Sensitivity analysis: non-MSI-H comparison. (A) Standardized mean difference is shown for features pre- and post- adjustment for the TMB10+ cohort. Propensity-adjusted comparison of ICI (tTMB≥10) vs. taxanes (any tTMB level) is shown for (B) TTNT and (C) OS among the cohort excluding MSI-H.

450 Background: Inactivating genomic alterations (GA) of FANCC gene are associated with genomic instability, DNA cross-linking, and homologous DNA repair deficiency (HRD). FANCC GA are most frequently associated with colon, lung, breast, and prostate cancers (0.5% frequency) with germline FANCC mutations linked to familial breast cancer. FANCC GA have been rarely associated with RT and are not currently linked to any hereditary renal cancer predisposition syndromes. We evaluated the incidence of FANCC GA and other genomic features across cancer types. Methods: 463,546 clinically advanced cancers (CAC) underwent hybrid capture-based comprehensive genomic profiling using the FDA-approved F1CDx assay to detect all classes of GA. MSI status, tumor mutation burden (TMB), gLOH, prediction of germline status, genomic ancestry, and genomic signature were determined with algorithm-based analysis. PD-L1 expression was tested by IHC (Dako 22C3 tumor proportional score; low positive (LP) 1-49%). Results: 1,993 (0.43%) CAC featured FANCC GA. 27 of these FANCC-mutated tumors (20 male, mean age 57) were RT (0.35% of 7,668 RT): 13 clear cell, 3 sarcomatoid, 3 urothelial, 3 chromophobe, 2 squamous cell, 2 medullary renal carcinomas (RCC), and 1 Wilm’s tumor. The primary tumor was sequenced in 9 cases and a metastatic site in 18 (5 lymph node, 4 soft tissue, 3 brain, 2 liver, 1 each lung, adrenal, eye, bone). Only 1 of 25 tested FANCC-mutated RT was MSI-high. The mean TMB was 5.7 mut/Mb while the median TMB was 2.5 mut/Mb, and 4 cases (15%) featured TMB ≥10 mut/Mb. 2 of 4 FANCC-mutated RT that were tested for PD-L1 were LP. The mean gLOH was 7.5%. Genomic ancestry evaluation revealed 21 EUR, 4 AFR, and 2 AMR patients. Genomic signature could be assessed in 5 cases: 4 were MMR deficient. The FANCC mutations included inactivating short variant mutations in 24 cases (10 nonsense, 10 frameshift, 2 non-frame and 2 splice-site mutations) and 3 truncating rearrangements ( FANCC: SUSD3, FANCC: FANCC, FANCC: C20orf24). Interestingly, 14 (52%) of the FANCC-mutated RT were predicted to be germline. Additional GA in the FANCC-mutated RT included VHL, TP53, CDKN2A, ARID1A, PBRM1, TERT, PTEN, and SETD2. Conclusions: Somatic and germline mutations in FANCC occur in an exceedingly small subset of clinically advanced RT but at similar rate to other cancers. RT with inactivated FANCC do not appear to have a different GA landscape from RT with wild-type FANCC. The high frequency of predicted germline status during somatic testing with FANCC alterations suggests the importance of further workup with confirmatory germline testing as it may affect counseling for other family members. Limitations of this study include lack of clinical and therapy data annotation.

ABSTRACT:Plasma cell leukemia (PCL) is rare and aggressive. Plasma cells from patients with PCL were examined using next-generation sequencing (NGS). We compared NGS data from 18 patients with PCL (peripheral blood, n = 10; bone marrow, n = 8) and 1742 multiple myeloma (MM) samples. Mutations of TP53, CCND1, and KRAS were commonly observed in both diseases. Alterations in DIS3 (17% vs 1%), CCND2 (22% vs 15%), PIK3R1 (6% vs 0%), and MAP3K1 (6% vs 2%) were more common in PCL than in MM (P< .05). Translocations occurred in 11 (61.1%) patients with PCL; IGH::CCND1 and IGH::MYC were more frequent in PCL than in MM (22% vs 14%, P = not significant and 11% vs 1%, P< .05, respectively). Druggable aberrations in BRAF, CCND1, PIK3R1, and RAS may be targeted in biomarker-driven therapeutic clinical trials.

Clinical and Pathological Features of Treatment Assignment by Treatment Assignment

Pre-adjustment and post-adjustment cohort Imbalances. Standardized mean difference is shown for features pre- and post- adjustment for the (A) TMB10+ cohort and (B) TMB < 10 cohort.

PURPOSE:Immune checkpoint inhibitors (ICI) have shown limited efficacy in unselected patients with metastatic castration-resistant prostate cancer (mCRPC). However, ICIs are approved for biomarker-defined subsets: microsatellite instability-high (MSI-H) and/or high tumor mutational burden (TMB-H). The efficacy of ICIs in TMB-H but not MSI-H disease remains unclear, and limited data exist evaluating ICI outcomes associated with blood-based MSI (bMSI) in mCRPC. EXPERIMENTAL DESIGN:This study used the United States-based deidentified Flatiron Health-Foundation Medicine prostate cancer Clinico-Genomic Database. Patients with tissue-assessed MSI (tMSI) and TMB (tTMB) status by an algorithm supporting an FDA-approved CDx for pembrolizumab were included if treated with single-agent ICI. Separately, outcomes on ICI associated with bMSI were assessed, including if treated with single-agent ICI or taxane. RESULTS:Among 2,965 patients with mCRPC, tMSI-H (3.2%) was nearly always also tTMB ≥10 mut/Mb (4.7%). In 84 ICI-treated patients, time to next treatment (TTNT) and overall survival (OS) were more favorable in tMSI-H with any TMB [TTNT HR, 0.18; 95% confidence interval (CI), 0.09-0.37 and OS HR, 0.32; 95% CI, 0.15-0.66] and tTMB ≥10 without tMSI-H (TTNT HR, 0.18; 95% CI, 0.04-0.48 and OS HR, 0.20; 95% CI, 0.05-0.77) compared with tTMB <10 without tMSI-H group. In intrapatient assessments, patients with tTMB ≥10 had more favorable TTNT with subsequent ICI versus prior taxane. Detection of bMSI-H was associated with more favorable TTNT on ICI (HR, 0.34; 95% CI, 0.14-0.83) and OS (HR, 0.21; 95% CI, 0.06-0.75) when tumor fraction ≥1%. CONCLUSIONS:These findings add support for tTMB and tMSI in predicting ICI monotherapy benefit in mCRPC and provide evidence supporting bMSI testing when tissue is unavailable.

Multivariable Outcomes by Biomarker Status on Chemo. Adjusting for potential known confounders, multivariable models evaluating (A) TTNT and (B) OS are shown.

Multivariable Outcomes by Biomarker Status on ICI. Adjusting for potential known confounders, multivariable models evaluating (A) TTNT and (B) OS are shown.

Genetic Ancestry and tTMB and tMSI associations. Associations between tMSI and tTMB categories are shown in relation to genetic ancestry.

618 Background: The successful outcome for most men with NSGCT of testis after surgery and, in some cases, chemotherapy is a milestone in oncology. However, a minority of patients may become platinum-refractory with relapse or progression after chemotherapy. Comprehensive genomic profiling (CGP) has the potential to uncover new therapy targets for these patients. Methods: Hybrid capture based CGP was performed using the FoundationOne CDx assay on 182 CANSGCT to identify all classes of genomic alterations (GA). The site of the analyzed sample was available in 161/182 patients: 12.4% locoregional lymph nodes, 16.8% testis, 70.8% distant metastases. Microsatellite instability status (MSI), tumor mutational burden (TMB) and Homologous Recombination Deficiency (HRDsig) were determined from the sequencing data. Genomic ancestry and Cosmic trinucleotide signatures were also identified. PD-L1 expression was determined by IHC using Dako TPS score (0% = negative; 1-49% = low positive and ≥50% = high positive). Most cases had relapsed after surgery and chemotherapy, either loco-regionally or with metastatic disease at the time of CGP. Results: 182 men with CANSGCT had median age of 33. Genomic ancestry revealed that 65.4% of patients were European, 29.1% were admixed American, 3.3% were African, 1.6% were East Asian and 0.5% were South Asian. There was a median of 3 GA per sequenced tumor. At 2.2% frequency, MSI-high status was rare and TMB was relatively low with 95% of cases having TMB < 10 mut/Mb. A positive HRD signature was identified in 4.7% of cases. Identification of specific genomic signatures was uncommon with 2.7% having MMR signature and 1.1% having a tobacco exposure signature. PD-L1 expression was low level in 18.1% and high level in 15.3% of cases. KRAS mutations were the most frequent individual GA with <1% KRAS G12C. Inactivation of PTEN (6.6%) and activation of KIT (4.9%) were GA with potentially associated targeted therapies. Conclusions: CGP of CANSGCT cases revealed potential therapy targets with opportunities for clinical trials, e.g. focused on PTEN/MTOR pathway and KIT. Although initial clinical experience with targeted agents and anti-PD1/L1 have yielded limited benefit in unselected patients with these tumors, further CGP-based study of refractory NSGCT is warranted to enable the development of biomarker-driven clinical trials designed to rescue men who exhibit platinum refractory disease. Study limitations include the retrospective nature, lack of comparison with the genomic landscape of platinum-sensitive tumors, lack of clinical outcomes data annotation, selection and confounding biases.

Treatment for gastrointestinal stromal tumor (GIST) focuses on tyrosine kinase inhibitors, the selection of which depends on specific mutations. We sought to determine the clinical use of liquid biopsy in advanced GIST. Liquid (n = 181) (FoundationOne Liquid CDx) and tissue (n = 2198) (FoundationOne and FoundationOne CDx) comprehensive genomic profiling of GIST were evaluated. The presence of circulating tumor DNA in liquid was determined via tumor fraction (TF), with an elevated TF defined as TF >= 1%. Liquid comprehensive genomic profiling revealed 30% (54/181) of samples had an elevated TF, among which the prevalence of KIT and PDGFRA alterations were 89% (48/54) and 2% (1/54), respectively. In patient-matched tissue/liquid samples (n = 49), the positive percent agreement of driver alterations in liquid with an elevated TF relative to tissue was 100%. Fifty-five percent (42/77) of liquid samples with a KIT driver mutation had a co-occurring imatinibresistant alteration; a minority of cases harbored non-KIT mechanisms of resistance such as FGFR2 fusions and BRAF or EGFR alterations. The relative prevalence of imatinib resistance KIT exon 13 and 17 mutations was enriched in liquid compared with tissue. Finally, in the liquid cohort, 2.2%, 1.7%, and 1.1% of patients were predicted to harbor germline KIT, SDHx, or NF1 mutations, respectively. In conclusion, known driver and tyrosine kinase inhibitor-resistant mutations were identified in liquid biopsies of patients with GIST with high concordance to tissue in the presence of an elevated TF. Liquid biopsy may be valuable in the molecular classification and medical management of GIST. (c) 2025 United States & Canadian Academy of Pathology. Published by Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

666 Background: CAUBC (any patients with node positive, metastasis positive or inoperable T4 disease) arising in YOUC is a challenging disease with significant need for improvements in systemic therapy especially for patients with surgically incurable disease. Methods: 9,411 cases of CAUBC underwent comprehensive genomic profiling (CGP) to examine all classes of genomic alterations (GA). MSI high status, tumor mutation burden (TMB) levels, genomic ancestry and trinucleotide mutational signatures were determined from the sequencing data. HRDsig status was calculated using a broad set of genome-wide copy number features (PMID 37769224). PD-L1 was determined by IHC using the Dako 22C3 tumor proportional score (TPS) system. Results were compared using the Fisher exact system with the Benjamini-Hochberg adjustment to correct for false discovery. Results: 291 (3.1%) CAUBC developed in YOUC patients under the age of 50 (CAUBC<50). The GA and biomarkers were compared with 9,120 CAUBC that developed in patients 50 years of age and older (CAUBC>50). The male gender preponderance in the CAUBC>50 was greater than in the CAUBC<50 cases (75.1% vs 69.4%; p=.069). The GA/tumor were higher in the CAUBC>50 (8.0 vs 7.0; p<.0001). Genomic ancestry distribution revealed significantly more Admixed American cases in the CAUBC<50 (8.9% vs 5.2%; P=.033) and more European cases in the CAUBC>50 (85.3% vs 74.9%; P<.0001). MSI high status was extremely uncommon (range 0.3% to 0.8%; NS). The frequency of TMB > 10 mutations/Mb was higher in the CAUBC>50 patients (35.1% vs 28.5%; P=.048). An APOBEC trinucleotide signature was more frequent in the CAUBC>50 patients (32.5% vs 25.8%; p=.041). PD-L1 low expression (1-49% TPS) was available in small subsets of cases and was similar in both groups (21.5% vs 28.5%; p=.84). GA in KMD6A (26.6% vs 14.8%; p=.0001), ERBB2 (18.4% vs 16.8%; p=.85.), ERBB3 (6.3% vs 6.5%; p=.91) MTAP (25.0% vs 20.3%; p=.27) and TERT (78.0% vs 68.4%; p=.005) were more frequent in CAUBC>50 cases and GA in HRAS (4.8% vs 2.1%; p=.11) and CCND1 (16.2% vs 13.1%; p=.35) were more frequent in the CAUBC<50 cases. GA in FGFR1 (4.1% vs 5.2%; p=.71), FGFR3 (14.1% vs 17.9%; p=.31), PIK3CA (20.3% vs 21.3%; p=.91) and PTEN (6.5% vs 4.5%; p=.31) were similar in both groups. Conclusions: In review of the CGP data, in contrast with CAUBC arising in older patients > 50 years of age, CAUBC arising in YOUC patients <50 years features a unique patten of biomarker and GA distribution that has potential to influence therapy selection and guide future clinical trials. CAUBC 0-49 years CAUBC >50 years P-value Sex (% male) 69.4% 75.1% .069 AMR ancestry 8.9% 5.2% .033 EUR ancestry 74.9% 85.3% <.0001 FGFR3 14.1% 17.9% NS KDM6A 14.8% 26.6% .0001 MTAP 20.3% 25.0% NS TERT 68.4% 78.0% .005 TMB>10 mut/Mb 28.5% 35.1% .048 APOBEC Signature 25.8% 32.5% .041

Introduction While renal AML is generally a benign tumor, malignant eAML is a rare form of renal malignancy of uncertain histogenesis and variable prognosis. This CGP-based study was designed to identify genomic alterations (GA) associated with potential therapeutic targets in this uncommon tumor. Methods Hybrid capture based CGP was performed on 38 cases of clinically advanced/metastatic renal eAML for all classes of GA. MSI status, gLOH, genomic ancestry, and gene signatures were determined by CGP. TMB was measured in mutations/Megabase (mut/Mb) of sequenced DNA. PD-L1 was measured by IHC using the DAKO 22C3 tumor cell proportional score (0%: negative; 1-49%: low positive). Results The primary tumor was used for CGP in 16/38 (42.1%) and a metastatic tissue in 22/38 (57.9%) cases (12 liver, 3 lung, 4 retroperitoneum, 2 peritoneum; 1 psoas muscle). In a subset when either MART1 or HMB45 IHC staining was performed all (100%) cases were positive for either biomarker. There were 21 (55.2%) female and 17 (44.8%) male patients with mean age of 50.5 years (median 53 years). There were no MSI-high cases. The mean (1.54 mut/Mb) and median (1.3 mut/Mb) TMB were uniformly low. Of 5 malignant eAML cases tested, 4 (80%) were negative and 1 (20%) case was low PD-L1 expression. Short variant mutations were found in TSC2 in 27 (71.1%) cases, the other 8 (28.9%) were TSC2-wild type. Germline mutations in the malignant eAML cases included 2 (5.3%) TSC2 GA and 1 (2.6%) each in the TP53, ARID1A, BRCA2, CHEK2, EGFR, MUTYH, CD36, FLCN and FANCC genes, all of uncertain roles in the pathogenesis of these tumors. In addition to the TSC2 mutatations, other GA potentially implicating MTOR pathway based targeted therapies included GA in PTEN (5.2%) and NF2 (2.6%). Other identified GA currently considered as “non-targetable” included TP53 (34.2%) and CDKN2A/B, ATRX and RB1 all at 13.2%. GA potentially impacting PARPi use included BRCA2 (2.6%). In 28 evaluable cases, the mean gLOH was 6.3% (0-37.9%) with 3/28 malignant eAML (10.7%) featuring gLOH of >16%. In 34 evaluable cases, 26 (76.5%) patients were of EUR ancestry, 5 (14.7%) of AMR ancestry, 3 (8.8%) of AFR ancestry. No specific genomic signature characterized these cases. Conclusions Renal malignant eAML also known as malignant PEComa of the kidney is an exceedingly rare tumor with propensity for malignant behavior that frequently displays a variety of germline mutations as well as GA indicative of potential efficacy of MTOR pathway inhibitors.

8039 Background: Lung cancers can present as multiple pulmonary tumors, representing either separate primary lung carcinomas (SPLCs) or intrapulmonary metastases (IPMs) arising from a single advanced cancer. Distinguishing SPLCs from IPMs is important for effective staging, prognosis, and treatment. Comparative molecular profiling (CMP) of paired tumors can elucidate their clonal relationship. We report real-world data on the use of CMP to distinguish SPLCs versus IPMs. Methods: Paired lung tumors from the same patients, submitted for FoundationOne CDx (F1CDx) comprehensive genomic profiling within a 12-month period from one another between 2014 and 2024, were centrally reviewed by a board-certified pathologist. After excluding clonal hematopoiesis and germline alterations, 1) tumors with different driver alterations were classified as SPLCs, 2) tumors with shared drivers and other alterations as IPMs, and 3) tumors with 1 common driver but different other alterations as SPLCs. The molecular landscapes of SPLCs and IMPs as defined by CMP were subsequently compared. Results: In all, 359 paired lung cancers were identified and analyzed using pathologist-directed CMP. Of these, 32.6% (117/359) were classified as SPLCs, 63.8% (229/359) as IPMs, and 1.1% (4/359) as misdiagnoses (i.e., not lung cancer). Among SPLC pairs, 89.7% (105/117) harbored actionable genomic alterations in one tumor but not the other. Classification of 2.5% (9/359) of pairs was inconclusive due to suboptimal sequencing quality control metrics. Misdiagnoses, based on the molecular results, included 2 metastatic HPV-associated carcinomas (supported by high-risk HPV reads), and 2 metastatic cutaneous squamous or basal cell carcinomas (supported by ultraviolet mutational signatures). Of 31 paired lung tumors with different histologies, 67.7% (21/31) were SPLCs, while 32.3% (10/31) were IPMs. Molecular landscape analysis showed that IPMs had significantly higher frequencies of CDKN2B (20.1% vs. 10.3%, P=0.03, OR=2.2) and ERBB2 (6.3% vs. 0.9%, P=0.03, OR=7.8) alterations, while SPLCs were more likely to have elevated tumor mutational burden (i.e., TMB ≥10 Muts/Mb) (39.8% vs. 21.3%, P<0.0001). Conclusions: Pathologist-directed CMP using F1CDx facilitates the classification of multiple lung cancers as SPLCs or IMPs and may identify diagnostic errors associated with conventional histopathological examination. The higher prevalence of ERBB2 alterations in IPMs suggests a role in their development and opportunities for targeted therapy, while the increased rate of elevated TMB in SPLCs indicates a field cancerization effect, driven by smoking or other exposures, and possible responsiveness to immunotherapy. Accurately classifying SPLCs and IPMs is important for proper staging and therapeutic planning in patients with lung cancer.

5592 Background: CH results from mutations in hematopoietic stem cells and can occur in clinically relevant genes that are detected in liquid biopsy (LBx) of solid tumor patients. TP53 and less frequently BRCA1/2 can be detected in tumor and as CH potentially confounding interpretation of LBx results. Using an algorithmic method for CH prediction in a cohort of tubo-ovarian carcinoma (OC) LBx, we evaluated the prevalence of CH and non-CH alterations in TP53 and BRCA1/2 genes, CH frequency by circulating tumor DNA tumor fraction (ctDNA TF) and concordance in samples with paired tissue biopsies. Methods: Patients (pts) with a diagnosis of OC and LBx via FoundationOne Liquid CDx were included. ctDNA was quantified via TF. A machine learning model incorporating fragmentomics and other sequencing features was trained using LBx samples with equal-depth sequencing of plasma and white blood cells for short variant origin prediction (VOP) with output probabilities of origin (germline, tumor-somatic, or CH). Oncogenic short variants (i.e. mutations [mut]) with probability of being CH >0.5 were classified as CH and with probability <0.5 as tumor. Detection of CH and tumor TP53 mut in paired tissue samples (FoundationOne CDx) was evaluated (n=355). Results: 1,405 pts met criteria for study inclusion. 498 (35%) had TF≥1%. Overall, TP53 mut was detected in 74% (origin 30% tumor only, 23% CH only, and 21% both CH and tumor TP53 mut), and 26% had no detected TP53 mut Prevalence of VOP TP53 mut groups varied by TF with TF<1% having more pts with no detected TP53 mut (35% vs 9%) or only CH TP53 mut (33% vs 4%) and fewer pts with only tumor TP53 mut (18% vs 51%) or both CH and tumor TP53 mut (13% vs 36%) than TF≥1%. The emerging drug target TP53 Y220C was predicted to be CH in 42/68 (62%) LBx samples. Of the 333 individual TP53 mut predicted to be CH on LBx, 310 (93%) were not detected in corresponding tissue. Of the 200 TP53 mut predicted to be tumor on LBx, 176 (88%) were detected in paired tumor tissue. Overall, 9% of pts had at least one germline BRCA1/ 2mut, 4% had no germline BRCA1/2 mut but had a tumor-somatic BRCA1/2 alteration (51 with mut, 4 with truncating rearrangements or copy number loss), and 1% had only CH BRCA1/2. Of patients with non-germline BRCA1/2 mut, 12/63 (19%) had only CH-derived BRCA1/2 mut. Conclusions: >60% of OC LBx with TP53 mut, including TP53 Y220C, had evidence of CH contributing to cell free DNA. TF <1% was associated with higher rates of CH only TP53 mut, but tumor-derived variants in TP53 and other genes were still detected. The majority (93%) of TP53mut predicted to be CH were not detected in tissue biopsies of paired samples. While the majority of LBx with a BRCA1/2 mut had germline or tumor-somatic muts, 6% of BRCA1/2 mut LBx only harbored BRCA1/2 mut predicted to be CH. Together, CH prediction and TF can be used to correctly contextualize LBx findings to support informed clinical decision making.

Targeted anti-HER2 therapy has been recently added to the standard treatment recommendations in endometrial serous carcinoma. Current eligibility requires testing for HER2 overexpression and/or gene amplification by immunohistochemistry and by fluorescence in situ hybridization. However, clinical trials have also demonstrated the efficacy of anti-HER2 drugs against activating ERBB2/HER2 mutations in a variety of solid tumor types, and fam-trastuzumab deruxtecan is now approved by the US Food and Drug Administration for HER2-mutant non-small cell lung cancer. This study aimed at evaluating the detailed clinical, histomorphological, immunohistochemical, and molecular characteristics of gynecologic malignancies with ERBB2/HER2 mutations. We identified 16 tumors with 19 ERBB2/HER2 mutations in our departmental archives: 11 endometrial primaries, 2 endocervical adenocarcinomas, 1 ovarian mucinous adenocarcinoma, 1 tubo-ovarian undifferentiated carcinoma, and 1 high-grade endometrioid adenocarcinoma of Mullerian origin. ERBB2/HER2 mutations most often involved the tyrosine kinase domain (52.6%), and the most frequent specific mutation was R678Q (31.6%), involving the juxtamembrane domain. More than half (54.5%) of endometrial carcinomas and half of all tumors were MMR-deficient, resulting from MSH6 loss in all but 2 tumors. None of the tumors (0%) were POLE-mutated, while 18.8% were TP53-mutated. HER2 IHC was negative (score 0 or 1+) in 12 tumors (67%) and equivocal (score 2+) in 4 tumors (33%), whereas none of the tumors were scored as HER2 3+. Score 2+ was associated with R678Q, L755S, I767M mutations, and ERBB2/HER2 rearrangement with a breakpoint in exon 23. Concurrent ERBB2/HER2 amplification was identified in 2 endometrial carcinomas, with HER2/CEP17 ratios of 3.1 and 3.5. We also queried the cBioportal database, which revealed 70 ERBB2/HER2-mutant gynecologic tumors with a total of 77 ERBB2/HER2 mutations, most often involving the active site of the tyrosine kinase domain (n=36; 46.8%), and the most common specific mutation was S310F (n=20; 26%), located in the extracellular domain. Our results provide important details regarding the clinicopathological and molecular associations of potentially actionable ERBB2/HER2 mutations in endometrial carcinoma and other gynecological cancer types and contribute to addressing clinical treatment needs and improving pathology testing recommendations in the future.