ABSTRACT
BACKGROUND: In the realm of in vitro fertilization (IVF), artificial intelligence (AI) models serve as invaluable tools for clinicians, offering predictive insights into ovarian stimulation outcomes. Predicting and understanding a patient's response to ovarian stimulation can help in personalizing doses of drugs, preventing adverse outcomes (eg, hyperstimulation), and improving the likelihood of successful fertilization and pregnancy. Given the pivotal role of accurate predictions in IVF procedures, it becomes important to investigate the landscape of AI models that are being used to predict the outcomes of ovarian stimulation. OBJECTIVE: The objective of this review is to comprehensively examine the literature to explore the characteristics of AI models used for predicting ovarian stimulation outcomes in the context of IVF. METHODS: A total of 6 electronic databases were searched for peer-reviewed literature published before August 2023, using the concepts of IVF and AI, along with their related terms. Records were independently screened by 2 reviewers against the eligibility criteria. The extracted data were then consolidated and presented through narrative synthesis. RESULTS: Upon reviewing 1348 articles, 30 met the predetermined inclusion criteria. The literature primarily focused on the number of oocytes retrieved as the main predicted outcome. Microscopy images stood out as the primary ground truth reference. The reviewed studies also highlighted that the most frequently adopted stimulation protocol was the gonadotropin-releasing hormone (GnRH) antagonist. In terms of using trigger medication, human chorionic gonadotropin (hCG) was the most commonly selected option. Among the machine learning techniques, the favored choice was the support vector machine. As for the validation of AI algorithms, the hold-out cross-validation method was the most prevalent. The area under the curve was highlighted as the primary evaluation metric. The literature exhibited a wide variation in the number of features used for AI algorithm development, ranging from 2 to 28,054 features. Data were mostly sourced from patient demographics, followed by laboratory data, specifically hormonal levels. Notably, the vast majority of studies were restricted to a single infertility clinic and exclusively relied on nonpublic data sets. CONCLUSIONS: These insights highlight an urgent need to diversify data sources and explore varied AI techniques for improved prediction accuracy and generalizability of AI models for the prediction of ovarian stimulation outcomes. Future research should prioritize multiclinic collaborations and consider leveraging public data sets, aiming for more precise AI-driven predictions that ultimately boost patient care and IVF success rates.
Subject(s)
Artificial Intelligence , Fertilization in Vitro , Ovulation Induction , Humans , Ovulation Induction/methods , Fertilization in Vitro/methods , Female , PregnancyABSTRACT
BACKGROUND: Research gaps refer to unanswered questions in the existing body of knowledge, either due to a lack of studies or inconclusive results. Research gaps are essential starting points and motivation in scientific research. Traditional methods for identifying research gaps, such as literature reviews and expert opinions, can be time consuming, labor intensive, and prone to bias. They may also fall short when dealing with rapidly evolving or time-sensitive subjects. Thus, innovative scalable approaches are needed to identify research gaps, systematically assess the literature, and prioritize areas for further study in the topic of interest. OBJECTIVE: In this paper, we propose a machine learning-based approach for identifying research gaps through the analysis of scientific literature. We used the COVID-19 pandemic as a case study. METHODS: We conducted an analysis to identify research gaps in COVID-19 literature using the COVID-19 Open Research (CORD-19) data set, which comprises 1,121,433 papers related to the COVID-19 pandemic. Our approach is based on the BERTopic topic modeling technique, which leverages transformers and class-based term frequency-inverse document frequency to create dense clusters allowing for easily interpretable topics. Our BERTopic-based approach involves 3 stages: embedding documents, clustering documents (dimension reduction and clustering), and representing topics (generating candidates and maximizing candidate relevance). RESULTS: After applying the study selection criteria, we included 33,206 abstracts in the analysis of this study. The final list of research gaps identified 21 different areas, which were grouped into 6 principal topics. These topics were: "virus of COVID-19," "risk factors of COVID-19," "prevention of COVID-19," "treatment of COVID-19," "health care delivery during COVID-19," "and impact of COVID-19." The most prominent topic, observed in over half of the analyzed studies, was "the impact of COVID-19." CONCLUSIONS: The proposed machine learning-based approach has the potential to identify research gaps in scientific literature. This study is not intended to replace individual literature research within a selected topic. Instead, it can serve as a guide to formulate precise literature search queries in specific areas associated with research questions that previous publications have earmarked for future exploration. Future research should leverage an up-to-date list of studies that are retrieved from the most common databases in the target area. When feasible, full texts or, at minimum, discussion sections should be analyzed rather than limiting their analysis to abstracts. Furthermore, future studies could evaluate more efficient modeling algorithms, especially those combining topic modeling with statistical uncertainty quantification, such as conformal prediction.
ABSTRACT
Effective modeling of patient representation from electronic health records (EHRs) is increasingly becoming a vital research topic. Yet, modeling the non-stationarity in EHR data has received less attention. Most existing studies follow a strong assumption of stationarity in patient representation from EHRs. However, in practice, a patient's visits are irregularly spaced over a relatively long period of time, and disease progression patterns exhibit non-stationarity. Furthermore, the time gaps between patient visits often encapsulate significant domain knowledge, potentially revealing undiscovered patterns that characterize specific medical conditions. To address these challenges, we introduce a new method which combines the self-attention mechanism with non-stationary kernel approximation to capture both contextual information and temporal relationships between patient visits in EHRs. To assess the effectiveness of our proposed approach, we use two real-world EHR datasets, comprising a total of 76,925 patients, for the task of predicting the next diagnosis code for a patient, given their EHR history. The first dataset is a general EHR cohort and consists of 11,451 patients with a total of 3,485 unique diagnosis codes. The second dataset is a disease-specific cohort that includes 65,474 pregnant patients and encompasses a total of 9,782 unique diagnosis codes. Our experimental evaluation involved nine prediction models, categorized into three distinct groups. Group 1 comprises the baselines: original self-attention with positional encoding model, RETAIN model, and LSTM model. Group 2 includes models employing self-attention with stationary kernel approximations, specifically incorporating three variations of Bochner's feature maps. Lastly, Group 3 consists of models utilizing self-attention with non-stationary kernel approximations, including quadratic, cubic, and bi-quadratic polynomials. The experimental results demonstrate that non-stationary kernels significantly outperformed baseline methods for NDCG@10 and Hit@10 metrics in both datasets. The performance boost was more substantial in dataset 1 for the NDCG@10 metric. On the other hand, stationary Kernels showed significant but smaller gains over baselines and were nearly as effective as Non-stationary Kernels for Hit@10 in dataset 2. These findings robustly validate the efficacy of employing non-stationary kernels for temporal modeling of EHR data, and emphasize the importance of modeling non-stationary temporal information in healthcare prediction tasks.
Subject(s)
Algorithms , Electronic Health Records , Humans , Disease ProgressionABSTRACT
BACKGROUND: Students usually encounter stress throughout their academic path. Ongoing stressors may lead to chronic stress, adversely affecting their physical and mental well-being. Thus, early detection and monitoring of stress among students are crucial. Wearable artificial intelligence (AI) has emerged as a valuable tool for this purpose. It offers an objective, noninvasive, nonobtrusive, automated approach to continuously monitor biomarkers in real time, thereby addressing the limitations of traditional approaches such as self-reported questionnaires. OBJECTIVE: This systematic review and meta-analysis aim to assess the performance of wearable AI in detecting and predicting stress among students. METHODS: Search sources in this review included 7 electronic databases (MEDLINE, Embase, PsycINFO, ACM Digital Library, Scopus, IEEE Xplore, and Google Scholar). We also checked the reference lists of the included studies and checked studies that cited the included studies. The search was conducted on June 12, 2023. This review included research articles centered on the creation or application of AI algorithms for the detection or prediction of stress among students using data from wearable devices. In total, 2 independent reviewers performed study selection, data extraction, and risk-of-bias assessment. The Quality Assessment of Diagnostic Accuracy Studies-Revised tool was adapted and used to examine the risk of bias in the included studies. Evidence synthesis was conducted using narrative and statistical techniques. RESULTS: This review included 5.8% (19/327) of the studies retrieved from the search sources. A meta-analysis of 37 accuracy estimates derived from 32% (6/19) of the studies revealed a pooled mean accuracy of 0.856 (95% CI 0.70-0.93). Subgroup analyses demonstrated that the accuracy of wearable AI was moderated by the number of stress classes (P=.02), type of wearable device (P=.049), location of the wearable device (P=.02), data set size (P=.009), and ground truth (P=.001). The average estimates of sensitivity, specificity, and F1-score were 0.755 (SD 0.181), 0.744 (SD 0.147), and 0.759 (SD 0.139), respectively. CONCLUSIONS: Wearable AI shows promise in detecting student stress but currently has suboptimal performance. The results of the subgroup analyses should be carefully interpreted given that many of these findings may be due to other confounding factors rather than the underlying grouping characteristics. Thus, wearable AI should be used alongside other assessments (eg, clinical questionnaires) until further evidence is available. Future research should explore the ability of wearable AI to differentiate types of stress, distinguish stress from other mental health issues, predict future occurrences of stress, consider factors such as the placement of the wearable device and the methods used to assess the ground truth, and report detailed results to facilitate the conduct of meta-analyses. TRIAL REGISTRATION: PROSPERO CRD42023435051; http://tinyurl.com/3fzb5rnp.
Subject(s)
Algorithms , Artificial Intelligence , Humans , Databases, Factual , Libraries, Digital , Mental HealthABSTRACT
BACKGROUND: Anxiety disorders rank among the most prevalent mental disorders worldwide. Anxiety symptoms are typically evaluated using self-assessment surveys or interview-based assessment methods conducted by clinicians, which can be subjective, time-consuming, and challenging to repeat. Therefore, there is an increasing demand for using technologies capable of providing objective and early detection of anxiety. Wearable artificial intelligence (AI), the combination of AI technology and wearable devices, has been widely used to detect and predict anxiety disorders automatically, objectively, and more efficiently. OBJECTIVE: This systematic review and meta-analysis aims to assess the performance of wearable AI in detecting and predicting anxiety. METHODS: Relevant studies were retrieved by searching 8 electronic databases and backward and forward reference list checking. In total, 2 reviewers independently carried out study selection, data extraction, and risk-of-bias assessment. The included studies were assessed for risk of bias using a modified version of the Quality Assessment of Diagnostic Accuracy Studies-Revised. Evidence was synthesized using a narrative (ie, text and tables) and statistical (ie, meta-analysis) approach as appropriate. RESULTS: Of the 918 records identified, 21 (2.3%) were included in this review. A meta-analysis of results from 81% (17/21) of the studies revealed a pooled mean accuracy of 0.82 (95% CI 0.71-0.89). Meta-analyses of results from 48% (10/21) of the studies showed a pooled mean sensitivity of 0.79 (95% CI 0.57-0.91) and a pooled mean specificity of 0.92 (95% CI 0.68-0.98). Subgroup analyses demonstrated that the performance of wearable AI was not moderated by algorithms, aims of AI, wearable devices used, status of wearable devices, data types, data sources, reference standards, and validation methods. CONCLUSIONS: Although wearable AI has the potential to detect anxiety, it is not yet advanced enough for clinical use. Until further evidence shows an ideal performance of wearable AI, it should be used along with other clinical assessments. Wearable device companies need to develop devices that can promptly detect anxiety and identify specific time points during the day when anxiety levels are high. Further research is needed to differentiate types of anxiety, compare the performance of different wearable devices, and investigate the impact of the combination of wearable device data and neuroimaging data on the performance of wearable AI. TRIAL REGISTRATION: PROSPERO CRD42023387560; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=387560.
Subject(s)
Anxiety , Artificial Intelligence , Humans , Anxiety/diagnosis , Anxiety Disorders , Algorithms , Databases, FactualABSTRACT
Objective This study aims to assess the knowledge and attitudes toward clinical trial (CT) participation among the adult population in the Eastern Province of Saudi Arabia. Material and methods This cross-sectional study was conducted among the population of the Eastern Province of Saudi Arabia. A self-administered questionnaire was distributed among the general population using an online survey. Results A total of 334 participants completed the questionnaire. Participants' ages ranged from 18 to 65 years, with a mean age of 31.2 ± 13.9 years, 56.6% of whom were males, 42.2% were employed, 29.6% were students, and 23.1% were unemployed. Surprisingly, only a small percentage of respondents (7.5%) were requested to participate in a randomized controlled trial (RCT), of which the majority did partake. Additionally, 25.4% of participants believe CTs are used to evaluate new drugs; others believe that CTs are used to understand diseases and human behavior. The data show that most participants believe that CTs improve patient care, welfare, and society. Also, participants were more likely to take part if they were aware of the study's purpose and findings and were given more time to consider their options. Conclusion Participants believed that the biggest obstacle was a lack of knowledge of CTs. It is crucial to educate patients more about CTs. Multimodal strategies such as improved patient-provider communication and online information for trial information sharing may be effective in boosting knowledge and CT recruitment.
ABSTRACT
Attention, which is the process of noticing the surrounding environment and processing information, is one of the cognitive functions that deteriorate gradually as people grow older. Games that are used for other than entertainment, such as improving attention, are often referred to as serious games. This study examined the effectiveness of serious games on attention among elderly individuals suffering from cognitive impairment. A systematic review and meta-analyses of randomized controlled trials were carried out. A total of 10 trials ultimately met all eligibility criteria of the 559 records retrieved. The synthesis of very low-quality evidence from three trials, as analyzed in a meta-study, indicated that serious games outperform no/passive interventions in enhancing attention in cognitively impaired older adults (P < 0.001). Additionally, findings from two other studies demonstrated that serious games are more effective than traditional cognitive training in boosting attention among cognitively impaired older adults. One study also concluded that serious games are better than traditional exercises in enhancing attention. Serious games can enhance attention in cognitively impaired older adults. However, given the low quality of the evidence, the limited number of participants in most studies, the absence of some comparative studies, and the dearth of studies included in the meta-analyses, the results remain inconclusive. Thus, until the aforementioned limitations are rectified in future research, serious games should serve as a supplement, rather than a replacement, to current interventions.
ABSTRACT
Depression is a prevalent mental condition that is challenging to diagnose using conventional techniques. Using machine learning and deep learning models with motor activity data, wearable AI technology has shown promise in reliably and effectively identifying or predicting depression. In this work, we aim to examine the performance of simple linear and non-linear models in the prediction of depression levels. We compared eight linear and non-linear models (Ridge, ElasticNet, Lasso, Random Forest, Gradient boosting, Decision trees, Support vector machines, and Multilayer perceptron) for the task of predicting depression scores over a period using physiological features, motor activity data, and MADRAS scores. For the experimental evaluation, we used the Depresjon dataset which contains the motor activity data of depressed and non-depressed participants. According to our findings, simple linear and non-linear models may effectively estimate depression scores for depressed people without the need for complex models. This opens the door for the development of more effective and impartial techniques for identifying depression and treating/preventing it using commonly used, widely accessible wearable technology.
Subject(s)
Artificial Intelligence , Depression , Humans , Depression/diagnosis , India , Neural Networks, Computer , Machine LearningABSTRACT
The integration of large language models (LLMs), such as those in the Generative Pre-trained Transformers (GPT) series, into medical education has the potential to transform learning experiences for students and elevate their knowledge, skills, and competence. Drawing on a wealth of professional and academic experience, we propose that LLMs hold promise for revolutionizing medical curriculum development, teaching methodologies, personalized study plans and learning materials, student assessments, and more. However, we also critically examine the challenges that such integration might pose by addressing issues of algorithmic bias, overreliance, plagiarism, misinformation, inequity, privacy, and copyright concerns in medical education. As we navigate the shift from an information-driven educational paradigm to an artificial intelligence (AI)-driven educational paradigm, we argue that it is paramount to understand both the potential and the pitfalls of LLMs in medical education. This paper thus offers our perspective on the opportunities and challenges of using LLMs in this context. We believe that the insights gleaned from this analysis will serve as a foundation for future recommendations and best practices in the field, fostering the responsible and effective use of AI technologies in medical education.
ABSTRACT
Given the limitations of traditional approaches, wearable artificial intelligence (AI) is one of the technologies that have been exploited to detect or predict depression. The current review aimed at examining the performance of wearable AI in detecting and predicting depression. The search sources in this systematic review were 8 electronic databases. Study selection, data extraction, and risk of bias assessment were carried out by two reviewers independently. The extracted results were synthesized narratively and statistically. Of the 1314 citations retrieved from the databases, 54 studies were included in this review. The pooled mean of the highest accuracy, sensitivity, specificity, and root mean square error (RMSE) was 0.89, 0.87, 0.93, and 4.55, respectively. The pooled mean of lowest accuracy, sensitivity, specificity, and RMSE was 0.70, 0.61, 0.73, and 3.76, respectively. Subgroup analyses revealed that there is a statistically significant difference in the highest accuracy, lowest accuracy, highest sensitivity, highest specificity, and lowest specificity between algorithms, and there is a statistically significant difference in the lowest sensitivity and lowest specificity between wearable devices. Wearable AI is a promising tool for depression detection and prediction although it is in its infancy and not ready for use in clinical practice. Until further research improve its performance, wearable AI should be used in conjunction with other methods for diagnosing and predicting depression. Further studies are needed to examine the performance of wearable AI based on a combination of wearable device data and neuroimaging data and to distinguish patients with depression from those with other diseases.
ABSTRACT
BACKGROUND: Learning disabilities are among the major cognitive impairments caused by aging. Among the interventions used to improve learning among older adults are serious games, which are participative electronic games designed for purposes other than entertainment. Although some systematic reviews have examined the effectiveness of serious games on learning, they are undermined by some limitations, such as focusing on older adults without cognitive impairments, focusing on particular types of serious games, and not considering the comparator type in the analysis. OBJECTIVE: This review aimed to evaluate the effectiveness of serious games on verbal and nonverbal learning among older adults with cognitive impairment. METHODS: Eight electronic databases were searched to retrieve studies relevant to this systematic review and meta-analysis. Furthermore, we went through the studies that cited the included studies and screened the reference lists of the included studies and relevant reviews. Two reviewers independently checked the eligibility of the identified studies, extracted data from the included studies, and appraised their risk of bias and the quality of the evidence. The results of the included studies were summarized using a narrative synthesis or meta-analysis, as appropriate. RESULTS: Of the 559 citations retrieved, 11 (2%) randomized controlled trials (RCTs) ultimately met all eligibility criteria for this review. A meta-analysis of 45% (5/11) of the RCTs revealed that serious games are effective in improving verbal learning among older adults with cognitive impairment in comparison with no or sham interventions (P=.04), and serious games do not have a different effect on verbal learning between patients with mild cognitive impairment and those with Alzheimer disease (P=.89). A meta-analysis of 18% (2/11) of the RCTs revealed that serious games are as effective as conventional exercises in promoting verbal learning (P=.98). We also found that serious games outperformed no or sham interventions (4/11, 36%; P=.03) and conventional cognitive training (2/11, 18%; P<.001) in enhancing nonverbal learning. CONCLUSIONS: Serious games have the potential to enhance verbal and nonverbal learning among older adults with cognitive impairment. However, our findings remain inconclusive because of the low quality of evidence, the small sample size in most of the meta-analyzed studies (6/8, 75%), and the paucity of studies included in the meta-analyses. Thus, until further convincing proof of their effectiveness is offered, serious games should be used to supplement current interventions for verbal and nonverbal learning rather than replace them entirely. Further studies are needed to compare serious games with conventional cognitive training and conventional exercises, as well as different types of serious games, different platforms, different intervention periods, and different follow-up periods. TRIAL REGISTRATION: PROSPERO CRD42022348849; https://tinyurl.com/y6yewwfa.
Subject(s)
Alzheimer Disease , Cognitive Dysfunction , Exergaming , Memory, Episodic , Aged , Humans , Cognitive Dysfunction/therapy , Exercise , LearningABSTRACT
[This corrects the article DOI: 10.2196/42672.].
ABSTRACT
BACKGROUND: Anxiety and depression are the most common mental disorders worldwide. Owing to the lack of psychiatrists around the world, the incorporation of artificial intelligence (AI) into wearable devices (wearable AI) has been exploited to provide mental health services. OBJECTIVE: This review aimed to explore the features of wearable AI used for anxiety and depression to identify application areas and open research issues. METHODS: We searched 8 electronic databases (MEDLINE, PsycINFO, Embase, CINAHL, IEEE Xplore, ACM Digital Library, Scopus, and Google Scholar) and included studies that met the inclusion criteria. Then, we checked the studies that cited the included studies and screened studies that were cited by the included studies. The study selection and data extraction were carried out by 2 reviewers independently. The extracted data were aggregated and summarized using narrative synthesis. RESULTS: Of the 1203 studies identified, 69 (5.74%) were included in this review. Approximately, two-thirds of the studies used wearable AI for depression, whereas the remaining studies used it for anxiety. The most frequent application of wearable AI was in diagnosing anxiety and depression; however, none of the studies used it for treatment purposes. Most studies targeted individuals aged between 18 and 65 years. The most common wearable device used in the studies was Actiwatch AW4 (Cambridge Neurotechnology Ltd). Wrist-worn devices were the most common type of wearable device in the studies. The most commonly used category of data for model development was physical activity data, followed by sleep data and heart rate data. The most frequently used data set from open sources was Depresjon. The most commonly used algorithm was random forest, followed by support vector machine. CONCLUSIONS: Wearable AI can offer great promise in providing mental health services related to anxiety and depression. Wearable AI can be used by individuals for the prescreening assessment of anxiety and depression. Further reviews are needed to statistically synthesize the studies' results related to the performance and effectiveness of wearable AI. Given its potential, technology companies should invest more in wearable AI for the treatment of anxiety and depression.
Subject(s)
Artificial Intelligence , Depression , Humans , Adolescent , Young Adult , Adult , Middle Aged , Aged , Depression/diagnosis , Depression/therapy , Anxiety/diagnosis , Anxiety/therapy , Anxiety Disorders , AlgorithmsABSTRACT
BACKGROUND: Early identification of pregnant women at risk for preterm birth (PTB), a major cause of infant mortality and morbidity, has a significant potential to improve prenatal care. However, we lack effective predictive models which can accurately forecast PTB and complement these predictions with appropriate interpretations for clinicians. In this work, we introduce a clinical prediction model (PredictPTB) which combines variables (medical codes) readily accessible through electronic health record (EHR) to accurately predict the risk of preterm birth at 1, 3, 6, and 9 months prior to delivery. METHODS: The architecture of PredictPTB employs recurrent neural networks (RNNs) to model the longitudinal patient's EHR visits and exploits a single code-level attention mechanism to improve the predictive performance, while providing temporal code-level and visit-level explanations for the prediction results. We compare the performance of different combinations of prediction time-points, data modalities, and data windows. We also present a case-study of our model's interpretability illustrating how clinicians can gain some transparency into the predictions. RESULTS: Leveraging a large cohort of 222,436 deliveries, comprising a total of 27,100 unique clinical concepts, our model was able to predict preterm birth with an ROC-AUC of 0.82, 0.79, 0.78, and PR-AUC of 0.40, 0.31, 0.24, at 1, 3, and 6 months prior to delivery, respectively. Results also confirm that observational data modalities (such as diagnoses) are more predictive for preterm birth than interventional data modalities (e.g., medications and procedures). CONCLUSIONS: Our results demonstrate that PredictPTB can be utilized to achieve accurate and scalable predictions for preterm birth, complemented by explanations that directly highlight evidence in the patient's EHR timeline.
ABSTRACT
BACKGROUND: Predictive modeling with longitudinal electronic health record (EHR) data offers great promise for accelerating personalized medicine and better informs clinical decision-making. Recently, deep learning models have achieved state-of-the-art performance for many healthcare prediction tasks. However, deep models lack interpretability, which is integral to successful decision-making and can lead to better patient care. In this paper, we build upon the contextual decomposition (CD) method, an algorithm for producing importance scores from long short-term memory networks (LSTMs). We extend the method to bidirectional LSTMs (BiLSTMs) and use it in the context of predicting future clinical outcomes using patients' EHR historical visits. METHODS: We use a real EHR dataset comprising 11071 patients, to evaluate and compare CD interpretations from LSTM and BiLSTM models. First, we train LSTM and BiLSTM models for the task of predicting which pre-school children with respiratory system-related complications will have asthma at school-age. After that, we conduct quantitative and qualitative analysis to evaluate the CD interpretations produced by the contextual decomposition of the trained models. In addition, we develop an interactive visualization to demonstrate the utility of CD scores in explaining predicted outcomes. RESULTS: Our experimental evaluation demonstrate that whenever a clear visit-level pattern exists, the models learn that pattern and the contextual decomposition can appropriately attribute the prediction to the correct pattern. In addition, the results confirm that the CD scores agree to a large extent with the importance scores generated using logistic regression coefficients. Our main insight was that rather than interpreting the attribution of individual visits to the predicted outcome, we could instead attribute a model's prediction to a group of visits. CONCLUSION: We presented a quantitative and qualitative evidence that CD interpretations can explain patient-specific predictions using CD attributions of individual visits or a group of visits.