C. difficile Outcome Model (version 1.13)

Age:
# of Prior CDI Episodes (0=initial occurrence):
Duration of Any Prior Hospital Episode (0 if unavailable):
Antibiotic Days (within prior 60d):
PCR Cycle Threshold:

Fever (Tmax > 100.4) (uncheck if unavailable):
Hypotension (SBP < 90):
Pressors:
WBC > 15 (uncheck if unavailable):
Creatinine > 1.5 (uncheck if unavailable):
Lactate > 1.9 (uncheck if unavailable):

Onset:

Community-Onset (Dx on hospital days 0-3 and no recent hospitalization (3mo.))
Community-Onset Healthcare-Associated (Dx on hospital days 0-3 plus recent hospitalization (3mo.))
Hospital-Onset (Dx on hospital day 4 or greater)

Initial Anti-CDI Therapy:

Vancomycin Monotherapy
Fidaxomicin Monotherapy
Metronidazole Monotherapy
Dual Therapy (e.g., vancomycin plus metronidazole)

CDI Outcome Predictions:

SHapley Additive exPlanations (SHAP)

Force Plot: 60-day Recurrent CDI Prediction

Force Plot: CDI-attributable Severe Outcomes (ICU, Shock, Surgery, or Death)

(SHapley Additive exPlanations (SHAP) values help explain how important each of this patient's inputs were in calculating predicted probabilities using the neural network model.)

Note. For boolean variables, zeroes indicate 'no' (vs. 1='yes'). Force plot 'base values' represent average model outputs over the training set and thus may differ slightly from expected probabilities from the entire (1,660 case) dataset, as shown in the table.

Disclaimer: This tool has not been adequately validated for clinical use and is intended for demonstration and research purposes only.

References

1. Madden GR, Boone RH, Lee E, Sifri CD, and Petri WA. Predicting Clostridioides difficile Infection Outcomes with Explainable Machine Learning. eBioMedicine. 2024 Aug 15;106:105244.

2. Lundberg SM and Lee SI. “A unified approach to interpreting model predictions.” Advances in Neural Information Processing Systems (2017).