Supporting code for the workshop in Process-Oriented Data Science for Healthcare (PODS4H), part of ICPM (International Conference on Process Mining) 2023. The code shows how to develop sample cohorts of hospitalised patients with COVID-19 and generate summary features describing rehabilitation need from routine contacts with an AHP (Allied Healthcare Professional). The Process Mining functionality uses bupaR (fuzzy mining) to generate an event log from the timestamped contacts data and visualise patient interactions (during Wave 1 or Wave 2) using frequency-based process maps. Other functions allow additional analytics of the event log, such as plotting the precedence matrix, dotted chart, throughput and idle time over patient groups.
This is a case study that uses dummy data, based on NHS Lothian linked routine data from Electronic Health Records. In our case study, all data was collected from EHRs and national registries previously anonymised by the DataLoch service (Edinburgh, United Kingdom) and analysed in a Secure Data Environment. Details on these datasets are available on the Metadata Catalogue. The provided dummy datasets describe:
- The COVID-19 linked episodes of care (episodes.rda)
- The performed lab tests for COVID-19 positivity (lab_tests.rda)
- TrakCare pathways describing routine timestamped hospital visits (trak_care.rda)
- AHP-related timestamped contact visits including session duration (contacts.rda)
- TrakCare admission episodes (trak_adm.rda)
- TrakCare discharge episodes (trak_disch.rda)
- Scottish Morbidity Records (SMR-coded phenotypes), describing previous long-term conditions (smr_phenotype.rda)
- Death records (deaths.rda)
This is an unofficial R library built in RStudio and Rtools 4.3.0 (recommended R >= 3.5.0). The required packages are listed in DESCRIPTION. Note that DiagrammeRsvg (>= 0.1), rsvg (>= 2.4.0) are typically required to produce pdf visualisations.
1. Clone the project from this repository to your local directory using:
git clone https://github.com/kgeorgiev42/cov19.rehab.analysis.git
2. Install the R library using RStudio or a similar interpreter. This can be done by loading this repository as a project and clicking Build -> Install and Restart, or:
Rcmd.exe INSTALL --no-multiarch --with-keep.source cov19.rehab.analysis
3.0. The R library is now ready to use. The simplest way of launching the experiments is to use the pre-built .rda files in the data/ folder. However, if you would like to tweak the data, you can do so by modifying the parameters in generateData.R and running:
cov19.rehab.analysis::gen_cov_data()
to re-create all datasets or run the respective function of the modified dataset. Each R function contains adjustable parameters to change the target number of records, patients, records per patient, timestamps and likelihoods of sampling for categorical data.
3.1. To produce the cohort files (patient summary file and rehabilitation contacts file) you can run:
cov19.rehab.analysis::gen_cohorts()
This will re-run the code to sample COVID-19 positive patients and extract summary rehabilitation needs data based on trak_care.rda and contacts.rda. This functionality can also be tweaked in generateEvalCohort.R.
3.2. To produce the event log required to create the process maps you can run:
cov19.rehab.analysis::pm.generate_event_log(out_path='data/pm_event_log.rda', positive_only=1)
This will extract the timestamps from only the COVID-19 positive hospital episodes and save this in data/.
4. The main functionalities include:
4.1. Generating a summary cohort table, stratified by COVID-19 wave.
cov19.rehab.analysis::pm.get_summary_table()
4.2. Plotting the frequent traces using a process map:
cov19.rehab.analysis::pm.plot_frequent_traces(out_path='results/freq_map.pdf', maf=5, minf=50, wave='Second')
The above will produce a process map describing traces within COVID-19 Wave 2, restricting the flows across the full map to at least 50 instances and requiring at least 5 patients with a specific interaction.
4.3. Plotting the pathways for a specific age group:
cov19.rehab.analysis::pm.plot_frequent_traces(out_path='results/freq_map_subage.pdf', maf=5, minf=50, wave='Second', age_group='80+')
The above will produce the same subset as the previous map but will show the traces present only in patients aged 80 and above.
4.4. Plot a precedence matrix for a specific wave:
cov19.rehab.analysis::pm.plot_precedence_matrix(out_path='results/prec_matrix.png', wave='Second')
The above will plot a table of antecedent-consequent pairs, representing the frequent interactions between specialist nodes, across those in Wave 2.
4.5. Other plots:
### Plot the total time required to complete the pathway across age groups
cov19.rehab.analysis::pm.plot_throughput_time(out_path='results/throughput_time.png', units='days')
### Plot the cumulative time without an AHP-related session in the hospital
cov19.rehab.analysis::pm.plot_idle_time(out_path='results/idle_time.png', units='days')
### Plot a cumulative dotted chart showing the accumulated population over time by COVID-19 wave
cov19.rehab.analysis::pm.plot_dotted_chart(out_path='results/dotted_chart.pdf')
AHP and healthcare provider codes in dataset:
- NURSE (care delivered by a specialist nurse)
- PT (physiotherapist)
- OT (occupational therapist)
- DT (dietician providing nutritional support)
- SPL (speech and language therapist)
- CP (clinical pharmacist)
- PC (specialist in palliative care)
- INFSV (infection prevention service)
- OTHER (unidentified care provider)
- Provide documentation for R functions.
- Modify time to therapy and length of stay to provide more realistic pathways and admission episodes.
- Add functionality for linear regression analysis.
- Add coding and functionality for rehabilitation-related activities.
We acknowledge the funding provided by the Sir Jules Thorn Charitable Trust PhD award (grant no. 21/01PhD).