Predicting benefit from immune checkpoint inhibitors in patients with non-small-cell lung cancer by CT-based ensemble deep learning: a retrospective study.

BACKGROUND: Only around 20-30% of patients with non-small-cell lung cancer (NCSLC) have durable benefit from immune-checkpoint inhibitors. Although tissue-based biomarkers (eg, PD-L1) are limited by suboptimal performance, tissue availability, and tumour heterogeneity, radiographic images might holistically capture the underlying cancer biology. We aimed to investigate the application of deep learning on chest CT scans to derive an imaging signature of response to immune checkpoint inhibitors and evaluate its added value in the clinical context. METHODS: In this retrospective modelling study, 976 patients with metastatic, EGFR/ALK negative NSCLC treated with immune checkpoint inhibitors at MD Anderson and Stanford were enrolled from Jan 1, 2014, to Feb 29, 2020. We built and tested an ensemble deep learning model on pretreatment CTs (Deep-CT) to predict overall survival and progression-free survival after treatment with immune checkpoint inhibitors. We also evaluated the added predictive value of the Deep-CT model in the context of existing clinicopathological and radiological metrics. FINDINGS: Our Deep-CT model demonstrated robust stratification of patient survival of the MD Anderson testing set, which was validated in the external Stanford set. The performance of the Deep-CT model remained significant on subgroup analyses stratified by PD-L1, histology, age, sex, and race. In univariate analysis, Deep-CT outperformed the conventional risk factors, including histology, smoking status, and PD-L1 expression, and remained an independent predictor after multivariate adjustment. Integrating the Deep-CT model with conventional risk factors demonstrated significantly improved prediction performance, with overall survival C-index increases from 0·70 (clinical model) to 0·75 (composite model) during testing. On the other hand, the deep learning risk scores correlated with some radiomics features, but radiomics alone could not reach the performance level of deep learning, indicating that the deep learning model effectively captured additional imaging patterns beyond known radiomics features. INTERPRETATION: This proof-of-concept study shows that automated profiling of radiographic scans through deep learning can provide orthogonal information independent of existing clinicopathological biomarkers, bringing the goal of precision immunotherapy for patients with NSCLC closer. FUNDING: National Institutes of Health, Mark Foundation Damon Runyon Foundation Physician Scientist Award, MD Anderson Strategic Initiative Development Program, MD Anderson Lung Moon Shot Program, Andrea Mugnaini, and Edward L C Smith.

Outcomes of breakthrough COVID-19 infections in patients with hematologic malignancies.

Patients with hematologic malignancies have both an increased risk for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections and higher morbidity/mortality. They have lower seroconversion rates after vaccination, potentially leading to inferior coronavirus disease 2019 (COVID-19) outcomes, despite vaccination. We consequently evaluated the clinical outcomes of COVID-19 infections in 243 vaccinated and 175 unvaccinated patients with hematologic malignancies. Hospitalization rates were lower in the vaccinated group when compared with the unvaccinated group (31.3% vs 52.6%). However, the rates of COVID-19-associated death were similar at 7.0% and 8.6% in vaccinated and unvaccinated patients, respectively. By univariate logistic regression, females, older patients, and individuals with higher modified Charlson Comorbidity Index scores were at a higher risk of death from COVID-19 infections. To account for the nonrandomized nature of COVID-19 vaccination status, a propensity score weighting approach was used. In the final propensity-weighted model, vaccination status was not significantly associated with the risk of death from COVID-19 infections but was associated with the risk of hospitalization. The predicted benefit of vaccination was an absolute decrease in the probability of death and hospitalization from COVID-19 infections by 2.3% and 22.9%, respectively. In conclusion, COVID-19 vaccination status in patients with hematologic malignancies was associated with a decreased risk of hospitalization but not associated with a decreased risk of death from COVID-19 infections in the pre-Omicron era. Protective strategies, in addition to immunization, are warranted in this vulnerable patient population.

Association between abnormal lipid profile and inflammation and progression of myelodysplastic syndrome to acute leukemia.

Clonal hematopoiesis of indeterminate potential (CHIP) is associated with a small risk of developing hematologic malignancies and a higher risk of cardiovascular diseases (CVD). We asked whether the reverse correlation exists and cardiometabolic risk factors have an impact on the progression of myelodysplastic syndrome (MDS) to acute myeloid leukemia (AML). We investigated the association between abnormal lipid profiles and inflammation in MDS, which shares many genetic mutations with CHIP, and the risk of developing acute leukemia. We examined the medical records of 11071 MDS patients. Among them, 5422 had at least one lipid profile or C-reactive protein (CRP) measurement. In univariate and multivariate analyses, elevated triglyceride and high-sensitive C-reactive protein (HS-CRP) were significantly associated with a diagnosis of acute leukemia in MDS patients. Next, we examined these associations in patients with available MDS prognostic scores (International Prognostic Scoring System, IPSS, or its revised version IPSS/R) (n = 2786 patients). We found that the statistical association between CRP and the progression of MDS to leukemia was independent of other variables in the scoring system. MDS patients with elevated CRP in both the high-risk and low-risk groups had a higher risk of progression to AML than those with a lower CRP. We speculate that inflammation might be a common denominator in developing hematologic malignancies and CVD in patients with clonal hematopoiesis.

External validation of the HIGH-2-LOW model: A predictive score for venous thromboembolism after allogeneic transplant.

In patients undergoing hematopoietic cell transplantation (HCT), venous thromboembolism (VTE) remains a serious complication that lacks validated risk assessment models (RAMs) to guide thromboprophylaxis. To address this dilemma, we performed a temporal and external validation study of the recently derived HIGH-2-LOW RAM. We selected adult patients undergoing allogeneic HCT from Fred Hutchinson Cancer Research Center (FHCRC) and MD Anderson Cancer Center (MDACC). Patients who died, received anticoagulation, or did not engraft platelets by day 30 were excluded. Primary outcomes were defined as overall VTE and pulmonary embolism ± lower-extremity deep venous thromboembolism (PE/LE-DVT) by day 180. Covariates were weighted according to the original model, except that grade 2-4 GVHD was substituted for grade 3-4. Discrimination and calibration were assessed. A total of 765 patients from FHCRC and 954 patients from MDACC were included. Incident VTE by day 180 was 5.1% at FHCRC and 6.8% at MDACC. The HIGH-2-LOW score had a c-statistic of 0.67 (0.59-0.75) for VTE and 0.75 (0.64-0.81) for PE/LE-DVT at FHCRC and 0.62 (0.55-0.70) for VTE and 0.70 (0.56-0.83) for PE/LE-DVT at MDACC. Twenty-five percent and 23% of patients were classified as high risk (2+ points) in the two cohorts, respectively. High versus low-risk was associated with odds ratio (OR) of 2.80 (1.46-5.38) for VTE and 4.21 (1.82-9.77) for PE/LE-DVT at FHCRC and OR of 3.54 (2.12-5.91) for VTE and 6.82 (2.30-20.16) for PE-LE-DVT at MDACC. The HIGH-2-LOW RAM identified allogeneic HCT recipients at high risk for VTE in both validation cohorts. It can improve evidence-based decision-making for thromboprophylaxis post-transplant.

Habitat Imaging Biomarkers for Diagnosis and Prognosis in Cancer Patients Infected with COVID-19.

OBJECTIVES: Cancer patients have worse outcomes from the COVID-19 infection and greater need for ventilator support and elevated mortality rates than the general population. However, previous artificial intelligence (AI) studies focused on patients without cancer to develop diagnosis and severity prediction models. Little is known about how the AI models perform in cancer patients. In this study, we aim to develop a computational framework for COVID-19 diagnosis and severity prediction particularly in a cancer population and further compare it head-to-head to a general population. METHODS: We have enrolled multi-center international cohorts with 531 CT scans from 502 general patients and 420 CT scans from 414 cancer patients. In particular, the habitat imaging pipeline was developed to quantify the complex infection patterns by partitioning the whole lung regions into phenotypically different subregions. Subsequently, various machine learning models nested with feature selection were built for COVID-19 detection and severity prediction. RESULTS: These models showed almost perfect performance in COVID-19 infection diagnosis and predicting its severity during cross validation. Our analysis revealed that models built separately on the cancer population performed significantly better than those built on the general population and locked to test on the cancer population. This may be because of the significant difference among the habitat features across the two different cohorts. CONCLUSIONS: Taken together, our habitat imaging analysis as a proof-of-concept study has highlighted the unique radiologic features of cancer patients and demonstrated effectiveness of CT-based machine learning model in informing COVID-19 management in the cancer population.

Effect of adjuvant/neoadjuvant trastuzumab on clinical outcomes in patients with HER2-positive metastatic breast cancer.

BACKGROUND: The purpose of the current study was to describe the outcomes of patients with human epidermal growth factor receptor 2 (HER2)-overexpressed/amplified (HER2+) early breast cancer who received adjuvant or neoadjuvant trastuzumab-based therapy and were subsequently retreated with trastuzumab for metastatic disease. METHODS: A total of 353 patients with metastatic HER2+ breast cancer who were treated with trastuzumab as part of their first-line treatment for metastatic disease were identified. A total of 75 patients had received adjuvant or neoadjuvant trastuzumab-based therapy for early breast cancer, and 278 had not. Clinical outcomes of patients who had or had not received prior trastuzumab were compared using Cox proportional hazards regression and logistic regression analyses. Survival was estimated using the Kaplan-Meier method. RESULTS: The clinical benefit (complete response, partial response, or stable disease of ≥ 6 months) rates were 71% in the group who did not receive prior trastuzumab and 39% in the group previously treated with trastuzumab. The adjusted odds ratios were 0.28 (95% confidence interval [95% CI], 0.13-0.59; P = .0009) for clinical benefit rates and 0.39 (95% CI, 0.18-0.82; P = .038) for objective (complete or partial) response rates. In the univariate analysis, the median overall survival rate was longer in the group who did not receive prior trastuzumab (36 months vs 28 months) (hazards ratio, 1.47; 95% CI, 1.07-2.01 [P = .022]). The multivariate analysis found no significant difference in overall survival. CONCLUSIONS: When treated with trastuzumab for metastatic disease, patients with HER2+ breast cancer without prior exposure to trastuzumab were found to have superior clinical outcomes to those with prior exposure. Prior trastuzumab exposure should be considered in treatment algorithms and in HER2-targeted clinical trial enrollment for metastatic disease.

Elevated serum P1NP predicts development of bone metastasis and survival in early-stage breast cancer.

Bone is the most common site of metastasis of breast cancer, affecting most women with advanced disease. Procollagen type I N-terminal propeptide (P1NP), osteocalcin, CTX, and IL-6 are markers of bone turnover. Our objective was to determine whether serum levels of these proteins have clinical utility as predictors of breast cancer metastasis to bone. Blood was collected before treatment from 164 patients with stage I-III breast cancer from September 2001 to December 2008. Serum levels of P1NP, CTX, IL-6, and OC were measured using an automated immunoassay system. Correlations of the levels of these markers with time to bone metastasis development and with overall survival (OS) rate were assessed using Cox proportional hazards regression analysis and the Kaplan-Meier method. Fifty-five patients with stage I-III disease at the time of blood sample collection subsequently experienced metastasis to bone. A baseline P1NP level of at least 75 ng/mL predicted increased risk of bone metastasis (hazard ratio, 2.7 [95 % confidence interval, 1.2-6.0]; P = 0.031) and a poor OS rate (P = 0.031). Serum P1NP levels at or above 75 ng/mL correlate with a short time to development of bone metastasis and low overall survival in patients with stage I-III breast cancer.

Mutation profiling identifies numerous rare drug targets and distinct mutation patterns in different clinical subtypes of breast cancers.

The mutation pattern of breast cancer molecular subtypes is incompletely understood. The purpose of this study was to identify mutations in genes that may be targeted with currently available investigational drugs in the three major breast cancer subtypes (ER+/HER2-, HER2+, and Triple Negative). We extracted DNA from fine needle aspirations of 267 stage I-III breast cancers. These tumor specimens typically consisted of >80% neoplastic cells. We examined 28 genes for 163 known cancer-related nucleic acid variations by Sequenom technology. We observed at least one mutation in 38 alleles corresponding to 15 genes in 108 (40%) samples, including PIK3CA (16.1% of all samples), FBXW7 (8%), BRAF (3.0%), EGFR (2.6%), AKT1 and CTNNB1 (1.9% each), KIT and KRAS (1.5% each), and PDGFR-α (1.1%). We also checked for the polymorphism in PHLPP2 that is known to activate AKT and it was found at 13.5% of the patient samples. PIK3CA mutations were more frequent in estrogen receptor-positive cancers compared to triple negative breast cancer (TNBC) (19 vs. 8%, p=0.001). High frequency of PIK3CA mutations (28%) were also found in HER2+ breast tumors. In TNBC, FBXW7 mutations were significantly more frequent compared to ER+ tumors (13 vs. 5%, p=0.037). We performed validation for all mutated alleles with allele-specific PCR or direct sequencing; alleles analyzed by two different sequencing techniques showed 95-100% concordance for mutation status. In conclusion, different breast cancer subtypes harbor different type of mutations and approximately 40 % of tumors contained individually rare mutations in signaling pathways that can be potentially targeted with drugs. Simultaneous testing of many different mutations in a single needle biopsy is feasible and allows the design of prospective clinical trials that could test the functional importance of these mutations in the future.

Agreement in risk prediction between the 21-gene recurrence score assay (Oncotype DX®) and the PAM50 breast cancer intrinsic Classifier™ in early-stage estrogen receptor-positive breast cancer.

PURPOSE: To compare risk assignment by PAM50 Breast Cancer Intrinsic Classifier™ and Oncotype DX_Recurrence Score (RS) in the same population. METHODS: RNA was extracted from 151 estrogen receptor (ER)+ stage I-II breast cancers and gene expression profiled using PAM50 "intrinsic" subtyping test. RESULTS: One hundred eight cases had complete molecular information; 103 (95%) were classified as luminal A (n = 76) or luminal B (n = 27). Ninety-two percent (n = 98) had a low (n = 59) or intermediate (n = 39) RS. Among luminal A cancers, 70% had low (n = 53) and the remainder (n = 23) had an intermediate RS. Among luminal B cancers, nine were high (33%) and 13 were intermediate (48%) by the RS. Almost all cancers with a high RS were classified as luminal B (90%, n = 9). One high RS cancer was identified as basal-like and had low ER/ESR1 and low human epidermal growth factor receptor 2 (HER2) expression by quantitative polymerase chain reaction in both assays. The majority of low RS cases were luminal A (83%, n = 53). Importantly, half of the intermediate RS cancers were re-categorized as low risk luminal A subtype by PAM50. CONCLUSION: There is good agreement between the two assays for high (i.e., luminal B or RS > 31) and low (i.e., luminal B or RS < 18) prognostic risk assignment but PAM50 assigns more patients to the low risk category. About half of the intermediate RS group was reclassified as luminal A by PAM50.

First generation prognostic gene signatures for breast cancer predict both survival and chemotherapy sensitivity and identify overlapping patient populations.

The aims of this study were to compare the performance of six different genomic prognostic markers to predict long-term survival and chemotherapy response on the same patient cohort and assess if clinicopathological variables carry independent prognostic and predictive values. We examined seven clinical variables and six previously described prognostic signatures on 228 tumors from patients who received homogeneous preoperative chemotherapy and had long-term follow-up information for survival. We used the area under the receiver operator characteristic curve (AUC) to compare predictors and also performed univariate and multivariate analyses including the genomic and clinical variables and plotted Kaplan-Meir survival curves. All genomic prognostic markers had statistically similar AUCs and sensitivity to predict 5-year progression-free survival (PFS, sensitivities ranged from 0.591 to 0.773, and AUCs: 0.599-0.673), overall survival (OS, sensitivities: 0.590-0.769, AUCs: 0.596-0.684) and pathologic complete response (pCR, sensitivities: 0.596-0.851, AUCs: 0.614-0.805). In multivariate analysis, the genomic markers were not independent from one another; however, estrogen receptor (Odds Ratio [OR] 7.63, P < 0.001) and HER2 status (OR: 0.37, P = 0.021) showed significant independent predictive values for pCR. Nodal status remained an independent prognostic, but not predictive, variable (OR for PFS: 2.77, P = 0.021, OR for OS: 3.62, P = 0.01). There was moderate to good agreement between different prediction results in pair-wise comparisons. First-generation prognostic-gene signatures predict both chemotherapy response and long-term survival. When multiple predictors are applied to the same case discordant risk prediction frequently occurs.