JCO Clinical Cancer Informatics最新文献

Purpose: The adoption of artificial intelligence (AI) in health care may afford new avenues for personalized and patient-centered care. This systematic review explored the role of AI in symptom monitoring for adult cancer survivors.

Methods: A comprehensive search was performed from inception to November 2023 in seven bibliographic databases and three clinical trial registries. This PROSPERO registered review (ID: CRD42023476027) assessed reports of empirical research studies of AI use in symptom monitoring (physical and psychological symptoms) across all cancer types in adults.

Results: A total of 18,530 reports were identified, of which 41 met review criteria and were analyzed. Included studies were predominantly published between 2021 and 2023, originated in the United States (39.0%) and Japan (14.6%), and primarily used cohort designs (80.5%), followed by cross-sectional designs (12.2%). The mean sample size was 617.14 (standard deviation = 1,401.37), with most studies primarily including multiple tumor types (31.7%) or breast cancer survivors (26.8%). Machine learning algorithms (43.9%) was the most used AI method, followed by natural language processing (29.3%), AI-driven chatbots (17.1%), and decision support tools (9.8%). The most common inputs to the AI algorithms were textual data, patient-reported symptoms, and physiologic measurements. The most examined symptom was pain (34.2% of studies), followed by fatigue and nausea (17.1% of studies each). Overall, the review showed increasing AI technology use in the prediction and monitoring of cancer symptoms.

Conclusion: AI is being used to enhance symptom monitoring in various cancer settings. When considering integration into clinical practice, standardization of data capture, the use of analytics, investing in infrastructure, and the end-user experience should be considered for successful implementation and monitoring the improvement of patient outcomes.

Purpose: Cancer-related pain is prevalent among people with advanced cancer. To improve accessibility and engagement with pain-cognitive behavioral therapy (pain-CBT), we developed and tested a serious game hosted within a mobile health intervention that delivers pain-CBT and pharmacologic support. The game focuses on teaching and practicing cognitive restructuring (CR), a central pain-CBT intervention component.

Methods: The pain-CBT game was developed through partnerships with commercial and academic game developers, graphic designers, clinical experts, and patients. Patients with metastatic cancer and pain participated in iterative, semistructured interviews. They described their experience playing each level and reflected on relevance, clarity, usability, and potential changes. Content codes captured patients' suggestions and informed game refinements.

Results: The final game includes five levels that prompt players to distinguish between adaptive and maladaptive thoughts that are pain- and cancer-specific. The levels vary in objective (eg, hiking and sledding), interaction type (eg, dragging and tapping), and mode of feedback (eg, audio and animation). Fourteen participants reviewed the game. Patients appreciated the pain- and cancer-specific thought examples, with a few noting that the thoughts made them feel less alone. Many stated that the game was fun, relatable, and an engaging distraction. Others noted that the game provided helpful CR practice and prompted reflection. For example, one 40-year-old woman said the game "brings [a thought] to the forefront so you can acknowledge it, and then maybe you could let it go or… do something about it."

Conclusion: Patients coping with cancer pain found the CR game helpful, enjoyable, and satisfactory. Serious games have the potential to increase engagement while facilitating learning and rehearsal of psychological skills for pain. Future testing will evaluate the efficacy of this serious game.

Purpose: Neoadjuvant chemotherapy (NAC) is increasingly used in breast cancer. Predictive modeling is useful in predicting pathologic complete response (pCR) to NAC. We test machine learning (ML) models to predict pCR in breast cancer and explore methods of handling missing data.

Methods: Four hundred and ninety-nine patients with breast cancer treated with NAC in two centers in Singapore (National Cancer Centre Singapore [NCCS] and KK Hospital) between January 2014 and December 2017 were included. Eleven clinical features were used to train five different ML models. Listwise deletion and imputation were evaluated on handling missing data. Model performance was evaluated by AUC and calibration (Brier score). Feature importance from the best performing model in the external testing data set was calculated using Shapley additive explanations.

Results: Seventy-two (24.6%), 18 (24.7%), and 31 (24.8%) patients attained pCR in NCCS training, NCCS testing, and KK Women's and Children's Hospital (KKH) testing data sets, respectively. The random forest (RF) base and imputed models have the highest AUCs in the KKH cohort of 0.794 (95% CI, 0.709 to 0.873) and 0.795 (95% CI, 0.706 to 0.871), respectively, and were the best calibrated with the lowest Brier score. No statistically significant difference was noted between AUCs of the base and imputed models in all data sets. The imputed model had a larger positive predictive value (PPV; 98.2% v 95.1%) and negative predictive value (NPV; 96.7% v 90.0%) than the base model in the KKH data set. Estrogen receptor intensity, human epidermal growth factor 2 intensity, and age at diagnosis were the three most important predictors.

Conclusion: ML, particularly RF, demonstrates reasonable accuracy in pCR prediction after NAC. Imputing missing fields in the data can improve the PPV and NPV of the pCR prediction model.

Purpose: Electronic patient-reported outcome (ePRO) tools are increasingly used to provide first-hand information on patient's symptoms and quality of life. This study explored how patients and health care providers (HCPs) perceive the use of a digital real-time ePRO tool, coupled with digital analytics at a cancer center located in a majority-minority county. Furthermore, we described the implementation barriers and facilitators identified from the participants' perspectives.

Methods: We conducted a qualitative substudy as part of a larger implementation study conducted at University of California Irvine Chao Family Comprehensive Cancer Center. Patients and HCPs completed semistructured interviews and a focus group discussion. Thematic analysis was used to identify key themes regarding perceived impact of the intervention on patient's care and implementation factors.

Results: A total of 31 participants, comprising 15 patients (67% English-speaking, 33% Spanish-speaking) and 16 HCPs (43.8% pharmacists, 37.5% physicians, 18.8% nurses), were interviewed. The utilization of real-time ePRO was perceived to beneficially affect patient care, improve patient-provider communication, and increase symptom awareness. Implementation facilitators included ease of comprehension and completion within the infusion center. Barriers included the need to incorporate results in electronic medical records and create real-time referral pathways to address patient's needs.

Conclusion: The use of real-time ePRO in a majority-minority population was perceived to enhance patient-centered oncology care, yet implementation barriers must be addressed for successful integration in clinical settings. The findings from this study may inform implementation strategies to reduce health disparities.

Purpose: Patients with cancer visit the emergency department (ED) frequently. While some ED visits are necessary, others may be potentially preventable ED visits (PPEDs). Reducing PPEDs is important to improve quality of care and reduce costs. However, a robust definition and the characteristics of patients at risk remain unclear. This study aimed to describe oncology-related PPEDs and identify characteristics of patients at the highest risk for PPEDs to help target interventions and minimize avoidable ED visits.

Methods: A retrospective study was conducted using four clinical and administrative databases. All ED visits by oncology patients between April 1, 2019, and April 1, 2021, were identified. A novel definition of PPEDs was explored, specifically visits that resulted in immediate discharge from the ED or admissions <48 hours. Trends in ED use, including PPEDs, were evaluated using descriptive statistics, logistic regression, and machine learning (ML) modeling.

Results: During the 2-year period, 6,689 oncology patients visited the ED (N = 13,415 visits). A total of 62.1% of visits were classified as PPEDs. PPEDs were most common among patients with stage I to III breast cancer and those on systemic therapy. Characteristics of patients at high risk for non-PPEDs included stage IV disease with either lung or GI carcinomas and shorter distances to the ED. The highest-performing ML model yielded an AUC of 0.819.

Conclusion: Our novel definition of PPEDs appears promising in identifying oncology patients who could avoid the ED with targeted interventions. This work demonstrated that patients with early-stage disease, those with breast cancer, and those on systemic therapy are at the highest risk for PPEDs and may benefit from proactive interventions to avoid the ED. Although our definition requires validation, using ML models for more robust predictive modeling appears promising.

The integration of large language models (LLMs) into oncology is transforming patients' journeys through education, diagnosis, treatment monitoring, and follow-up. This review examines the current landscape, potential benefits, and associated ethical and regulatory considerations of the application of LLMs for patients in the oncologic domain.

Purpose: Machine learning algorithms are used for predictive modeling in medicine, but studies often do not evaluate or report on the potential biases of the models. Our purpose was to develop clinical prediction models for readmission after surgery in colorectal cancer (CRC) patients and to examine their potential for racial bias.

Methods: We used the 2012-2020 American College of Surgeons' National Surgical Quality Improvement Program (ACS-NSQIP) Participant Use File and Targeted Colectomy File. Patients were categorized into four race groups - White, Black or African American, Other, and Unknown/Not Reported. Potential predictive features were identified from studies of risk factors of 30-day readmission in CRC patients. We compared four machine learning-based methods - logistic regression (LR), multilayer perceptron (MLP), random forest (RF), and XGBoost (XGB). Model bias was assessed using false negative rate (FNR) difference, false positive rate (FPR) difference, and disparate impact.

Results: In all, 112,077 patients were included, 67.2% of whom were White, 9.2% Black, 5.6% Other race, and 18% with race not recorded. There were significant differences in the AUROC, FPR and FNR between race groups across all models. Notably, patients in the 'Other' race category had higher FNR compared to Black patients in all but the XGB model, while Black patients had higher FPR than White patients in some models. Patients in the 'Other' category consistently had the lowest FPR. Applying the 80% rule for disparate impact, the models consistently met the threshold for unfairness for the 'Other' race category.

Conclusion: Predictive models for 30-day readmission after colorectal surgery may perform unequally for different race groups, potentially propagating to inequalities in delivery of care and patient outcomes if the predictions from these models are used to direct care.

Purpose: The integration of patient-reported outcomes (PROs) into electronic health records (EHRs) has enabled systematic collection of symptom data to manage post-treatment symptoms. The use and integration of PRO data into routine care are associated with overall treatment success, adherence, and satisfaction. Clinical trials have demonstrated the prognostic value of PROs including physical function and global health status in predicting survival. It is unknown to what extent routinely collected PRO data are used in the development of risk prediction models (RPMs) in oncology care. The objective of the scoping review is to assess how PROs are used to train risk RPMs to predict patient outcomes in oncology care.

Methods: Using the scoping review methodology outlined in the Joanna Briggs Institute Manual for Evidence Synthesis, we searched four databases (MEDLINE, CINAHL, Embase, and Web of Science) to locate peer-reviewed oncology articles that used PROs as predictors to train models. Study characteristics including settings, clinical outcomes, and model training, testing, validation, and performance data were extracted for analyses.

Results: Of the 1,254 studies identified, 18 met inclusion criteria. Most studies performed retrospective analyses of prospectively collected PRO data to build prediction models. Post-treatment survival was the most common outcome predicted. Discriminative performance of models trained using PROs was better than models trained without PROs. Most studies did not report model calibration.

Conclusion: Systematic collection of PROs in routine practice provides an opportunity to use patient-reported data to develop RPMs. Model performance improves when PROs are used in combination with other comprehensive data sources.

Purpose: Population-based cancer registries (PBCRs) collect data on all new cancer diagnoses in a defined population. Data are sourced from pathology reports, and the PBCRs rely on manual and rule-based solutions. This study presents a state-of-the-art natural language processing (NLP) pipeline, built by fine-tuning pretrained language models (LMs). The pipeline is deployed at the British Columbia Cancer Registry (BCCR) to detect reportable tumors from a population-based feed of electronic pathology.

Methods: We fine-tune two publicly available LMs, GatorTron and BlueBERT, which are pretrained on clinical text. Fine-tuning is done using BCCR's pathology reports. For the final decision making, we combine both models' output using an OR approach. The fine-tuning data set consisted of 40,000 reports from the diagnosis year of 2021, and the test data sets consisted of 10,000 reports from the diagnosis year 2021, 20,000 reports from diagnosis year 2022, and 400 reports from diagnosis year 2023.

Results: The retrospective evaluation of our proposed approach showed boosted reportable accuracy, maintaining the true reportable threshold of 98%.

Conclusion: Disadvantages of rule-based NLP in cancer surveillance include manual effort in rule design and sensitivity to language change. Deep learning approaches demonstrate superior performance in classification. PBCRs distinguish reportability status of incoming electronic cancer pathology reports. Deep learning methods provide significant advantages over rule-based NLP.