Data harmonization and federated learning for multi-cohort dementia research using the OMOP common data model: A Netherlands consortium of dementia cohorts case study.

BACKGROUND: Establishing collaborations between cohort studies has been fundamental for progress in health research. However, such collaborations are hampered by heterogeneous data representations across cohorts and legal constraints to data sharing. The first arises from a lack of consensus in standards of data collection and representation across cohort studies and is usually tackled by applying data harmonization processes. The second is increasingly important due to raised awareness for privacy protection and stricter regulations, such as the GDPR. Federated learning has emerged as a privacy-preserving alternative to transferring data between institutions through analyzing data in a decentralized manner. METHODS: In this study, we set up a federated learning infrastructure for a consortium of nine Dutch cohorts with appropriate data available to the etiology of dementia, including an extract, transform, and load (ETL) pipeline for data harmonization. Additionally, we assessed the challenges of transforming and standardizing cohort data using the Observational Medical Outcomes Partnership (OMOP) common data model (CDM) and evaluated our tool in one of the cohorts employing federated algorithms. RESULTS: We successfully applied our ETL tool and observed a complete coverage of the cohorts' data by the OMOP CDM. The OMOP CDM facilitated the data representation and standardization, but we identified limitations for cohort-specific data fields and in the scope of the vocabularies available. Specific challenges arise in a multi-cohort federated collaboration due to technical constraints in local environments, data heterogeneity, and lack of direct access to the data. CONCLUSION: In this article, we describe the solutions to these challenges and limitations encountered in our study. Our study shows the potential of federated learning as a privacy-preserving solution for multi-cohort studies that enhance reproducibility and reuse of both data and analyses.

Colorectal cancer health and care quality indicators in a federated setting using the Personal Health Train.

OBJECTIVE: Hospitals and healthcare providers should assess and compare the quality of care given to patients and based on this improve the care. In the Netherlands, hospitals provide data to national quality registries, which in return provide annual quality indicators. However, this process is time-consuming, resource intensive and risks patient privacy and confidentiality. In this paper, we presented a multicentric 'Proof of Principle' study for federated calculation of quality indicators in patients with colorectal cancer. The findings suggest that the proposed approach is highly time-efficient and consume significantly lesser resources. MATERIALS AND METHODS: Two quality indicators are calculated in an efficient and privacy presevering federated manner, by i) applying the Findable Accessible Interoperable and Reusable (FAIR) data principles and ii) using the Personal Health Train (PHT) infrastructure. Instead of sharing data to a centralized registry, PHT enables analysis by sending algorithms and sharing only insights from the data. RESULTS: ETL process extracted data from the Electronic Health Record systems of the hospitals, converted them to FAIR data and hosted in RDF endpoints within each hospital. Finally, quality indicators from each center are calculated using PHT and the mean result along with the individual results plotted. DISCUSSION AND CONCLUSION: PHT and FAIR data principles can efficiently calculate quality indicators in a privacy-preserving federated approach and the work can be scaled up both nationally and internationally. Despite this, application of the methodology was largely hampered by ELSI issues. However, the lessons learned from this study can provide other hospitals and researchers to adapt to the process easily and take effective measures in building quality of care infrastructures.

Predicting the risk of neurocognitive decline after brain irradiation in adult patients with a primary brain tumor.

BACKGROUND: Deterioration of neurocognitive function in adult patients with a primary brain tumor is the most concerning side effect of radiotherapy. This study was aimed to develop and evaluate Normal-Tissue Complication Probability (NTCP) models using clinical and dose-volume measures for 6-month, 1-year and 2-year Neurocognitive Decline (ND) post-radiotherapy. METHODS: A total of 219 patients with a primary brain tumor treated with radical photon and/or proton radiotherapy (RT) between 2019 and 2022 were included. Controlled Oral Word Association (COWA) test, Hopkins Verbal Learning Test-Revised (HVLTR) and Trail Making Test (TMT) were used to objectively measure ND. A comprehensive set of potential clinical and dose-volume measures on several brain structures were considered for statistical modelling. Clinical, dose-volume and combined models were constructed and internally tested in terms of discrimination (Area Under the Curve, AUC), calibration (Mean Absolute Error, MAE) and net benefit. RESULTS: 50%, 44.5% and 42.7% of the patients developed ND at 6-month, 1-year and 2-year timepoints, respectively. Following predictors were included in the combined model for 6-month ND: age at radiotherapy>56 years (OR=5.71), overweight (OR=0.49), obesity (OR=0.35), chemotherapy (OR=2.23), brain V20Gy≥20% (OR=3.53), brainstem volume≥26cc (OR=0.39) and hypothalamus volume≥0.5cc (OR=0.4). Decision curve analysis showed that the combined models had the highest net benefits at 6-month (AUC=0.79, MAE=0.021), 1-year (AUC=0.72, MAE=0.027) and 2-year (AUC=0.69, MAE=0.038) timepoints. CONCLUSION: The proposed NTCP models use easy-to-obtain predictors to identify patients at high-risk of ND after brain RT. These models can potentially provide a base for RT-related decisions and post-therapy neurocognitive rehabilitation interventions.

A distributed feature selection pipeline for survival analysis using radiomics in non-small cell lung cancer patients.

Predictive modelling of cancer outcomes using radiomics faces dimensionality problems and data limitations, as radiomics features often number in the hundreds, and multi-institutional data sharing is ()often unfeasible. Federated learning (FL) and feature selection (FS) techniques combined can help overcome these issues, as one provides the means of training models without exchanging sensitive data, while the other identifies the most informative features, reduces overfitting, and improves model interpretability. Our proposed FS pipeline based on FL principles targets data-driven radiomics FS in a multivariate survival study of non-small cell lung cancer patients. The pipeline was run across datasets from three institutions without patient-level data exchange. It includes two FS techniques, Correlation-based Feature Selection and LASSO regularization, and Cox Proportional-Hazard regression with Overall Survival as endpoint. Trained and validated on 828 patients overall, our pipeline yielded a radiomic signature comprising "intensity-based energy" and "mean discretised intensity". Validation resulted in a mean Harrell C-index of 0.59, showcasing fair efficacy in risk stratification. In conclusion, we suggest a distributed radiomics approach that incorporates preliminary feature selection to systematically decrease the feature set based on data-driven considerations. This aims to address dimensionality challenges beyond those associated with data constraints and interpretability concerns.

Assessing the role of GPT-4 in thyroid ultrasound diagnosis and treatment recommendations: enhancing interpretability with a chain of thought approach.

Background: As artificial intelligence (AI) becomes increasingly prevalent in the medical field, the effectiveness of AI-generated medical reports in disease diagnosis remains to be evaluated. ChatGPT is a large language model developed by open AI with a notable capacity for text abstraction and comprehension. This study aimed to explore the capabilities, limitations, and potential of Generative Pre-trained Transformer (GPT)-4 in analyzing thyroid cancer ultrasound reports, providing diagnoses, and recommending treatment plans. Methods: Using 109 diverse thyroid cancer cases, we evaluated GPT-4's performance by comparing its generated reports to those from doctors with various levels of experience. We also conducted a Turing Test and a consistency analysis. To enhance the interpretability of the model, we applied the Chain of Thought (CoT) method to deconstruct the decision-making chain of the GPT model. Results: GPT-4 demonstrated proficiency in report structuring, professional terminology, and clarity of expression, but showed limitations in diagnostic accuracy. In addition, our consistency analysis highlighted certain discrepancies in the AI's performance. The CoT method effectively enhanced the interpretability of the AI's decision-making process. Conclusions: GPT-4 exhibits potential as a supplementary tool in healthcare, especially for generating thyroid gland diagnostic reports. Our proposed online platform, "ThyroAIGuide", alongside the CoT method, underscores the potential of AI to augment diagnostic processes, elevate healthcare accessibility, and advance patient education. However, the journey towards fully integrating AI into healthcare is ongoing, requiring continuous research, development, and careful monitoring by medical professionals to ensure patient safety and quality of care.

[Implementing new interventions and indications: the possible role for real world data]. / De mogelijke rol van 'real world data'.

In advising the preferred therapy for the individual patient the expected results of the proposed intervention and possible side effects are the most relevant considerations. However, predicting the results of an intervention is difficult, especially when well designed randomized clinical trials (RCT's) are lacking or not conclusive. Artificial intelligence (AI) algorithms based on routine clinical data (real world data) can support clinical decision making, but in daily practice AI is still scarcely used. In this article one large radiotherapy facility and two health insurers describe their joint opinion on the possible role of AI based on real world data as an aid in clinical decision making when evidence from RCT's is not available. The introduction of proton radiotherapy in The Netherlands is being used as case model for AI model based clinical decision making.

Natural Language Processing Algorithm Used for Staging Pulmonary Oncology from Free-Text Radiological Reports: "Including PET-CT and Validation Towards Clinical Use".

Natural language processing (NLP) can be used to process and structure free text, such as (free text) radiological reports. In radiology, it is important that reports are complete and accurate for clinical staging of, for instance, pulmonary oncology. A computed tomography (CT) or positron emission tomography (PET)-CT scan is of great importance in tumor staging, and NLP may be of additional value to the radiological report when used in the staging process as it may be able to extract the T and N stage of the 8th tumor-node-metastasis (TNM) classification system. The purpose of this study is to evaluate a new TN algorithm (TN-PET-CT) by adding a layer of metabolic activity to an already existing rule-based NLP algorithm (TN-CT). This new TN-PET-CT algorithm is capable of staging chest CT examinations as well as PET-CT scans. The study design made it possible to perform a subgroup analysis to test the external validation of the prior TN-CT algorithm. For information extraction and matching, pyContextNLP, SpaCy, and regular expressions were used. Overall TN accuracy score of the TN-PET-CT algorithm was 0.73 and 0.62 in the training and validation set (N = 63, N = 100). The external validation of the TN-CT classifier (N = 65) was 0.72. Overall, it is possible to adjust the TN-CT algorithm into a TN-PET-CT algorithm. However, outcomes highly depend on the accuracy of the report, the used vocabulary, and its context to express, for example, uncertainty. This is true for both the adjusted PET-CT algorithm and for the CT algorithm when applied in another hospital.

The role of mHealth intervention to improve maternal and child health: A provider-based qualitative study in Southern Ethiopia.

INTRODUCTION: Maternal and child mortality remained higher in developing regions such as Southern Ethiopia due to poor maternal and child health. Technologies such as mobile applications in health may be an opportunity to reduce maternal and child mortality because they can improve access to information. Therefore, the main aim of this study was to explore the role of mHealth in improving maternal and child health in Southern Ethiopia. METHODS: This study employed a qualitative study design to explore the role of mHealth in improving maternal and child health among health professionals in Southern Ethiopia from December 2022 to March 2023. We conducted nine in-depth interviews, six key informants' in-depth interviews, and four focused group discussions among health professionals. This is followed by thematic analyses to synthesize the collected evidence. RESULTS: The results are based on 226 quotations, 5 major themes, and 24 subthemes. The study participants discussed the possible acceptance of mHealth in terms of its fitness in the existing health system, its support to health professionals, and its importance in improving maternal and child health. The participants ascertained the importance of awareness creation before the implementation of mHealth among women, families, communities, and providers. They reported the importance of mHealth for mothers and health professionals and the effectiveness of mHealth services. The participants stated that the main challenges related to acceptance, awareness, negligence, readiness, and workload. However, they also suggested strategic solutions such as using family support, provider support, mothers' forums, and community forums. CONCLUSION: The evidence generated during this analysis is important information for program implementations and can inform policy-making. The planned intervention needs to introduce mHealth in Southern Ethiopia. Planners, decision-makers, and researchers can use it in mobile technology-related interventions. For challenges identified, we recommend solution-identified-based interventions and quality studies.

Comparing the performance of a deep learning-based lung gross tumour volume segmentation algorithm before and after transfer learning in a new hospital.

Objectives: Radiation therapy for lung cancer requires a gross tumour volume (GTV) to be carefully outlined by a skilled radiation oncologist (RO) to accurately pinpoint high radiation dose to a malignant mass while simultaneously minimizing radiation damage to adjacent normal tissues. This is manually intensive and tedious however, it is feasible to train a deep learning (DL) neural network that could assist ROs to delineate the GTV. However, DL trained on large openly accessible data sets might not perform well when applied to a superficially similar task but in a different clinical setting. In this work, we tested the performance of DL automatic lung GTV segmentation model trained on open-access Dutch data when used on Indian patients from a large public tertiary hospital, and hypothesized that generic DL performance could be improved for a specific local clinical context, by means of modest transfer-learning on a small representative local subset. Methods: X-ray computed tomography (CT) series in a public data set called "NSCLC-Radiomics" from The Cancer Imaging Archive was first used to train a DL-based lung GTV segmentation model (Model 1). Its performance was assessed using a different open access data set (Interobserver1) of Dutch subjects plus a private Indian data set from a local tertiary hospital (Test Set 2). Another Indian data set (Retrain Set 1) was used to fine-tune the former DL model using a transfer learning method. The Indian data sets were taken from CT of a hybrid scanner based in nuclear medicine, but the GTV was drawn by skilled Indian ROs. The final (after fine-tuning) model (Model 2) was then re-evaluated in "Interobserver1" and "Test Set 2." Dice similarity coefficient (DSC), precision, and recall were used as geometric segmentation performance metrics. Results: Model 1 trained exclusively on Dutch scans showed a significant fall in performance when tested on "Test Set 2." However, the DSC of Model 2 recovered by 14 percentage points when evaluated in the same test set. Precision and recall showed a similar rebound of performance after transfer learning, in spite of using a comparatively small sample size. The performance of both models, before and after the fine-tuning, did not significantly change the segmentation performance in "Interobserver1." Conclusions: A large public open-access data set was used to train a generic DL model for lung GTV segmentation, but this did not perform well initially in the Indian clinical context. Using transfer learning methods, it was feasible to efficiently and easily fine-tune the generic model using only a small number of local examples from the Indian hospital. This led to a recovery of some of the geometric segmentation performance, but the tuning did not appear to affect the performance of the model in another open-access data set. Advances in knowledge: Caution is needed when using models trained on large volumes of international data in a local clinical setting, even when that training data set is of good quality. Minor differences in scan acquisition and clinician delineation preferences may result in an apparent drop in performance. However, DL models have the advantage of being efficiently "adapted" from a generic to a locally specific context, with only a small amount of fine-tuning by means of transfer learning on a small local institutional data set.

The effect of mHealth on childhood vaccination in Africa: A systematic review and meta-analysis.

INTRODUCTION: Vaccine-preventable diseases are the public health problems in Africa, although vaccination is an available, safe, simple, and effective method prevention. Technologies such as mHealth may provide maternal access to health information and support decisions on childhood vaccination. Many studies on the role of mHealth in vaccination decisions have been conducted in Africa, but the evidence needs to provide conclusive information to support mHealth introduction. This study provides essential information to assist planning and policy decisions regarding the use of mHealth for childhood vaccination. METHODS: We conducted a systematic review and meta-analysis for studies applying mHealth in Africa for vaccination decisions following the Preferred Reporting Items for Systematic and Meta-Analysis [PRISMA] guideline. Databases such as CINAHL, EMBASE, PubMed, PsycINFO, Scopus, Web of Science, Google Scholar, Global Health, HINARI, and Cochrane Library were included. We screened studies in Endnote X20 and performed the analysis using Revman 5.4.1. RESULTS: The database search yielded 1,365 articles [14 RCTs and 4 quasi-experiments] with 21,070 participants satisfied all eligibility criteria. The meta-analysis showed that mHealth has an OR of 2.15 [95% CI: 1.70-2.72; P<0.001; I2 = 90%] on vaccination rates. The subgroup analysis showed that regional differences cause heterogeneity. Funnel plots and Harbord tests showed the absence of publication bias, while the GRADE scale showed a moderate-quality body of evidence. CONCLUSION: Although heterogeneous, this systematic review and meta-analysis showed that the application of mHealth could potentially improve childhood vaccination in Africa. It increased childhood vaccination by more than double [2.15 times] among children whose mothers are motivated by mHealth services. MHealth is more effective in less developed regions and when an additional incentive party with the messaging system. However, it can be provided at a comparably low cost based on the development level of regions and can be established as a routine service in Africa. REGISTRATION: PROSPERO: CRD42023415956.

A PET/CT radiomics model for predicting distant metastasis in early-stage non-small cell lung cancer patients treated with stereotactic body radiotherapy: a multicentric study.

OBJECTIVES: Stereotactic body radiotherapy (SBRT) is a treatment option for patients with early-stage non-small cell lung cancer (NSCLC) who are unfit for surgery. Some patients may experience distant metastasis. This study aimed to develop and validate a radiomics model for predicting distant metastasis in patients with early-stage NSCLC treated with SBRT. METHODS: Patients at five institutions were enrolled in this study. Radiomics features were extracted based on the PET/CT images. After feature selection in the training set (from Tianjin), CT-based and PET-based radiomics signatures were built. Models based on CT and PET signatures were built and validated using external datasets (from Zhejiang, Zhengzhou, Shandong, and Shanghai). An integrated model that included CT and PET radiomic signatures was developed. The performance of the proposed model was evaluated in terms of its discrimination, calibration, and clinical utility. Multivariate logistic regression was used to calculate the probability of distant metastases. The cutoff value was obtained using the receiver operator characteristic curve (ROC), and the patients were divided into high- and low-risk groups. Kaplan-Meier analysis was used to evaluate the distant metastasis-free survival (DMFS) of different risk groups. RESULTS: In total, 228 patients were enrolled. The median follow-up time was 31.4 (2.0-111.4) months. The model based on CT radiomics signatures had an area under the curve (AUC) of 0.819 in the training set (n = 139) and 0.786 in the external dataset (n = 89). The PET radiomics model had an AUC of 0.763 for the training set and 0.804 for the external dataset. The model combining CT and PET radiomics had an AUC of 0.835 for the training set and 0.819 for the external dataset. The combined model showed a moderate calibration and a positive net benefit. When the probability of distant metastasis was greater than 0.19, the patient was considered to be at high risk. The DMFS of patients with high- and low-risk was significantly stratified (P < 0.001). CONCLUSIONS: The proposed PET/CT radiomics model can be used to predict distant metastasis in patients with early-stage NSCLC treated with SBRT and provide a reference for clinical decision-making. In this study, the model was established by combining CT and PET radiomics signatures in a moderate-quantity training cohort of early-stage NSCLC patients treated with SBRT and was successfully validated in independent cohorts. Physicians could use this easy-to-use model to assess the risk of distant metastasis after SBRT. Identifying subgroups of patients with different risk factors for distant metastasis is useful for guiding personalized treatment approaches.

Development and validation of multicentre study on novel Artificial Intelligence-based Cardiovascular Risk Score (AICVD).

OBJECTIVE: Cardiovascular diseases (CVD) are one of the most prevalent diseases in India amounting for nearly 30% of total deaths. A dearth of research on CVD risk scores in Indian population, limited performance of conventional risk scores and inability to reproduce the initial accuracies in randomised clinical trials has led to this study on large-scale patient data. The objective is to develop an Artificial Intelligence-based Risk Score (AICVD) to predict CVD event (eg, acute myocardial infarction/acute coronary syndrome) in the next 10 years and compare the model with the Framingham Heart Risk Score (FHRS) and QRisk3. METHODS: Our study included 31 599 participants aged 18-91 years from 2009 to 2018 in six Apollo Hospitals in India. A multistep risk factors selection process using Spearman correlation coefficient and propensity score matching yielded 21 risk factors. A deep learning hazards model was built on risk factors to predict event occurrence (classification) and time to event (hazards model) using multilayered neural network. Further, the model was validated with independent retrospective cohorts of participants from India and the Netherlands and compared with FHRS and QRisk3. RESULTS: The deep learning hazards model had a good performance (area under the curve (AUC) 0.853). Validation and comparative results showed AUCs between 0.84 and 0.92 with better positive likelihood ratio (AICVD -6.16 to FHRS -2.24 and QRisk3 -1.16) and accuracy (AICVD -80.15% to FHRS 59.71% and QRisk3 51.57%). In the Netherlands cohort, AICVD also outperformed the Framingham Heart Risk Model (AUC -0.737 vs 0.707). CONCLUSIONS: This study concludes that the novel AI-based CVD Risk Score has a higher predictive performance for cardiac events than conventional risk scores in Indian population. TRIAL REGISTRATION NUMBER: CTRI/2019/07/020471.

SwinHR: Hemodynamic-powered hierarchical vision transformer for breast tumor segmentation.

Accurate and automated segmentation of breast tumors in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a critical role in computer-aided diagnosis and treatment of breast cancer. However, this task is challenging, due to random variation in tumor sizes, shapes, appearances, and blurred boundaries of tumors caused by inherent heterogeneity of breast cancer. Moreover, the presence of ill-posed artifacts in DCE-MRI further complicate the process of tumor region annotation. To address the challenges above, we propose a scheme (named SwinHR) integrating prior DCE-MRI knowledge and temporal-spatial information of breast tumors. The prior DCE-MRI knowledge refers to hemodynamic information extracted from multiple DCE-MRI phases, which can provide pharmacokinetics information to describe metabolic changes of the tumor cells over the scanning time. The Swin Transformer with hierarchical re-parameterization large kernel architecture (H-RLK) can capture long-range dependencies within DCE-MRI while maintaining computational efficiency by a shifted window-based self-attention mechanism. The use of H-RLK can extract high-level features with a wider receptive field, which can make the model capture contextual information at different levels of abstraction. Extensive experiments are conducted in large-scale datasets to validate the effectiveness of our proposed SwinHR scheme, demonstrating its superiority over recent state-of-the-art segmentation methods. Also, a subgroup analysis split by MRI scanners, field strength, and tumor size is conducted to verify its generalization. The source code is released on (https://github.com/GDPHMediaLab/SwinHR).

Assessing the effect of mHealth on child feeding practice in African countries: systematic and meta-analysis.

INTRODUCTION: Poor child feeding practice is a public health problem in Africa. Mobile health (mHealth) is a supportive intervention to improve this problem; however, the evidence available in the current literature is inconsistent and inconclusive in Africa. Some studies state that exclusive breastfeeding is not different between controls and mHealth interventions in the first month. Other studies state that health providers need additional training for the success of mHealth interventions. OBJECTIVE: This systematic review and meta-analysis aims to provide the summarized effect of mHealth on child-feeding practices in Africa to improve future planning and decisions. METHOD: We conducted a systematic review and meta-analysis based on the published and unpublished evidence gathered from PubMed, Web of Science, Cochrane Library, and Embase databases between January 1, 2000, and March 1, 2022. Studies included were randomized control trials and experimental studies that compared mHealth to standards of care among postpartum women. Preferred Reporting Items for Systematic Review and Meta-analysis guidelines followed for the reporting. RESULTS: After screening 1188 studies, we identified six studies that fulfilled the study criteria. These studies had 2913 participants with the number of total intervention groups 1627 [1627/2913 = 56%]. Five studies were completed within 24 weeks while one required 12 weeks. We included two RCTs, two cluster RCTs, and two quasi-experimental studies all used mHealth as the major intervention and usual care as controls. We found significant improvement in child-feeding practices among intervention groups. CONCLUSION: This systematic review and meta-analysis showed that the application of mHealth improved child-feeding practices in Africa. Although the finding is compelling, the authors recommend high-quality studies and mHealth interventions that consider sample size, design, regional differences, and environmental constraints to enhance policy decisions. The place of residence, access, low socioeconomic development, poor socio-demographic characteristics, low women empowerment, and low women's education might cause high heterogeneity in the included regions and need consideration during interventions. REGISTRATION NUMBER: PROSPERO: CRD42022346950.

Real-time auto-segmentation of the ureter in video sequences of gynaecological laparoscopic surgery.

BACKGROUND: Ureteral injury is common during gynaecological laparoscopic surgery. Real-time auto-segmentation can assist gynaecologists in identifying the ureter and reduce intraoperative injury risk. METHODS: A deep learning segmentation model was crafted for ureter recognition in surgical videos, utilising 3368 frames from 11 laparoscopic surgeries. Class activation maps enhanced the model's interpretability, showing its areas. The model's clinical relevance was validated through an End-User Turing test and verified by three gynaecological surgeons. RESULTS: The model registered a Dice score of 0.86, a Hausdorff 95 distance of 22.60, and processed images in 0.008 s on average. In complex surgeries, it pinpointed the ureter's position in real-time. Fifty five surgeons across eight institutions found the model's accuracy, specificity, and sensitivity comparable to human performance. Yet, artificial intelligence experience influenced some subjective ratings. CONCLUSIONS: The model offers precise real-time ureter segmentation in laparoscopic surgery and can be a significant tool for gynaecologists to mitigate ureteral injuries.

Image based prognosis in head and neck cancer using convolutional neural networks: a case study in reproducibility and optimization.

In the past decade, there has been a sharp increase in publications describing applications of convolutional neural networks (CNNs) in medical image analysis. However, recent reviews have warned of the lack of reproducibility of most such studies, which has impeded closer examination of the models and, in turn, their implementation in healthcare. On the other hand, the performance of these models is highly dependent on decisions on architecture and image pre-processing. In this work, we assess the reproducibility of three studies that use CNNs for head and neck cancer outcome prediction by attempting to reproduce the published results. In addition, we propose a new network structure and assess the impact of image pre-processing and model selection criteria on performance. We used two publicly available datasets: one with 298 patients for training and validation and another with 137 patients from a different institute for testing. All three studies failed to report elements required to reproduce their results thoroughly, mainly the image pre-processing steps and the random seed. Our model either outperforms or achieves similar performance to the existing models with considerably fewer parameters. We also observed that the pre-processing efforts significantly impact the model's performance and that some model selection criteria may lead to suboptimal models. Although there have been improvements in the reproducibility of deep learning models, our work suggests that wider implementation of reporting standards is required to avoid a reproducibility crisis.