Fast Healthcare Interoperability Resources (FHIR) is gaining popularity as a standard framework for the exchange of electronic health record (EHR) data. Despite the advantages of FHIR, it is dificult for clinicians to understand the data in EHR. To support clinicians in accessing data about a patient, we created a pipeline that extracts, transforms, and visualizes patient data from FHIR. We employ a web-based timeline visualization that shows all clinical data recorded for the patient over their disease trajectory. This can help clinicians to use the patient data more eficiently and to get a clear picture of the patient's disease progress and physical condition more quickly, which could help them to develop the best treatment plan for their patients. The source code with an example synthetic, but realistic patient is available at https://github.com/rtg-wispermed/Patient_trajectory_public.
tools that i) process FHIR data, ii) select and transform it for the use case at hand, and iii) display the relevant data in a user-friendly manner.
Visualizations allow users to understand information more efectively, thereby improving
judgment and decision-making performance [
In this paper, we present a proof-of-concept for a visualization pipeline that obtains patient data from EHR in FHIR and apply it to a use case of melanoma patients. In place of the long-term and complex development of visualization tools, we utilize the publicly available, web-based charting framework AnyChart2.
Our pipeline consists of four steps (cf. Figure 1): i) patient selection, ii) relevance judgment of attributes by domain experts, iii) extraction of relevant attributes, and iv) visualization.
First, the patient cohort of interest is selected and exported from the hospital EHR database based on defined inclusion and exclusion criteria. This export is done using high-level FHIR APIs, such as FHIR-PYrate3 or fhircrackr 4. The primary stakeholder of our visualization are clinicians, each with specific information needs about their patients. EHRs contain comprehensive information about patients, including information that is not relevant for a first assessment. To identify the relevant information for the visualization, we let clinicians rate the available information. Clinicians rate each available attribute on a 3-point Likert scale (0: not important, 2https://github.com/AnyChart, last accessed 12.07.2023 3https://github.com/UMEssen/FHIR-PYrate, last accessed 07.06.2023 4https://github.com/POLAR-fhiR/fhircrackr, last accessed 07.06.2023 1: moderately important, 2: very important). We then selected the attributes that at least one clinician considered very important. In the whole process, domain experts only need to do the relevance judgment of attributes once for all the patients in one patient cohort.
The selected attributes are then extracted from FHIR and transformed into a simple JSON file . First, the resources were flattened, i.e. we changed the nested structure of the original FHIR resource such that each field in the newly transformed JSON only contains atomic values to make it easier readable and usable. Secondly, we removed unnecessary and redundant information based on the patient ID and date. Lastly, in cases where information stems from the same event but is distributed between multiple resources, the information is merged.
For the visualization of the extracted attributes as a timeline, we use the AnyChart JavaScript visualization library2. The AnyChart library provides interactive functionality designed specifically for temporal data and supports both, time point events (a specific point in time) and time range events. We map the patient information accordingly, e.g., a physical examination is encoded as a time point event, and a medication plan with start and end date to a time range event. AnyChart supports zoom, to adjust the time range showing either a global view or focusing on a particular time period only.
In this section, we will describe the application of our approach to a specific use case. Due to the confidentiality of patient data, we use a synthetic, but realistic patient to showcase the visualization.
We use the FHIR-based melanoma database from the University Hospital in Essen. As selection criteria, we used the primary tumor diagnosis to only include melanoma patients: Condition.category = |C0677930 (Primary Neoplasm) AND Condition.code= C43.*. Each file contains the complete tumor documentation for a patient and other relevant information (e.g. laboratory observations and progress notes). Overall, our cohort includes all possible patients with the start of treatment between 2001 and 2023, with the majority of the cases being from 2014 to 2023 (about 90%). In total, there are 1899 melanoma patients (46.18% female, 53.77% male) in the data set. The vast majority is from the North Rhine-Westphalia.
An example of the relevance judgment of patient attributes is shown in Table 1 (see Appendix A for the full list). Two experienced dermato-oncologists have rated the attributes based on the national guidelines for melanoma treatment5. We selected the attributes which at least one clinician rated as very important for inclusion in the patient trajectory visualization. An example of JSON output based on these selected attributes can be seen in Appendix B.
The visualization of the patient trajectory shows general patient information along with the timeline. Figure 2, top, shows the overview including all attributes selected in the relevance judgment step. It presents core diagnostic and treatment information for the patient. To further facilitate clinicians to select the information of interest without being distracted by other information, we also added a filter function, to show only one special attribute or one special combination of diferent attributes (cf. Figure 2, bottom). 5http://www.leitlinienprogrammonkologie.de/leitlinien/melanom, last accessed 31.08.2023
This work presents a method to visualize FHIR healthcare data as patient trajectory which aims to support clinicians in accessing information from EHR (available in FHIR) and in obtaining a comprehensive and concise picture of the patient’s disease progress and physical condition. The overall framework is extensible and applicable to other patient cohorts and use cases. However it also has some potential limitations, e.g., relevant attributes might need to be updated due to the changing of guidelines of treatment and ICD versions, and the choice of web technology, some browsers may not support JavaScript. An evaluation of the usability, usefulness and impact of this visualization is planned for future work.
Careplan
Pathological TNM staging (pTNM) Version of pTNM T, N, M Stage separately (pTNM) Residual-State (pTNM) Sentinel Lymph nodes positive (pTNM) Sentinel Lymph nodes examined (pTNM) Regional Lymph nodes positive (pTNM) Regional Lymph nodes examined (pTNM) Clinical TNM staging (cTNM) Version of cTNM T, N, M Stage separately (cTNM)
Results of genetic analysis, if genetic findings frequency and name of the mutation
Result of the regular check up (e.g., No tumor detection, Questionable findings, Progress, Decline, etc.) 2 2 2 2 2 2 2 2 2 2 1 1 2 1 2 1 1 0 2 1 2 1 1 1 1 2 2
Intention (e.g., adjuvant, curative) 2 Status (completed or not) 2 Reason of status 2 B. Example JSON File 1 1 Example data extract from a synthetic, but realistic patient (excerpt due to space constraints) { }, "stages": [ }, "patid": "Patient/01", "dt_record": "2014-02-06", "cat_examination_type": [ "Laboratory procedure", "physical exam" ], "cat_reasons": [ "Initial presentation", "Treatment planning" }, ... ], "radiotherapy": [ { "patid": "Patient/01", "dt_start": "2016-06", "dt_end": "2016-06-29", "cat_intention": [ "palliative", "adjuvant" ], "cat_status": "completed", "cat_reason_end": "regular ending"
}, ... ], ...