Citations

Importance Immune checkpoint inhibitors (ICIs) are efficacious in many cancer types but can produce immune-related adverse events (irAEs). As such, patients with preexisting autoimmune disorders are often excluded from clinical trials, although subsequent studies have shown that many of these patients have acceptable ICI tolerance. The safety and efficacy of ICIs among patients with preexisting neurologic autoimmune disorders (NAIDs) is not well characterized. Objective To evaluate the safety and clinical outcomes associated with ICI therapy among patients with NAIDs. Design, Setting, and Participants This multicenter retrospective cohort study included patients with cancer who were treated with ICIs between October 2013 and May 2023 and had preexisting multiple sclerosis (MS), myasthenia gravis (MG), Guillain-Barré syndrome (GBS), and other NAIDs as well as a control cohort of patients with Parkinson disease (PD). Exposure ICI therapy. Main Outcomes and Measures Demographic and clinical characteristics (neurologic disability, active or recent immunosuppression), ICI outcomes (response, progression-free survival [PFS], and overall survival [OS]), and safety outcomes (NAID exacerbation, irAEs) were collected. Results A total of 135 patients were included; the median (range) age was 72 (40-88) years, 84 (62%) were men, and 51 (38%) were women. A total of 45 patients had MS; 18, MG; 10, GBS; 5, another NAID; and 57, PD. Exacerbations occurred most frequently in MG (12 of 18 patients [67%]), often resulting in hospitalization (6 [50%]) or death (2 [17%]), with much lower rates in the MS cohort (8 of 45 patients [18%]). Ten patients with a history of GBS tolerated ICI without exacerbations, although 1 developed a fatal case of Lambert Eaton myasthenic syndrome following ICI treatment. No differences in response rate, PFS, or OS were observed between NAID groups. Conclusions and Relevance In this cohort study of ICI use in NAIDs, patients with MG had frequent and more severe exacerbations, while those with MS had few exacerbations. No obvious differences in survival between groups were observed. ICI may be an option for many patients with appropriate oncologic indications and preexisting NAIDs.

PURPOSE Tumor treating fields (TTFields) use alternating electric fields to disrupt cancer cell proliferation. Feasibility of TTFields therapy with gemcitabine/nab-paclitaxel was previously demonstrated in patients with advanced pancreatic adenocarcinoma. PANOVA-3 was designed to confirm safety and efficacy of TTFields in patients with unresectable locally advanced pancreatic adenocarcinoma (LA-PAC). METHODS In this global phase III trial, 571 patients with newly diagnosed LA-PAC were randomly assigned to receive gemcitabine 1,000 mg/m 2 and nab-paclitaxel 125 mg/m 2 by intravenous infusion once a day on days 1, 8, and 15 of a 28-day cycle with or without TTFields. The primary end point was overall survival (OS). Secondary end points included progression-free survival (PFS), local PFS, pain-free survival, and overall response rate (ORR). Distant PFS was analyzed post hoc. RESULTS OS was significantly prolonged using TTFields with gemcitabine/nab-paclitaxel versus gemcitabine/nab-paclitaxel (median, 16.2 months [95% CI, 15.0 to 18.0] v 14.2 months [95% CI, 12.8 to 15.4]; hazard ratio [HR], 0.82 [95% CI, 0.68 to 0.99]; P = .039). PFS, local PFS, and ORR were not improved. Pain-free survival was significantly prolonged with TTFields with gemcitabine/nab-paclitaxel (median, 15.2 months [95% CI, 10.3 to 22.8] v 9.1 months [95% CI, 7.4 to 12.7]; HR, 0.74 [95% CI, 0.56 to 0.97]; P = .027), as was distant PFS (median, 13.9 months [95% CI, 12.2 to 16.8] v 11.5 months [95% CI, 10.4 to 12.9]; HR, 0.74 [95% CI, 0.57 to 0.96]; P = .022). Device-related skin adverse events (AEs) were experienced by 76.3% of patients. Most device-related skin AEs were mild to moderate, with 7.7% of patients reporting a grade 3 AE. CONCLUSION This study demonstrated significant OS, pain-free survival, and distant PFS benefits for TTFields with gemcitabine/nab-paclitaxel versus gemcitabine/nab-paclitaxel in patients with unresectable LA-PAC, with no additive systemic toxicity.

6057 Background: The incidence rate of human papillomavirus associated oropharyngeal squamous cell carcinoma (HPV(+)OPSCC) has persistently increased over the past decade and is anticipated to rise further over the next decade. Cost effective, accurate, and less invasive screening options for HPV(+)OPSCC have remained stagnant leading to the interrogation of circulating tumor DNA in liquid biopsies such as plasma. We hypothesized that candidate methylated DNA markers (MDMs), previously described by our group, would be present in the plasma of patients with HPV(+)OPSCC and absent in controls. Methods: HPV(+)OPSCC cases were enrolled from a high-volume referral practice. Cancer-free control patients were identified from the surrounding 7-county catchment area and frequency matched to cases by age and alcohol use. cfDNA was extracted from 4 mL of plasma and bisulfite converted prior to Target Enrichment Long-probe Quantitative Amplified Signal assays for 15 MDMs. Strand counts for individual MDMs were normalized to a methylated genomic reference marker. An MDM score was derived via random forest modeling of the 15 MDMs. Area under the receiver operator characteristic curve (AUC) for discriminating cases from controls was evaluated for individual markers and for the MDM score. The sensitivity of the MDM score at 95% specificity was also evaluated. Results: The study consisted of 96 HPV(+)OPSCC and 100 cancer-free controls. The study population was predominately white (95%) with a median age of 61 yrs (IQR: 52-69 yrs). Eighty-nine percent of the cases were male with cancers affecting either the tonsil in 50 (52%) or base of the tongue in 46 (48%) patients; by design 50% of controls were male. Stage I:II:III:IV disease was present in 66%:22%:9%:3% of the cases. The median (IQR) AUC across the 15 MDMs was 0.83 (0.82-0.85) overall: 0.78 (0.77-0.83) for stage I disease and 0.92 (0.91-0.93) for stage II-IV disease. The MDM score for the combined panel of 15 MDMs achieved a cross-validated AUC of 0.93 (CI: 0.89-0.97) overall: 0.90 (CI: 0.84-0.96) for stage I disease and 0.98 (CI: 0.97-1.00) for stage II-IV disease. At 95% specificity, the cross-validated sensitivity of the MDM score for HPV(+)OPSCC was 80% (CI: 71-87%) overall: 73% (CI: 60-83%) for stage I disease and 94% (CI: 78-99%) for stage II-IV disease. Conclusions: We validated a set of previously discovered MDM markers in HPV(+)OPSCC in a robust case control cohort using plasma. The MDMs were detected in both early and late-stage disease. Additional prospective studies in larger intended use cohorts are needed to validate our results for clinical use.

612 Background: EOP, an I-SPY2 sub-study, evaluates the tolerability and activity of novel endocrine strategies in stage 2/3 breast cancer (BC) patients (pts) predicted to have lower chemotherapy benefit. Laso, a selective estrogen receptor modulator (SERM), has shown favorable toxicity profile and activity in HR+/HER2- endocrine-resistant metastatic BC. Methods: Pts with Stage 2/3 HR+/HER2-, MammaPrint (MP) low risk BC were enrolled. Pts with MP High1 BC were included if clinically node-negative. Pts received oral laso 5 mg daily for six 28-day cycles. Laso was continued until the day prior to surgery. Premenopausal pts received ovarian function suppression (OFS) starting C2D1. The primary endpoint was feasibility (>75% of patients completing >75% study therapy). Baseline (T0), 3-wk (T1) biopsies, and the surgical specimen (T3) was assessed centrally for Ki-67. Breast MRI functional tumor volume (FTV) was performed at T0, T1, 12 weeks (T2), and pre-operatively (T3). Blood was collected for tumor informed ctDNA at T0, T1, T2, T3. Advance event (AE) was assessed using CTCAE V5. Results: From 3/2023 to 5/2024, 20 pts were enrolled. Median age 50.5 years, 50% premenopausal, and 1 male pt. 60% cN0, 80% MP low-risk signature. 18 (90%) pts completed >75% study therapy. Two pts discontinued treatment due to pt preference. Median Ki67 at T0 was 14.7%. At T1, 87.5% of pts remained or suppressed Ki67 to <10% and 37.5% suppressed to <2.7%. Ki67 at T1 was similar between pre-and postmenopausal pts despite OFS (Table). The median MRI FTV was 8.4 cc at T0, and 3.4 cc at T3. Median % FTV reduction from T0 to T3 was -47.5%. 2/20 pts (10%) achieved a modified PEPI score of 0. No patients achieved completed pathological response. Of the 16 pts with RCB results, 2 (12.5%) RCB-1, 6 (37.5%) RCB-2, 8 (50%) RCB-3 disease. 14 pts had ctDNA available at T0. 4/14 were ctDNA+ at T0, 2 of whom became ctDNA negative, and 2 remained ctDNA+. 10/14 pts were ctDNA negative at T0, 8 of whom remained negative, 2 became positive at T1 then cleared. All AEs were grade(G) 1 except 1 pt with G2 hot flashes. Most common AEs include hot flashes (85%), constipation (50%), fatigue (50%), and nausea (35%). One pt had G3 hypersensitivity and hypertension, both unrelated to therapy. Conclusions: Neoadjuvant laso demonstrates a favorable AE profile and promising anti-tumor activity in suppressing 3-wk Ki67 and MRI FTV change in pts with HR+ HER2-negative early BC. Ki67 suppression in premenopausal pts was seen in the absence of OFS. Clinical trial information: NCT01042379 . Ki67 expression at pre-treatment, and 3-wk time point. All Patients(n=20) Premenopausal (n=10) Postmenopausal (n=9) Median Ki67 expression Baseline 10.0% 12.5% 10.0% 3-wk 5.1% 3.0% 6.0% Number of pts with Ki67 expression <10% at 3-wk 87.5% 1 87.5% 85.7% Number of pts with Ki67 expression <2.7% at 3-wk 37.5% 50% 28.6% 1 Include the one male pt.

610 Background: The adverse event (AE) landscape in oncology is changing due to the introduction of immunotherapy and antibody drug conjugates. These AEs come with both short and long-term symptoms that significantly impact patient quality of life. Monitoring for early onset of symptoms could optimize therapy for a particular patient, maximizing potential efficacy while mitigating toxicity. It is also possible that some toxicities are directly associated with drug sensitivity. We sought to identify symptoms associated with pathologic complete response (pCR) using patient-reported outcomes (PROs) in early-stage high-risk breast cancer patients. Methods: Our study population included 288 stage II/III high-risk breast cancer patients enrolled on the I-SPY2 trial from 2021-2024, who received novel neoadjuvant therapies ± standard paclitaxel. pCR was assigned if tumor was absent in breast and nodes at surgery following neoadjuvant treatment. Patients (n = 288, pCR rate = 29%, 89% administered immunotherapy) were sent electronic PROs. 33 patient-reported AEs were measured using NCI's Patient Reported Outcomes - Common Terminology Criteria for Adverse Events (PRO-CTCAE). Each symptom was evaluated using severity, frequency, and interference on a Likert Scale. Presence of early PRO symptoms (cycles 1-3 of treatment) were binarized (at least one of moderate or greater), and odds ratios were computed with pCR as outcome. To assess whether higher grade AEs were enriched in patients that achieved a pCR, we also performed the Wilcoxan rank sum test using maximum (worst) symptom severity. Results: Of 288 patients included in our analysis (median age = 48 years, range = 20-78), 203 (70.5%) were White, 17 (5.9%) were Asian, 33 (11.5%) were Black or African American, and 35 (12.2%) were Hispanic. PRO analysis revealed that patients that had moderate to severe muscle pain (27% vs 10% OR = 3.15, p < 0.05), joint pain (22% vs 8% OR = 3.23, p < 0.05), headache (27% vs 12.5% OR = 2.59, p < 0.05), or mouth/throat sores (16% vs 5% OR = 3.56, p < 0.05) within weeks 1-3 had higher odds of achieving a pCR. When we looked at maximum severity between weeks 1-3, patients that achieved a PCR had higher grade muscle pain (p = 0.04), heart palpitations (p = 0.035), and significantly lower grade numbness and tingling (p = 0.002). Beyond 6 weeks, associations were weaker or insignificant. Conclusions: Our study utilizes an analysis framework that was able to determine sentinel symptoms such as muscle and joint pain, mouth/throat sores and palpitations as early as weeks 1-3 associated with increased efficacy. This may suggest an early immune reaction in patients that eventually respond favorably to treatment. Our work can help provide earlier proactive monitoring to mitigate toxicities, treatment redirection if needed, and a potential symptom-based early understanding to personalize treatment efficacy. A similar analysis is underway to predict immune related AEs. Clinical trial information: NCT01042379 .

5064 Background: FoundationOneCDx (F1CDx) supports two FDA-approved biomarkers to guide treatment decisions for ICI for patients with mCRPC: MSI status and TMB. MSI-H and TMB-H (10+ mut/MB) have strongly overlapping prevalence. We sought to better characterize ICI outcome associations of TMB-H / non-MSI-H population and relative effectiveness of taxanes & ICI among patients who received these agents in sequence. Methods: Following a prespecified analysis plan, this study used the nationwide (US-based) de-identified Flatiron Health-Foundation Medicine mCRPC clinico-genomic database (FH-FMI CGDB), with data originating from ~280 US cancer clinics (~800 sites of care). Inclusion criteria included patients with mCRPC treated with single-agent anti-PD(L)1 therapy in the FH network between 1/1/2011 - 3/30/2024. This study used the MSI and TMB algorithms from the tissue based F1CDx. Time to next treatment (TTNT) and OS were assessed with Kaplan-Meier plots and in multivariable Cox models adjusted for ECOG performance score, socioeconomic status, prior treatment history, and baseline PSA. Among patients who received taxanes in a prior line of therapy, the effectiveness (TTNT1 vs. TTNT2) of taxane and ICI were compared. Results: Among 2995 prostate cancer tissue specimens in the database, 95 (3.1%) were MSI-H and 142 (4.7%) were TMB-H. 94 (3.1%) were MSI-H & TMB-H, 1 was MSI-H & TMB-L, and 48 (1.6%) were TMB-H & not MSI-H. Among these, 84 patients with mCRPC were treated with ICI and met inclusion criteria, including MSI-H & TMB-H (n = 30), non-MSI-H & TMB-H (n = 8), and non-MSI-H and TMB-L (n = 46). The respective median TTNT on ICI was 8.0 vs. 9.6 vs. 3 months. The respective median OS from initiation of ICI was 10.9 vs. not reached vs. 4.4 months. In multivariable models evaluating ICI only, compared to non-MSI-H & TMB-L, the MSI-H & TMB-H group had more favorable TTNT (HR: 0.20, 95%CI: 0.10 – 0.41, p < 0.001) and OS (HR: 0.33, 95%CI: 0.16 – 0.70, p = 0.004), and the non-MSI-H & TMB-H group also had more favorable TTNT (HR: 0.13, 95%CI: 0.04 – 0.45, p = 0.002) and OS (HR: 0.20, 95%CI: 0.05 – 0.75, p = 0.017). 50 of the 84 (60%) patients treated with ICI had prior mCRPC taxane treatment. Better TTNT2 on subsequent ICI vs. prior taxane was observed for MSI-H (HR: 0.49, 95%CI: 0.23 – 1.01, p = 0.051) and TMB-H (0.54, 95%CI: 0.30 – 0.98, p = 0.044), but the opposite was true for non-MSI-H and TMB-L subgroups, with significant treatment interactions for each (p = 0.0018, p = 0.00052). Conclusions: TMB-H (4.7%) is more prevalent than MSI-H (3.1%) by F1CDx in mCRPC. Non-MSI-H / TMB-H (1.6%) in the routine practice cohort have similar outcome associations on ICI to MSI-H. Both MSI-H and TMB-H by F1CDx are predictive of differential benefit for ICI vs. taxanes in later mCRPC treatment lines.

504 Background: Axillary surgery in breast cancer is used for staging and therapeutic purposes, but axillary lymph node dissection (ALND) confers high risk of complications including lymphedema. Accordingly, trials have focused on right-sizing axillary surgery, with options ranging from complete omission of surgery, sentinel lymph node (SLN) surgery, targeted axillary dissection (TAD), or ALND. Identifying a biomarker to reliably predict nodal burden would facilitate accurate selection of surgical management. We evaluated whether the presence or absence of ctDNA in the blood pre and post NAC is associated with residual nodal burden. Methods: I-SPY2 is a prospective, multicenter NAC trial for patients with clinical stage II-III high-risk breast cancer. Patients are randomized to novel NAC agents, with pathologic complete response being the primary endpoint. As part of the trial, serial ctDNA is assessed with a highly sensitive tumor-informed assay using up to 16 patient-specific tumor mutations (Signatera) at baseline, 3 weeks, 12 weeks, and post-NAC. We determined whether ctDNA positivity or negativity post-NAC, and the change in ctDNA status baseline/post-NAC (-/-, -/+, +/+, +/-) are associated with ypN category (N0, N1, N2). Results: ctDNA status was available post-NAC in 495 patients and change in ctDNA status from baseline was available in 493. At baseline, ctDNA was detected in 160/220 (72.3%) cN0 patients and 227/273 (83.2%) cN+ patients (p=0.006). Post-NAC, ctDNA was detected in 11/220 (5%) cN0 patients and 34/273 (12.5%) cN+ patients (p=0.004). While baseline ctDNA status was not associated with ypN category, there was a significant association between post-NAC ctDNA status and ypN category. For ctDNA + patients post-NAC, 33.3% were ypN0, 31.1% were ypN1, and 35.6% were ypN2 at surgery; in contrast, for ctDNA - patients post-NAC, 67.1% were ypN0, 23.1% were ypN1, and 9.8% were ypN2 (p<0.0001). Dynamic ctDNA changes were also associated with ypN category, with significantly more ypN0 patients among ctDNA -/- or ctDNA +/- cases, and more ypN2 patients in those who did not clear ctDNA (+/+) (p=0.0001, Table). Conclusions: To our knowledge this is the first study to demonstrate a significant relationship between ctDNA and ypN category, which has important surgical implications. CtDNA negativity was associated with low likelihood of ypN2 disease, making these patients excellent candidates for SLN surgery or TAD. Ongoing analyses will incorporate receptor subtype and timing of clearance of ctDNA. CtDNA in breast cancer may help tailor surgical management of the axilla, potentially reducing patient morbidity without compromising prognostic information. ctDNA baseline/post-NAC ypN0 (n=317) ypN1 (n=116) ypN2 (n=60) ctDNA -/- (n=105) 72.4% 19.1% 8.6% ctDNA -/+ (n=1) 0.0% 100.0% 0.0% ctDNA +/+ (n=44) 34.1% 29.6% 36.4% ctDNA +/- (n=343) 65.9% 23.9% 10.2%

e17617 Background: Abnormal uterine bleeding (AUB) is a cardinal symptom of endometrial cancer (EC). However, 90-95% of those with AUB will undergo invasive endometrial sampling with the finding of benign etiology. Our group has identified EC-related methylated DNA markers (MDMs) detectable in vaginal fluid obtained via self-collected tampon with high sensitivity to detect EC. To provide alternative vaginal fluid collection options, we compared total DNA recovery and detection of EC MDMs between tampon and swab collection methods. Methods: Patients ≥45 yrs with AUB or undergoing routine gynecologic care, and patients ≥18 yrs with biopsy-confirmed EC or atypical endometrial hyperplasia (AEH) were prospectively enrolled for paired vaginal fluid biospecimens collection. Participants self-collected vaginal fluid via tampon before provider-collected vaginal fluid sampling via swab immediately prior to clinically indicated endometrial sampling or hysterectomy. Cases (EC/AEH) and benign controls were determined by expert review. DNA was extracted from vaginal fluid. Bisulfite converted DNA was then used in long-probe quantitative amplified signal (LQAS) assays for 10 MDMs and 2 reference markers (RMs). Total DNA recovery using the RMs and detection of EC-specific MDMs were compared across sampling techniques. MDMs were standardized using the log of the ratio of the MDM and RM1. Standardized MDMs were compared between swabs and tampons using median and interquartile range (IQR). Results: A total of 40 eligible participants were evaluated in this study, 20 cases and 20 controls. Median time between tampon and swab collection was 27 hours (range 1-53). DNA quantity as measured using both RM1 and RM2 were comparable between swab and tampon in both cases and controls; MDM detection was also comparable between swab and tampon in both cases and controls (Table 1; 8 representative MDMs shown). Though not powered for sensitivity and specificity estimation, EC-related MDMs appear higher in EC cases vs controls, using either sample type. Conclusions: Sampling vaginal fluid with provider-collected swabs or self-collected tampons are both effective for detection of validated EC MDMs. Studies are ongoing to optimize an MDM panel to detect EC in vaginal fluid. Control Tampon Control Swab Case Tampon Case Swab RM1 3.5 [3.3-3.8] 4.2 [3.8-4.2] 3.6 [3.2-3.9] 4.2 [3.8-4.5] MDM1 0.5 [0.4-0.7] 0.4 [0.2-0.6] 1.3 [0.8-1.8] 1.1 [0.6-1.7] MDM2 1.5 [1.2-1.9] 0.9 [0.7-1.2] 3.2 [2.2-4.7] 2.9 [1.7-4.5] MDM3 0.5 [0.4-0.6] 0.4 [0.3-0.5] 1.5 [0.9-2.3] 1.0 [0.8-2.3] MDM4 0.9 [0.6-1.3] 0.5 [0.4-0.7] 2.3 [1.3-3.7] 1.9 [1.0-3.6] MDM5 1.3 [0.6-1.4] 0.8 [0.6-1.0] 3.3 [2.5-4.1] 3.1 [2.4-3.9] MDM6 1.6 [1.3-2.1] 1.2 [0.8-1.7] 3.3 [2.6-4.3] 3.1 [2.3-4.1] MDM7 1.5 [1.1-1.7] 1.0 [0.4-1.4] 3.5 [2.6-4.1] 3.3 [2.1-3.8] MDM8 0.5 [0.3-0.6] 0.5 [0.4-0.7] 1.8 [1.0-2.7] 1.7 [1.0-2.5] Median (IQR) for RM1 and 8 MDMs by case/control and sampling technique.

3008 Background: Numerous studies have demonstrated the prognostic value of ctDNA analysis after neoadjuvant chemotherapy for early-stage high-risk breast cancer. Few studies have characterized ctDNA in pts receiving NET for early-stage hormone receptor positive (HR+)/HER2- BC that is predicted to benefit less from chemotherapy. Methods: Cell-free DNA (cfDNA) was isolated from 432 plasma samples from 108 pts enrolled in the I-SPY2 EOP trial. Pts had Stage 2/3 HR+/HER2-, MammaPrint low or high risk 1 BC. Pts were randomized to one of 7 neoadjuvant-based treatment arms including arms containing AI, Z-endoxifen, Lasofoxifene, vepdegestrant (ARV-471), and Abemaciclib. Pts were treated for 6 months prior to surgery. Blood was collected at baseline (T0), 3 weeks (T1), 12 weeks (T2), and 6 months (T3). A personalized ctDNA test (Signatera) was designed to detect up to 16 patient-specific mutations (from whole-exome sequencing of pretreatment tumor) in cfDNA by ultra-deep sequencing. The chi-square test was used to assess associations between categorical variables, and the Wilcoxon rank-sum test was used to evaluate differences in medians. Results: ctDNA information was available for 101 patients at baseline (T0) (Table 1). At T0, 36 (35.6%) patients were ctDNA-positive. 23/36 (63.9%) became ctDNA-negative and 13/36 (36.1%) remained ctDNA-positive. At T0, 65/101 patients (64.4%) were ctDNA-negative. Of these, 57 (87.7%) remained negative, while 8 (12.3%) became ctDNA-positive at T1 before reverting to ctDNA-negative. A higher percentage of ctDNA-positive patients at T0 were cN+ compared to ctDNA-negative pts (p = 0.036, 64% vs. 40%). Additionally, ctDNA-positivity at T0 was strongly associated with higher Ki67 (p = 0.03) and larger functional tumor volume by MRI at baseline (p = 0.03). T3/T4 and high-grade tumors at baseline were also associated with having ctDNA positivity at baseline, though this was not statistically significant (p = 0.34 and 0.058, respectively). Conclusions: In this study of pts with Stage 2/3 HR+ HER2- BC with largely MammaPrint low risk signatures, over one-third of pts had detectable ctDNA at baseline. Detectable ctDNA at baseline was associated with cN+ disease, larger FTV, and high baseline Ki67. The majority of pts with positive ctDNA at baseline cleared the ctDNA on NET. Clinical trial information: NCT01042379 . Clinicopathologic characteristics of EOP patients. All Patients(N=101) Age at screening 55 (27-80) years* Clinical N Stage Node+ (cN+) 49 (48.5%) Node- (cN-) 52 (51.5%) SET status High 84 (83.2%) Low 14 (13.9%) Missing 3 (2.97%) MammaPrint (MP) risk** High risk 1 (H1) 15 (14.9%) Low risk 86 (85.1%) *Mean(min-max). **H1: MP score between 0 and -0.57; low: MP score between 0 and 0.355.

Introduction: Machine learning (ML) has become increasingly pivotal in health care, particularly in colorectal cancer (CRC) detection and diagnosis with the use of predictive models and artificial intelligence-assisted colonoscopies. This study evaluates the efficacy of ML models in predicting the risk for CRC and advanced colorectal polyps (ACP) before colonoscopy. Methods: A systematic literature review was conducted following PRISMA guidelines, focusing on studies using ML for CRC and ACP prediction. Data extraction regarding study type, ML methodology, quality of data, and bias assessment was in line with the CHARMS checklist. Meta-analysis was also performed to assess the performance of models for the prediction of CRC, adenoma, or both. Results: This systematic review included 14 studies with 3618 median patients (333 to 263,879). Our study demonstrated considerable heterogeneity in methodologies and outcomes, with area under the receiver operating characteristic (AUROC) ranging from 0.6 to 1. The derivation+validation cohorts showed a pooled sensitivity of 0.832 (95% CI: 0.755-0.889) and specificity of 0.802 (95% CI: 0.722-0.863), with an overall AUROC of 0.883. Conclusion: The review underscores the significant role of ML in CRC and ACP diagnosis and its routine use could efficiently direct high-risk patients to timely colonoscopies and spare the low-risk ones from unnecessary procedures. Despite the promise shown, the variability in methodologies and outcomes highlights the need for standardized approaches and further investigation in this field.