Claim Risk models built from variables collected before glucocorticoid exposure will predict Strongyloides-related treatment need and short-term complications better than models first assembled after steroids have started.

Rationale Glucocorticoids can rapidly alter eosinophil counts and symptom patterns, which may erase some of the screening signal clinicians depend on. A model trained on post-steroid variables may therefore be systematically miscalibrated even if it uses the same underlying predictors.

Testable prediction In autoimmune cohorts with paired pre-steroid and post-steroid data, the pre-steroid model will show superior calibration slope/intercept and lower Brier score for 90-day Strongyloides-related adverse outcomes than the post-steroid model.

Suggested study design

• Longitudinal cohort with baseline data before immunosuppression and repeat data 3-7 days after steroid initiation
• Outcomes: positive Strongyloides testing, empiric-treatment decision, hyperinfection/dissemination, or gram-negative sepsis within 90 days
• Compare logistic or Bayesian dynamic models using pre-steroid vs post-steroid covariates
• Prespecify subgroup analyses by eosinophilia, pulse steroids, and endemic exposure burden

Falsification condition If post-steroid models match or exceed pre-steroid models on calibration and clinical utility, the hypothesis fails.

Why this matters This is a biostatistics question with immediate bedside consequences: when steroid urgency prevents perfect workup, the timing of model inputs may determine whether the model is clinically trustworthy.

References

1. Requena-Mendez A, et al. Am J Trop Med Hyg. 2017;97(3):645-652. DOI: 10.4269/ajtmh.16-0923
2. Ming DK, et al. Trop Med Infect Dis. 2021;6(4):203. DOI: 10.3390/tropicalmed6040203
3. Fragoulis GE, et al. Ann Rheum Dis. 2023;82(6):742-753. DOI: 10.1136/ard-2022-223335
4. Buonfrate D, et al. Pathogens. 2020;9(6):468. DOI: 10.3390/pathogens9060468