Development of Machine Learning Models to Predict Candidemia in Hospitalized Adult Patients
Clinical Snapshot
PICO Framework
| P — Population | Adult inpatients admitted to a single Japanese tertiary hospital (Shin Koga Hospital) who underwent blood cultures at least 14 days after admission, April 2014 – March 2022 |
| I — Intervention | LASSO logistic regression model (and three comparator ML models) trained on electronic medical record variables available on or before the date of blood culture collection |
| C — Comparator | Blood culture as the reference standard for candidemia diagnosis; model performance compared across extreme gradient boosting, random forest, LASSO logistic regression, and standard logistic regression |
| O — Outcomes | Discriminative performance (AUROC) for predicting candidemia; optimism-corrected AUROC via 1,000-bootstrap internal validation |
Bottom Line
This single-centre Japanese study develops a LASSO logistic regression model to predict candidemia in hospitalised adults using routinely available EMR data, reporting AUROCs of 0.880–0.913 with bootstrap-corrected values of 0.851–0.876. While the discriminative performance is promising and the methodological approach (cross-validation, bootstrap optimism correction, comparison of multiple algorithms) is sound, the study has critical limitations that preclude clinical implementation. The cohort is small (47 candidemia events), single-centre, and restricted to patients already selected for blood culture after ≥14 days of admission — introducing substantial spectrum and verification bias. Crucially, no threshold-specific sensitivity, specificity, or likelihood ratios are reported, and no decision curve analysis is presented, making it impossible to assess whether the model would improve clinical outcomes over current practice. No external validation has been performed. For Australian clinicians, the model cannot be recommended for adoption without prospective external validation in local populations, specification of an actionable decision threshold, and demonstration of net clinical benefit. This study represents a methodologically reasonable proof-of-concept that warrants prospective multicentre validation before any clinical translation.
Key Findings
P Value: Not reported
Effect Size: AUROC 0.880 (Complete case dataset, 13 variables); AUROC 0.913 (Imputed full-variable dataset, 36 variables)
Primary Outcome: Discriminative performance (AUROC) of LASSO logistic regression for predicting candidemia in hospitalised adults using pre-culture EMR data
Nnt Or Sensitivity: Optimism-corrected AUROC: 0.851 (Complete case) and 0.876 (Imputed full-variable). Sensitivity, specificity, and likelihood ratios at specific thresholds not reported.
Confidence Interval: Complete case: 95% CI 0.836–0.924; Imputed full-variable: 95% CI 0.878–0.948
Clinical Application
The model uses only automatically extractable EMR variables, which is a practical strength. However, implementation requires local EMR integration, prospective validation in the target setting, and definition of an actionable decision threshold with acceptable sensitivity/specificity trade-offs. The absence of threshold-specific metrics in the paper is a significant barrier to immediate clinical deployment. Candidemia is a notifiable condition in several Australian states and carries mortality rates of 30–50% in Australian ICU cohorts. The RACGP and ACSQHC advocate for antimicrobial stewardship frameworks that include early identification of invasive fungal infections. Antifungal agents used in candidemia treatment (echinocandins such as caspofungin, anidulafungin; fluconazole) are listed on the PBS for specific indications. TGA-approved diagnostics for candidemia include blood culture and beta-D-glucan assays. This model, if externally validated in Australian hospitals, could complement existing antifungal stewardship programmes. However, direct application is premature without validation in Australian patient populations, where Candida species distribution and resistance patterns (particularly fluconazole-resistant Candida glabrata/C. auris) may differ from Japanese cohorts. Australian hospital EMR systems (e.g., Epic, Cerner, Meditech) would require local variable mapping. Adult inpatients at risk of candidemia who have been hospitalised for ≥14 days and are being considered for blood culture, particularly those with known risk factors (prolonged ICU stay, central venous catheters, broad-spectrum antibiotics, total parenteral nutrition, immunosuppression)
Abstract
PURPOSE: We aimed to develop a generalizable machine learning model that leverages electronic medical record (EMR) data to predict candidemia using information available on or before the date of blood culture collection. METHODS: This study included adult patients admitted to Shin Koga Hospital (April 2014-March 2022) who underwent blood cultures at least 14 days after admission. We prepared two datasets: a "Complete case dataset" (13 variables, no missing values) and a "Imputed full-variable dataset" (36 variables, single imputation). We evaluated the discriminative performance of four machine learning models, including extreme gradient boosting, random forest, least absolute shrinkage and selection operator (LASSO) logistic regression, and logistic regression using stratified 5-fold cross-validation by comparing their area under the receiver operating characteristic curve (AUROC). The model with the highest mean AUROC was selected, assessed on the entire dataset, and internally validated using 1,000 bootstrap replicates. RESULTS: A total of 919 patients were included, of whom 47 (5.1%) had candidemia. The LASSO logistic regression demonstrated the best performance and was selected as the final model. The AUROC (95% CI) was 0.880 (0.836-0.924) for the Complete case dataset and 0.913 (0.878-0.948) for the Imputed full-variable dataset. The optimism-corrected AUROCs were 0.851 and 0.876, respectively. CONCLUSION: We developed two versions of the LASSO logistic regression model using only information automatically extracted from EMR, considering real-world data availability. These models have the potential to serve as practical tools to support early diagnosis and therapeutic intervention for candidemia.
References
- 1.Hashimoto, K., Matsumoto, K., & Murotani, K. (2026). Development of machine learning models to predict candidemia in hospitalized adult patients. The Kurume Medical Journal. https://doi.org/10.2739/kurumemedj.MS721234017
Related Research
Journal of medical Internet research
Advancing Human-Centered AI in Clinical Decision Support: Sociocognitive Human-in-the-Loop Study in HIV Care
2 Aug 2026
JBJS reviews
One-Stage or Two-Stage Revision for Infected Shoulder Arthroplasty: A Head-to-Head Meta-Analysis of Comparative Studies
30 July 2026
Signal transduction and targeted therapy
Mucosal immunity and vaccine development
27 July 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service