Translation of AI into oncology clinical practice

Artificial intelligence (AI) is a transformative technology that is capturing popular imagination and can revolutionize biomedicine. AI and machine learning (ML) algorithms have the potential to break through existing barriers in oncology research and practice such as automating workflow processes, personalizing care, and reducing healthcare disparities. Emerging applications of AI/ML in the literature include screening and early detection of cancer, disease diagnosis, response prediction, prognosis, and accelerated drug discovery. Despite this excitement, only few AI/ML models have been properly validated and fewer have become regulated products for routine clinical use. In this review, we highlight the main challenges impeding AI/ML clinical translation. We present different clinical use cases from the domains of radiology, radiation oncology, immunotherapy, and drug discovery in oncology. We dissect the unique challenges and opportunities associated with each of these cases. Finally, we summarize the general requirements for successful AI/ML implementation in the clinic, highlighting specific examples and points of emphasis including the importance of multidisciplinary collaboration of stakeholders, role of domain experts in AI augmentation, transparency of AI/ML models, and the establishment of a comprehensive quality assurance program to mitigate risks of training bias and data drifts, all culminating toward safer and more beneficial AI/ML applications in oncology labs and clinics.

Health-related quality of life and vulnerability among people with myelodysplastic syndromes: a US national study

Health-related quality of life (HRQoL) and vulnerability are variably affected in patients with myelodysplastic syndromes (MDS) and other cytopenic states; however, the heterogeneity of these diseases has limited our understanding of these domains. The National Heart, Lung, and Blood Institute-sponsored MDS Natural History Study is a prospective cohort enrolling patients undergoing workup for suspected MDS in the setting of cytopenias. Untreated patients undergo bone marrow assessment with central histopathology review for assignment as MDS, MDS/myeloproliferative neoplasm (MPN), idiopathic cytopenia of undetermined significance (ICUS), acute myeloid leukemia (AML) with <30% blasts, or "At-Risk." HRQoL data are collected at enrollment, including the MDS-specific Quality of Life in Myelodysplasia Scale (QUALMS). Vulnerability is assessed with the Vulnerable Elders Survey. Baseline HRQoL scores from 449 patients with MDS, MDS/MPN, AML <30%, ICUS or At-Risk were similar among diagnoses. In MDS, HRQoL was worse for vulnerable participants (eg, mean Patent-Reported Outcomes Management Information Systems [PROMIS] Fatigue of 56.0 vs 49.5; P < .001) and those with worse prognosis (eg, mean Euroqol-5 Dimension-5 Level [EQ-5D-5L] of 73.4, 72.7, and 64.1 for low, intermediate, and high-risk disease; P = .005). Among vulnerable MDS participants, most had difficulty with prolonged physical activity (88%), such as walking a quarter mile (74%). These data suggest that cytopenias leading to MDS evaluation are associated with similar HRQoL, regardless of eventual diagnosis, but with worse HRQoL among the vulnerable. Among those with MDS, lower-risk disease was associated with better HRQoL, but the relationship was lost among the vulnerable, showing for the first time that vulnerability trumps disease risk in affecting HRQoL. This study is registered at www.clinicaltrials.gov as NCT02775383.

Enabling Precision Medicine in Cancer Care Through a Molecular Data Warehouse: The Moffitt Experience

PURPOSE

The use of genomics within cancer research and clinical oncology practice has become commonplace. Efforts such as The Cancer Genome Atlas have characterized the cancer genome and suggested a wealth of targets for implementing precision medicine strategies for patients with cancer. The data produced from research studies and clinical care have many potential secondary uses beyond their originally intended purpose. Effective storage, query, retrieval, and visualization of these data are essential to create an infrastructure to enable new discoveries in cancer research.

METHODS

Moffitt Cancer Center implemented a molecular data warehouse to complement the extensive enterprise clinical data warehouse (Health and Research Informatics). Seven different sequencing experiment types were included in the warehouse, with data from institutional research studies and clinical sequencing.

RESULTS

The implementation of the molecular warehouse involved the close collaboration of many teams with different expertise and a use case-focused approach. Cornerstones of project success included project planning, open communication, institutional buy-in, piloting the implementation, implementing custom solutions to address specific problems, data quality improvement, and data governance, unique aspects of which are featured here. We describe our experience in selecting, configuring, and loading molecular data into the molecular data warehouse. Specifically, we developed solutions for heterogeneous genomic sequencing cohorts (many different platforms) and integration with our existing clinical data warehouse.

CONCLUSION

The implementation was ultimately successful despite challenges encountered, many of which can be generalized to other research cancer centers.

Abstract A32: Process improvement in online consenting for the Moffitt Cancer Center Total Cancer Care biobanking protocol

Introduction: Broad-based institutional biobanking protocols are crucial for advancing personalized medicine in cancer research. However, in-person consenting is resource intensive and interrupts clinic flow. Ethical standards support online consenting for biobanking protocols, which has the potential to reach more participants for enrollment. Herein, we discuss challenges in implementing broad-based online patient consenting at the Moffitt Cancer Center (MCC) for our institutional biobanking protocol, Total Cancer Care (TCC), and subsequent steps to better integrate this process with clinical care on-boarding. We hypothesized that implementation of online patient consenting would broaden patient enrollment into TCC and the redesigned process would increase the number of online patient consents by at least 5-fold.

Methods: TCC started as the institutional biobanking protocol in 2006. TCC online consent was implemented in Feb. 2013 (Phase 1) and redesigned in Oct. 2018 (Phase 2). Phase 1 was developed with the guidance of ethicists and patient advisors, and consisted of (a) communication via email, (b) education online with text, video, and FAQs, (c) review and initialing of the Informed Consent form (ICF), and (d) typed name signature on ICF. The Phase 2 redesign created a more streamlined process that was integrated into the MCC New Patient To-Do list, a clinical on-boarding platform. The revised version used interactive material, a shorter animated video, and an electronically signed ICF. We compared the number of online patient consents in Phase 1 and 2 and have preliminary data on patient acceptance of Phase 2.

Results: Since 2006, 108,898 MCC patients enrolled into TCC. 342 patients enrolled online in Phase I between 2013–2018 (5/mo). In the initial 6 weeks of Phase 2, there was a 13-fold increase in online consenting (n=98, 65/mo) compared to Phase 1. According to web analytics, TCC pages in Phase 2 were viewed 991 times compared to 432 views over 6 weeks in Phase 1. In general, patient characteristics were similar between those who enrolled online and those who enrolled in person during the same timeframe, with a slightly higher percentage of online vs. in-person enrollees being female (57% vs. 53%, respectively). In a small survey of new patients who viewed the clinical on-boarding platform in Phase 2, 76% reviewed the TCC material without issue whereas 20% did not review the TCC material.

Conclusions: While online consenting has tremendous potential, we had very low uptake of patient education and consenting in our initial design (Phase 1). Challenges included a lack of visibility within the patient portal and a dependency on email as the primary mode of directing patients to online consenting. Alternatively, we have seen initial success with our second phase of online consenting that was directly integrated with patient on-boarding. Initial success highlights the potential of this modality to increase patient enrollment into biobanking studies. Further evaluation of this approach is ongoing.

Citation Format: Erin M. Siegel, Kyle P. Hawkins, Lynne Hildreth, Timothy Grose, David Stringfellow, Amanda Bloomer, Shelley Tworoger, Dana Rollison, Scott Gilbert, Thomas A. Sellers. Process improvement in online consenting for the Moffitt Cancer Center Total Cancer Care biobanking protocol [abstract]. In: Proceedings of the AACR Special Conference on Modernizing Population Sciences in the Digital Age; 2019 Feb 19-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(9 Suppl):Abstract nr A32.

Abstract 2101: Deep learning for automatic extraction of tumor site and histology from unstructured pathology reports

Introduction: Much of the information in electronic medical records (EMRs) required for the practice of clinical oncology is contained in unstructured text. While natural language processing (NLP) has been used to extract information from EMR text, accuracy is suboptimal. In late 2018 a powerful new deep-learning NLP algorithm was published: Bidirectional Encoder Representations from Transformers (BERT). BERT set new accuracy records and for the first time achieved human-level performance on several NLP benchmarks. Our goal was to train BERT to extract clinically relevant data from pathology reports with high accuracy.

Procedures: Like many cancer centers nationwide, Moffitt Cancer Center employs Certified Tumor Registrars (CTRs) to collect and report data about cancer patients to state and federal agencies. The CTR extracted data are labels that identify, with high accuracy, important information in each pathology report. Consequently, we used this data to tune BERT to perform a question-and-answering (Q&A) task. Our system sought the answers to 2 predetermined questions in each pathology report: “What organ contains the tumor?”, and “What is the kind of tumor or carcinoma?”

To achieve this, we matched surgical pathology reports created at Moffitt from January 1, 2007 onwards with structured data extracted by CTRs. The resulting dataset was randomly divided into training (80%) and testing (20%) subsets.

After Q&A training, model performance was assessed using the test dataset. Two metrics were calculated for each question: a true-or-false indication of a perfect word-for-word match between the BERT-extracted data and CTR-extracted data; and, the F1 statistic. The latter produces a value between 0% and 100% indicating the degree of overlap between words in the BERT-extracted data and words in the CTR-extracted data.

Results: The final dataset contained 14,143 pathology reports (11,520 for training, 2,623 for testing). This dataset included tumors from 228 organ sites involving 232 histological classifications. The three most common organ sites / histological classifications were: Prostate Gland / Adenocarcinoma (6.7%); Breast / Invasive Carcinoma (6.1%); and, Breast Overlapping Lesion / Invasive Carcinoma (5.9%).

Our BERT-based Q&A system searched for answers to both questions in each test report. Thus, a total of 5,246 answers were generated. Of these, 4,667 (89%) were a perfect word-for-word match with the corresponding CTR extracted phrases. The mean F1 statistic between the BERT answers and the CTR extracted phrases was 92%.

Conclusions: Future efforts will focus on improving performance via unsupervised training of the BERT language model using 484,000 Moffitt pathology reports. We will also extract additional data fields with CTR-matched ground truth labels. Ultimately new NLP transformer models could aid extraction of information from pathology reports and other EMR documents. This, in turn, could greatly facilitate personalized medicine.

Citation Format: Ross Mitchell, Rachel Howard, Patricia Lewis, Katie Fellows, Jennie Jones, Phillip Reisman, Brooke Fridley, Dana Rollison. Deep learning for automatic extraction of tumor site and histology from unstructured pathology reports [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2101.

Abstract 3226: Facilitating personalized medicine with cloud-based storage and analytics

The vast wealth of medical data collected over the last decade holds great promise for accelerating novel research, discovery, and clinical translation. Specifically, the rapid expansion of genomic testing provides new opportunities for the clinical management of cancer patients, influencing diagnosis, risk stratification, and treatment planning. Moffitt Cancer Center's Personalized Medicine Clinical Service integrates next-generation sequencing test results into patient care, using the data to guide individualized treatment plans. To maximize the efficiency and efficacy of this service, creative solutions for data harmonization, storage, and management are required. We implemented a commercial molecular data warehouse (MDW), directly linked to our existing clinical data warehouse, to store and manage molecular data ranging from genotypic alterations to annotations from public resources (HUGO, COSMIC, Ensembl) and clinically actionable targets (4,256 records currently loaded). A centralized, cloud-based data and analytics platform is also being implemented at Moffitt that will integrate a broad range of multi-modal data. In the cloud environment, the data from the MDW will be linked to typically siloed data streams from the electronic health record, cancer registry management system, biospecimen management system, billing and scheduling systems, patient-reported information and outcomes, and patient-generated health data, creating a unique and customized Personalized Medicine Curated Data Mart (CDM). In addition to describing the features of the MDW and the challenges faced during its implementation, we will provide an overview of the extensive data cleaning and curation required to facilitate such a CDM. This includes the extraction of disease characteristics from unstructured clinical text via natural language processing, creation of new derived data fields, approaches to extracting and managing complex treatment data, and the inclusion of detailed, manually-abstracted recurrence and outcomes data for historical patients from existing institutional datasets such as the Clinical Genomics Action Committee (CGAC) database. Finally, we will present prototypes of analytics dashboards that will interface directly with our CDM, facilitating intuitive data exploration for all members of our personalized medicine teams.

Citation Format: Rachel Howard, Kevin Hicks, Jamie Teer, Phillip Reisman, Mandy O'Leary, Steven Eschrich, Ross Mitchell, Howard McLeod, Dana Rollison. Facilitating personalized medicine with cloud-based storage and analytics [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3226.

Innovations in research and clinical care using patient-generated health data

Patient-generated health data (PGHD), or health-related data gathered from patients to help address a health concern, are used increasingly in oncology to make regulatory decisions and evaluate quality of care. PGHD include self-reported health and treatment histories, patient-reported outcomes (PROs), and biometric sensor data. Advances in wireless technology, smartphones, and the Internet of Things have facilitated new ways to collect PGHD during clinic visits and in daily life. The goal of the current review was to provide an overview of the current clinical, regulatory, technological, and analytic landscape as it relates to PGHD in oncology research and care. The review begins with a rationale for PGHD as described by the US Food and Drug Administration, the Institute of Medicine, and other regulatory and scientific organizations. The evidence base for clinic-based and remote symptom monitoring using PGHD is described, with an emphasis on PROs. An overview is presented of current approaches to digital phenotyping or device-based, real-time assessment of biometric, behavioral, self-report, and performance data. Analytic opportunities regarding PGHD are envisioned in the context of big data and artificial intelligence in medicine. Finally, challenges and solutions for the integration of PGHD into clinical care are presented. The challenges include electronic medical record integration of PROs and biometric data, analysis of large and complex biometric data sets, and potential clinic workflow redesign. In addition, there is currently more limited evidence for the use of biometric data relative to PROs. Despite these challenges, the potential benefits of PGHD make them increasingly likely to be integrated into oncology research and clinical care.

Nonmelanoma skin cancer and risk of all-cause and cancer-related mortality: a systematic review

Some reports suggest that a history of nonmelanoma skin cancer (NMSC) may be associated with increased mortality. NMSCs have very low fatality rates, but the high prevalence of NMSC elevates the importance of the possibility of associated subsequent mortality from other causes. The variable methods and findings of existing studies leave the significance of these results uncertain. To provide clarity, we conducted a systematic review to characterize the evidence on the associations of NMSC with: (1) all-cause mortality, (2) cancer-specific mortality, and (3) cancer survival. Bibliographic databases were searched through February 2016. Cohort studies published in English were included if adequate data were provided to estimate mortality ratios in patients with-versus-without NMSC. Data were abstracted from the total of eight studies from independent data sources that met inclusion criteria (n = 3 for all-cause mortality, n = 2 for cancer-specific mortality, and n = 5 for cancer survival). For all-cause mortality, a significant increased risk was observed for patients with a history of squamous cell carcinoma (SCC) (mortality ratio estimates (MR) 1.25 and 1.30), whereas no increased risk was observed for patients with a history of basal cell carcinoma (BCC) (MRs 0.96 and 0.97). Based on one study, the association with cancer-specific mortality was stronger for SCC (MR 2.17) than BCC (MR 1.15). Across multiple types of cancer both SCC and BCC tended to be associated with poorer survival from second primary malignancies. Multiple studies support an association between NMSC and fatal outcomes; the associations tend to be more potent for SCC than BCC. Additional investigation is needed to more precisely characterize these associations and elucidate potential underlying mechanisms.