastrointestinal endoscopy in 2758 Chinese hospitals. Br J Anaesth. 2021;127(1):56–64. doi:10.1016/j.bja.2021.01.028

30. Kollmann CM, Schmiegel W, Brechmann T. Gastrointestinal endoscopy under sedation is associated with pneumonia in older inpatients—results of a retrospective case-control study. United Eur Gastroenterol J. 2017;6(3):382–390. doi:10.1177/2050640617735059

31. Kilian Friedrich SGS, Beck S, Gotthardt D, Stremmel W, Rex DK, bng-Study-Group. Andreas Sieg Respiratory complications in outpatient endoscopy with endoscopist-directed sedation. J Gastrointestinal Liver Dis. 2014;23(3):255–259. doi:10.15403/jgld.2014.1121.233.kf1

32. Ng A, Smith G. Gastroesophageal reflux and aspiration of gastric contents in anesthetic practice. Anesthesia Analg. 2001;93(2):494–513. doi:10.1213/00000539-200108000-00050

33. Rezaiguia-Delclaux S, Bouleau D, Delchier JC, Dhonneur G, Meignan M, Duvaldestin P. Pulmonary scintigraphy for diagnosis of aspiration during intravenous propofol anaesthesia for colonoscopy. Br J Anaesth. 2001;87(2):204–206. doi:10.1093/bja/87.2.204

34. Goyal RK, Longo DL. Gastric Emptying Abnormalities in Diabetes Mellitus. N Engl J Med. 2021;384(18):1742–1751. doi:10.1056/NEJMra2020927

35. Manning BJ, McGreal G, Kirwan WO, Redmond HP, Redmond HP. Nasogastric intubation causes gastroesophageal reflux in patients undergoing elective laparotomy. Surgery. 2001;130(5):788–791. doi:10.1067/msy.2001.116029

36. Ohchi F, Komasawa N, Mihara R, Hattori K, Minami T. Effect of Cricoid Pressure on the Insertion Efficacy of Six Supraglottic Devices: a Crossover Randomized Simulation Trial. J Emergency Med. 2017;53(5):635–641. doi:10.1016/j.jemermed.2017.06.041

37. Charles E, Smith DB. Cricoid pressure decreases ease of tracheal intubation using fibreoptic laryngoscopy (WuScope System. Can J Anaesthesia. 2002;49(6):614–619. doi:10.1007/BF03017391

38. Piepho T, Cavus E, Noppens R, et al. S1-Leitlinie Atemwegsmanagement. Der Anaesthesist. 2015;64(11):859–873. doi:10.1007/s00101-015-0087-6

39. Morris J, Cook TM. Rapid sequence induction: a national survey of practice. Anaesthesia. 2008;56(11):1090–1115. doi:10.1111/j.1365-2044.2001.01962.x

40. Janda M, Scheeren TWL, Nöldge-Schomburg GFE. Management of pulmonary aspiration. Best Pract Res Clin Anaesth. 2006;20(3):409–427. doi:10.1016/j.bpa.2006.02.006

41. Beck-Schimmer B, Bonvini JM. Bronchoaspiration: incidence, consequences and management. Eur J Anaesthesiol. 2011;28(2):78–84. doi:10.1097/EJA.0b013e32834205a8

42. Perlas A, Van de Putte P, Van Houwe P, Chan VWS. I-AIM framework for point-of-care gastric ultrasound. Br J Anaesth. 2016;116(1):7–11. doi:10.1093/bja/aev113

43. Van de Putte P, Perlas A. Ultrasound assessment of gastric content and volume. Br J Anaesth. 2014;113(1):12–22. doi:10.1093/bja/aeu151