In 2020, the adoption rate of electronic medical records (EMRs) in general hospitals in Japan reached 57.2%. This rate was 91.2% for hospitals with 400 or more beds, demonstrating substantial progress in EMR implementation, especially among larger healthcare institutions [1]. However, electronic health record (EHR) databases are primarily designed for administrative purposes and may not be appropriately structured for clinical research. In Japan, clinical research often utilizes databases such as the Japan Medical Data Center, the Diagnosis Procedure Combination database (Ministry of Health, Labor and Welfare), the Medical Data Vision database, and the National Database. Consequently, relatively few studies have leveraged EMR or EHR data from single or multiple hospitals with smaller patient populations [2].

In leveraging EHRs, EMR patient data must be integrated with existing disease registries to support critical components and ensure optimal health outcomes. However, this carries several challenges, including issues with data standardization [3], difficulty measuring data quality [4], and the absence of a standardized approach for assessing EHR data quality [5]. One study identified factors either promoting (63%; 147 of 232) or hindering (37%; 85 of 232) the adoption of EMRs [6]. In cancer care, the use of EHRs to complement cancer registry data can help identify health behaviors, clinical characteristics, and other factors that may influence patient prognosis [7]. Notably, analysis of EMR data alone tends to overestimate survival time compared to analyses that combine EMR information with cancer registry data [8]. Therefore, when repurposing EMR data, one must consider the potential for false positives or missing cancer cases, underscoring the need for analysis by appropriately trained experts [9]. Furthermore, bias may complicate the use of EHRs to provide information for cancer surveillance [10]. EMRs in cancer care contain information on diagnosis, treatment, and decision-making about treatment, place of care, and genetic testing. This information is vital for evaluating support to improve the quality of life among patients undergoing cancer treatment. However, no survey reports have examined actual descriptions associated with decision-making support.

EHRs have gained prominence as valuable sources of real-world data (RWD). Studies have reported the use of natural language processing (NLP) and machine learning (ML) techniques in analyzing EHRs, with minimal variance (<8%) observed in the results across different data curation methods [11]. The patterns discerned in EHR data through ML can be extrapolated to real-world settings. However, these techniques often reflect the specific structures of local EHR systems and may not be generalizable to other systems [12]. Additionally, EHR data are frequently stored in an unstructured format, including a variety of documents such as clinical notes, surgical records, and prescriptions. To uncover meaningful relationships among variables that could influence patient outcomes, these data must be transformed into a structured, computable format [13].

NLP and ML techniques are increasingly used to process patient-reported outcome (PRO) data. Several algorithms have been identified, underscoring the potential of NLP/ML technologies [14]. Furthermore, various ML and deep learning techniques have been employed in named entity recognition for the International Classification of Diseases, 9th Revision; analysis of clinical records; and research into mental health disorders [15]. NLP techniques can aid in converting vast amounts of unstructured PRO data found in EHRs into actionable clinical insights for patients with cancer [16]. In this context, we focused on the palliative care and breast oncology departments, where patients with cancer frequently face complex decisions. Our aim was to use nursing records to clarify the content regarding decision support. Such support from nurses involves assisting patients in improving the quality of their decisions related to treatment options, genetic testing, and care approaches for cancer.

The decision-making support provided by the OCNSs (NR1) and a record form that detailed the decision-making situation (NR2) were obtained and analyzed.

In our analysis of palliative care and internal medicine SOAP records, we identified words associated with the future, such as “treatment,” “future,” “progress,” “alleviation,” and “hope,” in category A. In contrast, words like “pain,” “visit,” “care,” “outpatient,” “alleviation,” and “hope” were found in category P. This discrepancy may exist because “hospitalization in the PCU” often represents a pivotal decision-making point, in which patients find hope despite the progression of their illness. Recent advancements in model development, including OpenAI’s ChatGPT, have opened new avenues for the efficient curation of EHRs. Models that use ML extraction techniques for databases derived from EHRs do not predict future events or impute missing information. Instead, they offer a method for extracting clinically meaningful information—contextual details rather than just clinical keywords— from documents [17]. This technology has the potential to improve the validity of assessments by automatically and efficiently extracting information from EHRs indicating the need for medical intervention. This approach is particularly beneficial for patients whose cancer has progressed to the point where palliative care is more prevalent than active treatment, as the S and O (or SO) information in the EMR provides more contextual insight into their lives and emotions.

Regarding decision-making support, SO was used to identify patients’ concerns about their medications, physical condition, and hospitalization, and to present treatment options while offering support that respected the patient’s preferences. However, the narrative communication and mental health-related content exchanged between patients and nurses during the support process were not captured in the nursing records. Evaluations of EMRs in the mental health domain have highlighted benefits such as reduced adverse events and better quality of care [18]. Accessing EHRs and gaining a deeper understanding of mental health issues can facilitate the provision of appropriate support [19]. Additionally, the inclusion of a mental health care plan in EHR systems has been associated with a lower risk of attempted suicide [20]. However, challenges arise from the lack of mental health-related information in these systems, potentially disrupting workflows [21]. Another limitation is that EHRs do not capture information obtained through narrative communication, such as details about a patient’s life, social connections, and personal values [22], which can be burdensome to document and review [19]. Appropriate support is feasible when mental health information is available in the EHR; however, concerns persist regarding the labor-intensive process of converting and interpreting these data. Therefore, a prevailing strategy is to equip the EHR with artificial intelligence that can automatically analyze the mental health information contained within it, screen patients who need assistance, and alert healthcare professionals through the EHR. Depressive tendencies may arise in patients with cancer who require more palliative care. Early detection of depressive tendencies from the SO information recorded in the EMR is effective for early intervention by applying mental health knowledge to patients with cancer.

Even after receiving decision-making support, some individuals still face challenges in identifying pathways to achieve their goals, envisioning their future lives, or accepting their illness or symptoms. An EHR-embedded colorectal cancer (CRC) screening decision support program, termed “e-assist: Colon Health,” has been developed to facilitate informed decision-making in this context. This program is designed to effectively address common barriers and questions related to CRC screening, bridging the communication gap between healthcare providers and patients [23]. The integration of such a program into the EHRs of palliative care outpatient clinics is promising for establishing a systematic approach to advance care planning from the outset, such as at the time of cancer diagnosis. However, while bridging the gap may increase the likelihood of promoting cancer screening behavior, advance care planning requires addressing the emotional responses to the unexpected situations that arise with disease progression.

In an analysis of SOAP records from the breast oncology department, words such as “child,” “metastasis,” and “concern” may have been prominent due to the frequent inclusion of “genetic testing” and “treatment selection” in the decision-making scenarios. An EMR-based RWD collection framework has been designed to evaluate the efficacy and safety of anticancer drugs [24]. A retrospective observational analysis of EMR data in patients with small cell lung cancer showed that those who experienced myelosuppressive adverse events had a greater need for supportive care interventions and more frequent visits to inpatient and outpatient facilities [25]. Patient-generated health data, collected using smart devices or digital health technologies, are key to self-care and shared clinical decision-making. Studies have shown that free text and unstructured data can be processed and analyzed using NLP techniques to produce meaningful summaries and insights [26]. Thus, more efficient decisionmaking support can be achieved by leveraging information gathering and free text analysis methods focused on adverse event management. However, genetic tests performed in breast oncology departments may include information related to the risk of hereditary tumors, which requires counseling; hence, careful attention must be paid to information management. Notably, perceptions of family and individuals differ between Japan and the West; therefore, responses must be tailored to cultural backgrounds.

The observed improvement in patient reactions following the provision of decision-making support may be due to medical personnel’s recognition of the concerns and anxieties of patients and their families. This is supported by their provision of thorough explanations about breast cancer treatment options, including surgery, and the need for genetic testing. Previous research has shown that recording social determinants of health in the EHR and using NLP to extract this data from narrative clinical notes can be instrumental in creating screening tools, risk prediction models, and clinical decision support systems [27]. Therefore, carefully choosing the information to document in the EHR and utilizing NLP can represent an effective strategy for providing personalized care that addresses the needs of patients and their families. After receiving decision support, some individuals may still struggle to achieve their goals, manage symptoms, or accept their illness and symptoms. A system must be established that can identify such patients and connect them to advanced decision support resources. Incorporating patient-tailored interventions into the EHR to promote CRC screening can facilitate practice integration. However, the program’s effectiveness could be compromised without proper segmentation of the patient population [28]. Therefore, when implementing a decision support program within an EHR, it is essential to include a screening function that categorizes the patient population based on specific information.

The study’s limitations are numerous decision-making scenarios related to genetic testing within the breast oncology department and admission to the palliative care ward. Consequently, the nursing records may have contained more contextual expressions to describe the phenomenon rather than concise information. This might have obscured the results of the analysis and hindered the identification of trends.

In summary, decision-making within the palliative care department frequently revolved around admission to the PCU. This suggests that patients continue to find hope as their disease progresses. Regarding decision-making support, the SO framework was utilized to collect patients’ concerns regarding medication, physical condition, and hospitalization, and to present treatment options that respected patient preferences. Decision-making in the breast oncology department often focused on “genetic testing” and “treatment options.” In the context of decision support, SO clearly played a key role in providing information and facilitating communication related to genetics. However, the development of a decision-making program that leverages EMRs requires further research. This research should aim to categorize the patient population based on EMR data and to identify markers indicating patients who will require care.