Recurrent klebsiella pneumoniae pyogenic liver abscess: Developing a literature-informed 0–2 scoring framework from a solved index case

Background: Recurrent Klebsiella pneumoniae pyogenic liver abscess (KPLA) is clinically important because persistent or recurrent cavities may reflect ongoing infection, inadequate source control, or an occult gastrointestinal lesion. We aimed to develop a simple, literature-informed scoring framework that organizes high-yield features associated with recurrent or persistent KPLA, using a solved clinical case as the test case.

Methods: We performed a human-directed review of adult peer-reviewed literature on pyogenic liver abscess and KPLA, retaining studies that reported clinically relevant variables across comorbidities, laboratory findings, imaging, microbiology, treatment response, and evaluation for occult gastrointestinal sources. A large language model was used only to organize and summarize information from the retained studies; article screening, inclusion decisions, variable selection, thresholds, and score assignments were performed and verified by the student investigator and faculty mentor. Each variable was assigned an ordinal score (0 = absent/low concern, 1 = intermediate concern, 2 = high concern) using explicit thresholds defined in a companion scoring table. The index case was mapped to the framework only after completion of the literature review. A companion workflow figure illustrates the supervised development process.

Results: Fourteen variables were retained: hyperglycemia burden, leukocytosis burden, recurrent inflammatory relapse, total bilirubin elevation, alkaline phosphatase elevation, mild transaminase injury, largest abscess diameter, radiographic non-resolution or interval enlargement despite therapy, recurrence after drainage and antibiotics, abscess culture positive for K. pneumoniae, concordant urinary isolate, occult appendiceal-cecal/colorectal abnormality, source-control

escalation, and delayed gastrointestinal-source evaluation. The index case involved a 76-year-old woman with diabetes whose urine and abscess cultures both grew K. pneumoniae. Despite drainage and antibiotic therapy, she experienced recurrence, minimal radiographic improvement, repeat drainage, and subsequent identification of a cecal lesion suspicious for malignancy despite a benign biopsy. When mapped to the framework, the case accumulated multiple high-concern features and fell into the high conceptual-risk category.

Conclusions: A supervised AI-assisted workflow can transform a solved infectious disease case and published literature into a transparent, auditable scoring framework for recurrent KPLA. This framework is not a validated clinical prediction model but rather a research and educational tool that makes variable selection, thresholds, and case-to-score mapping explicit. Prospective multi-case evaluation and formal assessment of educational utility are the next steps.