Data Mining in Oncology: The ukCAT Project and the Practicalities of Working with Routine Patient Data

Evaluating distributed-learning on real-world obstetrics data: comparing distributed, centralized and local models

This study focused on comparing distributed learning models with centralized and local models, assessing their efficacy in predicting specific delivery and patient-related outcomes in obstetrics using real-world data. The predictions focus on key moments in the obstetric care process, including discharge and various stages of hospitalization. Our analysis: using 6 different machine learning methods like Decision Trees, Bayesian methods, Stochastic Gradient Descent, K-nearest neighbors, AdaBoost, and Multi-layer Perceptron and 19 different variables with various distributions and types, revealed that distributed models were at least equal, and often superior, to centralized versions and local versions. We also describe thoroughly the preprocessing stage in order to help others implement this method in real-world scenarios. The preprocessing steps included cleaning and harmonizing missing values, handling missing data and encoding categorical variables with multisite logic. Even though the type of machine learning model and the distribution of the outcome variable can impact the result, we reached results of 66% being superior to the centralized and local counterpart and 77% being better than the centralized with AdaBoost. Our experiments also shed light in the preprocessing steps required to implement distributed models in a real-world scenario. Our results advocate for distributed learning as a promising tool for applying machine learning in clinical settings, particularly when privacy and data security are paramount, thus offering a robust solution for privacy-concerned clinical applications.

Towards estimating the carbon footprint of external beam radiotherapy

PURPOSE

The National Health Service (NHS) in the United Kingdom (UK) is aiming to be carbon net zero by 2040 to help limit the dangerous effects of climate change. Radiotherapy contributes to this with potential sources quantified here.

METHOD

Activity data for 42 patients from within the breast IMRT and prostate VMAT pathways were collected. Data for 20 prostate patients was also collected from 3 other centres to enable cross centre comparison. A process-based, bottom-up approach was used to calculate the carbon footprint. Additionally, patients were split into pre-COVID and COVID groups to assess the impact of protocol changes due to the pandemic.

RESULTS

The calculated carbon footprint for prostate and breast pre-COVID were 148 kgCO2e and 101 kgCO2e respectively, and 226 kgCO2e and 75 kgCO2e respectively during COVID. The energy usage by the linac during treatment for a total course of radiotherapy for prostate treatments was 2-3 kWh and about 1 kWh for breast treatments. Patient travel made up the largest proportion (70-80%) of the calculated carbon footprint, with linac idle power second with ∼ 10% and PPE and SF6 leakage were both between 2 and 4%.

CONCLUSION

These initial findings highlight that the biggest contributor to the external beam radiotherapy carbon footprint was patient travel, which may motivate increased used of hypofractionation. Many assumptions and boundaries have been set on the data gathered, which limit the wider application of these results. However, they provide a useful foundation for future more comprehensive life cycle assessments.

Predicting cancer relapse following lung stereotactic radiotherapy: an external validation study using real-world evidence

Purpose

For patients receiving lung stereotactic ablative radiotherapy (SABR), evidence suggests that high peritumor density predicts an increased risk of microscopic disease (MDE) and local-regional failure, but only if there is low or heterogenous incidental dose surrounding the tumor (GTV). A data-mining method (Cox-per-radius) has been developed to investigate this dose-density interaction. We apply the method to predict local relapse (LR) and regional failure (RF) in patients with non-small cell lung cancer.

Methods

199 patients treated in a routine setting were collated from a single institution for training, and 76 patients from an external institution for validation. Three density metrics (mean, 90th percentile, standard deviation (SD)) were studied in 1mm annuli between 0.5cm inside and 2cm outside the GTV boundary. Dose SD and fraction of volume receiving less than 30Gy were studied in annuli 0.5-2cm outside the GTV to describe incidental MDE dosage. Heat-maps were created that correlate with changes in LR and RF rates due to the interaction between dose heterogeneity and density at each distance combination. Regions of significant improvement were studied in Cox proportional hazards models, and explored with and without re-fitting in external data. Correlations between the dose component of the interaction and common dose metrics were reported.

Results

Local relapse occurred at a rate of 6.5% in the training cohort, and 18% in the validation cohort, which included larger and more centrally located tumors. High peritumor density in combination with high dose variability (0.5 - 1.6cm) predicts LR. No interactions predicted RF. The LR interaction improved the predictive ability compared to using clinical variables alone (optimism-adjusted C-index; 0.82 vs 0.76). Re-fitting model coefficients in external data confirmed the importance of this interaction (C-index; 0.86 vs 0.76). Dose variability in the 0.5-1.6 cm annular region strongly correlates with heterogeneity inside the target volume (SD; ρ = 0.53 training, ρ = 0.65 validation).

Conclusion

In these real-world cohorts, the combination of relatively high peritumor density and high dose variability predicts increase in LR, but not RF, following lung SABR. This external validation justifies potential use of the model to increase low-dose CTV margins for high-risk patients.

Efficient and Reliable Data Extraction in Radiation Oncology using Python Programming Language: A Pilot Study

Background and Purpose

In recent years, data science approaches have entered health-care systems such as radiology, pathology, and radiation oncology. In our pilot study, we developed an automated data mining approach to extract data from a treatment planning system (TPS) with high speed, maximum accuracy, and little human interaction. We compared the amount of time required for manual data extraction versus the automated data mining technique.

Materials and Methods

A Python programming script was created to extract specified parameters and features pertaining to patients and treatment (a total of 25 features) from TPS. We successfully implemented automation in data mining, utilizing the application programming interface environment provided by the external beam radiation therapy equipment provider for the whole group of patients who were accepted for treatment.

Results

This in-house Python-based script extracted selected features for 427 patients in 0.28 ± 0.03 min with 100% accuracy at an astonishing rate of 0.04 s/plan. Comparatively, manual extraction of 25 parameters took an average of 4.5 ± 0.33 min/plan, along with associated transcriptional and transpositional errors and missing data information. This new approach turned out to be 6850 times faster than the conventional approach. Manual feature extraction time increased by a factor of nearly 2.5 if we doubled the number of features extracted, whereas for the Python script, it increased by a factor of just 1.15.

Conclusion

We conclude that our in-house developed Python script can extract plan data from TPS at a far higher speed (>6000 times) and with the best possible accuracy compared to manual data extraction.

Heart dose and cardiac comorbidities influence death with a cardiac cause following hypofractionated radiotherapy for lung cancer

Background

There is increasing evidence of cardiac toxicity of thoracic radiotherapy however, it is difficult to draw conclusions on cardiac dose constraints due to the heterogeneity of published studies. Moreover, few studies record data on cause of death. The aim of this paper is to investigate the relationship between conventional cardiac dosimetric parameters and death with cardiac causes using data from the UK national cause of death registry.

Methods

Data on cancer diagnosis, treatment and cause of death following radical lung cancer radiotherapy were obtained from Public Health England for all patients treated at the Christie NHS Foundation Trust between 1/1/10 and 31/12/16. Individuals with metastatic disease and those who received multiple courses of thoracic radiotherapy where excluded. All patients who received > 45Gy in 20 fractions were included. Cardiac cause of death was defined as the following ICD-10 codes on death certificate: I20-I25; I30-I32; I34-I37; I40-I52. Cardiac V5Gy, V30Gy, V50Gy and mean heart dose (MHD) were extracted. Cumulative incidence of death with cardiac causes were plotted for each cardiac dosimetric parameter. Multi-variable Fine and Gray competing risk analysis was used to model predictors for cardiac death with non-cardiac death as a competing risk.

Results

Cardiac dosimetric parameters were available for 967 individuals, 110 died with a cardiac cause (11.4%). Patients with a cardiac comorbidity had an increased risk of death with a cardiac cause compared with those without a cardiac comorbidity (2-year cumulative incidence 21.3% v 6.2%, p<0.001). In patients with a pre-existing cardiac comorbidity, heart V30Gy ≥ 15% was associated with higher cumulative incidence of death with a cardiac cause compared to patients with heart V30Gy <15% (2-year rate 25.8% v 17.3%, p=0.05). In patients without a cardiac comorbidity, after adjusting for tumour and cardiac risk factors, MHD (aHR 1.07, 1.01-1.13, p=0.021), heart V5Gy (aHR 1.01, 1-1.13, p=0.05) and heart V30Gy (aHR 1.04, 1-1.07, p=0.039) were associated with cardiac death.

Conclusion

The effect of cardiac radiation dose on cardiac-related death following thoracic radiotherapy is different in patients with and without cardiac comorbidities. Therefore patients’ cardiovascular risk factors should be identified and managed alongside radiotherapy for lung cancer.

Infrastructure platform for privacy-preserving distributed machine learning development of computer-assisted theragnostics in cancer

INTRODUCTION

Emerging evidence suggests that data-driven support tools have found their way into clinical decision-making in a number of areas, including cancer care. Improving them and widening their scope of availability in various differing clinical scenarios, including for prognostic models derived from retrospective data, requires co-ordinated data sharing between clinical centres, secondary analyses of large multi-institutional clinical trial data, or distributed (federated) learning infrastructures. A systematic approach to utilizing routinely collected data across cancer care clinics remains a significant challenge due to privacy, administrative and political barriers.

METHODS

An information technology infrastructure and web service software was developed and implemented which uses machine learning to construct clinical decision support systems in a privacy-preserving manner across datasets geographically distributed in different hospitals. The infrastructure was deployed in a network of Australian hospitals. A harmonized, international ontology-linked, set of lung cancer databases were built with the routine clinical and imaging data at each centre. The infrastructure was demonstrated with the development of logistic regression models to predict major cardiovascular events following radiation therapy.

RESULTS

The infrastructure implemented forms the basis of the Australian computer-assisted theragnostics (AusCAT) network for radiation oncology data extraction, reporting and distributed learning. Four radiation oncology departments (across seven hospitals) in New South Wales (NSW) participated in this demonstration study. Infrastructure was deployed at each centre and used to develop a model predicting for cardiovascular admission within a year of receiving curative radiotherapy for non-small cell lung cancer. A total of 10417 lung cancer patients were identified with 802 being eligible for the model. Twenty features were chosen for analysis from the clinical record and linked registries. After selection, 8 features were included and a logistic regression model achieved an area under the receiver operating characteristic (AUROC) curve of 0.70 and C-index of 0.65 on out-of-sample data.

CONCLUSION

The infrastructure developed was demonstrated to be usable in practice between clinical centres to harmonize routinely collected oncology data and develop models with federated learning. It provides a promising approach to enable further research studies in radiation oncology using real world clinical data.

The role of artificial intelligence in paediatric neuroradiology

Imaging plays a fundamental role in the managing childhood neurologic, neurosurgical and neuro-oncological disease. Employing multi-parametric MRI techniques, such as spectroscopy and diffusion- and perfusion-weighted imaging, to the radiophenotyping of neuroradiologic conditions is becoming increasingly prevalent, particularly with radiogenomic analyses correlating imaging characteristics with molecular biomarkers of disease. However, integration into routine clinical practice remains elusive. With modern multi-parametric MRI now providing additional data beyond anatomy, informing on histology, biology and physiology, such metric-rich information can present as information overload to the treating radiologist and, as such, information relevant to an individual case can become lost. Artificial intelligence techniques are capable of modelling the vast radiologic, biological and clinical datasets that accompany childhood neurologic disease, such that this information can become incorporated in upfront prognostic modelling systems, with artificial intelligence techniques providing a plausible approach to this solution. This review examines machine learning approaches than can be used to underpin such artificial intelligence applications, with exemplars for each machine learning approach from the world literature. Then, within the specific use case of paediatric neuro-oncology, we examine the potential future contribution for such artificial intelligence machine learning techniques to offer solutions for patient care in the form of decision support systems, potentially enabling personalised medicine within this domain of paediatric radiologic practice.

Implementation of the Australian Computer-Assisted Theragnostics (AusCAT) network for radiation oncology data extraction, reporting and distributed learning

INTRODUCTION

There is significant potential to analyse and model routinely collected data for radiotherapy patients to provide evidence to support clinical decisions, particularly where clinical trials evidence is limited or non-existent. However, in practice there are administrative, ethical, technical, logistical and legislative barriers to having coordinated data analysis platforms across radiation oncology centres.

METHODS

A distributed learning network of computer systems is presented, with software tools to extract and report on oncology data and to enable statistical model development. A distributed or federated learning approach keeps data in the local centre, but models are developed from the entire cohort.

RESULTS

The feasibility of this approach is demonstrated across six Australian oncology centres, using routinely collected lung cancer data from oncology information systems. The infrastructure was used to validate and develop machine learning for model-based clinical decision support and for one centre to assess patient eligibility criteria for two major lung cancer radiotherapy clinical trials (RTOG-9410, RTOG-0617). External validation of a 2-year overall survival model for non-small cell lung cancer (NSCLC) gave an AUC of 0.65 and C-index of 0.62 across the network. For one centre, 65% of Stage III NSCLC patients did not meet eligibility criteria for either of the two practice-changing clinical trials, and these patients had poorer survival than eligible patients (10.6 m vs. 15.8 m, P = 0.024).

CONCLUSION

Population-based studies on routine data are possible using a distributed learning approach. This has the potential for decision support models for patients for whom supporting clinical trial evidence is not applicable.

Deep representation learning of patient data from Electronic Health Records (EHR): A systematic review

OBJECTIVES

Patient representation learning refers to learning a dense mathematical representation of a patient that encodes meaningful information from Electronic Health Records (EHRs). This is generally performed using advanced deep learning methods. This study presents a systematic review of this field and provides both qualitative and quantitative analyses from a methodological perspective.

METHODS

We identified studies developing patient representations from EHRs with deep learning methods from MEDLINE, EMBASE, Scopus, the Association for Computing Machinery (ACM) Digital Library, and the Institute of Electrical and Electronics Engineers (IEEE) Xplore Digital Library. After screening 363 articles, 49 papers were included for a comprehensive data collection.

RESULTS

Publications developing patient representations almost doubled each year from 2015 until 2019. We noticed a typical workflow starting with feeding raw data, applying deep learning models, and ending with clinical outcome predictions as evaluations of the learned representations. Specifically, learning representations from structured EHR data was dominant (37 out of 49 studies). Recurrent Neural Networks were widely applied as the deep learning architecture (Long short-term memory: 13 studies, Gated recurrent unit: 11 studies). Learning was mainly performed in a supervised manner (30 studies) optimized with cross-entropy loss. Disease prediction was the most common application and evaluation (31 studies). Benchmark datasets were mostly unavailable (28 studies) due to privacy concerns of EHR data, and code availability was assured in 20 studies.

DISCUSSION & CONCLUSION

The existing predictive models mainly focus on the prediction of single diseases, rather than considering the complex mechanisms of patients from a holistic review. We show the importance and feasibility of learning comprehensive representations of patient EHR data through a systematic review. Advances in patient representation learning techniques will be essential for powering patient-level EHR analyses. Future work will still be devoted to leveraging the richness and potential of available EHR data. Reproducibility and transparency of reported results will hopefully improve. Knowledge distillation and advanced learning techniques will be exploited to assist the capability of learning patient representation further.

Systematic Review of Privacy-Preserving Distributed Machine Learning From Federated Databases in Health Care

Big data for health care is one of the potential solutions to deal with the numerous challenges of health care, such as rising cost, aging population, precision medicine, universal health coverage, and the increase of noncommunicable diseases. However, data centralization for big data raises privacy and regulatory concerns.Covered topics include (1) an introduction to privacy of patient data and distributed learning as a potential solution to preserving these data, a description of the legal context for patient data research, and a definition of machine/deep learning concepts; (2) a presentation of the adopted review protocol; (3) a presentation of the search results; and (4) a discussion of the findings, limitations of the review, and future perspectives.Distributed learning from federated databases makes data centralization unnecessary. Distributed algorithms iteratively analyze separate databases, essentially sharing research questions and answers between databases instead of sharing the data. In other words, one can learn from separate and isolated datasets without patient data ever leaving the individual clinical institutes.Distributed learning promises great potential to facilitate big data for medical application, in particular for international consortiums. Our purpose is to review the major implementations of distributed learning in health care.

Proton relative biological effectiveness (RBE): a multiscale problem

Proton radiotherapy is undergoing rapid expansion both within the UK and internationally, but significant challenges still need to be overcome if maximum benefit is to be realised from this technique. One major limitation is the persistent uncertainty in proton relative biological effectiveness (RBE). While RBE values are needed to link proton radiotherapy to our existing experience with photon radiotherapy, RBE remains poorly understood and is typically incorporated as a constant dose scaling factor of 1.1 in clinical plans. This is in contrast to extensive experimental evidence indicating that RBE is a function of dose, tissue type, and proton linear energy transfer, among other parameters. In this article, we discuss the challenges associated with obtaining clinically relevant values for proton RBE through commonly-used assays, and highlight the wide range of other experimental end points which can inform our understanding of RBE. We propose that accurate and robust optimization of proton radiotherapy ultimately requires a multiscale understanding of RBE, integrating subcellular, cellular, and patient-level processes.

The use of volunteers to implement electronic patient reported outcomes in lung cancer outpatient clinics

•

104 eligible lung cancer patients were approached, 86 (83%) consented to take part.

•

At 1st attempt 69% of patients completed the ePRO questionnaire without assistance.

•

Assistance was defined as verbal/physical help to complete the ePRO questionnaire.

•

Most patients requiring help had a companion that could have provided assistance.

•

More patients preferred electronic than paper questionnaires.

Background

Treatment related toxicity is common after chemotherapy and radiotherapy. Our group has developed and validated an electronic Patient Reported Outcome questionnaire (ePRO) to assess symptoms and toxicity in lung cancer patients receiving (chemo)radiotherapy treatment. We assessed the need for volunteer support in clinics to assist patients in completing ePROs.

Methods

Lung Cancer patients attending outpatient or radiotherapy clinics at The Christie NHS Foundation Trust, Manchester were consented and asked to complete a Patient Reported Outcomes questionnaire using an electronic device (a touchscreen). Trained volunteers were available if patients required help such as verbal or physical assistance. The primary objective was to determine the need for volunteers to assist lung cancer patients in completing ePROs.

Results

27/86 (31.4%) of patients who consented to this study required assistance to complete the ePRO. After questioning, we found that only 7/86 (8.1%) would have relied on volunteers for assistance as the majority of patients had a companion that could have provided help. 81/86 (94.2%) of patients were satisfied with the use of a touchscreen tablet to complete the ePRO.

Conclusion

Our results demonstrate that the introduction of ePROs in lung cancer outpatient clinics is feasible, even without the use of volunteers for the majority of patients. The implementation of ePROs would allow large volumes of high quality (chemo)radiotherapy toxicity data to be collected accurately and quickly. This is essential for the development of predictive models of outcome using population-based data, which could allow the personalisation of (chemo)radiotherapy treatment for lung cancer patients.