Modified PRISMA Protocol

• Modified PRISMA protocols are tailored adaptations of standard guidelines that integrate domain-specific modifications, AI support, and enhanced reporting steps.
• They add extended reporting items such as dataset transparency, inter-rater reliability, and quality appraisal metrics to ensure rigorous and reproducible reviews.
• These protocols streamline workflows in fields like non-profit healthcare and AI-assisted literature reviews, reducing screening times and improving result validity.

A modified PRISMA protocol refers to a systematic adaptation or extension of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to suit nonstandard systematic evidence synthesis contexts. These adaptations frequently arise in methodological frameworks (e.g., scoping reviews, AI-assisted SLRs) characterized by new demands on transparency, reproducibility, or computational integration. Notably, two prominent examples from recent literature are (1) the theory-driven adaptation of PRISMA-ScR for AI-enabled knowledge sharing in non-profit healthcare (Ongala et al., 10 Mar 2025), and (2) the PRISMA-DFLLM extension for systematic literature reviews leveraging finetuned domain-specific LLMs (Susnjak, 2023).

1. Rationale and Emergence of PRISMA Modifications

Standard PRISMA guidelines primarily target traditional systematic reviews and meta-analyses. However, new domains and workflows—such as scoping reviews, interdisciplinary health informatics, and AI-augmented synthesis—expose limitations related to labor-intensity, domain heterogeneity, and increasing demands for computational reproducibility. The principal motivations for modifications are:

• Accommodating nontraditional evidence types or broad exploratory aims (as in scoping reviews).
• Integrating computational tools or AI-based automation within the review pipeline.
• Providing explicit, transparent documentation of novel processes or domain-specific considerations.

In both (Ongala et al., 10 Mar 2025) and (Susnjak, 2023), modifications enforce protocol-level rigor where standard PRISMA falls short, introduce custom reporting items, and embed additional steps (e.g., quality appraisal, technical ML details, ethical mapping) to ensure that methodological heterogeneity does not compromise replicability or inferential trust.

2. Protocol Architecture and Workflow Specifics

PRISMA-ScR Adaptation: AI-Enabled Knowledge Sharing (Ongala et al., 10 Mar 2025)

The adapted protocol is a theory-driven scoping review pipeline traversing standard PRISMA-ScR phases with additional domain-driven steps. The workflow includes:

1. Identification: Searches across PubMed, Scopus, IEEE Xplore, and first 200 Google Scholar hits. Supplementary sourcing involves reference lists and organizational repositories (WHO, USAID). Duplicates are purged via EndNote 20 (automatic/manual).
2. Screening: Two independent reviewers use Rayyan.io; inclusion requires mention of AI plus knowledge sharing in non-profit healthcare, with exclusion of for-profit, purely technical, or opinion categories.
3. Eligibility: Full-text review by two independent reviewers, discrepancies by consensus or third-party.
4. Inclusion: All studies are charted and mapped explicitly to the Resource-Based View (RBV), Dynamic Capabilities Theory (DCT), or Absorptive Capacity Theory (ACT).

PRISMA-DFLLM Extension: LLM-Augmented SLRs (Susnjak, 2023)

PRISMA-DFLLM inserts three AI-centric pillars into the reporting architecture:

1. Finetuning Dataset Construction: Detailing preprocessing, format, curation, and composition of instructional data extracted from the SLR corpus.
2. LLM Finetuning Specifications: Technical reporting on base model choice, parameter-efficient finetuning (e.g., LoRA, QLoRA), hyperparameters, and calibration.
3. Model Evaluation and Alignment: Pre/post-finetuning benchmarking, error analysis, alignment analysis, and metric justification.

A modified living review process (Item 26e) and explicit legal/ethical availability checklist items (Items 30, 31) are also integrated.

3. Reporting Innovations and Additional Checklist Items

Domain-Thematic Adaptations (Ongala et al., 10 Mar 2025)

• Theoretical Mapping: Mandatory coding of each included paper to RBV, DCT, or ACT for structured cross-paper synthesis.
• Inter-rater Reliability Assessment: Statistical quantification of reviewer agreement, using Cohen’s $\kappa = \frac{P_o - P_e}{1 - P_e}$, where $P_o$ is observed, $P_e$ expected agreement by chance.
• Methodological Quality Appraisal: Brief risk-of-bias commentary contextualizes diverse research designs.
• Ethics Signaling: Explicit reporting of ethical or data governance content.
• Google Scholar Limit: Search hits capped at 200 due to diminishing relevance.

Technical and Computational Extensions (Susnjak, 2023)

• Dataset Transparency (Items 16a–e): Detailing raw-text cleaning, automated/manual extraction, augmentation, and statistical composition.
• Model Training (Items 17a–d): Requiring disclosure of model details, PEFT method, optimizer, dropout, and temperature scaling.
• Evaluation and Alignment (Items 18a–f): Stipulating pre-finetuning baselines, reproducibility protocols, qualitative/quantitative analyses, and concrete evaluation metrics (e.g., ROUGE, F1).
• Ongoing “Living” Review Mechanism (26e): Mandating processes for incremental model updating as new literature emerges.
• Data/Code/Model Availability (30): Requiring open sharing of training data and model artifacts.
• Ethical/Legal Compliance (31a–c): Documenting potential model harms, copyright, and compliance.

4. Data Extraction, Processing, and Quality Assessment

The modified protocols utilize highly structured data extraction templates:

Data Category	PRISMA-ScR Adaptation (Ongala et al., 10 Mar 2025)	PRISMA-DFLLM Extension (Susnjak, 2023)
Bibliographic Metadata	Authors, year, journal, country, non-profit type	Included as source; forms base finetuning corpus
Methodology	Study design, AI method, domain, theory mapping	Dataset preprocessing, split, augmentation
Quality Assessment	Methodological rigor, inter-rater reliability notes	Benchmarking, evaluation stability
Ethical Considerations	Dedicated field for each record	Explicit reporting, legal and privacy compliance

This suggests that both adaptations foreground the need for rich metadata and evidence-strength contextualization, either for theory synthesis or for model accuracy and alignment reproducibility.

5. Impact, Challenges, and Future Directions

Efficiency and Scalability: PRISMA-DFLLM enables automation of screening, extraction, and synthesis, with documented reductions in labor and cycle times—e.g., pilot studies report screening time reductions from 12 to 4 hours per 1,000 records, with precision/recall above 95% (Susnjak, 2023). In non-profit healthcare synthesis, the adaptation ensures comparability across diverse organizational theory frames (Ongala et al., 10 Mar 2025).

Challenges: Persistent issues include file heterogeneity, fine-grained PDF-to-text processing, ambiguity in theory coding, legal restrictions (copyright), and AI-related risks (bias, hallucination). Alignment metrics and validation of automated extractions are highlighted as critical open areas.

Future Directions: Adoption of modified PRISMA protocols—especially with embedded AI reporting—sets a precedent for living, incremental systematic reviews and transparent, open-source artifact sharing. This development is poised to influence adjacent reporting standards (e.g., CONSORT, STROBE) and expand reproducible evidence synthesis across computational and organizational research domains.

6. Summary of Key Modifications

Modification/Extension	Purpose	Reference
Theoretical Mapping	Ensures framework-based comparability	(Ongala et al., 10 Mar 2025)
Inter-rater Reliability Calculation	Documents reviewer consistency	(Ongala et al., 10 Mar 2025)
Quality and Risk-of-Bias Appraisal	Contextualizes heterogeneous evidence	(Ongala et al., 10 Mar 2025)
Dataset/Model Reporting Items	Full ML transparency and reproducibility	(Susnjak, 2023)
Living Review Process	Enable ongoing, incremental synthesis	(Susnjak, 2023)
Legal/Ethical Documentation	Mitigate risks of model harm or misuse	(Susnjak, 2023)

These modifications define the new state-of-the-art in systematic review reporting and operational transparency when integrating advanced computational or theoretical methodologies.

7. Conclusion

Modified PRISMA protocols instantiate transparent, reproducible, and context-appropriate systematic review workflows equipped to manage heterogeneity—whether theoretical, technical, or computational. The dual cases of theory-driven scoping review in non-profit healthcare and LLM-powered SLRs exemplify this evolution, demanding precise theoretical mapping, explicit computational reporting, and rigorous assessment of quality, ethics, and reproducibility (Ongala et al., 10 Mar 2025, Susnjak, 2023). These frameworks collectively signal a transition toward systematic reviews that are not only more efficient but also auditable and adaptable to rapid advances across both organizational and machine learning-driven evidence domains.