Cost-effectiveness analysis of artificial intelligence-aided colonoscopy for adenoma detection and characterization in Spain

Abstract

Background and study aims

The aim of this study was to assess the cost-effectiveness of an intelligent endoscopy module for computer-assisted detection and characterization (CADe/CADx) compared with standard practice, from a Spanish National Health System perspective.

Methods

A Markov model was designed to estimate total costs, life years gained (LYG), and quality-adjusted life years (QALYs) over a lifetime horizon with annual cycles. A hypothetical cohort of 1,000 patients eligible for colonoscopy (mean age 61.32 years) was distributed between Markov states according to polyp size, location, and histology based on national screening program data. CADe/CADx efficacy was determined based on adenoma miss rates and natural disease evolution was simulated according to annual transition probabilities. Detected polyp management involved polypectomy and histopathology in standard practice, whereas with CADe/CADx leave-in-situ strategy was applied for ≤ 5 mm rectosigmoid non-adenomas and resect-and-discard strategy for the rest of ≤ 5mm polyps. Unit costs (€,2024) included the diagnostic procedure and polyp and colorectal cancer (CRC) management. A 3% annual discount rate was applied to costs and outcomes. Model inputs were validated by an expert panel.

Results

CADe/CADx was more effective (16.37 LYG and 14.32 QALYs) than standard practice (16.33 LYG and 14.27 QALYs) over a lifetime horizon. Total cost per patient was €2,300.76 with CADe/CADx and €2,508.75 with colonoscopy alone. In a hypothetical cohort of 1,000 patients, CADe/CADx avoided 173 polypectomies, 370 histopathologies, and 7 CRC cases. Sensitivity analyses confirmed model robustness.

Conclusions

The results of this analysis suggest that CADe/CADx would result in a dominant strategy versus standard practice in patients undergoing colonoscopy in Spain.

Keywords

Endoscopy Lower GI Tract - Colorectal cancer - Polyps, Adenomas - CRC screening - Endoscopic resection (polypectomy, ESD, EMRc) - Computer-assisted detection (CADe) - Computer-assisted characterization (CADx) - Leave-in-situ - Resect-and-discard

Introduction

Colorectal cancer (CRC) is the third most common malignant neoplasm, accounting for approximately 10% of all cancer cases, and the second leading cause of cancer-related deaths worldwide [1]. In Spain, CRC is the most commonly diagnosed tumour, with an estimated 42,721 new cases in 2023 [2]. Population-based CRC screening programs have been implemented to resect precursor lesions and diagnose CRC in earlier stages, decreasing incidence and mortality of CRC and leading to better prognosis [3]. In Spain, these programs are carried out in individuals aged 50 to 69 years through a biennial fecal immunochemical test (FIT), which, if positive, is followed by a colonoscopy [4].

The efficacy of population-based CRC screening programs depends on the rate of detection of precursor lesions during colonoscopy [5], which is far from perfect. A recent meta-analysis reported adenoma miss rates (AMR) of 26%, 9%, and 27% for adenomas, advanced adenomas, and serrated polyps, respectively [6]. Similarly, lesions missed during colonoscopy account for 50% to 60% of interval CRC (iCRC) cases [6].

On the other hand, an increased detection rate results in a higher number of polypectomies and histopathology analyses. However, more than 90% of polyps found in screening colonoscopy are less than 10 mm in size, and 85% are less than 6 mm [7]. In addition, prevalence of advanced histology in diminutive polyps is less than 0.5% [8]. Within this context, introduction of resect-and-discard and leave-in-situ strategies has been proposed to decrease risks and costs of polypectomy and histopathology analyses of diminutive polyps [9]. Nevertheless, these strategies have not been implemented in clinical practice, mainly because of the great variability in efficacy, diagnostic accuracy, and failure to meet standards [10].

In recent years, artificial intelligence (AI)-based systems have been developed that may help overcome these issues. Integration of AI systems can enhance efficiency and accuracy of adenoma detection and characterization through real-time pattern recognition [11]. Indeed, it has been consistently shown that use of computer-assisted detection (CADe) systems can increase the adenoma detection rate [11] [12]. In addition, several studies have described the good performance of computer-assisted characterization (CADx) in terms of differentiating diminutive polyps, with a negative predictive value (NPV) for diagnosing adenoma exceeding 90% [13]. To date, only one study has evaluated performance of an AI system to fulfil the criteria to apply the resect-and-discard and leave-in-situ strategies, which has shown that the GI Genius CADe/CADx technology meets the Preservation and Incorporation of Valuable Endoscopic Innovations (PIVI) requirements for implementation of these strategies [9] [14].

Implementation of a CADe system is likely to increase healthcare costs in the short term as more lesions are detected, but this could be balanced in the long term by CRC management savings [15]. In addition, implementation of a CADx system could further reduce costs by reducing polypectomies and histopathological assessments [14]. However, there is little evidence of the cost-effectiveness in this context. Thus, it is important to explore economic benefits of implementing CADe and CADx solutions to determine feasibility of adopting such technology in healthcare settings with limited resources.

Therefore, the present analysis aimed to assess cost-effectiveness of GI Genius CADe/CADx technology compared with standard clinical practice in patients eligible for colonoscopy in Spain.

Methods

A Markov model was conceptualized and designed to represent the clinical pathway of patients undergoing colonoscopy for primary CRC screening, for polypectomy surveillance, to follow up on a positive FIT result, or because they present any suspicious symptoms or signs. Eight mutually exclusive main Markov states were established: colonoscopy, no polyp, polyp, CRC I, CRC II, CRC III, CRC IV, and death ([Fig. 1]). In addition, within the polyp state, different substates were considered to incorporate characterization of these polyps, and they were classified according to their size (≤ 5 mm, 6–9 mm, or ≥ 10 mm). Furthermore, polyps ≤ 5 mm were also classified according to their location (rectosigmoid [RS], or no rectosigmoid [No-RS]) and histopathology (adenoma [A] or no adenoma [No-A]).

Fig. 1 Markov model diagram. A: Adenoma; CRC, Colorectal cancer; iCRC, Interval colorectal cancer; RS, Rectosigmoid.

A cost-effectiveness model was developed in Microsoft Excel 365 MSO (version 2409) to estimate, over a lifetime horizon with annual Markov duration cycles, total cumulative costs, and total health outcomes, in terms of life-years gained (LYG) and quality-adjusted life-years (QALYs). The population considered in the analysis consisted of a hypothetical cohort of 1,000 adult patients eligible for colonoscopy, with a mean age of 61.32 years at model entry.

At the beginning of the simulation, the assessed cohort of patients, who presented an initial distribution of different clinical situations based on the available evidence (absence of polyps, presence of polyps of different sizes, locations, and histology, or CRC stage), underwent an index colonoscopy (i.e., the first colonoscopy or a follow-up colonoscopy). Detection rates differed between patients who underwent colonoscopy with standard practice and patients who underwent colonoscopy with GI Genius. Patients with undetected lesions could progress to larger lesions or even to iCRC in the period until the next follow-up colonoscopy. Patients in whom a lesion was detected received the appropriate management strategy according to their diagnosis and were then considered to constitute a healthy population. However, during the follow-up period, the healthy population could develop new ≤ 5mm lesions, which could also evolve into larger lesions and could even progress to cancer. At the time of the next follow-up colonoscopy in both groups of patients, the detection rate corresponding to each type of lesion was reapplied, and corresponding management and follow-up strategies were repeated for each case.

Patients diagnosed with CRC were classified according to CRC stage (CRC stage I, CRC stage II, CRC stage III, and CRC stage IV) and could progress from early to advanced stages. At any time during the simulation, there was a risk of death either due to general causes or due to the clinical situation of the patient.

This study was conducted from the perspective of the National Health System (NHS) of Spain to illustrate the efficiency of incorporating AI-assisted colonoscopy for detection and characterization of polyps and CRC. An annual discount rate was applied to both costs and health outcomes, in line with national guideline recommendations for development of a cost-effectiveness analysis [16]. The model considered a 3.00% annual discount rate according to published recommendations for Spain [16].

Panel of experts

The model structure and all the input values necessary for development of the analysis were validated and agreed upon by a panel of three endoscopists with extensive expertise, experience, and knowledge of the disease and the CRC screening program. To this end, a structured questionnaire was developed with all the parameters identified in the scientific literature that were proposed for use in the model. This questionnaire was individually completed by the experts, and subsequently, two face-to-face consensus meetings were held to validate and agree on values and assumptions when needed.

Interventions assessed

The analysis compared two different interventions for early detection of malignancies: GI Genius Intelligent Endoscopy Module-assisted colonoscopy versus the current standard clinical practice (i.e., colonoscopy alone).

Clinical data

At the beginning of the simulation, based on data reported in national screening programs and available evidence identified in the scientific literature [14] [17] [18] [19] [20] [21], the population was stratified based on different health states and substates, absence of polyps (33.15%), presence of lesions of different sizes, locations and histology (≤ 5 mm RS No-A [14.80%], ≤ 5 mm RS A [2.25%], ≤ 5 mm No-RS No-A [2.66%], ≤ 5 mm No-RS A [7.80%], 6–9 mm [21.61%], ≥ 10 mm [13.91%]) and presence of CRC (CRC stage I [0.84%], CRC stage II [1.03%], CRC stage III [1.09%] and CRC stage IV [0.86%]).

Efficacy of GI Genius was determined for each lesion size based on lesion miss rate (LMR) described as AMR in a prospective randomized study carried out in eight centers in Italy, the United Kingdom, and the United States ([Fig. 1]) [12]. The population with undetected lesions was calculated using the LMR, whereas the number of patients with detected lesions was calculated using the complementary value of the LMR.

Table 1 Clinical data.

A, adenoma; CRC, colorectal cancer; iCRC, interval colorectal cancer; RS, rectosigmoid.
Lesion miss rate		GI Genius		Standard practice	Reference
≤ 5 mm RS No-A		15.85%		35.75%	[12]
≤ 5 mm RS A		15.85%		35.75%	[12]
≤ 5 mm No-RS No-A		15.85%		35.75%	[12]
≤ 5 mm No-RS A		15.85%		35.75%	[12]
6–9 mm		20.69%		22.86%	[12]
≥ 10 mm		6.06%		15.79%	[12]
Detected lesions’ management		GI Genius		Standard practice
≤ 5 mm RS No-A		Leave-in-situ		Polypectomy + histopathology	[9] [22]
≤ 5 mm RS A		Resect-and-discard		Polypectomy + histopathology	[9] [21]
≤ 5 mm No-RS No-A		Resect-and-discard		Polypectomy + histopathology	[9] [21]
≤ 5 mm No-RS A		Resect-and-discard		Polypectomy + histopathology	[9] [21]
6–9 mm		Polypectomy + histopathology		Polypectomy + histopathology	[9] [21]
≥ 10 mm		Polypectomy + histopathology		Polypectomy + histopathology	[9] [21]
Transition from healthy patient to patient with ≤ 5mm adenoma
50–54 years old		0.80%			[23]
55–59 years old		1.00%			[23]
60–64 years old		1.20%			[23]
65–69 years old		1.30%			[23]
≥ 70 years old		1.50%			[23]
Evolution of lesions
From ≤ 5 mm to 6–9 mm		3.50%			[23]
From 6–9 mm to ≥ 10 mm		2.20%			[23]
From ≥ 10 mm to iCRC		5.00%			[24]
Evolution of CRC
CRC stage I to CRC stage II		23.80%			[23]
CRC stage II to CRC stage III		48.50%			[23]
CRC stage III to CRC stage IV		30.20%			[23]
Annual recurrence risk
CRC stage I		5.80%			[25]
CRC stage II		5.80%			[25]
CRC stage III		18.80%			[25]
CRC stage IV		18.80%			[25]
Distribution at follow-up colonoscopy
≤ 5 mm RS No-A		53.78%			[14]
≤ 5 mm RS A		8.19%			[14]
≤ 5 mm No-RS No-A		9.66%			[14]
≤ 5 mm No-RS A		28.36%			[14]
CRC annual mortality rate		< 65 years	65–75 years	> 75 years
CRC stage I	No recurrent	3.00%	5.00%	10.50%	[25]
	Recurrent	56.00%	56.00%	87.00%	[25]
CRC stage II	No recurrent	3.00%	5.00%	10.50%	[25]
	Recurrent	56.00%	56.00%	87.00%	[25]
CRC stage III	No recurrent	5.00%	8.50%	16.50%	[25]
	Recurrent	68.00%	67.00%	93.50%	[25]
CRC stage IV	No recurrent	5.00%	8.50%	16.50%	[25]
	Recurrent	68.00%	67.00%	93.50%	[25]

Management of the lesions was performed based on a consensus document on the follow-up of patients after lesion resection during colonoscopy [22]. Polypectomy and histopathology for any lesion were considered standard practices regardless of its size, histology, or location. When GI Genius was used, based on PIVI criteria endorsed by American guidelines, the leave-in-situ strategy was applied for ≤ 5 mm RS No-A lesions, and the resect-and-discard strategy was applied for the other lesions ≤ 5 mm (≤ 5 mm RS A, ≤ 5 mm No-RS No-A and ≤ 5 mm No-RS A) ([Table 1]) [9]. Finally, management of lesions > 5 mm involved both polypectomy and histopathology. Regarding post-colonoscopy follow-up surveillance, a 10-year interval was considered for patients without lesions and patients with lesions < 10 mm, and a 3-year interval was considered for patients with lesions ≥ 10 mm ([Table 1]) [22].

Natural evolution of the disease was simulated according to annual transition probabilities between the different health states identified in the literature [23] [24] [25]. All lesions except ≤ 5 mm RS No-A lesions could grow in size [23] or progress to iCRC [24]. Moreover, within the CRC state, patients could also experience recurrences associated with worsened outcomes ([Table 1]) [25].

Mortality

All-cause mortality data were considered in the model to reflect annual probability of death and stratified by age and sex. Standardized rates for the Spanish population were obtained from the National Statistics Institute [26]. Given that CRC patients have a higher risk of death than the general population, the analysis considered specific CRC-related mortality from available literature data ([Table 1]) [25].

Utilities

To estimate cumulative QALYs, different utility values reported in the scientific literature [23] [27] for each of the health states were considered to assess the impact on patient quality of life. Health state-specific utility values were derived from EuroQol 5 Dimension (EQ-5D) questionnaire scores [27]. All the utilities are described in detail in [Table 2].

Table 2 Utilities and costs.

*This cost includes the cost of histopathology (€ 64.08) and the cost associated with a facultative visit (€ 88.15) CRC, colorectal cancer.
Utilities			References
No polyp		0.88	[27]
Polyp		0.88	[27]
CRC stage I		0.74	[23]
CRC stage II		0.74	[23]
CRC stage III		0.59	[23]
CRC stage IV		0.25	[23]
Costs
Colonoscopy		€ 326.98/per colonoscopy	[28]
GI Genius		€ 7.59/per colonoscopy	Assumption
Polypectomy		€ 133.55/per lesion	[28]
Histopathology*		€ 152.23/per lesion	[28] [29]
CRC management	Stage I	€ 4,211.61/per annum	[17]
	Stage II	€ 4,700.04/per annum	[17]
	Stage III	€ 4,714.16/per annum	[17]
	Stage IV	€ 7,833.94/per annum	[17]

Resource consumption and costs

In line with the perspective adopted, only direct healthcare costs were considered in the analysis, including diagnostic procedures and disease management costs per health state. All costs are expressed in euros, 2024-year value (€, 2024).

Regarding diagnostic procedure cost, colonoscopy and acquisition of GI Genius were considered. Cost of colonoscopy (€ 326.98) was derived by averaging unitary tariffs identified through a national database that collects health costs from different sources ([Table 2] [28]). The cost of GI Genius for each colonoscopy (€ 7.59) was estimated by considering an average cost of € 45,000 per intelligent endoscopy module, which included 3 years of software updates and support and allowed for 1,976 colonoscopies per year (assuming use in a room that is active 8 hours per day with an average duration per colonoscopy of 1 hour over the 247 working days estimated per year) ([Table 2]).

Cost of disease management was established by the health state. For estimation of cost of lesion management, polypectomies, and histopathology analyses were considered, as well as facultative visits per colonoscopy associated with communication of histopathology findings. Polypectomies and facultative visit unitary costs were estimated as the average of the unitary tariffs identified [28], and the cost of histopathology was obtained from the literature ([Table 2]) [29]. Annual costs associated with CRC management at each stage of the disease were derived from a retrospective observational study carried out in Spain ([Table 2]) [17]. These costs were updated to 2024 values with the general consumer price index reported by the Spanish National Statistics Institute [26].

Cost-effectiveness analysis

For each of the interventions assessed in the model, total costs and QALYs gained were estimated. Efficiency was expressed as an incremental cost-utility ratio (ICUR) in terms of cost per additional QALY with GI Genius-assisted colonoscopy versus colonoscopy performed without AI according to current standard of clinical practice.

Although in Spain, there is no officially established threshold of willingness to pay, a strategy is usually considered cost-effective when the ICUR versus the alternative option is below a cost-utility threshold of € 25,000/QALY gained [30]. Moreover, when a strategy is as or more effective and less costly than the alternative option, it is considered to be a dominant strategy [31].

Sensitivity analyses

Sensitivity analyses (SA), including one-way SA (OWSA) and probabilistic SA (PSA), were performed to assess robustness of the model and uncertainty around the parameter values considered in the analysis.

To carry out OWSA, the following parameters were varied individually: discount rate, mean age of the initial cohort, initial distribution of patients, LMR, probability of a healthy person developing a ≤ 5mm adenoma, lesions, and CRC evolution, management strategy of ≤ 5 mm RS No-A lesions, utility values, colonoscopy cost, GI Genius cost, lesion management cost and CRC management cost (Supplementary Material 1).

PSA was performed through 10,000 Monte Carlo iterations to assess the impact of model parameters by simultaneously varying their values. A beta distribution was chosen to modify the LMR, probability of healthy patients developing a ≤ 5mm adenoma, evolution of lesions and CRC, risk of CRC recurrence, and utilities. Distribution of patients at follow-up colonoscopy was modified with a Dirichlet distribution, and a gamma distribution was used for unitary resource costs (colonoscopy, GI Genius, polypectomies, and histopathology) and annual CRC management costs (Supplementary Material 2).

Results

Base case

During the simulation, for a cohort of 1,000 persons eligible for undergoing a colonoscopy, GI Genius was associated with performance of 2,879 colonoscopies, 574 polypectomies, 377 histopathology analyses, and detection of 44 CRC cases. In comparison, standard clinical practice was associated with 2,863 colonoscopies, 747 polypectomies, 747 histopathology analyses, and detection of 51 CRC cases. Therefore, the use of GI Genius at colonoscopy avoided 173 polypectomies, 370 histopathology analyses, and 7 cases of CRC due to early-stage detection of lesions.

Over a lifetime horizon, GI Genius was associated with 16.37 LYG and 14.32 QALYs and resulted in a more effective option compared with current standard practice (16.33 LYG and 14.27 QALYs).

The total cost per patient at the end of the simulation was € 2,300.76 with GI Genius-assisted colonoscopy and € 2,508.75 with colonoscopy alone.

Based on these results, GI Genius was considered a dominant option, i.e., it is more effective and less costly than standard clinical practice in a population undergoing colonoscopies for CRC detection in Spain ([Table 3]).

Table 3 Base case results.

ICER, incremental cost-effectiveness ratio; ICUR, incremental cost-utility ratio; LYG, life-years gained; QALY, quality-adjusted life year.
	GI Genius	Standard practice	Incremental (GI Genius vs. standard practice)
Total LYG	16.37	16.33	0.04
Total QALYs	14.32	14.27	0.05
Total costs	€ 2,300.76	€ 2,508.75	– € 207.99
Diagnostic cost	€ 724.59	€ 704.09	€ 20.50
Disease management cost	€ 1,576.17	€ 1,804.66	– € 228.49
ICER (cost/LYG gained)	GI Genius resulted in a dominant option (more effective and less costly than current standard practice)
ICUR (cost/QALY gained)	GI Genius resulted in a dominant option (more effective and less costly than current standard practice)

Sensitivity analyses

The results of SA confirmed the robustness of the model´s base case results in the simulations. GI Genius was a dominant strategy compared with standard clinical practice for all OWSAs performed. The parameter that most influenced the results was the LMR, followed by the discount rate. Variations in the LMR resulted in variations in the ICUR ranging from – € 4,513/QALY gained to – € 7,789/QALY gained, indicating that this parameter influences results of the analysis. A decrease in the discount rate (0%) caused an increase in the ICUR up to – € 3,597/QALY gained. On the other hand, an increase in the discount rate (5%) caused a reduction in the ICUR up to – € 5,267/QALY gained ([Fig. 2]).

Fig. 2 Deterministic sensitivity analysis. Tornado diagram. A, Adenoma; CRC, Colorectal cancer; iCRC, Interval colorectal cancer: ICUR, Incremental cost-utility ratio; LMR, Lesion miss rate; QALY, Quality-adjusted life years; RS, Rectosigmoid

PSA results showed that GI Genius was a dominant strategy in 92.91% of the 10,000 simulations performed, with lower costs (average savings of € 204.22) and greater effectiveness (average gain of 0.04 LYG and 0.04 QALYs). Results of each of the simulations are shown in the cost-effectiveness plane ([Fig. 3]).

Fig. 3 Probabilistic sensitivity analysis. Cost-effectiveness plane. PSA, Probabilistic sensitivity analysis; QALYs, Quality-adjusted life years

Discussion

This analysis shows that although implementation of AI-aided colonoscopy requires an initial investment, use of GI Genius over a lifetime horizon results in increased effectiveness and a reduction in costs for the NHS. The increase in effectiveness is explained by increased survival, with better quality of life (16.37 LYG and 14.32 QALYs with GI Genius compared with 16.33 LYG and 14.27 QALYs with current standard practice per patient). The feasible rationale is that a reduced LMR results in a better diagnostic yield and treatment of precursor lesions before they become cancerous. The reduction of € 207.99 per patient can be explained by a reduction in CRC management costs due to earlier detection of CRC cases and a reduction in CRC cases because of increased detection of lesions before they can progress to carcinoma. In addition, these differences in total costs can also be justified by avoidance of polypectomies and histopathology analyses; although more lesions are detected with use of AI, introduction of the CADx module enables cost-saving strategies to be implemented (leave-in-situ and resect-and-discard), resulting in a reduction in the number of polypectomies and histopathology analyses performed [14]. Therefore, the economic benefits of AI may also extend beyond lesion detection to lesion characterization.

In the current context of the NHS in Spain, where resources available for health care are limited, cost-saving strategies based on optical diagnosis of colorectal lesions are required to reduce the economic burden of polypectomy and histological diagnosis, as well as risks associated with endoscopic resection. Nevertheless, implementation of the leave-in-situ strategy for ≤ 5mm RS hyperplastic lesions and the resect-and-discard strategy for more proximal ≤ 5mm lesions has been hindered partly by suboptimal accuracy reported by the endoscopy community [10] and because minimum cut-off values required to incorporate this paradigm in clinical practice have not been reached [9]. AI programs thus have the potential to improve overall prediction of histology based on endoscopic imaging, thereby democratizing access to enhanced diagnostic results. Consequently, this may lead to improved health equity and increased NHS efficiencies while also yielding potential cost savings [14]. So far, as a previous cost-effectiveness analysis focused on the detection of lesions, reduction in costs derived from use of AI during colonoscopies was only centered on avoidance of CRC, implying long-term savings. However, in our analysis, after considering AI-guided characterization of lesions and thus applying leave-in-situ and resect-and-discard strategies, a cost reduction per patient was observed from the first year of the simulation, when colonoscopies were performed, due to management of detected lesions and reduction of polypectomies and histopathological analyses. Furthermore, in the era of climate change and global warming, the opportunity to greatly reduce greenhouse gas emissions related to gastrointestinal pathology processing would be an additional benefit associated with use of AI [32].

To the authors’ knowledge, this is the first cost-effectiveness analysis exploring benefits of adding both the detection and characterization modules to AI-assisted colonoscopies, as previously published economic evaluations studying the addition of AI to colonoscopies only included the consequences of using the CADe module. In addition, this study is the first economic evaluation in the Spanish context comparing the use of GI Genius with current standard practices in patients undergoing colonoscopy. Previously published analyses have evaluated efficiency of GI Genius detection in other countries, such as Italy and Canada [33] [34]. Findings from this study are in line with other cost-effectiveness analyses previously published in the literature, indicating that incorporation of the detection module of GI Genius results in a more effective and less costly alternative than use of standard clinical practice [33] [34]. Interventions that are more effective and less expensive than their comparator, known as dominant alternatives, should always be accepted in decision-making, because their adoption generates better clinical outcomes while saving system resources [31].

There are some limitations in the present model that should be considered when interpreting the results. First, there are potential limitations inherent to the nature of this type of economic evaluation. For instance, the theoretical nature of cost-effectiveness analyses may not be an exact representation of clinical practice. In this sense, the influence of the number of lesions presented per patient was not considered in the analysis, because it added more complexity to the model and thus uncertainty. Nevertheless, available guidelines on CRC screening and lesion management were followed to design the structure of the model [4] [22]. Similarly, the model considered GI Genius to be 100% accurate for polyp characterization to avoid introducing more complexity into the simulation. The accuracy of GI Genius in real life will certainly be lower, but data are scarce. An Italian study showed an overall accuracy of 86.8% for lesions ≤ 5 mm, with a slight increase in the rectosigmoid (91.8%) and an NPV of 97.6% for the rectosigmoid lesion [14]. More reliable data are needed before introducing this variable into an economic analysis. In addition, adverse events (AEs) related to polypectomies were not considered because of their low incidence, difficulty involved in quantifying this incidence due to the large variation in the literature, and the complexity it would have added to the model. Nonetheless, if considered, costs associated with AEs after polypectomies should have been less with GI Genius, as fewer polypectomies are performed.

In addition, due to a lack of available data, assumptions related to natural evolution of the pathology need to be made. LMR and the evolution of lesions depend only on their size and not on their location or histopathology. Moreover, it was assumed that all 6 to 9 mm and ≥ 10mm lesions were adenomas, because, according to the experts’ opinion, the percentage of non-adenomatous lesions of these sizes was very low. For lesion detection, it was assumed that there were no false positives. Moreover, it was assumed that there was no risk of discarding invasive cancer, because prevalence of invasive cancer in diminutive polyps is very low [35] and considering it would have added more complexity to the model. Finally, when no data were available for the Spanish context in the literature, values from other countries were selected. However, all parameters and assumptions included in the analysis were validated by a panel of clinical experts with experience and expertise in CRC screening.

Despite the above limitations and the assumptions considered in the analysis, results of the SA confirm the robustness of the model, because uncertainty associated with the parameters used in the modelling did not show a significant deviation from results obtained in the base case, with the use of GI Genius being a dominant strategy in all simulations.

Conclusions

In conclusion, results of the present model suggest that use of the GI Genius CADe and CADx modules could be considered a dominant strategy (i.e., it is more effective and less costly) compared with standard clinical practice in patients undergoing colonoscopies in Spain. Moreover, this analysis confirmed that use of a computer-aided colonoscopy with GI Genius for CRC screening can help to avoid polypectomies and histopathology analyses and improve detection of CRC precursor lesions.

Conflict of Interest

MB-B has received research grants from Medtronic Ibérica, S.A., payment from Medtronic Ibérica, S.A. for consultant activities, and speaker's fees from Olympus and Medtronic Ibérica, S.A. BMR and LBP declare no conflict of interest. IO, MM, and SDP are full-time employees of PORIB, a consultancy firm which has received economic funding by Medtronic Ibérica, S.A. to conduct this project. MA, JM, and NvL are full-time employees of Medtronic Ibérica, S.A. The authors hereby declare that this economic support has not interfered with development of this project. The authors state that the sponsor did not participate or influence analysis of the present study or interpretations of its results. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

• References

• 1 Global Cancer Observatory. International Agency for Research on Cancer. Cancer Today: Spain. 2021. https://gco.iarc.fr/today/fact-sheets-populations
• 2 Red Española de Registros de Cáncer (REDECAN). Estimaciones de la incidencia del cáncer en España, 2023. 2023. https://redecan.org/storage/documents/02d62122-9adb-4d35-b6d0-551435dbe4ae.pdf
• 3 Ponti A, Anttila A, Ronco G. et al. Cancer Screening in the European Union (2017). Report on the implementation of the Council Recommendation on cancer screening. 2017. https://health.ec.europa.eu/document/download/911ecf9b-0ae2-4879-93e6-b750420e9dc0_en
• 4 Red de Programa de Cribado de Cáncer. Evaluación programas de cribado de cancer colorectaL-2019. 2019. https://cribadocancer.es/
• 5 Gallastegui E, Palma-Vasquez C, Estrada M. et al. Inteligencia artificial para la detección y caracterización de lesiones precancerosas colorrectales en la colonoscopia. Barcelona: Agència de Qualitat i Avaluació Sanitàries de Catalunya; 2023 (Informes de Evaluación de Tecnologías Sanitarias.
• 6 Zhao S, Wang S, Pan P. et al. Magnitude, risk factors, and factors associated with adenoma miss rate of tandem colonoscopy: A systematic review and meta-analysis. Gastroenterology 2019; 156: 1661-1674 e11
• 7 Rex DK, Helbig CC. High yields of small and flat adenomas with high-definition colonoscopes using either white light or narrow band imaging. Gastroenterology 2007; 133: 42-47
• 8 Gupta N, Bansal A, Rao D. et al. Prevalence of advanced histological features in diminutive and small colon polyps. Gastrointest Endosc 2012; 75: 1022-1030
• 9 Rex DK, Kahi C, O’Brien M. et al. The American Society for Gastrointestinal Endoscopy PIVI (Preservation and Incorporation of Valuable Endoscopic Innovations) on real-time endoscopic assessment of the histology of diminutive colorectal polyps. Gastrointest Endosc 2011; 73: 419-422
• 10 Vu HT, Sayuk GS, Hollander TG. et al. Resect and discard approach to colon polyps: real-world applicability among academic and community gastroenterologists. Dig Dis Sci 2015; 60: 502-508
• 11 Repici A, Badalamenti M, Maselli R. et al. Efficacy of real-time computer-aided detection of colorectal neoplasia in a randomized trial. Gastroenterology 2020; 159: 512-520 e7
• 12 Wallace MB, Sharma P, Bhandari P. et al. Impact of Artificial Intelligence on Miss Rate of Colorectal Neoplasia. Gastroenterology 2022; 163: 295-304 e5
• 13 Mori Y, Kudo S, Misawa M. et al. Real-Time use of artificial intelligence in identification of diminutive polyps during colonoscopy: A prospective study. Ann Intern Med 2018; 169: 357
• 14 Hassan C, Balsamo G, Lorenzetti R. et al. Artificial intelligence allows leaving-in-situ colorectal polyps. Clin Gastroenterol Hepatol 2022; 20: 2505-2513 e4
• 15 Areia M, Mori Y, Correale L. et al. Cost-effectiveness of artificial intelligence for screening colonoscopy: a modelling study. Lancet Dig Health 2022; 4: e436-e444
• 16 Comité Asesor para la Financiación de la prestación Farmacéutica del Sistema Nacional de Salud. Guía de Evaluación Económica de Medicamentos. 2023. https://www.sanidad.gob.es/areas/farmacia/comitesAdscritos/prestacionFarmaceutica/docs/20240227_CAPF_Guia_EE_definitiva.pdf
• 17 Ibarrondo O, Lizeaga G, Martínez-Llorente JM. et al. Health care costs of breast, prostate, colorectal and lung cancer care by clinical stage and cost component. Gaceta Sanitaria 2022; 36: 246-252
• 18 Xunta de Galicia. Dirección Xeral de Saúde Pública. Memoria 2016–2017. https://libraria.xunta.gal/es/direccion-xeral-de-saude-publica-memoria-2016-2017
• 19 Conselleria de Sanitat Universal y Salud Pública. Servicio de cribados poblacionales. Dirección General de Salud Pública y Adicciones. Programa de Prevención de cáncer colorrectal de la Comunitat Valenciana. 2022. https://www.gva.es/es/inicio/atencion_ciudadano/buscadores/departamentos/detalle_departamentos?id_dept=16284
• 20 Consejería de Sanidad. Junta de Castilla y León. Programas de prevención de salud pública. 2021. https://www.saludcastillayleon.es/transparencia/es/transparencia/informacion-datos-publicos/salud-publica/programas-prevencion-salud-publica.ficheros/2189606-Programas%20de%20Prevenci%C3%B3n%20de%20Salud%20P%C3%BAblica%202021.pdf
• 21 Gobierno Vasco. Departamento de Salud. Programa de cribado de cáncer colorrectal. 2019. https://www.osakidetza.euskadi.eus/programa-de-prevencion-del-cancer-colorrectal/webosk00-oskenf/es/
• 22 Mangas-Sanjuan C, Jover R, Cubiella J. et al. Vigilancia tras resección de pólipos de colon y de cáncer colorrectal. Actualización 2018. Gastroenterología y Hepatología 2019; 42: 188-201
• 23 Coretti S, Ruggeri M, Dibidino R. et al. Economic evaluation of colorectal cancer screening programs: Affordability for the health service. J Med Screen 2020; 27: 186-193
• 24 Frazier AL. Cost-effectiveness of screening for colorectal cancer in the general population. JAMA 2000; 284: 1954
• 25 Gilard-Pioc S, Abrahamowicz M, Mahboubi A. et al. Multi-state relative survival modelling of colorectal cancer progression and mortality. Cancer Epidemiol 2015; 39: 447-455
• 26 Instituto Nacional de Estadística (INE). https://www.ine.es/
• 27 Janssen MF, Szende A, Cabases J. et al. Population norms for the EQ-5D-3L: a cross-country analysis of population surveys for 20 countries. Eur J Health Econ 2019; 20: 205-216
• 28 Oblikue Consulting. eSalud – Información económica del sector sanitario. 2023. https://esalud.oblikue.com/
• 29 Loras C, Mayor V, Fernández-Bañares F. et al. Study of the standard direct costs of various techniques of advanced endoscopy. Comparison with surgical alternatives. Digest Liver Dis 2018; 50: 689-697
• 30 Vallejo-Torres L, García-Lorenzo B, Serrano-Aguilar P. Estimating a cost-effectiveness threshold for the Spanish NHS. Health Econ 2018; 27: 746-761
• 31 Sacristán JA. How to assess the value of low-value care. BMC Health Serv Res 2020; 20: 1000
• 32 Gordon IO, Sherman JD, Leapman M. et al. Life cycle greenhouse gas emissions of gastrointestinal biopsies in a surgical pathology laboratory. Am J Clin Pathology 2021; 156: 540-549
• 33 Hassan C, Povero M, Pradelli L. et al. Cost-utility analysis of real-time artificial intelligence-assisted colonoscopy in Italy. Endosc Int Open 2023; 11: E1046-E1055
• 34 Barkun AN, Von Renteln D, Sadri H. Cost-effectiveness of artificial intelligence-aided colonoscopy for adenoma detection in colon cancer screening. J Canadian Assoc Gastroenterology 2023; 6: 97-105
• 35 Vleugels JLA, Hassan C, Senore C. et al. diminutive polyps with advanced histologic features do not increase risk for metachronous advanced colon neoplasia. Gastroenterology 2019; 156: 623-634 e3

Publication History

Received: 01 August 2024

Accepted after revision: 17 December 2024

Accepted Manuscript online:
02 January 2025

Article published online:
14 March 2025