# Data, Scripts, and Output for the use-case **Perioperatieve Stress Response**
This repository contains the raw data used for the paper "XXX" (YYY). The repository also contains the scripts and output for the simulation results re-enacting the screening phase using active learning in ASReview ([Van de Schoot et al., 2021]).
## Background information
Glucocorticoids, especially cortisol, are essential in the adaption to stressful situations. Patients suffering from adrenal insufficiency are not able to produce cortisol adequately themselves and are reasonably administered extra in demanding circumstances, such as in surgical interventions. There is
currently, however, no consensus regarding peri-operative cortisol
administration in these patients [Salem et al., 1994]). This becomes clear giving the variation in practical guidelines differing not only in the type of corticosteroid, but also in timing, dose, and duration of administration. The common thread is, however, to supply relatively very high doses in order to minimize the chance
of having an adrenal crisis. Since exposure to high glucocorticoids levels can induce post-operative comorbidities and impair the recovery process ([Chilkoti, et al., 2019], [Nguyen et al., 2014]), it is essential to develop more rational based regimens. It is therefore necessary to understand the physiological alterations in cortisol levels
during and after surgery. Hence, in this study we systematically evaluated previous literature for peri-operative cortisol changes in different categories of surgery to answer the question: What are the physiological changes in cortisol secretion during and after different types of surgical procedures?
Due to changing operation techniques over time ([Kehlet and Wilmore, 2002]) which presumably influence the impact of surgery on the stress response, we arbitrarily limited the search for publications from 2010 onwards.
To increase the quality of a systematic review on Perioperatieve Stress Response, Savas et al. applied a 2-step screening. First, a search was performed identifying 1.155 records published >2010 of which 30 were deemed as relevant. Then, the search was conducted for papers published <2010 identifying 4.305 records. The labels of the first search were used as training data for the second search using ASReview. Then, 310 records were
screened and the first 30 additionally found relevant records were inspected for full-text in detail and 5 appeared to be relevant.
The search terms for the original search are available in the file `Search.strategy.pdf` as used for the publication [**ADD REFERENCE**].
## Raw data
The following data are available:
- `perioperatieve_stress_response_original_data_before_2010` containing 4.305 unlabelled records;
- `perioperatieve_stress_response_original_data_2010_onwards` containing 1.155 labelled records;
The data before and after 2010 has been combined into one dataset.
The results are stored in `raw_data/output` and contains the file `perioperatieve_stress_response_all_data.csv` containing labelling decissions for the the subset >2010 while the subset <2010 is unlabbeled.
The raw data `perioperatieve_stress_response_all_data.csv` contains, among many others, the columns:
- `title` / `abstract note` of 5460 records;
- `include` with 30 relevant and 1125 irrelevant records, and 4305 unlabelled records.
Wordcloud of the included titles+abstacts:
![enter image description here]
## Installation requirements
Run the following code in the CLI:
pip install -r requirements.txt
The data files in the data-folder were used for running a simulation study. To run the simulation, run
The results are stored in `output/simulation`.
The simulation was conducted on the original data with 93 runs with each relevant record being a prior inclusion and 10 randomly chosen irrelevant records. In each run the same 10 irrelevant records have been used. To extract this information run
The results are stored in `output/tables`.
The dataset characteristics are obtained with
and stored in `output/tables`.
The metrics resulting from the simulation study per run, can be obtained with
and are stored in `output/tables`.
The recall plot:
![enter image description here]
On average, after screening 19% of the records (n = 209), you would have found 95 % of all the relevant records. In other words, the time that can be saved using active learning
expressed as the percentage of records that do not have to be screened is 78% (sd=0.77), while still identifying 95% of the relevant records. This metric is also known as the Work Saved over random Sampling at 95% recall (WSS@95).
Another way to interpret the results is the RRF metric. The RRF@10 is 73% (sd=1.9), meaning that after screening 10% of records, already 73% of the relevant records have been identified.
The relevant record that was in row 1100 of the dataset was the easiest to find over all trials (tITLE: *Influence of spinal and general anesthesia on the metabolic, hormonal, and hemodynamic response in elective surgical patients*). Discovering this record took screening 14 records on average, that is screening 1.19% of all records in the perioperatieve_stress_response_all_data dataset. The record that was most difficult to find was in row 1101 of the dataset, which was discovered after screening on average 254 records, 22.24% of all records in this dataset
(*Effect of electroacupuncture at Zusanli (ST36) and Sanyinjiao (SP6) acupoints on adrenocortical function in etomidate anesthesia patients*).
The file `../data/perioperatieve_stress_response_all_data.csv` was uploaded in ASReview and the 30 relevant records plus the irrelevant records screened in the >2010 dataset, were used as prior knowledge for the <2010 dataset.
210 records were screened via the default model settings in version 0.16, see the projecct file `peri-operative-serum-cortisol-levels.asreview`. In total, 131 were denoted as relevant, see the file
The first 30 additionally found relevant full text articles were
screened for relevance, see the file
`asreview_result_peri-operative-serum-cortisol-levels_MS.xlsx`. 17 papers were not accessible, 10 papers with an English abstract were written in a non-English language and 5 of these papers might be worth translating. 3 studies need in depth discussion.