Eligibility criteria

Table 1 presents the inclusion and exclusion criteria, developed in accordance with the PICOTS framework, across six domains: population, intervention/index test, comparator, outcomes, timing, and setting [30].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Eligibility criteria for the systematic review framed according to the PICOTS system.

https://doi.org/10.1371/journal.pone.0342524.t001

Population

Outcome

Commonly used instruments to assess compassion fatigue include the Compassion Fatigue Self Test Scale (CFSTS), the Professional Quality of Life Scale (ProQOL), and the Compassion Fatigue Short Scale (CFSS) [31–33]. The CFSTS comprises two dimensions (compassion fatigue and burnout) with a total of 40 items. Figley subsequently revised the scale by adding a compassion satisfaction dimension, increasing the total number of items to 66. Cronbach’s alpha values for compassion satisfaction, burnout, and compassion fatigue were 0.87, 0.90, and 0.87, respectively [31]. The ProQOL contains 30 items across three dimensions: compassion satisfaction, burnout, and STS. Cronbach’s alpha values for the three dimensions were 0.88, 0.75, and 0.81, respectively [32]. The CFSS includes two dimensions (secondary trauma and burnout) with 13 items, and its Cronbach’s alpha is 0.90 [33]. Given that other instruments are also used to measure compassion fatigue, no restrictions will be placed on the measurement tools in this review. All studies reporting prediction models for compassion fatigue will be included, and the outcome definitions, measurement instruments, and reference criteria used to ascertain compassion fatigue will be summarized in the final table. The primary outcome is compassion fatigue, with prediction models for burnout and STS also included as related outcomes.

Timing

There is no limit on the predicted time point, and any study predicted at any time will be included.

Setting

All nursing practice settings, including but not limited to hospitals, community care, home care, hospice care, and long-term care facilities.

Exclusion criteria

Studies will be excluded if they: a) involve nursing students, physicians, or other healthcare personnel (i.e., populations other than licensed/registered nurses); b) assess compassion fatigue using measurement instruments without established reliability and validity evidence; c) are not peer-reviewed (e.g., conference abstracts); d) have no accessible full text; e) do not report key information required to extract or evaluate prediction models (e.g., predictors and model performance measures); f) are published in languages other than English or Chinese.

Types of studies and limits

No restrictions will be placed on study design. The search will cover the period from database inception to October 31, 2025.

Search methods for the identification of studies

Search strategy

The search strategy will have three stages. First, a pilot search in PubMed will examine the characteristics of relevant records, such as titles, abstracts, and keyword patterns. Based on these results, we will consult nursing experts to finalize search terms and build the Boolean search string. Second, two researchers will independently run the searches across the databases using the specified strategies. Third, we will screen reference lists of included studies through backward citation searching. The PubMed search strategy appears in Table 2. Preliminary search strategies for all databases are in S5 Appendix.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Full search string for the database “Pubmed”.

https://doi.org/10.1371/journal.pone.0342524.t002

Data collection and analysis

Selection process

We will use dual independent screening to ensure the accuracy and reliability of study selection. Two reviewers with nursing expertise will independently screen the titles and abstracts of all retrieved records against prespecified eligibility criteria. Records deemed potentially relevant or with uncertain eligibility will be carried forward to full-text assessment. The same two reviewers will then independently assess the full texts of the retained articles. Any disagreements will be resolved through discussion and, if necessary, adjudication by a third reviewer. The study selection process will be presented in a PRISMA flow diagram (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart of study selection.

https://doi.org/10.1371/journal.pone.0342524.g001

Data extraction

Two researchers will independently extract data using a standardized data extraction form (e.g., in Microsoft Excel) and will then cross-check all extracted information to ensure accuracy. Any discrepancies or inconsistencies will be resolved by re-examining the source articles and reaching a consensus.

The data extraction table was developed with reference to the CHARMS checklist and comprises 11 fields, including study objective, data source, participants, predicted outcome(s), candidate predictors, and other relevant information [27]. Model performance measures will include calibration and discrimination, as well as Decision Curve Analysis (DCA). DCA facilitates identification of the optimal clinical decision strategy by visually comparing net benefit across a range of threshold probabilities [34,35].

Because compassion fatigue is assessed using standardized scales, we will extract both continuous compassion fatigue scores and the prevalence (i.e., the proportion of participants classified as having compassion fatigue) within each sample. We will also extract the criteria used to define compassion fatigue (e.g., for the Chinese version of the ProQOL, compassion fatigue will be described if any one of the three dimensions—compassion satisfaction (<37 points), occupational burnout (>27 points), or secondary traumatic stress/emergency response (>27 points)—meets the specified threshold).

Critical appraisal

Two reviewers will independently assess the risk of bias and applicability of the included studies using the Prediction model Risk Of Bias Assessment Tool (PROBAST) [28]. Assessments will be performed in accordance with the PROBAST guidance, including evaluation of relevant signalling questions and assignment of domain-level judgments. Risk of bias will be assessed across four domains: participants, predictors, outcome, and analysis, whereas applicability will be assessed across three domains: participants, predictors, and outcome.

For each domain, risk of bias and applicability concerns will be rated as low, high, or unclear, as appropriate. An overall risk-of-bias judgment will be assigned to each prediction model in accordance with the PROBAST recommendations. If a study reports multiple prediction models, external validations, outcomes, time points, or analytical approaches, each eligible prediction model or validation analysis will be assessed separately. Decisions regarding which model, outcome, time point, or analysis to extract and appraise will be prespecified according to the eligibility criteria and data-extraction plan, rather than based on the reported results.

Disagreements between the two reviewers will be resolved through discussion and, when necessary, adjudication by a third reviewer. To minimize the risk of selective analytical decisions and potential bias, all appraisal rules and any deviations from the planned procedures will be documented. Assessment results will be summarized in tables that present domain-level and overall judgments on risk of bias and applicability.

Qualitative data synthesis of prediction models

Extracted data will be presented in tabular format, summarizing study characteristics, participant demographics, and model details. Descriptive analyses will be performed on the prevalence of compassion fatigue (mean scores and prevalence rates), the frequency of reported predictors, and model performance metrics. Performance metrics will include discrimination (e.g., sensitivity, specificity, AUC, R²), calibration (calibration plots and Hosmer–Lemeshow test results), and clinical utility (decision curve analysis, DCA). Given the anticipated heterogeneity in model types (e.g., logistic regression for binary outcomes vs. linear regression for continuous scores) and study settings, a meta-analysis is unlikely to be feasible. To account for this, we will conduct a structured narrative synthesis. Studies will be grouped by key characteristics—including clinical setting, sample size, and validation status—to compare findings across these categories. Within each group, findings will be synthesized descriptively to explore patterns and potential sources of heterogeneity. Compassion fatigue, burnout, and STS will be considered the three outcomes of this study and reported separately, in line with the analytic approach described above.

Reporting and presentation of findings

The systematic review will follow the PRISMA 2020 statement [29]. Upon completion, all extracted datasets, assessment tables, and supporting materials will be made publicly available as supplementary materials with the final publication.

Strengths and limitations

A key strength of this study is that it systematically reviews existing prediction models for compassion fatigue in nurses, describing their model characteristics, predictors, predictive performance, and other relevant metrics. This review may help nursing managers select appropriate prediction models and serve as a reference for future research on models for compassion fatigue. However, this study has limitations. Owing to differences in nurses’ baseline characteristics across studies and the inconsistent use of compassion fatigue measurement tools, substantial heterogeneity was present. Therefore, we did not undertake a meta-analysis; we conducted only a descriptive synthesis.