RESUMEN

BACKGROUND: Ductal carcinoma in situ (DCIS) is a non-obligate precursor to invasive breast cancer (IBC). Studies have indicated differences in DCIS outcome based on race or ethnicity, but molecular differences have not been investigated. METHODS: We examined the molecular profile of DCIS by self-reported race (SRR) and outcome groups in Black (n = 99) and White (n = 191) women in a large DCIS case-control cohort study with longitudinal follow up. RESULTS: Gene expression and pathway analyses suggested that different genes and pathways are involved in diagnosis and ipsilateral breast outcome (DCIS or IBC) after DCIS treatment in White versus Black women. We identified differences in ER and HER2 expression, tumor microenvironment composition, and copy number variations by SRR and outcome groups. CONCLUSIONS: Our results suggest that different molecular mechanisms drive initiation and subsequent ipsilateral breast events in Black versus White women.

Asunto(s)

RESUMEN

PURPOSE: Germline genetic mutations in women with phyllodes tumors (PT) are understudied, although some describe associations of PT with various mutations. We sought to determine the prevalence of pathogenic/likely pathogenic (P/LP) variants in women with PT. METHODS: A 6-site multi-center study of women with a PT was initiated, then expanded nationally through an online "Phyllodes Support Group." All women underwent 84-gene panel testing. We defined eligibility for testing based on select NCCN (National Comprehensive Cancer Network) criteria (v1.2022). Logistic regression was used to estimate the association of covariates with the likelihood of a P/LP variant. RESULTS: 274 women were enrolled: 164 (59.9%) through multi-center recruitment and 110 (40.1%) via online recruitment. 248 women completed testing; overall 14.1% (N = 35) had a P/LP variant, and over half (N = 19) of these individuals had a mutation in genes associated with autosomal dominant (AD) cancer conditions. The most common AD genes with a P/LP variant included CHEK2, ATM, and RAD51D. A quarter of participants (23.8%) met NCCN criteria for testing, but we found no difference in prevalence of a P/LP variant based on eligibility (p = 0.54). After adjustment, the presence of P/LP variants was not associated with age, NCCN testing eligibility, or PT type (all p > 0.05). CONCLUSION: Our study demonstrates that 7.7% of women with PT harbor germline P/LP variants in genes associated with AD cancer conditions. Early identification of these variants has implications for screening, risk reduction, and/or treatment. National guidelines for women with PT do not currently address germline genetic testing, which could be considered.

RESUMEN

Progression from pre-cancers like ductal carcinoma in situ (DCIS) to invasive disease (cancer) is driven by somatic evolution and is altered by clinical interventions. We hypothesized that genetic and/or phenotypic intra-tumor heterogeneity would predict clinical outcomes for DCIS since it serves as the substrate for natural selection among cells. We profiled two samples from two geographically distinct foci from each DCIS in both cross-sectional (N = 119) and longitudinal cohorts (N = 224), with whole exome sequencing, low-pass whole genome sequencing, and a panel of immunohistochemical markers. In the longitudinal cohorts, the only statistically significant predictors of time to non-invasive DCIS recurrence were the combination of treatment (lumpectomy only vs mastectomy or lumpectomy with radiation, HR = 12.13, p = 0.003, Wald test with FDR correction), ER status (HR = 0.16 for ER+ compared to ER-, p = 0.0045), and divergence in SNVs between the two samples (HR = 1.33 per 10% divergence, p = 0.018). SNV divergence also distinguished between pure DCIS and DCIS synchronous with invasive disease in the cross-sectional cohort. In contrast, the only statistically significant predictors of time to progression to invasive disease were the combination of the width of the surgical margin (HR = 0.67 per mm, p = 0.043) and the number of mutations that were detectable at high allele frequencies (HR = 1.30 per 10 SNVs, p = 0.02). These results imply that recurrence with DCIS is a clinical and biological process different from invasive progression.

RESUMEN

BACKGROUND: Breast-conserving surgery alone, breast-conserving surgery with adjuvant radiation treatment, and mastectomy are guideline-concordant treatments for ductal carcinoma in situ. The aim of this study was to compare survival outcomes between these treatment options. METHODS: A stratified random sample of patients diagnosed with pure ductal carcinoma in situ between 2008 and 2014 was selected from 1330 sites in the USA. Data on diagnosis, treatment, and follow-up were abstracted by local cancer registrars. Population-averaged marginal estimates of disease-specific survival and overall survival for breast-conserving surgery alone, breast-conserving surgery with radiation treatment, and mastectomy were obtained by combining sampling and overlap weights. RESULTS: A total of 18 442 women were included, with a median follow-up of 67.8 (interquartile range 46.1-93.5) months. A total of 35 women died from breast cancer, at a median age of 62 (interquartile range 50-74) years. Population-averaged 8-year rates of disease-specific survival were 99.6% or higher for all treatment groups, with no significant differences between groups (breast-conserving surgery alone versus breast-conserving surgery with radiation treatment, HR 1.19 (95% c.i. 0.29 to 4.85); and mastectomy versus breast-conserving surgery with radiation treatment, HR 1.74 (95% c.i. 0.53 to 5.72). There was no difference in overall survival between the patients who underwent a mastectomy and the patients who underwent breast-conserving surgery with radiation treatment (HR 1.09 (95% c.i. 0.83 to 1.43)). Patients who underwent breast-conserving surgery alone had lower overall survival compared with the patients who underwent breast-conserving surgery with radiation treatment (HR 1.29 (95% c.i. 1.00 to 1.67)). This survival difference vanished for all but one subgroup, namely patients less than 65 years (HR 1.86 (95% c.i. 1.15 to 3.00)). CONCLUSION: There was no statistically significant difference in disease-specific survival between women operated with breast-conserving surgery alone, breast-conserving surgery with radiation treatment, or mastectomy for ductal carcinoma in situ. Given the low absolute risk of disease-specific mortality, these results provide confidence in offering individualized locoregional treatment without fear of compromising survival.

Asunto(s)

RESUMEN

BACKGROUND: Older breast cancer patients represent a heterogeneous population. Studies demonstrate that sentinel lymph node biopsy (SLNB) omission may be appropriate in some clinical scenarios, yet patients with triple-negative breast cancer (TNBC) are often excluded from these studies. This study evaluated differences in treatment and survival for older patients with TNBC based on SLNB receipt and result. METHODS: Patients 70 years old or older with a diagnosis of cT1-2/cN0/M0 TNBC (2010-2019) who underwent surgery were selected from the National Cancer Database. Logistic regression estimated the association of SLNB with therapy, and Cox proportional hazards models estimated the association of SLNB with overall survival (OS) after adjustment for select factors. RESULTS: Of the 15,167 patients included in the study (median age, 77 years), 13.02% did not undergo SLNB, 5.14% had pN1 disease, 0.12% had pN2 disease, and 0.01% had pN3 disease. Most of the patients (83.9%) underwent surgery first, and 16.1% received neoadjuvant chemotherapy. Of those who underwent surgery first and SLNB, 6.2% had pN+ disease. Receipt of SLNB was associated with a higher likelihood of chemotherapy (odds ratio [OR] 1.85; 95% confidence interval [CI] 1.55-2.21), regardless of pN status. Compared with those who did not undergo a SLNB, a negative SLNB was significantly associated with lower mortality (hazard ratio [HR] 0.68; 95% CI 0.63-0.75), although there was no difference for a positive SLNB (HR 1.14; 95% CI 0.98-1.34). The patients receiving chemotherapy first showed no difference in survival based on SLNB receipt or result (p = 0.23). CONCLUSIONS: Most older patients with TNBC do not have nodal involvement and do not receive chemotherapy. The receipt and results of SLNB may be associated with outcomes for some who undergo surgery first, but not for those who receive neoadjuvant chemotherapy.

RESUMEN

BACKGROUND: Patients with inflammatory breast cancer (IBC) have worse survival compared with stage III non-IBC matched cohorts; however, the prognostic significance of achieving pathologic complete response (pCR) in the setting of IBC is not well described. We evaluated overall survival (OS) between IBC patients and non-IBC patients who achieved pCR. METHODS: Adult females diagnosed in 2010-2018 with clinical prognostic stage III unilateral invasive breast cancer treated with neoadjuvant chemotherapy (NAC) followed by surgery were selected from the National Cancer Database. Unadjusted OS from surgery was estimated using the Kaplan-Meier method, and log-rank tests were used to compare groups. Cox proportional hazard models were used to estimate the association of study groups with OS after adjustment for available covariates. RESULTS: The study included 38,390 patients; n = 4600 (12.0%) IBC and n = 33,790 (88.0%) non-IBC. Overall pCR rates were lower for IBC compared with non-IBC (20.7% vs. 23.3%; p < 0.001). Among those achieving pCR, 5-year mortality was higher for IBC patients (16.4%, 95% confidence interval [CI] 13.9-19.1%) versus non-IBC patients (9.1%, 95% CI 8.4-9.8%; log-rank p < 0.001). Among all patients achieving pCR, IBC remained associated with worse OS compared with non-IBC (hazard ratio 1.48, 95% CI 1.19-1.85; p < 0.001). CONCLUSION: We found a lower pCR rate and worse OS in IBC patients compared with non-IBC stage III patients. Despite effective systemic therapies, achieving a pCR for IBC patients may not carry the same prognostic impact compared with non-IBC stage III patients.

RESUMEN

BACKGROUND: This study compared incident breast cancer cases in the National Cancer Database (NCDB) and Surveillance, Epidemiology, End Results Program (SEER) to a national population cancer registry. METHODS: Patients with malignant or in situ breast cancer (BC) 2010-2019 in the NCDB and SEER were compared to the US Cancer Statistics (USCS). Case coverage was estimated as the number of patients in the NCDB/SEER as a proportion of USCS cases. RESULTS: The USCS reported 3,047,509 patients; 77.5% patients were included in the NCDB and 46.0% in SEER. Case ascertainment varied significantly by patient sex (both registries, p < .001). For males, 84.1% were captured in the NCDB, whereas only 77.5% of females were included. Case coverage in SEER was better for females than males (46.1% vs. 43.5%). Registries varied significantly by race/ethnicity (both p < .001). Case coverage in the NCDB was highest for non-Hispanic White (78.2%), non-Hispanic Black (77.7%), and non-Hispanic Asian or Pacific Islander (72.5%) BC patients, and lowest for Hispanic (56.4%) and non-Hispanic American Indian/Alaska Native (41.1%) patients. In SEER, case coverage was highest for non-Hispanic Asian or Pacific Islander (78.1%) and Hispanic (69.6%) patients and it was significantly lower for all other subgroups (non-Hispanic Black, 44.8%; non-Hispanic White, 42.4%; and non-Hispanic American Indian/Alaska Native, 36.6%). CONCLUSIONS: National US tumor registries provide data for a large sampling of breast cancer patients, yet significant differences in case coverage were observed based on age, sex, and race/ethnicity. These findings suggest that analyses using these data sets and interpretation of findings should account for these meaningful variances.

RESUMEN

BACKGROUND: The utility of sentinel lymph node biopsy (SLNB) in older patients remains controversial. Advancements in human epidermal growth factor receptor 2 (HER2)-directed therapy have revolutionized disease response rates and prognosis, supporting efforts to re-evaluate the utility of SLNB. We aimed to assess the differences in treatment and overall survival (OS) in older patients with HER2-positive breast cancer based on SLNB. METHODS: Using the National Cancer Database (2010-2020), patients ≥ 70 years of age diagnosed with cT1-2/cN0/M0, HER2-positive breast cancer were identified. Logistic regression assessed associations with SLNB, systemic therapy, and radiation. Cox proportional hazard models were used to identify factors associated with OS. Analyses were stratified by treatment sequence, i.e. upfront surgery or neoadjuvant therapy (NAT) followed by surgery. RESULTS: Of the 17,609 patients included, 94% underwent upfront surgery (n = 16,492) and the remaining underwent NAT (n = 1117). Those who underwent SLNB were more likely to receive adjuvant therapy, irrespective of nodal status {upfront surgery/systemic therapy (odds ratio [OR] 2.82, 95% confidence interval [CI] 2.17-3.67); upfront surgery/radiation (OR 3.97, 95% CI 3.03-5.21); NAT/radiation (OR 5.69, 95% CI 1.83-17.69)}. The breast pathologic complete response (pCR) rate was highest among the hormone receptor (HR)-negative/HER2-positive subtype (50.0%), of which none were found to be ypN+. Comorbidity burden was associated with significantly lower rates of adjuvant systemic therapy and worse OS. CONCLUSIONS: Patients who underwent SLNB, regardless of pN status, were more likely to receive adjuvant therapy. Nodal positivity is exceedingly rare for patients with a breast pCR following NAT, especially among the HR-negative/HER2-positive subtype. It is reasonable to consider omission of SLNB in select subgroups of older patients with HER2-positive breast cancer.

RESUMEN

BACKGROUND: Proliferative breast atypical lesions, including atypical ductal hyperplasia (ADH) and lobular intraepithelial neoplasms (LIN), represent benign entities that confer an elevated risk of ductal carcinoma in situ (DCIS) and invasive breast cancer (IBC). However, the timing of disease progression is variable and risk factors associated with the trajectory of disease are unknown. METHODS: Patients diagnosed with ADH or LIN from 1992 to 2017 at an academic center were identified. Early progression was defined as DCIS or IBC diagnosed within 5 years following the initial atypia diagnosis. Unadjusted cancer-free survival was estimated using the Kaplan-Meier method. Demographics, clinicopathologic features, and use of chemoprevention were compared between the early and late development groups. RESULTS: Overall, 418 patients were included-73.7% with ADH and 26.3% with LIN. Over a median follow up of 92.1 months, 71/418 (17.0%) patients developed IBC (57.7%) or DCIS (42.3%). Almost half (47.9%, 34/71) were diagnosed within 5 years of their initial atypia diagnosis, and 52.1% (37/71) were diagnosed after 5 years. Patient and atypia characteristics were not associated with rate of events or time to events. There was a trend of early events being more often ipsilateral (76.5% early vs. 54.1% late; p = 0.13) versus contralateral. CONCLUSIONS: In a large cohort of patients with breast atypia and long-term follow up, 17% experienced subsequent breast events, with approximately half of the events occurring within the first 5 years following the initial atypia diagnosis. Clinical features were not associated with the trajectory to subsequent events, supporting that atypia signals both local and overall malignancy risk.

RESUMEN

Ductal carcinoma in situ (DCIS) is a heterogeneous breast disease that remains challenging to treat due to its unpredictable progression to invasive breast cancer (IBC). Contemporary literature has become increasingly focused on extracellular matrix (ECM) alterations with breast cancer progression. However, the spatial regulation of the ECM proteome in DCIS has yet to be investigated in relation to IBC. We hypothesized that DCIS and IBC present distinct ECM proteomes that could discriminate between these pathologies. Tissue sections of pure DCIS, mixed DCIS-IBC, or pure IBC (n = 22) with detailed pathological annotations were investigated by multiplexed spatial proteomics. Across tissues, 1,005 ECM peptides were detected in pathologically annotated regions and their surrounding extracellular microenvironments. A comparison of DCIS to IBC pathologies demonstrated 43 significantly altered ECM peptides. Notably, eight fibrillar collagen peptides could distinguish with high specificity and sensitivity between DCIS and IBC. Lesion-targeted proteomic imaging revealed heterogeneity of the ECM proteome surrounding individual DCIS lesions. Multiplexed spatial proteomics reported an invasive cancer field effect, in which DCIS lesions in closer proximity to IBC shared a more similar ECM profile to IBC than distal counterparts. Defining the ECM proteomic microenvironment provides novel molecular insights relating to DCIS and IBC.

Asunto(s)

RESUMEN

Importance: Data on oncological outcomes after omission of axillary lymph node dissection (ALND) in patients with breast cancer that downstages from node positive to negative with neoadjuvant chemotherapy are sparse. Additionally, the best axillary surgical staging technique in this scenario is unknown. Objective: To investigate oncological outcomes after sentinel lymph node biopsy (SLNB) with dual-tracer mapping or targeted axillary dissection (TAD), which combines SLNB with localization and retrieval of the clipped lymph node. Design, Setting, and Participants: In this multicenter retrospective cohort study that was conducted at 25 centers in 11 countries, 1144 patients with consecutive stage II to III biopsy-proven node-positive breast cancer were included between April 2013 and December 2020. The cumulative incidence rates of axillary, locoregional, and any invasive (locoregional or distant) recurrence were determined by competing risk analysis. Exposure: Omission of ALND after SLNB or TAD. Main Outcomes and Measures: The primary end points were the 3-year and 5-year rates of any axillary recurrence. Secondary end points included locoregional recurrence, any invasive (locoregional and distant) recurrence, and the number of lymph nodes removed. Results: A total of 1144 patients (median [IQR] age, 50 [41-59] years; 78 [6.8%] Asian, 105 [9.2%] Black, 102 [8.9%] Hispanic, and 816 [71.0%] White individuals; 666 SLNB [58.2%] and 478 TAD [41.8%]) were included. A total of 1060 patients (93%) had N1 disease, 619 (54%) had ERBB2 (formerly HER2)-positive illness, and 758 (66%) had a breast pathologic complete response. TAD patients were more likely to receive nodal radiation therapy (85% vs 78%; P = .01). The clipped node was successfully retrieved in 97% of TAD cases and 86% of SLNB cases (without localization). The mean (SD) number of sentinel lymph nodes retrieved was 3 (2) vs 4 (2) (P < .001), and the mean (SD) number of total lymph nodes removed was 3.95 (1.97) vs 4.44 (2.04) (P < .001) in the TAD and SLNB groups, respectively. The 5-year rates of any axillary, locoregional, and any invasive recurrence in the entire cohort were 1.0% (95% CI, 0.49%-2.0%), 2.7% (95% CI, 1.6%-4.1%), and 10% (95% CI, 8.3%-13%), respectively. The 3-year cumulative incidence of axillary recurrence did not differ between TAD and SLNB (0.5% vs 0.8%; P = .55). Conclusions and Relevance: The results of this cohort study showed that axillary recurrence was rare in this setting and was not significantly lower after TAD vs SLNB. These results support omission of ALND in this population.

Asunto(s)

RESUMEN

Breast cancer is a significant global health concern affecting millions of women each year. Digital health platforms are an easily accessible intervention that can improve patient care, though their efficacy in breast cancer care is unknown. This scoping review aims to provide an overview of existing research on the utilization of digital health platforms for breast cancer care and identify key trends and gaps in the literature. A comprehensive literature search was conducted across electronic databases, including Ovid MEDLINE, Elsevier EMBASE, and Elsevier Scopus databases. The search strategy incorporated keywords related to "digital health platforms", "breast cancer care", and associated terminologies. After screening for eligibility, a total of 25 articles were included in this scoping review. The identified studies comprised mobile applications and web-based interventions. These platforms demonstrated various functionalities, including patient education, symptom monitoring, treatment adherence, and psychosocial support. The findings indicate the potential of digital health platforms in improving breast cancer care and patients' overall experiences. The positive impact on patient outcomes, including improved quality of life and reduced psychological distress, underscores the importance of incorporating digital health solutions into breast cancer management. Additional research is necessary to validate the effectiveness of these platforms in diverse patient populations and assess their impact on healthcare-resource utilization.

RESUMEN

PURPOSE: We sought to assess survival outcomes of patients with de novo metastatic breast cancer (dnMBC) who did not receive treatment irrespective of the reason. METHODS: Adults with dnMBC were selected from the NCDB (2010-2016) and stratified based on receipt of treatment (treated = received at least one treatment and untreated = received no treatments). Overall survival (OS) was estimated using the Kaplan-Meier method, and groups were compared. Cox proportional hazards models were used to identify factors associated with OS. RESULTS: Of the 53,240 patients with dnMBC, 92.1% received at least one treatment (treated), and 7.9% had no documented treatments, irrespective of the reason (untreated). Untreated patients were more likely to be older (median 68 y vs 61 y, p < 0.001), have higher comorbidity scores (p < 0.001), have triple-negative disease (17.8% vs 12.6%), and a higher disease burden (≥ 2 metastatic sites: 38.2% untreated vs 29.2% treated, p < 0.001). The median unadjusted OS in the untreated subgroup was 2.5 mo versus 36.4 mo in the treated subgroup (p < 0.001). After adjustment, variables associated with a worse OS in the untreated cohort included older age, higher comorbidity scores, higher tumor grade, and triple-negative (vs HR + /HER2-) subtype (all p < 0.05), while the number of metastatic sites was not associated with survival. CONCLUSIONS: Patients with dnMBC who do not receive treatment are more likely to be older, present with comorbid conditions, and have clinically aggressive disease. Similar to those who do receive treatment, survival in an untreated population is associated with select patient and disease characteristics. However, the prognosis for untreated dnMBC is dismal.

Asunto(s)

RESUMEN

BACKGROUND: The predictive and prognostic value of the recurrence score (RS) has emphasized the importance of tumor biology and has reduced the prognostic implications of limited nodal burden in post-menopausal women with HR+/HER2-invasive breast cancer (IBC). It is unclear whether routine axillary staging has a continued role in the management of small, clinically node negative (cN0) HR+/HER2- IBC. We sought to estimate the association of RS with pN stage. METHODS: Patients >50yo diagnosed with cN0, HR+/HER2- IBC (2015-2019) with an available RS were identified from the National Cancer Database. The clinicopathologic characteristics and rates of pN-stage (pN0, pN1, pN2/3) were compared for RS of ≤25 vs. >25. RESULTS: The median patient age was 64.1 (IQR 58-69) and the majority (75%) of tumors displayed ductal histology. Most (81.6%) were cT1 on presentation and pT1 (74.7%) on final pathology. There were 130,568 (86.2%) with a RS â€‹≤ â€‹25 and 20,879 (13.8%) with a RS â€‹> â€‹25. On final pathology, 128,995 (85.2%) were pN0 and 21,991 (14.5%) pN1. Of the pN1, 2699 (12.3%) yielded a RS â€‹> â€‹25. There were 461 (0.3%) patients with pN2-pN3 disease. Of those, 57 (12.4%) had RS â€‹> â€‹25. CONCLUSION: In our analysis, pN0 and pN1 tumors are biologically similar by gene expression assay in postmenopausal patients with similar proportions of high RS. These data support the notion that tumor biology examined via RS may have more prognostic and predictive value than metastatic dissemination to limited lymph nodes. These findings support the ongoing evaluation of routine axillary staging in postmenopausal patients with HR+/HER2- IBC.

Asunto(s)

RESUMEN

PURPOSE: Black women have higher rates of death from triple-negative breast cancer (TNBC) than White women. We hypothesized that pathologic complete response (pCR) to neoadjuvant chemotherapy (NAC) and overall survival (OS) may vary by race/ethnicity in patients with TNBC. METHODS: We identified women 18 years and older with stage I-III TNBC who received NAC followed by surgery from the National Cancer Database (2010-2019). We excluded patients without race/ethnicity or pathology data. Primary outcomes were pCR rates and OS on the basis of race/ethnicity. RESULTS: Forty thousand eight hundred ninety women with TNBC met inclusion criteria (median age [IQR], 53 [44-61] years): 26,150 Non-Hispanic White (64%, NHW), 9,672 Non-Hispanic Black (23.7%, NHB), 3,267 Hispanic (8%), 1,368 Non-Hispanic Asian (3.3%, NHA), and 433 Non-Hispanic Other (1.1%, NHO) patients. Overall, 29.8% demonstrated pCR (NHW: 30.5%, NHB: 27%, Hispanic: 32.6%, NHA: 28.8%, NHO: 29.8%). Unadjusted OS was significantly higher for those with pCR compared with those with residual disease (5-year OS, 0.917 [95% CI, 0.911 to 0.923] v 0.667 [95% CI, 0.661 to 0.673], log-rank P < .001), and this association persisted after adjustment for demographic and tumor factors. The effect of achieving pCR on OS did not differ by race/ethnicity (interaction P = .10). However, NHB patients were less likely (odds ratio [OR], 0.89 [95% CI, 0.83 to 0.95], P = .001) and Hispanic patients were more likely (OR, 1.19 [95% CI, 1.08 to 1.31], P = .001) to achieve pCR than NHW patients. After adjustment for patient and disease factors, including achievement of pCR, Hispanic (hazard ratio [HR], 0.76 [95% CI, 0.69 to 0.85], P < .001) and NHA (HR, 0.64 [95% CI, 0.55 to 0.75], P < .001) race/ethnicity remained associated with OS. CONCLUSION: Odds of achieving pCR and OS in patients with TNBC appear to be associated with race/ethnicity. Additional research is necessary to understand how race/ethnicity is associated with rates of pCR and OS, whether related to socioeconomic factors or biologic variables, or both.

Asunto(s)

RESUMEN

OBJECTIVES: To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study. METHODS: Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n = 400) and test cases (n = 300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features. RESULTS: Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance. CONCLUSIONS: In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings. ADVANCES IN KNOWLEDGE: Performance bias can result from model testing when using limited datasets. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Asunto(s)